THE COMPLETE

TECHNICAL PAPER PROCEEDINGS

FROM:
 A COMPARISON OF PON ARCHITECTURES
James O. “Jim” Farmer
Wave7 Optics, Inc.

Abstract ONT

Several Passive Optical Network (PON)

standards have been proposed as new architec-
Splitter or taps
tures for delivering video, voice, and data to

...
(16-128X)
homes. PONs are being built in large numbers
in Asia, and in increasing numbers in the Amer- Headend
icas and Europe. Several cable operators are
starting to deploy PONs in selected greenfield
applications, typically in situations where re- Figure 1. Basic PON
quired by the developer.
cal Network Terminal (ONT). In many cases
This paper shows the most popular the ONT is located on the outside of the home at
forms of PONs in use today. We compare the the utility entrance. Alternate locations include
performance of the PONs, and talk about how inside the home and in a purpose-built niche in
and when one may want to consider PON archi- the outside wall.
tectures.
Frequently the splitting is done in a cen-
tral location as shown. In other cases the split-
ting may be replaced by a tapped architecture
WHAT IS A PON? more like that used in HFC architectures. The
number of homes served by one PON is limited
PONs, or passive optical networks, are by the loss budget. While PONs are built with
just that: fiber optics all the way to the home, more or fewer subscribers, 32 subscribers is
with only passive (non power-consuming) de- considered the “sweet spot” in PON sizing to-
vices in the field. With no powered devices in day. We show up to 128-way splitting, but the
the field, you save on power costs, and mainten- optics available today don’t support this high a
ance is much lower than with hybrid fiber-coax split ratio.
(HFC). Since the network is all glass (usually
called “all dielectric”), you eliminate problems Done correctly, the advantages of PONs
such as sheath current. Lightning issues are include much lower operational expenses, high-
generally limited to anything that comes into the er quality, elimination of leakage and the resul-
home over the power line and, through sub- tant measurement requirements, and incredible
scriber equipment, jumps to your equipment. bandwidth. Data bandwidth of at least 1 Gb/s in
each direction, shared over just 32 subscribers is
Figure 1 illustrates the basic PON. A the norm today. This bandwidth is delivered
single fiber optic strand extends from the head- over separate wavelengths from that used for
end to an optical splitter located near a group of broadcast video, so the entire 54-1,000 MHz RF
homes. Outputs of the splitter supply optical band is available for video.
signals to a group of homes. Signals are termi-
nated on each home in a device called an Opti-

2008 NCTA Technical Papers - page 163

 TYPES OF PONS them with the likely RFoG architecture. We say
“likely” architecture because work on the RFoG
We shall describe several types of PONs standard has just started this year, and while
in this paper, including BPON (Broadband Pas- there are some pre-standard systems entering the
sive Optical Network, approaching end-of-life), market, the standard system has not been de-
GPON (Gigabit Passive Optical Network), and fined. Thus, what is described herein is the au-
GE-PON (Gigabit Ethernet Passive Optical thor’s conjecture of what the system may be.
Network). We shall mention a variant used in
some places, called an active optical network. PHYSICAL LEVEL ARCHITECTURE
We’ll also describe an emerging adaptation of
an HFC network to extend fiber deeper. It is Figure 2 illustrates the physical layer ar-
called RFoG (Radio Frequency over Glass), and chitecture of PONs. Figure 2a illustrates the
is an option to consider when a developer re- BPON/GPON/GE-PON architecture, and Figure
quires fiber-to-the-home (FTTH). 2b illustrates a possible RFoG architecture. In
each case, the headend comprises what head-
GPON and GE-PON systems (and ends usually comprise in the way of video,
BPON) share a common physical layer architec- voice, and data equipment, except that in the
ture, with some differences in optical levels and standard PONs of Figure 2a, there is no CMTS
speeds, so we will cover them together while – this will be explained later. Downstream RF
discussing the physical layer. We’ll compare signals are supplied to a downstream optical

Figure 2. Physical Layer Architectures

2008 NCTA Technical Papers - page 164

transmitter, usually an externally-modulated lines (POTS – plain old telephone service) will
transmitter and always at 1550 nm because am- be supplied. Other options are shown below.
plification of the optical signal is needed. While
you can amplify other wavelengths, amplifica- RFoG
tion at 1550 nm is the most mature and econom-
ical process today. All of the standards use A possible RFoG system is shown in
1550 nm for downstream RF broadcast. Figure 2b. The headend is identical to that of an
HFC system, because RFoG is really an HFC
GPON/GE-PON node serving one subscriber. The downstream
is again a 1550 nm transmitter, because you will
Unlike HFC networks, the network inter- need to amplify the optical signal. The up-
face for data is not a CMTS – none is used – but stream receiver is similar to that used in up-
rather is an analogous device called an Optical stream paths today. The upstream may be ana-
Line Terminal, or OLT. It serves the same func- log or it may be digital; this has not been de-
tion as the CMTS in that it converts data (usual- cided in the standardization effort as of this
ly delivered as gigabit Ethernet) into the format writing.
needed for PON transmission. That conversion
includes conversion to the particular PON pro- An optical node in the field is shown as
tocol being used, and conversion to light. The optional. Of course, if used, the network is no
downstream signals are carried at 1490 nm and longer completely passive. If used, the optical
the upstream at 1310 nm. These are combined node will likely contain optical amplification in
with the 1550 nm broadcast signal in a wave the downstream direction, and combining (in the
division multiplexer, or WDM. The WDM op- optical and/or RF domains) in the upstream.
erates analogously to a diplex filter in the HFC Some proposals convert the upstream to digital.
world.
Again, for RFoG we show a 32-way
Typically, the OLT includes many PONs split, though in practice, some may elect to go
in one chassis, density being very important. with different split ratios. Optical budgets will
There are some cases in which you may need a lead to these answers, and as of this writing,
less-dense solution for outlying pockets of sub- optical budgets for RFoG have not been de-
scribers, and some manufacturers have accom- cided.
modated this. While we show only one PON,
typically many PONs feed into an area and all The RFoG upstream wavelength issue is
splitters may be located at a common point interesting. One naturally gravitates to 1310 nm
called a local convergence cabinet. We can as an upstream wavelength, based on wide-
show that this architecture, particularly in green- spread availability of low-cost lasers and the
fields, results in a very economical deployment zero-dispersion wavelength of standard cable.
of equipment. Since this is the nominal zero-dispersion wave-
length of the fiber, it may be possible to use Fa-
After splitting, individual fibers supply bry-Perot lasers, at least for shorter distances.
optical signals to the ONTs at individual homes. On the other hand, there are applications in
An ONT may have one RF output that looks just which you may want to have some GPON or
like the downstream signals from an HFC net- GE-PON and some RFoG ONTs on the same
work, and it may have one or more data connec- network. For instance, you may want to serve
tions, usually 10/100Base-T and sometimes some businesses with GPON and some nearby
1000Base-T. Also, several analog telephone residences with RFoG. Or, you may someday

2008 NCTA Technical Papers - page 165

want to upgrade from RFoG to GPON or GE- THE ONT
PON. Since GPON and GE-PON use 1310 nm
for upstream data transmission, you cannot put Figure 3 illustrates the Optical Network
RFoG with a 1310 nm upstream on the same Terminal (ONT) at the home. In Figure 3a we
PON. illustrate a fully-featured GE-PON or GPON
ONT, and in Figure 3b we illustrate a possible
These considerations would lead to a dif- RFoG ONT. In Figure 3a we show the optical
ferent wavelength choice for RFoG upstream input to the ONT coming from a 32-way split-
signaling. 1590 nm is a candidate, but the next ter, common practice today. In Figure 3b we
generation of GE-PON (and perhaps GPON) has are showing a tapped architecture. While
already staked out this wavelength for faster up- people deploying FTTH today tend to favor the
stream. Any other wavelength that can be splitter architecture, some in the cable TV
passed through the fiber with low attenuation community are leaning toward a tapped archi-
could be used, so the RFoG working group may tecture.
choose some other wavelength. While the lasers
might be more expensive at first, presumably Experience has shown that centralizing
with volume and competition the cost will drop. splitters from a common point within the net-
Of course, it is more likely that DFB lasers will work and dedicating fiber to each home in a star
have to be used since we are well away from the configuration provides the most cost effective
zero dispersion wavelength of the fiber. deployment option. An additional benefit cen-
tralized splitters provides is the ability to scale
OLT ports and splitters in accordance with sub-

Figure 3. Optical Network Terminations

2008 NCTA Technical Papers - page 166

scriber penetration. In comparison a tapped to- gateway to provide the data over coax solution,
pology necessitates provisioning the network for while others use an internal bridge as illustrated.
100% of homes passed. A typical serving area This can be used for delivery of data to a com-
for centralized splitters is 250 homes. Either to- puter or home network, or it can be used for de-
pology will work. livery of IPTV (Internet Protocol Television). It
can be used for both.
GPON/GE-PON ONT
In greenfield applications, it is common
Figure 3a illustrates a fully-featured practice today to include cat5 data wiring, so for
GPON or GE-PON ONT. These are three- greenfield applications, it may not be necessary
wavelength systems. The broadcast down- to use data over coax at all.
stream, 54 – 1,002 MHz, comes in on a 1550
nm carrier. A WDM in the front end of the The ONT includes an RF receiver for the
ONT routes the wavelength to an RF receiver 1550 nm broadcast wavelength. As shown, it
not unlike those in HFC nodes, except that it has includes circuitry to convert the upstream RF
been optimized for low cost. Since there are transmission from set tops to digital for trans-
fewer sources of noise and distortion in FTTH mission back to the headend. Other systems
plant compared with HFC plant, more contribu- may use a separate analog transmitter for this
tion can be allocated to the ONT than to an HFC function, or it may not be available.
node.
RFoG ONT
This ONT was described in a paper by
this author presented at the 2007 NCTA Con- Compare Figure 3a, a fully-featured
vention,i so the detailed description will not be GPON or GE-PON ONT, with Figure 3b, a
repeated here. We shall review enough detail to stripped-down RFoG ONT. Again, we don’t
compare with the RFOG ONT of Figure 3b. know yet what standard RFoG ONTs will have
The Figure 3a ONT includes a data transceiver in them, so we start with the simplest possible
interfacing with a PON protocol chip. This is an solution and we’ll discuss possible upgrades.
ASIC (application-specific integrated circuit)
built by merchant silicon vendors for the appro- As with the GPON/GE-PON OLT, the
priate PON standard. It can be thought of as fiber is connected to a WDM, which separates
roughly analogous to a DOCSIS modem. the downstream RF on a 1550 nm carrier, from
Processing on the output of the PON protocol the upstream RF (not data) signal on whatever
chip converts the data into voice lines and data wavelength is chosen. The downstream receiver
lines, as well as providing control to the ONT. could be identical to that in Figure 3a.

Typically, two or more voice lines are A diplexer separates the downstream
provided, with the internal processing support- from the upstream RF signals. Inside the home,
ing any of the common VoIP protocols in use RF wiring is exactly as it is for HFC, including
today. Data is usually presented on the use of a cable modem and, for voice, an
10/100Base-T ports, or sometimes on a MTA, either embedded in the cable modem or
1000Base-T port. Many manufactures have a separate as shown here.
way to put data on coax in order to reduce the
amount of wiring that must be done at a home. The RFoG upstream transmitter presents
Two technologies dominate today: HPNA and an interesting situation. Analogous to the way
MoCA. Some manufacturers use an external upstream RF signals are combined, the upstream

2008 NCTA Technical Papers - page 167

optical signals from many transmitters will be top gateway (DSG), an internal modem, for
combined before being detected in a common their upstream. This would work, but restricts
receiver. If we allowed the upstream transmit- you on the set tops you can use. Due to cost, it
ters to be on all the time, we would have unac- is not likely that low-end set tops will use DSG.
ceptable interference at the upstream receiver.
Thus, each transmitter must be turned on only Of course, the RFoG upstream optical
when something in the house, be it a set top or a transmitters will need to work with DOCSIS
cable modem, is transmitting. The RF detector 3.0, which can have multiple upstream data
of Figure 3b detects RF signals coming from the channels in use at the same time. This adds to
house and turns on the upstream transmitter, the performance required of the upstream opti-
turning it off when the RF transmission ceases. cal transmitter. DOCSIS 3.0 is likely to work
better with RFoG than with HFC because there
A concern is based on the fact that there are fewer sources of distortion, and the RF de-
could be two or more independent systems us- tector in the ONT will prevent noise funneling.
ing the upstream path. The most common situa-
tion being a set top upstream transmitter and a Since RFoG utilizes optical combining
DOCSIS upstream transmitter. There is no way in the upstream direction the architecture will
to coordinate when the two disparate systems only support one upstream DOCSIS domain per
come on, so it is possible to have a set top in serving group. The upstream bandwidth capaci-
one home transmitting at the same time that a ty is now limited by the capacity of a single
DOCSIS modem in another home is transmit- DOCSIS domain rather than being frequency
ting. If the two optical transmitters are close limited.
enough in wavelength, it is possible that they It is logical that the RFoG specification,
will interfere, resulting in neither transmission when complete, will have a specification for the
getting through. Retransmitting routines may RF level threshold at which the transmitter is
mitigate this to an extent, but if a voice packet is turned on. This threshold would logically be set
affected, there will be a noticeable customer as high as possible in order to improve immuni-
event. ty against noise generated in the house. It is de-
sirable to force the highest possible upstream
Some people assume that the probability levels, because this puts operation as far above
of the above situation is sufficiently small that the noise level as possible.
the industry can live with it if the upstream wa-
velength utilized is 1310 nm and FP lasers are Possible Enhancements to the RFoG ONT
utilized. Others are not so sure. The assump-
tion is that FP lasers utilize a wide wavelength We have shown a basic RFoG ONT in
spectrum with a variance between devices, and Figure 3b. Some have suggested putting a
with 32 devices being combined statistically this DOCSIS modem in the ONT. This is possible,
would be ok. The center wavelengths of these but deviates from current cable TV practices. If
devices tend to drift with temperature so deter- the market likes the idea of outside ONTs, as are
mining the statistical frequency in which two or commonly used with GPON and GE-PON now,
more wavelengths will overlap is rather un- this would require a wider operating tempera-
scientific. As set tops are used for more appli- ture range of the modem, again driving up cost.
cations, it is likely that the percentage of time
they transmit will go up, and we know that An advantage of having some sort of
DOCSIS modems are transmitting a lot. A solu- communications in the ONT is that it would al-
tion would be to use set tops using DOCSIS set low management of the ONT, something that is

2008 NCTA Technical Papers - page 168

not possible with the simple configuration
shown in Figure 3b. A DOCSIS modem in the In an active Ethernet system, a switch is
ONT would allow two-way communication, placed in the field close to a cluster of subscrib-
permitting the ONT to report on it’s health and ers. An individual fiber is run from the switch
environment, something that is standard with to each home, as shown in Figure 4. The IEEE
GPON and GE-PON. Lacking two-way com- Ethernet standard has a section that standardizes
munications, a one-way communications path this configuration. The speed on the fiber to the
would permit remote disconnect, a standard home can be either 100 Mb/s or 1 Gb/s. How-
function of GPON and GE-PON ONTs. Of ever, there is typically no speed advantage with
course, there would be no confirmation, but that active Ethernet, because the common fiber to
may not be seen as too great a price to pay for the left of the remote Ethernet switch has li-
reducing the cost of the ONT. mited bandwidth, depending on what the opera-
tor wants to provide.
ACTIVE ETHERNET
Active Ethernet systems are difficult to
Before we change the subject, we’ll provision with RF video, because the video
mention one non-PON FTTH architecture that is would have to be WDM’ed into each individual
popular in certain places. This is variously subscriber’s fiber. A few such systems have
called Active Ethernet or Point-to-Point (P2P) been built with a second fiber system for video,
FTTH. but for the most part, active Ethernet systems

Optical Network
Termination
(ONT)

Remote
Usually one fiber per
Ethernet
direction
switch
...

Ethernet
switch

Headend or
Hub

Figure 4. Active Ethernet FTTH System

2008 NCTA Technical Papers - page 169

carry only IPTV or no video at all. EPON (Ethernet Passive Optical Network),
802.3ah (after the IEEE designation of the
ORGANIZING THE OPTIONS working group that developed it), or EFM
(Ethernet in the first mile – someone wanted to
We’ve talked a lot about the physical ar- emphasize that this applied close to the sub-
chitectures of PONs. Now we need to try to scriber, so it was considered to be the first, ra-
make some sense of the various types of PONs, ther than the last, mile). The active Ethernet
organizing them so we can understand what architecture of Figure 4 is also a part of this
each does and where they fit with each other. standard, as is a version operating on twisted
Figure 5 diagrams the options under discussion. pair, at much lower data rates.
Starting on the right, we have the ongoing de-
velopment of RFoG. This standardization effort The specification was approved in 2004,
is ongoing within the SCTE, in the fiber optics and volume quantities of ASICs became availa-
working group of the Interface Practices Sub- ble about 2006. GE-PON is very popular in
committee. It will be an option for cable opera- Asia, which is currently leading the world in
tors to consider when required to install FTTH. FTTH deployment, so most of the PONs in the
world are GE-PON. It is also being used in
In the center of the figure is the IEEE ef- North America and in Europe.
fort, which has been incorporated into the
Ethernet specification, managed by the IEEE Currently GE-PON operates at 1 Gb/s in
802.3 committee. The standard is referred to in both directions. The wire speed, or speed on the
this paper as GE-PON, but it is also known as fiber, is actually 1.25 Gb/s, but 8b/10b codingii

ITU G.98x series AT M-based Ethernet-based IEEE 802.3ah SCTE

APON
GE-PON (EPON, EFM )
1550 nm data down - ATM -only - Ethernet only
1310 nm data up - no broadcast
- effectively - Available 2006
abandoned - Security is not integral to standard, Point-to-point
Move downstream but AES is commonly practiced Ethernet on
data to 1490 nm - Aux  (broadcast) fiber or
to open 1550 nm twisted pair
for broadcast BPON 1 Gb/s
- Still ATM symmetrical
Add Ethernet - Aux  (broadcast)
transport,
increase speed, RFoG
GPON “Lite”
other GPON (RF over Gass)
- (Unofficial subset of G-PON -
improvements - ATM , Ethernet, Proposed
still called G-PON)
TDM - Ethernet
- Aux  (broadcast) - Available 2007
- Complicated - Aux (broadcast)
-No fully - Less complicated than G-PON
operational
systems 2.488 Gb/s down
1.2 Gb/s up
Remove
ATM
Figure 5. Comparison of PON Types

2008 NCTA Technical Papers - page 170

is used in order to ensure frequent transitions for compass all three layer 2 technologies. The
clock recovery and other purposes, so the net problem was that implementing the complete
speed is 1 Gb/s. The IEEE is currently working standard was exceedingly complex. By the time
on a new version of the standard that will oper- people started considering implementing G.984,
ate at 10 Gb/s downstream and either 1 Gb/s or it had become clear that Ethernet was the choice
10 Gb/s upstream. technology for the last mile (or first mile if you
use IEEE-speak).
To the left in Figure 5 are the ITU stan-
dards. The first ITU standard, ca. 1995, was Thus, the real implementation of GPON
called APON for ATM Passive Optical Network. is based on the Ethernet portion of the standard,
It used 1550 nm for downstream data and 1310 with the ATM portion not implemented. The
nm for upstream. It was replaced shortly by author has called this “GPON Lite,” but this is
BPON (Broadband PON), which moved the not an official designation – it is still known as
downstream data to 1490 nm to make room for GPON. The currently-favored version of
a broadcast overlay at 1550 nm. This is the ver- GPON has a downstream wire speed of 2.488
sion of PON that Verizon is currently deploying, Gb/s and an upstream speed of 1.2 Gb/s. It is
though they have announced an eventual switch specified to work with splits to 128 ways, but
to the next standard in the ITU series, GPON current optics don’t support this many splits
(Gigabit PON). over any meaningful distance. The ITU’s an-
nounced plan for future enhancement has been
GPON, ITU’s G.984 series, was ap- to use wave division multiplexing, where either
proved in parts, in 2003 and 2004. It started as each subscriber or a group of subscribers gets a
a combined standard that would encompass different wavelength. However, this tends to be
ATM, Ethernet, and TDM (time division multip- expensive, and there is some talk in the industry
lex, in this context referring to DS-1 or E-1 about revisiting the strategy.
transmissions). The standard is written to en-

Table 1. Comparison of PON Capabilities

Standard: RFoG GE-PON GPON

Year standard available: Not yet 2004 2004

Year of product general availa- Not yet (pre-standard now) 2006 2008
bility
Field actives? Optional Exceptional cases
Downstream wavelength 1550 nm 1550 nm (broadcast, optional), 1490 nm
data
Upstream wavelength Probably 1310 nm and one 1310 nm (possibly going to 1590 nm in
longer wavelength next generation)
RF Bandwidth 54 – 1,002 MHz, depending on manufacturer
Downstream data DOCSIS 1 Gb/s (after removing 2.488 Gb/s
8b/10b)
Upstream bandwidth DOCSIS 1 Gb/s (after removing 1.2 Gb/s
8b/10b)
Headend data interface CMTS OLT
IPTV ready? DOCSIS Yes
Service disconnect? Not decided Yes (depends on manufacturer)
ONT management? Not decided Yes
Upstream interference potential? Maybe No

2008 NCTA Technical Papers - page 171

 COMPARING THE PONS
Data is where we see the major differen-
Table 1 list comparative features of the tiation between RFoG and the other standards.
PON technologies being discussed. We’ve RFoG data uses DOCSIS for transport and is
listed the year that product started to be general- limited to DOCSIS speeds. At four channel
ly available to the marketplace, though there DOCSIS 3.0 bonding, you have the potential for
could have been limited deployments earlier. roughly 160 Mb/s of downstream data spread
Usually, GE-PON and GPON are built strictly over, using common practice, 32 subscribers.
as passive networks, with all active equipment This is an average data rate per subscriber of 5
being restricted to the headend or hub. Howev- Mb/s per subscriber, assuming one DOCSIS
er, some manufacturers have made provisions channel per node. Absent IPTV, this is a lot of
for a smaller field-mounted OLT for several data, because of the statistics of data sharing, a
scenarios in which this configuration is opti- subject in which the cable TV industry has de-
mum. veloped a lot of expertise. Yet it pales when
comparing with the other two standards, which
Everyone carries downstream broadcast offer, respectively, average data rates per sub-
on 1550 nm in order to provide for economical scriber of 31.25 Mb/s and 77.5 Mb/s.
optical amplification, and because good optical
transmitters are available for that wavelength. DOCSIS 3.0 upstream bonding should
This is the only downstream wavelength in work better in RFoG than in HFC because of the
RFoG, but the other two standards carry all data lack of noise funneling, but the difference in
(including voice) on a 1490 nm optical carrier. upstream bandwidth is more dramatic than in
Thus, they don’t loose any of the downstream the downstream direction. Developers demand-
RF band for data – you have up to 158 RF ing FTTH often employ telecommunications
channels exclusively for analog and digital vid- consultants who are familiar with GE-PON and
eo. If you used them all with 256 QAM, you GPON, and how they will react to a solution
would have on the order of 6 Gb/s broadcast to offering less bandwidth is not known yet. We
all homes. are certainly talking about a lot of bandwidth
with any of these PON solutions. Yet the histo-
The upstream wavelength for GE-PON ry of data communications is that there has nev-
and GPON is currently 1310 nm for economy. er been enough data bandwidth for long. With
There is talk in the industry of using 1590 nm all the over-the-top video and peer-to-peer traf-
for the next generation of GE-PON (and maybe fic today, it is not clear how long the old band-
for GPON, though this is conjecture). RFoG width sharing statistical models will hold true.
may provide an option of 1310 nm and some-
thing else, but this is not decided yet. The trick IPTV is certainly on everyone’s mind
is to allow interoperability between RFoG and today. Both GE-PON and GPON come ready to
the other standards, while keeping cost low. implement IPTV, and a fair number of users are
Interoperability will allow you to deploy RFoG doing so, some in North America, more over-
now, and migrate to something else later if you seas. While there are IPTV solutions designed
wish. Alternatively, you might deploy RFoG to for DOCSIS on the market, the case for putting
residences, but need to serve a few businesses IPTV over DOCSIS is not as clear as it is with
from the same PON, using either GE-PON or other PON technologies – with DOCSIS/RFoG
GPON. Obviously you cannot do this if you are you still have the broadcast infrastructure, and
using 1310 nm for the RFoG upstream and the switched digital video seems to have the poten-
other standard is using it for digital upstream. tial for doing the same thing as IPTV, using

2008 NCTA Technical Papers - page 172

more mature set top technology and likely mi- could also operate in mixed mode for an indefi-
nimizing overhead. nite time. You will have DOCSIS on the down-
stream that is not used in the GE-PON or GPON
More and more subscribers are stream- area, but having the signal there will not hurt
ing IPTV from internet web sites as the amount except for the four RF channels you loose for
of content available from major networks con- video (DOCSIS 3.0, four channel bonding).
tinues to increase. In essence your subscribers
have already launched you into over-the-top CONCLUSION
IPTV distribution.
FTTH systems are ready to be deployed
SO WHEN DO YOU CONSIDER FIBER? now and may make sense for greenfield dep-
loyment. The widely-recognized standards are
We have not addressed the question of GE-PON and GPON, which are similar in capa-
when a cable operator should build a PON. As bility from a user perspective, except for speed.
we look at the competitive landscape, HFC is in If you are not ready to make that leap, you can
much better shape than is DSL, so the urgency derive some of the benefits of FTTH by deploy-
is not what it is for someone with twisted pair ing RFoG, though the standard is not complete
plant. yet.

While cable is in better shape than it’s Author contact: [email protected]

competition, bandwidth demands always go up.
Your competition is starting to build FTTH. A References:
wise decision today is to build greenfield areas i
with your choice of fiber technologies, while Jim Farmer, Making FTTH Compatible with HFC, 2007
NCTA Technical Papers.
continuing to operate HFC plant where it exists. ii
Walter Ciciora et. al., Modern Cable Television Tech-
Some developers are demanding FTTH because nology : Video, Voice, and Data Communications, 2nd ed.,
they have learned that it improves the salability San Francisco : Morgan Kaufman, 2004, Chapter 19
of homes.

Conversion of HFC to fiber may make

sense when contemplating upgrading old plant
to higher bandwidths. This is particularly true
when contemplating use of bandwidth above 1
GHz, where massive plant modifications are
frequently required. But this conversion can be
done only on an as-needed basis, in areas of
high demand (and presumably high revenue).

If you start with RFoG and later convert

to either GE-PON or GPON, you would need to
convert an entire PON (normally 32 or fewer
subscribers) at one time. Alternately, if you
elected to use a non-interfering upstream wave-
length in RFoG, with suitable headend modifi-
cation and taking loss budgets into account, you
could convert one customer at a time. You

2008 NCTA Technical Papers - page 173

 ADVANCES IN DWDM ROADM TECHNOLOGY
USING PHOTONIC INTEGRATED CIRCUITS
Gaylord Hart
Infinera

Abstract To address these shortcomings, add/drop

multiplexers were developed which essentially
Modern photonic integrated circuits (PICs) combined the two back-to-back platforms into
integrate multiple optical subsystems on a single the same chassis. In these systems, an OEO
chip, which greatly reduces the traditional cost conversion is performed by the multiplexer at
structure for DWDM ROADMs (reconfigurable each node so services can be digitally added,
optical add/drop multiplexers). This allows dropped, groomed, or switched. Services merely
ROADMs to be cost-effectively architected for transiting the node still undergo the OEO
the first time using an Optical-Electrical-Optical conversion, but are digitally directed to the next
(OEO) conversion for every wavelength at every node. This type of system is typified by SONET
node, thereby allowing the use of digital multiplexers and offers the benefit that any
electronic switches for reconfigurability instead service can be digitally groomed or reconfigured
of all-optical, wavelength-only switches. The at any time for add/drop or pass-through.
resulting digital ROADM enables new
capabilities and yields significant advantages As the demand for increased transport
over analog all-optical ROADMs. capacity has grown, these systems have not
scaled well because they require a fiber pair for
each pair of connected multiplexers, resulting in
INTRODUCTION fiber exhaust in many cases. To address this
shortcoming, WDM systems were developed
Optical transport systems for delivering which allow multiple wavelengths to be carried
digital services have evolved significantly over on a single fiber pair, connecting multiple
the last several years. At each stage of this multiplexers at either end.
evolution, advances have been driven by
economics, as well as the need for greater While these WDM systems relieved the fiber
capacity and scalability. Inflection points in this exhaust problem, they do not economically scale
evolution have typically occurred when well because they utilize an expensive OEO
technological breakthroughs have enabled a conversion based upon discrete optical
paradigm shift that allowed significant cost components for every wavelength at every node,
reductions or new, advanced capabilities, or even for wavelengths which are not adding or
both. dropping services at the node. To squeeze
additional costs out of these platforms, fixed
The first optical transport systems deployed optical add/drop multiplexers (FOADMs) were
were point-to-point, single-wavelength systems. developed so that only those wavelengths adding
To create larger networks, these platforms were or dropping traffic at a node undergo an OEO
often placed back-to-back at a node and conversion. These systems utilize transponders
electrically interconnected for traffic to transit to add or drop a specific service on a specific
the node. This is an inefficient and expensive wavelength at a specific node. All other
method for building optical networks. wavelengths are passed through the node in the
optical domain.

2008 NCTA Technical Papers - page 174

 FOADMs relieve fiber exhaust and lower transponders with tunable lasers to provide
CapEx costs by reducing the number of OEO enhanced re-configurability.
conversions in a network, but they also created
new challenges. Since some wavelengths now Optical ROADMs are a great improvement
pass through nodes in the optical domain, optical over FOADMs. They accelerate and simplify
engineering for the network is no longer a single- network turn-up and service changes by
span engineering problem, but a multi-node, automatically balancing optical power
network-wide challenge. Additionally, analog throughout the network. They also allow
optical impairments now accumulate for those wavelengths to be remotely switched for optical
wavelengths that must transit through multiple add/drop or pass-through operation at each node.
nodes before undergoing an OEO conversion. However, they still have many of the limitations
This can limit the size of networks and may of FOADMs, including complex optical layer
require periodic re-gen of these wavelengths to engineering and a dependence on transponders,
remove the impairments. which tie services to wavelengths and hence
make service layer engineering dependent upon
FOADMs also present operational challenges. optical layer engineering.
When a typical FOADM network is initially
turned-up, it requires manual power balancing of One way to address these optical ROADM
every wavelength at each node in the network to limitations is to perform an OEO conversion for
ensure optimal operation. Moreover, these every wavelength at every node, then process all
networks typically require rebalancing whenever services digitally at each node. This would
services are added or deleted. If the initial reduce the optical network engineering problem
network design was not carried out guaranteeing to a series of simple single-span designs, even
any-to-any connectivity, but only based upon for mesh networks. Using 3R (regenerate,
initial services, it is possible the entire network reshape, retime) OEO conversions at each node,
may require a redesign and reconfiguration when optical impairments would not accumulate, and
services are changed. This is inconvenient and this would allow networks of essentially any
disrupts existing services on the network. arbitrary size to be built. Finally, digital
processing at each node would allow replacing
In all but the simplest of networks, FOADMs the all-optical wavelength selective switch of the
are time-consuming and complex to engineer and traditional ROADM with a digital switch which
operate. To address some of these limitations, can support advanced capabilities not possible
reconfigurable optical add/drop multiplexers with an optical layer switch. Such a digital
(ROADMs) have been developed. Optical ROADM would allow for the first time the
ROADMs still use transponders for the OEO creation of a flexible digital optical network
conversion for adding and dropping services at a where the service layer is independent of the
node, and they still pass through a wavelength in optical layer
the optical domain when no services are being
added or dropped locally from that wavelength. In the past, digital ROADMs were not
But a ROADM offers three capabilities a economically feasible due to the many discrete
FOADM does not typically provide: auto power optical components required to perform an OEO
balancing, optical wavelength switching, and a conversion for every wavelength at every node.
communications control plane to automatically However, recent advances in PIC technology
and remotely reconfigure the network when allow complete optical subsystems to be
necessary. Some ROADMs also support economically placed on a pair of chips (TX and
RX) less than 5 mm square and supporting 100

2008 NCTA Technical Papers - page 175

Gb/s per chip in a 10 x 10G DWDM ROADM ARCHITECTURE COMPARISONS
configuration. These PICs integrate over 60
discrete optical components (lasers, detectors, SONET has lost favor with most MSOs as
mux/demuxes, etc.) on a single pair of chips, they have migrated to Ethernet, but SONET is
eliminating all the discrete optical packaging and still widely deployed in the Telco world and in
the fiber jumpers formerly required to some CATV commercial services environments
interconnect the devices (see Figure 1, below). where some customers require SONET transport.
The resulting cost-reduction enables digital FOADMs are also widely deployed, but are
ROADMs to be built cost-effectively for the first finding more favor in applications where the
time, and commercial digital ROADMs built on networks are smaller and less complex. For
PIC technology have now been available for metro core, regional, and national DWDM
three years. PIC technology yields other networks, most cable operators are now
benefits, as well: reduced power consumption, deploying ROADMs to lower total cost of
network ownership and to accelerate new service
and bandwidth turn-up. In these networks,
digital ROADMs now compete with all-optical
ROADMs for market share and technical
leadership.

Optical and digital ROADMs typically share

many common attributes, including the ability to
provide dispersion compensation, amplification,
and automated optical power balancing at each
node. They both typically implement an
intelligent control plane (preferably using
GMPLS) that allows remote and/or automated
Figure 1 – Large Scale Photonic Integration Places configuration as well as other features. Many
Several Optical Components on a Single IC support automated topology and inventory
discovery and optical layer turn-up. Where
smaller footprint, reduced heat generation, fewer optical and digital ROADMs primarily differ is
modules, fewer fiber jumpers, and higher in the way they provide core re-configurability:
reliability. whether they switch in the optical or digital
domain. To understand these differences, it is
necessary to examine the way optical and digital
ROADMs are architected. These differences
Just as large-scale integration in digital have implications in the cost, engineering, and
electronics has enabled radical cost reductions operation of a network, and a digital ROADM,
even while increasing processing power, PICs as we shall see, provides significant advantages
now enable the digital paradigm shift to digital in each of these areas.
optical networks. Even more exciting, Moore’s Optical ROADM Architecture
law indicates significant gains are possible in
future generations of PIC technology. Fully Optical ROADMs (just as FOADMs do) only
functional PICs have now been built in the lab perform an OEO conversion at a node for
which support 10 x 40G configurations, and wavelengths being added or dropped at that
these 400 Gb/s PICs are scheduled to be node. Wavelengths not being added or dropped
available in 2009.

2008 NCTA Technical Papers - page 176

at the node are “expressed” through the node in directly through the ROADM and out the
the optical domain. reciprocal east-facing TX optical mux, or it
drops it to a west transponder for local service
Unique to the optical ROADM is a handoff. The TX signals originating from these
wavelength selective switch (WSS), or west transponders are usually passively added to
equivalent functionality implemented with the pass-through channels travelling from east to
wavelength blockers or other technology, which west and sent out the west-facing TX fiber.
allows individual wavelengths received at the
node to be switched for local add/drop or for The ROADM also provides reciprocal
“express” transit through the node. Figure 2, functionality for the TX and RX optical signals
below, shows the high-level architecture of a originating from the network’s east-facing fiber
typical 2-degree optical ROADM. interfaces. For redundancy purposes, this is
implemented in the ROADM with separate east
Looking at signal flow from west to east, the and west optical modules with each containing
RX optical signals originating from the an optical switch and any other required optical
network’s west-facing fiber interface are muxing, combining, or splitting components.
presented to an optical demux which breaks out Some ROADMs also provide optical
the individual wavelengths for processing by a performance monitoring points here for
software configurable optical switch. The switch “express” wavelengths, as well.
then individually passes each RX wavelength

Optical ROADM Architecture

All-Optical

1 1
.
WEST .
. WSS .
Optical Mux / Demux
Optical Mux / Demux

. .
RX n (Optical Switch) n TX

TX RX
1 EAST 1
. .
. WSS .
. .
n (Optical Switch) n
WEST EAST

West East
OEO

OEO

OEO

OEO

Add / Drop 1 …  n 1 …  n Add / Drop

Transponders Transponders

Local Client Service Interfaces

Figure 2 – Typical 2-Degree Optical ROADM Architecture

2008 NCTA Technical Papers - page 177

 It should be noted that some optical are typically only installed at a node when
ROADMs are banded, requiring add/drop or services are actually required there.
express treatment for a group of wavelengths as
a whole (banding is usually implemented in Digital ROADM Architecture
wavelength groups that are an integer multiple of
four). Because banded ROADMs must treat all Digital ROADMs perform an OEO
wavelengths in the banded group the same, this conversion for every wavelength arriving at
can result in stranded bandwidth. For example, every node, both east and west. Wavelengths not
if only two wavelengths need to be dropped off being added or dropped at the node are still
at a node, and the banding group size is four, the “expressed” through the node, but in the digital
ROADM will strand two wavelengths at the domain before the conversion back to optical
node. when exiting the ROADM.

For the wavelengths being dropped off Unique to the digital ROADM is a core
locally, the transponders perform an OEO digital cross-connect switch connected to every
conversion and provide client optical interfaces digital signal derived from every optical
for connection to local equipment. Between the interface, including the line interfaces and the
client and line-side optical interfaces, services client interfaces. This switch allows any input to
are processed digitally on the transponder, and it be connected to any output. Because this switch
is here that bit-error-rate (BER) evaluation, resides in the middle of the OEO conversion
forward error-correction (FEC), digital wrapper (unlike with an optical ROADM, where the OEO
processing (usually G.709), and digital takes place in the transponder after optical
performance monitoring takes place. The switching), the line-side optical layer is
transponder also typically provides performance completely segregated from the client side
monitoring for any service-specific attributes. service layer. Figure 3, below, shows the high-
Optical performance monitoring points for the level architecture of a typical 2-degree digital
add/drop signals are frequently provided here as ROADM constructed with PICs.
well.
At each line-side interface, both east and
Each transponder at the node links a west, an optical band multiplexing module
particular line wavelength to a particular client segregates groups of incoming wavelengths into
service interface (if tunable laser transponders Optical Carrier Groups (typically comprised of
are used, this provides a greater degree of 10 wavelengths at 10G each) for handoff to the
reconfigurability, but at a higher cost than fixed RX PICs. The same band multiplexing module
wavelength transponders). Because the same aggregates the reciprocal group of outgoing
wavelength may arrive from the east and west, wavelengths from the TX PICs for handoff to the
and for redundancy purposes, the ROADM line-side fiber. In turn, each PIC simultaneously
supports both east and west-facing transponders, processes multiple line-side wavelengths in
and these typically can only receive wavelengths parallel, performing O-E conversions on the RX
from the direction of the optical switch module side and E-O conversions on the TX side. Each
they are attached to. When protected services are PIC processes 10 WDM wavelengths at 10 Gb/s
required, two transponders must be used, one each. A typical initial ROADM deployment
each for east and west. Service growth is starts with a pair of TX/RX PICs on each of two
implemented by adding transponders at the digital line cards facing east and west,
service source and destination nodes, and these respectively. This provides an initial transport
capacity of 100Gb/s in both directions. Services

2008 NCTA Technical Papers - page 178

are then added until all 10 wavelengths on the performance monitoring. Once inside the core
initial PICs are consumed, and then an additional switch, these signals may be directed in any
pair of digital line cards is installed to support direction, to any interface (line or client), and to
further service growth. any wavelength (east or west, any color). In this
manner, express services are switched directly
The electrical interfaces on each PIC are through the ROADM, and add/drop services are
connected to the core digital cross-connect directed to local client interfaces.
switch, which is also connected to tributary
adapters for local service handoff on the client Services are added or dropped at the node
side. This switch is under software control and through a tributary adapter. The tributary adapter
may be remotely configured. It is implemented converts the digital transport stream from the
in a redundant configuration to eliminate any core switch for native service handoff and
single points of failure. provides this service on a client optical interface
for local use. The tributary adapter also provides
Internal to the ROADM, all services are performance monitoring for any service-specific
digitally processed. As the PIC’s electrical attributes and loopback capabilities for test
signals interface with the core digital switch, purposes.
they are processed for BER monitoring, FEC,
digital wrapper manipulation, and other

Digital ROADM Architecture

Optical Electrical Optical
Tx Tx
PIC PIC
10 x 10G 10 x 10G
Optical Band Mux / Demux

Optical Band Mux / Demux

Rx Rx
PIC PIC
RX Core Digital TX
OCG1 OCG1
. Cross-Connect .
. .
TX . Switch .
RX
. .
OCGn OCGn
(Any Port to Any Port)
Tx Tx
PIC PIC
10 x 10G 10 x 10G

WEST Rx Rx EAST
PIC PIC
Video
10GE

OC-n
GbE

Tributary Adapters

Local Client Service Interfaces

Figure 3 – Typical 2-Degree Digital ROADM Architecture Using Photonic Integrated Circuits

2008 NCTA Technical Papers - page 179

 OPTICAL LAYER ENGINEERING optical transport parameter (OSNR, dispersion,
power levels, etc.) has to be calculated east and
The optical layer engineering process and west for TX and RX for every wavelength for
requirements are very different for optical and every possible combination of end-nodes (see
digital ROADMs. Optical ROADMs only Figure 4, below). In a ring network, this
perform an OEO conversion for wavelengths requires 2(N2-N)WP calculations (where N is the
locally adding or dropping services at a node. number of nodes in a network, W is the number
All other wavelengths are passed through the of wavelengths supported by the network, and P
node in the optical domain. For these pass-
through wavelengths, optical impairments such
as dispersion and cascaded filter losses Optical ROADM – Optical Layer Design
accumulate from span to span. Since the source Typical Optical Parameters:
OSNR
and destination nodes may be anywhere in the Optical Power (max/min)
Dispersion
network, each wavelength must be individually Non-linearities

engineered for its particular path through the Total Optical Parameter Calculations:
network. But since different wavelengths often Total measurements: 2(N2-N)WP
• N = number of nodes
share common spans in the network, this is not a • W = number of wavelengths
• P = number of optical parameters
simple process. The final design must represent
a common denominator that works for all Figure 4 – Optical Layer Design for an Optical ROADM
wavelengths in the network, regardless of their
paths, and this may result in a less than optimal
design that may limit node counts or span is the number of transport parameters to be
distances. In some cases, it may be necessary to verified).
perform a 3R re-gen in the network, and this In a large ring network with a large number of
requires back-to-back transponders for every wavelengths, thousands of calculations may be
wavelength requiring re-gen. required to verify the optical layer design when
optical ROADMs are used. If the initial design
A digital ROADM performs an OEO assumptions prove unworkable, redesign and
conversion for every wavelength at every node. recalculation may be iteratively required to find a
In this case, all wavelengths on a span share the workable combination of all optical layer
identical path regardless of whether they are parameters. For a mesh network with multi-
being added or dropped at a node, and only degree nodes, the engineering problem becomes
individual spans need to be engineered between vastly more complex.
nodes. Moreover, because a 3R OEO conversion
takes place at each node, optical impairments do Digital ROADMs limit the optical layer
not accumulate from span to span in the network. design to a series of independent span designs
This essentially allows networks of any arbitrary where the design of one span does not impact the
size to be built. design of any other span (see Figure 5, below).
In a ring network, this requires a total of 2NWP
Ideally, an optical network’s initial design calculations (where N is the number of nodes in
should allow any service at any node to be a network, W is the number of wavelengths
transported to any other node in the network at supported by the network, and P is the number of
any time without requiring any re-engineering or transport parameters to be verified). To
re-configuration of the optical layer to do so. To guarantee any-to-any connectivity in the
guarantee this any-to-any connectivity in an network, one only has to guarantee that each
optical ROADM network, every relevant analog individual span between a pair of nodes has been

2008 NCTA Technical Papers - page 180

 Digital ROADM – Optical Layer Design intermediate nodes on the network unless back-
Typical Optical Parameters
to-back transponders are used where this
OSNR functionality is required. This is because the
Optical Power (max/min)
Dispersion sub-rate services can only be processed digitally,
Non-linearities
and an optical ROADM only has access to
Total Optical Parameter Calculations:
Total measurements: 2NWP
digital signals at the transponders, which are
• N = number of nodes only located at the source and destination points
• W = number of wavelengths
• P = number of optical parameters for a service.
Figure 5 – Optical Layer Design for a Digital ROADM
While some transponders provide some
integrated switching capability, this is typically
limited to within the transponder itself or to an
properly engineered. And this is true even for
adjacent back-to-back transponder for digital
large, complex mesh networks.
add/drop capabilities. It is not uncommon when
service grooming or switching is required at a
node to accomplish this with manually installed
OPTICAL vs. DIGITAL PROCESSING
fiber jumpers between the client interfaces on the
back-to-back transponders. In these cases, one is
Aside from the major difference in these
often faced with the economic tradeoff between
ROADMs between switching in the optical or
stranding bandwidth or paying for additional
digital domain, another key difference is where
transponders.
the switching is actually performed. An optical
ROADM switches above the OEO conversion,
In WDM optical networks, especially mesh
and the transponder which actually performs this
networks, it is quite common to have wavelength
conversion tightly couples the line-side optical
contention (or even blocking) between services
transport layer to the client side service layer. In
being carried over the network. This occurs
contrast, a digital ROADM separates the OEO
when two independent services using the same
conversion process into distinct line-side optical
wavelength need to travel over the same fiber on
layer and client-side service layer interfaces, and
at least one span. Of course, one may simply use
does its switching in the middle of these.
a different wavelength for one of the services
when this occurs, but this may in turn create
Having examined the respective architectures
contention with another service on another span.
for optical and digital ROADMs, we can now
To fully utilize available network capacity, it
examine the specific capabilities supported by
may be necessary to perform a wavelength
each.
conversion for a service for transport over the
contended path. Because the transponders on an
Switching
optical ROADM tightly couple a service to a
wavelength between the transponders, and
Optical ROADMs, of course, switch entire
because an optical ROADM cannot switch
wavelengths, which are typically transported
services between wavelengths, the only way to
today at a nominal 2.5 Gb/s or 10 Gb/s line rate.
perform this wavelength conversion is to use
If any of these wavelengths have lower rate
back-to-back transponders using different
services multiplexed up to the line rate by the
wavelengths on the contended path. This is an
transponder (muxponder), as is common to
expensive solution, but may be the only option
maximize the use of transport capacity, these
when blocking occurs.
sub-rate services cannot be individually
switched, groomed, or added/dropped at

2008 NCTA Technical Papers - page 181

 Digital ROADMs provide simple and At the switching level, the ROADM’s
inexpensive solutions to these switching integrated digital switch works at the 2.5G sub-
challenges, and provide additional capabilities as rate granularity, so sub-rate switching is fully
well. As already noted, these ROADMs have supported along with switching at the 10G line
digital access to all services and wavelengths at rate. This supports add/drop multiplexing of any
every node and also have an integral digital service at any node at any time. In fact, even the
switch interconnecting all interfaces at every tributary interfaces at a node may be switched to
node. This combination yields very powerful face east or west in the network. Unlike with
capabilities for switching services and transponder based optical ROADMs, digital
maximizing bandwidth usage in the network. An switching makes service transport completely
examination of these capabilities follows, but a independent of the transport wavelengths: full
more detailed discussion of how a digital grooming and switching of every service at every
ROADM processes services is necessary first. node (including sub-lambda services) is possible
to and from any wavelength, and the service bit
To maximize transport capacity, advanced rate is independent of the transport line bit rate.
digital ROADMs utilize PICs providing 10 x 10 Figure 6, below, shows several of the processing
Gb/s wavelengths for transport across the capabilities provided at every digital ROADM
network. A 10 Gb/s digital transport frame node.
(DTF) using a G.709 digital wrapper or an
enhanced version of it is used on each
wavelength to transport native client services Digital ROADM – Nodal Processing

end-to-end. The DTF also provides forward error Digital Electronics

correction (FEC) and performance monitoring, WEST Integrated Photonics EAST

Integrated Photonics
& Software

not only between intermediate transit nodes, but • 3R Signal Regeneration

• Error Correction
between end-to-end service points as well. The • Performance Monitoring
• Sub-l Add / Drop
DTF in turn has four 2.5 Gb/s digital signals • Sub-l Grooming / Switching
• Multi-Service Multiplexing
asynchronously multiplexed into it, thus the DTF • Digital Protection

may transport one 10G or four 2.5G services. Optical Electrical Optical

Figure 6 – Digital ROADMs Process All Services

The DTF and its 2.5G sub-rate signals are
Digitally at Each Node
transparent to the client signals and may carry
Ethernet, SONET or other protocols. A 10G
DTF can simultaneously support any
combination of client signals mapped into its Digital switching is particularly effective in
2.5G signals, up to the full 10G rate. A 2.5G creating multi-degree networks, especially more
signal may in turn have 2 GbEs mapped into it. complex mesh networks, where the complexities
Thus a 10G DTF may carry one OC-192, one of optical layer engineering and wavelength
10GE, four OC-48s, 8 GbEs, or some planning become great. With a digital ROADM,
combination of these that does not exceed the the optical layers of each path at a junction node
10G line rate. Other mapping possibilities exist, are independent, and traffic is simply switched
and other protocols can be supported as well. between them digitally. Services may then be
The DTF’s flexible mapping capabilities allow routed end-to-end through any available path in
each wavelength to be used efficiently by the network, and services may be created or torn
multiplexing any combination of sub-rate down on demand without any optical layer
services into it until the wavelength is fully engineering. Mesh networks provide enhanced
utilized. protection and bandwidth management options

2008 NCTA Technical Papers - page 182

by providing multiple paths between endpoints Because a digital ROADM provides 3R OEO
in the network. conversions for all wavelengths at every node
and because optical layer engineering is then
Digital ROADMs are also quite effective in limited simply to spans between these OEO
building hierarchical networks, where a single nodes, service layer engineering is now made
digital ROADM can serve, for example, as a fully independent of optical layer engineering.
junction site between a regional and metro core Once a span has been engineered for the first
network. In this configuration, a digital digital line module, additional modules may be
ROADM can provide north and south interfaces installed without the need for any additional
for one network, and east and west for the other. optical layer engineering or reconfiguration.
Traffic may then be switched digitally in any This greatly reduces the time and effort required
direction between or within the networks, and to turn up new bandwidth and services.
protection is supported for both networks.
Because a digital ROADM provides 3R OEO Bandwidth virtualization enables flexible
conversions at every node, there are inherently reconfiguration of the network at any node at any
no distance or node count limitations in a digital time without any optical layer re-engineering and
network. It is actually possible to build a provides significant additional capabilities that
national backbone with integrated regional cannot be provided with an all-optical ROADM.
networks and metro core rings using a scalable First, unlimited add/drop capabilities are
digital ROADM. supported for any service at any node at any
time, including sub-lambda services. Second,
Bandwidth Virtualization wavelength or sub-lambda services may be
groomed or switched from any wavelength to
Whereas transponders in an all-optical any other wavelength at any node. This enables
ROADM tightly couple services to wavelengths, wavelength conversion for wavelength services
a digital ROADM for the first time allows the and inter-wavelength grooming for sub-lambda
service layer to be fully segregated from the services, so a service may actually be transported
optical layer. This allows wavelengths (and on several different wavelengths in the network
2.5G sub-lambdas) to be treated as a pool of between its source and destination points. This
virtual bandwidth, that is, as an allocatable flexibility allows existing wavelengths in the
resource to be assigned to services when and network bandwidth pool to be fully utilized
only as needed. As long as sufficient bandwidth before having to add more capacity and provides
exists between any two nodes in a digital a simple solution to wavelength contention or
ROADM network, any service may be turned up blocking.
between these nodes at any time. If sufficient
bandwidth does not exist, digital line modules Bandwidth virtualization also supports super-
may simply be added in those spans where more lambda services. For example, a 40G service
bandwidth is needed to support the new service. can be delivered using tributary adapters that
Since PIC based line modules add bandwidth in concatenate 40G service transport over four 10G
groups of 100 Gb/s, any unallocated bandwidth wavelengths from the available bandwidth pool.
on the new line module is added to the pool for In this manner, 40G services may be delivered
future use. In this way, capacity usage is over a 10G network without any additional
optimized on individual wavelengths and optical layer engineering or reconfiguration.
between individual nodes. Since the client handoffs at either end of the
service are standard 40G interfaces, service
delivery is fully transparent.

2008 NCTA Technical Papers - page 183

 Digital processing enables other features as is used for this application. Digital multicast
well, many of which support advanced services may be digitally protected for increased
maintenance capabilities or unique architectures. reliability.
A digital ROADM supports hair-pinning
services, where a service is brought into the Performance Monitoring
ROADM on one client port and is “hair-pinned”
out a different client port at the same node. All ROADMs provide some degree of
Bridge-and-roll is also supported. Using bridge- performance monitoring, and the information
and-roll, an alternate path is created for a service derived from this is used for many purposes,
between the two endpoints. A bridge is then including guaranteeing service level agreements
created to duplicate the service over both paths. (SLAs) with end customers, establishing base
Finally, a roll operation is executed which network operating parameters so any observed
transfers the end-to-end service over to the new degradation can be used to solve problems pro-
path in under 50 ms. The original path is then actively before an outage occurs, and to diagnose
free for maintenance operations. Non-obtrusive and sectionalize an outage if it does occur. All
digital test ports may also be created at any node ROADMs provide performance monitoring
in the network to observe traffic on any service. points (PMs) for both optical and digital
In this case, a service is digitally replicated at parameters, but significant differences exist
any node in its path, and this copy is then sent to between the number of parameters monitored
another client port at any node in the network, and where they are monitored.
where it may then be attached to test equipment
or otherwise examined. Finally, digital Optical PMs typically consist of optical
processing enables a wide range of loopback power levels measured at various points in the
options to be used for circuit verification and ROADM, but may include other parameters as
fault-isolation purposes, including the ability to well, such as laser bias current. Power
perform loopbacks not only at the service measurements may be aggregate power
endpoints, but at intermediate nodes in the (combined power of all wavelengths at the PM
service path as well. All of these maintenance point) or individual power (power level of an
and test mechanisms are remotely provisionable. individual wavelength at the PM point). An
aggregate power level PM can indicate a
Using digital multicast, digital ROADMs also problem exists (level too high or low), but it
make unidirectional drop-and-continue digital can’t tell you which wavelengths are
video broadcast architectures simple and contributing to the problem and therefore is
inexpensive to implement at Layer 1. In this usually less helpful diagnosing and locating a
application, the ROADM’s digital switch particular problem.
replicates the broadcast digital video service at Optical ROADMs typically provide optical
any required node, drops it locally, and then PMs at various TX and RX points in the
passes it on to the next node. At each node, up ROADM, but how and where these are measured
to three output multicasts may be created, varies considerably. Optical PMs should be
supporting local drop and multicast branching at provided for both TX and RX signals on both the
junction nodes. Since the multicast is handled east and west line-side interfaces. Ideally, these
digitally, no optical splitting or specialized should provide aggregate and per-lambda data
transponders are required, and there is no limit to for express wavelengths as well as those
the number of drop or branching sites. Digital originating from local transponders. This
multicast can also be used to support switched provides the greatest amount of useful
digital video when an edge-switched architecture information for diagnosing and pinpointing any

2008 NCTA Technical Papers - page 184

problems. However, some ROADMs only transponders are only used at the service
measure optical power levels on the line-side endpoints, digital PMs are only available there,
interfaces in aggregate and only look at and not at any intermediate nodes (see Figure 7,
individual wavelength levels on the local below). This makes diagnosing and localizing a
transponders. problem more difficult since intermediate nodes
in the path cannot provide PMs which might
Digital ROADMs, because they perform an indicate which section is responsible for the
OEO conversion for every wavelength at every problem.
node, typically provide a full complement of
optical PMs, measuring TX and RX power levels
for every wavelength on both the east and west
line-side interfaces. Aggregate optical power is Optical ROADM – Digital PMs
also usually monitored on the line interfaces.
This provides rapid and robust fault diagnosis
and location capabilities at the optical layer.
All-optical end-to-end service path

Optical power level monitoring is quite

useful, but there are only a limited number of O-E-O
O-E-O
O-E-O
O-E-O
O-E-O
O-E-O
O-E-O
O-E-O
optical layer problems which can be diagnosed O-E-O
O-E-O
O-E-O
O-E-O
O-E-O

with power levels. For example, optical power

levels can help diagnose fiber cuts or laser G.709 PM and
OAMP only at
failures, but not dispersion problems. Moreover, service end-points
optical PMs are of no help in diagnosing digital
or service layer problems, and these are usually Figure 7 – Digital PMs on an Optical ROADM
assessed using digital PMs.

Digital PMs typically provide a large amount Digital ROADMs, on the other hand, perform
of information derived from a large number of an OEO operation at every node for every
monitoring points. Typical information would wavelength, so digital PMs are available at every
include loss of frame (LOF), loss of signal node for every signal that transits the node (see
(LOS), uncorrected BER, corrected BER, errored Figure 8, below). Digital ROADMs provide
seconds, severely errored seconds, and numerous PMs for both the transport section (transport
other parameters. This information is usually between each node) and the transport path (end-
gathered on the optical transport path for each to-end service transport). This allows rapid
service and/or wavelength being transported on pinpointing of any problems to an individual link
the network, depending on the particular between two nodes.
ROADM. BER data is usually gathered for the
optical transport path via a G.709 or enhanced To facilitate testing, some digital ROADMs
digital wrapper, which provides FEC, as well. also directly incorporate a pseudo-random bit
Additional service-specific PMs (e.g., Ethernet stream (PRBS) generator for direct BER testing
frame and errored frame counts) are also between any two nodes without the need for any
typically available. external test equipment. In PIC based digital
ROADMs, a PRBS stream is run continuously
Optical ROADMs, by their nature, only on those 10G wavelengths which are installed on
provide digital PMs where signals are processed operational digital line cards but which are not
digitally (i.e., only at the transponders). Since yet carrying services. This provides the MSO

2008 NCTA Technical Papers - page 185

 the protection switching, the clients on each of
Digital ROADM – Digital PMs the two transponders are connected directly to
reciprocal clients on the end device.

Digital access to full G.709/DTF O/H Digital ROADMs can also be configured for
dedicated protection. In this case, both working
and protection paths are pre-defined through the
network between the service endpoints. These
paths are set up through each node using the
ROADM’s integrated digital switch and any
available wavelengths through the network.
Full PM data and OAMP access at Route diversity is used for path protection. At
every node in the service path the endpoints, a pair of tributary adapters is used
to provide the protected clients. As with optical
Figure 8 – Digital PMs on a Digital ROADM ROADMs, the paths and tributary adapters are
dedicated and cannot be used for other services.
A Y-cable may be used to provide a single
with an operational history and track record even protected client interface, just as with optical
for those wavelengths not yet turned up with ROADMs, or the ROADM’s integrated switch
services and provides a higher degree of may be used to deliver the best signal from either
confidence in successful service turn-up when path to a single tributary adapter client. This
the time comes to activate these wavelengths. second method saves the cost of one tributary
adapter and the Y-cable. Dedicated 1+1
Protection protection is handled with two tributary adapters
just as with an optical ROADM.
Optical ROADMs typically protect a service
by using two route-diverse paths between the Digital ROADMs, however, support shared
endpoints, with a pair of transponders on each protection modes which are not supported by
path. One transponder pair serves for the most optical ROADMs. With shared protection,
working path, one for protection. This the protection wavelengths are left uncommitted
configuration provides effective equipment and throughout the network, and these remain in the
path protection, but the protection wavelength pool of allocatable bandwidth until actually
and its transponders are dedicated and cannot be needed. In this way, a small shared bandwidth
used for other services. A single client handoff pool can be used to protect many services across
at either end is usually provided by a Y-cable the network, resulting in much lower bandwidth
connected to the clients on the working and consumption when compared to dedicated
protect transponders. Some optical ROADMs protection.
support a digital backplane interface between the
transponder protection pairs, which allows a When a failure occurs in a shared protection
single client interface on one of the transponders network, the GMPLS control plane finds a path
to be used for the protected service without through the network with sufficient bandwidth to
requiring a Y-cable. In this case, the best signal restore service, allocates this bandwidth from
from either of the transponders is presented at end-to-end, then switches the service over to the
the single client. For 1+1 protection, where the new path. If sufficient paths and bandwidth are
local end device connected to the ROADM available, protection may be provided over
(typically a switch or router) actually performs multiple failures at multiple points in the

2008 NCTA Technical Papers - page 186

network. Since multiple failures in different ten 10 Gb/s wavelengths, on a pair of chips (TX
segments of the network are unlikely to occur and RX) no more than 5 mm square. This allows
simultaneously, shared protection provides an a low-cost OEO conversion to be used for every
effective means of conserving bandwidth while wavelength at every optical multiplexer in a
providing a high degree of confidence in network. The PIC cost-savings in turn support
protecting against network failures. Shared integrating a full digital cross-connect switch
protection may be used in ring or mesh into the core of the multiplexer, creating for the
networks, but mesh networks typically provide first time a cost-effective digital ROADM.
more paths through the network between service
points and therefore may provide more options The architecture of the digital ROADM (and
for protection routing than would be available in hence its name) is such that all services are
a ring network. processed digitally, rather than optically. Since a
digital ROADM performs an OEO operation on
Unlike dedicated protection, shared protection every wavelength at every node, its integrated
frees up significant bandwidth in the network for digital switch has unrestricted access to every
other uses, and the network operator has full service entering or leaving the multiplexer, and
control over how much spare bandwidth to therefore has unrestricted ability to groom,
provide for shared protection. However, there is switch, or add/drop services at the node. This
a tradeoff in using shared protection. A delivers a much wider range of reconfiguration
dedicated protection path, because it is always options than an all-optical ROADM can provide,
live, provides protection switching in under 50 but it also enables a completely new set of
ms. A shared protection path must be found and features and capabilities.
routed after a failure, and this can take a few
seconds. In a digital ROADM, protection may Digital ROADMs greatly simplify optical
be provisioned as dedicated or shared on a layer engineering, and their 3R OEO architecture
service by service basis, so this is typically not a supports networks of essentially any size or
problem since any sensitive services can always shape to be built and provisioned easily.
be configured with dedicated protection. Because digital ROADMs segregate the optical
layer from the service layer, turning up new
services is quick and requires no optical layer
SUMMARY AND CONCLUSIONS engineering whatsoever. Digital ROADMs
support wavelength conversion, sub-lambda
Photonic integrated circuits represent a major grooming, inter-wavelength switching, and are in
inflection point in optical networking evolution, general more bandwidth efficient than their
enabling the digital paradigm shift to digital optical counterparts.
optical networking and delivering the scalability
that will be required for next-generation Digital ROADMs bring significant benefits to
networks. all phases of network ownership, from network
engineering, to network turn-up, to service
Major advances in photonic integrated growth, and to network evolution. They offer a
circuits have resulted in commercial production lower total cost of network ownership while
of inexpensive, highly reliable photonic ICs accelerating network operations, bandwidth
integrating all the components required to deliver expansion, and service delivery.

2008 NCTA Technical Papers - page 187

 BUILDING LARGE VOD LIBRARIES WITH NEXT GENERATION ON DEMAND
ARCHITECTURE

Weidong Mao
Comcast Fellow
Office of the CTO
Comcast Cable

Abstract scalability, reliability, and expandability to

future on demand services.
The paper presents an integrated Video On
Demand (VOD) content library platform that
supports virtually an unlimited amount of media MANAGED NETWORK VS OVER
content such as movies, TV shows, Internet THE TOP
video, and user generated content. This
approach combines the advantages of the Increasingly cable operators are using Video
existing Managed Network approach and the On Demand (VOD) as a key competitive
emerging Over the Top approach in offering advantage. Alternative video delivery methods
VOD services. such as movie download or video streaming via
the Internet are also becoming more practical and
Specifically, this paper describes the overall feasible as service providers deploy either
requirements and architectural evolution of DOCSIS 3.0 wideband or Fiber to the Home
Video On Demand (VOD) infrastructures to technologies.
support large content libraries. The content
libraries will enable a large amount of VOD Figure 1 illustrates comparisons between the
content (SD and HD) as well as Internet Video “Managed Network” and the “Over the Top”
content. The solution is based on the Comcast approaches for providing on demand video to
Next Generation On Demand (NGOD) subscribers. In the existing Managed Network
architecture with a key extension of the Content approach that is adopted by Cable and Telco
Delivery Network (CDN) that utilizes national network operators, VOD content is usually
and regional IP network and library storage. encoded in MPEG-2 format and distributed along
with metadata via a Satellite or IP backbone to
Several key architectural building blocks and the local VOD systems. The content is usually
technology options are include content encoding “pushed” and replicated in every local VOD
and transcoding, real time and non real time system. The VOD client on the digital set-top
content ingest, asset metadata and rights box will be able to setup sessions and perform
management, content library and asset stream control functions such as Play, Pause, Fast
propagation management, VOD backoffice Forward and Rewind.
integration, streaming server, as well as shared
edge resources. In contrast, the emerging Over the Top
approach uses the broadband Internet as the
Finally, this paper also discusses content content distribution and streaming platform.
formats, open interfaces, performance, Content aggregators / integrators license and

2008 NCTA Technical Papers - page 198

publish movies and TV shows at the Internet • Leverage existing digital STB at home
website. PC or CE devices such as TV set-tops and digital cable ready TV through
are able to access the video content via the tru2way
Internet using the broadband pipe such as cable • Achieve high VOD usage with the large
modem, DSL, or Fiber to the Home network. installed subscriber base
Content for the Over the Top services is typically • No buffering at the client is required.
encoded using advanced codec such as H.264 This enables easy navigation and access
with lower resolutions. Content distributions of the content.
within the Internet are usually driven by the
“pull” requests coming from the subscriber. Over the Top features and capabilities:

Managed Network Over the Top

• Access to a vast amount of Internet video
User
Content Content
Providers
Generated and user generated video
Providers Content
• Feature rich navigation user interface and
Integrators search capability using open Web
Satellite or IP
Backbone technology
• Benefit from bandwidth competition at
VOD
Infrastructure Internet access network
• Benefit from advanced codec technology
Access Nework
using PC or new CE appliance
CE Device
• Potential to offer the same VOD service
to any device with Internet access

There are several limitations of the Over the

Figure 1 Comparisons of Managed Network and Over
the Top VOD Services Top approach:

• Challenge to achieve high concurrency

The following lists some of the unique for HD video streaming
features and capabilities of Managed Network • Utilizes public Internet infrastructure that
and Over the Top VOD services: imposes quality of service constraints
(e.g. congestion)
Managed Network features and capabilities: • Lack of end to end network resource
management
• Extend linear TV programming service • Inconsistent premium content offering
with free and premium VOD content due to lack of programming agreements
• Utilize managed IP networks for VOD with content providers
content distribution with better quality of • Requires subscriber to purchase a
service separate CE appliance for viewing VOD
• Manage bandwidth expansion at access on TV
network to satisfy high concurrency (> • Does not yet have a large subscriber base.
10%) and HD VOD Fragmented market with too many
• Build VOD server streaming players
infrastructure for streaming capacity • Long buffering time at the client may be
required

2008 NCTA Technical Papers - page 199

 Free/Premium Operator Internet
Content VOD Portals Video
Content
A HYBRID VOD LIBRARY Providers

APPROACH Operator IP
Networks Internet
National /
Regional

Today’s Managed Network VOD system CDN / Library

architectures helped network operators bring a

compelling product to market. There are Headend
VOD
Infrastructures

significant opportunities for the network HFC Network

operators to expand the current VOD architecture
in order to support large VOD content libraries
that provide an expansive amount of content Subscriber
Portable

including the Internet video. The other

opportunity is to provide the VOD offering to Figure 2 A Hybrid VOD Library Approach
devices other than STBs. Most of these are IP
enabled devices such as PCs and portable CE
In this approach, the existing Managed
devices.
Network VOD infrastructure is expanded with
the Content Delivery Network (CDN). The large
In order to keep and expand the competitive
content libraries within CDN are connected via
advantages in providing VOD services, network
operators’ IP backbone and regional networks to
operators are embracing the vision to give
both content providers and local headends.
customers the ability to watch any movie,
television show, user generated content or other
The CDN content library will ingest and store
video that a content provider wants to make
content coming from traditional free or premium
available through Video On Demand. The
VOD sources. The CDN content library will also
service would offer the following:
be able to ingest video content published from an
Internet portal and external Internet video content
• More HD content:
including user generated content.
• More library VOD content
• Time shifted TV (StartOverTM)
Typically, all the VOD content is stored in the
• Personalized, video rich navigation with
CDN content library. Popular VOD content can
better search
be replicated and propagated ahead of time to the
• Internet video content
local VOD systems via CDN. Upon a
• Cross platform video services (TV, PC,
subscriber’s request for VOD content, the VOD
portable devices)
system can start streaming if the content is
• Extensions for addressable advertising
already available at the local VOD system. The
VOD system will pull the content from CDN
In order to support these goals of any video
content library if it is not available at the local
content, at any time, to any device, a hybrid
VOD system. The content may be cached at the
approach using VOD content libraries based on
local VOD system for a period of time to serve
the Content Delivery Network (CDN) that will
other subscribers’ requests.
extend the existing Managed Network VOD
infrastructure is proposed as shown in Figure 2.
The same CDN content library will serve
multiple devices including STBs, PCs, and
portable devices. Content may be transcoded to

2008 NCTA Technical Papers - page 200

multiple formats upon ingest to the CDN content – Real time and non real time ingest
library. – Performance and scalability of
streaming from library storage
The proposed hybrid VOD library approach – Asset propagation management
has several significant advantages, compared • Streaming Capacity and Bandwidth
with the traditional Managed Network and the – Additional streaming server capacity
emerging Over the Top approaches, they include: is required
– Increased edge QAM and unicast
• Content: bandwidth is required
– Free or premium VOD content • VOD Navigation User Interface
– Offer vast amount of movie and • Internet Video Model
television shows, most in HD – Pull versus push
– Enable access to Internet video such – Metadata format
as user generated video – Transcoding
• Infrastructure:
– Utilize a managed IP backbone and Next Generation On Demand (NGOD)
regional networks for VOD content
distribution with better quality of Comcast has developed the Next Generation
service On Demand (NGOD) architecture framework to
– Manage bandwidth expansion at address both the feature expansion and capacity
access network for high concurrency expansion of the VOD infrastructure to support
of HD VOD multiple on demand services (see [1]).
– Expand existing VOD content
distribution, management, The Next Generation On Demand architecture
entitlement, and streaming platform will continue to be used as a foundation to
• Devices: support the VOD library expansion based on the
– Leverage existing digital STBs and following principles:
digital cable ready TVs through
tru2way  Open Interfaces: The reference architecture
– Support PCs and other portable is developed with logical functional
devices with Internet access components. Standardized open interfaces
between the different components in the
ARCHITECTURAL BUILDING BLOCKS architecture are also developed. This will
enable multiple vendors to innovate in the
Main Challenges areas of their expertise and allow seamless
integration among the various components.
With the advent of new technologies in IP For example, NGOD has specified key
networking and high performance storage and components and interfaces for the Session
streaming servers, it becomes feasible to evolve Manager, On Demand Resource Manager,
the current VOD architecture to support large Edge Resource Manager, and Edge QAM.
scale content libraries. However, there are
several challenges that need to be addressed:  Shared Resources for Multiple Services:
Today’s architectures are typically
• Content Library (CDN) customized for a limited set of services.

2008 NCTA Technical Papers - page 201

 Unfortunately, a significant re-engineering and its extension to next generation CDN are
effort is required to support the addition of illustrated in Figure 3.
new services. The NGOD architecture
enables the sharing of storage, streaming, The architecture is partitioned functionally
network, and edge resources among multiple into a number of logical components. Each
services. It is an extensible, on-demand component is defined in such a way that the
platform that allows multiple services to interchangeable module implementing the
share the same underlying infrastructure. It common interfaces can be introduced to work
will create significant cost efficiencies and cooperatively with the rest of the system. It is
make it possible to quickly provide new possible that implementations may integrate
services more quickly and easily. several components into a single product or
solution.
 High Performance, Scalability, and
Reliability: The NGOD architecture is Each logical entity described in the reference
designed to achieve high performance and architecture may represent one or many physical
scalability by allowing each component to be entities in an actual implementation.
scaled and optimized independently. In
addition, redundancy is built in various Content Ingest Content Delivery Network

Asset Library Service

VOD System Client

STB
Metadata &
sources and components to provide high
Client
Rights
Navigation &
Management
Entitlement

reliability. Transcoding Asset Propagation

PC & CE
Devices
Manager

Session &

Content Delivery Network (CDN) Trick File

Library
Resource
Manager
Storage

Edge

One of the key enabling technologies for large

Real Time
Ingest
QAM

VOD content libraries is the next generation

Streaming
Encryption /
IP Networks Server
DRM CMTS

Content Delivery Network (CDN). The next

generation CDN will support VOD content and
other media files. The national and regional IP Figure 3 Key Architectural Building Blocks
networks connect multiple VOD content libraries
in various locations such as national media The key architectural building blocks include:
center, regional centers, and local VOD systems.
The CDN will enable operators to provide a large Metadata & Rights Management – manage the
amount of VOD content cost effectively by asset metadata and rights for content from
serving them from the national and regional various content providers and aggregators, such
libraries instead of replicating all content to the as licensing windows.
local VOD systems. Intelligent caching can be
adapted at the CDN and the local VOD system Transcoding – transcode the content into various
based on the popularity and actual usage of the formats based on codec, resolution, and bitrate.
content to further reduce the network bandwidth
usage and enhance the overall performance. Trick File – generate fast forward and rewind
trick files from the original content.
The overall architectural building blocks for
VOD libraries based on the NGOD architecture Real Time Ingest – ingest the real time content
streams from content providers and aggregators.

2008 NCTA Technical Papers - page 202

 PC & CE Devices – PC and CE devices that
Encryption / DRM – perform encryption and communicate with the VOD system and typically
digital rights management packaging on the receive video over IP via CMTS.
content.
OTHER CONSIDERATIONS
Asset Library Service – maintain and update a
directory of content locations in the CDN / Content Formats and Metadata
libraries.
The content can be realized as various types
Asset Propagation Manager – manage content of media files and real time streams targeting the
replication and movement among various library end devices. Some of the most popular content
storage locations based on the external business formats are described in the following Table:
rules and actual content usage.
Table 1. List of Content Formats
Library Storage – provide persistent content
Format Resolution / End Device Delivery Use Cases
storage or temporary content caching at various Bit Rate (Typical)

library locations. MPEG-2 SD / 3.75 Mbps

HD / 15 Mbps
STB QAM VOD, StartOver

MPEG-4 / H.264 SD / 2 Mbps STB / PC QAM / IP VOD, StartOver

IP Networks – provide national and regional IP VC-1

HD / 8 Mbps
SD / 2 Mbps STB / PC QAM / IP VOD, StartOver
networks that connect multiple library locations HD / 8 Mbps
Window Media 800 – 1500 kbps PC / Portable IP Internet
and local VOD systems. Streaming
Flash Streaming 400 – 1500 kbps PC / Portable IP Internet

Navigation & Entitlement – provide navigation Audio (Dolby AC-3) 192 – 384 kbps STB QAM Music Choice

and entitlement functions to content requested Audio (Window

Media, MP3)
64 – 128 kbps PC / Portable IP Internet

from client devices. Image (JPEG, PNG,

Bitmap)
Various PC / Portable / STB IP Graphics, Photo
Sharing
Files Various PC / Portable / STB IP Application
Download

Session & Resource Manager – manage session

life cycle and its associated resources for on
demand video services requested by subscribers. In addition, Advertising content with similar
content formats can be delivered via the CDN.
Streaming Server – store and outputs content SCTE 35 parsing and Ad splicing will be
and enables stream control. performed at the Streaming Server under the
direction of the VOD BackOffice that interfaces
Edge QAM – perform re-multiplexing and QAM with external Ad decision system.
modulation.
Traditional VOD content is identified using
CMTS – Cable Modem Termination System for the CableLabs Asset Distribution Interface (ADI)
DOCSIS enabled devices. Provider ID and Asset ID. In addition, the
metadata for the content is described in the
STB Client – digital set-top box and its CableLabs ADI 1.1 or ADI 2.0 standard. The
client software that communicate with the VOD content identifier and metadata structure need to
system and typically receive video over MPEG-2 be extended to support Internet video that uses
transport via an Edge QAM. Internet media publishing standard such as Really
Simple Syndication (RSS).

2008 NCTA Technical Papers - page 203

 The locations for all content within the CDN
Content Ingest is maintained and updated by the Asset Library
Service (ALS). Upon a session setup request
The CDN can support ingest of content files from the subscriber, the VOD session and
from traditional satellite based catchers as well as resource manager will interface with the
content files and real time streams via an IP Streaming Server for the selected content. If the
backbone from VOD content providers or content is already pre-positioned or cached at the
Internet Video providers. Content processing Streaming Server, it will stream the content to
may be required upon the content ingest: the subscriber. If the content is not available at
the Streaming Server, it will query the ALS for
• Content may need to be transcoded to a the locations of the requested content within the
different resolution, bitrate, and codec CDN in order to fetch the content from the
upon ingest to the CDN content library and stream to the subscriber.

• Trick files (Fast Forward, Rewind) may

need to be created upon content ingest Content Streaming
into the CDN
When the Streaming Server fetches the
• Content with multiple formats (e.g. HD content from Content Library and streams to the
vs. SD) for the same title are treated as subscriber, it will use the Content Transfer
different content assets with different Protocol from Content Library to Streaming
content metadata Server. The Content Transfer Protocol may be
extensions to one of the existing standard
• The metadata and rights management are protocols such as:
performed on the content. This includes
content life cycle management such as the • NFS
licensing window • CIFS
• FTP
• Encryption and Digital Rights • HTTP
Management (DRM) are performed on
the content upon ingest. It is also Open standards, scalability, and performance
possible that tier based or session based are some of the key criteria for the selection and
encryption can be performed upon design of the Content Transfer Protocol.
streaming
Asset Propagation Management Operations and Reporting

The CDN contains multiple Content Library An operation model for system monitoring
nodes connected via national and regional IP and management is required. Specifically, it will
networks. The Asset Propagation Manager include the following aspects:
(APM) is responsible to replicate and/or move
the content through the storage nodes of the CDN • Component level fault monitoring and
dynamically based on the content popularity and management
usage. • Content level status monitoring
• Network level monitoring

2008 NCTA Technical Papers - page 204

 • Video quality monitoring SUMMARY

In addition, key reporting metrics need to be This paper describes an architecture

defined: framework for large VOD libraries based on the
• Viewing patterns for content Next Generation On Demand (NGOD)
• Network bandwidth usage architecture and its extension to the Content
• Storage bandwidth usage Delivery Network (CDN). The architecture
• Peak number of streams and concurrency combines the benefits of Managed Network and
rate Over the Top approaches. It will enable
• Measurement of efficiency of caching distribution of any content to any device, at any
algorithm time. The architectural building blocks are
presented and some of the key challenges and
Scalability, Performance, and Reliability design considerations are discussed. On going
work includes detailed architecture and interface
Scalability, reliability, and performance specifications as well as performance and
requirements may include the following: scalability analysis.

• Daily hours of content ingest REFERENCES

• Library storage sizing
• Target streaming capacity [1] Evolution of Video On Demand
• Jitter Architectures, Weidong Mao & Kip Compton,
• Distribution metrics (delay/latency) May 2004, NCTA Technical Papers
– Initial Request
– Peak Utilization [2] CableLabs Video On Demand Content
• Caching versus network bandwidth Specification Version 1.1, August 31, 2006
tradeoff
• [3] CableLabs Asset Distribution Interface
VOD for Multiple Devices Specification Version 1.1, May 5, 2006

It is highly desirable to share the large VOD [4] CableLabs ADI 2.0 Specification Asset
libraries for multiple end devices such as PCs Structure, January 5, 2007
and other portable media players.
[5] ISO/IEC 13818-6: MPEG-2 Digital Storage
There are several aspects which should be Media – Command and Control (DSM-CC)
considered when expanding the architecture to
support any content to any device. [6] ISO/IEC 13818-1: MPEG-2 System

 Content format transcoding [7] ISO/IEC 13818-2: MPEG-2 Video

 Session and resource signaling
 Digital Rights Management (DRM) [8] IETF RFC 2326, Real Time Streaming
 Home networking Protocol (RTSP)
 Subscriber and device authentication
 Cross application platform [9] ISO/IEC 14496-10, Advanced Video Coding

2008 NCTA Technical Papers - page 205

 CABLE’S MOBILE FUTURE: WHICH TECHNOLOGY AND WHY?
Jay Bestermann
Director Product Development, ARRIS
[email protected]

Tarun Chugh
Sr. Software Engineer, ARRIS
[email protected]

Abstract Vulcan Ventures), and Bend Broadband

winning additional spectrum. On March 26,
This paper examines the wireless market and 2008, the Wall Street Journal reported Comcast,
network technology, WiMAX or Long Term Time Warner, Brighthouse and others are
Evolution (LTE), cable operators will likely considering a WiMAX joint venture with
deploy in the not too distant future. Key market SPRINT and Clearwire.
trends such as wireless growth, competitive
threats and incumbent carrier landline erosion The previous events are noteworthy
are explored. Technologies are compared because they highlight a clear interest from the
based on expected mobile device availability, MSO community in bringing a credible wireless
access technology, core network architecture, offering to market. This paper examines the
and roaming capability. The value of converged prevailing market trends and concludes with a
services is identified as a cornerstone to cable solution recommendation.
operators’ wireless strategy. Finally a network
technology recommendation is made based on WIRELESS GROWTH
the previous market and technology analysis.
According to recent press, and the latest
INTRODUCTION Quarterly reports from AT&T and Verizon,
wireless subscriber growth is driving their
Over the past 3 to 5 years North overall company revenue growth. The graph
American cable operators have invested below shows the revenue growth of the wireless
significant resources to add cellular service to organizations within Verizon and AT&T. Each
their portfolio with varied results. The first of these corporations have experienced year
major step was the announcement of the over year growth in excess of 11.8% for the past
SPRINT Joint Venture in November of 2005. 3 years.
This was a very celebrated development, but it
hasn’t provided an effective Quad Play offering. Wireless Revenue

The next step was when major cable operators 45

Billions

in the US, via Spectrum Co, purchased a

43
40

nationwide footprint of AWS spectrum in

38 38
35
Revenue

34 Verizon

September of 2006. This was again a major 30

33 ATT

development, but the companies have yet to

28

25

officially announce plans for the use of this 20

spectrum. More recently, several major MSOs

2005 2006 2007
Year

have entered the January 2008 FCC 700Mhz

auctions with Cox, Charter (via Paul Allen /

2008 NCTA Technical Papers - page 50

 Over the past few years revenue growth with Finland leading the charge at an incredible
within wireless business units has been driven 54%.
by voice, but as this market matures, revenue
growth will be driven by value added data
services such as SMS, email, and MMS. For
instance, AT&T wireless has experienced
exponential growth rates of mobile data usage.
Usage of this service on their network has at
least quadrupled for each of the last 4 years.

According to CTIA-The Wireless In recent mobile history, Europe has

Association®, the average US cellular been a leading indicator for wireless trends in
subscriber spends $50 per month and there were the US for services such as SMS and data.
243.4M wireless subscribers in June of 2007. Assuming this applies to users cutting the cord
This is a staggering $146B dollar annual as well, the US is going to follow the European
revenue source for carriers. Comparing this to trend that is taking place today. To back this
MSOs’ most popular and highest ARPU service theory the table below from In-Stat4 shows very
today, the National Cable Television strong interest in users migrating to a wireless
Association (NCTA) reports US cable operators only paradigm in the US.
have 65.1M video households as of September
2007 with an approximate ARPU of $60 for Interest in Going All Wireless
video service (Comcast 2007). This represents a
market of $46.9B. The point of this comparison
is to illustrate, using current statistics, that
offering a cellular service is a potential growth
opportunity for MSOs. Cable Operators are According to NCHS National Health
already reaping the fruits of their landline Interview Survey, there were 13.6% wireless
efforts, but this opportunity will certainly only households in the United States in 2007.
diminish in the coming years. A viable and
even obvious next step is to target a sliver of the 16
US Wireless Only Households

wireless market representing a powerful new 14

13.60

ARPU growth engine. This growth engine for 12

% of Households

12.6
11.8
10
MSOs will also serve to neutralize the predicted 8
7.7
9.6

negative growth due to competition from the 6

4 4.4
5.4
6.7

wide deployment of carrier-based services such 2

as AT&T U-verseSM and Verizon FiOSTM.

0
1/2004 – 7/2004 – 1/2005 – 7/2005 – 1/2006 – 7/2006 – 1/2007 – 6/2007 –
6/2004 12/2004 6/2005 12/2005 6/2006 12/2006 6/2007 12/2007*

Year

In recent years subscribers have been

transitioning to a wireless only voice Assuming a continued average of 2.25%
communication paradigm. This transition is growth per year until 2010, the US will have
fueled by the improved reliability and approximately 20% wireless only households.
affordability of wireless communication and This prediction will directly impact the wireline
subscribers’ passion for mobility. As shown in market going forward. Below is a chart of this
the graph below from OECD, the European prediction.
average for wireless only households is 22%

2008 NCTA Technical Papers - page 51

 US Wireless Only Households ( Predicted )
25

20 20.50
% of Households

19.35
18.20
17.05
15 15.90
14.75
13.60
11.8 12.6
10
9.6
7.7
5 6.7
5.4
4.4

0
1/2004 – 1/2005 – 1/2006 – 1/2007 – 1/2008 – 1/2009 – 1/2010 –
6/2004 6/2005 6/2006 6/2007 6/2008 6/2009 6/2010

Year

The predictions above are a conservative

estimate when the following trend is considered. This growth has been the result of the
The younger a subscriber is, the more likely he significant landline erosion that has been a
or she is to be a wireless only consumer. The market force over the last few years in the
key message in the next graph is this trend will traditional fixed line operator space. The
likely shift to the right as each demographic traditional landline incumbents have taken the
grows older versus remaining as an age-defined brunt of landline erosion as well as market
wireless only demographic. The younger competition from cable operators and pure play
generations will be more comfortable with VoIP providers such as Vonage. The graph
mobile technology and thus more willing to rely below shows the landlines lost by the traditional
on it as their sole communication device, but as landline carriers over the past 4 years. One
they grow older their demand for mobile detail to notice in the graph is the only growth
services will not diminish. realized by either of the major carriers in the
United States was via acquisition. In particular
Wireless Only Subscribers SBC acquired Bellsouth in 2005 which resulted
in significant landline growth for the new
35 corporation, but it is still very clear that the
30
overall trend is landline erosion within the
25
incumbent landline carriers. For instance,
20
Verizon has lost 26.5% of its residential landline
Percent subscriber base over the analyzed time frame
15
below.
10

Jan-07
5 Jul-06
Jan-06 Residential Wireline Subscribers
Jul-05
0 Jan-05 70
Year 66
Millions

18-24 Jul-04 62
25-29 Jan-04 60
30-44 52
Age 45-64 49
> 65 50
Subscribers

40
AT&T
34
31 Verizon
30 28
25

LANDLINE MARKET
20

10

0
2004 2005 2006 2007

In recent years cable operators have Year

benefited significantly from strong subscriber

landline growth as shown in the graph below Despite cable operators’ significant
from NCTA. landline growth in the last couple of years the
overriding industry trend is landline erosion as
shown in the diagram below. As shown by
March 2008 data from the FCC Wireline

2008 NCTA Technical Papers - page 52

Competition Bureau, during the past 5 years, the These trends are being driven by the
total landline market compressed by 15.1%. consumer’s passion for wireless connectivity
This doesn’t highlight other overriding trends and continuous connectivity to friends, family
such as price erosion, but it certainly shouldn’t and associates. Wireless Minutes of Use
be ignored. actually surpassed residential landline usage in
2003 and the population continues to become
US Landlines more and more reliant on wireless devices. The
195
diagram below shows current and historical and
Millions

192 192

wireless Minutes of Use (MOU). There is

190 189
189

185 185
183
certainly a peak to this trend, but for the
Access Lines

180 180
178 178
175 175
172
foreseeable future landline erosion and wireless
170

165
167 MOU growth will continue.
163
160
O 1

O 2

O 3

O 4

05

06

07
02

03

04

05

06

07
1

Wireless Minutes of Usage

0

0
-0

-0

-0

-0

-0

-0
n-

b-

n-

b-

n-

b-

n-

b-

n-

b-

n-

b-

n-
ct

ct

ct

ct

ct

ct
Ju

Fe

Ju

Fe

Ju

Fe

Ju

Fe

Ju

Fe

Ju

Fe

Ju
O

Year
1200

1000

Using the data from the growth plot 800

Minutes (B)
above and assuming continued -5% annual 600

growth, the projected carrier landline market is 400

shown in the next graph. On this same graph an 200

annual growth rate of 15% for MSO landline 0

2000 2001 2002 2003 2004 2005 2006 2007

growth is assumed for the next 2 years. After Year

this period the growth rate reduces 2% per year

until the MSOs reach 40% penetration of their To further accelerate the previously
123M current homes passed. Based on the discussed trends, incumbent wireless operators
previous assumptions cable operators will likely are increasing the pressure on landline Minutes
experience flat landline growth in the 2012 of Use and thus the profitability of landline
timeframe. operators’ business. Verizon Wireless, AT&T,
T-Mobile, and SPRINT all announced unlimited
Landline Subscriber Growth calling plans early in 2008. These unlimited
180
plans will likely drive further growth of wireless
Millions

160

140
MOUs and further commoditize landline voice.
120

T-Mobile is the most aggressive in its

Landlines

100
Wireline
Cable

offering of the Hotspot@HomeTM and Talk

80

60

40
ForeverTM services. The T-Mobile
20

0
Hotspot@Home service is a $9.99 per month
2007 2008 2009 2010 2011 2012
Year
2013 2014 2015 2016
service add-on to a minimal wireless plan that
provides unlimited voice and data using UMA
enabled dual-mode handsets over Wi-Fi. The
The key takeaway from this analysis is
Talk Forever service is a UMA enabled Analog
as follows: Once MSO landline growth stops it
Terminal Adapter that offers a Vonage-like
is likely that the overriding trend of landline loss
service to T-Mobile customers. This service is
will then start to affect cable operators as it has
available in limited markets at the time of this
affected the incumbents for the last few years.
writing, but is planned for a nationwide launch
at $9.99 as an add-on to an existing wireless
plan.

2008 NCTA Technical Papers - page 53

 The goal of these new services is best contrast the CDMA Development Group reports
stated by T-Mobile USA CEO, Robert Dotson 431M CDMA subscribers in 97 countries
in a June 27, 2007 release, “More people than resulting in approximately 13% of the global
ever are looking to drop their home landline wireless market share in 4Q2007.
phone and pocket the savings. However, they
don’t want to use all their wireless minutes The above comparison can then be
talking from home. Our new service solves this applied to the current handset market for each of
dilemma once and for all. T-Mobile’s these network technologies. The CDMA
HotSpot@Home is a first-of-its-kind service Development Group reports that 1,950 devices
that helps people simplify their lives, save have been introduced in to the market including
money, and enjoy great call quality on one 509 1xEV-DO Rel. 0 and 48 1xEV-DO Rev A
device — their mobile phone — at home”. Joe devices. This is a historical figure over the
Sims, VP of T-Mobile Broadband Products was lifetime of CDMA 2000 technology. GSM
quoted by Wi-Fi Net News stating “T-Mobile is Arena reports current GSM device availability
looking to address the remaining reasons people at 1159 in March 2008. This shows that the
were reluctant to cut the cord.” current number of commercially available GSM
handsets in the market today is over half of the
The expectation is that unlimited calling lifetime total number of CDMA2000 devices.
plans will force the major wireless operators to This comparison is very useful when thinking
offer solutions similar to the ones T-Mobile has about the next technology choices that MSOs
in its portfolio for various reasons. With will make or are already making. 3GPP Long
unlimited calling plans, subscribers will become Term Evolution (LTE) and WiMAX handset
more and more reliant on their mobile devices, availability, variety, and cost will clearly be
thus demanding better in-home coverage and driven by the addressable market size. Below is
bandwidth. Assuming uptake is high, the a subscriber growth prediction using data from
unlimited services will require operators to add Senza-Fili and Analysys. The Senza-Fili data
additional network capacity to keep customer was extended to project 2014 and 2015 for this
satisfaction at high levels. Another key driver paper. The current projection is LTE will
of these new converged services is the quickly overtake WiMAX (including Total
undeniable cost advantage of landline vs. Fixed and Mobile) deployments in the
mobile MOUs. beginning of 2013.

NETWORK TECHNOLOGY Predicted WiMAX vs LTE Growth

500
Millions

450

The key consideration during the 400

selection of a new mobile access technology is

350

300
Subscriber

the future expectations of access device cost and

LTE
250
WiMAX ( Mobile and Fixed )
200

availability. Both of these factors are typically 150

tied to subscriber bases with the capability to

100

50

drive the largest handset volumes. Today GSM 0

2009 2010 2011 2012 2013 2014 2015

and CDMA are the prevalent network Year

technologies and handset costs and varieties

follow the afore-mentioned trend. According to Technology decisions will also affect
3GAmericas.org, as of 3Q2007, there are 2.7B roaming relationships that are possible between
GSM subscribers in 220 countries resulting in networks and operators. LTE has a clear
86% of the global wireless market share. In advantage over WiMAX in this area because of
the clear evolution path as technologies evolve.

2008 NCTA Technical Papers - page 54

3GPP GSM networks are evolving from GPRS roaming. For example, if WiMAX devices are
to EDGE to HSPA today and the next evolution deployed by SPRINT they will be the only
to LTE is an extension of that experience. The established US operator deploying a
primary enabler of this evolution is a healthy WiMAX/CDMA2000 technology combination.
handset ecosystem that will build handsets that As previously mentioned, Verizon Wireless has
support the upcoming network deployments. A made public commitments to deploying a LTE
handset that supports EDGE, HSPA, and LTE network in its next round of technology
will be able to get network access virtually upgrades. This implies SPRINT will not have
anywhere in the world in the coming years. an established nationwide WiMAX roaming
This is a critical capability that must not be carrier with a large network footprint or the
ignored as operators build out “greenfield” subscriber base to support a nationwide network
networks that have limited coverage areas. deployment. SPRINT will have to rely
Technology inclusion and thus roaming primarily on its network deployments for
relationships for WiMAX devices are very WiMAX coverage or utilize CDMA2000 based
unclear. SPRINT and Clearwire have roaming. This also implies SPRINT will
announced WiMAX network(s) in the US but receive little WiMAX based roaming revenue
primarily for laptops and fixed wireless access from other US based operators except
with a transition to mobile devices. The next Clearwire. On the other hand, Verizon’s LTE
step is offering mobile WiMAX in portable approach will immediately expand its roaming
devices. Given the limited WiMAX coverage partner ecosystem in North America and abroad.
areas, a combination WiMAX/ CDMA2000 Verizon Wireless’ 50% owner, Vodaphone, has
handset will likely be required. This announced intentions of deploying LTE as have
combination will have far less volume and a AT&T, China Mobile and NTT DoCoMo. It is
much smaller ecosystem than the 3GPP driven likely that many of the over 200 3GPP operators
LTE standard. This is driven by the uncertainty will follow this clear evolution trend. This
or non-linearity in the technology evolution roaming ecosystem will provide Verizon access
path. Because SPRINT has an existing to a worldwide roaming based revenue engine
CDMA2000 network they will require a as well as offer its customers a much better
CDMA2000 /WiMAX handset solution until a service availability. Verizon isn’t a true
nationwide WiMAX network is in place. greenfield operator as most cable operators are
This is in direct contrast to the recent decision today, but its decision to deploy LTE
that Verizon has made to utilize LTE technology makes it very similar because of the
technology in its next generation network technology discontinuity.
upgrade. This decision will divide the US
CDMA2000 subscriber base between Technically speaking, WiMAX and
WiMAX/CDMA2000 and LTE/CDMA2000 in 3GPP’s LTE are very comparable access
the United States. There are many other technologies. Some of the access technology
potential handset technology combinations such highlights are as follows: OFDM-based, similar
as WiMAX/EDGE/HSPA that could be modulation techniques, theoretical throughput
deployed by existing 3GPP or greenfield and capacity. One notable exception to this rule
operators. This decision again would result in a is that WiMAX is typically a TDD solution
divided market in comparison to the clear 3GPP whereas LTE is typically a FDD solution.
evolution path and global scale. WiMAX does have a FDD profile, but this
hasn’t been deployed. The data in the table
The technology decision made by MSOs below gives a more thorough comparison.
will also affect the opportunity for inter-carrier

2008 NCTA Technical Papers - page 55

 Bit-rate/site(DL)
3GPP LTE
100Mbps(MIMO 2TX, 2 RX)
WIMAX
75Mbps(MIMO 2TX, 2RX) as Email, IM/presence, Web browsing, etc.
Bit-rate/site(UL) 50Mbps 25Mbps

Base Standard E-UTRAN IEEE 802.16e

This single statistic puts the cable operator with
Duplex Method
Downlink
FDD
OFDMA
TDD(FDD optional)
OFDMA
a wireless offering in a very good position to
Uplink Multiple Access
Channel BW
SC-FDMA
1.25 - 20Mhz
OFDMA
Scalable:4.375, 5,7,8.75,10 Mhz offer compelling new service bundles and
Modulation DL
Modulation UL
QPSK/16QAM/64QAM
QPSK/16QAM/(64QAM opt)
QPSK/16QAM/64QAM
QPSK/16QAM capabilities required to compete in a hyper-
Cell Radius
Spectral Efficiency
5km
5[bits/sec/Hz]
2-7km
3.75[bits/sec/Hz]
competitive marketplace. In 2007 there were
Cell Capacity
>200 users @ 5Mhz
>400 users for larger BW 100-200 users
123 Million households passed by cable
operator networks. If we assume an average of
From a core network architecture 2 persons per household, MSOs have the ability
perspective both WiMAX and LTE have been to offer converged services to 246 Million
designed to be very flat IP-based solutions that subscribers. Compelling new services such as
interconnect with an IP Multimedia Subsystem, these will be required to create an impetus for
but there is a key difference. LTE has been change and drive the current 50% service
architected from the beginning to support penetration enjoyed today even higher.
seamless handover and global roaming to
LTE/2G/3G networks. WiMAX mobility is Device convergence is the concept of
based on mobile IP and hasn’t addressed inter- embedding both Wide Area Network and Local
Radio Access Technology handover or global Area Network technology in mobile devices.
roaming scenarios. The most common example today is Wi-
Fi+GSM in a single handset, but in future
The current standardization, deployment deployments this will likely become Wi-
and mass market timelines show WiMAX Fi+WiMAX+CDMA2000 or Wi-
reaching market in 2007, which has already Fi+LTE+GSM. Femtocells can also be viewed
happened. WiMAX mass market adoption is to as device convergence at a slightly different
begin in 2009. LTE is trailing WiMAX with layer. The femtocell combines traditional Wide
expected deployments to start in 2010 with mass Area Network technology such as CDMA2000
market adoption in 2012. with high speed local area network backhaul.
The purpose of all Device Convergence is to
CONVERGENCE OPPORTUNITY make use of high bandwidth, low cost local area
networks when they are available, but make the
As cable operators become more serious user experience very simple, cost effective, and
about bringing a wireless offering to market, truly next generation.
they must consider the clear advantage of
adding a flavor of either device or service Service Convergence is the idea of
convergence to their solution. They must blending physically independent device types on
leverage the next generation access and one or multiple carrier networks utilizing
transport networks that have been put in place to intelligent core networks. This blending occurs
serve their other business needs. This through simultaneous ringing of the independent
investment can be further exploited with an device types and allows the call to be moved
offering that addresses both device and service between different terminals very easily. For
layer convergence. This need becomes obvious example, imagine walking into your home on an
when examining the calling patterns of wireless active AT&T cell call and with a single key
subscribers. In the US the average consumer press moving that active call from your mobile
makes at least 40% of calls from an indoor device to your cable operator managed landline.
home or work environment. This doesn’t even This would allow the user to select the current
consider continuous data usage for services such device of convenience, best performance, least

2008 NCTA Technical Papers - page 56

cost, or comfort as they prefer. This type of
blending would offer a compelling reason for
consumers to keep a landline in the home and
thus reduce the current trend of landline erosion.

Demand for these services has been

shown by recent studies. Of those surveyed,
49% of subscribers rate coverage at home and
42% rate voice plan pricing as the most
important factors when selecting a cellular
carrier. These are the top two factors which
drive a consumer’s wireless decision. This
DOCSIS 3.0 has shown the ability to improve
clearly shows that a device convergence strategy
download speeds at least 4X (40 Mbps to 160
which attacks both of these issues head on, is a
Mbps) over the current DOCSIS 1.x/2.0
win-win offering for the consumer and operator.
technology that is deployed in today’s networks.
This is proving to be a competitive advantage
One of the key issues in deciding to
because of the relatively low capital investment
deploy a wireless network is developing an
required to deliver this level of bandwidth to a
offering that is going to be compelling enough
consumer’s door. In the not too distant future
to entice a subscriber to switch providers.
converged devices play a significant role to
Based on a recent report, a converged solution is
further leverage this investment as part of the
extremely or somewhat interesting to 92.7% of
cable operator’s wireless strategy. Using their
subscribers.
HSD infrastructure cable operators can vastly
improve the user experience and decrease the
cost per bit significantly.

A device convergence strategy addresses

an obvious problem MSOs will encounter with
indoor coverage as they deploy their new
networks. As an illustration of the issue the
diagrams below show indoor penetration of 2G
From this same report subscribers are Voice, 3G Voice, 3G 64K data, and 3G 384K
demanding a converged voice experience. Of data by cellular base stations.
particular interest to the cable industry is the
perception that a fixed-only life is very
undesirable to the study group. This indicates a
user faced with a forced selection would prefer a
mobile-only life rather than a tethered one, but
the most desirable solution by far is a mixed
offering.

Notice the higher the data rate the less

effective in-home coverage becomes. This is a
well known characteristic of higher bandwidth

2008 NCTA Technical Papers - page 57

and/or higher frequency channels driven by the customer satisfaction and revenue generation
laws of physics. It may not be intuitive but this capability. One potential path to LTE to satisfy
phenomenon will also affect outdoor users in operators need for accelerated deployment is a
terms of effective available bandwidth. The staged approach of deploying 3G HSPA today
indoor users will consume more of the cell site’s and upgrading that network to LTE in the
resources because adaptive modulation coming years. This strategy would allow for
techniques are utilized in mobile networks. The early market entry with field proven technology
indoor user’s device will be using lower that is comparable in performance to today’s
modulation schemes and hence require more of WiMAX solutions.
the available bandwidth than outdoor users to
transmit the same amount of data. These facts Finally, a convergence strategy is a
show a converged solution will alleviate differentiator that will help insure a successful
pressure on the outside network, therefore market entry for MSOs by significantly
reducing the network capital investment reducing capital expenditure and providing
required to offer best in class coverage and subscribers with yet another reason to migrate to
bandwidth to the end user. cable operators’ networks.

CONCLUSIONS References:

1. "Statistical Trends in Telephony 2007 Report”, FCC

From the market analysis above it is 2. "A Comparison of Two Fourth Generation
clear that cable operators must enter the wireless Technologies: WiMAX and 3GPP LTE",
space to, at a minimum, prevent contraction of http://www.comsysmobile.com/pdf/LTEvsWiMax.pdf
their existing business. Furthermore a much 3. "2008 Corporate Brochure", GSMA
larger opportunity exists to accelerate revenue 4. "HSPA and mobile WiMAX for Mobile Broadband
Wireless Access", gsmworld
growth by entering and capturing a small 5. "EDGE,HSPA and LTE The Mobile Broadband
portion of the wireless subscriber base that is Advantage",
searching for a true Quad-Play service offering. http://www.rysavy.com/Articles/2007_09_Rysavy_3GAm
Cable operators have a lead in bandwidth to the ericas.pdf
home and a best in class content offering today, 6. CTIA mid year surver 2007, CTIA
7."Wireless Quick Facts", CTIA
but the carrier community is attacking this safe 8."Industry Statistics", NCTA
haven with relentless vigor. With the addition 9."Verizon Communication Inc., Form 10-K"
of wireless it is quite clear that the MSOs will 10."AT&T Inc. Form 10-K"
continue to be a formidable competitor. 11."Comcast Corporation. Form 10-K"
12."Time Warner Cable Inc. Form 10-K"
13."Wireless Substitution: Early Release of Estimates
Analyzing market data and comparing From the National Health Interview Survey,January-June
technology capabilities leads to a 2007, NCHS
recommendation of LTE as the network 14.“T-Mobile’s VoIP Home Service: Goodbye to the
technology best fit for MSO deployment. It is PSTN”, In-stat
evident that the global scale of LTE insures 15.”3G Femtocell, Fixed-Mobile Convergence”, Light
Reading Femtocell Webinar.
competition as well as innovation in the handset 16. "Global Mobile Broadband: Market potential for 3G
and network equipment space via a vast vendor LTE", Analysys
ecosystem. Its technical capability is second to
none with the key element being handover and
global roaming capability with not only LTE but
legacy 3GPP and even CDMA EV-DO Rev-A
networks. This is paramount when it comes to

2008 NCTA Technical Papers - page 58

 COST EFFECTIVE WATERMARKING IN THE SET TOP BOX
Joseph Oren
Cinea Inc. a Dolby Company

Abstract DISCLAIMER

Providing Cable consumers with premium The author of this paper, Joseph Oren, is
(e.g. HD or early window) content via Video on employed by Cinea Inc., a Dolby company.
Demand services is projected to become a key Cinea offers commercial products utilizing
revenue source for system operators. Yet the certain technologies described herein.
Hollywood studios insist that before this content
will be made available, enhanced content
protection technologies must be deployed within INTRODUCTION
the content distribution infrastructure i .
Specifically, forensic watermarking, defined as DRM and CA technologies have made great
the binding of unique traceable information to strides toward system recognition of the rules
the video streams, is increasingly mentioned as agreed upon by content owners and consumers.
an essential content protection layer, one that The available mechanisms to enforce those rules,
complements existing conditional access and termed content protection, is, however, limited
digital rights management solutions. ii This to encryption during transmission and storage.
paper describes how this new business Once the content is rendered in a consumable
requirement can be technically and form, its digital and analog representations
economically fulfilled by watermarking become subject to copying and subsequent
technologies now reaching the market. Our unauthorized redistribution (piracy). Figure 1
focus will be on watermarking technology shows a simplified receiving device, with
implemented in the consumer’s equipment, vulnerabilities identified. A real-world home
commonly called the Set Top Box (STB). network may spread these functions over several
devices, each with analogous vulnerabilities.

2008 NCTA Technical Papers - page 65

 Digital Probe Circumvention Device

PROTECTED DIGITAL LINK ENCRYPTION

CONTENT

DIGITAL MONITOR
DIGITAL
DRM / DESCRAMBLER DECODE
PLAIN-TEXT
BASEBAND

Resampling

RECEIVER
(STB)
D TO A
CONTENT
LOCAL DRM STORAGE

ANALOG MONITOR

CAMCORDER

Figure 1 – Receiver with vulnerabilities identified

These vulnerabilities have fed the interest in While the DRM acts to constrain the user,
forensic watermarking technology, whereby actually challenging him to circumvent the
each instance of a content item is individuated technology, forensic watermarking deters piracy
by information to facilitate the tracing of the by introducing risk of exposure. In the case of
content back to its last legitimate holder. large scale re-distribution, forensic
Tracing produces valuable evidence in watermarking facilitates identification of the
identifying copyright violators. Further, since point of compromise. With both DRM and
forensic watermarking is an investigative tool, watermarking available, a more balanced and
as opposed to a control tool, it offers the appealing approach to content protection
potential to obviate some of the more complex becomes possible.
and consumer hostile aspects of strong DRMs.

2008 NCTA Technical Papers - page 66

 Figure 2 – Forensic watermarking identifies the source of unauthorized distribution

S c r a m b le d
C o n te n t

HOM E NETW ORK

S e r ia liz e d
( In d iv id u a te d )
C o n te n t
D IG IT A L C O N T E N T
D IS T R I B U T O R
C O N T E N T A C Q U IS IT IO N
D E V IC E ( e g . S T B )

PO RTABLE
D E V IC E

H O M E M E D IA
SERVER

U n a u th o r iz e d
P IR A T E S E R V E R
R e d is tr ib u tio n

Id e n tifie s
STB

W ATERM ARK
RECOVERY

Figure 2 illustrates how forensic technology provides a means of embedding such

watermarking is used to investigate media information into content, through subtle
piracy. In this example, the STB binds an alterations to the content itself.
identifying watermark to the content it acquires
from the network. The watermark does not It is important to differentiate watermarking
interfere with the consumer’s enjoyment of the from simply appending identifying metadata to
content. But, if the consumer chooses to the content. Identifying metadata can be
circumvent the DRM and distribute transparently excised from the content, while
unauthorized copies, the watermark can be used erasing a watermark requires specific
to determine his identity. Awareness, on the manipulation of the content itself.
part of the consumer, of the possibility of Watermarking does not impair the content, but
exposure serves to discourage careless removal of a forensic watermark without
redistribution of the content. impairing content quality is, by design, a very
difficult task. We will examine the
Also called media serialization or content requirements for an effective forensic
tracing, this paradigm is analogous to the use of watermarking implementation and how those
serial numbers to track physical machinery or requirements can be addressed within the
other valuable products. Watermarking constraints of the STB.

2008 NCTA Technical Papers - page 67

 WATERMARKING CONCEPTS AND Perceptibility
TERMINOLOGY
Watermarks may either be apparent to the
In the broadest sense, forensic watermarking viewer, when the content is rendered, or
is a steganographiciii technique, one that embeds disguised in such a way that the viewer is
data into an instance of a cover work, in such a unlikely to notice the presence of the watermark.
way that the data can subsequently be read Perceptible watermarks are commonly used to
(recovered) from copies of the watermarked proclaim ownership, exemplified by the visible
cover work. The cover work may be any logo appearing in many network broadcasts. In
communication medium, but we will focus on general, watermarks fall along a continuum of
digital video entertainment content. These perceptibility, according to the needs of the
principles may, however, be adapted for other users and the capabilities of the technology.
media, such as the audio channels. The field of steganography, the technology of
hiding messages in content such that the casual
The process of binding the watermark to a observer is unaware of the message’s existence,
content item is termed watermark embedding, includes imperceptible watermarking.
and the additional data, in its embedded form, is
the watermark itself. The process of reading the Readable vs. Detectable Watermarks
watermark from a copy of the content is termed
watermark recovery. A watermark may carry only a single bit of
information, that is, it is significant only in its
In a data communications model, the presence or absence. Such watermarks are
watermark information is data to be classified as detectable. A readable watermark,
communicated, and the cover work is a carrier on the other hand, contains a more complex
signal. Indeed, the cover work carrying the message, typically many bits of information.
watermark is often identified as the host signal. Mathematically, a readable watermark with N
In the communications channel, the perceptible bits of information could be conceptualized as
features of the content constitute noise that having been chosen from a set of 2N detectable
interferes with the watermark’s information watermarks. For a message of useful length, the
signal. It is important to recognize that number of marks in such a set becomes
watermark itself is embodied in changes to the unmanageable, so a practical readable
cover work features, as opposed to just being watermark implementation must include a
ancillary data. Any faithful reproduction of the means of decomposing the watermark to
content (the carrier) will also carry the reconstruct the message from independent parts.
watermark data. The watermark signal may
thus be viewed as modulating a noisy carrier Forensic watermarks must carry a message: a
signal, and thus becomes part of the cover work readable watermark, or a series of detectable
itself. watermarks is required to identify the particular
source of the content instance. If detectable
There are numerous watermarking watermarks are used, the message is treated as a
technologies, with varying degrees of suitability series of independent parts, each of which is
for specific applications. The attributes represented by a single detectable watermark.
commonly used to characterize a watermarking
technology are as follows:

2008 NCTA Technical Papers - page 68

Bandwidth terms, robustness tends to increase with the
amplitude of the watermark signal.
Bandwidth refers to the of data carrying Paradoxically, if the signal intensity level
capacity of the watermark, in proportion to the reaches the threshold of perceptibility, its nature
amount of content carrying the watermark. For and location become apparent to the attacker,
multi-media content, bandwidth is commonly thereby facilitating the attack. Consequently,
expressed in terms of message bits per second. the watermark intensity must be carefully
In interpreting bandwidth metrics, however, it is calibrated to achieve the required level of
important to distinguish between the original robustness.
message and an encoded message. Forensic
watermarking implementers may apply multiple As mentioned previously, error control
layers of error control coding (ECC) to the coding is an important contributor to robustness.
message, to compensate for the “noise” in the Alterations to the content may erase or distort
channel. Such coding can expand the message significant portions of the watermark signal.
several fold, thereby reducing the effective Effective recovery must include mechanisms to
bandwidth of the watermarking technique by the compensate for missing or erroneous signal
same factor. It is also common to embed elements. The watermark system communicates
several copies of the message into the content. over an extremely noisy channel, requiring
For a robust implementation, the bandwidth aggressive error control.
required is many times that which is implied by
the message length alone. FORENSIC WATERMARKING
REQUIREMENTS
Robustness
The primary requirement of a Forensic
Sometimes termed “survivability”, Watermarking application is the placement of
robustness is the degree to which the watermark the watermark embedder at a point in the
can withstand the various transformations the distribution network where the legitimate
content may undergo before reaching the recipient of the content instance is known. In
recovery process. An effective forensic uni-cast or download distribution models, the
watermark can tolerate operations such as content instance can be watermarked as it is
rescaling, resampling of analog signals, transmitted to the consumer. The consumer thus
recompression, cropping, rotation, resolution receives a unique copy of the content,
changes, deinterlacing, gamma changes, and individuated by the watermark that identifies
temporal averaging, all of which may occur in that consumer’s identity, account, or purchase
the course of pirating the content. Additionally, transaction. Any reproduction of the content
a pirate may undertake targeted attacks to instance can thereby be traced to the consumer
directly suppress the watermark by filtering, when the Forensic Watermark is recovered.
noise addition, collusion or other video
processing techniques. In broadcast or multi-cast systems, however,
each consumer receives an identical copy of the
Although no watermarking technique is content. It is thus only possible to individuate
unconditionally robust, an effective technique the content instance in the consumer’s content
requires the adversary either to apply an receiving device. Consequently, these
unreasonable amount of effort to suppress the distribution models require authentication of the
watermark, or to unacceptably impair the receiver, and sufficient security in the receiver
content in the process. In signal processing

2008 NCTA Technical Papers - page 69

to ensure that the watermark is correctly
embedded. Figure 3 – Forensic watermarking in a broadcast
distribution model

the content path also ensures that all outputs of

Figure 3 illustrates forensic watermarking in a the device are protected.
broadcast distribution model. A watermark is
applied following decryption, under the control Embedding Performance
of the DRM. The content made available to
subsequent processes will have been A closely related requirement is performance.
individuated by the forensic watermark. The STB platform typically lacks substantial
spare processing and memory resources.
Watermark Integrity Watermarking in a receiving device must take
place at rendering speed for real-time streaming
To prevent compromise, and thus obtain the content. Devices supporting background
maximum benefit from forensic watermarking, downloads may require watermarking at
the watermark should be embedded immediately network speeds exceeding real-time. More
following decryption. The physical security sophisticated home network devices may
envelope in the device must enclose all require simultaneous watermarking of multiple
processing inclusive of the DRM and the content streams. Thus the watermark embedder
watermark embedding to prevent interception must only minimally impact the STB
prior to watermarking or other circumvention. computational resources.
Any accessible data paths carrying unprotected
content invite interception and can be targeted Imperceptibility
by pirates. Once the forensic watermark is
embedded, however, the content becomes As noted above forensic watermarks must be
traceable. Traceable content is less attractive to imperceptible. The system objective is to
the pirate, due to the increased risk of exposure, preserve the value of the content, so significant
and, consequently, is somewhat less demanding quality impacts are unacceptable.
of physical protection. Watermarking early in Imperceptibility is particularly important for

2008 NCTA Technical Papers - page 70

high definition content, which the consumer targeted watermark technology will be rendered
expects to be of the highest quality. ineffective. Renewing the watermark system
forces the pirate to analyze a new technique and
Robustness adapt his countermeasures. Thus the ability to
renew watermarking techniques, by varying the
Robustness is, of course, critical in forensic watermark signal, is a hallmark of an effective
watermarking. The system is only effective in system.
exposing pirates to the extent that the watermark
information can be extracted from the Consistency
unauthorized copy of the content. Pirated
content is often degraded in the capture of the Uniform quality is important in an
initial copy, as well as trans-coded iv for the entertainment offering. Similarly, uniform
pirate’s distribution channel. The initial capture robustness is important in a security system.
technique may range from a perfect digital copy, Both reputation for quality and content security
to resampling of analog signals, or even a are only as strong as the system’s weakest links.
camcorder directed at a rendered image. Unless Similarly, when an unauthorized content
the pirate captures a compressed digital signal, instance is discovered, recovery requires
recompression - possibly accompanied by knowledge of the technology used to embed the
cropping, frame rate changes, (de)interlace, watermark. If the content has been marked
and/or resolution changes – will be necessary to inconsistently, it becomes more difficult to
re-distribute the content. And finally, the pirate effect recovery, and, if no watermark is detected,
may attack the watermark by injecting noise, very difficult to determine which watermarking
dropping frames, filtering, or collusion v . It technology has failed. An ideal watermark
should be made difficult for the pirate to verify system deployment should, therefore, include a
that s/he has successfully removed the mechanism to ensure that all instances of a
watermark. given content title are watermarked in a
consistent manner.
Cost Effectiveness
Flexibility
Economical implementation is of paramount
importance in the consumer domain. Security Analogous to renewability, flexibility
features are of minimal apparent benefit to the describes the ease of adapting watermarking to
consumer, so it is generally not possible to the needs presented by specific content items.
recover a significant cost increment for the The content universe features broad ranges of
material and licensing cost of the watermark exposures to piracy, as well as sensitivity to
embedder in each STB. quality. Content providers are likely to prefer
watermark perceptibility-robustness tradeoffs
Renewability that differ from one content item to another.
Ideally, a watermark system should provide a
Another requirement is that of renewability. means of control, to conform to the content
Content pirates have unfailingly adapted to new provider’s preferences and policies.
media security technology. Watermarking will
not be spared. As watermarking technology is For example, Theatrical content may require
deployed, adversaries will build tools to very low watermark perceptibility with a
suppress the watermark signal. As such tools corresponding decrease in robustness.
are perfected and become widely available, the Alternatively for the purposes of identifying

2008 NCTA Technical Papers - page 71

service theft, a higher degree of watermark point of watermark imperceptibility can impact
perceptibility may be tolerated in order to robustness.
achieve an increase in robustness.
Informed Embedding
Bandwidth
Certain watermarking techniques favorably
Forensic watermarking makes only modest shift the perceptibility-robustness tradeoff.
bandwidth demands: DCI requires only 35 Watermark placement and composition can be
message bits in each 5 minute segment of a optimized to take advantage of host signal
motion picture (~.117 bit/sec).vi As mentioned (content) characteristicsvii. Numerous studies of
above, allowance for error control coding human perception have determined that
increases the raw bandwidth requirement. sensitivity to a particular sensory input varies
according to context (i.e. background). In the
FORENSIC WATERMARKING watermarking paradigm, the watermark is the
ENGINEERING CONSIDERATIONS sensory input that should be disguised, and the
background context is the content itself.
An effective watermarking system for
forensic watermarking (or any watermarking A particular watermark signal will, thusly, be
application for that matter) must perform three more or less likely to be perceived over various
basic functions. First, it must decide where in backgrounds. The effectiveness of a given
the content to place the watermarks. Secondly, background in disguising a feature is called its
it must generate the watermark signal used to “masking” property. Masking is a function of
modify content features such that the signal can the characteristics of both the background host
be detected and recovered from a copy of the signal and the disguised feature. It is thus
content. Thirdly, it must convey information in possible to reduce watermark perceptibility by
the watermark signal. The choice of a method choosing watermark signal characteristics and
to perform these functions greatly impacts the placements that leverage the masking properties
performance and cost of the system. of the host signal.

Meeting the application requirements Exploiting the host signal masking properties
discussed in the previous section, some in direct accommodates more watermark signal energy at
opposition to one another, is a non-trivial a given degree of perceptibility, thereby
undertaking. It is illuminating to examine the improving the perceptibility-robustness tradeoff.
major issues individually: Robustness can also be enhanced by choosing
watermark characteristics and placements that
Perceptibility vs. Robustness optimize recoverability. Watermark robustness
depends on the watermark and background
Both perceptibility and robustness are image characteristics, as well as the recovery
directly related to watermark signal strength. technique being used. Recoverability analysis
As the signal amplitude increases, other factors evaluates interference between the background
held constant, the watermarks become both image and the watermark signal. The technique
more robust and more perceptible. A desired is analogous to “dirty paper coding” where the
level of robustness can thus be achieved by signal is positioned to sidestep interference.
increasing the signal strength, at cost of quality. Optimal watermark composition is often a
Conversely, decreasing signal strength to the tradeoff between perceptibility and
recoverability, as an image area with a high

2008 NCTA Technical Papers - page 72

level of masking energy may also interfere with the receiver. An optimal security architecture
the watermark. for forensic watermarking in the receiver applies
the watermark immediately following content
Watermark embedding that conforms to the decryption. Both processes should occur within
content background characteristics is called the device’s physical security envelope, so that
“informed embedding”viii. Properly employed, both encoded and baseband plain-text
informed embedding significantly and favorably (deciphered) content is protected from
shifts the perceptibility-robustness tradeoff. eavesdropping prior to forensic watermarking.
This advantage comes at significant A serious complication arises, however, due to
computational cost, however. Analysis of the the requirement of conventional watermarking
masking and recoverability properties of motion techniques for access to
video signals requires complex algorithms to be the uncompressed (baseband) digital video
applied to several successive frames. The task signal.
is particularly challenging at the data rates
required to support high definition content in The baseband digital video is required for
real time. informed embedding analysis, and typically for
the composition and embedding of the
Sequencing processors or programmable gate watermark signal. Consequently, a secure
arrays capable of this task can add significant architecture in the receiving device requires that
per unit costs. ASICs are an option, but only at the physical security envelope enclose both the
high volumes, and are difficult to renew. decode and watermark processes, in addition to
the DRM and decryption blocks. Both the
Receiver Watermarking Security decoding and masking analysis require complex
logic, and thus force a potentially costly
For broadcast or multi-cast distribution expansion of the physical security envelope.
models, as discussed above, forensic
watermarks must be applied at

Figure 4 – Forensic watermarking post-decode

2008 NCTA Technical Papers - page 73

 Figure 4 diagrams a receiver implementation process requires access to baseband content, the
in which the forensic watermark is applied to receiver must either defer watermarking until
the baseband video, subsequent to decode. As the content is decoded and rendered; or decode,
shown in the red shaded area, a substantial watermark, and re-encode prior to storage. The
processing block requires physical security to former choice weakens security by distancing
protect the unmarked data. watermarking from the initial decryption, in
both time and space. Figure 5 illustrates this
Another security issue arises when the design, including a very large requirement for
receiver imports and stores content, as opposed the physical security envelope.
to rendering in real-time. If the watermark

Figure 5 – Late Watermarking Model

Figure 6 – Watermarking Prior to Storage

2008 NCTA Technical Papers - page 74

 The latter choice adds a video encoder to the REPLACEMENT WATERMARKING IN THE
receiver, at significant component and potential COMPRESSED DOMAIN
licensing cost, as well as exposure to content
degradation due to multiple encodings. This So far, we have discussed forensic
architecture is illustrated in Figure 6. The watermarking assuming a monolithic
security envelope appears to have been reduced, implementation – one in which the entire
but additional decode and encode blocks have watermarking process takes place in a single
appeared as components of the forensic device or component, apart from any other
watermarking process. device or process. In such architectures, the
watermarking process requires access to (at least
Integration Issues partially) decoded content. As discussed above,
the more sophisticated informed embedding
A requirement for access to baseband video techniques require full access to baseband video.
introduces integration issues for existing This requirement has caused some system
equipment designs. It may be difficult to designers to question the feasibility of
arrange access to the baseband video in a format watermarking in a practical consumer device.
that is compatible with the watermarking
process. Access to several successive frames It is clear, on the other hand, that several of
for sophisticated informed embedding analysis the engineering issues could be resolved if the
is even more challenging. Additionally, frame watermark embedder were capable of operating
latency in the watermarking process could directly on the encoded content. The much
necessitate adjustments to audio synchronization. lower data rate of encoded content translates to
Large scale integration in media processors can smaller frame buffers and a lesser processing
raise formidable barriers to watermark resource requirement. Encoded content is
implementation by limiting accessibility to the available immediately following decryption,
video frame buffers. where watermarking could be more securely
integrated with the DRM. In use cases where
Renewability and Flexibility the content is stored or downloaded subsequent
to watermarking, the costly decode-encode steps
As mentioned above, both ease of renewal could be avoided. Thus there is ample
and flexibility are desirable features in a motivation to develop a technique for
watermarking system. It is important for the watermark embedding in the compressed
operator to be as nimble as the pirates, so the domain.
overhead of a change to the watermarking
technique must not constrain his ability to Encoded content is, however, highly
respond to new challenges. complex to modify directly. MPEG, the most
common video codec family in the consumer
Consistency domain, contains numerous interdependencies
such as inter and intra frame references,
The need for consistency in forensic differential coding, and entropy coding. Such
watermarking was discussed above. dependencies make it impossible to interpret a
Consistency is most difficult to achieve when single frame, much less part of a frame, or to
complex algorithms are deployed, particularly in make a localized modification. The only
renewal scenarios. obvious approach appears to be the cumbersome
decode-watermark-reencode paradigm.

2008 NCTA Technical Papers - page 75

 generates compressed, watermarked video
An enhancement to the watermark system fragments, such that each fragment can be
architecture can, however, circumvent these inserted into the encoded content stream, at a
problems and actually permit watermark specific location, in place of pre-existing video
embedding to take place in the compressed data. Now, package these watermarked
domain. Assume that an upstream watermark fragments with the content, and a downstream
processing step operates on a single content watermark embedder need only choose
master instance. This process performs the watermarks from this watermark “metadata” to
analysis required for informed embedding, and substitute for parts of the encoded content.

Figure 7 – Replacement Watermarking in the Set Top Box

2008 NCTA Technical Papers - page 76

 This architecture, diagrammed in Figure 7, is The replacement architecture greatly
termed replacement watermarking. The facilitates renewal. Since all of the watermark
embedding process has become a simple switch, placement and composition decisions are made
capable of selectively replacing segments of the in the upstream watermark processor, these
content stream with the watermarked video attributes of the watermark system can be
fragments from the metadata. A substantial altered without affecting the operation of the
efficiency gain is realized by performing downstream watermark embedders. Such
complex video analysis and watermark changes are reflected in the watermark metadata
composition only once, where sufficient created by the upstream watermark processor,
processing resources can be easily applied, and and become effective immediately when the
preserving the results for reuse. Most content is processed for replacement
importantly, decode and encode processing in watermarking. Significantly, there is no need to
the watermarking block is completely obviated. update software in widely deployed watermark
Fielded implementations of the replacement embedders.
embedder require little or no additional
processing power than what is commonly The same techniques used to achieve renewal
available in existing STB designs. The can be used to increase system flexibility.
perceptibility-robustness-cost dilemma is Watermarking placement and composition are
thereby alleviated, and the possibility exists that controllable upstream, at the point of watermark
deployed STBs can be upgraded with the light metadata creation.
weight embedder.
The replacement architecture ensures
The STB receiver can utilize replacement consistency across the fielded devices,
watermarking to watermark encoded content alleviating concerns about heterogeneous
immediately following decryption. The very fielded watermarking technologies and versions
light-weight embedder can be easily integrated applying watermarks of differing perceptibility
with the DRM and decryption processes, and and robustness, and requiring diverse recovery
reside within the same physical security techniques. Effectively, the control of
perimeter. Whether the resulting content is watermarking is centralized, and less subject to
decoded and rendered immediately, or is stored variations in fielded devices.
for later viewing, it will have been individuated
by the forensic watermark. The marked content REPLACEMENT WATERMARKING
can subsequently be moved throughout a home ENGINEERING CONSIDERATIONS
network, with no requirement for a watermark
embedder in each playback device. Several factors must be considered in the
implementation of replacement watermarking.
Since the embedder is essentially a simple An obvious issue is how to incorporate the
switch, it operates on streams of content data. watermark metadata into the content package,
Frame buffers are not required. Very modest such that the metadata is available to the
memory demands facilitate integration within watermark embedder. There are several
the security perimeter inside a consumer requirements affecting this mechanism.
electronics device. Integration of the
replacement embedder into a STB is Foremost, the bandwidth must be available to
straightforward, with reduced impact on time- deliver the metadata to the embedder, properly
to-market. synchronized with the content. Prior to each
frame being processed, any of the watermarked

2008 NCTA Technical Papers - page 77

fragments in the metadata affecting that frame
must be available to the embedder. A second approach is to utilize a side
Watermarked metadata volume in existing channel. When a side channel is employed, it is
implementations is minimal, but is subject to a necessary to secure the side channel to prevent
number of factors, principally the density of the tampering with the data that might disrupt the
marks (e.g. marks per second) and the size of watermarking process.
the marks.
On the upstream process side, the metadata
To secure the watermark embedding process, must be created in such a way that a valid
particularly when it takes place in the encoded content stream results when the
potentially hostile environment of the watermarked fragments are embedded. The
consumer’s receiver, it is necessary to ensure techniques for doing this are largely dependent
that the watermark metadata is bound to the on the codec in use.
content such that it cannot be identified and
removed, prior to decryption. If an adversary SUMMARY
were able to isolate the metadata stream, it
would be a simple matter to delete or corrupt the Forensic watermarking involves mass
metadata and thus suppress forensic production of individuated content instances.
watermarking. The process is very challenging to securely
implement in a large distribution system, using
Two techniques for conveying the metadata autonomous watermark embedders. By
stream have been explored for commercial centralizing computationally intensive tasks,
implementation. One is to simply utilize the distributed watermark embedding can be
codec “user data” features to carry the metadata accomplished through simple operations in the
within the compressed frame structure to which compressed domain. The perceptibility and
it refers. This approach has the advantage of robustness advantages of informed embedding
transparency in distribution, since the metadata can be realized with minimal cost impact at
simply appears to be part of the encoded content, scale, along with improved security,
secured by the same encryption wrapper. renewability, consistency, and flexibility.

i
Fred Dawson: Studios Eye 1st-Run Service As IPTV Security
Advances ScreenPlays January 2008
ii
Mark Kirstein: MultiMedia Intelligence Identifies Digital
Watermarking & Fingerprinting As Key New Opportunity,
http://multimediaintelligence.com/index.php?option=com_conte
nt&task=view&id=49&Itemid=1
iii
Steganography is defined as the technology of hiding a secret
message inside of a larger cover work, such that the existence
and content of the secret message are hidden.
iv
Media pirates often reencode content at lower data rates or
resolutions to suit their preferred distribution channels.
v
Collusion attacks are those that involve combining content
from separately captured instances of the same content in order
to dilute the watermark signals.
vi
Digital Cinema Systems Specification V1.0, July 20, 2005
vii
Barni, Bartolini, and DeRosa Perceptual Data Hiding in Still
Images, Idea Group Publishing 2005
viii
Barni, Bartolini Watermarking Systems Engineering, Marcel
Dekker, Inc. 2004

2008 NCTA Technical Papers - page 78

 DYNAMIC INSERTION FOR SHORT-FORM VIDEO ON DEMAND ADVERTISING
John Chandler-Pepelnjak and Brent Roraback
Advertiser and Publisher Solutions, Microsoft

Abstract Microsoft in 2007, makes software for agencies

and advertisers to plan, manage, deploy, track,
Historically Video On Demand (VOD) report on, and optimize online marketing
advertising has required that programming campaigns. For the dynamic insertion VOD trials
content and advertising content be encoded and outlined here, we used this technology to
delivered together. In 2006, we deployed two facilitate campaigns in two cable markets.
field trials of “dynamic insertion,” the run-time
assembly of advertising and programming A brief history of VOD advertising
content.
Historically, VOD advertising has been very
This paper details the execution of these similar to linear television advertising, at least
trials from an ad-server perspective. We explore from the perspective of the advertiser. Typically
the requirements of an ad-serving solution, on ad creative is produced, sent to a network where
both the execution and reporting aspects of a it is encoded with programming and distributed
campaign. Sections 1 and 2 provide introduction to operator VOD systems. Views of the
and background material. Section 3 details the programming content run via a request from the
technical requirements for campaign set-top box (STB) and the collective viewing of
management and execution. Section 4 details the the ad asset and programming content is
data that is reported on and best practices for its recorded. At the end of the campaign advertisers
analysis. and agencies receive some data detailing number
of views and the reach (the unique count of
subscribers or STBs viewing the content).
Critically, views of ad content are not separated
INTRODUCTION from programming content views, obscuring the
critical information on whether or not the
What Atlas does advertisements were fast-forwarded or even
viewed at all. Moreover, modifying the
Video On Demand (VOD) advertising has scheduled creative mid-campaign is impractical,
been labeled an “emerging media” channel for if not impossible, due to the lead times required
far longer than one would think it takes to to re-encode and distribute the updated content.
“emerge.” The delay has been due, in part, to the
impressively complicated systems that underlie For example, movie studios often take their
the delivery of cable television. Another part has creative material from the finished version of the
been the reluctance of cable operators to innovate film. These shots are sometimes ready only one
and thereby risk disturbing service. Finally, VOD or two weeks before the film is released. With
advertising has been challenging for the agencies typical VOD lead times of six weeks, VOD
and advertisers that would be likely to use it— advertising becomes untenable for these
markets are small, deployment is tricky and advertisers. Additionally, it is often advantageous
creative requires long lead times. for these advertisers to change creative after the
opening weekend, touting reviews from critics or
In this paper we detail a solution that solves other achievements (e.g., “Number one movie of
many of these agency and advertiser issues 2008”). Given that the ad content and
through the Atlas On Demand Media Console. programming content are joined, simply
Atlas, founded in 2001 and acquired by swapping creative is impossible. There are many

2008 NCTA Technical Papers - page 128

other industries for whom the ability to change impressions to be delivered, date ranges,
creative depending on external circumstances is acceptable types of inventory (e.g., length of
valuable. spots), etc.

The final drawback of the current VOD Along with information describing the
system is perhaps the most damaging in the long- inventory purchased, the second important aspect
run. The internet’s quick ascension in the of campaign management relates to the ads
marketing mix is partially due to the ability to associated with the campaign. Information about
target. Meaning, when an ad impression is called the ads (e.g., identification codes such as ISCI or
for, the content provider or network can use AdID, the names of the assets, the asset
information about the viewer to determine the durations, etc.) is specified along with the
most effective creative messaging. In the current inventory with which they are associated.
incarnation of VOD, this is not possible as all
viewers of a piece of programming content will Campaign Management Solution
receive the same ads. The ability to make
television addressable through dynamic insertion In dynamic advertising scenarios, campaign
is a critical feature both for advertisers and for management becomes an active exercise: users
the medium itself. receive frequently updated information regarding
the status of their campaigns and have the ability
The requirements for a new solution are to make changes to the campaign while the
clear. Advertisers must have the flexibility to campaign is being delivered. In a non-dynamic
insert creative into placements on short notice. scenario, such as broadcast television, the same
This allows creative swaps, different ads targeted degree of active management and detailed
to different viewers, and evaluation of creative reporting does not exist.
performance independent of programming
performance. Atlas has created a solution for agencies and
advertisers to present instructions to ad execution
and management systems directing these systems
CAMPAIGN MANAGEMENT AND to display specific ads when a particular piece of
EXECUTION content or inventory is delivered. Atlas also
collects information about the viewing of
Background advertising content, calculating metrics based on
this viewership along with the business terms
“Campaign Management” refers to the set of and goals under which the inventory was
activities concerned with the definition and purchased.
management of advertising campaigns. Sellers of
inventory describe the format and characteristics Our solution, Atlas On Demand, interfaces
of media they have for sale. Agencies and with the ad execution and management systems
advertisers record the inventory they have over secure connections via APIs defined with
purchased from various sellers and the terms our technology partners in this space (e.g., Aaris,
under which it was purchased. These business SeaChange; Atlas is also a provider of inventory
terms include information such as the cost and ad management solutions for sellers of
method used for describing the unit of media inventory, integrated with other partner systems).
(e.g., CPM or “cost-per-thousand” impressions, The set of services and message structure for
time-based costing), the cost per unit (or “cost managing this communication leverages
basis”), target or guaranteed quantity of

2008 NCTA Technical Papers - page 129

emerging industry standards such as SCTE 130 complete package of program and ad assets is
(DVS 629). distributed appropriately.

Within Atlas, campaigns are created and ads To insure the viewing experience in dynamic
assigned to purchased inventory. Instructions are VOD is of the same quality as other on demand
then published to the ad execution and viewing, content encoding standards must be
management systems. When ad-supported rigorously followed. In our trial campaigns,
content is requested by subscribers the content is CableLabs OD encoding standards formed the
assembled based, in part, on these instructions. basis for theses specifications but extra care was
Ad and program content are “seamlessly spliced” taken to insure that bit and frame rates,
together and streamed down to the subscriber’s resolutions, and audio were identical for all
set-top box (STB). assets. Assets were also required to start and end
with silent frames of black to ease the transition
Campaign Execution between assets. The “seamless splicing” of
assets, mitigating any remaining discontinuities
For a variety reasons – disparate, closed between MPEG files, was either accomplished in
network systems; manual or semi-automated software by the VOD system or in the edge
processes; emerging standards; pre-existing device (i.e., the QAM) level, depending on the
workflows; etc. – executing dynamic VOD VOD system provider’s approach.
advertising campaigns is still a very complex
process. Tight coordination across a range of Aside from system configurations and
partners at multiple levels, from senior sales content preparation, the standard campaign
executives and content owners to network workflow generally begins with media
engineers and ad operations personnel, is negotiation: sellers of media (content networks,
required. operators) offer packages of inventory to buyers
of media (agencies, advertisers). Rates, schedule
In order for Atlas to be able to communicate and other terms are negotiated and agreed to
with the ad execution and management system through terms and conditions, insertion orders,
deployed at the operator a secure connection etc.
must be established, such as through a VPN
concentrator or other secure web service After the contract is finalized, ad campaign
connection. The operator’s endpoint, transport information is configured in the respective
and protocol are managed through configuration inventory and campaign management systems of
settings in Atlas. the buyers and sellers (information may already
exist in the seller’s system, enabling the seller to
As noted above, program content and ad forecast and book inventory). Ad assets are
content have traditionally been encoded together distributed, generally to the inventory seller but
as a single on demand asset. With dynamic VOD in some cases to the operator directly. Ad and
advertising, programs and ads are treated as program assets are encoded per the on demand
separate assets, “seamlessly” spliced together at specifications and distributed to the operator and
runtime and streamed to the viewer. In such a headend systems, usually through existing “pitch
model, content providers and distributors must – catch” mechanisms but potentially through IP-
account for “ad free” versions of the content based distribution.
suitable for dynamic VOD ad campaigns, as well
as the ad assets themselves, insuring that the Once campaigns have been configured and
ad and program content distributed, ad

2008 NCTA Technical Papers - page 130

instructions are published by the REPORTING AND DATA ANALYSIS
agency/advertiser to the inventory/ad
management system at the operator. Instructions Data description
are validated by the operator system and
acknowledgements returned to Atlas. If the As mentioned previously, one of the principal
instructions are valid they will be referenced benefits of dynamic insertion is the ability to
when ad-supported content is assembled. measure ad performance separately from
programming performance. In other words, as
Viewers request ad-supported VOD content, users fast-forward (FF), rewind (RW) or pause
initiating sessions with the VOD system. The their ad programming, we collect data on each
Atlas instructions are referenced in the assembly trick-play and can use that to measure
of the on demand content, ad and program viewership.
content is seamlessly spliced together and
streamed down to the viewers. Viewer The data that Atlas collects come in two
interactions with the content (“trick mode” different styles. The simplest conceptually is
activity, such as fast-forwards, pauses, rewinds) what we call “event-level records”. These
are recorded by the VOD system. We collect records detail every trick play event and capture
detailed information regarding viewing and the following pieces of data (or meta-data)
playback of ad assets. Data is imported in Atlas’ associated with the event:
reporting system and metrics are calculated.  Date
 Time
Atlas users view statistics related to their  Operator
campaign’s performance through online reports.  Headend
Statistics are updated several times per day. This  Masked MAC address
granular viewing data allows Atlas to calculate  Ad Asset Name
and display multiple metrics describing
 Ad Asset ID
campaign performance. Impressions (i.e., views
 Event (Setup, Play, FF, RW, Pause,
irrespective of playback speed), “Brand
Teardown)
Exposure Duration” (BXD) (i.e., viewing
duration at normal playback speed), completed  Event Speed (1 for play, positive for play
or FF, negative for RW)
plays, reach, trick mode counts, and more may
used to compare the performance of ads and/or  Programming Content Name
their associated inventory. Campaign  Programming Content ID
performance is assessed by analyzing the metrics
corresponding most closely to the advertiser’s In data of this format, one row of data
campaign goals. Our users may then apply this represents one event of ad viewing. A viewing
information by “optimizing” their campaign: session is defined as beginning with a Setup
changing the ads assigned to their campaigns event followed by a Play and ending with a
and/or the business rules governing ad rotation to Teardown. There can be any number of
maximize performance. New instructions are interstitial trick play events (FF, RW, Pause).
then published to the VOD system and enforced The most complicated data field is masked MAC
during subsequent viewing sessions. address. Typically the unique identifier of a STB
must be masked for privacy reasons. It is
important that the masking algorithm be 1-1 so
that no two MAC addresses can be mapped to
one masked MAC address so that reach and
frequency can be accurately calculated.

2008 NCTA Technical Papers - page 131

 campaign you have “Total BXD”. While simple
Occasionally the detailed data format is not at first glance, Total BXD is a powerful omnibus
available. In that case, there is an alternative data metric, combining length of asset, asset average
format (“aggregate-level”) where one row of data view time, and total number of views. Increasing
represents one (potentially partial) view of an ad. asset length has the effect of decreasing the
In order to use this data, additional fields must be average view time (jumping from a 15 second to
added. These include the following: a 30 second spot diminishes viewership) but our
 Start Date research indicates that net viewership typically
 Start Time increases with increasing asset length. We can
 End Date also look at BXD in several different ways.
 End Time Average asset BXD can be used to optimize
 View time (amount of time asset was creative—if this method is followed total
viewed in normal playback speed) viewing time will be maximized. Alternatively, if
 Fast Forward Count all creative are in rotation in a placement (say,
the first commercial break of an ad-supported
 Rewind Count
VOD program) then average BXD will
 Pause Count
determine the value of the placement and the
suitability of that program’s audience to the
The key metrics that can be derived from
creative message. All else being equal, BXD will
these reporting fields are worth pointing out
tend to choose longer commercial assets. From
explicitly. All traditional campaign measures—
the data we have seen across VOD on television
GRPs, impressions, reach, and frequency—are
and video on the web, doubling an asset’s length
available. Additionally, we can look at
rarely cuts the average percentage of the asset
viewership patterns by time of day and day of
viewed by half. Finally, BXD is a useful cross-
week. Finally, a variety of user-level reporting
platform measurement. BXD can be calculated
metrics are available and these will be detailed
for video across any screen and is a useful
later in the section.
measure of engagement across platforms. Note
that the efficacy of any video metric diminishes
Measures of Performance
if the ability of users to FF is disabled.
The VOD landscape is crowded with many
The alternative to BXD that emerged during
metrics used to evaluate performance across
our trials was “completed plays”. This metric can
many dimensions. Ads, placements, asset
be defined in multiple ways but the simplest is a
lengths, and content providers are judged by a
view of an asset with no FF activity. (An
variety of yardsticks. Though our dynamic
alternative is to look for every second of an asset
insertion trial, two metrics emerged as the most
being viewed at least once in normal play mode,
critical for evaluating these campaign attributes.
though this is more complicated.) Assets are
Because VOD is an accountable media, we take
compared with each other based on the
the time to highlight these two different
percentage of views resulting in a completed
performance metrics and detail their
play. Whereas BXD tended to reward longer
implications.
assets, completed plays unequivocally skew
results towards short assets. Completed plays are,
The first measure has already been
however, the only metric that makes sense for
mentioned: Brand Exposure Duration (BXD).
certain classes of creative assets. Commercials
This is simply the amount of time an asset is
where the call-to-action or brand message is
viewed in normal playback speed. If you sum up
delivered in the closing seconds require
all the BXD values for every impression on a

2008 NCTA Technical Papers - page 132

optimization based either on completed plays or analyze regarding pre- and post-roll advertising.
on a “weighted BXD” where certain portions of Are longer pre-rolls fast forwarded more often
the asset are worth more than others. than shorter pre-rolls? Do viewers stick around to
watch post-roll advertising?

Results from Data Analysis It goes without saying that in our data the
pre-roll ads received a higher impression count
The two trials detailed in this paper had and higher completed plays. This is nearly
different compositions. The first had two pre-roll tautological. Due to the greater length of the
creative assets for one advertiser running over post-roll, however, the post-roll commercials
two different time periods across one content resulted in longer BXD. Again, BXD is the
provider. This trial allowed us to prove the average or aggregate duration in minutes that an
concept in a simple environment. The second ad or a brand (if multiple ads) is watched. This
trial was much richer from a data analysis mirrors research we typically see with digital
perspective: two advertisers and two content video: longer assets perform better from a BXD
providers running multiple assets of varying perspective, shorter assets perform better from a
lengths in both pre-roll and post-roll positions. In completed play perspective.
the results that follow we focus on this second
trial. It is easier to follow the results when Viewership on post-rolls was surprisingly
speaking about a specific campaign and the high, including the number of completed plays.
results themselves are much deeper with these This would indicate an undervaluation of post-
data. roll ads given the common assumption of little to
no viewership. One factor contributing to the
As mentioned above, two advertisers took longer BXD was the longer durations of the
part in the second VOD trial. The first advertiser creative used in the post-roll positions compared
had several assets in rotation, all 30-second to the creative used in the pre-roll positions. On
spots, running in both pre-roll and post-roll. The average post-rolls were viewed approximately
second advertiser had pre-rolls of both 15 36% of the way through (versus 42% for pre-
seconds and 30 seconds, followed by post-rolls rolls).
of 60 seconds and 120 seconds. These assets
were all in rotation on both content providers. Unsurprisingly, our analysis revealed that the
30 second pre-rolls were fast forwarded more
When reporting on short-form, dynamically- than the 15 second pre-rolls. Since one advertiser
inserted VOD campaigns, there are a number of had only 30 second spots, it is possible there was
standard metrics that barely need mentioning in some burn-out, although this behavior (more FF
this forum. Although fundamental to campaign activity on longer spots) is not atypical.
evaluation, metrics like reach, frequency, and
impressions (divided into various time ranges There was evidence to suggest that
and publisher and placement groups) are viewership of post-roll ads can be augmented by
straightforwardly defined elsewhere. Instead we pre-roll ads from the same advertiser. In other
will focus on a series of analyses we performed words, "bookended" placements (with a single
during the trials that were unique to the dynamic advertiser in both positions) are more valuable
VOD environment. than pre- or post-roll placements alone or from
distinct advertisers.
Initially there were two pieces of
“conventional wisdom” that we wished to

2008 NCTA Technical Papers - page 133

 From a viewer perspective, post-roll and content provider? A display of the data
viewership of bookended placements might be appears in Figure 1.
perceived in much the same way as mini-long-
form advertisements. Indeed, our research This figure shows the interaction between the
indicates that, if maximizing BXD is the goal, advertiser’s commercial content and the
advertisers would do well to treat post-roll programmer’s content adjacent to which the
viewership as a “conversion” and focus on pre- commercials run. On the y-axis we see BXD
rolls that are most likely to achieve viewership of expressed as a percentage of asset length—
the post-roll. Given the viewers discretion to higher numbers indicate more of an asset was
watch or FF in this context post-roll viewership watched. As we can see, Programmer 2
starts to look like a click or performance-type performed better overall, but there is an
media on the web. Our research also indicates interaction between the advertiser’s content and
that longer pre-rolls (:30s over :15s) make the programmer’s content (indicated by the
viewers more tolerant of longer post-rolls. crossing of the lines). This is important: not all
programming is optimal for all advertisers and
Another question we asked of the data was, Is advertisers may reasonably value different pieces
there an interaction effect between advertiser of VOD inventory differently. Currently many

BXD by Advertiser and Content

60%

50%

40%
BXD (% of Asset)

Advertiser 1
30% Advertiser 2

20%

10%

0%
Programmer 1 Programmer 2

Figure 1: In this figure we see the interaction between the advertiser’s commercial content and the programmer’s
content adjacent to which the commercials run. On the y-axis we see Brand Exposure Duration (BXD) expressed as a
percentage of asset length—higher numbers indicate more of an asset was watched. As we can see, Programmer 2
performed better overall, but there is an interaction between the advertiser’s content and the programmer’s content
(indicated by the crossing of the lines).

2008 NCTA Technical Papers - page 134

marketers believe that pre-roll video viewing is CONCLUSION
dependent entirely upon the content following
the spot. This research contradicts that view. There are two critical components in the
deployment of a short-form VOD campaign. The
There was, however, a tradeoff to consider first is the ability to package ad content
between reach and impressions on the one hand separately from programming content. This is
and increased duration on the other. fundamental to many needs of advertisers
including creative management, creative
As for asset lengths, longer advertising assets swapping, decreasing creative lead times,
generally resulted in longer viewing durations accurate reporting. The second is detailed
and more total minutes viewed, shorter measurement of individual ad viewing duration.
commercials enjoyed more completed plays and The first requirement is a near-prerequisite for
were watched in their entirely a higher the second, but it is only with this measurement
percentage of time. that the power of the addressable television
medium is achieved.
Although you might assume that the average
BXD percentage might decrease as the length of
the commercial increases, it was found that the
percentage of :120 spot viewership was higher
than that of the :60s.

Post-rolls were watched approximately the

same percentage of the time regardless of asset
length (roughly 35%). This confirms previous
research conducted by Atlas on long-form VOD
advertising. One interesting effect of post-roll
viewership was noted. For one advertiser,
viewership of :120s in the post-roll position
increased by 20% if the pre-roll was a :30 instead
of a :15.

The :30 commercials of one participating

advertiser performed better in the pre-roll than in
the post-roll in terms of both BXD average
minutes and as a percent. Some of this effect
could be due to the repetition of the same ads
given that if the same spot appeared in the pre-
roll position as in the post-roll position a viewer
would be predisposed to fast forwarding through
the second appearance.

2008 NCTA Technical Papers - page 135

 ECONOMIC IMPACTS OF VIDEO FOR THE BROADBAND WIRELESS OPERATOR
Paul Steinberg, Mark Shaughnessy, Lloyd Johnson, Philip Fleming
Motorola, Inc.

Abstract anywhere. We conclude with a brief look at

some anticipated future trends and applications
The ability to offer compelling wireless video of mobile video.
services over a wide area has been a long
awaited goal for consumers and operators alike.
Finally, 4G wireless technologies such as INTRODUCTION
WiMAX and Long Term Evolution (LTE) offer
adequate performance characteristics to support Our world is increasingly mobile, and this is
IP based video services to a large number of driving the demand for easier access to content
consumers simultaneously with a quality that and services from any location at any time.
most will find attractive. However, there are a Existing wireline networks have made universal
number of economic and technical factors that access to the Internet possible, but only for those
the operator will need to consider when it comes that have a physical cable or short-range WiFi
to actually implementing video services in a connection to the network.
wireless network.
The underlying demand is for untethered
In this paper we begin by summarizing the wide-area mobility, i.e. accessing content and
history of broadband wireless technologies and services without wires. To partially satisfy this
their shortcomings relative to the desired demand, a vast market has grown in support of a
consumer experience and operator economics. “cache and carry” model. Since digital
Next we review the key attributes and technology has enabled the efficient storage of
performance implications of 4G broadband vast amounts of content in small portable
wireless technologies and relate those to the devices, content can now be downloaded from a
capabilities necessary for a service provider to multitude of public and personal sources and
deliver a viable mobile video service offering. stored (or “cached”) in a portable device that is
carried wherever a person goes.
The paper then describes an overall network
architecture and essential elements to deliver an However, “cache and carry” requires the
end-to-end video solution. The mobile wireless step of pre-loading, which doesn’t solve the
environment enables the operator to tailor the problem of access from any location whenever
offered content, services, and advertising, the need arises. Nor does it support streaming
dependent on user location and context. real-time content or social interaction services
Operators that offer both wireless and wireline like real-time communications. Fortunately,
access networks have the opportunity to wireless mobile data networks, coupled with a
integrate them under a common IPTV video converged content and delivery network, are a
headend which provides not only consistent solution to these mobility aspects. However,
consumer experience and content access, but only recently have wireless technologies become
also the ability to provide mobility between these fast enough to support delivery of bandwidth
networks. Video streams in progress can be heavy content, in particular the type of content
moved among wired and wireless devices, and that is the focus of this paper: video.
content can be made available to any device,

2008 NCTA Technical Papers - page 275

 EARLY WIRELESS DATA versions of the services never materialized.
TECHNOLOGIES While there were undoubtedly multiple reasons
for these commercial failures, the three most
Since they were first deployed, there have significant were high cost of data delivery, low
been attempts to use wide-area wireless mobile bit rate performance, and limited mobile device
data networks to provide some level of video functionality.
service. Early examples of wireless mobile data
networks were in the military and public safety The device limitations in particular, such as
realms [1]. Early public mobile data networks, very little memory, slow processors, and small
such as Cellular Digital Packet Data (CDPD), low resolution displays, were the pivotal factors
Mobitex, and DataTac offered data rates of 19.2 that drove the data services offered for 2/2.5G
Kbps or less, and suffered from relatively low cellular phones to be primarily limited to
subscriber interest. The first public wireless data browsing of cellular-aware web sites and
networks to gain significant subscriber uptake specialized email programs; and even these were
were extensions to the 2G cellular telephony painfully slow. There have been some 2/2.5G
networks, primarily GSM (3GPP) and CDMA data commercial successes in controlled system
IS95A (3GPP2). Early technology environments. A notable example was the i-
demonstrations of video services on these Mode service offered by NTT DoCoMo in Japan
networks centered on video telephony and the [5]. Success was achieved through creation of
exchange of video clips via the multimedia unique applications and experiences that were
message service (MMS). optimized for low bit rates, by a large
development community.
2G . . . 3G 4G

• 1992 • 1994 • 1998 • 2000 • 2002 • 2004 • 2006 • 2007 • 2009 • 2010
A wider range of services could be offered
3GPP GSM GPRS UMTS HSDPA HSUPA LTE using PCs with 2/2.5G access cards. In theory,
3GPP2
IS95A 1X DO DO Rev A UMB
subscribers could even attempt video downloads
16e
using these cards, assuming they had enough
IEEE 802.11 b/g 802.16d WiMAX 802.16m
time and money. The predominant business
model employed by operators was to charge for
usage (bytes transferred). Download of even a
small 5MB highly compressed video file at low
Mobile Data Mobile Data Mobile Video Mobile Video
Emerges Enabled Emerges Enabled resolution with a typical data rate 50 Kbps or less
Figure 1. Timeline for the Introduction of would have taken over 13 minutes; and given the
Mobile Data Functionality tariffs of the time (often as much as $1 per 100
KB for GPRS), cost as much as $50.
Figure 1 summarizes the evolution of the
commercial wireless broadband standards [2]. The introduction of 3G and 3G+ technologies
With the advent of 2.5G technologies – such as – UMTS, HSDPA and EV-DO – during the
GSM GPRS and CDMA 1xRTT, basic data middle of the current decade provided a marked
services such as email and web browsing were increase in wireless data capabilities, with typical
offered commercially and were somewhat data rates on the order of 0.5 to 1 Mbps
successful. There was much discussion of achievable for HSDPA and EV-DO. These new
extending the cellular data offerings to include technologies also provided a dramatic decrease
some form of video service, with attention in the cost of delivering data services, as is
focused on video telephony and multimedia shown in Table 1.
messaging. Technical feasibility was
demonstrated for these services but commercial

2008 NCTA Technical Papers - page 276

 Peak Realistic Cost to Deliver Cost to Deliver
Figure 2 compares fixed/wireline and wireless
Speed Speed MB ($) 5GB/Month ($) broadband technologies in terms of throughput
HSDPA 14 Mbps 900 Kbps 0.021 $105
capability over time [3]. In general, we see that
1xEV-DOr0 2.5 Mbps 300-500 Kbps 0.022 $110

UMTS 2.0 Mbps 150-200 Kbps 0.069 $345

wireline bandwidths fairly consistently exceed
1xRTT 625 Kbps 60 Kbps 0.059 $295
those of the contemporary wireless technologies
EDGE 384 Kbps 30 Kbps 0.138 $690
by approximately two orders of magnitude. The
GPRS 120 Kbps 12 Kbps 0.415 $2075 dotted line approximates our view of
Source: CSFB, HSDPA Networks Group
expectations for “good enough” mobile wireless
Table 1. Data Delivery Costs for 2G/3G performance. We see that it’s not until 4G that
Wireless Technologies wireless throughput reaches a level that is
satisfactory.
“Unlimited” data services (where unlimited is 10000

defined as a few GB per month) were now being

offered by several operators at less than $100 per 1000

month. During this same time period, dramatic Wireline GPON*

100 Mbps*

advances were made in the quality and cost of 100 VDSL2

25-50 Mbps

Bandwidth (Mbps)
the cameras and displays that can be incorporated ADSL Cable LTE*
10-20 Mbps 50+ Mbps
into a mobile phone. With the combination of 10
ADSL, Cable
4-6 Mbps
WiMAX, HSPA+*
10+ Mbps
1-2 Mbps
the new functionality and reduced cost, non-text HSPA*
3-7 Mbps

services finally began seeing substantial growth. 1 HSDPA*

1.8 Mbps
Wireless
Multimedia messaging incorporating both still 0.1
UMTS*

and moving images is now in wide use; and, as is

0.384 Mbps
GPRS* ‘Good Enough’ Performance for
0.075 Mbps Mobile / Wireless Expectations * Bandwidth of shared media
demonstrated almost daily on the evening news, 0.01
are per-subscriber

the uploading of embarrassing film clips from 2000 2005

Year
2010 2015

mobile phones to YouTube has become a world-

wide hobby. Figure 2. Technology Bandwidth Comparison

Mobile Video Needs

BROADBAND WIRELESS
TECHNOLOGIES BECOMING AVAILABLE Table 2 illustrates typical bandwidth
NOW demands for video services using the current
generation of mobile terminals and video
4G wireless broadband in the form of displays. It is reasonable to assume that these
WiMAX and 3GPP’s Long Term Evolution values are the minimum acceptable; going
(LTE)1 (see [7], [8], [10]) provide a further forward, higher resolutions and faster frame rates
significant step forward in consumer experience will become the norm since consumers will
and network capacity, as well as network price expect the better picture quality that they
performance. experience at home to be duplicated in the
mobile environment.
Before diving into the specifics of 4G
wireless performance for video, it’s useful to
compare to wireline performance since that’s the
more traditional reference for carrying video.

1
3GPP2 is also defining a 4G air-interface standard called
Ultra-Mobile Broadband (UMB), which is technically very
similar to LTE.

2008 NCTA Technical Papers - page 277

 Resolution Picture size H.264 (MPEG-4 part 10) HSPA WiMAX LTE
Parameter 5+5 MHz 10+10 MHz
Name (& frame rate) Bit Rate FDD
10 MHz TDD
FDD

QCIF 176 x 144 64 to 80 kbps DL Peak Rate (Mbps) 14 32 60

(10 to 15 fps) DL Peak SE
(bps/Hz/Sector) 2.8 6.3 6
CIF 352 x 288 192 to 240 kbps DL Sector
(7.5 fps) Throughput (Mbps) 3.3 7.9 16.7
DL 5%ile User
QVGA 320 x 240 192 to 240 kbps Throughput (Kbps) 120 210 450
(10 fps) DL SE
(bps/Hz/Sector) 0.66 1.30 1.67
Table 2. Example Data Rates for Mobile UL Peak Rate (Mbps) 5.8 5.0 20.0
Video
UL Peak SE
(bps/Hz/Sector) 1.15 1.0 2.0
4G Air Interface Advantages UL Sector
1.5 1.4 7.6
Throughput (Mbps)
UL 5% User
43 52 192
Table 2 shows that QVGA resolution video Throughput (Kbps)

(currently used by YouTube) at 10 frames per UL SE

(bps/Hz/Sector) 0.31 0.37 0.76
second can be supported with a data throughput
Table 3. Wireless Broadband Performance
of 240 Kbps (max). Using the performance
Comparison
projections in Table 3 and assuming a 10 MHz
carrier deployment, we estimate that WiMAX
The estimates in Table 3 are based on
can deliver a sustained downlink streaming video
technology simulations performed by Motorola
session with a throughput of 240 kbps to about
and other major suppliers of wireless broadband
60 randomly scattered video users
2 access points and network equipment [11], [12].
simultaneously in an area of roughly 3 km . This
An important metric used in these comparisons is
results in the capability of supporting 20 video
Spectral Efficiency (SE). SE is calculated by
sessions per square kilometer with a good (i.e.
taking a ratio of the throughput as measured in
QVGA) user experience. LTE, using the same
bits per second (bps) and the amount of radio
spectrum allocation and cell size, can support
spectrum allocated (Hz). So the units are
slightly more than 1.5 times as many video
typically given in bps/Hz (or equivalently,
sessions. Furthermore, in the near future, LTE
Mbps/MHz). Higher spectral efficiency translates
and WiMAX, may exploit increased bandwidth
to lower cost per subscriber for the operator. 4G
allocations up to 40 MHz. Support of 100 or
wireless technologies are able to achieve higher
more simultaneous streaming video users per
spectral efficiency through the use of advanced
square kilometer will then be within reach.
transmitter and receiver designs, multi-antenna
arrays, adaptive coding and modulation
The uplink is typically the weaker link,
techniques, and smart packet scheduling
supporting 20% to 50% of the data throughput of
methods, all of which take advantage of the
the downlink. For uplink streaming applications
changes in the radio environment during a data
such as See-What-I-See that require about 120
session. While earlier technologies possessed
kbps we can expect today’s WiMAX to support
some of these methods they were not able to
about 10 such sessions per square km and LTE
utilize them all in concert over a wider spectrum
and future WiMAX to support up to 50.
allocation to achieve the high data rates of
WiMAX and LTE. In addition, the improvement
in uplink spectral efficiency possible with LTE is
primarily enabled by the use of an advanced
coherent detection scheme [13].

2008 NCTA Technical Papers - page 278

 interworking between the wireless network and
As the relative total cost of ownership curves the operator’s IP backbone.
presented Figure 3 show, LTE and WiMAX,
which are based on OFDMA (Orthogonal The architectures of the 4G WiMAX and
Frequency Division Multiple Access) radio LTE networks are contrasted with that of the 3G
technology, provide a substantial improvement in UMTS network in Figure 4.
network price performance (Total Cost of
Ownership, or TCO), especially as the amount of = control element
= bearer element Internet IMS
data consumption increases per device. Based on Legacy
the above QVGA estimation, a heavy mobile Circuit
Internet IMS Core UMTS
video user (say one who averages an hour of GGSN
viewing per day) would consume on the order of WiMAX LTE
.
PDN-GW SGSN
7 GBytes per month of data service. CN (HA)

Examination of Figure 3 indicates that it would ASN-GW (EP) * SGW

. RNC

cost 2 to 4 times as much to operate an HSDPA

Packet Circuit Packet
or EVDO network when providing this level of .
Bearer
.
Bearer Bearer
Path Path Path
per user capacity, as it would to operate an Access Points Access Points
equivalent LTE or WiMAX network. *Associated control element (not shown):
WiMAX = ASN-GW DP LTE = MME

Figure 4. Flatter, Data-Only Architecture for

14
12 DoRA LTE and WiMAX
Relative TCO

10 HSxPA
LTE
8
6
WiMAX The WiMAX and LTE network architectures
4 were specifically created to support packet data
2
0
services and are optimized for those services.
0.5 1 2 5 10 20 Voice services for WiMAX and LTE are based
GB/Month/Subscriber on VoIP technology and are treated just like
Figure 3. Relative Total Cost of Ownership other data applications. Both WiMAX and LTE
for Broadband Data Delivery2 make extensive use of Internet concepts and
protocols. This allows them to limit the amount
4G Network Technologies & Topologies of domain-specific equipment that must be used
and leverage the volume production of
Prior to 4G, macro-area wireless networks components designed for Internet use. In
were first and foremost cellular voice systems contrast, the 3G UMTS network has both a data
optimized for carrying narrow circuit voice style architecture and a legacy circuit architecture.
traffic. These systems had packet data facilities Since legacy circuit and packet data are
“glued” on almost as an afterthought. WiMAX dramatically different concepts, this combination
and LTE are pure, broadband data systems that architecture gives rise to complex, domain-
have no specific circuit voice provisions (other specific components that limit network
than as a constant bit rate QoS class) – they are performance and drive up infrastructure costs.
IP packet based access technologies. This results
in simpler and more decentralized network The WiMAX and LTE architectures are
architectures relative to earlier networks and flatter (i.e. more distributed, with fewer layers of
eliminates the need for complex protocol system elements) than the UMTS data
architecture. In WiMAX and LTE there are two
2
For this illustration: a population density of 1000/km2 is levels of components in the bearer path between
assumed with a 15% subscriber penetration rate. The the BTS / AP and the application core network,
spectrum usage is normalized across the technologies.

2008 NCTA Technical Papers - page 279

while in UMTS there are three. In networks (MBMS) for LTE. In principle, MBMS provides
designed to support legacy voice services this facilities in the network that define sets of base
additional level was useful for a variety of stations over which a given service or media
reasons including aggregation of low speed stream should be broadcast. There are provisions
circuit traffic, scheduling of timeslots on circuit to control when a particular service is broadcast
transport facilities, and setup/teardown of including facilities to dynamically enable
transport bearer circuits. In a high speed data broadcast based upon user demand in a particular
network however, these circuit-related functions location. IP Multicast will be the network level
are not needed, and this additional layer distribution method to direct content from the
accomplishes nothing other than driving up cost content source via a gateway function (MBMS-
and slowing performance. The flatter GW) to specific base stations as required.
architecture of LTE and WiMAX also aids in
simplifying network operations, as there are Mobile
Terminal eNodeB
fewer different types of components to be
SFN
managed. Sync

One final characteristic of the high-level SFN

MBMS
MBMS
Service
Sync
architectures of WiMAX and LTE that Gateway Center

contributes to their cost advantage over 3G SFN SFN

Content

networks is better separation of wireless control Sync Sync

functions from data plane functions. By

definition, wireless control functions are unique SFN
Sync

to the wireless domain. Separating the unique

wireless control functions from the data plane SFN
Sync
functions makes it possible to leverage some
network elements that can be produced in higher Figure 5. LTE Multicast Broadcast
volumes with lower costs. This separation also Multimedia Service (MBMS) Operation
simplifies network sizing and expansion. For
example, wireless control functions are MBSFN (Multicast Broadcast Single
influenced by the frequency with which Frequency Network) further optimizes broadcast
subscribers enter and leave the network and the operations of the network by synchronizing the
speed with which they move around in the delivery of specific content across multiple base
network. And, the data plane processing load is stations so that the media can be transmitted
determined by the number and size of data simultaneously by each station, as shown in
packets that are sent between terminal devices Figure 5. This provides a much improved RF
and applications servers. The ability to scale environment and corresponding signal/noise ratio
wireless control network elements and data plane by allowing the mobile terminal to combine
elements separately in LTE and WiMAX enables reinforcing signals from multiple adjacent base
operators to size their networks to their exact stations.
needs and to focus capacity increases on those
functions that are under stress. Mobile Device Advances

WiMAX and LTE will also include specific Device technologies will continue to improve
provisions in support of video broadcast. For with the same technology advances that drive the
example, 3GPP is in the process of defining the desktop computing environment. Moore’s law
Multicast Broadcast Multimedia Service coupled with substantial improvements in power
management and memory density advanced the

2008 NCTA Technical Papers - page 280

functionality available on the handheld A FULL SERVICE SOLUTION IS
computing platforms. Just as the laptop NEEDED
computer became a staple among business
professionals and college students, now the A high-performance wireless access network
handheld mobile device is as well. is important, and generates revenue on its own.
But access networks are really just an enabler for
The potential for new and expanded video providing content and services to consumers.
services is also being impacted dramatically by The Internet model has taught consumers to
technological advances in the mobile device expect universal access to hundreds, if not
domain. Advances in materials and Micro thousands of valuable applications and content
Electro-Mechanical Systems (MEMS) sources. For a network operator to maximize
technology are spawning a new generation of revenue and profitability, it‘s desirable to
miniature advanced antennas systems that enable participate in every way possible in the business
devices to fully exploit the capabilities of LTE of delivery of content and applications. Two key
and WiMAX. MEMS is also a key to new ways to do this include supporting an open
designs that leverage the DLP™ technology environment in client devices, and establishing
developed for HDTV to provide high-quality, unique value added applications.
reflective-light displays that are well suited for
mobile applications. These new displays Open Client Environment
promise to provide much improved visibility is
high ambient light environments and extremely The open client environment is the first tier
low power consumption, ameliorating two of the in the application value chain. This allows 3rd
most serious challenges to supporting video party application providers to create and deploy
applications in a truly mobile environment. 12 applications quickly, constrained only by
GB flash memory cards will be available for competitive capitalism. Even if these
mobile devices this year. One card has enough applications are hosted outside of the operator’s
storage to hold 1500 songs, 3600 photos and network, the open client environment benefits the
over 24 hours of video at the same time3. One network operator by increasing ‘stickiness’ since
manufacturer has even announced a 120 fpm consumers know (and expect) that they can
video capability in one of its mobiles to support enhance what their device can do, freely and at
slow motion video functionality comparable to their own discretion – said another way, they
today’s in-home DVD players. By early 2009, don’t need to move to a competitor’s network to
wireless devices are expected to have built-in gain access to new services.
projectors. These devices have the potential to
eliminate the most frequent cited inhibitor to Operator Provided Applications
video on mobile devices – their small screen.
Finally, advances in lens, flash lighting, memory The 2nd tier up the application value chain is
cards, and other technologies are making certain when the network operator provides some
that the explosion in bandwidth utilization for applications themselves. Again, these are
video will not be a one way street. supported within the same open client
environment, and may be developed by 3rd
parties or by the operator, but they are hosted
within the operator’s own network. Now the
3 operator gains revenue not only from the use of
Approximation based on 4 minute songs using 128 kbps
MP3, pictures taken with 2Mpixel camera and MPEG-4 the access networks, but also by charging for the
video at 384 kbps. Pictures and video assume typical services themselves. Of course the goal for an
compression and resolution.

2008 NCTA Technical Papers - page 281

operator is to identify as many compelling Smart-
Activity iPhone Market
applications as possible and bring them into an phone*

integrated environment in their network, rather Any news or info via

84.80% 58.20% 13.10%
browser
than passively watching as others gain revenue
Accessed web search 58.60% 37.00% 6.10%
for those applications.
Watched mobile TV
30.90% 14.20% 4.60%
and/or video
Over time, consumers have come expect to Watched on-demand
access applications and content from the Internet video or TV 20.90% 7.00% 1.40%
programming
as well as from the operator. However, there are
Accessed Social
still many ways that operator hosted applications Networking Site or 49.70% 19.40% 4.20%
Blog
can differentiate and more easily provide
Listened to music on
capabilities that Internet hosted applications mobile phone
74.10% 27.90% 6.70%

can’t. Here are some examples: Table 4. Mobile Content Consumption:

iPhone, Smartphone and Total Market,
 Integrating multiple applications by linking January 2008
data between them
 Sharing user preferences
 Sharing user identity information Video content and applications are most cost
 Improving performance through application effectively delivered by a comprehensive end-to-
linkage to QoS enforcement in the access end network architecture. This architecture
network should leverage the all-IP nature of video and
 Tailoring content and functionality based on voice services to provide a converged set of
client geographic location (note that Internet functionality which can support service delivery
hosted applications can do this too – if the across both wireline and wireless networks.
client knows its location.)

Video (streaming, VOD, and linear) will be a

major component of the future service offerings
for the mobile and converged service provider.
Table 4 shows the results of a survey done by
M:Metrics showing the percent of users of a
given device type who performed a number of
popular data activities in the month of January
2008 [6]. There is a clear trend towards
significantly increased video usage by consumers
who have an easier to use device (and typically a
flat-rate data service plan).

2008 NCTA Technical Papers - page 282

 CONVERGED VIDEO DELIVERY
ARCHITECTURE

Figure 6. Functional Blocks of the Converged Video Architecture

before the content is available. This supporting

The primary functional blocks of the information is referred to as meta-data. Some
converged video architecture are shown in Figure examples of information contained in the meta-
6. data include program title and synopsis, length,
encoding, points where ads can be inserted, etc.
Content Processing – Includes direct video
manipulation such as content ingestion, storage, Linear (real-time) content is ingested,
and stream playout, ad insertion, logo insertion, processed, and streamed out immediately. Video
linear (real time) and off-line encoding / on Demand content processing includes delivery
transcoding, and off-line production tools. of the VOD program (asset) to the system,
ingestion of the asset meta-data into the VOD
Most video content has associated catalog, and storage of the content itself into the
informational data files that are delivered to the VOD stream pump.
system either coincident with or at some time

2008 NCTA Technical Papers - page 283

 Digital Rights Management (DRM/IPRM) Supporting applications, such as Presence,
and Conditional Access functions include linear Location/Mapping, and user context services are
and non real-time (offline) encryption, and Key / also ideally integrated here with the other
Certificate Management. application servers in an orchestration
environment. This approach allows for sharing
Middleware (not shown) – Middleware of data and linking of multiple applications
typically describes a software layer in the client together to create more feature rich offerings.
and network that enables and supports video
applications. The open client environment SIP Servers – The converged architecture
described earlier is embodied as part of the takes advantage of the capabilities defined within
middleware of the solution. Middleware a SIP environment to provide a consistent
typically supports many or all of the following mechanism for managing control of applications
functions: that are naturally session based such as Video on
- Retrieval of the meta-data information Demand or Voice over IP. Operator
described above from the Electronic Program implementations will likely rely on SIP for
Guide or store front portal server, and providing converged session control of voice,
formatting that for display. This includes a multi-media, and video streaming services. This
list of what video programs are available for convergence allows for rapid deployment of
viewing, what’s being played now, what can compelling combined services, such as
be purchased on demand, and what has been concurrent voice and video (talk to your friend
stored locally. while you both watch the same content,
- Actions related to viewing, including Picture integration of presence for dynamic
in Picture, control of video display (play, personalization of content, and so on. Many
pause, rewind, fast forward) carrier class operators are expected to use
- DVR functions such as recording video for variations of 3GPP’s IP Multimedia Subsystem
later viewing, or starting playback of content (IMS) architecture, which is SIP based [9], to
already recorded. enable this convergence.
- Emergency Alerts and display of desired
informational streams Service Delivery – Includes Content
- Content Advisory / Parental Control Management functions such as VOD asset
management, and the Electronic Program Guide /
Motorola’s preferred implementation for store front portal servers, as well as Service
middleware is to use a ‘thin-client’ strategy, such Management functions such as service bundling
that the centrally managed servers deliver and merchandising, access rule management, and
“display ready” user interfaces to the clients, mediation functions.
from which the viewer makes a selection. This
allows having the same user experience across Policy Management & Quality of Service -
multiple devices, which will be described in Includes Quality of Service (QoS) mechanisms
more detail in the next section entitled Media which coordinate the assignment of capacity to
Mobility. individual clients based on session needs. The
converged architecture proposed here uses
Applications – Includes the multitude of information within the Session Description
application servers that provide operator Protocol (SDP) structures of SIP messaging to
managed video related services to client devices determine the session needs, thus the SDP
(as well as voice services, if offered). Some contains information on flow rate and the end
examples of these will be given shortly. points which are used to establish a connection

2008 NCTA Technical Papers - page 284

from video source to client. The underlying approach limits video mobility to content that has
system is then expected to enforce the requested been previously stored, whereas the Motorola
connection during the session, or to work with Media Mobility solution applies to live broadcast
the Policy Management and Application servers (linear) content as well as personal video sharing
to adapt the video stream to match revised among consumers.
system capabilities.
The flexibility of the converged applications
Access Networks – In previous sections, this architecture means that the underlying
paper has focused on wireless broadband access mechanisms that have been developed to allow
networks such as LTE or WiMAX. However, video transfer among devices can be readily
the real power of this converged architecture is applied to other applications as well. Presence
the applicability of the same video application notifications, news feeds, home caller-id, and
and content delivery environment to devices in many others are all examples of information that
any broadband IP access network, whether could easily have their target context changed
wireline or wireless. from one device to another, as a user wishes.

Media MobilityTM Benefits to the Operator

An important class of applications that this A comprehensive network architecture which

converged architecture enables is what Motorola allows an operator to ubiquitously deliver video
calls Media MobilityTM. and multimedia content to their subscribers
offers significant business advantages, such as:
Media Mobility applications enable a
commonality of service between all of the Network efficiencies from a common video
content access and display devices used by headend and converged session control -
consumers. These devices include PCs, Operators with fixed and wireless networks can
televisions (and set-top boxes), and mobiles, especially benefit from a converged architecture.
including phones and PDAs. Drawing on the It is a single video headend delivering content
lessons that content providers have learned independent of access technology. Significant
through the growth of the web, Media Mobility savings in CAPEX and OPEX are realized by
promises to provide consumers with entirely new this approach, as opposed to deploying individual
forms of video entertainment that combine independent video delivery solutions for each.
traditional program styles with emerging social
network-driven entertainment modes across a The SIP mechanisms used to move the
wide range of consumer devices. content from one device to another also enable
the network to be aware of what is being viewed,
Benefits to the Consumer by whom, and on what device. The Motorola
solution that supports the gathering of these
The converged SIP-based solution offers kinds of information, including service brokering
consumers access to advanced applications that and service orchestration, is being designed as a
provide a personal video experience. For modular architecture that allows for easy
example, the ability to pause a video stream on integration into existing IMS environments.
one device and pick it up on another allows
consumers to seamlessly carry their video with Ubiquitous access and meeting increased
them wherever they choose to view it. As expectations - By having ‘always on’ access
discussed earlier, the simple cache and carry regardless of location, there are more

2008 NCTA Technical Papers - page 285

opportunities created for pay-per content access, Stickiness – Providing consistent access,
or advertising ‘eyeballs’ reached. Additional management of preferences, as well as look and
ARPU should be possible simply by providing feel (e.g., EPG, storefront) to the user in a mobile
consistent service across multiple environments. environment as well as the home/fixed locations
For example, consumers clearly find YouTube to increases the stickiness of the subscriber to the
be an interesting way of consuming video at their carrier.
PC and this is already becoming desirable in a
mobile environment (it is reported that in 2007,
YouTube video generated more Internet traffic in THE FUTURE – EMERGING SERVICE
the United States than all of the Internet traffic TRENDS
combined, worldwide, in the year 2000). As
consumer expectations grow, it becomes a Many services and capabilities that appear in
competitive imperative for a carrier to meet the wireline networks (i.e. the Internet) tend to make
common denominator. their way to wireless, gated by the ability of the
wireless access technologies to support them.
Taking advantage of location – Targeted 4G wireless technologies are clearly able to
advertising is possible based upon an support most activities people do with the
individual’s past behaviors as well as the content Internet. Also, in the past there tended to be a
that is currently available (e.g., when a particular slower pace of client development for wireless.
piece of media is about to be distributed by a However, given the inevitable trend toward web
MBMS broadcast, the user can be alerted). With based applications and open wireless client
feedback of current subscriber location environments discussed earlier, as well as the
information back into the video headend, it is convergence of networks, this development gap
possible to customize individual streams (VOD is also expected to disappear.
for example) based upon a particular user’s
location. This enables location based advertising A few examples of new and emerging video
with a fine degree of granularity (e.g., for related applications that are likely to be popular
individual shopping malls, restaurants, events, among wireless consumers are:
etc.).
Social TV – This is a range of enhancements
Additional public services – Enhanced to today’s one-way video viewing experience
emergency communications services are also which will provide people with the ability to
possible such as broadcast alerts and/or updates share their viewing experience, regardless of
on surrounding context (e.g., traffic, etc.) based location. It is envisioned to include things like
upon location. Minimally, this becomes a crucial integration of presence so you see on your screen
part of a user’s bundle and dependency on the not only which of your “buddies” are online, but
mobile media environment. Many of these also what they’re watching, and give an easy
information services can be subscription based. ability to switch to that content yourself. Unlike
It is also possible that public carriers may be able today’s passive experience, the ability to have a
to “wholesale” selected broadband capabilities to voice or text conversation with your buddies
public safety agencies.4 while viewing the content, or sending your rating
on whether you like it or not, is supported by the
device and 4G wireless network being inherently
4 bi-directional.
Note that the FCC’s recent attempt to formalize this
with the D-block spectrum auction failed in that no carrier
was willing to accept the business terms associated with
operating in this spectrum.

2008 NCTA Technical Papers - page 286

 See-What-I-See – This is a real-time linking will have significant impact on wireless
of people’s visual experience. One-to-one video networks: most of them require vastly more
telephony is a basic form of this. However, it uplink bandwidth (from client to network, or
can be extended to share video among a small from client to network to client) than has been
group, such as sharing what I see with my fellow needed in the past. Earlier we discussed the
construction or firefighting team. Or it might be improved uplink capacity of 4G wireless
offered as a fee based service for enhancing the technologies over previous generations. This
viewing of public events, such as broadcasting will help to ameliorate the network cost impacts
point-of-view video from a sports star to fans of widespread uptake of these new services.
who have subscribed to it. However, since the uplink is the limiting
direction for wireless, its utilization will need to
User Generated Content – Revenues for the remain a parameter that is closely watched.
total user generated content market have been
estimated to grow between 66% and 99% per
year on average over the next 5 years [4]. The CONCLUSION
success of social networking and blogging sites
reflects the desire for people to make, publish, With the advent of 4G wireless technologies,
and view their own video content. The ability to operators will have the flexibility and network
do this from a wireless device will accelerate the robustness to economically deliver a truly mobile
quantity of content produced. Video blogging video and multimedia experience. WiMAX and
will become an extension to today’s simple video LTE offer significant performance and economic
file upload, giving consumers the ability to easily advances over their 2G and 3G predecessors.
create multi-media narratives for public or These networks are based on OFDMA radio
private viewing. Revenue opportunities exist all technology and use a flat, IP-based network
along the user generated content value chain, architecture. WiMAX deployments are well
from the creation of easy to use clients for underway and their pace continues to increase.
creating the content, to the application servers LTE will follow, and bring with it even more
which store catalog, and transmit it. advances and capabilities that are applicable to
video delivery (e.g., MBMS). Mobile device
Access to Personal Content – Mobile technology has advanced commensurately to
wireless enables access to personal content provide a reasonable platform for mobile
libraries from anywhere. A potential future multimedia.
revenue opportunity for the operator is to provide
an easy to use hosting service for storing and To best exploit the opportunities for revenue
managing personal content (perhaps for a flat fee that these Radio Access Network technologies
or ad based), as well as integration of a network enable, a comprehensive end-to-end service and
based Personal Video Recorder (PVR) content delivery architecture is essential, and is
capability. of particular benefit in a converged
wireline/wireless environment. In addition to the
Peer-to-Peer Sharing – Like in the wired radio access infrastructure, the architecture needs
Internet, peer-to-peer sharing of content among to include the following:
wireless users will likely grow to dominate
traffic on the 4G network.  An open client environment enabling
consumer installation of new services and
When examining these new and emerging applications, which are deployed either by
applications, we find an interesting trend that

2008 NCTA Technical Papers - page 287

 the network operator, or by 3rd parties in the Enhancements for Evolved Universal Terrestrial Radio
Access Network (E-UTRAN) Access (Release 8)].
Internet. [8] 3GPP TS 23.402 [3rd Generation Partnership Project;
 Content Processing: Encoding and Device Technical Specification Group Services and System
Aspects; Architecture Enhancements for non-3GPP
Adaptation, Ingest and Storage, Metadata Access (Release 8)].
processing, DRM. [9] 3GPP TS 23.228 [3rd Generation Pertnership Project;
Technical Specification Group Services and System
 Middleware: Consistent implementation Aspects; IP Multimedia Subsystem (IMS); Stage 2].
among all clients and network servers of key [10] Long Term Evolution (LTE): A Technical Overview,
supporting capabilities such as EPG, content http://www.motorola.com/mot/doc/6/6834_MotDoc.p
df, Motorola, 2007.
merchandising mechanisms, and interactive [11] F. Wang, A. Ghosh, C. Sankaran and P. Fleming
features. WiMAX Overview and System Performance, IEEE
Vehicular Technology Conference (VTC), Sept. 2006.
 SIP Session Management: Consistent [12] F. Wang, A. Ghosh, C. Sankaran and S. Benes,
orchestration of multimedia sessions. “WiMAX System Performance with Multiple
Transmit and Multiple Receive Antennas”, IEEE
 Service Delivery: Management of Vehicular Technology Conference (VTC), April. 2007.
application introduction, deployment, and [13] Y. Sun, W. Xiao, R. Love, K. Stewart, A. Ghosh, R.
Ratasuk, B. Classon, Multi-user Scheduling for
charging for services. OFDM Downlink with Limited Feedback for Evolved
 Policy and QoS: Determination of a UTRA, IEEE 64th VTC Conf. Fall 2006.
session’s QoS needs and entitlements, and
orchestration of this across the end-to-end
network via content selection/adaptation, ACKNOWLEDGMENTS
bearer flow establishment, etc.
The authors wish to thank the following
A unified mobile media solution offers many people for their valuable input to this paper:
carrier revenue generation opportunities
including ubiquitous always-on access to existing Kishore Albal
services, location based derivatives such as Pete Armbruster
location based advertising, public services, and Samantha Buechele
increased dependency (stickiness) on existing Herb Calhoun
operator services and content. Amitava Ghosh
John Harris
Marie-José Montpetit
REFERENCES Mike Needham
[1] Robert E. Kahn, Steven A Gronemeyer, Jerry
Burchfiel and Ronald C. Kunzelman, Advances in
Packet Radio Technology, Proceedings of the IEEE
Vol 66, No 11 November 1978.
[2] Motorola Strategy and Business Development, internal
analysis.
[3] Bandwidth evolution prediction by Motorola,
extrapolated from publications and presentations from
Arther D. Little, Motorola, Ericsson, and Alcatel.
[4] User Generated Content--More than Just Watching the
YouTube and Hangin' in MySpace, In-Stat,
http://www.instat.com/Abstract.asp?ID=212&SKU=I
N0602976CM, September 2006.
[5] I-Mode Overview. MobileInfo.com,
http://www.mobileinfo.com/imode/index.htm.
[6] M:Metrics: iPhone Hype Holds Up,
http://www.mmetrics.com/press/PressRelease.aspx?art
icle=20080318-iphonehype, March 2008.
[7] 3GPP TS 23.401 [3rd Generation Partnership Project;
Technical Specification Group Services and System
Aspects; General Packet Radio Service (GPRS)

2008 NCTA Technical Papers - page 288

 FEMTOCELLS—THE GATEWAY TO THE HOME

Sheriff Popoola
Senior Manager, Product Line Management
Motorola Connected Home Solutions

Abstract several key areas; a combination of expertise not

previously required from a single cellular
Femtocells are small, low-cost cellular base infrastructure product.
stations optimised for use in the home and small
businesses. This paper discusses this exciting  It must perform in a hostile RF
new market and concludes that femtocell and Wi- environment.
Fi technology will be co-existing, rather than
competing, to deliver a comprehensive digital  It must meet the high expectations of a
home experience. mature cellular subscriber base.

It describes how femtocells will enhance the  It must integrate seamlessly with existing
delivery of telecommunications services in the HFC access networks.
home and the new possibilities arising from the
integration of femtocells with home gateways  It must be capable of being deployed and
and set-tops. It will also point out technical supported in high volume.
challenges cable operators must assess and
outline the opportunities for cable operators  It must extract maximum performance from
complementing cable access infrastructure with HFC backhaul and it must be capable of
femtocells to enhance market share and customer being remotely managed without excessive
retention through enhanced triple play and quad operator effort.
play services.
These “must haves” attributes demand a
skillset that cable operators can rely on for
delivering end-to-end femtocell solutions that
A FEMTOCELL OVERVIEW increase Average Revenue Per User (ARPU),
grow market share, and enable innovative
Femtocells—miniature cellular base stations partnerships with wireless service providers that
that connect via cable infrastructure to provide increase brand value and allow cable operators to
enhanced 3G signal within the home – represent develop innovative triple play and quad play
arguably the most exciting development in home services.
networking since the arrival of Wi-Fi®.
STANDARDS INTEGRATION
Interest in femtocell technology is reflected
by growing activity among telecom operators Femtocell technology is very new and
and hardware vendors alike. Research firm IDC, developments are fast moving and exciting. As a
predicts that spending on femtocell-enabled consequence, product development is far ahead
services will grow to $900 million by 2011. of standardization. Cable operators require that
new femtocell deployments rely on industry
The unique demands of a high-performing standards and enable smooth integration and the
femtocell ecosystem demands competence in

2008 NCTA Technical Papers - page 266

ability to deliver seamless mobility of voice and THE NEED FOR FEMTOCELL
data services. DEPLOYMENTS

If the femtocell market is to grow as Femtocells will deliver home broadband

expected, the evolution of common standards is communication like never before by enabling
the best way to achieve this. The interface personal devices to perform seamlessly in and
between the femtocell Customer Premises out of the home. Connection to the mobile
Equipment (CPE) and the femtocell aggregator network via a gateway over an existing HFC
(sitting in the operator’s headend) is key to connection, femtocell solutions make indoor
interoperability. coverage for mobile communications truly
pervasive while delivering additional benefits to
This is also the interface that’s most open to both the cable operator and the end-user.
interpretation, so it is important to monitor the
evolution of industry standards. Motorola is very A femtocell is a small, low-power, self-
active in the standards bodies to move this debate installed cellular base station optimized to
forward but in the meantime is supporting two deliver cost-effective coverage in the home and
subtly different versions of the CPE-aggregator small office environment. Once installed in a
interface within the company’s femtocell product customer's home, it enables the operator to
portfolio. Femtocell trials are being conducted provide higher-quality and higher-performance
this year, with commercial launches expected to wireless voice and real-time data services to their
begin in late 2008 or 2009. customers inside their homes. Today, 3G is the
focus for femtocell technology. The full 3G
service set can be delivered in the home from a
small stylish device, which is connected to the
mobile operator's core network using open 3GPP
based standards via the consumer's HFC
connection

2008 NCTA Technical Papers - page 267

 Femtocell products require extensive to provide an enhanced 3G signal within the
experience in collapsed architectures, RF home.
techniques, home gateways, CPE management,
and fixed-mobile solutions to enable cable Femtocells—which may look like a stand-
operators to launch new services with fast time- alone consumer device sitting on a kitchen
to market and low risk. Femtocells are aimed at counter—actually function as part of the
providing high-performance 3G voice and data provider’s network infrastructure. Consumers
communications in and around the immediate primarily use their mobile phones at home, even
home environment. Connected to the operator’s when they have a fixed-line telephone. People—
mobile network over existing broadband especially younger consumers—have become
connections in the home, femtocells have the accustomed to the mobility conveniences of a
potential to make indoor coverage for mobile single communication device.
communications truly pervasive while delivering
additional benefits to both the operator and end- Additionally, more people would prefer one
user. number and one device to handle all their
communications needs, whether they are in the
The femtocell network architecture provides home or at work or play. An October 2007
operators with a complete indoor coverage survey commissioned by mobile content backup
solution, all in a small, low-cost, low-power, services provider, FusionOne, Inc., found that
easy-to-install base station that can be seamlessly more than half of respondents indicated that their
integrated into existing mobile networks and social lives would “suffer” if their mobile phone
provisioned for service within minutes of were to go missing.
switching it on. Such platforms enable a host of
new applications and revenue opportunities, and Most wireless operators agree that a
provide cable operators with the means to significant proportion of all calls made from
prevent the loss of subscribers to carriers offering mobile phones are initiated indoors, so it
bundles of wired and wireless access services. becomes understandable why providing good
indoor coverage is essential to provisioning cost-
Femtocell allows cable operators to effective, high-quality and higher-performance
aggressively enter the fixed/mobile convergence wireless voice and data services to consumers.
market. Femtocell deployments will address the
driving need for seamless mobility of voice and Allowing subscribers at home to connect to
data service. the wireless operator’s mobile network over
existing HFC infrastructure allows cable
ADDRESSING SUBSCRIBER PAIN POINTS operators to make mobile communications truly
pervasive, creating long-term bonds with
Every consumer of wireless voice services subscribers that minimize churn and enabling
has had the experience of being on an important new revenue opportunities from bundles mobile
wireless phone call and losing their cell signal. In and HFC service packages.
light of that consumer pain, femtocell technology
is arguably one of the most exciting FEMTOCELL BENEFITS FOR THE CABLE
developments in home networking since the OPERATOR
arrival of Wi-Fi – both are enabling operators to
better meet consumer demands for seamless Femtocells enable cable operators to provide
connectivity. The low-power, wireless femtocell higher-quality and higher-performance wireless
access points are optimized for use in the home voice and real-time data services to their
and small businesses, connecting via broadband residential and small home office customers.

2008 NCTA Technical Papers - page 268

They will be able to offer subscribers high-  Lower-cost voice calls from within the
quality 3G services at lower costs while they are home.
in their homes.
 Consolidated billing for voice and data
In addition, they enable a lower cost of services.
delivery of wireless traffic in comparison to the
macro cell network. Femtocells can be used as  The convenience of using a mobile handset
part of integrated triple or quad play services, with its personal phonebook and other cool
which meet consumer communication needs— handset features, without the concerns of
increasing the competitiveness and customer poor call quality or additional cost.
retention for the cable operator.
ADDRESSING TECHNOLOGY
The integration of cable and femtocell CHALLENGES
technologies will allow cable operators to fend
off attacks from carriers, create longer-lasting The successful deployment of femtocell
relationships with subscribers, and drive new technology can bring great rewards, but require
revenue growth by offering attractive seamless that cable operators overcome diverse technology
mobility services. challenges.

Femtocells have an important role to play in Operators face significant challenges with the
driving premium mobile service adoption, finally deployment of new technology, new applications
turning the 3G service vision into a reality by and the ever-increasing usage demands placed
encouraging a culture of usage through low-cost upon mobile networks. At the forefront of these
high-performance mobile data services. challenges is making 3G coverage as “near
ubiquitous” as practically possible, both
FEMTOCELL BENEFITS FOR THE outdoors and indoors.
CONSUMER
The traditional 3GPP 3G network
For consumers, the benefits of femtocells architecture, made up of numerous macro base-
include: stations, with its centralized RNC function and
ATM backhaul was designed to provide wide-
 A seamless communication experience as area coverage. It was not however designed to
they roam from inside to outside their scale, physically or economically, to provide
homes. effective coverage for individual
indoor/residential situations.
 Greater convenience via effective fixed-
mobile substitution by removing the need Cable operators can make a compelling value
for users to have separate home phones and proposition for partnering with mobile operators.
offering the flexibility for consumers to It is common knowledge within the mobile
rely on a single phone for access on the telecommunications industry that the use of
road or at home. outdoor macro-cells to provide indoor/residential
coverage quite simply does not provide an
 Reduced in-home call charges. effective solution, from both a coverage and
economic perspective. It also impacts practical
 Excellent indoor coverage. matters, such as site acquisition, which is
becoming increasingly problematic.

2008 NCTA Technical Papers - page 269

 Not only is site acquisition costly, forming a 3G signals, operating at very high
major proportion of network build-out costs, it is frequencies and high bandwidths have a poor
also very time consuming with many local ability to penetrate through structures. This often
authorities closely regulating the sale and usage leads to service quality and service experiences
of potential cell-sites. Assuming suitable sites that do not meet end-user expectations and can
can be acquired, increasing cell-site density lead to dissatisfaction, reduced minutes of use,
through the use of smaller cells, may not increased customer churn and ultimately, lost
overcome all coverage issues but will lead to revenues.
increased backhaul costs and other practicality
issues. Most end-users of 3G services invariably
have to settle for the coverage provided by the
People are becoming increasing reliant on macro base-station serving their location at that
their handset device to the extent that it forms point in time, whether stationary out in the
“part of their identity”; similarly more and more world, in a building or while on the move. The
people would prefer one number and one device issues associated with providing coverage for
to handle all their communication needs be it in indoor situations from macro base-stations are
the home or elsewhere. Many end-users prefer to well known; 3G and buildings, or to be precise
use their mobile phone when in the home, even their fabric, are inherently not a good mix.
where a fixed-line telephone is available. People
have become accustomed to and take for granted Since late 2006, interest in femtocell
the convenience that the mobile phone provides solutions has increased to the extent that most
in terms of mobility and in having a single device industry analysts suggest femtocell deployment
to communicate that includes their contacts and will become widespread in the coming years.
even takes and stores their messages in a variety
of formats.

2008 NCTA Technical Papers - page 270

 FEMTOCELL CONSIDERATIONS AND power levels than it might otherwise have to
CHARACTERISTICS when on the macro network.

Having considered the major drivers for Capacity—Femtocells are aimed at

femtocell deployment, this section looks at some delivering dedicated 3G coverage to a household
of their practical aspects. Femtocells overcome and in doing so can provide a very good end-user
the issue of providing effective indoor coverage experience within the home environment. As a
from the 3G-macro layer by their placement in result, femtocells have a design “capacity” of up
the end-users’ homes. to 20 registered users and 4 simultaneously
active calls.
Once installed in an end-user’s home a
femtocell will enable the cable operator to Low-cost—There is significant competition
provide higher-quality and higher-performance for access solutions in the home space. Wi-Fi is
wireless voice and 3G data services in and commonplace, and easy to install/configure.
around the immediate vicinity of the home Femtocell platforms in the home should compare
environment. favorably with Wi-Fi base stations in cost and
performance.
Femtocell products are in many ways similar
to Wi-Fi access points in that they enable access Low-power consumption—Clearly if the
through an unobtrusive device; however end-user is to foot the bill for the electrical
femtocells enable full 3G service delivery in the energy consumed by the femtocell base-station
home. Similar in size to a cable modem, a then this figure must be low enough not to raise
femtocell is a low-capacity base-station, concerns as to its impact on the fuel bill.
radiating only sufficient power to cover the area
of a home environment. The femtocell connects Easy end-user installation—Like cable
to the operator’s core network using open 3GPP modems and DSL routers, femtocells will be
based standards through the end-user’s installed by consumers and activated through
household broadband Internet connection rather service providers. This means that the cable
than traditional cellular backhaul methods. operator will not have to employ installation
Accordingly femtocells must also fulfill a teams or have a truck-roll every time a new
number of other criteria: femtocell is deployed. From the end-user
perspective the unit must be a simple “plug and
Low-impact—Space may be limited for play” installation with a minimal amount of
some households. As a result femtocells must be intervention required.
physically small, aesthetically pleasing and easy
to position. Furthermore, they should also be Interference—The use of femtocells in
silent in operation, generate low levels of heat spectrum also currently used by the macro layer
output, and be inexpensive to run in terms of on- may, if not managed correctly, give rise to issues
going electricity costs. with interference between cells; macro with
femtocell and in the instance of close proximity
Low RF power—The transmit RF power of two or more units, femtocell with femtocell.
output of femtocells is low, typically less than 10 Operators will likely want to launch femtocells
mW. Put in perspective, this is a lower power on the same channel as macro cell networks for
level than many Wi-Fi access points, which capacity reasons.
transmit at 100 mW of output power.
Additionally, by being close to the femtocell the Handovers—Current macro RF planning
3G handset itself is able to transmit at lower techniques are inappropriate for femtocells

2008 NCTA Technical Papers - page 271

because of the sheer potential numbers of monitoring via a centralized Operations,
femtocells. Also the potential to “ping-pong” Administration, Management, and Provisioning
between layers, especially as an end-user moves (OAM&P) system.
around the home and enters into areas where the
signal strength from the macro-cell is greater New services and applications—Femtocells
than that of the femtocell, must be considered are likely to become an integral part of managing
very carefully to ensure that the networks all communications in and out of the home
provide the best overall coverage without issue. environment. They will enable cable operators to
Femtocells introduce new complexities in macro cost-effectively offer in-home pricing and
to Femtocell, Femtocell to macro, and Femtocell integrate mobile services into triple-play / quad-
to Femtocell handover scenarios play service offerings. Femtocell architectures
will need to include provisioning for a complex
Security—Given the requirements for low- service environment on which applications may
cost and easy installation, the use of the be added, thereby facilitating new revenue
broadband Internet as the network interface opportunities.
becomes very easy to understand. However this
raises security risks in that broadband Internet Service Assurance—Remote management is
has open access. There are various approaches to needed to enable an operator to provide the end-
address this issue including the embedding of the user quality of service needed at the edge of the
interface within the IP signaling itself while network.
network security is managed by the IP security
(IPSec) protocol. INDUSTRY TRIALS

Worldwide cellular network standards– Cable operators can turn to the following
Understandably femtocell products are likely to URLs for up-to-date information on femtocell
appeal to many end-users around the world. As a technology:
result differing models will be developed and
offered to satisfy the various needs from the  www.femtoforum.org
different regions. Products should offer support
for their respective and existing (3GPP) UMTS  www.motorola.com/femtocell
and (3GPP2) CDMA standards, as well as
While the previously mentioned criteria and
emerging standards such as Imax, UMB and
challenges are being addressed throughout the
LTE.
industry, femtocell testing and deployment
Support for existing 3G handsets and continue to happen around the world. Selecting
devices—Support for existing handsets and equipment vendors with experience in femtocell
devices is a very important consideration for the trials is essential, and trials are now underway
end-user and operator alike. In each technology primarily in Europe and North America.
market, femtocells will support existing handsets
These developments and the industry’s early
and devices, further helping to drive uptake of
groundwork are leading toward the realization
3G services and femtocells in particular.
that as technical and commercial challenges are
Operator control—Femtocells operate in resolved, a femtocell in every home could
licensed spectrum and as such cable operators become a reality.
must ensure that they comply with regulatory
requirements. Femtocells need to feature client
software that enables remote configuration and

2008 NCTA Technical Papers - page 272

 platforms at customer locations and activate and
FEMTOCELL MANAGEMENT
support residential subscribers without bearing
Cable operators need software platforms that the burden of unnecessary truck rolls.
will allow them to remotely access, configure,
END-TO-END FEMTOCELL SOLUTIONS
and troubleshoot the full portfolio of consumer
devices – including mobile phones, CPE and Cable operators implementing femtocell
femtocells. This helps to lower operators’ costs
connected home solutions can deliver highly
by reducing truck rolls and lowering operational
differentiated services that drive revenue growth
expenses. It also helps to increase revenues by and lead to longer-term relationships with
accelerating new service introduction. subscribers. Considering that the majority of
mobile calls originate in the home and end-users
Scalable, carrier-grade systems are needed
that can manages devices, home networks, and prefer to use a single handset, cable operators can
now turn to emerging solutions available to them
services. MSOs deploying femtocell solutions
that overcome the issues of poor in-building
will need to an excellent end-user experience,
and they need centralized OAM&P systems that coverage.
enable the efficient provisioning, management Until now, providing good mobile coverage
and operations of femtocell solutions.
for homes has largely been overlooked. That is
Centralized control is essential so that cable changing. Femtocells will provide a one-box
operators can deliver an excellent user solution: a small, low-cost, low power unit that
experience while minimizing support costs and
can be self-installed to provide mobile 3G
swiftly generating revenues from mobile
coverage to the home.
services.
For the end-user femtocell solutions will
MSOs need the ability to centrally automate
provide dedicated and reliable mobile 3G
service provisioning, upgrade femtocell

2008 NCTA Technical Papers - page 273

coverage in the home with opportunities for By relying on proven technologies already
preferential tariffs. For the cable operator, tested in femtocell trials, cable operators can
femtocells deliver cost-effective coverage and safely explore the value of seamless mobility by
new revenue and customer satisfaction offering integrated voice and data services
opportunities. available throughout the home over wireless
cellphones and PDAs.
Once in the home, the femtocell is likely to
encourage end-users to use their mobile as their Connected home solutions that leverage
single communications device irrespective of emerging femtocell technology will enhance the
their location. Femtocell solutions are also likely user experience, allowing the operator to increase
to increase minutes of use and ARPU and also ARPU through better home coverage and new
open up brand-new revenue streams for cable multimedia and location-based applications.
operators through the integration of mobile Selecting vendors experienced in delivering
services into triple-play and quad-play service high-volume CPE for both mobile and fixed
offerings. Cable operators need to minimize the networks is crucial, and the ability to develop
risk of deploying new technologies by relying on and manage end-to-end femtocell solutions will
vendors that already have proven expertise in the allow cable operators to prosper by enabling the
many technology areas required for successful future of the connected home.
femtocell deployment.
ABOUT THE AUTHOR
Solutions however are not simply about only
technology; they are about capabilities and Sheriff Popoola is Senior Manager, Product Line
delivery. That’s why cable operators need access Management with product responsibility for
to expert professional services to deploy Motorola’s Femtocell CPE offering covering UMTS,
GSM, xDSL and cable technologies. He has been with
femtocell solutions that leverage existing
Motorola for 13 years, and his experience includes
wireless standards. Networks in Public Safety, the Japan CDMA group,
Advanced Radio Technology group, iDEN Dispatch
End-to-end femtocell solutions will allow and Packet Data services group, and the Aspira
cable operators to differentiate their service Network Products group, in functional areas
offerings by providing seamless mobility and spanning Systems Engineering, RF Development
allowing subscribers to access voice and data Engineering, Digital Hardware Development
services as they move throughout the connected Engineering (where he received a patent in CDMA),
home. and Product Line Management.
Prior to joining Motorola Popoola worked for
While femtocell technology offers great Cellutech Communications as the Cellular Service
promise, selecting equipment from the right Manager running the Chicago-North facility. He has
vendor is key to successfully launching new a BSEEE degree from Obafemi Awolowo University
services. Cable operators should rely on fully in Nigeria, a MSEECS degree from University of
integrated and tested end-to-end solutions based Illinois, and an MBA degree from Northwestern
on open standards, which includes: University’s Kellogg Graduate School of
Management. He can be reached at
 A range of low-cost, easy to deploy CPE. [email protected].

 A core network concentrator.

 A centralized management and

provisioning system.

2008 NCTA Technical Papers - page 274

 FUTURE DIRECTIONS IN CABLE BROADBAND BANDWIDTH CAPACITY
John M. Ulm
Motorola, Home & Networks Mobility Technology Office

Abstract DRIVING BANDWIDTH NEEDS

This paper takes a long term look at This section discusses several drivers to
cable’s broadband bandwidth needs over a forecast bandwidth needs down the road.
10+ year horizon. It discusses several drivers Some of these may appear to conflict with
that forecast bandwidth needs down the road. each other, but each gives a unique
We compare HFC potential against FTTP perspective on bandwidth needs.
architectures and shows that there is plenty of
capacity left in coax to compete with GPON Moore’s Law continues
or even co-exist with it. HFC also has the
Over the years, Moore’s Law has driven
significant advantage in that it can
data bandwidth growth and it looks to
incrementally expand bandwidth through this
continue for the foreseeable future. Simply
entire process without requiring a massive
stated, Moore’s Law is the doubling of
infrastructure overhaul like pure FTTP
transistors per device every 18-24 months.
providers.
This increased technology capacity has
How far can the MSOs take coax looking resulted in a corresponding improved
way out on the 10+ year horizon? With the performance, density and power. Some
recent evaluations of the RF overlay systems, studies have shown that high speed data
we take a closer look at the theoretical broadband service offerings have closely
maximum capacity of coax. The paper tracked Moore’s Law over the last 8-10 years.
discusses how HFC can offer 1+ Gbps The implication that it will continue to track
services over coax and what is needed beyond means that we’ll need to offer 100 times
our current DOCSIS systems. today’s bandwidth in another 10-15 years.
With broadband providers offering data
Introduction services around 10Mbps today, this means
that the broadband providers should be
HFC has long enjoyed the position as the prepared to make 1 Gbps services generally
leader in providing broadband content to the available in 10-15 years.
home for both video & high speed data. Now
it’s being besieged by FTTP and satellite HD There are several aspects of data
technologies and many have claimed its days bandwidth growth that need to be examined.
are numbered. Our analysis shows that there In Andrew Odlyzko’s paper “Internet traffic
is plenty of capacity left in coax to compete growth: Sources and implications,” he
with GPON and even 10G PON technologies, contends that the data traffic growth will
given the appropriate investments. HFC also continue to follow Moore’s Law and is
has the significant advantage in that it can primarily driven by file transfers as opposed
incrementally expand bandwidth through this to streaming traffic like video. As bandwidth
entire process without requiring a massive growth is modeled, it may be necessary to
infrastructure overhaul like FTTP. separate traffic that is streamed and requires
constant bit rate service from the more

2008 NCTA Technical Papers - page 39

general, best effort high speed data traffic that Video dominates & HD becomes Mainstream
is following the Moore’s Law growth.
Among Video, Voice and Data services, by
But there are some questions with this far the one with the largest bandwidth
argument. Will Moore’s Law continue requirements is video. Not only does video
forever? Will technology improvements require a high data rate of multiple Mbps, it
continue in the density and power areas but streams for extremely long periods of time,
less so in the performance area? We may not maybe even several hours. We are also
know the answer for another decade, but must reaching the point where High Definition
prepare in case it does continue to track. (HD) is becoming main stream. HD has a
significant impact on bandwidth requirements,
Supply Side Economics increasing the video bandwidth by a factor of
2-4X over Standard Definition (SD) video.
Is the right question to ask: “What is the
demand for bandwidth?” Some insist it is not Many indications over the last year have
and instead we must look at the bandwidth shown that HD adoption is accelerating. As
supply rather than the demand. Internet HD content rolls out, it will cause a
demand is (nearly) infinite and users have significant impact on short term bandwidth
been shown to try and consume everything needs. But what are the longer term impacts
given to them. So the operators should assume once the majority of video content is delivered
that any high speed data offering could as HD? To deliver one or two narrowcast (i.e.
probably be 100% utilized. personalized) HD streams along with a couple
of SD streams to each subscriber would
Instead of looking at demand, the right require ~20Mbps per home using MPEG-4.
question should be “How much bandwidth
should I supply?” And the answer to that is However, this may actually be understating
based primarily on competition. In the early the bandwidth needs. As HD penetration
days, cable modems were competing with becomes truly pervasive, then we may need to
dial-up services and only needed to offer support 3-4 HDTV’s per home with a couple
1Mbps service, even though the channel extra HD streams going to a DVR for
supported almost 40 Mbps. As DSL recording. On top of this, some of these
penetration increased, cable ratcheted up its streams may also be high quality 1080p HD
data rates to maintain a sufficient performance content requiring additional bandwidth. This
edge. As FTTP penetrations like Verizon’s longer term scenario would require closer to
FIOS increase, competitors will be able to flip 40-50Mbps per home to support streaming
the tables on cable and it will be up to cable to HD video. The bottom line is that Service
race and keep up with FTTP. Providers will need to continue to offer more
bandwidth capacity to the home as the
The difference between FTTP and HFC is percentage of HD content increases along
most notable in the upstream direction. Since with the percentage of HDTV penetration.
the HFC upstream capacity is significantly
less than the downstream, competitors could Burst speeds vs. Sustained rates
cause an immediate disruption in cable
operator’s business by emphasizing services Throughout computer history, data
that use significantly more upstream capacity. networks have been useful because of the
We are just starting to see this in current nature of statistical multiplexing. By offering
marketing efforts. a shared resource with high burst rates, users
get the impression that they have high data

2008 NCTA Technical Papers - page 40

rate services. In reality, data usage is very Peer-to-Peer (P2P)
bursty and average data rates are significantly
lower than the burst rate. As discussed previously, Internet demands
appear to be virtually unlimited and one of the
An example broadband system might be driving applications behind this is Peer-to-
engineered today to provide 1% concurrency Peer (P2P) applications. From its start in
for a 10Mbps data service. This means that music file sharing, P2P is branching out into
during peak busy hour each user will get an many more main stream applications. As P2P
average of 100Kbps. As network speeds video file sharing becomes prevalent, it will
increase, the statistical gain also increases. significantly increase the bandwidth demands
This implies that the concurrency can be on the system, especially in the upstream.
reduced as burst speed increases. So in this
example, the operator could decide to offer a Home Generated content
100Mbps service with 0.25% concurrency.
The result is that each user would get an Consumer devices and in particular mobile
average of 250Kbps during peak busy hour. devices have made tremendous progress in
recent years and look to continue in coming
The significance of this is that the data years. They have all become rich multimedia
service rate increased by 10-fold while the devices. Mobile devices some day will
actual bandwidth provided by the operator include 10 Megapixel digital cameras that can
only had to increase 2.5x. This factor will stream video clips. Digital camcorders are
become especially important as broadband becoming HD capable and inexpensive. All of
providers look to start offering Gbps services. this combined with the ease of use being
Note, this analysis applies only to bursty data introduced by UPnP/DLNA will create a
applications and does not apply to streaming tremendous amount of user generated content
voice or video. Care must be taken with that will be shared over the Internet.
concurrency if the bulk of internet traffic
becomes streaming video. In addition to these consumer devices, the
introduction of low cost video cameras (a.k.a.
Not All Subscribers are created equal webcams) will increase the number of
applications like video surveillance or “nanny
Another important aspect of understanding cams” that can be shared over the Internet.
bandwidth needs is to look at the usage across Video telephony will also increase the amount
the many different subscribers. Many of traffic coming from the home, and as
operators have noted that a relatively few previously mentioned, these applications will
users consume a proportionately large amount have a much larger and earlier impact on the
of the bandwidth. One piece of data indicated upstream bandwidth needs.
that 5% of the users consume two thirds of the
total bandwidth and that 25% of the users Wireless Backhaul
consume 95% of the bandwidth.
Mobile Devices are evolving from their
This has a major implication in how an roots as a telephony device to a full
operator rolls out its bandwidth capacity multimedia device supporting streaming
increases. If it can increase bandwidth video. This evolution is going to create a
incrementally to a small number of power tremendous demand for 4G capable wireless
users, then it can avoid costly upgrades that networks like WiMax and LTE as well as
go across the board to all subscribers. technologies like Metro-Wi-Fi. As the number
of mobile devices continues to rise along with

2008 NCTA Technical Papers - page 41

improved video quality and screen sizes, the developed that offers even higher video
bandwidth needs will continue to escalate. resolutions, with corresponding higher
Cell sizes will need to continually shrink to bandwidth demands.
accommodate the wireless bandwidth density.
Cable operators have a further incentive for As we think about enhancing the user
creating wireless networks to maintain a video experience, the simple 2-way video call
competitive position with Telcos who already may be improved through the use of tools
own spectrum and provide cellular services. such as avatars and facial recognition. Beyond
this, we may get into other enhancements to
As 4G cell sizes are reduced to several the user video experience with things like
hundred meters, the number of cell sites visualization and holography. At this point,
increases significantly and the site become there is no hard data on the bandwidth
more geographically dispersed. This creates a capacity impact, but we should continue to
great opportunity for the Broadband Service monitor and keep an eye for killer bandwidth
Providers to supply the backhaul over their applications. The only thing we do know is
existing infrastructure. This may put that bandwidth needs have always increased.
tremendous additional bandwidth demands on
their access networks.
EXISTING HFC CAPACITY UPGRADES
3-D and Multi-view technologies
Current techniques for expanding HFC
Since video is the main bandwidth hog bandwidth are well known and include: RF
today, it is important to understand future Upgrades (1GHz), node splitting and deep
variations that will be hitting the market and fiber expansion, Switched Digital Video
potentially creating the next big impact to the (SDV), reclamation of analog channels,
industry. At consumer shows like CES, we are MPEG-4 encoding, DOCSIS 3.0 & M-CMTS.
starting to see High Definition technology that
supports both 3-D and multi-view Node Splits and Deep Fiber
technologies. The initial thrust for this
technology will be the gaming world, studios In today’s HFC plants, there are many that
starting to create 3-D movies as well as sports still have 500, 750 or even >1000 Households
casting. Eventually, this technology may Passed (HHP) per fiber node. These plants
become main stream just like HD. The 3-D were designed and optimized for a broadcast
technology may cause a 50-100% increase in system. As we evolve to a completely on-
the bandwidth required to deliver the content. demand system, one of the most effective
The multi-view causes an even larger means of increasing bandwidth capacity is to
increase, as it needs to deliver separate HD reduce the node size by a factor of 2, 4, 8 or
streams for each view provided to the user. even more.

Future User Experiences Typical Fiber Nodes have 2 to 4 coax

outputs. This allows the operator to split a
As we gazed into a very foggy crystal ball, node 2-way to 4-way by replacing electronics
we tried to imagine what applications of the inside the existing housing without pulling
future might drive a new paradigm in any additional fiber. The operator also has the
bandwidth requirements. There may be other option of splitting the upstream independent
technologies in addition to 3-D and multi- of the downstream. Splitting the downstream
view just discussed that will impact the user does require additional narrowcast
experience. An ultra-HD technology may be wavelengths to be sent to the node.

2008 NCTA Technical Papers - page 42

 To reduce node size further, the operator mechanism to allow the transition to new
can push the fiber deeper into the HFC plant. technologies, such as migrating to 1 GHz
This can often be done cost effectively using tuners, or from SD to HD as well as from
smaller satellite nodes. While many of today’s MPEG-2 to MPEG-4. Over time, once the
HFC systems support a six amplifier cascade new technology becomes the dominant
(N+6), newer deep fiber systems may installed based, then any remaining broadcast
eliminate (N+0) or have a single (N+1) technology can also be converted to the newer
amplifier. A deep fiber system may reduce technology while the older technology will be
node size to 125 HHP or even less. completely switched using SDV.

A key issue with node splitting and Analog Reclamation

pushing fiber deeper is available fiber count.
The number of fibers in an HFC plant can Analog TV channels consume 50-75% of
vary dramatically from plant to plant. For the spectrum in today’s typical HFC system.
those plants with low fiber count, Wave Analog TV channels are also extremely
Division Multiplexing (WDM) becomes a inefficient with the use of spectrum. A single
critical technology in providing additional 6-MHz TV channel can be replaced by a
narrowcast wavelengths and hence additional digital QAM channel delivering 10-15 SD
bandwidth capacity. A good example of this programs. Digital QAM channels are also
technology is Motorola’s Enhanced Coarse very versatile and can be dynamically
WDM (E-CWDM) system that was assigned to VOD, SDV or high speed data.
announced at the 2007 SCTE Cable-TEC
show last June. So why aren’t cable operators just dumping
all of their analog channels and going all
A key advantage for HFC systems is that digital? Supporting analog TV channels has
they only need to split nodes that need the become a competitive advantage for cable
extra capacity. For example, if an entire operators over satellite providers. This will
community is configured with 750 HHP, but become even more critical after 2009 when
only one neighborhood has exhausted its on- the Over-the-Air analog channels are no
demand bandwidth, the operator only needs to longer broadcast. Many homes have a 2nd, 3rd
split that one node to expand bandwidth and even 4th TV in the house. Satellite
capacity to meet demand. providers must add a new STB for every
additional TV in the house. Once the Over-
SDV the-Air analog channels are removed, cable
service will be the only way for consumers to
Switched Digital Video (SDV) is different get basic local services to these other TVs.
than the other bandwidth approaches. Rather
than increasing bandwidth capacity, SDV Over time, those operators that want to
provides a mechanism to better utilize continue to offer an analog service can still
existing capacity. SDV allows the operator to reclaim a large portion of the analog channels
convert a fixed number of broadcast channels to get a significant increase in bandwidth
into a potentially unlimited number of video capacity while still offering consumers a
channels within existing spectrum. This reasonable analog service. For example, a
becomes even more important as the amount HFC system with 125 6-MHz channels could
of HD content being offered increases. reduce the number of analog channels from 75
to 25. This would double the bandwidth
In addition to increasing the number of capacity available for digital QAMs, while
offered channels, SDV is an important

2008 NCTA Technical Papers - page 43

still offering a basic analog service with all advanced technologies like SCDMA and
local and major network TV channels. improved ingress cancellers. SCDMA allows
the operator to re-coup data bandwidth from
DOCSIS 3.0 and M-CMTS the lower 5-15 MHz spectrum. All told, an
operator can get a total of 140 Mbps of
DOCSIS 2.0 systems are currently in their DOCSIS upstream bandwidth with
prime, but will be running out of steam over appropriate improvements.
the next several years as it tries to compete
with FTTP. The DOCSIS 2.0 cable modem After improving the existing upstream,
can support up to 38 Mbps downstream and another common method of increasing
30 Mbps upstream before it hits a brick wall. upstream capacity is to split it into smaller
DOCSIS 3.0 will be coming on line shortly node sizes. This can be done independent of
and its channel bonding feature will allow splitting downstream node sizes and can also
significantly improved data rates to the reduce the noise per upstream. Most fiber
subscriber. A DOCSIS 3.0 cable modem with nodes support two to four coax legs. Each
8 downstreams and 4 upstreams could enable coax leg is potentially a separate upstream
a 300 Mbps downstream and 100 Mbps return spectrum.
upstream service.

While DOCSIS 3.0 enables a bigger IP HFC VS. GPON COMPARISON

pipe to the subscriber, it does not impact the
additional headend costs relative to video But how does HFC today stack up against
bandwidth associated with today’s integrated- GPON in raw bandwidth capacity? It turns out
CMTS. A new Modular-CMTS (M-CMTS) quite well . . . in the downstream. Up until
architecture has been defined by DOCSIS to now, we’ve discussed many mechanisms that
address this. It decouples the upstream from HFC may use to incrementally increase
downstream and separates the RF technology bandwidth. At this point, we will compare an
from the CMTS core. This allows commodity HFC system using these available
driven Universal Edge QAM modulators to be enhancements to a GPON system. The HFC
shared between VOD, SDV and CMTS system under consideration has 1GHz RF
resources. This is the first major step in with deep fiber nodes [125 House Holds
reducing the cost of delivering IP packets over Passed (HHP), 100 subs]. We also assume
DOCSIS and should reduce its cost relative to about half of the analog channels have been
delivering video from 10-20X in the early reclaimed, so the system reserves ~40 analog
days down to ~2X. Over time, the gap should channels and 8 digital simulcast QAM
continue to shrink as improvements are made channels (i.e. 80-120 digital video broadcast
to the all-digital CMTS core. streams). This means about 100 QAM
channels are available for switched
Upstream splitting and stacking VOD/SDV/Data. This is approximately 1
QAM per sub or slightly less than 40Mbps
The current HFC bandwidth capacity is downstream capacity per sub.
extremely asymmetric, with the downstream
spectrum occupying 54 MHz to 1GHz while This is almost identical switched
the upstream spectrum is limited to 5-42 MHz downstream capacity to a GPON system with
in North America. The upstream bandwidth 64 subscribers (i.e. 2.4 Gbps / 64 subs = 37.5
capacity is further hampered by operating in a Mbps per sub). The implication here for the
much noisier environment. To get the most HFC system is a significant increase (almost
out of an existing upstream requires the use of 100-fold) in the number of QAM modulators

2008 NCTA Technical Papers - page 44

to achieve this. Edge QAM devices will need RFoG & CablePON – FTTP for cable
to continue to make significant improvements
in cost, density and power to achieve this. MSOs may have certain targeted areas
However, the undesirable alternative for where it makes economic sense to install a
MSOs is to pull fiber to all 100 homes. It complete fiber based solution. This may
should be noted that some FTTP vendors are include support for an industrial park or FTTP
deploying GPON with 32 subscribers per in a new housing development. If the existing
PON. The HFC can continue to match this by bandwidth requirements allow, the cable
pushing the fiber even deeper (e.g. N+0 with operator may initially support existing cable
50-65 HHP). services over the FTTP, while having the fiber
in place for future expansion. This approach is
The GPON system still has an edge in RF called RF over Glass (RFoG) and is currently
bandwidth, burst speed and upstream being standardized by the SCTE.
capacity. The GPON system has an optional
750MHz RF carrier while the previous HFC There may be some applications where the
system had set aside about 350MHz for current HFC bandwidth solutions are
analog and digital simulcast channels. For inadequate. Some of these include Gbps
certain applications like large file transfers, Commercial Services or 4G Cellular & other
the burst speed of the network is critical and wireless backhaul. For these, Motorola’s
GPON still has a large edge in this category. CablePON solutions allow operators to offer a
For upstream bandwidth, GPON supports GPON solution where needed but within their
~1Gbps while the DOCSIS 3.0 system only existing legacy cable equipment
supports about 100Mbps which it will still fall infrastructure. This solution is different than a
short of GPON’s upstream capacity. traditional GPON deployment in that the
MSO leverages the fiber portion of its HFC to
transport the services. This is also different
UPCOMING HFC BANDWIDTH from traditional GPON in that the devices fit
CAPACITY UPGRADES into the MSOs back office infrastructure.
Another important aspect is that this solution
Modulation improvements fits within the cable operators existing router
and M-CMTS infrastructure without requiring
As HFC systems continue to improve, they expensive B-RAS equipment.
will eventually be able to support higher order
modulations like 1024 QAM. Increasing the RF Overlays
QAM constellation density increases the
MPEG-2 transport bit rate almost 28% to Other technologies like RF Overlay
almost 50 Mbps. However, this higher systems offer the prospect of increased RF
throughput comes at a cost: the required bandwidth capacity. Some recent technologies
threshold signal-to-noise ratio (SNR) to offer a 2-3 GHz system adding additional
achieve the same reliability (bit error rate) is downstream and upstream capacity. These
at least 6 dB higher using conventional J83.B systems are being evaluated from both a
FEC coding. Using 1024 QAM with advanced technology and business perspective. While it
FEC can yield up to 23% higher throughput may not make business sense to apply this
with a moderate increase of ~3dB in threshold technology to an entire plant, it may prove
SNR. useful to bring additional bandwidth to a
particular site (i.e. a surgical strike) where it is
not feasible to extend the fiber portion of the
plant. RF Overlay technology will most likely

2008 NCTA Technical Papers - page 45

be used in conjunction with other plant Hybrid PON Coax (HPC) systems
upgrade techniques.
As bandwidth needs for residential users
DOCSIS 3.0 Mid-Split Systems continue to increase, it will eventually make
sense to take the small percentage of users
In addition to channel bonding, DOCSIS who consume a significant portion of the total
3.0 also supports an option for a mid-split available bandwidth on existing HFC systems
HFC system with additional upstream and offer them Gbps services thru an FTTP
capacity. The upstream spectrum is increased solution. In this scenario, it would be highly
from 5-42MHz to a 10-85MHz range. The desirable that an MSO have the capability to
added frequency range is also significantly drop fiber to any existing individual
less affected by impulse and ingress noise, so subscriber.
the net effect is that upstream bandwidth
capacity may almost quadruple (e.g. from One method of accomplishing this is with a
27MHz usable at 16-QAM average to 70MHz concept called Hybrid PON Coax (HPC). The
usable at 64-QAM average). The mid-split HPC system is basically a CablePON/GPON
upgrade needs to be done in conjunction with overlay on top of existing HFC plant. HPC
reclaiming the lower analog TV channels allows the MSO to install pockets of FTTP
(channels 2-6) and a plant upgrade that within its existing HFC infrastructure. Having
replaces/eliminates the old diplex filters. the ability to drop a fiber connection to a
single home without upgrading its entire plant
DIBA – CMTS By-Pass technology can give MSOs marketing leverage to combat
its telco rivals.
As the world migrates to an all IP
environment for the delivery of video, the Statistics have shown that a small
additional cost of delivery over a DOCSIS percentage of users consume a large portion
network becomes more important, even with of the data bandwidth. The MSO can
an M-CMTS approach. Motorola has potentially move 5-10% of its subscribers to
pioneered a concept called DOCSIS IPTV FTTP and free up two thirds of its high speed
Bypass Architecture (DIBA) that allows HFC data bandwidth. This means that with a
session oriented IP traffic to bypass the relatively small fiber plant investment, the
CMTS core and go directly from its server to MSO can significantly extend the life of its
the Edge QAM device. This greatly reduces HFC. The HPC also allows the MSO to
the needed CMTS core capacity which in turn combat the telco marketing in being able to
reduces the relative cost of delivering IP offer FTTP to anyone, but with the huge
packets over DOCSIS by more than two advantage compared to them that it only needs
thirds. to pull fiber to the select few that need the
Gbps service.
Over the long term, DIBA holds the
promise on economically converting cable Headend Impacts from Bandwidth Increases
systems to an all IP infrastructure all the way
to the home. Motorola has published several Increasing HFC bandwidth capacity is
white papers on this topic, including a paper more than just upgrading cable plants, it is
at the 2007 SCTE Emerging Technology providing equipment at the access edge to
conference. These papers provide a detailed deliver this capacity. For HFC to compete
description of DIBA. with GPON, it will require a significant
investment in additional Edge QAMs and
CMTS core capabilities. Other video

2008 NCTA Technical Papers - page 46

capabilities like transcoding and encryption area to understand completely how far we can
will need to be scaled as well. As technology push cable.
progresses these functions will get pushed
from the core further out to the access edge. For RF designs, the complexity is often
related to the number of octaves (i.e. doubling
In addition to the video components just of frequency) that the design must cover.
described, there will need to be new advances Current 1GHz systems must cover more than
in the high speed data transport beyond the 4 octaves since they start at 50MHz. Going
current M-CMTS developments to from 1 to 4 GHz adds two more octaves. As
economically scale data traffic for tens or investigation continues on multiple GHz
hundreds of Gbps data rates. Current CMTS systems, the lower couple octaves (e.g. 50-
architectures provide high touch, per 200MHz or -400MHz) should be considered
subscriber services similar to B-RAS dropped for new systems to help minimize
equipment. In order to economically get a cost and complexity.
hundred fold increase in data bandwidth, we
will need to start migrating to Class based As deep fiber architectures eliminate other
services and other technologies that can active components, we will reach the day
leverage standard Ethernet switching where the coax is the final limitation on
equipment. bandwidth capacity to the home. If the cable
hardline is ultimately replaced by fiber, the
There will also be the need to converge coax drop line has a much higher RF
future generations of DOCSIS CMTS and frequency limit due to its much smaller
GPON OLT products. As just discussed cable diameter. This could allow cable to the home
operators may need to start deploying HPC to even exceed 5 GHz. Other areas of
systems. With M-CMTS decoupling the PHY investigation could look at ways of increasing
layer in DOCSIS, a similar approach can be the theoretical limit of the cable as well as
taken with GPON to create a common packet using different waveguide mode(s) along with
processing core for both technologies going the lower order mode to increase the cable’s
forward. Both share similar capabilities in capacity.
routing, traffic shaping and policing as well as
seamless mobility support. This platform Next Gen Coax System
should also be extended to support other
access technologies such as WiMax, LTE and How far can the MSOs take coax looking
metro-Wi-Fi. out on the 10+ year horizon before it hits the
brick wall? From above, the RF hardline cable
might be capable of supporting systems in the
NEXT GEN COAX – FUTURE OF HFC 3-5GHz range. However, no standard exists
yet for devices above 1GHz. With the
RF Upgrades – 3GHz & beyond increased bandwidth needs, the old 6MHz
channel size no longer makes sense in this
With the recent evaluations of RF Overlay range. This gives us the opportunity to define
equipment, we have taken a closer look at the a new Next Gen Coax system above 1GHz.
theoretical maximum capacity of coax. It
turns out that the typical hardline coax being To be competitive with FTTP, Next Gen
used today has a limit of ~5GHz before Coax must support extremely wideband
waveguide effects take over. There may be channels (e.g. >100MHz wide) with dense
other effects that limit the total coax capacity, advanced modulations (e.g. 1024/4096 QAM
but we should continue investigation in this or equivalent) that are capable of delivering

2008 NCTA Technical Papers - page 47

greater than 1 Gbps symmetrical bandwidth to Next Gen Coax development should start
individual subscribers in a single or a small now so it can keep pace with 10G PON
number of bonded channels. The new development. As MSOs upgrade to N+0 and
standard should also eliminate MPEG-2 N+1 cable architectures, we should consider
transport and DOCSIS layers and define a enabling at least 3GHz RF, if not the full
simple all IP infrastructure that unifies all 5GHz as feasible. When you combine these
devices, including cable modems and STBs changes with the Hybrid PON Coax
with end-to-end IP connections. An all IP architecture that allows select users to migrate
infrastructure will also help with integrating to FTTP if needed, then the HFC system
the Next Gen Coax system with other FTTP appears to have a very long life ahead of it.
and wireless access technologies. And most important, all bandwidth increases
are incremental and invested as needed.
These wideband channels would be of
great value in existing HFC as well. The
ability of offering symmetric Gbps services HOME NETWORKING IMPACTS
within the existing 1 GHz spectrum would
significantly level the playing field with FTTP The Broadband pipe into the home is
competitors. It may make sense to roll the becoming a fire hose. The prospect of offering
wideband channel support out initially within 100’s Mbps or even >1Gbps to a home only
existing frequency ranges and then allow for makes sense if the home network can handle
1-5 GHz operation in future releases. that in addition to all of its local LAN traffic.
This will place a burden on existing
Next Gen Coax vs. 10G PON multimedia home networks. To meet this need
will require multiple wired and wireless home
Given the expected timeframe, a Next Gen networks to be interconnected.
Coax system would need to compete with a
10G PON system. How does this compare? For home networking over cable, next
Assuming the 1-5GHz range is split 2:1 in generation MoCA will be needed to scale to
favor of downstream traffic with a 1024 QAM several hundred megabits per second to match
modulation or equivalent, the Next Gen Coax or exceed DOCSIS 3.0 speeds. This next
system could theoretically support a total of generation MoCA will become critical for
20Gbps downstream and 10Gbps upstream cable operators to deliver its DOCSIS 3.0
bandwidth capacity. This is roughly bandwidth throughout the home and compete
equivalent to a pair of 10G PON systems. with FTTP.
This means that MSOs could keep their node
sizes roughly twice the size of the PON group Another important piece of the future home
(e.g. 125 HHP) and still provide the networking scene will be 802.11n. This next
equivalent bandwidth per subscriber as a 10G generation Wi-Fi network supports more than
PON. 100 Mbps and provides a number of features
that will improve robustness and range. We
If it turns out that there are other also expect to see smart antenna
limitations in reaching a 5GHz system, the developments coupled with 802.11n to
MSO can still provide a 3GHz system that improve performance. Work on the following
would provide an additional 10Gbps generation to 802.11n has started and may be
downstream and 5Gbps upstream capacity. a step towards approaching Gbps wireless
This along with smaller node sizes (e.g. 50 rates in the home.
HHP) can still keep coax competitive with
10G PON systems.

2008 NCTA Technical Papers - page 48

 Looking even further out there is the to downstream traffic. This bodes well for
possibility of 60 GHz technology evolving to offering personalized HD video services. As
provide higher bandwidth home networking. new applications start to drive upstream
The major issue with 60GHz technology will capacity, this will expose the upstream as the
be the ability to propagate through walls to cable operator’s Achilles heel. Increasing
provide whole home coverage. It is possible upstream capacity is an area that needs
that 5 GHz technology may be the limit in continued research and development.
terms of whole home wireless coverage.
In the near term, there may be scenarios
Next Gen Coax as Home Network technology such as offering Gbps Commercial Services,
wireless backhaul, or residential green field
As the Next Gen Coax technology is builds where the cable operator needs an
developed as a multi-gigabit access FTTP solution. For these, Motorola’s
technology, it should also be extended within CablePON solutions allows MSOs to offer
the home as the first multi-gigabit in-home FTTP where needed while operating
network. No other home networking completely within a legacy cable equipment
technology seems poised to address this environment. As the bandwidth race
challenge. While GPON currently provides continues, it may become necessary for MSOs
2.4 Gbps to the side of the house, actual burst to be able to offer Gbps service to select
data rates to consumer devices will be limited power users on existing HFC thru a Hybrid
to the home networking technology, which PON Coax (HPC) system. Moving the heavy
today is on the order of 100 Mbps or less. users off the HFC network extends the HFC
useful life and having the ability to drop a
If cable operators can develop multi- fiber connection to any home without
gigabit technology delivered all the way to upgrading its entire plant can give MSOs
consumer devices throughout the home over marketing leverage to combat its telco rivals.
coax, it can once again leapfrog its Telco
rivals. Another advantage with this approach Looking far out on the 10+ year horizon,
is the economies of scale from the consumer there is a possibility of a new Next Gen Coax
devices sharing the same technology as the system. Capable of operating up to 3-5 GHz
access devices. We’ve seen the benefits of with very wideband channels, the cable
this in the Wi-Fi world. operator can offer symmetric Gbps services to
the user while matching the overall bandwidth
CONCLUSION of a 10G PON system while maintaining 50-
125 HHP node architectures. This Next Gen
We have seen that the bandwidth needs Coax architecture may also be a catalyst to
will continue to increase for the foreseeable enable Gbps services throughout the home
future and that cable operators will need to network as well.
extensively support Gbps services within 10-
15 years. The cable operators will also get With all told, the future does not look
severe competitive pressure from FTTP bleak for HFC, but it looks to have a long and
providers like Verizon FIOS further healthy life. And the most important piece of
accelerating the need to increase bandwidth. this is that the HFC can grow incrementally as
needed without the need for a forklift upgrade
With existing upgrades and continued like its competitors.
technological developments in devices like
MOTOROLA is registered in the US Patent & Trademark Office.
Edge QAMs, the cable operator is in excellent DOCSIS is registered trademark of Cable Television Laboratories Inc
position to compete with GPON with respect Wi-Fi is a registered trademark of the Wi-Fi Alliance. All other marks
are the property of their respective owners. All rights reserved.

2008 NCTA Technical Papers - page 49

 HOW TO MONETIZE OVER-THE-TOP (OTT) VIDEO
Author Name: Eitan Efron
VP Marketing & Business Development, Oversi

Xbox 360 movies, and MySpace, already

Abstract generate more traffic than the entire US
backbone in the year 2000.i
The MSO role in the new media ecosystem is
under scrutiny. How can cable operators In a joint declaration by the Broadband
financially benefit from the demand for over- Services Forum in January 2008, it stated that
the-top (OTT) video services and become key “current projections on the growing popularity
players in the content delivery supply chain? At of OTT video predict that service provider
present, most operators are mere bandwidth networks are going to be operating at near or
conduits which do not receive any financial gain complete capacity by 2010.”
from the increasing amounts of Internet video
flowing through their broadband networks. Even with service providers spending
billions of dollars to build better, faster and
This paper will outline how edge content more reliable pipes, OTT video could bring
distribution networks (CDNs) deployed at the many networks to a grinding halt in the near
cable operator core network can accelerate future.ii
Internet video delivery and improve quality of
service through content distribution platforms. The changes noted in Internet traffic patterns
This opens up new monetization opportunities and OTT video reflect the customer’s desire for
for cable operators with both content owners convergence. Customers want access to endless
that are willing to pay for guaranteed service varieties of content on multiple devices, or
levels, and customers who can choose from a “anything, anytime, anywhere.” However, due
variety of tiered service packages. to high levels of congestion, service providers
can’t keep up, and customers are experiencing
ever lower levels of quality of experience
MARKET OUTLOOK (QoE), leading to customer dissatisfaction and
increased churn.
Growth in OTT Traffic

Content is traveling through the Internet in OTT Delivery

ever growing quantities. The most ubiquitous
segment is OTT video, which is increasing by Because of the importance of the user's video
volume and quality. The growth of video on the experience, QoE is becoming a number one
Internet is so marked it is predicted that very priority for content providers. Content providers
soon it will account for the vast majority of all have traditionally been using the services of
traffic. CDNs (Content Distribution Networks), such as
Akamai and Limelight, in order to expedite the
A recent white paper from Cisco showed that delivery of content and ensure high levels of
Internet video sites, such as YouTube, QoE.
In recent months, OTT video providers, such
as the BBC and NBC, are using peer-to-peer

2008 NCTA Technical Papers - page 59

(P2P) or hybrid P2P/CDN technology to OTT video.
distribute high quality content to their
customers. P2P is helping them to 2. Charge content providers, including OTT
simultaneously reduce their distribution costs providers, for the delivery of content with
and scale their services to serve more customers assured QoE. In this context, MSOs are
with higher quality video content (flash complementing CDNs and providing service to
crowds). the last and most important segment in the
content delivery value chain: the end user.
However, while CDNs are paid for content
delivery, MSOs continue to deliver content 3. Establish mega video portals, thereby
across their networks without receiving becoming part of the video distribution value
compensation from content owners or CDN chain and leveraging their existing relationship
operators. with their customers. Under this model, the
MSO joins the OTT value chain through shared
It is clear that MSOs must find new ways to revenues with the content provider.
manage video traffic of all kinds through their
networks. 4. Generate advertising revenue sharing with
content providers for ad-supported content. As
cited in a recent report, the largest revenue
Pain with No Gain opportunity for online video will come from
advertising, which could reach $4.4 billion in
To meet the rising demand of OTT video, the US by 2011.iii Under this model, the
infrastructure costs for the MSO are steadily consumer receives the content at no charge, but
increasing, while the revenues that MSOs can advertising has been added to the content by the
enjoy from their subscribers as a result of this MSO. Advertising can be general or
investment is partial at best. personalized to specific user groups.

Because of this changing landscape, 5. Introduce a host of services that can

monetization of OTT video has become a must increase the average revenue per user (ARPU)
for MSOs. based on subscriptions or advertising. These
new services could include nPVR, catch-up TV,
and so on.
THE OPPORTUNITY
6. Sell tiered services to users, which also
From Pain to Gain increases the ARPU. MSOs can offer enhanced
delivery over specific time periods for special
In order to close the gap between rising deals. For example, a customer can purchase a
expenses and decreasing revenues, and to movie for download for $3 and pay $1 extra for
capitalize on market opportunities, MSOs can a faster download, or access their favorite OTT
adopt a combination of the following business video with accelerated service for an extra $10
models: per month. This service improves the customer
QoE and reduces churn.
1. Manage the surge of OTT video on their
networks, without hurting the user
experience and creating net neutrality issues, by
implementing caching platforms which support

2008 NCTA Technical Papers - page 60

A New Content Delivery Infrastructure

To actualize these opportunities, MSOs need MSOs need a solution that can deliver
to implement an infrastructure that will: content in both the open and the private
network. While open Internet traffic travels on a
 Cost-effectively manage and deliver their best efforts basis, using a private network, the
OTT video over the network while maintaining MSO can provide content providers/OTT
QoE, including high definition TV (HDTV). partners with quality of service guarantees and
enhance the services it offers its customers.
 Enable users to view content on all
devices, whether on their TV, home computer or
portable. THE CDP SOLUTION

 Support monetization schemes for content The content distribution platform (CDP)
delivery, such as local advertisement injection. solution is designed to meet the needs and
requirements of MSOs in the delivery of rich
In essence, MSOs need their own internal media content through their networks and
content distribution platforms (CDPs), housed enable monetization opportunities.
within their existing network. Today’s CDNs
are deployed outside the MSO network and can The CDP is an edge content delivery
only deliver content in the open Internet platform deployed within the MSO’s network.
environment, where they cannot guarantee
quality of service inside the MSO network.

The following figure illustrates the overall

CDP concept:

Figure 1: CDP Solution

2008 NCTA Technical Papers - page 61

 CDP Conceptual Design

The CDP solution consists of a two-tiered The CDP is based on the smart deployment
content delivery network: of a robust and scalable caching and
acceleration system. A proven way to ensure the
1. A best effort freeway, which is today’s delivery of popular, high quality content is to
open broadband connection for P2P, user- cache it close to the user. The caching of content
generated content (UGC), and other OTT traffic, avoids randomly created network bottlenecks,
that provides equal (net neutral) service on a saves on bandwidth and ensures prompt content
best efforts basis to all incoming traffic. delivery upon request.

2. A QoE tollway, which is designed as a The CDP is a multi-protocol caching

gated garden that allows for service level platform that provides all of the necessary
agreement (SLA) based delivery of premium functionality from content collection and smart
content for users and OTT content providers. caching to the delivery of huge quantities of the
cached content. The CDP supports all of the
Both tiers are supported by the CDP major and most popular protocols that are being
infrastructure at different service levels, which used for content delivery today and anticipated
is the key to both a rational network/traffic in the future, including: P2P, HTTP, RTSP, etc.
management model and an effective content
monetization solution. The following diagram illustrates the CDP
overall architecture:

Figure 2: CDP Conceptual Design

1. Content Ingestion Unit (CIU): Located

A CDP typically consists of the following within the MSO or at the Content Provider (CP)
main units in its architecture: facility, the CIU enables content providers to
upload content and define content-related

2008 NCTA Technical Papers - page 62

policies. Uploaded content has business data  Modular architecture that supports
attached to it, which describes the business multiple access protocols, including HTTP,
model between the MSO and the CP. RTSP, P2P, etc.

2. Content Caching Unit (CCU): A caching  Modular solution: Enables gradual

unit that is characterized by strong bandwidth transition from single multi-server node to
generation capacity (it can deliver the same managed network of multi-server nodes.
content to many concurrent users
simultaneously). Once content is ready for  Virtual CDP architecture enables
caching, it is uploaded into the CCUs. CCUs are providers to offer managed CDN services to
also typically installed in the access network third parties.
since they need to support high demand
delivery, such as streaming content. The SUMMARY
subsystem keeps track of all of the content
handled by the system using smart caching To avoid becoming dumb pipe operators, and
technology and is responsible for sourcing and carrying the costs of delivering OTT video,
delivering content. MSOs must adopt strategies that enable them to
become part of the OTT video distribution value
3. Advertising Ingestion Unit (AIU): chain.
The AIU enables the CDP to deliver content and
advertisements in accordance with policies A CDP deployment enables the cable
established in the external system. The unit operator to monetize OTT video by:
interfaces with external advertisement control
systems or ad networks. The CDP’s ad-insertion 1. Generating revenue from video traffic
system is located close to the end user. passing through its network. MSOs can charge
content providers for hosting and delivering
their content.
4. End Unit Client (EUC): This optional
MSO branded client is installed at the
2. Delivering virtual video portals for their
customer’s premises, providing branding
customers, using a pay-per-view model or
opportunities and advertising functionalities and
subscriptions.
enables quality of service guarantees.
3. Participating in advertising revenue
Functional CDP Requirements
sharing with content providers for ad-supported
content directed at an MSO’s existing customer
The CDP solution meets the following key
base.
functional requirements:
4. Selling tiered services and different
 Works in managed and unmanaged
advertising/subscription packages to customers,
environments, and gated garden (B2B) modes.
thereby increasing the ARPU.
 Scalable: Supports millions of users and
5. Taking advantage of personalized
assets.
advertising opportunities through the MSO’s
own content portal.
 Personal: Per-subscriber SLA and
accounting/charging.

2008 NCTA Technical Papers - page 63

 Additional MSO benefits of a CDP
deployment include:  Improved QoE for end users generated
through accelerated content delivery and a better
 Full control over video content. overall experience of all online services, which
will ultimately increase the MSO customer base
 Reduced MSO bandwidth costs. and reduce churn.

i
Statistics from Figure 4 from The Exabyte Era
White Paper (based on the paper: Traffic
Forecast and Methodology 2006-2011), Cisco
Systems, 2007.
ii
Joint Declaration of the Broadband Services
Forum, January 2008.
iii
Report: IPTV Competitors are Over-the-Top
(Quoting James Crawshaw’s report: Internet
TV, Over-the-Top Video & the Future of IPTV
Services, Heavy Reading), Ryan Lawler, Light
Reading, June 28, 2007.

2008 NCTA Technical Papers - page 64

 IMPLEMENTING ADDRESSABLE ADVERTISING IN LINEAR NETWORKS
Steve Riedl, Principal Architect, Time Warner Cable
Doug Jones, Chief Architect, BigBand Networks

Abstract  Instantiation of an overall control

mechanism to coordinate these activities –
Although the potential for addressable the SCTE DPI committee has begun the
advertising to increase revenues is widely process of developing standardized interfaces
accepted, the infrastructural changes needed to to meet this need.
support it are still being examined. Existing
network functionality needs to be enriched to The delivery of advertisements should include
allow the following abilities to be implemented: extensions of switched digital video known as
microcast and unicast, and set-top based ad
 Replication of program streams such that insertion mechanisms. These methods are
more than one copy can be simultaneously complimentary because content can be delivered
supported on the network – since bandwidth- either over the network or from a local hard disk
intensive HDTV is increasingly being offered drive in the set-top box. This paper unifies these
to subscribers the ability to place its topics, describes an architecture designed to
replication closer to the edge of the HFC satisfy the requirements listed above, and
network could conserve IP distribution explains its operation. The intent of the authors
network capacity; is to prove a technology framework for
 Managing the multiple copies within the converged resource management and
current systems without causing issues with addressable advertising for emerging marketing
the large increase in sources; services.
 Implementing IGMPv3 to insure only viewed
networks are distributed on the IP sections of The paper also provides relevant context on
the operator network; the addressable advertising service including an
 Association of a particular program stream architecture for a next-generation advertising
with viewers of like demographics to enable system. The authors show how ads can be stored
relevant ads to be delivered to those viewers and how the developing standard interfaces will
– grouping these viewers can be enable next-generation ad delivery systems.
accomplished either when theviewer initiates Additionally, it identifies a variety of parameters
a channel change or by force-tuning that that can be used to size the advertising system by
viewer before the ads are to be played; leveraging projections of initial deployments.
 Addressing of program streams to a single
user in order to deliver a particularly
relevant ad to them;
 Selective insertion of advertising into a INTRODUCTION OF ADDRESSABLE
program stream to match viewers’ ADVERTISING
demographics – ad insertion can occur either
through seamless splicing or a playlist; Addressable advertising is the selective
 Interoperability between the HFC resource insertion of advertising into a program to match
management system and the advertising viewers’ interests, thereby making that
decision manager to ensure that the available advertising more relevant to the viewer. Ad
capacity of the HFC network is taken into relevancy is important, because if presented with
consideration; a choice many viewers would likely opt-in to

2008 NCTA Technical Papers - page 144

minimize or eliminate certain types of homes. Commercial databases are available
advertising that does not match their interest; for which will provide specific demographics for the
example, parents without kids might prefer not to households within all ZIP+4 areas and
see another baby diaper ad. advertisers already use these for mass mailings.
This advertising principle can easily be carried
With addressable advertising, sets of viewers over to cable if the proper advertising
can be targeted at home watching a program like infrastructure were in place to provide
ESPN’s SportsCenter and delivered different ads advertising to just a few dozen homes, or the
more relevant to there interest at precisely the viewers associated with those homes.
same time. For example, one set of viewers who
prefer full-size high-performance vehicles can be Prizm Codes are associated with several
shown a commercial for the new Chevy Blazer demographic parameters. There are 66 defined
and another set of viewers who prefer mid-size Prizm Codes and one or more can be assigned to
fuel-efficient vehicles can be shown a each cable customer. In fact, one of the 66 Prizm
commercial for the all-new Chevy Equinox. This Cluster Codes is already assigned to every
type of viewer ad targeting was pre-planned by address in the U.S. Prizm Codes are based on one
the advertiser and tools are becoming available in of four urbanicity categories which is determined
the network to make it a reality based on by the population density of an area and it’s
aggregate geographic, demographic, neighboring areas. Within each urbanicity
psychographic or other characteristics of the category, segments are further sorted into groups
consumers residing within specific areas. based on affluence, another powerful
demographic predictor of consumer behavior.
Making advertising more relevant to viewers
can have several benefits. First, viewers will CABLE PRIVACY ACT
come to see advertising as part of the
programming, and will most likely be happier A discussion of addressable advertising
and more interested in viewing it. Additionally requires an up front disclosure on the
the relevancy will become more valuable to the implications on customer privacy. Addressable
advertisers making cable programming a better advertising implies a knowledge about a
place to spend their advertising dollars. Finally, customers demographics and with all the
intelligently choosing advertising benefits the information available about consumers, a major
advertiser by minimizing the placement of issue in addressable advertising is that cable
irrelevant advertising in front of viewers. operators must comply with some very strict
privacy guidelines, including the Federal Cable
While there are many methods to address Privacy Act of 1984 and other specific laws
advertising to viewers, this section will introduce passed by individual states. Nothing in this paper
two widely available methods; ZIP+4 and Prizm should be taken as legal advice. Always check
Codes. Both methods are based on the premise with your company privacy attorneys before
that "birds of a feather flock together," and that implementing any system.
people with similar demographic traits tend to
behave in the same way in the marketplace.

ZIP+4 codes are specific from a single to a

few dozen households, which clearly is more
specific than the average cable service group
which can pass several hundred to a thousand

2008 NCTA Technical Papers - page 145

 TODAY’S CABLE ADVERTISING the typical Lifetime Movie Network viewer.
LANDSCAPE Others networks like CNN attract a wider
demographic of viewers. Even a network that
This paper primarily focuses on Linear attracts a niche demographic will have some
programming but many of the concepts will also variation and if third-party data can be used, like
apply to On Demand programming. Linear if the car lease is expiring, then targeting on
programming is traditionally a broadcast those networks has real value.
medium. On a cable plant, broadcast bandwidth
is a very precious resource. Narrowcast Today the most sophistication associated with
bandwidth can be created by reducing node size advertising is geographic zoning, where
and/or adding additional QAM bandwidth to a customers in one part of a city will see different
node by removing analog networks. Once the ads than customers in another part of the city.
node size reaches about 500 tuners, then it is The degree of geographic zone subdividing
usually more economical to expand the number within the cable system would generally be
of QAMs per node. measured in tens of square miles associated with
different regions of a city. Figure 1 shows an
Since unicast services such as VOD and example of carving a city into five geographic
multicast services such as SDV share this zones.
narrowcast bandwidth, an Edge Resource
Manager needs to be put in place to allocate
these resources. This paper show how this
service plays an important role in utilizing this
bandwidth for advertising purposes.

Not all channels are insertable; it generally

depends on the programming contract. There are
about 60 ad-insertable networks today. Most
operators insert on 40 channels, four 30 second
spots per hour is normal – and only 12 hours of
the day matter. So while there are about 4,000
insertion opportunities each day, only about
2,000 are useable for any significant revenue.
The top ten networks provide about 80% of the
ad revenue and the top 20 networks provide over Figure 1 – Geographic Zoning
90% of the revenue.
To deliver this type of zoning, the ad-
Given that cable local / spot advertising is insertable programming is replicated, generally
estimated to be a $5 billion business in 2008 in the headend, and each copy of the
(NCTA), the last 10% is not small change at programming is sent through equipment which
$500 million using current methods. By making inserts ads for a particular geographic region.
that smaller number of eyeballs addressable, Then that programming is distributed to the
measurable and interactive, this 50% of the avail appropriate geographic zones and is delivered to
space could bring in well over $5 billion. the viewers.

Some programs are addressed by their very Migrating to more personalized addressability
nature. The typical ESPN viewer is different than requires pushing the functions of program

2008 NCTA Technical Papers - page 146

replication and ad insertion further out to the processing can be moved out to the hub to
edge of the network, and coupling these manage the metro network bandwidth utilization.
functions to a system which can make decisions If the transport capabilities are inexpensive
about which streams to which viewers to enough and can handle the required streams, then
personalize and how. centralizing the resources makes more sense. If
the transport is more expensive or just not
ARCHITECTURES FOR NETWORK-BASED possible, then functions can be pushed to the
ADDRESSABLE ADVERTISING edge, which is a key component of the emerging
system to be described following the rest of this
This section introduces both a method to do example.
addressable advertising with current network
infrastructure as well as an emerging architecture Figure 2 shows one logical flow through this
to deliver addressable advertising. architecture. The network signals are received off
the satellite which then goes to a stat-mux/splicer
The architecture being developed with today’s to be rate-shaped either as a feed for SDV,
network components for linear addressable network PVR or as part of a multi-program
advertising is based on modifications to the On multiplex. The splicer portion is used to insert
Demand system. Based on how VOD is deployed advertisements at every available opportunity.
today, most of the resources are centralized at
one or more super headends while other

Figure 2 – Linear Addressable Advertising Based on VOD Modifications

2008 NCTA Technical Papers - page 147

 These will typically be the advertisements for for creation of personal mosaics.
the largest geographic ad zone. By performing
this initial splice, the network stream is prepared The media services platform allows the
for easier content segmentation. This feed is then operator to grow the new addressable advertising
distributed onto the metro ring to the hubs to service on purpose-built equipment without
provide a constant reliable network feed source. impacting existing services such as VOD. In this
This processed feed also then goes into the On example, the media services platform ingests
Demand server complex for real-time program streams, national or local feeds which
acquisition. If additional ad zones are required, come with default ads, and has the capability to
real-time feeds can be set up out of the On personalize these streams by replicating them
Demand server and new ads can be inserted by and inserting advertising specific for the viewers.
using playlist techniques from the On Demand
server. The media services platform interfaces with a
personalization engine which is the decision
Figure 3 shows the emerging method to maker that instructs the media services platform
implement an addressable advertising system when and how to personalize streams. The
which is not based on the On Demand system. personalization can include inserting addressed
With this new method, the equipment is located advertising, or a particular enhanced
closer to the viewers such as in a Distribution programming or to create a mosaic. The
Hub. In this example, a new category of network personalization engine can make decisions on
component, the media services platform, is what streams to personalize based on the
capable of the traditional stream replication and household viewing of that content, their
ad insertion, but also more advanced forms of geographic location, assigned Prizm Codes or
personalization including managing bound any number of other factors known about that
applications (Enhanced Television, or ETV, and household including specific opt-in
OCAP bound applications) as well as the opportunities.
possibility

Figure 3 – Personalization System

2008 NCTA Technical Papers - page 148

 When either a customer changes channels or for personalization. It is only at peak viewing
an advertising avail comes up in a program times that the QAM service group can become
stream, the personalization engine needs to make full and if the operator wants to continue high
a real-time decision to decide if: levels of personalization during those times,
additional QAM capacity should be considered.
a) The program is ad-insertable;
b) There is an available campaign for this With a system as show in Figure 3, the
subscriber on this network; options for delivering addressable advertising in
c) Is there enough QAM capacity on the service linear programs are expanded from just
group to do something like support geographic zoning to include three basic types of
replicating the stream for just that one addressability which should cover the broad
viewer. spectrum of advertising models to be developed
in the foreseeable future; broadcast, groupcast
Separating the decision making from the and unicast linear ad insertion models.
actual delivery system allows for open systems.
The SCTE (Society of Cable Network-based addressable advertising uses
Telecommunications Engineers) DVS (Digital SDV (Switched Digital Video) as an enabling
Video Subcommittee) is developing an interface technology for several reasons. First, SDV has
standard, SCTE 130, which supports this every channel change request go through a server
separation of functions and will allow innovation which allows subscribers to receive the
to occur within both the personalization engine appropriate personalized programming each time
and media services platform while allowing they change channels. Additionally, tuning
operators to choose the components independent information from set-top boxes, including both
of each other. switched and broadcast networks, can be
collected and used to understand how to apply
Finally as shown in Figure 3, the addressable demographics to viewers.
advertising system has to interface with an ERM
(Edge Resource Management) system which is Broadcast is the current ad environment and is
used to manage the digital QAM bandwidth to addressed through the specific network, the
subscribers. Because the personalization system show, time of day, etc. As shown in Figure 4, a
can cause program streams to be replicated on a single version of the program stream is delivered
service group, it uses edge QAM capacity more to all viewers who all see the same ads. This
so than a geographic zoned ad system. As such, solution is the basis for today’s advertising
the personalization system needs feedback from business and cable is continually adding better
the ERM system as to how much QAM capacity audience measurement capabilities that will
is available for personalization. During most allow us to better utilize this segment.
parts of the day, the QAM service group is
underutilized and there is excess QAM capacity
available

Figure 4 – Broadcast Advertising

2008 NCTA Technical Papers - page 149

 Groupcast is a solution where viewers of like which each carry ads specific for that
demographic are “grouped” onto a common copy demographic. In this case there are copies of the
of a program which is then enhanced for them. program generated for demographics A and B
Take a program such as CNN which can have a and a third copy of the program is available for
wide variety of viewers, the personalization viewers which do not fit these demographics.
system would create several copies of CNN, one
for each demographic which an advertiser is Groupcast is very effective at demographic
trying to reach and each copy of CNN would targeting-based techniques such as ZIP code
then be personalized with ads for that particular or Prizm code. A household is either in a ZIP or
demographic. Prizm cluster and doesn’t move (at least during
the duration of the program). Groupcast is not as
An example of groupcast is shown in Figure 5 effective as unicast at hyper-targeting households
where there are three copies of the program for certain ads based on a specific household
stream created in the media services platform and (such as a specific credit card holder or not).
viewers of particular demographic are grouped
onto those copies

Figure 5 – Groupcast Advertising

Before a new groupcast feed is created for a frequent occurrence and the operator might
particular demographic, bandwidth availability consider adding QAM capacity to those
can be determined by polling the ERM. If particular service groups.
bandwidth is not available then the additional
program feed is not created which can result in a Unicast is a technology where a single
slightly lower yield, but requires no additional program stream is customized for a single
edge bandwidth spending. Reporting can indicate viewer, as show in Figure 6.
if this is a

Figure 6 – Unicast Advertising

2008 NCTA Technical Papers - page 150

 With unicast, whenever a viewer changes customers can watch as much programming as
channels, the personalization system has to desired.
determine if a new unique copy of that program
should be created just for the viewer. This The goal is to add addressable streams based
solution is narrowcast, where there is a stream on bandwidth, demographics and ad availability.
for each viewer and can be used to fill available At startup, the system sets up an initial stream for
capacity on a QAM service group. every insertable network and assigns a base
demographic to that stream. The system could
While unicast means that every subscriber also create a “spare” stream of every ad
could have a dedicated stream, it does not insertable network with no assigned
necessarily mean that they will have a dedicated demographic. This spare stream could be
stream, just that if their demographics were such switched onto the service group if the
that they matched a specific set of available ads personalization system determines it is
to play and bandwidth is available on the service appropriate to customize it for a new viewer (at
group, a custom stream could be created for that which time a new “spare” stream would be
single viewer. created). If on the other hand there is no
bandwidth available (or above a limit set for ad
Groupcast is a good example where the purposes) the new viewer gets tuned to an
system can dynamically deliver advertising using existing groupcast which best fits their
both broadcast and unicast, as well as the specific demographics.
case of groupcast. When used fully as a
bandwidth saving multicast service, groupcast SDV ADVERTISING PHASED
resembles broadcast. When QAM capacity is IMPLEMENTATION APPROACH
available, the demographic parameters used to
replicate programs can approach unicast mode The following is a proposed phased approach
where it may be possible to create completely to addressable advertising based on groupcast
customized programs for each viewer. using switched digital video. The first three
phases can be done with existing equipment and
The migration to unicast needs to be managed small modifications to the SDV session manager.
by taking advantage of available service group Note that in this example transition the edge
capacity, as more QAM capacity is added over QAM resources will be managed through the
time. Unicast does not have to imply that a SDV system to ensure the additional copies of
separate stream is available for every viewer. programs do not overflow the QAM resources
Unicast may mean that separate streams are available on the service group. A more detailed
available for only some viewers. For example the discussion on QAM resource management is
QAM capacity of a given city may support presented following the phased implementation
80,000 narrowcast SDTV slots, but the peak TV approach.
viewing population of that city may surpass that
at certain times of the day. When this happens, it Step 1 – Study the current ad insertable
is simply not possible to deliver all unicast networks to determine which demographics are
streams and some blend of broadcast, groupcast needed. Also learn which demographic groups
and unicast will be most efficient. It is the watch the networks and during what times. For
personalization engine which makes these example if the goal is to personalize
decisions to best optimize the delivery of programming for Senior Citizens, don’t advertise
advertising and personalization while ensuring on Nickelodeon.

2008 NCTA Technical Papers - page 151

 there is an available ad campaign. Unicast
Step 2 – Use the data from step 1 to pick a advertising most closely resembles the Internet
small number of networks that typically have advertising model and promises the greatest
four demographics watching them. For a revenue per ad.
deployment of this scope, the SDV server can
perform the demographic selection based on a With respect to QAM service group
look-up table with set-top box IDs, networks and bandwidth, the following group of figures show
the demographic groups. Ad selection in this the utilization of the QAMs in a typical SDV
case can be done with current traffic and billing service group over a week, note these figures do
systems by generating four schedules or, the not yet include addressable advertising but they
system could choose to use an SCTE 130-based can still be instructive to understand how viewers
system. Alternatively in Step 2 one could use are using the linear lineup. While only about 50
fixed demographics such as ZIP+4 or political networks are ad inserted, typically 150 or more
party affiliation (democrat, republican, networks are carried and only a percentage of
libertarian, unknown) as a stable demographic viewers are watching ad inserted networks that
selector. would use the additional bandwidth if available.
The non ad inserted networks would always use
Step 3 – Allow the demographics on any multicast and only one version per service group.
given network to change during the day to better
represent both the viewing patterns and the Figure 7 shows how the amount of traffic on
available ad inventory. This additional degree of the switched tier is allocated between unicast
flexibility will allow better addressability, albeit traffic (a single unique viewer on a program) and
the combinations can become more specific and multicast traffic (2 or more viewers on a
migrating to a true personalization engine may be program) and indicates that there is a significant
prudent at this time. percentage of time when there is but a single
viewer on a program which could provide an
Step 4 – Extend this groupcast example to all opportunity for unicast advertising without
ad-insertable networks. This will maximize the having to replicate an additional version of the
revenue by bringing addressability to all the program on the service group.
networks where the rights are available to insert
ads.

Step 5 – Extend groupcast to unicast when

Figure 7 – Percent Unique Streams per Service Group

2008 NCTA Technical Papers - page 152

 Figure 8 shows the number of active viewers Viewers can be reached at any time during the
on a service group. Note the periodicity of the day with addressable advertising; however, the
graph with peaks building up through during largest numbers of viewers are present during
prime-time each day. prime-time.

Figure 8 – Active Viewers per Service Group

Figure 9 shows the percentage of bit-rate used advertising uses more capacity in the service
on a service group during the day. Again note the group then otherwise switched digital video
periodicity of the graph with peaks building up would. Hence, overlaying addressable
through during prime-time each day. As noted advertising onto this service group would require
earlier in the paper, the stream personalization more bandwidth throughout the day
needed for addressable

Figure 9 – Active Bit Rate per Service Group

And when coupled with the information from engineered today. The addressable advertising
Figure 8, that more viewers are on the system system can make more efficient usage of service
during prime-time, it should be clear that to reach group capacity by filling up an otherwise empty
the most viewers with addressable advertising service group with revenue generating
that additional service group capacity will be addressable advertising during the times of the
needed. However, because there is extra service day when the QAMs might otherwise sit empty
group capacity available during off-peak times, it and un-used.
is possible to get started with addressable
advertising with a switched digital system as

2008 NCTA Technical Papers - page 153

 To more fully reap the benefits of addressable This viewer should get the demographic 7
advertising as the system begins to pay for itself version of the program. Since this is a tuning
with additional ad revenues, the operator can request for a program which is not already on the
grow additional QAM capacity into the service service group, the SDV session manager will add
group to deliver additional addressable that program with a demographic 7 version and
advertising during the times of peak service then tune the viewer to that program.
group usage. And once those QAMs are
available, they can be used for other types of Because the addressable advertising system
personalization as well including personalized interfaces with the edge resource management
guides and mosaics. system, the net effect of addressable advertising
on edge bandwidth therefore should be
In general, a working philosophy for edge effectively zero. By monitoring service group
QAM utilization is to pick a value that if the usage, if over-time there is not enough edge
QAM group is above the threshold, additional QAM bandwidth to run the scheduled ad
streams will not be used for advertising (unless campaigns, then the operator should consider a
the required network is not already in the Service bandwidth expansion on specific service groups
Group). The current number is believed to be where capacity is an issue. But at this point in
between 80 and 90% utilization. The following time, there should be definite revenue numbers
examples illustrate how the operator can get associated with the addressable advertising
started with addressable advertising with the which will offset the bandwidth expansion.
currently available QAM service group capacity.
SUMMARY
Examples
Cable local and spot advertising is currently a
Scenario 1: A viewer tunes to a program and $5 billion business, and it is widely accepted that
there is already a version of that program on that adding addressability will grow that number.
service group assigned to demographic 3, but the This paper presents an architecture for linear
addressable advertising system decides that this addressable advertising which builds upon
viewer should get demographic 7 in order to play existing linear splicing by adding Switched
different ads for additional revenue. The service Digital Video as a means of personalization.
group is currently only 20% full, so the SDV Using SDV allows the operator to offer
manager allows the creation of a new feed of the addressable advertising in a number of contexts
program assigned to demographic 7, places that including both switched groupcast and switched
program onto the service group and feeds the unicast.
tuning info to that viewer.
The paper identifies several areas where
Scenario 2: Same as scenario 1, except that planning may be necessary for addressable
the service group is now 93% full meaning there advertising including reviewing transport
is no additional capacity to create a new feed of network capacity and switched service group
the program. In this case, the SDV server will usage. Planning can help ensure that the service
join the viewer to a program that already exists meets the needs to better target advertising while
and most closely matches their demographic. giving the customer a better viewing experience,
more HDTV and a more user friendly guide
Scenario 3: The viewer selects a program experience.
which is not currently switched onto the service
group; however, the service group is 93% full.

2008 NCTA Technical Papers - page 154

INFRASTRUCTURE CAPABILTIES SUPPORTING CABLE’S NATIONAL PLATFORM
James Mumma, Sr. Director of Video Product Development, Comcast Cable
Doug Jones, Chief Architect, BigBand Networks

Abstract  Capability to dynamically add or drop

individual PIDs associated with bound
One of the major initiatives for the cable programs;
industry is the introduction of functionality  Protocol interfaces to manage the
giving subscribers opportunities to interact manipulations of identifiers associated
with applications and services through their with bound programs;
televisions. Doing so will enhance viewing  Interoperability between the HFC
experiences, usher in new revenue resource management system and the PID
opportunities and provide competitive insertion function to account for the
differentiation to satellite broadcasters and additional bandwidth used on a QAM as
the telephone companies. bound applications are managed;
 An overall control mechanism to
The ETV and the tru2way family of coordinate the management of bound
specifications available at CableLabs applications with programmers, both
describe how applications can be bound to national and local.
programming allowing cable to deliver a
national platform for advertising and other With a proper management framework
services. While there are industry bound applications will provide both a
specifications for delivering bound platform for national services as well as
applications to a set-top box, there are no personalized services. The ETV and OCAP
specifications defined on the infrastructure toolset provides for a plethora of services, but
capabilities needed to manage these bound the management and control architecture
applications. This paper proposes a technical needs to be designed in order to achieve the
architecture and capabilities that can be used full potential for innovation of which it is
to manage and deliver bound applications (in capable. The authors examine the
both ETV and OCAP formats) capable of requirements associated with management
providing operators with a flexible platform and control and explain how present
for advanced services delivery. capabilities can evolve to satisfy them.

ETV and OCAP applications are bound to THE NATIONAL PLATFORM

individual programs by carrying those
applications on MPEG-2 PIDs (Program Until recently, implementing interactive
Identifiers) that are included along with the subscriber services on a national basis was not
programming. There can be multiple PIDs feasible due to the lack of implementation
associated with a bound program and the standards in the cable industry. The splintered
paper proposes a flexible architecture to approach of proprietary technologies was cost
manage them. These include: prohibitive for content providers and
distributors. Today, with new and emerging
 Passing bound applications, which specifications such as ETV, which includes
include extra PIDs, through headend the Enhanced TV Binary Interchange Format
equipment; (EBIF) and the tru2way, which includes the
Open Cable Application Platform (OCAP),

2008 NCTA Technical Papers - page 30

specifications developed by CableLabs, the bound applications garner the most interest
proliferation of interactive TV in North from content providers because of the
America is closer to becoming reality. economic potential. Currently bound
However deployment challenges still remain applications can be supported in linear and
for cable operators. VOD content and will be supported in time-
shifted content in the near future.
The benefits of interactive TV to
subscribers, programmers, advertisers and Unbound applications are not associated
cable operators is mutual, as figure 1 shows. with specific programming and are not
The shared benefits that interactive TV offers implemented as part of a programming
each of these key stakeholders provides fertile service; they are, instead, resident in the set-
ground for a nascent ecosystem, with the top box and can be run regardless of the
potential to improve the viewing experience, programming being watched. Examples of
while driving new corporate revenue streams. unbound applications include the standard
guide or CallerID to the TV, both of which
can be rendered on the TV regardless of the
currently tuned linear broadcast.

The delivery of applications, specifically

bound applications, remains in its infancy.
The CableLabs’ specifications define how the
bound applications should be interpreted at
the set-top box but stops short of governing
Figure 1 – Enhanced Programming Flow supporting infrastructure components;
justifiably so, since this area needs to evolve
Interactive TV application will come in to ensure the most efficient and innovative
two forms: ways to manage bound applications.

 Synchronous, or bound to the The focus of this paper is a suite of

programming; capabilities that will support the national
 Non-synchronous, or unbound to deployment of interactive TV by streamlining
programming. specific components that enable the
localization and personalization of bound
Bound applications are associated with interactive TV applications.
specific programming. While a subscriber is
watching a program, the ETV application will BUSINESS DRIVERS OF
run, resulting in an enhanced viewing INTERACTIVE TV
experience. A bound application provides the
ability to interact with a program or with Interactive TV offers advantages to key
advertising. For example, a subscriber would stakeholders including subscribers, content
be enabled to vote during a reality TV providers, programmers, advertisers, and
programming directly through the use of the cable operators through interaction,
remote control rather than through a personalization and localization of content.
secondary device such as a phone or PC. In
another scenario a subscriber can request Subscribers
more information about a product or service
promoted during an advertisement. While One does not need to look far to find
clearly enhancing the subscriber experience, evidence of subscribers’ interest in

2008 NCTA Technical Papers - page 31

participating, shaping and enhancing content. additional information to the subscriber.
Personalized websites, video outlets such as Consider the local car dealership that can
youtube.com and facebook, and the popularity promote its latest campaign during a national
of reality TV in which subscribers can advertisement for the car chain by ‘piggy-
determine the outcome of the programming backing’ onto that national advertisement.
align with the principles that interactive TV Interactive TV also provides incentive to
will enable. The adoption of these, and subscribers to watch time-shifted
similar, phenomena indicate that subscribers advertisements once this functionality is
should quickly embrace interactive TV as supported.
well.
As the CableLabs specifications are
Programmers adopted and implemented among MSOs, an
unrivaled national platform will emerge,
With increasing alternatives to broadcast providing advertisers and cable operators
TV, programmers continue to vie for alike with a robust opportunity.
‘eyeballs’ of live programming. Interactive
TV differentiates their programming and Cable Operators
strengthens their brand with enhancements.
Straightforward opportunities to enhance the Since subscribers, programmers and
programming with bound applications include advertisers can all be counted as customers or
the capability for voting and trivia questions partners, cable operators will benefit as the
while providing near real-time feedback about enabler of interactive TV. They can capitalize
how other subscribers responded at either the on their scale versus that of DBS and/or telco
national or local level, thereby creating a providers as well as their established
sense of community. relationships with programmers and
advertisers. In addition, operators can use
Additionally, programmers can use enhanced advertisements to promote their
interactive TV to keep subscribers ‘on brand’ own offers and services. For example, a cable
before tuning away to alternative networks by operator could enhance a linear promotion to
offerings opportunities to view their VOD telescope to a VOD clips to learn more about
content, perhaps at the conclusion of program, the available On Demand services, how to
or even offer content provided by sister interpret their cable bill, or even sign up for a
networks (i.e., NBC, Bravo, USA). The service offering directly from their TV.
programming community is well-positioned
to provider subscribers with personalized and Each of these key stakeholders shares a
localized experiences that are compelling and common ecosystem that powers the television
difficult for the competitors to mimic. business today. The benefits afforded by
interactive TV across these key stakeholders
Advertisers provide a recipe for the broad adoption and
consequent success of interactive TV that will
As broadcast and cable advertising introduce a new era in the TV viewing
continue to be threatened by internet or experience.
mobile device alternatives, interactive TV
provides advertisers with an effective UNDERSTANDING THE
response. Advertisers can enhance ads to SPECIFICATIONS
fulfill requests for information about products
or by using interactivity to telescope directly Within a proper framework, bound
to VOD clips about their products to provide applications can provide both a platform for

2008 NCTA Technical Papers - page 32

national services as well as individually standard MPEG-2 transport stream. These
addressed services. But in order to architect include two new descriptors in the MPEG-2
the right framework one must first fully Program Map Table (PMT), an EISS Table
understand the applications, the available (ETV Integrated Signaling Stream) containing
specifications and how they affect the existing applications signaling and timing information
architectures. and a Data Carousel for carriage of the
application itself. When the receiver tunes to
The industry specifications for ETV and the transport stream that contains the ETV
tru2way are developed by CableLabs. This set application, the receiver reads the PMT and
of specifications provides a basis for product determines there is an ETV application
interoperability. The specifications were present and alerts the ETV User Agent to run
designed to be non-proprietary and open in the application.
order to support the national reach for the
platform. The guiding principle varies little ETV is well beyond lab testing and has
from the ‘write once, run everywhere’ model entered field trials that are important to
common in computer programming today. validate the technologies in use. ETV is
generally recognized as a "fast track" item,
Enhanced TV garnering deployment priority as soon as
feasibly possible.
ETV provides a way in which interactive
TV applications may be deployed to legacy tru2way
set-top-boxes (STBs), such as the Motorola
DCT-2000 and Scientific-Atlanta Explorer tru2way has several components, including
2000. Since it is estimated there are millions a host specification, CableCard interface
of deployed legacy STBs, ETV was created to specification and the OCAP middleware
allow operators to deploy interactive specification. Middleware is software that
applications across this large footprint of provides an interface between applications
STBs. ETV applications will also run on and whatever system software a manufacturer
tru2way host devices. chooses for a host device. The middleware is
based on the widely accepted Java™
It is important to note that ETV is technology. By abstracting away the various
supplemental to tru2way. In fact, an ETV consumer electronics device operating
User Agent could be implemented as a systems to a common set of middleware APIs,
tru2way application to support ETV application developers can write an
applications on tru2way hosts. application only once and it will run on all
models of tru2way devices. Cable subscribers
ETV applications are set-out in an EBIF, with tru2way-enabled digital televisions, retail
for use in decoding and rendering ETV set-top boxes, and other interactive digital
constructs on the TV screen. Applications cable products will be able to receive all of
consist of a collection of one or more the cable operator's services just as if the
partitions containing resources and subscriber was leasing a comparable set-top
programmatic data. ETV applications are box from the operator.
interpreted by a User Agent resident in the
set-top box. On a tru2way host, the ETV User In terms of transmission from the headend,
Agent is a bound application. tru2way bound applications require similar
additions to a standard MPEG-2 transport
In terms of transmission from the headend, stream as does ETV. The application and the
ETV requires a number of additions to a data files that it accesses are packaged into an

2008 NCTA Technical Papers - page 33

Object Carousel (OC) format, which is an The applications originators are the
extension of the MPEG-2 transport programmers and advertisers who choose to
environment that exposes a file system to the enhance their programming. These groups
device at the other end of the network. In may create the application in-house or
addition, an Application Information Table outsource the application development to
(AIT) is required to tell the tru2way host both third-party developers.
that there is an application present and where
to find it. This collection of files is then Third-party developers often provide
multiplexed into the MPEG program stream. authoring software tools to create the
When the receiver tunes to the transport enhancements for the bound applications. For
stream that contains the application, the the purposes of this paper, the word
tru2way system reads the AIT and launches “enhancement” refers to an MPEG-2 program
the application; if the receiver tunes away that contains an ETV or OCAP application.
from the service, the application is terminated. The authoring tools generally include easy-to-
use interfaces and simulation tools to aid in
OVERVIEW OF THE the design and development process.
NATIONAL PLATFORM Ultimately, these tools are used to put the
applications in the correct format to be
While specifications exist for the client considered compliant to the industry
(i.e., how the set-top box is supposed to specifications.
receive and handle bound applications), the
server-side infrastructure requirements are With bound applications there will be
more loosely defined. This is analogous to the variations on how the enhancements will be
VOD infrastructure in which there are passed from a programmer to the headend and
numerous variations on how the service can on down to the market level. In addition to
be deployed and managed. Like VOD, this enhancements being originated directly from
scenario presents opportunities and challenges the programming studio, the ecosystem
to the broad deployment of a national includes programming distributors such as the
interactive TV platform. The goal is to create Comcast Media Center (CMC), TVN
an infrastructure framework that supports the Entertainment, Headend In The Sky (HITS)
national platform using industry specifications and others. Additionally, there is always the
for defined interfaces. possibility to swap enhancements at the local
level either in the headend or deeper in the
In creating this framework, there are at network, closer to the consumer.
least five different technical components of
the National Platform that must be considered, Another component of the National
as shown in Figure 2. Platform is the stream generators which put
the applications into the proper format to be
included with a digital program stream and
the devices which actually place that
enhancement into the program stream so it
can be delivered to a digital set-top box.
These stream generators will need to develop
the capabilities to add and drop different
enhancements based on the needs for
localization. These stream generators need to
Figure 2 – Enhanced Programming Ecosystem adhere to a number of industry specifications
(e.g., SCTE-130) to ensure the digital

2008 NCTA Technical Papers - page 34

program streams are of the proper type and enhanced ads into either enhanced or non-
format to be decoded by a set-top box. enhanced programming. In addition, the
coordination of the control of bound
Finally there are the digital set-top boxes applications and local enhancements needs to
which need User Agent to read the interactive be coordinated between the programmer /
signals in the broadcast stream. The User advertiser and the infrastructure. These
Agent is the software programs in the set-top additional control interfaces need to be
box that interprets the interactive applications. developed to support more sophisticated use
of the tools. This includes a general
CHALLENGES OF THE operational readiness from a business and
NATIONAL PLATFORM technical level between the operator and
programmer or advertiser.
Broad deployment of the National
Platform faces many challenges surrounding Bandwidth Management
the management and control of interactive TV
applications. Some of the issues to be Adding enhancements to the programming
considered are listed below. has the effect of increasing the amount of
bandwidth needed for that program. Since
Data PID Integrity those extra PIDs associated with the PMT
carry data, those extra bits and bytes need to
Data PID integrity includes successfully be accounted for by the infrastructure to
passing and maintaining bound applications ensure the complete bit rate through a QAM
through the National Platform without adverse modulator does not exceed the capacity.
impacts. The bound applications create
additions to the PMT associated with that Return Path Capacity
program which needs to be handled by a
number of pieces of equipment in the Since the programming enhancements rely
infrastructure. on interactivity, the capacity of the return path
needs to be managed. If the enhanced
Data PID Control program is widely viewed there can be bursts
of activity when all those subscribers respond
The capability to dynamically manage to an enhancement. The goal here is to not
individual or multiple PIDs, including the overwhelm the return path. Legacy boxes will
ability to administer and manipulate use the existing back-channel which is
identifiers associated with the bound relatively low capacity compared to newer
application, is key. If there is no business technologies such as DOCSIS®/DSG. Since
agreement between the programmer and the there are different tiers of set-top boxes, the
operator, the infrastructure needs the enhancements can be different and look better
capability to recognize and remove the with more advanced boxes with a higher
enhanced PIDs from the programming. The capacity back-channel.
infrastructure needs the ability to insert non-
enhanced ads into enhanced programming to Data Collection & Reporting
support regular advertising capabilities.
Since interactivity is managed across a
Localized Operations national platform, there has to be a method for
that interactivity to be aggregated on a
To support enhanced advertising, the massive scale. For example, some popular
infrastructure needs the capability to insert programs can garner tens of millions of

2008 NCTA Technical Papers - page 35

concurrent subscribers and if a significant nationally originated programming.
portion of the subscribers vote
simultaneously, the ensuing avalanche of data Figure 3 proposes several new components
needs to be handled, aggregated and acted to the local cable infrastructure including a
upon in a scaleable and quick fashion. new category of edge, video processing
platform, known as a media services platform.
A PROPOSED ARCHITECTURE This platform interfaces to the personalization
engine.
Bound applications can be inserted into
programming at either the national or local Media service platforms are responsible
level, including personalizing the bound for personalizing streams for subscribers and
application at the local level. Therefore, a can selectively insert specific ads into the
dynamic and flexible architecture is required programming, and specific bound applications
to manage, control and deliver these services into both the programming and advertising.
while accounting for national and local The media services platform will offer the
footprints. data PID control interfaces as well as ensure
data PID integrity for the program streams.
Figure 3 shows an architecture which Since personalization is done based on
supports the National Platform including specific subscribers of the programming, it is
personalization at a local level. National best done toward the edge of the local cable
programming originates on the left side of the network, as close to those subscribers as
figure, including the insertion of national possible.
advertising. At this point both the
programming and the advertising can include The bandwidth management function will
enhancements. When the programming be handled by the “last mile” network. The
reaches the local cable operator (on the right assumption for cable is that this last mile is an
side of the diagram), local personalization HFC network hence the Edge QAM plays a
systems can further modify the enhancements significant role in bandwidth management,
and direct the programming to groups of ensuring that the enhanced programming does
subscribers and potentially even individual not overrun the capacity of the QAM channel.
subscribers. The local personalization is There are other last miles networks, including
implemented in conjunction with the wireless, where the wireless access point
programmer or advertiser who wants to craft a would have the responsibility to ensure the
custom experience for their subscribers which wireless channel is not overrun.
otherwise would not be possible with just

Figure 3 – Proposed Architecture

2008 NCTA Technical Papers - page 36

 The personalization engine makes This figure represents how a program
decisions about what personalizations should delivered to a customer can have both local
occur in the programming. The and national enhancements for both the
personalization engine has separate interfaces program itself and advertisements associated
to both the programmers and advertisers to with that program. Note that the
make close to real-time decisions about how enhancements associated with the program are
to personalize programming based on available during the program, but not during
available campaigns, operator business the advertisement where different
agreements and viewership. enhancements can be available. The media
services platform must not only be able to
These personalization decisions are passed insert the correct enhancements, but must
to the media service platform over the DVS enforce the boundaries between the program
629 interface. DVS 629 has recently been and advertisements to ensure that the proper
developed by the Society of Cable enhancement is presented to the proper
Telecommunications Engineers Digital Video subscriber, with the proper enhancement at
Subcommittee (DVS) specifically for the the proper time during the program or
purpose of personalizing programming. It is advertisement.
expected that DVS 629 will be ratified into an
SCTE standard later this year and then be Considerations
known as SCTE 130.
Bound applications are actual software
Finally, the figure shows the data programs and data associated with a TV
collection and analysis function. Since the program or advertisement. The bound
programming is interactive, the user responses application is inserted either at the national
need to be collected, aggregated and passed origination of locally, and that application is
back to the programmers and advertisers. The run in the set-top box. Technically the bound
analysis can be either real-time, such as applications are inserted into the MPEG-2
voting which can be provided as feedback program stream which represents that
during the programming, or non real-time if program. The bound applications are inserted
there is no impact on the current as additional data on specific PIDS associated
programming. with the digital programming.

The cable industry has been working Since these bound applications represent
toward personalization and the DVS 629 real data, the media services platform must
interface allows separation of the account for them as they are multiplexed onto
personalization engine and the media services the last-mile network, which for this paper
platform to allow innovation to occur around could be considered a 256 QAM modulator
that interface. Figure 4 shows additional detail able to carry approximately 38.8 mbps of
around the insertion of enhancements to data. Traditionally the programming is
programming. statistically multiplexed to best fit within the

Figure 4 – Enhancements to Programming

2008 NCTA Technical Papers - page 37

38.8 mbps; the bound applications, however, CONCLUSION
represent additional bit rate overhead which
must also be carried by that QAM but which Interactive television offers the cable
cannot be statistically multiplexed. Clearly industry opportunities to improve subscribers’
inserting bound applications will impact the viewing experiences, reduce churn and
amount of bits through the QAM and this enhance advertising revenues. By providing
information will need to be signaled back to distinctive benefits to subscribers and content
the edge resource manager associated with the partners, each stakeholder will gain a
QAM resources in order to accurately account significant advantage as the ‘handshake’
for QAM usage. This information will allow among them is redefined.
the edge resource manager to efficiently
utilize the resources in that QAM. Cable operators can address the
infrastructure challenges described in this
Additionally, media services platforms paper by leveraging the architecture proposed
should be able to manipulate the Program by the authors. This can also provide the
Map Table associated with the program such potential to launch a migration towards a
that the additional PIDs associated with the more flexible national platform for
enhancements (both national and local) are personalized services and advertising.
accurately represented such that the set-top
box is presented with a valid MPEG-2 bit The necessary standard interfaces are
stream to decode. The integrity of the PMT becoming available, as is the equipment
and the MPEG-2 bit stream has to be needed to implement the required services and
maintained even though locally the functionality. At that point, the programming
enhancements can be added or dropped in experience will increasingly become limited
near real-time based on instructions from the only by the creativity of the application
personalization engine. developers. Such an evolution will usher-in
new viewing experience for cable subscribers.
The personalization engine will require
interfaces back to the programmers and
advertisers in order to manage the
personalization. These interfaces are yet to be
defined but could be considered a next phase
to the work being done to create DVS 629.

2008 NCTA Technical Papers - page 38

 IS IMS THE ANSWER?
Bruce P. McLeod
Cox Communications, Inc

Abstract seamless evolution has transpired because the

design of VoIP technology has followed a path
Is IMS the answer? The 3GPP IP of replication of telephone service to a handset.
Multimedia Subsystem is a topic of hot debate In the future, the traditional telephone handset
among technologists in the MSO community as will remain part of the Voice service network
to its validity as a service integration platform but it will not be the only interface as it has
and core technology. It is regarded by many as been for over 100 years. The voice interface of
a solution looking for a problem and by others the future may be a video screen, mobile PDA, a
as a panacea for simplifying the rapid utility within a web page, or possibly something
introduction of new service types that have that is difficult to imagine today. IP Multimedia
voice as a key component. This paper discusses Subsystem is a technology that lends itself to
the real world learning garnered by Cox integrating Voice to other application types. But
Communications during our technology is it the best answer for how to do this? This
research and prototype development beginning discussion analyzes the known requirements for
in 2006 and throughout 2007. Specifically voice services and how IMS addresses those.
addressed will be the strengths and weaknesses
of Session Initiation Protocol as an enabling IS IMS THE ANSWER?
integration technology and the challenges of
providing next generation voice services in a If IMS is the answer, what’s the question?
world where the rules of the Public Switched Its simple and it has nothing to with feature
Telephony Network still define much of what abstraction, common network core, service
can (and can’t) be done with new voice services. ubiquity, or any of the flashy promises we have
all heard much about. There are many ways to
achieve service enrichment goals and just as
Communications Technology has been on an many advocates and pundits about the right way
evolutionary path to convergence since the need to do it. So if the question is re-phrased a bit to,
to transmit computer data from one computer to “What does IMS do better than any other
another arose in the mid twentieth century. The possible service architecture?”, the answer is
telephone network was adapted to support “Take care of the guy on the other end of the
transmission of data. At the same time pure data line while all this neat multimedia feature stuff
network technology evolved and over time. With is going on in my network for my subs.”
the advent of internet technology it was
recognized that voice service could be From an architectural perspective,
considered as just another data type and in simplification of the call handling must be
many ways could be transmitted within a data achieved by minimizing the number of times
network as effectively as any other data type that call control must be shifted to different
and Voice over IP was born. applications. In the legacy telephony world the
Advanced Intelligent Network service
Voice over IP technology has begun to invocation mechanisms that allow applications
replace traditional Public Switched Telephone to manage call state are able to work flawlessly
Network elements across the globe. In most because the rules are very well defined and
successful cases this transformation has gone rigidly inflexible. Unfortunately those same
un-noticed by the end user. This apparently rules limit the communication types supported

2008 NCTA Technical Papers - page 261

to standard voice user scenarios and are not between a Cox customer and the far end PSTN
extensible to other media types or session type switch that can result in failed or dropped calls.
descriptors such as presence based routing The reality we observe today is that current
policies and lack effective web integration generation application environments and
capability. Voice over IP service invocation and systems want to handle service requests by
call control based on Session Initiation Protocol taking over control of the call.
lacks the rigor of AIN and consequently shifting
call control makes things complex. SIP is a A way to visualize this general issue is think
simple and very flexible protocol. It is flexible about having tried to manually set up a three-
almost to a fault and the specifications are often way call and then disengage yourself to allow
interpreted differently by vendors. That’s one of the other two people talk. These often fail
the biggest reasons issues still arise in SIP VoIP because you end up with a different count of
with features that have worked for decades in call setups versus lines in use. The PSTN
the TDM world. switches expect those counts to match so one
will initiate a teardown. We have been looking
Issue Legacy Interim IMS Comment for a way to replicate a trunk release in VoIP the
Solution Solution Solution same way a TDM PRI does it for years to no
Advanced AIN or PRI SIP Re- SIP App The IMS SIP avail because the VoIP protocols are unable
Voice based call direct and Server Application Server
Features forward call control provides features,
how to tell the far end its okay for the counts not
handoff feature interaction to match the way ISDN does this with a channel
management, and
web integration
transfer. This is one of the basic problems of
PSTN SS7 and TDM SIP CORE CSCF and An effective SIP PSTN replication with VoIP.
Routing and Interconnection using MGCF CLASS 4 Tandem
Call integrity ENUM easily evolves to
MGCF and can Cox encountered this issue when executing
provide some CSCF VoIP interoperability with a Directory
functions.
MGCF maintains call Assistance provider at the VoIP level (SIP –
state with the PSTN SIP) for directory assistance call placement by
and masks the
multimedia functions making it a three way call with a dormant leg
occurring in the IP since the softswitches involved can’t agree how
domain.
Fixed Mobile Dedicated FMC SIP “Call Presence In IMS the to release it. We pay a direct price there using
Convergence call control pull” via re- and user association of a SIP ports that don’t do anything but that’s not
agent to move a direct using preferences specific phone
call from one Application tell the number to each voice the real issue. The real problem is this heavy
phone number Server network endpoint isn’t handed approach breaks the general
to another how to required.
connect assumptions about call state and subsequent
with the feature invocation becomes clumsy at best,
user
because there is this third call leg involved that
Table 1
started the whole thing which, from a
subsequent call treatment point of view, has no
An ideal scenario for an operator is to map
business being there. That is the general
an inbound call to SIP only one time when it has
workaround for releasing a trunk today with
entered their VoIP core and anchor its
VoIP. You don’t release.
relationship to the PSTN is managed there. This
activity is referred to in IMS terms as Media
Simple features become complex because a
Gateway Control Function (MGCF).
call leg exists that shouldn’t be there.
At Cox we have realized that this approach
minimizes the possibility of state mismatches

2008 NCTA Technical Papers - page 262

Fixed Mobile Convergence today is a form of agents(Class 5, Tandem, and FMC)are involved
trunk releasing that encounters this same general for feature delivery and call control.
issue of trunk releasing compromise. This isn’t
to say that call control based FMC systems Subsequent expansion of the feature
don’t work, rather, that by not being able to capabilities on one or the other platform is now
execute a release with a base protocol, it is a required and portends painful integration until
heavy handed way to do it that results in making these systems have been augmented past their
things other than just moving the call from one original design capability. The ultimate
endpoint to the other incredibly complex. A consequence is that the operator ends up with
subsequent consequence of employing a another service platform that doesn’t perform
standalone FMC system is that eventually you well but is costly to replace because of all the
end up having to host feature applications on investment made to integrate to full
this system because it was never designed to functionality needed. All this in the name of
support services from another application taking care of the guy on the other end of the
environment. Providing feature transparency line. The issues that IMS technology directly
between mobile and fixed endpoints on your addresses with respect to the PSTN are listed in
existing messaging system systems (or other Table 1
apps) is where operators will struggle with
integration to FMC because up to three call

2008 NCTA Technical Papers - page 263

 challenge for them because when you closely
IMS however; addresses this directly by not inspect virtually any current generation VoIP
allowing any one application to dominate call application server you will probably see a
control with respect to the PSTN. It doesn’t device originally designed as a softswitch that
matter because IMS routes the inbound PSTN tends to behave like one by handling connection
call to a subscriber(fully qualified domain state of that far end PSTN switch.
name), not an endpoint(phone number).Fig 1
There are exceptions of course, some which
The concept of the far end PSTN switch that rely on AIN and the TDM Class 5 switch it is
has to abide by PSTN rules doesn’t exist inside attached to and some telephony application
of IMS, only at its border(MGCF). A good servers that function well being treated as a
MGCF looks like just another route for a service giant VoIP PBX until IMS liberates them totally
to IMS applications. Many of today’s available from the PSTN. Good pre-IMS app servers at a
“stovepipe” application systems promising IMS minimum mask control of the call by staying in
compliance in the future but this will be a real synch with the MGCF.

Figure 2

2008 NCTA Technical Papers - page 264

 This opens the door to a form of FMC that solutions. Essentially a doubling of effort is
can be utilized until IMS application integration required to pursue both paths.
has reached the level of maturity needed to
extend those services to a highly reliable voice An additional consequence is that to get to
offering IMS, engineering teams will have to expend
effort to undo the connected to legacy
This does limit the offering only to infrastructure adding yet more development
endpoints capable of supporting SIP. Inside IMS cycles. Our experience at Cox has been that the
the distinction of fixed/mobile disappears so effort to migrate systems rivals or surpasses
FMC based on call control handoff is replacing them entirely.
meaningless. Getting applications to integrate
seamlessly now becomes a design quality issue If you as an operator ask “Ok, what should I
no longer restricted by the limitations imposed do instead?” the answer today is focus effort on
at the call routing layer. Feature interaction building the MGCF that allows non-possessive
management is becoming a discipline unto itself application servers to work in your network and
that VoIP engineers will be practicing for as start identifying and integrating those
long as can be imagined. application servers. Fig 2 Does that get an
operator to wow factor features as quickly as
The bottom line is that from a technology stovepipe systems directly connected to legacy
standpoint any FMC or Unified Messaging infrastructure? Probably not but it will get you
design using available application servers that to where you really want to be faster than taking
directly interface to the PSTN CLASS 5 switch a big detour and it will allow you to stay there
is a big diversion away from IMS. As such any when you get there. Implement a uniform and
focus on nailing up a design for FMC and UM reliable method of taking care of the guy on the
under conventional terms becomes a full other end of the line and there are only a few
diversion away from IMS development because steps beyond that needed to realize the service
the core skills focused on IMS or pre IMS today rich environment of IMS.
are the same ones that must be used for interim

2008 NCTA Technical Papers - page 265

 MIGRATING DIGITAL AD-INSERTION APPLICATIONS
FROM MPEG-2 TO AVC (H.264)
Mukta Kar, Ph.D., Cable Television Laboratories, Inc.
Sam Narasimhan, Ph.D., Motorola, Inc.

Abstract for generating an AVC Video [1] and MPEG-2

Systems [2] conformant output by such splicing
Splicing is the fundamental technique used equipment so that a seamless or near-seamless
to insert commercials or short programs in splicing can be achieved.
channels, for editing audio/video content in
post-production houses, and for channel
switching in headends and other broadcast INTRODUCTION
stations. Splicing is currently used in US Cable
networks for digital ad-insertion based on The opportunity for Local commercial
MPEG-2 video [3], SCTE and ITU-T standards insertion has been created to benefit the
[4] and there are plans to migrate these communities and its businesses on a local, zonal
applications and develop associated standards or regional basis since the days of Analog
based on the emerging H.264/AVC video [1] in Television. Local commercial / advertisement
the near future. (Ad) opportunity provides the US broadcast
television industry over 35 billion dollars in
In these new applications, the splicing revenue. The revenue from this opportunity for
equipment (or function) combines two the cable industry has grown from a few million
independently encoded AVC streams and is to a few billion dollars with 5-6 billion in
expected to produce a stream for receiving revenue expected in 2008. As it provides a
equipment that conforms to both AVC Video [1] significant cash flow for our MSOs, increasing
and MPEG-2 Systems [2]. To achieve this revenue further has a prime importance to
significantly higher compression efficiency than the industry. In the days of analog television,
that of MPEG-2 video while providing same or most local Ad-insertion equipment was
better quality video, AVC compression standard proprietary in nature and hence non-
has introduced several new tools, reference interoperable. In moving from the era of analog
picture structures and enhanced MPEG-2 tools television to a digital one, the cable industry
all of which can be used adaptively based on the understood the problems in using proprietary
nature of the content. All these make AVC more equipment and the advantages in using inter-
complex compared to prior compression operable equipment from a multi-vendor market
schemes in addition to being not backward place. To create such a competitive multi-
compatible with MPEG-2 video. vendor market place, the cable industry took
initiative in standardization efforts in both the
Many of the Standards Development international (ITU/ISO) and national levels
Organizations (SDOs) such as DVB/ETSI, (SCTE) that covered not only Audio-Video but
SCTE, DVD and ARIB have completed the other areas such as cable modem, VOIP, etc.
specifications related to the adoption of AVC in One such application area is the local
broadcast, VOD and PVR applications. This program/commercial insertion.
paper outlines issues related to splicing between
two independently coded AVC streams for local Figure 1 displays a block schematic of a
Ad-insertion applications and proposes schemes typical local ad-insertion system. A timing

2008 NCTA Technical Papers - page 136

signal known as Cue-tone (in analog) or Cue- headend equipment separates Cue-tone/Cue-
message [4] (in digital) is embedded with a message from the program. Based on the timing
program and then distributed to the headends or signal in the cue-tone/cue-message, a splicer
broadcast affiliates via satellite. MSOs receive and Ad-server replaces the national ad with a
such a program using an IRD (Integrated local ad. This process is known as local ad
Receiver Decoder) at their headend. Then the insertion.

Combiner
Figure 1. Schematic Block Diagram of Digital Program/Ad Insertion System

As shown in Figure 2, Local Ad Insertion uncompressed video to consumer homes is very

technology in the analog video domain was inefficient in the usage of bandwidth in addition
simple in nature as the transmit order of frames to many other limitations. Digital video
is same as the display order. Splicing the digital compression technology coupled with digital
uncompressed video is also simple for the same modulation provides significant efficiency,
reasons. The process involves frame accurate flexibility and other benefits in delivering digital
timing signals indicating the beginning and end video and audio to consumer homes. Also
of a national advertisement in a program that digital technology provides superior video
need to be replaced with a local ad or perhaps quality compared to analog as the former is less
with an updated ad. Delivery of analog video or prone to noise.

2008 NCTA Technical Papers - page 137

 Figure 2. Simplified Diagram of Local Ad Insertion

However, digital video compression one splices out at any of the B pictures, then this
technology introduces problems for some of will introduce gaps in the display. For seamless
these broadcast applications (e.g. splicing and near seamless splicing between two
between two MPEG-2 streams or between two compressed video streams, MPEG-2 TSTD
AVC streams) primarily due to two reasons – buffer conformance must also be maintained
(a) Order of the video frames gets modified in where decoder buffer (buffer size of 1.8Mbits
compressed domain (transmission order does for MP@ML) must not overflow or underflow
not maintain display order) and (b) compressed as this may result in artifacts. Typical buffer
frames depend on other frame or frames (called behavior in a MPEG-2 decoder is shown in
as reference frames) for decoding / Figure 4 assuming the decoder receives a
decompressing them in the decoder. constant bitrate channel. MPEG-2 provided
tools to achieve seamless or near seamless
SPLICING BETWEEN STREAMS WITH splicing but it does not tell on how to achieve it.
MPEG-2 VIDEO MPEG left it to MPEG-2 product designers for
innovation and product differentiation. To
Figure 3(a) illustrates a segment of video achieve seamless or near seamless splicing,
where frames are in display order. Figure 3(b) some constraints may have to be maintained
depicts the same segment when compressed in while creating the streams to be spliced. Such
compliance with the MPEG-2 video standard constraints may include GOP structure, an
and sent over a transmission channel. One may anchor frame at the out-point of the first stream
notice that the transmit/decode order is not same and an I frame with a sequence header and
as the display order and hence a splice cannot be closed GOP at in-point of the second stream.
done at all picture boundaries. For example if

2008 NCTA Technical Papers - page 138

 Figure 3. Typical Prediction Methodology used in MPEG-2 Video Compression.

Figure 4. Simplified diagram of an MPEG-2 decoder and level of bits in the decoder buffer.

An additional function that needs to be met stream conditioning and matching impose too
by digital ad-insertion systems is the matching much of a burden and constraint on the uplink
of decode delays between the network and encoders and ad-servers. Hence SCTE
splice stream as shown in Figure 4. developed specifications such as SCTE 35 [4]
(digital cue-message standard) to signal the
Even though it may be easy to splice splice opportunities in the compressed video
between two ‘well conditioned’ MPEG-2 stream and splicers were developed to perform
transport streams (standards such as SMPTE the tasks outlined above so that splicing can
312M specify this stream conditioning), the occur without imposing too many constraints on

2008 NCTA Technical Papers - page 139

the uplink or ad-servers. Majority of the splicers transition between network program and
deployed currently perform the following advertisements.
functions:
AVC VIDEO CODING AND
 Continuous bitrate transcoding to HIERARCHICAL GOP STRUCTURES
maintain an average compressed bitrate
between the two video streams. It has been mentioned earlier that to achieve
 Time base adjustment to maintain a better compression efficiency than that of
common PCR, PTS and DTS between MPEG-2 video standard [3], AVC [1]
the streams without introduction of any introduced many new tools and enhanced some
discontinuities. MPEG-2 tools. One of these tools is in the use
 Matching the vbv-delay between the two of B-pictures as reference and hierarchical use
streams so that decode and presentation of such B pictures.(MPEG-2 video does not
are continuous when the output reaches allow B frames to be used as reference). This
the settop units at consumer premises. particular tool introduces an additional
 Splice between content in film-mode and complexity for splicing and this will be
non-film-mode by maintaining field discussed later. The MPEG/JVT committee also
parity. structured AVC in a very flexible way so that it
can be implemented in a wide range of
In addition, the splicers also maintain applications that includes broadcast, video
conformance to MPEG-2 video and systems telephony, video conferencing, and video
standards in their output and make sure that PSI streaming.
information does not change across the splice so
that settop units can present a seamless

Figure 5. Typical Predication Methodology used in AVC Video Compression.

Use of B frames for reference in AVC and Frames are in display order in Figure 5(a),
the associated GOP structures that use this are where as Figure 5(b) shows a typical
sometimes called ‘Hierarchical GOP structures’. compression structure of AVC where
Figure 5(b) shows a typical GOP structure in hierarchical GOP structures are used with B
AVC video where I, P and B frames are used. pictures as reference. One may notice that the

2008 NCTA Technical Papers - page 140

difference in transmit/decode order of pictures ‘one frame period’ for MPEG-2 sequences that
in Figure 3(b) and Figure 5(b). The use of B use different values of m and hence MPEG-2
pictures as reference and GOP hierarchy makes splicer’s were able to concatenate any two
decoding AVC coded stream more complex sequences and still maintain conformance to T-
compared to MPEG-2 coded video with respect STD, VBV and constant display rate at their
to the management of reference pictures in the output without any difficulty.
decoder memory. AVC also introduced IDR
picture and changed traditional definition of I In AVC this is not true as the ‘display
picture which is used in MPEG-2 video. These latency’ for different video sequences vary
advanced tools and flexibility of GOP structures based on the hierarchical GOP structures which
make seamless or near-seamless splicing using use B pictures as reference pictures. The display
AVC video [1] for local ad-insertion more latency can range from ‘one picture period’ (for
challenging compared to MPEG-2 video [3]. non-stored-B structures or MPEG-2 like
structures) to several picture periods based on
ADDITIONAL SPLICING ISSUES WITH the levels of hierarchy. In addition, unlike
AVC VIDEO COMPARED TO MPEG-2 MPEG-2 video AVC mandates the maintenance
of constant display rate (constant DPB output)
AVC splicers have to implement all the when coded video sequences are concatenated.
functions that are implemented by MPEG-2 This makes concatenation or splicing two such
video splicers. These include bitrate video sequences with different display latencies
transcoding, time base adjustment, CPB delay difficult as the output cannot conform to AVC
matching and maintaining field parity between specifications (I.E; maintain constant delta in
film and non-film modes. In addition, AVC the CPB removal time and DPB output time). If
splicers need to manage another function to the display latencies do not match, then one will
accommodate the ‘Hierarchical GOP’ structure see a missing picture at DPB output time or see
variations between the streams being spliced. 2 access units with the same DPB output time.
The following illustrates the issue and proposed One solution to this is to only combine
solutions. sequences that have the same GOP structure and
this mandate is not attractive in Cable networks
MPEG-2 provides relatively simple GOP for applications that use splicing. Based on
structure involving I, P, and B pictures. inputs from the US Cable community, AVC has
Depending on a scene the parameter m (number agreed to modify the standard allowing the use
of B picture between two anchor frames) varies. of a marker called end_of_stream NAL unit to
MPEG-2 allows only two reference pictures at splice between two AVC coded video sequences
any time which are managed in the decoder or streams with different display latency where
memory using FIFO method (also known as the requirement to maintain constant DPB
bumping process) where arrival of a new output does not apply. The next section covers
reference picture pushes out the older reference proposed solutions for splicing between AVC
picture out of the decoder memory. In MPEG-2 streams based on this action by AVC.
(for non-low-delay mode) the delay between the
decode time of first access unit in the sequence PROPOSED SOLUTIONS FOR BOTH
and the display time of first access unit in the STREAM CONDITIONING AND SPLICER
sequence is always ‘one frame period’ for both FUNCTIONS
closed GOP (I,P,B,B,P,B..) and open GOP
(I,B,B,P,B..) structures. Let us call this as the 1. In order to enable seamless splicing between
‘display latency’. This display latency is always two video sequences at different horizontal

2008 NCTA Technical Papers - page 141

 resolutions (at same frame rate and vertical highest horizontal resolution for the level
resolution), the application standards such as such as 720 for SD and 1920 for HD). This
SCTE 128 [5] are mandating the same allows the decoders to keep the same DPB
number of pictures in the DPB for all the memory management across the resolution
horizontal resolutions (determined by the change and hence produce seamless output.

O
Figure 6. Seamless / near-seamless splicing without the use of end_of_stream NAL unit at the splice point (a) a case where
no_output_of_prior_pics_flag=0, (b) a near-seamless case where no_output_of_prior_pics_flag=1.

2. AVC also agreed to loosen the requirement latency difference can be adjusted by the use
for decoders to infer the DPB management of Picture timing SEI message with an
using the no_output_of_prior_pics_flag appropriate value for pic_struct. This SEI
(I.E., infer this to be ‘1’ and clear the DPB with the pic_struct value allows repetition of
when there is a resolution change). The last displayed picture and this can be used to
change allows the application standards to splice a stream with a higher display latency
mandate that receivers process this flag into a stream with a lower display latency.
correctly so that DPB is managed per the This also requires the first stream to be
transmission systems intent. Splicers can set coded using frame pictures. For all other
this flag correctly at the transition points to combinations of streams, the end_of_stream
achieve ‘seamless’ splicing as shown in NAL unit should be used with the correct
Figure 6(a). setting of the no_output_of_prior_pics_flag
at the transition point to manage the DPB
3. The third proposal is the appropriate use of buffer and make sure that two pictures with
end_of_stream NAL unit at the splice the same display time are precluded.
transition points (called Out or In-Point) so Seamless splicing with end_of_stream NAL
that seamless or near-seamless splicing is unit can also be achieved by offsetting the
possible. This is shown in Figure 7. In some decode time of the pictures in Ad Stream
combinations, seamless splicing can be appropriately. This mode is not
achieved without the use of end_of_stream recommended as most receivers expect the
NAL unit. The first example is where the decode time to be contiguous between
display latencies match between the streams. network and ad-streams.
The second example is where the display

2008 NCTA Technical Papers - page 142

 X
X

O
O

Figure 7. Near-seamless splicing using end_of_stream NAL marker at the splice point (a) a case where no_output_of_prior_pics_flag=1 (b)
a case where no_output_of_prior_pics_flag=0.

SUMMARY
BIBLIOGRAPHY
In this paper the technology of splicing and
its importance to achieve local ad-insertion for 1. ITU-T Rec. H.264 | ISO/IEC 14496-10,
the broadcast industry have been discussed. The (2005), “Information Technology – Coding
local ad-insertion application provides of audio visual objects – Part 10: Advanced
significant amount of revenue to the industry, in Video Coding.”
particular, the cable MSOs. The challenges of
splicing two video streams in the compressed 2. ISO/IEC 13818-1, (2007), “Information
domain (e.g. MPEG-2 video) for seamless and Technology – Generic coding of moving
near-seamless viewing experience have been pictures and associated audio – Part 1:
discussed. It has been also shown that splicing systems.”
of two AVC video streams is much more
difficult than MPEG-2 video as AVC video 3. ISO/IEC 13818-1, (2000), “Information
coding uses more advanced video tools and Technology – Generic coding of moving
complex coding structure to achieve higher pictures and associated audio – Part 2:
compression efficiency. This paper proposes a video.”
few methods/solutions to splice AVC streams to
achieve seamless or near-seamless ad-insertion 4. ANSI/SCTE 35-2004 | ITU-T J.181, Digital
as needed in the broadcast industry. Program Insertion Cueing Message for
Cable.

ACKNOWLEDGEMENTS
SCTE 128 (2007), AVC Video Systems and
Transport Constraints for Cable Television
The authors wish to thank the management
of CableLabs and Motorola for their support and
encouragement in performing this work.

2008 NCTA Technical Papers - page 143

 MOBILE TV: A TECHNICAL AND ECONOMIC COMPARISON OF BROADCAST,
MULTICAST AND UNICAST ALTERNATIVES AND THE IMPLICATIONS FOR CABLE
Michael Eagles, UPC Broadband
Tim Burke, Liberty Global Inc.

Abstract We provide a toolkit for the MSO to assess the

technical options and the economics of each.
The growth of mobile user terminals
suitable for multi-media consumption, combined Mobile TV is not a "one-size-fits-all"
with emerging mobile multi-media applications opportunity; the implications for cable depend on
and the increasing capacities of wireless several factors including regional and regulatory
technology, provide a case for understanding variations and the competitive situation.
facilities-based mobile broadcast, multicast and
unicast technologies as a complement to fixed In this paper, we consider the drivers for mobile
line broadcast video. TV, compare the mobile TV alternatives and
assess the mobile TV business model.
In developing a view of mobile TV as a
compliment to cable broadcast video; this paper EVALUATING THE DRIVERS FOR MOBILE
considers the drivers for future facilities-based TV
mobile TV technology, alternative mobile TV
distribution platforms, and, compares the Technology drivers for adoption of facilities-
economics for the delivery of mobile TV based mobile TV that will be considered include:
services.
 Innovation in mobile TV user terminals - the
We develop a taxonomy to compare the feature evolution and growth in mobile TV
alternatives, and explore broadcast technologies user terminals, availability of chipsets and
such as DVB-H, DVH-SH and MediaFLO, handsets, and compression algorithms,
multicast technologies such as out-of-band and  Availability of spectrum - the state of mobile
in-band MBMS, and unicast or streaming broadcast standardization, licensing and
platforms. spectral harmonization,
 Evolution of network technology – the
INTRODUCTION increasing capacity of wireless bandwidth
the emerging mobile return path and channel
Cable MSOs operate in an increasingly change improvements,
competitive market with incumbent Telcos and  Usage context and prospects –
independent wireless operators. Cable's early demographics, viewership, and subscriber
victories in the voice market led to an aggressive willingness to pay.
response to offer video products by the Telcos.
1. Innovation In Mobile TV User Terminals
The next area for intense Telco competition will
likely be mobile television. The addition of As a key driver for mobile TV, advances in user
television to their mobile voice and data products terminals enable new features and usage models
may be the logical next step ... but it may not be that enhance the mobile TV experience. We
for cable. believe this trend will result in a wide availability
of handsets capable of receiving mobile TV over
time.

2008 NCTA Technical Papers - page 238

In particular, increasing screen sizes, resolutions, decision regarding support for full frame rate
and decreasing power consumption, support video, as the bandwidth required is around 1.5-
longer usage period and usage scenarios and 1.8x higher than today’s 256Kbps bit rates, at
enable a greater number of radio and network around 400 Kbps.
alternatives.
Some common data rates for mobile TV are
We note that mobile TV user terminals highlighted below. The analysis that follows in
supporting mobile TV are predominantly this paper will focus on the Class B/Medium data
targeting QVGA resolutions today. However, rate.
increasing mobile screen resolutions may drive a
need for higher bandwidth in the future. Table 2: Common Data Rates for Mobile TV 4
Data rate Frames per
Table 1: User Terminal Resolutions 1 (Kbps) second
Example Width Height Total (fps)
User Pixels
Terminal Class A 128 10 – 12
VGA Nokia 640 480 307,200 Class B – Low 256 15 – 20
N800 Class B – Medium 256 30
HVGA Apple 480 320 153,600
iPhone Class B – High 384 20 – 25
QVGA Samsung 320 240 76,800 Class C 768 30
P910
QCIF Motorola 176 144 25,344
V8
How do we know what users consider the
How much bandwidth is required to support minimum acceptable quality when viewing
QVGA at 20 frames per second? mobile TV user terminals?

We can estimate this by considering QVGA Several studies have been conducted into the
resolution of 76,800 x 24 bit colour x 20 frames acceptability of mobile TV content at varying
per second = 36,864,000 bits per second. resolutions and varying bit rates.
Assuming a compression rate of 1412 provides an
approximate video bandwidth of 256 Kbps. Figure 1: Mobile TV User Terminal
Acceptability of Video 5
Can today’s multi-media user terminals support
full frame rate broadcast video?

Over time we believe all mobile TV user

terminals will be able to support full frame rate
video. Early mobile TV user terminals could not
process QVGA resolution at 25 fps or higher3,
typical of most broadcast systems. For example,
the Nokia N92 and N77 could not support this
frame rate due to processing limitations. This is
changing with the new Nokia N96 being capable
of up to 30fps at QVGA resolution. Figure 1, above, indicates that QVGA mobile
video acceptability for football reached a plateau
With mobile TV user terminal processing for bitrates of 332 Kbps and greater than 84%.
capability improving, it will become an operator In contrast news and weather delivered an

2008 NCTA Technical Papers - page 239

acceptable service to 90% of people at bandwidth held 40 centimeters from the eyes, as seen in
of just 112Kbps6. figure 3 below.

What compression improvements are possible Figure 3: Visual Saturation for Fixed and Mobile
with advances in mobile TV user terminals? TV

As full frame rate handsets become available,

requiring higher bandwidth, operators will look
to advances in compression technology in order
to maximize use of finite mobile TV bandwidth.

Table 3: Mobile TV Compression Improvements

2008 2009 2010
Compression MPEG4 MPEG4 MPEG4
Profile AVC / VC1 AVC / VC1 AVC / VC1
enhanced improved
Percent Today 10 – 15% 10- 15% However, research indicates there are some key
Improvement perceived issues with mobile TV user terminals
Possible
for viewing TV, other than ‘general detail’ and
‘image size’. These included ‘fatigue’ and
‘effort’, perhaps associated with correct
Does a smaller screen size make mobile TV user positioning of the user terminal.
terminals less attractive to the viewer than
cable’s typical fixed-line TV? Table 4: Mobile TV User Terminal Problems
across All Content 8
Perhaps counter-intuitively, studies indicate that Problem % of General
standard television is much closer to the limits of Comments
human perception than mobile TV user General Detail 20%
terminals.7 Insufficient Image Size 18%
Fatigue 10%
Figure 2: Mobile User Terminal Viewing Ratios Effort 8%

What is the availability of mobile TV chipsets

and user terminals ?

The availability of user terminals capable of

delivering mobile TV is increasing over time.
Variations in different regional approaches to
standardization and service characteristics have
resulted in broader user terminal availability in
Asia.
Interestingly, if we represent this visually in a For example, mobile TV user terminals in Japan
fixed and mobile scenario, we can see that a have reached 20 million units shipped in just
typical living-room fixed widescreen TV at 3 under two years since launch, noting that Japan’s
meters can be visually similar to a QVGA screen mobile digital TV service ISDB-T (OneSeg)
offers a simulcast of Japanese terrestrial TV

2008 NCTA Technical Papers - page 240

stations at no cost to the end user. The Table 5: Current & Future Standards
broadcasts are not secured, which facilitates the USA W. Eu Japan Global
Current MediaFLO DVB-H "one-seg" T/S-DMB
production of low-cost compatible devices and Most & T-DMB ISDB-T Korea
handset diversity9. Popular
Standard
Options MPH, DVB- TDtv, MediaFLO, MediaFLO,
Other markets appear to be challenged with H, MBMS MBMS, MBMS, MBMS,
DVB-SH DVB-H MPH,
respect to handset diversity due to the emergence DVB-H,
of multiple differing distribution standards (i.e. CMMB
(China)
DVB-H, MediaFLO), the need for diversity and Expected MediaFLO DVB-H ISDB-T MediaFLO
potentially filters (i.e. TDtv), and the need for "Winning" dominates will dominates may
security support (i.e. OMA Bcast standard) Standard until unicast dominate until emerge in
over 4G with unicast Japan &
capabilities to protect content. TDtv, over 4G Hong Kong
MBMS & with large
DVB-SH CMBB
We believe that as the number of mobile TV emerging. volumes
distribution standards proliferate, the emergence expected in
China
of handsets with the ability to support multiple
technology options will emerge. With a view to assessing the availability of
spectrum for mobile TV we survey the typical
2. Availability of Spectrum frequency bands available in the summary below
and Table 6.
Another key driver for mobile TV includes the
availability of spectrum, including the state of VHF Band: In some European and Asian
mobile broadcast standardization, licensing and countries (Korea) narrow slices of the 200 MHz
spectral harmonization. VHF band has become available for terrestrial
broadcasters to provide Mobile TV services. T-
Considering the standards for mobile TV, we DMB technology was used in these allocations.
note that today there are five worldwide
broadcast TV standards (DVB-H™10, UHF (470 to 870 MHz): In relation to UHF
11
MediaFLO™ , ISDB-T, T-DMB, S-DMB) and spectrum, we believe that the long wait plus the
three more broadcast standards planned (MHP, uncertainty on how much spectrum will be made
DVB-SH and CMBB). available for Mobile TV and who will get the
spectrum complicates the technology selection
When the two most widely known multicast for operators13. Unfortunately, in many countries
standards (TDtv & MBMS) and the entire this spectrum will not get released until the
category of unicast (in band cellular) are added digital TV transition (Digital Dividend) in the
to the mix it is apparent that the world of mobile 2012/2013 timeframe.
TV technology is extremely fragmented12.
L-Band (1.452 – 1.492 GHz): Alternatively, the
L Band is slated to be made available in some
countries (U.K.) in the near future and could
offer an alternative for broadcast mobile TV
services18.

UMTS-Bands (1.7 to 2.5 GHz): Because of the

broadcast spectrum issue and lack of alternative
frequency options it is highly likely that

2008 NCTA Technical Papers - page 241

multicast options like TDtv and MBMS will get
deployed in Europe in existing UMTS 3G Power levels that are permitted in each market
spectrum bands to begin to relieve unicast have a substantial impact on the number of sites
capacity problems. required in a given frequency. For example,
MediaFLO in the USA transmit at 50 kW14,
S-Band (2.17 – 2.20 GHz): One interesting DVB-H in Europe transmit at 5 kW15. This can
alternative is the S-Band satellite spectrum have an impact on the number of sites required
planned for allocation across the entire European and hence the economic viability of a mobile TV
continent in 2008. This spectrum will be network.
available earlier and offer a uniform frequency
and technology across a large region. The DVB- For example, in the USA a typical cell radius of
SH standard is being positioned to serve this the MediaFLO network operating at 50kW
frequency range. In the USA, ICO Satellite is transmit power from 150 to 300 meter towers is
looking to promote a similar spectrum and 19 Km to 27 Km while providing equivalent
technology allocation17. indoor coverage over similar terrain16,17.

Table 6: Possible Spectrum for Mobile TV By comparison, DVB-H technology in Europe

Band Name Status has 5 kW power limits imposed due to EMF and
2500 – 2690 3G Extension Technology neutral, usable for
MHz Bands 3G, DVB-SH, WiMAX, etc. interference regulations severely restricts cell site
2170 – S-band (usable EC decision & Selection radius. In a Belguim trial an average 3 Km cell
2200MHz with DVB-SH) process,
1900 – 2170 UMTS TDD Usable with MBMS. Possible radius was typical in suburban locations from 60
MHz interference with 3G FDD meter towers18.
1785 – 1805 UMTS FDD Used for mobile TV already
MHz (3G streaming) today in unicast mode.
1452 – 1492 L-Band Possible T-DMB, MediaLFO, Additionally, the available heights of transmitter
MHz DVB-SH.
470 – 860 UHF (usable Subject to broadcast license
sites will increase or decrease the total broadcast
MHz DVB-H, others) laws, used by DTT, analog. mobile TV site counts and ultimately mobile TV
economics19.
Considering the alternative spectrum options for
mobile TV, it is clear that in-band unicast over 3. Evolution of network technology
cellular has a time-to-market advantage.
Another driver of mobile TV is the evolution of
Other frequency bands are currently either network technology.
subject to ongoing regulatory approval,
competing with alternative technologies or Wireless network technologies continue to
services, or at risk of interference from evolve, with increasing capacity of wireless
neighboring services. bandwidth, support by new technologies such as
OFDMA modulation and MIMO (Multiple Input
The spectrum availability issue may cause Multiple Output) antenna technology, improved
technology fragmentation in the near term. compression algorithms, and greater cell
Some standards bodies are eager to prevent this densities of mobile operators
outcome by promoting a single specification as
the official approved standard for mobile TV. Additionally, in-home devices such as femto-
Other regulatory bodies seem to be taking a more cells and Wi-Fi™20 allow the wireless operator
technology neutral stance. to off-load capacity from its wide area radio
network, which will help reduce the need for an
What other regulatory factors have an impact on overlay network to support mobile TV.
the business case for mobile TV?

2008 NCTA Technical Papers - page 242

4. Usage Context and Prospects Recent studies 25 indicate that while a majority of
the people were interested in viewing mobile TV
We consider the usage context for mobile TV; 80% of the respondents said they would not pay
characterizing the demographics and viewership, $15/month for it. The study also concluded that
willingness to subscribe to a pay mobile TV subscribers are more willing to watch mobile TV
service, and elasticity to the prospects for mobile that is essentially the kind of programming they
TV. get on their TV now. Certainly this second
conclusion is a positive indication for cable
What mobile TV viewership and demographics companies regarding the importance of mobile
can an operator expect? TV to their future business.

A review of literature reveals that perceived Figure 4: Willingness to Pay for Mobile TV 26
mobile TV viewership differs significantly by Willingness to Pay for Mobile TV (USD Per Month)

region. For example, of the markets that have Vodafone-Nokia (DE)

launched mobile TV, France is reported to have HPI Research Group

Nokia (DE)

the lowest usage with 70 minutes per week, O2 (UK)

BMC (DE)
whereas Korea is reported to have the highest Booz Allen Hamilton (DE)

consumption with 160 minutes per week, or IPDC (EU)

RI (FI)

about 20 minutes per day21. Pyramid Research (UK)

Nokia/SFR (FR)
BT Movio (UK)

Research is showing that consumers tend to Nokia (ES)

watch mobile video in the home more than

0 5 10 15 20 25
USD Per Month

previously thought … despite the presence of big

screen TV’s 22. In addition, content executives Other studies in European countries assessing the
have been surprised in the performance of long- propensity for consumers to pay for mobile TV
form content on mobile devices 23. content indicate a range of US$10 to US$20.
It is a common hypothesis that consumers will For example, in Italy 3 Italia has 800,000 DVB-
use mobile TV to “kill time” leading to the H mobile TV subscribers (out of 8 million
consumption model that mobile TV viewership is mobile subs). 3 claims, their mobile TV offering
a “snacking” phenomena. Data from early has been instrumental in raising their ARPU 60%
research studies24 indicate otherwise. Namely, over the last year where one-third of the increase
that a high proportion of mobile TV viewing has been driven by mobile TV and the remaining
(30% to 50% in 3 out of the 4 surveys) is in the two-thirds by voice & data services. 3 Italia may
home. have discovered one willingness to pay pricing
model as they offer a popular all inclusive
Additional research and improved viewership package (voice, data & mobile TV) for US$42
statistics and a better understanding of mobile per month 27.
demographics would assist in refining the
technology choices and business model for Japan and Korea also offer an interesting
mobile TV. benchmark for a mobile TV subscriber’s
willingness to pay. The ISDB-T (One-Seg) amd
What will the mobile TV subscriber be willing to T-DMB services, make up a majority of the
pay for a subscription service? current mobile TV subscribers worldwide yet are
free service offerings, delivering free-to-air
content.

2008 NCTA Technical Papers - page 243

 operators need to be aware of differing research
We have provided early evidence of operator- definitions as to what constitutes mobile TV, (b)
based research into end user willingness to pay methodology of today’s forecasts and, (c) as
for premium mobile TV. However, it is not clear mobile TV is at an early stage of development,
at this stage whether a free simulcast model, a variation in forecasts can be expected.
pay TV model, or ad supported model will
dominate Europe, the United States and Asia. Overall, we note that where a pay TV
The type of model that emerges could be subscription model is the focus (i.e. USA,
expected to have a significant impact on mobile Europe), the penetration of mobile TV services is
TV’s prospects. lower than in markets where the service is
bundled as a free offering (i.e. Japan). We
What does this mean in terms of the prospects for believe that mobile TV is therefore very price
mobile TV? elastic, and significant penetration will most
likely come from bundling and cross
Based on industry information, the projected subsidization with other core mobile or
take up of mobile TV is estimated at 150 million entertainment services.
users or 4% of all mobile users globally by 2012
from a base of almost 4 billion mobile users, COMPARING THE MOBILE TV
with developed economies expected to ALTERNATIVES
experience higher penetration rates. For
example, by 2012, mobile TV is expected to Which mobile TV technology should a cable
have 35 million users or 7% mobile user MSO consider and what platforms pose the
penetration in Western Europe, 35 million users largest threat or present the greatest opportunity?
or 12% penetration in the United States and 65
million users or 64% mobile user penetration in For the purposes of this paper we are focusing on
Japan by 2012. facilities-based mobile TV technologies and
setting to one side alternative non-facilities based
Figure 5: Prospects for Mobile TV 28 alternatives (i.e. in-home radio technologies such
as Wi-Fi or storage-based PC to user terminal
100%
Prospects for Mobile TV
4,500
file transfers).
Total Mobile Users 2012
90% 3,943 Total Mobile Users 2007 4,000
% Total Mobile Users 2012
Percentage of All Mobile Subscribers

80% % Total Mobile Users 2007

There are several competing facilities-based
Base of Mobile Users (Millions)

3,500
70% 2,983

60%
3,000

2,500
platforms for mobile TV. We consider an overall
50%
2,000
taxonomy based on classifying the technical
40%

30%
1,500 alternatives into (1) Broadcast, (2) Multicast and
20%
517
451
1,000
(3) Unicast;
10% 500
288 245
102 98
0% 0
Global USA Western Europe Japan

We chose a conservative study summarized in

Figure 5 to highlight mobile TV’s prospects,
focused on linear TV programming, rather than
all types of video content29.

Observing the range of industry forecasts, i.e. up

to 465 million by 2010 30 we note that (a) cable

2008 NCTA Technical Papers - page 244

Table 7: Mobile TV Delivery Alternatives 31
Broadcast Multicast Unicast OFDM (Orthogonal Frequency Division
Network Broadcast Cellular Cellular Modulation) was chosen as the most effective
Topology One-many Mixed One-one
way to meet these design goals. Fortunately, it
Return path No Yes Yes
Bandwidth Dedicated Mixed Shared was able to leverage other standards such as Wi-
Throughput Fixed Mixed Variable Fi™34, ADSL, DTV, UWB, WiMAX™ 35, LTE
Zap speed 1-3 secs 2-5 secs 5 – 8 secs and DVB-H that all employ OFDM technology.
32

Technology DVB-H TDtv WiMAX

Just as TDMA separates communication
Example DVB-SH MBMS LTE
MediaFLO HSPA, channels and end user conversations with time
HSPA+ division and CDMA segments channels with
3G
(UMTS/
codes (orthogonal spreading codes), OFDM
WCDMA) utilizes frequency. It differs from 1st generation
Advantages Cost Re-use of Variety, analog cellular frequency division techniques by
structure, existing on-demand
performance spectrum
using very tightly spaced frequencies without
Disadvantages Variety, Price, Price, overlapping and interfering. It does this by
additional performance performan forcing the narrowband FDM carriers (called
network ce
subchannels or tones) to appear unique or
independent from each other. The mathematic
We take a closer look at the technology concept of orthogonality is the key to
alternatives to determine what the advantages maintaining separate communication channels
and disadvantages are, and what this means for even though the subchannels are very narrow and
the cable MSO. spaced close to each other.
1. Broadcast Qualcomm effectively incorporated time slicing
into their specification so that mobile devices
Looking at a typical broadcast architecture for used as little power as possible. This technique
facilities-based mobile TV we note that there are transmits chunks of data in bursts so that the
quite a number of similarities to the traditional receiver could be turned on and off during
cable MSO broadcast architecture; including the inactive time periods. The result is substantial
need for encoders, and electronic program guide, power savings (90%) over traditional broadcast
and conditional access systems. technologies using fixed high power receivers.
Because MediaFLO was designed without the
Exploring the broadcast alternatives in more need for compatibility with legacy standards by a
detail we consider MediaFLO, DVB-H, DVB-SH company with relevant experience in mobile
and T/S-DMB. devices it utilized some very effective
techniques. MediaFLO uses a more frequent
(a) MediaFLO transmit time interval than DVB-H, which helps
in having quicker channel change speeds and
The MediaFLO specification was developed by improved power saving36.
Qualcomm specifically for broadcast mobile TV
applications. Consequently, it was optimized for The most unique design aspect of the MediaFLO
high bandwidth (many simultaneous video implementation is the ability to have layered
channels), high speed mobility, single frequency modulation. Basically, the data stream bursts are
networks, low power drain CPE devices, large divided into base and enhanced layers. The base
cell radius and fast channel changing layer supports the widest coverage area using
capability.33

2008 NCTA Technical Papers - page 245

lower quality (15 fps) video that subscribers and Consequently, time slicing and forward error
receivers in poor signal areas (such as in- correction elements were added to the
building) can decode. The enhancement layer specification. Physical layer changes included
supports high quality 30 fps video and is decoded the use of 4,096 OFDM subcarriers (DVB-T
by the receiver in high SNR (signal to noise) allowed for just 2K or 8K options), better
areas. The MediaFLO handset dynamically flexibility in using all modulation formats
adapts the video quality based on the signal (QPSK, 16 QAM, 64QAM), creating a 5 MHz
strength. Consequently, there is a smoother channel bandwidth and expanded bit interleaving
degradation of service as the signal strength options38.
varies37.
For the most part these layer 1 and 2
An important design tradeoff for a broadcast specification changes put MediaFLO and DVB-
mobile TV architecture is determining the H at a similar capability. Overall, MediaFLO
optimum number of OFDM subcarriers or tones. has more beneficial performance, coverage and
On the one hand, a large number of subcarriers capacity technical characteristics. Conversely,
(8000 in a 5 MHz channel bandwidth) will the DVB-H standard is much better positioned as
provide for higher capacity and a larger single a uniform worldwide standard because of its
frequency network (avoids handover to different strong backing in Europe.
frequencies) but will negatively impact high
speed mobile performance. Qualcomm and the (c) DVB-SH
DVB-H specification both settled on 4,096
subcarriers as the optimum compromise for A European wide allocation of satellite spectrum
mobility, capacity and large single frequency and the vision of a uniform continent wide
networks. roaming capability has prompted the creation of
the DVB-SH standard. This architecture will
A final unique aspect of the MediaFLO provide direct outdoor coverage to handhelds and
specification is the use of a variable bit rate and vehicles (with outdoor antennas) from a satellite.
statistical multiplexing allocation for the video Indoor coverage will require a large number of
services. This feature provides a bandwidth repeater sites located at existing cellular sites.
efficiency gain of about 30% translating into a Utilizing an existing wireless carrier’s dense cell
higher number of video channels at comparable site network will be critical for this service to be
quality in a channel. effective.

(b) DVB-H DVB-SH provides two key improvements to

DVB-H: (1) 3GPP2 Turbo Codes, that improves
The Digital Video Broadcast standard for the quality of reception in tough conditions and
handheld devices is based on the existing DVB-T (2) Physical layer time interleaving that improves
standard for fixed digital TV reception. the quality of reception while in motion.

Most changes were made to the layer 2 portion of The net result is that, under the same conditions
the specification and focused on making (frequency, channel size, data-rate) signal
improvements so that video transmission would reception requirements (carrier to noise) are a
be robust enough for a severe multipath mobile minimum of 5 to 6 dB lower39 and up to 6 to 8
environment and low power mobile devices. dB lower40 for DVB-SH relative to DVB-H.

2008 NCTA Technical Papers - page 246

Additionally DVB-SH is able to leverage cellular only mobile TV service S-DMB is extended
sites to down-convert S-Band Satellite mobile indoors with terrestrial repeaters.
TV content transmissions to maximize coverage
and minimize distribution/backhaul costs. All Essentially, T-DMB and S-DMB are very similar
other terrestrial based broadcast technologies specifications. The biggest difference is the RF
require backhaul transport of the mobile TV planning and implementation of the network
content to every site. DVB-SH requires a more associated with S-DMB as the interference
economical satellite dish and regenerator between satellite and terrestrial transmission
equipment. On the down side, DVB-SH requires makes it complicated to design the broadcasting
a very small cell radius to get sufficient transmit network.
power at 2.2 GHz to penetrate buildings.
Overall, the success of both T and S DMB
An overall comparison of the three standalone technology has been very limited because of
broadcast alternatives indicates performance capacity constraints (associated with narrow
advantages for DVB-SH. bandwidth allocations), limited CPE and
performance/quality issues. The major take-up
(d) T/S-DMB has occurred where the service offers free to air
content to mobile devices (T-DMB).
DMB technology was first developed in South
Korea and was designed to operate as either a 2. Multicast
satellite (S-DMB) or terrestrial (T-DMB) mobile
TV transmission system. Multicast distribution of mobile TV services
provides point to multipoint transmission of
In some countries, DTT and DAB broadcasters video and TV media from a single source to a
were allowed to utilize narrow bandwidths (1.5 group of users in a specific area. The key
MHz) of their spectrum for mobile TV. To distinction between broadcast and multicast is
accommodate these opportunities in the VHF that the end users must have joined the particular
spectrum (200 MHz), the T-DMB broadcast multicast group while in broadcast technology all
mobile TV standard was created by making users obtain the content. A classic illustration of
modifications to the terrestrial broadcasters DAB multicast is the delivery of radio station content
specification. over the internet.

Besides the much smaller channel bandwidth T- Figure 6: Multicast Network Architecture42
DMB does not allow for higher modulation
formats (16-QAM or 64 QAM) and has less
robust coding schemes (lacks either MPE-FEC or
turbo coding). Additionally, T-DMB lacks the
device power saving advantage of a full time
slicing architecture of other broadcast
technologies (DVB-H & MediaFLO)41.

The S-DMB system concept is based on a

combination of satellite and terrestrial
architecture for the delivery of broadcasting
digital multimedia services to mobile end users.
Because the satellite coverage provides outdoor

2008 NCTA Technical Papers - page 247

Figure 6 is a visual representation of a typical Dynamic channel broadcasting is not envisaged
cellular multicast architecture such as TDtv or yet but could increase the effectively available
MBMS. Only mobiles in a cell site interested in number of channels to 90 in the future with the
viewing a particular channel (dotted line) join the limitation then based on contribution capacity
multicast group in that cell. Other cells obtain (i.e. E3 at 34,368 Kbps at 75% utilization or
additional channels (dashed & solid lines) 25,776Kbps divided into 256 Kbps of video and
because of the desire by the mobiles in that cell 32Kbps of Audio or 286 Kbps ).
to view different content at the same time.
(b) MBMS
A multicast enabled network ensures that content
is solely distributed over those links that are The 3GPP Release 6 specification created the
serving receivers which belong to the Multimedia Broadcast Multicast Service
corresponding multicast group. This is a very (MBMS) standard. Only minor changes were
resource efficient way of delivering services to made to the existing radio and core network
larger user groups21. protocols. A new physical bearer channel that
carries the media content was created along with
(a) TDtv logical scheduling and control channels.

Many mobile operators in Europe were awarded The key for MBMS is that it can use all or a
5 or 10 MHz of TDD unidirectional spectrum portion of an existing 5 MHz HSPA radio
(1.9 to 2.0 GHz) as part of the 3G licenses won channel. TDtv is a multicast configuration that
through auctions and “beauty contests”. To date, requires dedicated TDD spectrum, cell site
very few operators have used this spectrum equipment and chipset enhancements to the
because it is unidirectional and little equipment is mobiles. MBMS requires none of that additional
available. equipment. But portions of the existing HSPA
network must be dedicated to MBMS services.
TDtv is a multicast technology that uses the For instance, if 256 Kbps mobile TV channels
existing TDD spectrum in a 3G license. Since are planned for then 32 TV channels can be
there are no TDD capable transmitters on base created in a single 5 MHz radio channel. If
stations, these have to be added, but they can go desired, only a portion of the 5 MHz is allocated
on the same towers and use the same power to MBMS while the remaining amount is used
supplies and antenna of the existing base station. for voice and data services. Additionally, 128
Using two antenna in the handset means that Kbps or 64 Kbps mobile TV channels can be set.
only 30% to 50% 43 of base stations require
transmitters. Two signals from any base stations Overall, MBMS has a capacity advantage over
in reach can combine through the antenna to give unicast when several subscribers reside in the
in-building penetration. At present operators same sector of a cell and are watching the same
have proven 15 channels in 5 MHz of TDD, but mobile TV channel. When there are very few
claim they can stretch to 28. users in a sector then a unicast architecture may
make more sense21.
Potential interference issues exist as the spectrum
sits next to existing 3G spectrum, meaning that 3. Unicast
not all of the 10 MHz may be available for use
and expensive filters may be required in related Unicast mobile TV technologies stream video to
handsets. mobile devices over various 3G wireless
technologies. Streaming video to handsets in

2008 NCTA Technical Papers - page 248

unicast has some inherent limitations that present The ongoing evolution of the 3G UMTS family
challenges relating to performance. ` of technologies, which builds on the foundations
of GSM, are listed in Table 8.
The most limiting problem for unicast mobile TV
has been the overall capacity constraints and end The original promise of UMTS/WCDMA, to
user speeds possible over existing cellular provide high speed broadband connectivity,
network technologies. Many network upgrades never occurred. Although speeds of 2 Mbps
and advancements have been made in recent were hoped for, in reality 256 to 384 Kbps were
years that begin to break the capacity limits of more typical9. Recently downlink and uplink
unicast mobile TV over 3G cellular technology. software enhancements (HSDPA and HSUPA)
have been adopted by operators worldwide that
Additionally, channel zap performance can be provides much improved performance.
challenging in relation to unicast mobile TV. In
particular, an inherent 15 to 20 seconds delay to A number of technologies have been deployed to
move from one specific channel to another make these improvements including adaptive
because the current session must be closed and a modulation and coding, fast packet scheduling
new one must be opened. and Hybrid Automatic Request (HARQ).
Adaptive modulation software analyze each end
Advances have been made in this area, that user for signal strength and determines which
enable the player to remain “alive” when modulation format (16QAM, QPSK…) and
switching from one channel to another; and coding scheme will work the best. Fast packet
keeping the video displayed when switching; and scheduling allows communication between the
finally, optimization for network conditions, that mobile device and cell site to make the most
allows for zap speeds between 3 to 8 seconds44. efficient use of the bandwidth available. In
CDMA systems the use of orthogonal CDMA
(a) 3G (UMTS/WCDMA, HSPA and HSPA+) spreading codes and time slots is critical in
allowing a device to attain its maximum data
For the purposes of this paper we will focus on rate. In the case of HSPA, devices using 5 codes
the use of mobile TV over the group of mobile allows for a maximum theoretical peak speeds of
standards to come out of the 3GPP (3rd 3.6 Mbps, 10 codes correlates to 7.2 Mbps and
Generation Partnership Project) standards body. 15 codes can theoretically attain 14.1 Mbps.

Table 8: 3GPP Specification Releases45 The combination of HSDPA and HSUPA (called
Version Released Description HSPA) is reflected in the 3GPP Release 6
Release 99 2000 Original UMTS/WCDMA specification. Only 7.2 Mbps capabilities are
3G air interface
currently available and most operators claim
Release 4 2001 Added new features from 5.0 to 6.0 Mbps speeds are attainable in
including all IP core
“real world” operation. More appropriately, the
Release 5 2002 Added HSDPA (improved
Downlink) and IMS average cell throughput capacity of a sector is
the critical design metric as this is the capacity to
Release 6 2004 Added HSUPA, (improved
Uplink) and MBMS … be shared among all users simultaneously
release is called HSPA accessing the network. For HSPA the average
Release 7 2007 Added downlink MIMO, sector throughput will range from 4 to 6 Mbps.
improved QOS and VoIP
… Release is called HSPA As is expected, much depends on the cell radius
Evolved or HSPA+ design, indoor or cell edge coverage and signal

2008 NCTA Technical Papers - page 249

strength assumptions. If it is assumed that a With GSM exceeding 2.6 billion mobile
large % of users will get great signal strength (for connections worldwide and global subscriptions
example; 200 meters from site with line of site, to all mobile network technologies exceeding 3.3
outdoor coverage, 16QAM modulation) then the billion, the 3GPP family of standards
sector capacities will be higher46. (GSM/UMTS/WCDMA/HSPA) represents over
80% of all cellular connections worldwide.
As an illustration of the variability, a single end
user in the sector (no contention with other HSPA Evolved or HSPA+ will enhance the
subscribers) with outdoor coverage could get downlink and claims to provide a theoretical peak of
between 2 and 4 Mbps service depending how 42 Mbps by utilizing 64QAM modulation and the
far away and if there is line of site to the cell uplink to 11.5 Mbps through 16QAM. A further
location. A single user indoors will typically get enhancement to help in achieving the increase data
between 800 Kbps and 2 Mbps, again depending rates is the addition of MIMO antennas, usually
on distance from the site, type of building deployed to enhance the system performance.
material and how far inside the building the user MIMO increases downlink sector capacity by
is located. For this reason most HSPA networks implementing a technique that transmits multiple
(7.2 Mbps & 10 codes type devices) are designed desired signals via separate antennas. This has the
and offer an average 1 to 3 Mbps product to effect of multiplying the amount of data that is able
subscribers. The peak rates advertised are to be transmitted over a single radio channel.
typically marketing buzz as the peak speed is
only obtainable if there is a single end user in Other features include reducing latency by keeping
service on the sector and is operating at the the devices in a different state when inactive.
strongest possible signal strength. Although the lab simulations of MIMO technology
are positive, this advancement still needs to be
Mobile broadband networks with sufficient proven out in the challenging RF environment of
capacity to provide for some unicast mobile TV mobile devices in a live network.
applications are becoming more prevalent
throughout the world. While HSPA+, with its MIMO capability, can
provide a capacity improvement over 3G and
For instance, there are 165 HSPA networks in HSDPA, the upgrade is not cost free to the
place and a device ecosystem of 465 different mobile operator. In order to support MIMO
devices from 102 suppliers currently available additional capital costs associated with antenna
around the world10. According to industry and installation, radio planning; in addition to
research, UMTS/WCDMA/HPSA is the world’s additional operational expenditure for the site
most popular 3G cellular technology as it rental associated with the additional antenna.
represents more than 200 million customers Additionally the upgrade from 3G to higher
worldwide. Almost 20 million are already network bandwidths requires the buildout of
subscribers to high-speed HSPA mobile larger capacity backhaul.
broadband networks and this number is expected
to double by the end of 200811. Nevertheless, (b) 4G (LTE and WiMAX)
penetration of high bandwidth cellular
connectivity is still emerging and needs to Long Term Evolution or LTE (3GPP Release 8)
develop further. For example, only 13% of and Mobile WiMAX (802.16e) are emerging as
homes in the U.S. are 3G capable today. 47 4th generation standards being specified to offer
very large average sector throughputs (20 to 40
Mbps) and as such could be impactful to the

2008 NCTA Technical Papers - page 250

application of Unicast Mobile TV applications
over broadband cellular technologies. Both The measurement of the cell radius calculations
technologies utilize OFDMA technology, assumed a common modulation scheme being
incorporate MIMO antenna and transmission received (16QAM ½) by the device, a fixed
gains and wide channel bandwidths (10 and 20 maximum quantity of mobile TV channels and
MHz) that contribute to large capacity an overall probability (90%) of obtaining a
improvements. minimum signal level. The conclusion from the
trial is that coverage can vary widely (from 1.45
Some research forecasts that LTE should be Km to 9.25 Km) as different types of mobile
going commercial by 2010 and represent around devices, their location and speeds are changed.
24 million subscribers globally by 201215. To illustrate further the complexity of the
wireless tradeoff’s typically encountered we
The research also predicts that HSPA will changed the transmit power and antenna height
dominate mobile broadband network for the Flanders test. Reducing the transmit
deployments by 2012, consistently accounting power to 2 kW and the antenna height to 30
for about 70% of the total mobile broadband meters for an indoor handheld moving at 3
subscriber base. LTE and Mobile WiMAX are Km/hr caused a severe reduction in cell radius
expected to achieve only a small proportion of from 3.35 Km to 1.65 Km54.
the 1.2 billion total mobile broadband subscriber
base48. How is the capacity of a mobile TV technology
4. Technology comparison across coverage, determined?
capacity, and mobility variables
The ability of the mobile TV device to receive a
As is typical with wireless technologies, strong enough signal in the presence of
understanding the tradeoffs among coverage, interference is a critical determinant of capacity.
capacity and mobility characteristics is critical Qualcomm has conducted numerous tests
when assessing mobile TV technologies. The illustrating the relationship of sufficient signal
key to comparing these characteristics is having strength received by the mobile TV device and
knowledge of the key assumptions that drive overall system capacities for their MediaFLO
coverage, capacity and mobility performance technology. Figure 7 below shows the
results. relationship between the maximum numbers of
256 Kb/s H.264 mobile TV channels possible for
For instance, the coverage claims of broadcast a given received signal at the mobile handheld.
technologies such as MediaFLO and DVB-H The signal strength is represented as a Carrier to
vary widely as a number of assumptions are Noise ratio (C/N) in decibels.
changed. In our Flanders DVB-H field trial
many scenarios were tested. The results of one Figure 7: Capacity vs. Signal Strength49
scenario are illustrated in Table 9 below.
MediaFLO QVGA TV Channels with Stereo Audio
vs. Planning C/N in 6 MHz Allocation
54
Table 9:DVBH Coverage in Flanders Trial 30

Suburban Location, CPE Type & Mobility/Portability

25
terrain, 5kW
Channel Count

transmit Outdoor, Indoor, In-car, 20

power & 60m Handheld, Handheld, 3 Handheld,

antenna 3Km/hr Km/hr 70 Km/hr 15

heights 10 Aggres sive

Cell Radius Cons ervative

9.25 3.35 1.45
(Km) 5
3 4 5 6 7 8 9 10 11 12 13

C/N

2008 NCTA Technical Papers - page 251

Extensive testing has revealed that a 10 dB C/N 400 MHz). Mobile TV standards such as
ratio is an attainable signal strength under a MediaFLO and DVB-H attempt to minimize this
number of typical mobile TV conditions. As can problem by using various advanced error
be seen from the graph a 10 dB C/N results in a correction coding, modulation formats and
capacity of 22 Mobile TV channels. OFDM sub carrier schemes.
Consequently, the MediaFLO network is
designed for a capacity of 22 mobile channels A considerable amount of high speed mobility
operating at 256 Kb/s per channel. To meet this testing has been conducted at 850 MHz for
capacity figure the key design criteria used to DVB-H mobile TV networks and is illustrated in
determine transmitter locations for the mobile one design scenario in Figure 8. Figure 8
broadcast network will be to meet the 10 dB illustrates the relationship of mobile speeds along
signal strength level in a majority (90%) of the the x-axis (represented as Doppler shift
locations the network serves. frequencies in Hz) and mobile receive signal
strength (C/N ratios) along the y-axis for the
A common question asked of mobile TV services DVB-H mobile TV specification.
and technologies is what do we really mean by
mobility?

Mobile TV terminals are expected to be usable in

stationary, pedestrian walking speeds, and high
speed in-vehicle applications that encompass
both outdoor and indoor environments.
Delivering a consistent high speed video signal
to a low power, low gain handheld device in
motion is a difficult technical challenge58.

One of the key issues with fast moving mobile

devices is the concept of multipath propagation.
In effect, the radio signal from the cell site Figure 8: Mobile speeds vs. signal strength57
transmitter takes multiple paths to reach the
mobile device which results in “echoes” that The graph shows that as long as a minimum
make it difficult for the mobile to recover the signal strength can be maintained (11.2 dB) then
video transmission. These echoes cause the mobile devices speeds from 3 Km/hr (at 10 Hz
digital video information (or symbols) to “blur” on the y-axis) to 126 Km/hr (corresponds to the
across each other creating severe problems called 100 Hz point on the y-axis) will support mobile
inter-symbol interference. TV services. It is interesting to note that at the
Doppler frequency of 174 Hz (equivalent to ~200
Unfortunately, handheld devices in high speed Km/hr) a sharp increase in C/N signal strength is
motion amplify the inter-symbol interference required. This means that at speeds greater than
reception issue. This effect is called the Doppler 200 Km/hr a nearly impossible signal strength
shift. Simply put, the faster a mobile is moving would be required to be received by the fast
the more of a Doppler shift occurs which moving mobile.
translates into a greater interference effect on the
mobile device receiver. The Doppler shift The major mobile broadcast technologies (DVB-
interference effect is more noticeable at mobile H, MediaFLO and DVB-SH) are designed to
systems operating at higher frequencies (e.g.- > operate in a high speed mobile environments (<
200 Km/hr). Likewise, multicast and unicast

2008 NCTA Technical Papers - page 252

technologies, being required to serve traditional Table 10:Unicast Mobile TV Cost/Sub/Month 53
mobile voice and data applications, are capable HSPA+ HSPA UMTS/
of operating at high speed vehicle and train WCDMA
speeds as well. 256 Kbps $7.66 $24.42 $30.21
@ 20m
THE MOBILE TV BUSINESS MODEL: A 128 Kbps $1.28 $4.07 $5.04
COMPLIMENT TO FIXED LINE @ 6m
BROADCAST 128 Kbps $0.38 $1.22 $1.55
@ 2m
What is the cost of extending fixed-line We assumed quality and viewership parity when
broadcast to mobile broadcast? Are the costs comparing unicast mobile TV to multicast or
different for a mobile operator? We explore the broadcast mobile TV, with all platforms
economics across multiple mobile TV platforms. delivering 256Kbps encoded video and 32Kbps
encoded audio with average viewing times of 20
Developing the network “Pain Threshold” minutes to match the viewership studies
referenced earlier.
Using revenue and operating assumptions from
earlier sections of this paper and reference Under differing quality and viewership
literature we can determine the network “pain conditions unicast mobile TV results in varying
threshold” for Mobile TV network economics. degrees of network congestion depending on the
network evolution technology deployed. We
We assume that the average revenue from mobile assumed an average sector throughput of 2.5
TV is $20 per month per subscriber50. Of this Mbps for UMTS/WDMA, 6.0 Mbps for HSPA
$20 per month we assume that 45%51, or $9 per (7.2) and 9.5 Mbps for HSPA+ technologies.
month, goes to mobile TV content costs and $8 is Additionally, an urban market density of 1,500
required for sales and marketing, billing and Pop’s per Km2, 0.57 Km cell radius and 3 sector
G&A per subscriber per month, about half of the cell sites assumptions were used.
amount for mobile data 52. This leaves $3 per
month per subscriber to cover all network related Clearly an early 3G network using
costs. UMTS/WCDMA would be unable to support
mobile TV services at 256Kbps and 20 minute
1. Unicast economics average view times! Although the evolution of
3G to HSPA, HSPA+ and ultimately LTE reduce
Initially considering the economics of unicast the probability of congestion, we need to take
mobile TV we find that the economics quickly into consideration that other services are also
pass the pain threshold of $3! Unicast operating on the same network. Consumption of
economics appear suited to low quality, short- 3G and HSPA networks for high speed data is
clip, long tail content that have low bit-rates and increasing. In the United Kingdom, Vodafone is
short view times rather than high quality reporting 50% to 60% of its available 3G
premium content with longer average view times. capacity is being used for data in dense urban
areas54. Therefore in this analysis we assume
50% of the sector capacity is allocated to data
services.

2008 NCTA Technical Papers - page 253

Table 11: Unicast Mobile TV Blocking TV service at 40-60% penetration.
Probability @ 15% Penetration
HSPA HSPA UMTS/ Our analysis of blocking probabilities also
+ WCDMA provide insights in relation to capacity advances
384 Kbps @ 0.00 0.26 0.65 for mobile TV. We can see in Figure 10 below
20m that new wireless technologies can enable Telcos
256 Kbps @ 0.00 0.00 0.49 and independent wireless providers to deliver
20m mobile TV at greater penetration rates.
128 Kbps @ 6m 0.00 0.00 0.00
128 Kbps @ 2m 0.00 0.00 0.00 Figure 10: Mobile TV Capacity
Capacity For Unicast Mobile TV
256Kbps - 20min Avg Viewing Time - 50% Reserved for Data

We can see from this analysis that low bit-rate, 500%

LTE
short clips have less impact on unicast mobile WIMAX
HPSA+
3G
HSPA

network congestion that high bit –rate, long form 400%

HSPA
3G

content, indicating that the mobile network can 300%

Maximum

HSPA+

Capacity
support short clips such as you-tube like rich-
data content. 200%
WiMAX

LTE
100%

Because the traffic is unicast as the penetration Maximum Capacity

increases to 30% the blocking probability 0%

10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

increases to the point where our 256k, 20 minute Mobile TV Penetration

scenario has blocking probability of 0.47 for

HSPA and 0.74 for 3G … put another way 74% It is also important to point out that if a mobile
of 3G mobile TV subs could not get the service, operator has sufficient spectrum to add another
in addition to there being no room for growth in 10 MHz channel to their network and reasons
data services! that there is sufficient return to allocate it to
video services then these capacities can be
Figure 9: Unicast Blocking Probability increased further and therefore provide for
Blocking Probability for Unicast Mobile TV improved mobile TV capacities.
256Kbps - 20min Avg Viewing Time - 50% Reserved for Data
1

The importance of a dense cellular network for

LTE
0.9 WIMAX
HPSA+
0.8
HSPA
3G
3G
delivering unicast mobile TV is illustrated in
0.7 HSPA
figure 11 below. Using LTE as a base we
Blocking Probability

0.6

0.5
HSPA+ illustrated the impact of larger cell sizes, noting
0.4
WiMAX
that an increase in cell radius from 0.57 to 0.80
0.3

0.2
LTE
kilometers results in a halving of non-blocking
0.1
mobile TV penetration potential!
0
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Mobile TV Penetration

As seen in figure 9, as wireless technologies

evolve, reasonable penetration levels of mobile
TV could be expected if 15 Mb/s (WiMAX) and
20 Mb/s (LTE) sector capacities are obtainable.
Our analysis indicates future WiMAX and LTE
technology could provide non-blocking mobile

2008 NCTA Technical Papers - page 254

Figure 11: Mobile TV Blocking by Cell Radius bearer requires more radio capacity. We assume
Mobile TV Blocking Probability for LTE by Cell Radius that about 13% of the Node-B (Base Station)
1
256Kbps - 20min Avg Viewing Time - 50% Reserved for Data
power is required for Multicast, assuming that
0.9 soft combining is not enabled (possible reduction
0.8
2.00Km to 6%)55.
0.7
Blocking Probability

1.50Km
0.6
1.00Km
0.5 Table 12: Multicast Mobile TV Blocking
0.80Km
0.4 Probability (256Kbps, 20 minutes view time,
0.3

0.2
30% penetration)
HSPA+ HSPA UMTS/
0.57K
0.1 m

0
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
WCDMA
Mobile TV Penetration Unicast 0.01 0.47 0.74

50% view 0.0 0.0 0.31

2 Channels
We can also show this in capacity potential, 60% view 0.0 0.0 0.21
where a cell size of 0.57Km exhausts capacity 2 Channels
for mobile TV at 60%, a cell size of 0.80Km 70% view 0.0 0.0 0.11
exhausts capacity for mobile TV at just 30% 2 Channels
penetration (assuming a constant market
density). Figure 13: MBMS Blocking Probability
Blocking Probability for Unicast and MBMS
256Kbps - 20min Avg Viewing Time - 50% Reserved for Data

Figure 12: Mobile TV Capacity by Cell Radius 0.9

Unicast HSPA
0.8 MBMS HSPA - 50% view 2 CH
Capacity for Mobile TV for LTE by Cell Radius
MBMS HSPA - 70% view 2 CH Unicast HSPA
256Kbps - 20 Min Avg View Time - 50% Reserved for Data 0.7
400%
Blocking Probability

0.6
2.00K
m 1.50K 0.5
m
300% MBMS HSPA - 50% view 2CH
1.00K 0.4
m
0.3
Capacity

0.80K MBMS HSPA - 70% view 2CH

200% m 0.2

0.1
0.57K
0
100% m
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Mobile TV Penetration

0%
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Mobile TV Penetration

As seen in Figure 13 above, unicast HSPA

mobile TV begins to suffer from blocking when
penetration exceeds 15%, where as adding
2. In-band multicast economics
MBMS can enable support higher penetration
rates.
Considering In-band MBMS, or Multicast
assumptions to the cellular models without
However, because enabling MBMS increases
extending available spectrum, we assumed that
power usage, and the reduction in blocking
50% of the users viewing time was for 2
probability is mitigated by the existence of other
channels, and accounted for up to 80% of the
services (i.e. mobile data, with its speed and
viewing time for those channels.
subscriber growth) it appears to be inefficient to
deploy in-band with other IP services until LTE
We also assume that for in-band multicast over
becomes available.
3G/HSPA/LTE networks that a mobile broadcast

2008 NCTA Technical Papers - page 255

Based on our analysis of network congestion, we Table 13: Network Cost of Mobile TV
assume that mobile operators would not consider Network Cost Channels
enabling MBMS as an in-band capability for Per Sub Per Supported @
some time, but would rather focus on deploying Month @ 256Kbps
MBMS capabilities in separate dedicated unicast 256Kbps @
spectrum (i.e. TDtv). 15% pen.
MediaFLO $3.10 22
3: Broadcast overlay network economics DVB-H $3.72 18 58
DVB-SH $1.70 18t & 9ts59
We used a specific propagation model based on a TDtv $1.87 15 – 28, Uni.
trial in Gent, Belgium 56 57to determine the site HSPA+ $7.66 Unlimited
radius for typical European DVB-H HSPA $24.42 Unlimited
deployments, and validated this information with 3G $30.21 Unlimited
other reference deployments in Europe.
Figure 14: Mobile TV “Pain Threshold” and
Our MediaFLO analysis was based on a Network
reference architecture for Chile. Results were Costs
reduced to a cost per kilometer square basis to Mobile TV Network Cost Per Subscriber Per Month

determine the comparable cost of coverage to $35

Network Cost Per Sub Per Month @ 256Kbps @ 15% pen.

Europe. $30
US$ Cost Per Subscriber Per Month

Pain Threshold

$25

4. Hybrid network economics $20

$15

With emerging hybrid mobile TV architectures

$10

we used information based on a reference UK

$5

$0
model for DVB-SH and assumed a 6dB to 8dB DVB-SH TDtv MediaFLO DVB-H LTE HSDPA 3G

gain over DVB-H providing approximately a

factor x2 improvement in coverage area. CONCLUSION
Out TDtv analysis was based on industry Based on our analysis we believe that the market
information including the assumption that 40% for and penetration of mobile TV is relatively
of cell sites require the TDtv transmitter, and elastic to price, with pay markets seeing 10% to
used industry literature to determine the capital 20% penetration and free markets seeing about
cost of the network extension. 60% penetration.
Comparing the alternatives, what technologies If Mobile Network Operators (MNO) or
are below out pain threshold of $3 per subscriber Multiple Service Operators (MSO) gave away
per month? mobile TV as a bundled offering the experience
in Japan tells us we would require capacity for
60% penetration and require broadcast overlay
solutions such as DVB-H and DVB-SH, or need
to wait for high capacity mobile pipes (i.e. LTE)
and/or multicast technologies.

On the other hand, cable MSOs or Telcos/MNO

considering charging for mobile TV as a pay

2008 NCTA Technical Papers - page 256

offering, could expect penetration between 10% have a large advantage over MSO’s and TV
and 20% by 2012. This type of penetration rate broadcasters.
is more suited to in-band HSPA, HSPA+,
MBMS and ultimately LTE alternatives for It is highly unlikely that mobile TV content and
mobile TV. Subsequently, today’s Telcos with viewership will steal subscribers from the in-
mobile operations, are in a better position to home TV viewing revenues of cable operators
deliver a mobile TV pay service as this but TV viewing time could be impacted. Any
alternative does not require a separate overlay opportunity of competitors to gain a foothold on
network. video and TV viewing time and habits, content
aggregation and user interfaces could be deemed
We can present the alternatives for the cable a viable mid to long term threat. One area to
MSO as follows: monitor closely in this area may be the
Table 14: Cable MSOs Mobile TV Choices technology advancements such as pico-projectors
Mobile TV Free Mobile TV Pay being built into mobile phones.
Bundle Service
Penetration ~ 60% ~10 – 20%
Therefore, given the tough barriers to entry in the
Technology S/T-DMB UMTS, HSPA mobile TV space, combined with advances in
Options (Limited (Limited mobile technology, it is the opinion of the
Eliminated Capacity) Capacity) authors that mobile TV is more of a threat than
Technology MediaFLO, HPSA+ opportunity for cable operators. In particular,
Options DVB-H/SH TDtv, MBMS MNOs that serve the low end cable TV base with
Today (Spectrum) (MNO/Telco)
Technology WiMAX, WiMAX,
free to air mobile TV and are able to complement
Options LTE, LTE, this with premium content delivered using either
Emerging (MNO/Telco) (MNO/Telco) broadcast or higher capacity in-band wireless
could be threatening.
In view of Table 14 it is less clear how the many
mobile TV distribution choices are suited to a On the other hand, MNO’s may need
cable operator that does not own suitable considerable assistance on the technical and
spectrum or a wireless network asset, in view of economic aspects of content acquisition,
challenging economics and fragmentation of management, aggregation, rendering &
spectrum options. distribution. An MSO’s ability to leverage this
strategic advantage into an effective partnership
A summary of the economics of the different with MNO’s is the most viable option for cable
facilities-based mobile TV alternatives operators. This may be particularly attractive to
determined that the economic margin for error is the MNO where the significant cost for mobile
low, flexibility for an sustainable deployment is TV content can be reduced through partnering
limited and that the selection of an economically with the MSO.
viable distribution technology for the delivery of
mobile TV is critical. Should MNO partnerships prove challenging,
cable operators that strongly desire a mobile TV
Given the inherent low margin aspects of the strategy need to consider the possibility of
mobile TV business, the tight linkages to unicast acquiring or building out a multi-service cellular
video (and data) consumption over existing wireless network to facilitate sustainable mobile
cellular data networks and the device centric TV economics. The build or buy path appears
nature of mobile TV, Mobile Network Operators overly aggressive if mobile TV is the only
economic driver.

2008 NCTA Technical Papers - page 257

An MSO going it alone in the broadcast area 9. Resolution Requirements for Mobile
(MediaFLO & DVB-H) appears risky as well. At TV”, 2005, p.836
the end of the day mobile TV using broadcast 10. The Online Reported, 23-29 Feb 2008,
technology requires devices. Therefore broadcast Issues 577-I, p.14.
technologies need a mobile operator’s network 11. DVB-H is a registered trademark of the
and mobiles. An MNO operator is a formidable DVB-H Project
competitor in the mobile TV market, with many 12. MediaFLO USA, and FLO are
distribution options in the toolkit, including trademarks of Qualcomm Incorporated
established multicast and unicast delivery options 13. Crawford, “Spectrum for Multimedia
that may not be perfect, but allow the MNO to Services” 2006, page 4
evaluate what content works for mobile TV as 14. Rethink Research, Faultline: European
the market emerges. CellCo’s act rather than wait for DVB-H
spectrum”, March 2008
REFERENCES 15. Lombardi,Qualcomm, “Presentation to
U.S. Subcommittee on Telecom &
Internet”, May 10, 2007, Pages 6 & 7
1. Yoram Soloman, “The Economics of 16. Joseph, Plets, Martens “DVB-H
Mobile TV”, 2007, p.1 Broadcast Network Design for indoor
2. Yoram Soloman, “The Economics of reception of DVB-H in Flanders” 2007,
Mobile TV”, 2007, p.2 page 3
3. The "frame rate" of interlaced systems is 17. Walker, Qualcomm, Presentation to
usually defined as the number of ATSC, “Technical and Business
complete frames (pairs of fields) Challenges of Mobile TV” May 17, 2007,
transmitted each second (25 or 30 in most Page 3
broadcast systems). For example 25p is a 18. Lombardi,Qualcomm, “Presentation to
video format which runs twenty-five U.S. Subcommittee on Telecom &
progressive (hence the "P") frames per Internet”, May 10, 2007, Pages 6 & 7
second. This frame rate is derived from 19. Joseph, Plets, Martens “DVB-H
the PAL television standard of 50i (or 25 Broadcast Network Design for indoor
interlaced frames per second). reception of DVB-H in Flanders” 2007,
4. Yoram Soloman, “The Economics of page 3
Mobile TV”, 2007, p.2 20. Erkki Aaltonen, Nokia, “DVB-H Radio
5. Hendrick Knoche & John D. McCarthy, Network Aspects”, July 2005, page 24
“Design Requirements for Mobile TV”, 21. Wi-Fi® is a registered trademark of the
2005, p.75 Wi-Fi Alliance
6. Hendrick Knoche & John D. McCarthy, 22. Perceived usage rather than measured
“Design Requirements for Mobile TV”, usage. Ericsson paper, “Changing The
2005, pp.74-75 Way We Look at Television”, p.22
7. Hendrick Knocke, John D. McCarthy, M. 23. Mitch Feinman, Fox Mobile
Angela Sasse, “Can Small Be Beautiful? Entertainment, “NATPE 2008
Assessing Image Resolution Conference Sessions”, “Mobile Content:
Requirements for Mobile TV”, MM05, What’s Hot, What’s New? What’s next?”
Nov 6-11 2005, p.837 24. Salil Delvi, NBC Universal, “NATPE
8. Hendrik Knoche, John D. McCarthy & 2008 Conference Sessions”, “Mobile
M. Angela Sasse, “Can Small Be Content: What’s Hot, What’s New?
Beautiful? Assessing Image What’s next?”

2008 NCTA Technical Papers - page 258

25. Crawford, “Spectrum for Multimedia (260 million) of the 2012 number. The
Services” 2006, page 4 Cable & Satellite Broadcast Association
26. Kaufhold , “US Consumers Attitudes (CSBAA) provides better granularity to
About Mobile Communications & these numbers as they track broadcast
Entertainment” , In-Stat, September, 2007 only mobile TV numbers. Their studies
27. All values originally in Euro, converted project 76 million broadcast subscribers
to USD assuming 1.44 USD/EURO. by 2012. From these two projections one
Urban, “Mobile Television: Is it just can conclude that a vast majority of Asia
Hype or a real Consumer Need” Pacific (and probably worldwide) Mobile
Observatorio Journal,3 (2007) 045-058, TV usage will be over unicast
Page 53 technology.
28. DVB-H Web Site, “3 Italia – Italy 32. Adapted from Yoram Soloman, “The
Services”, May 2007, http://www.dvb- Economics of Mobile TV”, 2007, p.10,
h.org/Services/services-Italy-3Italia.htm. figure 8.
Euro 29 per month at 1.44 EUR to USD 33. Herbert Mittermayer, “Fast Channel
29. David Sidebottom, Understanding & Switching Description”, Alcatel-Lucent,
Solutions, Mobile TV Forecast Update p.1
based on 3GSM 2008 Review, Feb 22, 34. Gallouzi “Deal with OFDM, a new old
2008. technology” July, 2007, Page 2
30. Understanding & Solutions define mobile 35. Wi-Fi is a registered trademark of the Wi-
TV in two ways, dedicated mobile TV Fi Alliance
whereby a dedicated broadcast 36. WiMAX is a trademark of the WiMAX
technology chip is required in the handset Forum.
to receive programming and includes 37. Gallouzi “Deal with OFDM, a new old
DVB-H, Mediaflo, ISDB-T and S/T- technology” July, 2007, Page 2
DMB standards (plus others). The second 38. Gallouzi, Brew 2005
method is cellular mobile TV delivery, Conference,“MediaFLO 101: FLO
which is linear TV programming Technology” June 2005, Pages 23 -25
delivered direct to the mobile handset via 39. Faria, Henriksson, Stare, Talmola
the operators cellular networks. The data “DVB-H: Digital Broadcast Services to
and analysis was complied via a mixture Handheld Devices” 2006, Page 2 & 3
of primary and secondary research. 40. Herbert Mittermayer, “Unlimited Mobile
Company financials and reports and trade TV, DVB-SH A Natural Evolution of
press sources are supplemented with DVB-H”, Alcatel-Lucent, Jan 2008, p.14
dedicated face to face and telephone 41. Discussion with Juan-Pablo Torres,
research with the key players in these Alcatel-Lucent, Feb 2008
sectors. Total market estimates are 42. Digital Video Broadcasting Project,
derived from a “bottom-up” approach, “System Comparison of T-DMB vs.
with individual services totaled to create DVB-H, 2006, Pages 3-5
a total market estimate. 43. Bakhuizen, Horn “Mobile
31. A late 2007 ABI research study projects broadcast/multicast in mobile networks”,
462 million mobile TV subscribers by 2005
2012. This number assumes a 44. Rethink Research “Wireless Watch: In-
combination of Unicast and Broadcast depth Analysis of WLan, Cellular and
technologies. Interestingly Asia Pacific Broadband Wireless Markets”, Vol. 5,
has the most growth and represents 56% issue 46., Feb 15th 2008, p.19

2008 NCTA Technical Papers - page 259

45. Herbert Mittermayer, “Fast Channel 58. D. Plets, W. Joseph, E. Tanghe, L.
Switching Description”, Alcatel-Lucent, Verloock, L. Martens, “Analysis of
Feb 2008 propagation of actual DVB-H signal in a
46. Ericson Technology Paper, “Technical suburban environment”, IEEE
Overview and Performance of HSPA”, International Symposium on Antennas
June 2007, Page 6 and Propagation, Honolulu, Hawaii,
47. Ericsson Presentation, “Mobile Features USA, Paper No. 1386, 10 – 15 June 2007.
to increase HSPA performance”, July 59. DVB-H channel based on 16QAM, 18
2007, Page 1 channels in 8MHz of UHF spectrum.
48. Louis Gump, The Weather Channel 60. Configuration based on (a) MUX of
Interactive, “NATPE 2008 Conference 5MHz both via satellite and terrestrial,
Sessions”, “Mobile Content: What’s Hot, using QPSK = 9channels, nationwide;
What’s New? What’s next?” and either (b) MUX of 5MHz via
49. Juniper Research, “Mobile Broadband terrestrial, using QPSK = 18 channels
Markets WIMAX, EV-DO, HSPA & (9channels each MUX), terretrial indoor
Beyond 2007 -2012” November 28, 3G-like coverage; or (c) MUX of 5MHz
2007, Page 7 via terrestrial, using 16QAM = 36
50. Walker, Qualcomm, Presentation to channels (18channels each MUX),
ATSC. terretrial indoor 3G-like coverage. We
51. Ericsson paper, “Changing The Way We assume QPSK in our example to create a
Look at Television”, p.23 comparison to DVH-H. Same number of
52. Michel Grech, Alcatel-Lucent Mobile TV channels with DVB-SH using half of the
Business Case (DVB-SH) UK Model” repeaters.
Jan 2008, p.14 61. TeamCast and DiBcom, Mobile
53. Qualcomm paper “The Economics of Performance Calculator, 2005,
Wireless Data” p.16 http://www.teamcast.com/en/maj-
54. Economics are based on a snapshot and e/c2a2i12445/support/dvb-h-
do not take into account potential calculator/dvb-h-calculator.htm
declines over time or volume discounts. 62. IBC Show, TeamCast, “DVB-H: Digital
55. Iain Morris, “Telecommunication TV in the Hands”, Sept. 2005, page 2
Magazine”, Jan 29, 2008
56. Frank Hartung, Uwe Horn, Jorg Huschke,
Markus Kampmann, Thorsten Lomhar,
Magnus Lundevall, “Delivery of
Broadcast Services in 3G Networks”,
IEEE Transactions on Broadcasting, Vol
53, No. 1, Mar 2007, p.194
57. D. Plets, W. Joseph, L. Martens, E.
Deventer, and H. Gauderis, “Evaluation
and Validation of the Performance of a
DVB-H Network”, 2007 IEEE
International Symposium on Broadband
Multimedia Systems and Broadcasting,
Orlando, Florida, USA, March 28 – 29,
2007.

2008 NCTA Technical Papers - page 260

 MPEG-4 TRANSITION USING SWITCHED DIGITAL VIDEO
John Schlack
Motorola, Inc.

Abstract BACKGROUND

During the transition from MPEG-2 to Cable operators currently deliver a mix of
MPEG-4, it will not be possible, due to analog video and digital video on the cable plant.
bandwidth constraints, to broadcast all channels The digital video includes a large number of
in both MPEG-2 and MPEG-4 format. However, standard definition (SD) channels as well a much
the switched digital video (SDV) system can be smaller number of high definition channels (HD).
used to manage the transition from MPEG-2 to The operators also provide other services such as
MPEG-4 and to minimize the bandwidth needed high speed data, video on demand (VOD), and
by the system until the transition completes. voice over IP (VOIP).

The switched digital video system delivers The cable industry standardized delivery of
streams with an encoding type based on the digital video content using the MPEG-2 format.
decoding capabilities of the settop boxes tuned to This format requires about 3.75 Mbps of
that channel. The SDV system may need to force bandwidth for SD programs and 15-19 Mbps for
tune settop boxes or use transcoding to handle HD programs. A 6 MHz QAM channel
transitions between different encoding types. modulated using QAM256 can carry 10 SD or 2
HD programs. Cable operators may use
TERMINOLOGY statistical multiplexing to groom the channels.
This allows a 6 MHz QAM channel carry 11-12
The terms H.264, AVC (advanced video SD programs or 2-3 HD programs.
coding), and MPEG-4 Part 10 all refer to the
same standard for video compression. These Figure 1 shows an example 860 MHz HFC
terms may be used interchangeably in this plant carrying these different services. This
document. Additionally, this document uses the example system delivers 50 HD channels.
term MPEG-4 to refer to MPEG-4 Part 10. However, many cable plants are 750 MHz or
MPEG-4 provides high quality video at less. These cable systems deliver much less HD
substantially lower bit rates than its predecessor, programming.
MPEG-2.

Upstream Downstream

5 MHz 42 54 MHz 504 MHz 654 MHz 774 MHz 858 MHz

Data MPEG-2 SD MPEG-2 HD

Analog TV Broadcast
& 275 Channels 50 Channels
75 Channels @ 6 MHz each
VOIP 25 QAMs 20 QAMs

PPV Music Data & VOIP VOD SD+HD

1 QAM 1 QAM 4 QAMs 8 QAMs
Figure 1: Example 860 MHz HFC Plant

2008 NCTA Technical Papers - page 118

 Cable operators are being pressured to deploy number by the end of 2008. Cable operators
additional HD channels. Subscribers are currently deliver 20-50 HD channels [2]. The
replacing standard definition television with HD cable operators need to deploy more HD
sets. This is one source driving the need for channels to keep pace.
additional HD content. Figure 2 shows the
current and projected growth of HD sets [1]. The industry has begun a shift towards
encoding video using MPEG-4 as a method to
reduce bandwidth for delivering HD content via
70
satellite. As an example, a major content
60 provider has announced that it will deliver the
HDTV versions of all 26 channels to cable
HDTV Units (millions)

50
headends using an MPEG-4 encoding [3].
40 MPEG-4 video requires about 30-50% less
30
bandwidth than comparable MPEG-2 content.

20 The settop box manufacturers are also

10 beginning to deliver settop boxes that can decode
MPEG-4 video. Using MPEG-4 instead of
0
2004 2005 2006 2007 2008 2009
MPEG-2 to encode all SD and HD programs will
Year
produce significant bandwidth savings. It will be
possible to deliver 100 channels of HD
Figure 2: HDTV Growth programming and 275 channels of SD
programming on an 860 MHz cable plant while
Competition is also driving the need to deploy still keeping a large analog channel lineup for
more HD content. Satellite providers currently subscribers without digital settop boxes. The
deliver 70-90 HD channels and plan to increase bandwidth calculations are shown in Figure 3.
the number to over 100 channels by the end of
2008. Telco operators plan to deliver a similar

Format Channels BW
Analog Analog 75 450
Digital SD MPEG-2 275 150
Digital SD AVC 275 78
Digital HD MPEG-2 100 240
Digital HD AVC 100 150
VOD SD + HD MPEG-2 48
VOD SD + HD AVC 36
Data + VOIP 24
Music MPEG-2 6
PPC MPEG-2 6
Total 1188

Format Channels BW Format Channels BW

Analog Analog 75 450 Analog Analog 75 450
Digital SD MPEG-2 275 150 Digital SD AVC 275 78
Digital HD MPEG-2 50 120 Digital HD AVC 100 150
VOD SD + HD MPEG-2 48 VOD SD + HD AVC 36
Data + VOIP 24 Data + VOIP 24
Music MPEG-2 6 Music AVC 6
PPV MPEG-2 6 PPV AVC 6
Total 804 Total 750

Figure 3: MPEG-4 Transition without SDV

2008 NCTA Technical Papers - page 119

 However, most digital settop boxes deployed
today in a cable system can only decode One can also view MPEG-4 encoded channels
MPEG-2 encoded content. It is not realistic to as long tail content while the number of MPEG-4
replace millions of digital settop boxes with capable settop boxes is less than the number of
MPEG-4 capable settop boxes in a short time MPEG-2 capable settop boxes. Thus, moving
frame. A transition period is required where the the MPEG-4 channels onto the switched tier will
cable system delivers both MPEG-2 and prove to be a very effective way to introduce
MPEG-4 content. During the transition, it will MPEG-4 onto the cable plant.
not be possible, due to bandwidth constraints, to
broadcast all channels in both MPEG-2 and From the example in Figure 1, an operator
MPEG-4 format. The “Transition Period” table may choose to move the least watched 100
in Figure 3 shows the bandwidth required to dual standard definition (SD) channels and 20 high
carry MPEG-2 and MPEG-4 versions of each definition (HD) channels from digital broadcast
channel as well as expands the HD channel to the switched digital video tier. These channels
lineup to 100 channels. originally required 16 QAMs to broadcast. SDV
requires approximately 8 QAMs to deliver this
Many of the cable operators are turning to content, meaning that 8 QAMs have been
switched digital video (SDV) as a way to reclaim reclaimed.
bandwidth in order to deliver new services,
particularly additional high definition content. Assume that the 8 reclaimed QAMs will be
Switched digital video replaces traditional used for delivering more HD content on the SDV
broadcast programs with a system that only tier. These 8 QAMs can be used to deliver 30-40
transmits a channel to a service group when additional “long tail” HD channels. Transmitting
requested by a subscriber. The system realizes both MPEG-2 and MPEG-4 format for all
bandwidth savings since only a subset of the channels on the switched tier provides additional
available channels is being watched by bandwidth savings as MPEG-4 capable settop
subscribers at any given time. boxes are deployed. Figure 4 shows the example
cable plant using SDV to deliver MPEG-2 and
The operator usually places niche or low take MPEG-4 encoded content. The actual mix of SD
rate channels on the SDV tier. When using this and HD to deliver on the SDV tier will vary by
“long tail” content, the SDV system can operator and region based on popularity of the
effectively offer at least twice the number of content.
channels than can actually be delivered in a
given QAM. The reclaimed QAMs from digital
broadcast can be used to offer additional
channels or can be assigned to other services.

Upstream Downstream

5 MHz 42 54 MHz 504 MHz 600 MHz 672 MHz 768 MHz 852 MHz

MPEG-2 SDV
Data MPEG-2 SD
Analog TV Broadcast HD MPEG-2/4
& 175 Channels
75 Channels @ 6 MHz each 30 Chan SD + HD
VOIP 16 QAMs
12 QAMs 16 QAMs

PPV Music Data & VOIP VOD SD+HD

1 QAM 1 QAM 4 QAMs 8 QAMs
Figure 4: 860 MHz HFC Plant with SDV

2008 NCTA Technical Papers - page 120

 The switched digital video system is an ideal discover the settop capabilities and assign the
platform for launching MPEG-4 channels. correct stream. Alternatively, when the settop
However, challenges exist. The SDV system box registers with the SDV system, it may send a
must deliver multiple encoding formats for the message to the SDV system to report its
same content, ensure that the settop boxes tune to capabilities. The capabilities may include the
the correct format, and minimize bandwidth number of tuners, the video and audio codecs
usage when there are demands for competing supported, and the communications methods, to
formats of the same content. name a few parameters.

MANAGING MULTIPLE ENCODINGS As the content providers begin delivering

MPEG-4 content, the cable system will need to
Architecture transcode these signals into an MPEG-2 version
so that legacy settop boxes will be able to decode
The SDV system has the ability to deliver a the stream. The system may also transcode
stream to the settop box based on the settop box MPEG-2 content into MPEG-4 content so that
capabilities. For example, if a settop supports additional lower bandwidth content will be
decoding only MPEG-2 signals, the SDV system available for the MPEG-4 capable settops. This
will ensure that the settop is only directed to tune will reduce the overall QAM bandwidth usage on
to streams that are encoded using MPEG-2. the system if only MPEG-4 capable settops are
tuned to a specific channel.
The SDV system may be pre-configured with
the settop capabilities. For example, the Figure 5 shows an example architecture of a
capabilities may be statically tracked by settop cable plant designed to deliver a stream with
model. The operator may track the settop model multiple encoding types. Transcoders are used to
and settop ID for each fielded settop box. The create MPEG-2 and MPEG-4 encodings for the
channel change request from the settop includes various channels.
the settop ID, which allows the SDV system to

OOB Mod /Demod

GbE Switch

SDV Server

Edge Manager

Receiver Service
Group
MPEG-2 AVC Combiner
HFC Network
Transcoder
MPEG-2 AVC
GbE Switch

GbE Switch

Rate Shape
GbE Edge QAM Settop
Bulk Encryptor
Distribution
Settop

Analog Content Settop

Figure 5: SDV System with MPEG-4 and MPEG-2 Transcoders

In this example, the transcoders are placed in network since it will carry multiple copies of
the headend with the other grooming equipment. some stream to the hubs.
This increases the load on the distribution

2008 NCTA Technical Papers - page 121

 The transcode function is shown performed encoding of that channel in order to conserve
by a stand alone server. However, it may also be bandwidth.
integrated in the edge device, VOD server,
Network Encryptor, or any system that ingests When a settop box capable of only decoding
and outputs GigE content. The key is that both MPEG-2 content requests an SDV channel that is
the MPEG-4 and MPEG-2 versions of the currently not being delivered to the service
content are available on the plant so that the SDV group, the SDV system directs the MPEG-2
system can deliver the appropriate version to stream of that SDV channel to that service group
settop boxes based on their capabilities. and it returns the channel tuning information to
the settop box. If an MPEG-4 capable settop box
Stream Delivery from the same service group subsequently
requests the same SDV channel, the SDV system
The SDV System directs channels to a service simply directs the settop box to tune the MPEG-2
group based on requests from the settop box. stream that is already being delivered to the
The SDV system uses the settop capabilities and service group. This assumes that the MPEG-4
policy information when determining the capable settop can also decode MPEG-2 content.
encoding format of the stream to deliver to that
settop box. When a settop box capable of decoding
MPEG-4 content requests an SDV channel that is
Although it is possible to transcode all content not currently being delivered to the service
encoded in MPEG-2 into an MPEG-4 encoding group, the SDV system directs the MPEG-4
to save bandwidth, it is likely that the SDV stream of that SDV channel to that service group
system will have some content that is only and it returns the channel tuning information to
encoded using MPEG-2. In this case, the SDV the settop box. If a settop box capable of only
system always delivers the MPEG-2 version of decoding MPEG-2 content that is from the same
the stream when requested by an MPEG-2 or service group subsequently requests the same
MPEG-4 capable settop box. This assumes that SDV channel, the SDV system must direct the
the MPEG-4 settop box can decode MPEG-2 MPEG-2 version of the SDV channel to that
encoded content. service group and return the tuning information
for the MPEG-2 stream to that settop box. Thus,
The SDV System may have content that is the channel is being delivered twice to the same
only encoded using MPEG-4. In this case, the service group using different encoding formats.
SDV system can only deliver such content to
MPEG-4 capable settop boxes. MPEG-2 capable Carrying both MPEG-2 and MPEG-4 versions
settop boxes will be denied service. However, of the same SDV channel is inefficient, since it
the channel map for the MPEG-2 capable settop requires at least 50% more bandwidth than
box should prevent subscribers from accessing carrying the MPEG-2 version of the channel
the content if the settop box cannot view it. As alone.
an example, the cable operator may have created
a special HD tier that is available as a premium Managing Multiple Copies
package. A subscriber receives an MPEG-4
capable settop box when subscribing to that The SDV system can employ several
package so that the subscriber can view that HD strategies for handling delivery of multiple
content. encodings of the same SDV channel to the same
service group. The basic choices are to leave
For content that is encoded using MPEG-2 multiple copies of the stream or to force tune all
and MPEG-4, the SDV system will likely be settop boxes onto a single copy of the stream.
configured to attempt to deliver only a single The chosen strategy depends on the bandwidth

2008 NCTA Technical Papers - page 122

available to the system and the impact of MPEG-2 settop is channel surfing. Thus,
reducing the number of copies of the stream. subsequent tune requests from MPEG-4 capable
settops may be placed into the MPEG-4 encoded
The main reason that the SDV system might stream in anticipation of the MPEG-2 settop
not always choose to deliver a single encoding of channeling off shortly. If the MPEG-2 settop
the stream is the force tune operation can be stays on the channel or other MPEG-2 settops
disruptive to the viewing experience. Directing a tune to that channel, the SDV system may revert
settop box from one stream to another requires to directing all new requests to the MPEG-2
the settop box to tune to a different frequency channel as discussed above.
and/or MPEG program number.
If resource availability in a service group is
Tuning to a different MPEG program number low and both MPEG-4 and MPEG-2 versions of
on the same frequency may take several hundred the same channels are being delivered to that
milliseconds while the tuner waits for the video service group, the SDV system may choose to
and audio data to arrive. This can cause jitter or force tune viewers to the MPEG-2 stream and
blocking on the display. Tuning to a different recover the resources assigned to the MPEG-4
frequency requires possibly 1-2 seconds while stream. This force tune operation may be
the tuner waits for information describing the disruptive, but it is the fastest way to reduce
stream contents and then waits for the video and resource usage in a service group.
audio data to arrive. This causes either a frozen
image or a black screen for the duration of the The SDV system executes the force tune
tune. The tune operation may also impact the operation by sending a message to each settop
DVR causing recordings to fail. box currently tuned to the MPEG-4 version of
the stream to tune a different frequency and/or
If the SDV system has sufficient unused program number. If the settop box is still tuned
QAM bandwidth for that service group, the SDV to the stream in question, it will tune to the
system may simply allow multiple encodings of MPEG-2 version of the stream using a frequency
the channel to remain on the cable plant. The and MPEG program number embedded in the
SDV system will still have the ability to service request message. The settop box returns an
new channel requests, therefore it would be best acknowledgement indicating whether the tune
not to disrupt the viewing experience by merging operation was performed. If the settop was
the streams. The SDV system can defer action already tuned to a different channel, the settop
until available bandwidth for that service group box returns an acknowledgment indicating that
drops below a “consolidation threshold”. the tuner is no longer on the channel in question.
The SDV system will retry the force tune
In the case where the SDV system elects to operation for settop boxes that fail to
leave both streams on the plant, the SDV system acknowledge its request.
may direct all new tune requests for that channel
onto the MPEG-2 version of the stream in hopes The SDV system has the option of either
of recovering resources when the settop boxes immediately recovering the resources assigned to
originally tuned to the MPEG-4 version of the the MPEG-4 stream or waiting until all settop
channel tune off. This mechanism allows the boxes have been tuned to the MPEG-2 stream.
SDV system to gradually recover resources Recovering the resources immediately is even
without force tuning settop boxes to the MPEG-2 more disruptive than simply the force tune
version of the channel. operation since the SDV server directs the edge
QAM to stop streaming the MPEG-4 encoded
In another case, the SDV System may use content. This likely occurs before the settop
recent channel change activity to detect that the boxes begin their tuning operation to the

2008 NCTA Technical Papers - page 123

MPEG-2 channel. This results in additional time DYNAMIC TRANSCODING
with a frozen picture on the screen or a blank
screen. Overview

The more graceful method of handling the The previous section described a static
resource recovery would be to wait until all transcoding method in which the headend
active tuners have been directed off the MPEG-4 network carries both MPEG-2 and MPEG-4
stream. Although this delays resource recovery copies of the same SDV channel to support both
by several seconds, it minimizes the impact on legacy settop boxes and MPEG-4 capable settop
the subscribers and provides an opportunity for boxes.
the settop box to use some advanced features in
an attempt to minimize the impact of the force Another method would be dynamic
tune. transcoding. In dynamic transcoding, the SDV
System directs a single MPEG stream onto the
The SDV system can further attempt to service group for each requested SDV channel.
minimize the impact of force tuning by following The MPEG stream is dynamically transitioned
these guidelines. The SDV system makes every between MPEG-2 and MPEG-4 based on the
effort to place both the MPEG-4 and MPEG-2 types of settop tuned to that SDV channel.
versions of the same channel on the same QAM.
This greatly reduces force tune times. The SDV Dynamic transcoding should be possible to
system may schedule force tunes to coincide with implement. The transport stream structure for
transitions between content and advertising. MPEG-4 encoded content can be similar to that
These force tunes would not only require of MPEG-2. However, the video and audio
knowledge of the expected time for the packets would carry MPEG-4 encoded data
advertising, the SDV system would need to instead of MPEG-2 encoded data. Additionally,
receive a trigger from the ad server indicating the the stream type identifier will signal that the
exact time of the advertising since the goal content carried in the stream is MPEG-4 [4].
would be to place the force tune on the transition
into or out of the ad pod. This would minimize MPEG decoders currently exist that can
the impact on both the subscriber and advertiser decode both MPEG-2 and MPEG-4 content.
since the transition typically includes a black These are being deployed into the new MPEG-4
frame or a fade to black sequence. capable settop boxes. However, it is likely that
these decoders cannot dynamically transition
Recent developments in settop box-based ad between the two encoding types. New decoders
splicing are providing the capability that the may need to be deployed.
settop box can perform a seamless splice
between programs carried on the same Decoders capable of handling the dynamic
frequency. If the SDV system is able to keep transition between encoding types will be alerted
both the MPEG-2 and MPEG-4 stream on the to the transition by a change in the PMT and a
same QAM and it does not recover resources change in the stream type identifier. The
until all settop boxes have acknowledged the transition will occur on an I-frame boundary for
force tune operation, then the settop box may be MPEG-2 and IDR frame for MPEG-4.
able to jump from the MPEG-4 stream to the
MPEG-2 stream with little or no noticeable Architecture
impact on the subscriber viewing experience.
The transcoding device itself can be a stand
alone server, part of an edge QAM, embedded in
a video server, or embedded in a network

2008 NCTA Technical Papers - page 124

encrypting device. Note that the transcoding a single copy of each channel. Each channel
device cannot work on encrypted streams. may be encoded in MPEG-2 or MPEG-4 based
Therefore, encryption must be applied after the on how it was received from the satellite. Then,
transcode. This can be done by a bulk encryptor on a service group by service group basis (i.e.
in the headend or encryption logic in the edge QAM by QAM basis), the edge QAM can be
QAM in the hub. directed to output either the MPEG-2 version of
that channel or the MPEG-4 version of that
Figure 6 shows an example system where channel.
dynamic transcoding occurs in the edge QAM.
The headend and GbE distribution network carry
OOB Mod /Demod

GbE Switch
SDV Server

Edge Manager

Service
Group
Combiner
HFC Network
Receiver
MPEG-2 AVC
Edge QAM
GbE Switch

GbE Switch
Rate Shape
GbE Transcode Settop
Bulk Encryptor
Distribution
Encrypt Settop

Settop
Analog Content

Figure 6: Dynamic Transcoding in the Edge QAM

A drawback of transcoding in the edge QAM

is that the stream cannot be encrypted until after Dynamic transcoding may also occur in a
the transcode. This means that the stream must video server. The video server ingests the
be both transcoded and encrypted at the edge. MPEG-4 or MPEG-2 source content. It can then
Performing both transcoding and encryption delivers an MPEG-2 or MPEG-4 version of the
operations at the edge may be costly as it stream for each service group with a settop box
requires dedicated processing power for each tuned to that channel. Figure 7 shows dynamic
service group. transcoding using a video server.

OOB Mod /Demod

GbE Switch

SDV Server

Edge Manager

Service
Receiver Group
Combiner
HFC Network
MPEG-2 AVC
Rate Shape
GbE Switch

GbE Switch

VOD Server
GbE Edge QAM Settop
Transcode Distribution
Settop
Bulk Encryptor
Analog Content Settop

Figure 7: Dynamic Transcoding in the Video Server

2008 NCTA Technical Papers - page 125

 By delivering a unicast stream for each service In another case, a particular SDV channel is
group, the Video Server is free to transition the being delivered in MPEG-4 format to a set of
stream between the MPEG-2 and MPEG-4 MPEG-4 capable settop boxes. When a legacy
versions as requested by the SDV system. Note settop box requests that SDV channel, the SDV
that this method requires significantly more Server must immediately direct the transcoding
bandwidth on the GbE distribution network than device to transcode the channel to MPEG-2
edge QAM based transcoding. format. The transcoding device will begin
transcoding at the next I-frame / IDR-frame
Stream Management boundary. The transcoding device then sends a
response to the SDV Server when the MPEG-2
Stream management when using dynamic transition has completed. The SDV Server
transcoding is greatly simplified. The SDV returns the tuning information to the legacy
system simply manages the transitions between settop box.
the MPEG-2 encoding and the MPEG-4
encoding based on the capabilities of the settop The SDV Server must wait for confirmation
boxes that tune to the channel. In most cases, no that the transcoding device transitioned to the
stream merging concerns exist. That said, there MPEG-2 version of the stream before the SDV
are still issues that need to be handled. Server returns tuning information to the legacy
settop box. This may take several hundred
In one case, a particular SDV channel is being milliseconds or more. By waiting for
delivered in MPEG-2 format to a mix of legacy confirmation, the legacy settop box will be
and MPEG-4 capable settop boxes for a guaranteed it is attempting to tune an MPEG-2
particular service group. If all legacy settop stream and not an MPEG-4 stream.
boxes tune to other channels, the SDV Server has
the opportunity to transition the stream from an Note that the SDV Server may deny the
MPEG-2 encoding to an MPEG-4 encoding. The request of the legacy settop box to tune to the
SDV Server sends a message to the transcoding SDV channel if there is insufficient output
device directing it to transition from the MPEG-2 bandwidth on that QAM to switch from MPEG-4
version of the stream to the MPEG-4 version of to the MPEG-2. Alternatively, the SDV System
the stream for the service group in question. may need to establish the MPEG-2 version of the
stream on a different QAM that has sufficient
The SDV System may not immediately bandwidth. In this case, one returns to the case
transition an MPEG-2 stream to an MPEG-4 where the SDV System delivers multiple streams
stream the instant all legacy settop boxes tune for the same channel.
off. The SDV Server may use heuristics to
determine the appropriate transition time, since COMPLETING THE TRANSITION
viewer behavior may indicate that some legacy
settop boxes are only channel surfing and will Initially with a small number of MPEG-4
return to the original SDV channel shortly. For capable settop boxes, SDV will be an enabling
example, the SDV System may have a database technology for deploying MPEG-4 into the cable
of advertising times for that SDV channel. If plant, allowing delivery of additional HD content
legacy boxes tune off during commercial breaks, and potential bandwidth savings. As the number
the SDV System may wait until the commercial of MPEG-4 capable settop boxes grows beyond
break ends before determining whether to the deployed legacy settop box count, the cable
transition to the MPEG-4 encoding. plant will transition to broadcasting more
MPEG-4 content while delivering the MPEG-2
content on the switched tier. This will provide

2008 NCTA Technical Papers - page 126

further bandwidth savings. As the number of [4] Robuck, M. (2006, July 1) MPEG-2 Stays
legacy settop boxes dwindle due to replacements Center Stage While MPEG-4 Waits in the
and upgrades, the cable operator may find it Wings, Cable360.net.
advantageous to replace the remaining legacy
settop boxes and remove MPEG-2 content from CONTACT
the cable plant. This would complete the
transition to MPEG-4 encoded content. John Schlack
Distinguished Member of Technical Staff
REFERENCES [email protected]

[1] Goetzl, D. (2007, November 26) HDTV Set

Sales to Soar, Broadcasting & Cable; and B&C
Staff (2007, December 28) CEA: DTV 101 Tournament Drive
Penetration Tops 50% Mark in U.S., Horsham, PA 19044
Broadcasting & Cable. Phone: (215) 323-1000

[2] Patterson, T. (2008, February) Fighting for

HD bragging rights, CED Magazine, 34. MOTOROLA and the Stylized M Logo are
registered in the US Patent & Trademark Office.
[3] Ellis, E. (2007, August 1) The MPEG All other product or service names are the
Transition, CEDMagazine.com. property of their respective owners. © Motorola,
Inc. 2008. All rights reserved.

2008 NCTA Technical Papers - page 127

 OPTICAL SEGMENTATION TECHNOLOGY ALTERNATIVES AND ARCHITECTURES

Phil Miguelez – Director, BAN Advanced Technology

Fred Slowik – Director, ANS Systems Marketing
Motorola Access Networks Solutions - 051808

Abstract small offices well but larger businesses require

GigE data rates and dedicated fiber.
Fiber non-linearity presents serious
challenges to fielding multi-wavelength optical Operators are also challenged to minimize
systems capable of transporting full broadcast / CapEx spend and limit system down time. This
narrowcast channel loads. Equipment vendors is especially true in the current unforgiving
are rising to meet this challenge with new economic environment. Pulling new fiber is not
technologies that will allow MSO’s to cost an option except in green fields and point to
effectively segment nodes and harvest fiber for point business access situations where the
new services or future network segmentation. revenue opportunities justify the expense. For
all other applications a means to increase
In this paper we briefly review the key fiber capacity using existing fiber is required. Multi-
optic challenges that are driving innovation and wavelength broadcast + digital transport is an
examine the different technology choices that ideal solution to meet this challenge.
are being offered today. We also look at a few
implementation models of multi-wavelength on Multi-wavelength transport allows node
the HFC plant for a variety of different segmentation with minimal touching of the
applications. physical plant. More importantly, the increased
BW capacity of fibers carrying multiple
wavelengths allows surplus fiber to be harvested
INTRODUCTION for other uses. These repurposed fibers can be
used for business access or further network
Market Drivers: segmentation needs.

Cable operators are faced with a wide range WDM solutions for digital transport are
of opportunities for expansion into adjacent commonplace. CWDM and DWDM network
markets such as commercial access and cell architectures for baseband digital and QAM data
tower backhaul. On top of this, competition and delivery have been in place for 10 years or
customer expectations are creating the need for more. The major barrier for realizing these
ever increasing bandwidth capacity in the networks as part of the HFC downstream
traditional CATV network. In order to meet the broadcast system has been the transport of
needs of both markets additional fiber or analog video carriers. Early attempts to
increased capacity of existing fiber is required. transport analog broadcast services over a
In most cases operators prefer to keep business CWDM network yielded poor results. Analog
services on a separate fiber network from video is extremely susceptible to noise and
residential video and data. Cable modems serve distortion. Fiber induced distortions add directly
to the native distortion of the source laser

2008 NCTA Technical Papers - page 188

transmitter. Additionally, fiber and passive solutions. Both approaches permit some of the
device nonlinearities create a host of potential fiber nonlinearity issues to be minimized while
impairments that must be avoided, minimized, making other optical impairments more difficult
or overcome. to correct.

Obstacles to Multi-Wavelength Transport: All of the various solutions have a few

common requirements. First among these is the
When multiple wavelength signals propagate necessity of having identical analog broadcast
through optical fiber an array of impairments channel lineups on each wavelength. Analog
come into play, the most significant of these are carriers are the most susceptible to crosstalk
Raman crosstalk, four wave mixing, dispersion, distortion. If the signal modulation on each
and cross phase modulation (XPM). The channel is identical, crosstalk susceptibility
magnitude of each of these impairments is a significantly reduced. Broadcast QAM will also
function of the laser chirp, the optical launch benefit from this same effect. Narrowcast QAM
power, the length of the fiber link, and the by definition is unique to each wavelength.
dispersion properties of the deployed fiber. Narrowcast modulation channels will
experience increased noise impairments due to
Additional impairments are also possible due crosstalk but the nature of digital modulation is
to interactions with passive elements in the more robust to these impairments.
system such as optical mux and demux filter
components. Optical power levels for each wavelength
should be roughly equal. Mixing high and low
Detailed descriptions of each of these optical power lasers creates the potential for Raman
nonlinearities have been presented in numerous scattering issues.
articles, technical whitepapers and previous
conference presentations on emerging multi- Another requirement in common is that the
wavelength technology. This paper will provide native laser distortion at each wavelength be as
a brief explanation for each of the critical low as possible. Many of the fiber induced
distortion generators where appropriate to distortions will magnify the raw distortion of the
emphasize their impact to analog or digital laser transmitter. Each distortion parameter in a
QAM performance. multi-wavelength system is a composite of the
laser distortion plus the distortion generated
Broadcast + Narrowcast Multi-Wavelength within the fiber and passive elements as shown
Solutions: in the following example equation to calculate
end of line (EOL) CSO performance.
Different techniques to mitigate the
numerous fiber induced distortions listed above EOL (cso) = 20 log (10 (Laser CSO /20) + 10 (Fiber CSO
/ 20)
have driven each vendor to create unique, + 10 (Mux CSO / 20) + 10 (Demux CSO / 20))
proprietary solutions. In order to take advantage
of the wide availability of proven analog The following sections will discuss
capable lasers and keep the complexity low, differences between DWDM and CWDM
most vendors have elected to operate in the solutions for multiple wavelength transport of
1310 nm region. Some vendors have chosen to analog broadcast + narrowcast channel loading.
pursue ITU standard coarse wavelength spaced
(CWDM) solutions. Other vendors have
promoted dense wavelength spaced (DWDM)

2008 NCTA Technical Papers - page 189

1310 DWDM Solution for Multi-Wavelength
Transport: The selected wavelengths must be located
away from the zero dispersion point of the fiber.
Stimulated Raman Scattering (SRS) effects The ZDP of SMF28 and SMF 28e fiber
have always been the most difficult impairment typically falls near 1310 nm but can vary from
to conquer. However, Raman gain is predictable fiber lot to fiber lot over a range of +/- 10 nm.
based on the optical power, link length, and The newest version of fiber that Corning plans
wavelength spacing. The plot below shows the to introduce this year (SMF28e+) will shift the
Raman gain coefficient versus wavelength typical ZDP to 1317 nm. Balancing the choice
spacing in THz. Operating with close spaced of wavelength selection to avoid FWM and the
wavelengths (left side of the plot) is an effective ZDP of the deployed fiber is one of the reasons
way to minimize SRS. for the different proprietary schemes of the
vendors supporting the DWDM approach.

Perhaps the most challenging issue facing

DWDM multi-wavelength solutions is related to
the optical passives. Mux and demux devices
are constructed using thin film optical filters.
The broadband response of these filters is
usually quite flat but as the filter bandwidth
becomes narrow as required for DWDM
wavelength spacing the pass band ripple
response can increase significantly. This higher
ripple creates sloped or tilted regions in the
bandpass response which interacts with laser
ITU standards do not exist for DWDM in the chirp to generate additional CSO beat products.
1310 nm O-Band spectrum but translating the At tilts larger than a few tenths of a dB / nm the
200 GHz or 100 GHz channel spacing CSO generated in the filter will begin to
commonly used at 1550 nm wavelengths to dominate the end of line distortion performance
1310 nm is easily done. depending on the chirp level of the laser used.
To avoid the problem of passband ripple, mux
While DWDM spacing helps to solve Raman and demux filters must be selected to very tight
crosstalk it enables another impairment, Four specifications.
Wave Mixing (FWM). This impairment acts in
an analogous manner as composite triple beat 1310 CWDM Solution for Multi-Wavelength
distortion. With equally spaced wavelengths the Transport:
interaction of three wavelengths will create a
beat that falls on the fourth wavelength. Custom Maintaining ITU standard CWDM spacing
wavelength selection avoiding equally spaced simplifies a number of the challenges that face
wavelengths is part of the solution to FWM. vendors of DWDM O-Band systems. Four
Distortion from Four Wave Mixing is most Wave Mixing issues are eliminated since the
pronounced as the wavelengths used are phasing of the optical wavelengths are de-
operated near the zero dispersion point (ZDP) of correlated by fiber dispersion. Optical passives
the fiber. Additional crosstalk and CSO with as with 20 nm channel spacing provide flat
few as two DWDM wavelengths has been passband response with measured ripple slope
reported when the optical channels were of < 0.1 dB / nm. The filter bandwidth is much
operated in the zero dispersion region.

2008 NCTA Technical Papers - page 190

greater than the worst case wavelength variation low chirp laser transmitters allows extended link
of the cooled laser transmitter, so stability over reach of up to 30 km.
environmental conditions is generally assured.

The major challenge to CWDM broadcast

transport is Stimulated Raman Scattering.
CWDM wavelengths are based on 20 nm
spacing defined by ITU standards. At this
spacing, Raman gain is

a significant factor and peaks in systems with 3

to 4 sequential channels. CSO distortions As shown in the block diagram above, the
generated in the RF and Optical domain by the broadcast input channels are conditioned and
laser are magnified by Raman gain interactions split to feed the individual CWDM lasers.
within the fiber and can dominate the overall Narrowcast channels are fed directly to each
system performance. High fiber dispersion such laser. No custom equipment is required at the
as occurs at 1550 nm would tend to de-correlate node. This solution can be configured with
the modulated signals (walk off effect) and help separate forward and return path fibers or
reduce the magnitude of Raman crosstalk. Near combined with a 1310 / 1550 WDM to provide
1310 nm dispersion is low so walk off is a single fiber solution for upstream and down
minimal. Optical launch power strongly stream loading.
contributes to the magnitude of the Raman
induced CSO distortion. Therefore, limiting For short reach applications (<15km) RF
laser output levels will minimize the effects of conditioning is not a requirement. We have
Raman at the expense of link reach. found in these cases that it is possible to reuse
currently deployed 1310 transmitters as long as
Adding wavelengths increases the optical the output power is equal or padded to match
power into the fiber and therefore increases the the added CWDM lasers.
magnitude of Raman crosstalk proportionate to
the additive optical level. Without using some Multi-wavelength solutions are extremely
external means of correction, multiple CWDM cost effective compared to the capital expense
wavelengths with Broadcast + Narrowcast of pulling new fiber. However, this technology
loading muxed onto a single fiber will produce does have limits. Fiber and passive component
unacceptable CSO distortion in optical links that insertion loss reduces link reach compared to a
exceed 12 to 15 Km. single transmitter. Since many of the fiber
distortions are optical level sensitive, cranking
Enhanced Coarse Wavelength Division up the power is not effective. Distortion
Multiplexing (E-CWDM) performance is a few dB lower than comparable
single transmitter distortion particularly CSO
Enhanced CWDM is a patented technology which is the most vulnerable to degradation
developed by Motorola to mitigate Raman from Raman and dispersion. Even with these
impairments in multi-wavelength systems. A restrictions, multi-wavelength solutions can
unique method of conditioning the RF broadcast provide sufficient performance to meet the
carriers minimizes Raman distortion along the requirements of typical N+6 cascade
fiber path. RF conditioning in conjunction with architectures.

2008 NCTA Technical Papers - page 191

 The next portion of this paper reviews the
applications of multi-wavelength technology for Early in the evolution of HFC network
Fiber Deep network migration strategies. deployment, cable operators had the choice of
designing their coaxial plant emanating from
Multi-Wavelength Applications optical nodes in either a balanced or unbalanced
fashion. Balanced means that all homes
The next wave of network migration for serviced from that node were equally divided
cable operators seems to be focusing upon among all feeder legs from the node. Due to
creating smaller node serving areas in order to topology, this balancing often required adding
provide increased bandwidth capacity to and express coaxial cables for segmentation
from fewer numbers of subscribers. Whether purposes. Although this approach provides a
accomplished by creating “smaller virtual smoother migration path for future node
nodes” via adding physical node segmentation segmentation, cable operators were hesitant to
capabilities at existing node locations, or by invest in the added material and construction
deploying additional satellite nodes deeper into cost to balance their node serving areas.
the network, one fact remains - there may not be Consequently, many operators chose to opt for
sufficient fiber available to support this the less expensive unbalanced approach where
migration strategy. the number of homes passed per feeder leg was
random.
Previous sections of this paper address some
of the various multi-wavelength technologies Ideally, if the original network design had
that are becoming available to operators to help followed the balanced approach, then virtual
alleviate fiber constrains. Because node sizes, node segmentation could occur rather smoothly
deployment depth and fiber counts can vary at existing node locations using segmentation
from operator to operator and system to system, capable nodes. Experience to date seems to
this paper refers to node migration in terms of a indicate that only about 20% of existing nodes
size reduction factor as opposed to absolute are sufficiently balanced to permit this ideal
house count per node. In this way, the reader form of segmentation. The remaining 80% of
can obtain an appreciation of available multi- existing nodes may be so unbalanced that some
wavelength options to meet their end goals. For combination of segmentation capable nodes plus
example, an operator with existing node sizes of the addition of new satellite nodes or adding
1200 HP might desire a 4X reduction factor express coaxial cabling may be required. The
whereas existing node sizes of 500 HP may only latter approach does drive fiber deeper in certain
desire a 2X reduction factor to meet the end areas,
goals.
Tables 1 & 2 illustrate a logical example of
Node Segmentation vs. Fiber Deep node segmentation for both balanced and
unbalanced scenarios using a hypothetical 512
Perhaps we should clarify that node home passed node and migrating fiber all the
segmentation and fiber deep architectures, both way to the home. Note the unbalanced node
candidates for multi-wavelength solutions, have creates the need for new fiber deployment
distinct differences and are not always during the initial migration process.
synonymous.

2008 NCTA Technical Papers - page 192

 Table 1
Motorola HFC Network- Balanced Node Migration Path

HFC 1X1 5t2 HP Original Node Location

HFC4X4 128 HP 128 HP 128 HP 128 HP

64HP 64 HP 64 HR 6lii HP

Fiber Deep

FTLA

or

FTTH
- - - • Fiber

Table 2
Motorola HFC Network- Unbalanced Node Migration Path

HFC 1X1 Sj 2 HP Original Node Location

HFC 4 (1X11 128 HP

128 HP 128 HP 128 HP

FTLA

or

FTTH

'---------+- 8---···G

2008 NCTA Technical Papers - page 193

Cascade Reduction merely chooses to drop-in new nodes at all
existing amplifier locations.
Although node segmentation and fiber deep
architectures reduce the serving area size with It is important to understand the cascade
respect to the number of homes/users per virtual impact of fiber migration since different multi-
node, the amplifier cascade length often remains wavelength technologies offer different
unchanged. This, due to the fact that certain performance characteristics at the node.
portions of the segmented node fed from the Combined optical and RF performance becomes
original node retain their existing footprint an important consideration in determining
while cascade reductions usually take place in which technology will support end-of-line
those areas where satellite nodes are added. network performance goals. Depending upon
RF amplifier cascades, one optical technology
Figure 1 Existing Node
might mesh better with reduced amplifier
cascades as opposed to another that might be
better positioned to support longer amplifier
cascades.

What About Adding Fiber?

If existing fiber counts were unlimited,

network migration would be a much simpler
task. Unfortunately, fiber counts are
constrained in many systems, so operators need
to understand new fiber deployment costs versus
Figure 2 Node Segments
alternative options such as multi-wavelength
7
technology. A very simple example illustrates.
2

1
Aerial
Aerial Make Total
3
Fiber Material Labor per Ready Per Aerial Per
Count Per Foot Foot Foot Foot
6
6 $ 0.27 $ 0.60 $ 1.00 $ 1.87

Using this as an average aerial constructed

4
5 price per foot, we can easily understand just
how expensive installing new fiber can be
Intentional attempts to drive fiber deeper ($1.87 X 5,280 = $9,873/mile). This cost is far
with the goal of strict cascade limitations often more than the cost of the opto-electronic
lead to very expensive migration solutions. elements required at the headend / hub, and
Maybe a better way of looking at fiber deep node location. Multiply this cost by the total
would be to size the node to a desirable house distance required to reach an existing fiber
count and ignore the cascade length. Pulling starved node location and the cost can become
fiber to a Node + 0 architecture for example, prohibitive. This fiber installation cost does not
without a lot of re-plumbing becomes consider more complex installations such as
tremendously expensive especially if one underground or areas where significant make-
ready costs could arise.

2008 NCTA Technical Papers - page 194

 excluding installation and new fiber if needed)
The following graph illustrates fiber and extend upwards based upon specific
installation cost on a per km. basis versus the application needs.
opto-electronic cost per virtual link for various
multi-wavelength solutions. WDM

Approximate Fiber Deployment vs. Opto-Electronics Cost The least complex of the multi-wavelength
$300,000 solutions, this approach simply uses a 1310nm
full band (54-1002MHz) downstream
$250,000
wavelength and a 1550nm upstream wavelength
$200,000
placed upon a single fiber. These wavelength
Total Cost

$150,000 directions can be reversed in some applications.

$100,000

$50,000

$0
1 10 20 30 40
Distance (km.)
Fiber Deployment Cost Opto-Electronics Cost

Node segmentation is accomplished by

simply lighting up one fiber per wavelength
Depending upon the distance and fiber pair. Assuming up to 6X migration is desired
counts to existing nodes, and whether feeder and sufficient fibers exist, this method is
legs are balanced as previously discussed, being generally a low cost least complex means to
able to expand bandwidth capacity via adding achieve node area segmentation, and can
wavelengths on existing fiber to existing nodes achieve distances greater than 40 km. with
is advantageous. Less significant are new fiber excellent performance in the area of 51/-70/-66
extensions to satellite nodes that may be dB CCN/CTB/CSO.
required beyond existing node locations. Since
these links are usually less than 2 km., the cost E-CWDM
becomes much more tolerable, again, depending
upon the extent of deployment. Considered advantageous for fiber
constrained applications, this approach,
What is the Right Solution? although a bit more complex, enables full band
(54-1002MHz) downstream 13xx nm
It now becomes clear that alternative wavelengths upon a single fiber. Depending
technology is needed to be able to cost upon distance requirements, upstream
effectively drive fiber deeper into the network. wavelengths may also be deployed upon the
Depending upon the particular situation, several same fiber or a second fiber may be required as
multi-wavelength solutions exist or are illustrated in the example below.
emerging that may co-exist in the same network.
Three basic types of solutions are presented
below. All of these options offer significant
benefits. These are WDM, E-CWDM, and
Broadcast/Narrowcast Overlay. Cost of these
bi-directional solutions begin in the $5000/link
range (including forward, reverse and nodes,

2008 NCTA Technical Papers - page 195

 E-CWDM BC/NC Overlay
Fwd. BC
Fwd. BC

To Node
1550 nm BC
E-CWDM

E-CWDM
15xx nm DWDM Rev.
13xx nm
1 1

Rev.

40 40

1 1
15xx nm DWDM Fwd. Fwd. NC

CWDM
CWDM

15xx nm Fwd. NC

40 40

Distance, channel loading and performance

requirements dictate whether a single or dual
Based upon the number of wavelengths downstream optical receiver is required. This
deployed upon a single fiber, this approach can solution is also well suited for applications
cover distances of up to 30km, providing requiring optical path redundancy. Reach of up
performance in the area of 50/-68/-60 dB to 80+ km. are possible producing performance
CCN/CTB/CSO. This solution may also permit in the area of (49/51)/-66/-66 dB
the ability to re-purpose existing fibers for other CCN/CTB/CSO.
business applications.
Which Solution is Best?
BC/NC Overlay
Applications vary and so too does the answer
A bit more complex than the two previous to this question. A network analysis is generally
solutions, this solution offers advantages in required to determine the best fit and in certain
networks requiring longer reach as optical instances, more that one solution may be
amplification is possible. Generally, a two fiber required. Some generic guidelines however,
solution, it consists of a single broadcast may be helpful in determining where to begin.
wavelength typically operating in the 54-
550MHz pass band placed on one fiber which A starting point would be to identify existing
also can accommodate upstream node sizes in the plant and determine the
CWDM or DWDM wavelengths. ultimate node size desired. This is determined
by network operator bandwidth requirements
based upon service offerings. Once this goal is
established, dividing the existing node size by
A second narrowcast fiber is used to the new desired node size establishes a node
transport up to 40 wavelengths typically used reduction factor.
for QAM signals in the operating pass band of
550-1002MHz. This factor, when considered with the fiber
counts to the existing node, the distances
required to be covered, and link performance
goals, enables a high level selection of
which multi-wavelength technologies are most
applicable. In some circumstances it may be
wiser to just utilize spare fibers if available or
convert an existing two fiber solution to a 2X
WDM solution (1 DS and 1 US wavelength per
fiber).

2008 NCTA Technical Papers - page 196

 particular application is required to establish a
Ongoing network analysis seems to indicate rational migration strategy.
that the E-CWDM solution will become a
dominant short to mid-range tool in the HFC It is important to note that many operators
network bandwidth expansion tool kit. approach the need to migrate their optical
networks as an all or nothing proposition, basing
Once this analysis is accomplished, and a their strategy and CAPEX requirements on an
few options are selected, it becomes time to put entire network optical migration. In reality, the
pencil to paper and validate the chosen solution migration process can and should take place in a
on the network design. At this time, additional phased approach addressing those areas of
decisions may be made to provision for immediate or impending node congestion and
additional future levels of migration should a deferring migration of those less endangered
staged approach over time be desired. nodes to some point in the future if and when
needed.
Conclusion
Numerous tools are evolving to expand HFC
There are many factors to consider when networks in order to provide increased
deploying multi-wavelength solutions for node bandwidth. The tools are growing and are of
segmentation and fiber deep applications in great interest to the cable industry.
order to increase network bandwidth.
Deployment of these various technologies
This paper only presents a high level and architectures can only help in the battle
discussion of some of the technology and against the competitive forces that threaten the
options available. Much more detailed analysis current market.
of which solution(s) make the most sense for a

2008 NCTA Technical Papers - page 197

 SCALING MOUNT EVEREST: DELIVERING MULTI-SCREEN VIDEO IN AN ‘INFINITE
CONTENT’ WORLD
John Pickens, Chief Technical Strategist VCNBU
Cisco Systems, Inc.
Sree Kotay, Chief Software Architect
Comcast

Abstract TV (even popular linear video becomes

unicast), the growing libraries of high quality
The consumption paradigm for TV is rapidly commercial video (movies, original cable
changing from pure broadcast to time-shifted shows, made-for-TV, and straight-to-video) and
unicast. This behavioral model is the driver for user generated content is accelerating this
the new formula, “Cached Unicast equals paradigm shift, thereby stretching the limits of
Multicast”. Supporting this trend is the rapid existing multicast and pitcher/catcher video
evolution of the network paradigm from a delivery systems to be competitive.
classic siloed broadcast dominated spectrum to
a shared spectrum with converged usage of IP The formula of "cached unicast = multicast",
transport for all applications including video. as embodied by Content Distribution Networks
The long range vision is tens of thousands of (CDNs) like Akamai, becomes more and more
channels, hundreds of millions of assets, and desirable as usage patterns change and different
orders of magnitudes more content producers – device types proliferate. This proposed shift
all delivered to the device of the consumer’s enables the video delivery system to deliver an
choosing. This paper identifies key extreme scale of available assets, including
characteristics of the next generation solution multiple formats, rates, and resolutions for the
architecture, such as real time enabled cache same asset, with little economic or operational
distribution hierarchies, in order to deliver an impact. However, traditional Internet CDNs
infinite world of content and unlimited scale of lack the proper control semantics (e.g. you
subscribers and consumption modalities, while never need to "rewind" a web page), and scale
delivering many of the economic benefits of of solution (latency, throughput and cost
today’s architectures. scalability).

OVERVIEW This paper identifies the next generation

technologies and paradigms in real-time media
The increasingly rapid user adoption of
delivery that enable cable operators to migrate to
time shift TV, new HD content (requiring multi-
this new world. Points highlighted include a
carry) and interactive video services [like video
massively scaleable authoritative storage
on demand (VOD)], coupled with the exploding
network, transition to more distributed
popularity of blogging and audio/video
architecture designed for media, dynamic caching
podcasting, along with higher delivery data rates
in the interior and at the edge of segmented
(e.g. DOCSIS 3.0) and two-way connectivity,
content, and n-screen enabling application
requires a revolution in service delivery for
paradigm, where resource management and
media content. The initiative of Switched
authorization enforcement is built into the next-
Digital Video (SDV) for linear video channel
generation network (NGN) media delivery
delivery is an early recognition of the emerging
infrastructure.
paradigm of long tail consumption and niche
programming in the core TV market. Time shift

2008 NCTA Technical Papers - page 212

 VIDEO CONSUMPTION shifting, place shifting, and device shifting. This
phenomenon is portrayed within
Consumer video consumption behaviors are Figure 1.
undergoing a paradigm change, encapsulated by
the concepts of time

Streaming
Broadband
VoD
Cellular Place EoD
Wi-Fi, … Shifting
Time DVR
Shifting nDVR, …
IP-Everything
out-of-the-bottle!

Device
Shifting DVR
PDA
iPod
Audio/Video, …
Figure 1 – Consumption Paradigms

For TV, the time shifting phenomenon to another device and resume consumption. TV
gained mainstream acceptance with the (HD downstairs, SD upstairs), PC (in office or
widespread adoption of digital video recording in hotel), and mobile (while riding the shuttle to
(DVR). Though a large percentage of homes the airport) are three well known device types
do not yet have DVR, and those that have a for converged consumption of media.
DVR do not have it for all TVs (or a media
center), a high percentage of consumers are now Device shifting is an increasingly discussed
very familiar and accustomed to DVR time shift paradigm where, in addition to streaming
consumption, whether in their own home or in content from the network, it is possible to
homes of family and friends. download the content to different devices, and
view the content on those devices. MP3/MP4
Place shifting is a new emerging trend players, mobile phone with storage, and laptop
increasingly being promoted by service PCs are three examples of such devices.
providers, typified by “multi-room DVR” and
“whole home VoD”, where an individual may Use cases demonstrating the combination of
choose a content to view on one device, then all three follows. Within the home the content
pause the content (bookmark), and then migrate is consumed on an HDTV, paused, and then

2008 NCTA Technical Papers - page 213

resumed upstairs on an SD TV. Alternatively, for the picture to display after the start of
the subscriber may travel to a hotel and prefer to streaming, the inability to seamlessly transition
view the content on his/her PC. Another into trick mode behavior, forward or rewind,
example includes a subscriber in a limousine/car and experiencing random display pauses while
who prefers to view it on his/her mobile device. repairing under-runs of the elasticity buffer in
Or the content might be downloaded to his/her the PC.
laptop and he/she views it on the airplane. In all
these cases the content is resumed from For TV quality consumption, by contrast, the
wherever it was paused or bookmarked, user experience delivered by the network is
independent of the type of device or location of expected to be extremely high quality. An
consumption. example is the requirement for low latency
(subsecond) delay from stream event to stream
All three consumption paradigms can already action. Examples of stream events are stream
be seen in the internet web browsing model for start, trick modes (fast forward, rewind), and
video consumption. The community of Internet interactivity (e.g. pause  pause-ad).
users familiar with these models has grown to a
staggering numbers, with over 10 billion unique Subscriber Infinite Content World
video views consumed monthly, with YouTube
accounting for approximately 30 percent of that The content universe is growing. Whereas a
number. typical library size for Video on Demand (VoD)
used to be a few thousands of hours, it is now
Also driving the change in user experience is targeted to be much larger, on the order of
the explosive growth of HD content and hundreds of thousands of hours and eventually
exponential expansion in the number of content millions of hours. [1]. In early 2007 [2,3]
producers – a consumption feedback cycle Netflix announced an online library of 70,000
highlighted by the early trends of blogging and titles. Now the estimated library size is well
podcasting, now extending to video. over 90,000 titles, and a high percentage of
newly added titles are HD Blu-ray format
All these paradigm shifts require a revolution reflecting the popularity of high definition
in mechanisms for enabling service delivery of programming. Comcast in January 2008
media content, because traditional multicast announced plans for Project Infinity to grow the
pub/sub models lack the cost and operational On Demand library to 6000 titles (3000 in HD)
scalability to compete effectively. in 2009, with that number expected to scale
dramatically thereafter [4].
Subscriber Quality of Experience
One of the key design differences between
The TV consumption experience of users today’s video delivery systems and those of
served by content within the “broadcast” tomorrow is the split between content discovery
network is significantly different and higher (asset metadata, availability, and associated
quality that that of today’s users who are served information) and content distribution (the
by internet or mobile services. physical movement of the asset from source to
consumer.
In the internet model (e.g., YouTube) the
consumer has been conditioned to accept lower As the number of available assets grows, it
quality consumption experiences. Experiential becomes untenable (and undesirable)
examples include long latencies while waiting economically and operationally to scale edge

2008 NCTA Technical Papers - page 214

capacity against the number of assets. Instead, distribution. Third is a transformation to make
edge capacity must scale against the number of the content format, place, and device
unique assets consumed. This design criterion independent in order to deliver n-screen
demands a separation of data flow from media delivery.
flow.
Convergence Initiative
An interesting number foreshadowing future
content volume growth pertains to the amount In order to deliver the universe of infinite
of user-generated video content on the Internet. content, and assure the DVR-like experience
While the quality is not as good as from within the network, a number of
professionally produced content, its growth is convergence initiatives are underway.
explosively accelerating. Based upon
unpublished monitoring done by search Perhaps most enabling convergence activity
companies, in early 2007 the estimated number is the rapid evolution to a wideband all-IP
of titles was in the order of 40,000,000. By the infrastructure. This transition was
beginning of 2008 the number of titles had foreshadowed in the Video-QAM universe by
grown to around 120,000,000. the migration toward IP enabled QAMs (IP to
QAMs, traditional MPEG to home). Switched
In addition, operators are beginning to offer Digital Video (SDV) [5] for linear video
managed services that enable users to generate channel delivery was the next step recognizing
their own content and make it broadly available of the need to rapidly evolve infrastructure in
either downloaded online or as part of user- order to free up bandwidth for next generation
generated channels. services. The key insight of SDV, versus
traditional broadcast to the home, is that it is
Increasingly, professional content producers desirable to scale content against consumption,
are opening up their content archives to instead of against the total corpus of
consumers, both directly (called over-the-top) availability. It is now accelerating with the
and via managed relationships with service evolution toward DOCSIS 3.0 (wideband all the
providers (assured quality of experience). way to the home) and universal QAMs, which
allow service channel sharing across VOD, high
IP NGN VIDEO ARCHITECTURE speed data, and video services.

Three key initiatives for achieving a video As currently portrayed in Figure 2, a DOCSIS
enabled IP NGN architecture are defined. First enabled wideband infrastructure will enable 6
is a series of infrastructure convergence Gbps aggregate IP enabled spectrum
initiatives required in order to increase the downstream – competitive with other service
diversity of content delivery and user providers – on a 950 MHz plant.
consumption experiences. Second is a real time
enabled caching architecture for content

2008 NCTA Technical Papers - page 215

 UPSTREAM DOWNSTREAM
5 42 52 550 600 650 750 880 1000

Unified Data,Voice Unified Data, Voice, Video

~50Mhz ~950Mhz
Figure 2 – 6Mbps DOCSIS convergence

A second enabling initiative is the continuing Other convergence initiatives not addressed
effort to reduce fiber node size. Two driving in this paper include bandwidth management,
factors are service enablement and the metadata, standard advertising interfaces, digital
competitive need to offer higher broadband rights management, real time streaming
bitrates. protocol, Digital Living Network
Alliance/Universal Plug and Play, conditional
For service enablement, even at 250 homes access systems, etc. Ultimately these issues
passed with 100 percent subscriber penetration, need to be addressed as challenges abound.
a 6 Gbps infrastructure can serve 750 MPEG-4 Unlike video services of the past, new services
HD streams (8Mbps per stream) – 3HD streams must be delivered to all types of devices in
per home. The reality is that subscriber myriad locations -- with high quality and to
penetration is less than 100 percent, and all TVs massive scale. A new architectural approach is
are not HD, and thus sufficient bandwidth exists needed.
even at higher HHP ratios.
Real Time Caching Initiative
For addressing competition a motivation for
reduction of fiber node size is the need to Web caching is a well understood and widely
further increase peak bandwidth offered per deployed paradigm which features the transient
subscriber. For example, optical fiber storage of web objects such as HTML
technologies such as EPON are migrating from documents for subsequent retrieval. Caching
today’s 2 Gbps:1 Gbps:32 to tomorrow’s 10 enables reduced bandwidth consumption,
Gbps:10 Gbps:32 (ratio of down:up:homes). A reduced load on servers with the authoritative
6 Gbps DOCSIS® downstream is already higher storage of content, and reduced interactive
than the 2 Gbps downstream offered in EPON latency. Overall it increases the user quality of
architectures today and is well in the league of experience, and reduces network infrastructure
Ethernet Passive Optical Network (EPON) 10 cost. [6, 7]
Gbps downstream architectures. The only
significant difference between DOCSIS and Web caching can be deployed in a variety of
EPON will be the number of homes sharing the modes, from client, to proxy, to arrays of front
bandwidth, and the amount of spectrum offered ending servers. In this paper we focus on
for DOCSIS® enabled converged IP delivery. caches placed within the network.

2008 NCTA Technical Papers - page 216

 Delivery of real time video via caching has reduced to approximately 8 Mbps for MPEG-
similar benefits to delivery of web objects via 4/AVC. Furthermore, the size of objects (the
caching. Consumption characteristics exhibit a sum of all object segments) can be in the
Zipf curve phenomenon [8] where more popular n*Gigabyte size range. Second the service level
content (e.g., a show now playing, though time expectation of the consumer is higher than it is
shifted) is viewed by more people. The first for web content. The jitter requirement in real
person to consume a video causes it to be time content delivery is much smaller than it is
downloaded from the authoritative source into for web services and requires that the consumer
cache, and the next person who consumes the experience no visible artifacts or delays. Third,
video accesses it from cache. No subsequent there are multiple correlated object segments
network transport is consumed upstream of the being delivered in real time video consumption
cache, and the access latency is shorter (by a – the 1x media stream, multiple fast forward
few hundred milliseconds in worst case). [9] renditions of the media stream, and multiple
rewind renditions of the media stream.
Given the rapid migration from real time
consumption to time shifted consumption, and Therefore additional characteristics are
the existence of the real time caching function, required within the cache delivery infrastructure
the benefits of caching derive similar benefits to for video. These are outlined below.
multicast distribution at the edge, with the
difference that consumers no longer need to Tiered Hiearchies
consume content at the same real time timeline.
The characteristic that one copy of the content A caching hierarchy is defined for real time
(first user) is distributed across the backbone content distribution. Figure 3 highlights several
toward the edge is like multicast. The possible configurations. The number of tiers
characteristic that subsequent consumers of the deployed is arbitrary – it can be minimal
content generate no backbone traffic is also like depending on the consumption characteristics of
multicast. devices downstream (e.g. number of subscribers
signed up for video service). As consumption
Three significant differences between web demand grows, additional cache storage can be
caching and real time video caching are deployed either in parallel or in hierarchies in
identified in this paper. First the bandwidth and order to manage the tradeoffs between
size consumed by “objects” is substantially concurrent usage and latency and resource
higher. For MPEG-2 HD, the average consumption.
bandwidth is about 15 Mbps – though it is

2008 NCTA Technical Papers - page 217

 Inte
CONTENT rme
di ate
VAULT hop
ss
eg m
O (“Infinity” ) STREAM en t
cac
CACHE he
fi ll

STREAM Las
CACHE t ho
p re
al-t
Cache I/O Reduction im e

Cache I/O Reduction

I/O proportional to STB
Number of active subs

Figure 3 – Real time Caching Hierarchy

At each tier of the hierarchy the latency

introduced is minimal – O(n*10ms). Also the In the push category, (typified by MPEG
ability to transition between consumption modes transport streams sent over either UDP or RTP
(1x, n*FF, n*REW, pause) with low latency – transport) the content source is bandwidth
O(250ms) is enabled. A key requirement is aware, and maintains the rate of transfer in order
that, with each of the transition modes, the to meet the bandwidth delivery characteristics of
awareness of frame by frame semantics of the the content. The push model avoids significant
content type is necessary so that the video bidirectional overhead (other than adaptation to
segments within cache storage can be accurately network resource constraints) and enables
managed. assurance of stream rate for real time content
objects.
Pull Versus Push Distribution
Either style can be utilized. The pull category
Caching object distribution protocols can be requires new mechanisms that assure that the
divided into two categories. real time caching servers (potentially different
servers) being contacted for unpredictable mid-
The pull category typified by web caching stream distribution operations have ways of
depends on the client that is issuing the requests learning that the previous operation has been
and intelligently managing the transfer canceled by the downstream cache, and initiate
operations. No awareness of server state is the abort procedure. The push category benefits
communicated to the client. Usually no from this mechanism since communication of
awareness of the bitrate of the content is client state already exists.
communicated either (though it can be
communicated by a separate control path).

2008 NCTA Technical Papers - page 218

Segmented Object Distribution time caching that is not present in web caching.
This is the capability to transition to whatever
Traditional video distribution models exhibit object distribution mode is being consumed by
the characteristic that the entire video object the client, assuming that the new object
must be distributed to the entity that streams the segments from the new mode are not yet cached
video toward the client. This generates several on the caching entity. It should be noted that
systemic deficiencies. First is that a significant this is based on client behaviors driven by
delay is incurred while awaiting distribution. operations in the control protocol, e.g., RTSP.
Second is that the percentage of content
consumed is less than 100 percent (especially In order to deliver correlated object caching,
for long-tail where segments, e.g. famous some structure, such as an indexing database,
scenes, are of primary interest). By distributing needs to be conveyed between the authoritative
the entire object the cache is utilized in a non- source and downstream caching entities so that
optimized manner. all cache entities have accurate awareness of the
object content segments contained within the
Therefore one of the characteristics of TV cache storage.
cache distribution is that object segments are
transferred, on demand, if the correlated content Static-Object Verses Dynamic-Object
segments are not already cached locally. This
has the advantage of optimizing bandwidth
Two different types of objects are to be
consumption and cache storage consumption.
distributed by the caching infrastructure. The
first category, here called static-object, pertains
Another advantage of segmented object to a content item that has been completely
distribution is that it enables new services such ingested into the authoritative source prior to
as remixing, where arbitrary segments of distribution towards the streaming server client.
content can be remixed into a new virtual asset. This is typically called VoD, but is not
An example is all the goal shots of the world constrained to VoD objects. In this paradigm all
famous soccer star Pele combined into one ingest and other processing of the content object
segment. This can be achieved without is completed prior to initiation of cache
distributing the dozens of entire full-game video distribution. In one use case this object is not
objects to real time caching servers. Only the identified as available to clients until full ingest
relevant scenes need to be real time cache filled. is complete.
The object granularity needs to have the ability
to identify frame level semantics in all cases of
The second category, here called dynamic-
segmented object distribution. Methods for
object, is a type of object that is dynamically
learning and communicating such semantics
created and ingested by the authoritative source,
range from control plane extensions identifying
and is concurrently distributed into the caching
offsets to embedded descriptors highlighted by
infrastructure. In this paradigm the authoritative
standards such as TV Anytime [10].
source is concurrently performing processing on
the object (e.g., computing trick files, if
Correlated Object Caching required) and making the object available for
concurrent distribution toward the destination.
The functionality of transitioning from 1x One well known use case for the dynamic-
content to rewind and fast forward modes of object paradigm is time-shift of linear content.
consumption highlights another feature for real

2008 NCTA Technical Papers - page 219

 It should be noted that dynamic-object types constraints. The caching protocol must be
have an impact on functionality of the correlated designed so as not to induce either packet loss
object caching indexing database, i.e., dynamic or excessive buffering jitter in the aggregate
updates concurrent with ingest by the number of streams being concurrently delivered.
authoritative server.
Opportunistic Resource Utilization
Source & Sink State Synchronization
The cache fill protocol should be aware of
The web service caching model exhibits a resources of the source and sink, and also be
lack of state synchronization between the client capable of adapting transfer behavior in
and the server. Each side estimates the projected response to dynamically changing resource
behavior of the other side. Neither side is aware behaviors. If for example, the source, sink, and
of any average or instantaneous bottlenecks or intervening path are lightly loaded from the
constraints of the other side. The real time cache perspective of resources, then an optimization is
fill protocol should support a method of to enable content to be transferred at higher
communicating instantaneous load and state rates opportunistically. Such transfers are not
change of both source and client. directly correlated to the actual play out state of
the content with respect to the subscriber.
One example of state synchronization is the
awareness of bandwidth. Each source and sink Elasticity Assurance
has a finite aggregate I/O bandwidth limit.
Examples of such bandwidth constraints are on- The real time cache fill protocol should also
board bus bandwidth, bandwidth to associated optimize management of cache fill buffer
storage, and bandwidth between memory and elasticity, while maintaining a short maximum
adapters. The real time object caching service latency for stream event transitions. This
should exhibit bilateral awareness of I/O requires a distribution mode where content is
constraints of the source and sink so that initially transferred at a rate higher than stream
unnecessarily high latencies or jitter behaviors rate, and then, after a short time window, settles
are not introduced. down to transfer at stream rate. The transition to
higher rates occurs at any point that new content
Ingest Overrun Avoidance is transferred and short streaming startup latency
is required. Examples include session start,
Each caching node in the distribution path splice points (different content objects),
from the authoritative source to the client has interactive transitions to new content, and trick
finite bandwidth ingest constraints. Content mode transitions (also where different content is
distribution must not overrun the ingest transferred). The elasticity buffer accommodates
bandwidth with the aggregate maximum number reasonably bounded jitter behavior and
of active session. This implies the ability to retransmission of dropped packets without
maintain tight tolerances on smooth delivery of disrupting streaming behavior to the subscriber.
the stream, and avoidance of unnecessary
bandwidth bursts. N-Screen Initiative

Network Bandwidth Optimization A system architecture for enabling N-screen

delivery is required. Key enabling
The path from the content source to characteristics include decoupling of the
downstream caches has finite bandwidth awareness of delivery infrastructure from the

2008 NCTA Technical Papers - page 220

application layer, and embedding all distribution the unique assets being consumed, not with the
and resource management into the delivery number of streams being watched.
infrastructure.
The characteristics of the application layer
Because the model is (a) inherently a “cache- are expected to be like web services in nature.
on-demand” model, (b) separates delivery from Figure 4 shows a sample configuration.
metadata, and (c) enables “real time ingest” Subscriber interfaces will be provided for
from external storage, sparsely populated media navigation, business logic (purchases, rentals),
consumption formats (format transcoding) may service configuration, entitlements, etc. Each
be generated on-demand, or opportunistically. device type will have control and transport
As with the demands on the central storage interfaces that are specific to the device, but
systems themselves, this load scales only with which are not seen by the application layer.

Web Services SOA, …

Middleware
Catalog/Purchase/…
A
Asset C
Metadata
C C
Ingest Web style authorization and reporting
E L
Workflow Asset
Control S I
Nationwide S E
Real Time CDN
RTSP
N
Storage
Setup and Control T
Bandwidth Control
Distribution Streaming
FTP, …
Catcher Ingest UDPRTP
Asset
Transport Transport

MGR SNMP

SP
EMS *
Manager

Figure 4 – “n-screen” Convergence

Real Time Streaming Protocol (RTSP) is the enable the broadest class of device
dominant control plane signaling for session consumption.
management. Capabilities discovery and on-
path session resource management will be The Web services style of interfaces can be
utilized to identify the appropriate version defined to allow all streaming infrastructures to
(encoding resolution, bit rate, codec type) of an
consult the authoritative business logic of the
asset to stream to the device, and is appropriate application layer. This logic should be
for more advanced media control. Systems authorization oriented, not authentication
should also expect to provide simplified HTTP
oriented, as the criteria for playback may be
semantics (with potentially degraded user-centric (commercial entitlements or user
performance and feature characteristics) to

2008 NCTA Technical Papers - page 221

sharing permissions) or publisher-centric (rights
management or web availability). References

CONCLUSION [1] Real Time Video Services & Distributed

Architectures: Irreconcilable Differences or
This paper identifies the rapid shift in user a Marriage Made in Heaven, John R.
consumption behavior from traditional Pickens, SCTE 2006
consume-on-broadcast-timeline (on TV only) to [2] Netflix 2007 press release
consume-on-subscriber-timeline with the ability http://www.netflix.com/MediaCenter?id=53
to pause, rewind, and fast forward content (on 84
any device). Infrastructure convergence toward [3] Netflix Current selection -
all-IP, wideband edge network transport, and http://www.netflix.com/BrowseSelection
unicast enabled real time cache distribution [4] Comcast Project Infinity press release
paradigms are highlighted. 2008 CES -
http://www.comcast.com/ces/infinity_hd.asp
The benefit of having authoritative sources x?section=hd
for the content (permanent library storage), as [5] An Open Architecture for Switched
exhibited in the web object distribution model, Digital Services in HFC Networks, Luis
plus the insertion of real time enabled caching Rovira, Lorenzo Bombelli, SCTE 2006
servers in the path between the authoritative Conference on Emerging Technologies.
source and the destination client, enables the [6] Web Cache Wiki article -
service to be scaled to an unlimited number of http://en.wikipedia.org/wiki/Web_cache
consumers, consuming an unlimited library of [7] A Survey of Web Caching Schemes for
content (both on-demand and time-shifted live), the Internet, Jia Wang, ACM SIGCOMM
while preserving the user expectation of DVR- Computer Communication Review, Volume
like consumption delivered by the network. 29 , Issue 5 (October 1999)
[8] ZIPF Wiki article -
Also highlighted is the decoupling of the http://en.wikipedia.org/wiki/Zipf's_law
application layer from the real time content [9] VOD Servers - Equations and Solutions,
delivery layer. No specific protocols are Glen Hardin, W. Paul Sherer, NCTA 2005
detailed in this paper. The primary focus is [10] Metadata - the role of the TV-Anytime
establishing a framework for scaling to the specification, Morecraft, C. Storage and
world of infinite content and infinite number of Home Networks Seminar, 2004. The IEEE,
subscribers. Volume , Issue , 3 Nov. 2004.

2008 NCTA Technical Papers - page 222

 SYSTEM OVERVIEW AND TECHNICAL DATA RESULTS AND ANALYSIS
FROM HBO’S FIELD TEST OF DVB-S2 AND MPEG-4 HD DEPLOYMENT
Andrew Levine
Home Box Office, Inc.

Abstract equipment from several vendors for testing at

On June 12, 2007, Home Box Office the HBO satellite uplink facility. Once the
announced it would make all 26 HBO and equipment selection was made, an engineering
Cinemax channels available to HBO field test plan was written and implemented so
distributors in high definition using MPEG-4 “real world” scenarios could be simulated and
compression technology by the end of the observed. The engineering field test provided
second quarter of 2008. The HBO engineering valuable information and validated the
team was tasked with finding a system that performance of the new system.
would meet HBO’s aggressive time frame for
deployment. The system would have to be SYSTEM SUMMARY
capable of high quality encoding, and robust
enough to be able to meet all of HBO’s HBO will have three C-band, 36 MHz
technical requirements. Taking advantage of transponders carrying the 26 HD MPEG-4
the newest MPEG-4 (AVC) compression and channels.
DVB-S2 satellite modulation, HBO felt it would
be able to deploy an efficient and cost effective HBO selected Motorola as its MPEG-4
way to make the 26 HD feeds available. encoding, multiplexer, encryption and satellite
modulation vendor. HBO tested several AVC
This paper will describe the technical encoders from several vendors, most of which
architecture of the MPEG-4 compression system produced good results, but ultimately Motorola
that HBO has chosen to implement. It will also was chosen for its ability to provide a complete
outline the field test plan that HBO developed system including MPEG-4 satellite receivers
for the system, as well as the results of that field that would be available for HBO affiliates in
testing. The field test results and observations early 2008. HBO has a successful history with
will show that this system is a viable alternative Motorola as its provider of analog encryption
to traditional MPEG-2 compression with QPSK (VideoCipher) and MPEG-2 compression and
modulation. encryption (DigiCipher II) equipment.

INTRODUCTION HBO will employ Motorola branded

Modulus encoders to perform the MPEG-4
In order to meet the aggressive time line for compression.
deployment, HBO needed to move rapidly in its
selection of a new system. HBO solicited

2008 NCTA Technical Papers - page 108

 HBO’s MPEG-4 HD/DVB-S2 Encoding and Transmission System
Basic System Diagram
SDI ASI ASI
DVB-S2
To Galaxy
Transport Modulator
Satellite
Multiplexer IF
(TMX)
MPEG-4 IF Switch
Encoders
IF
C-Band
HPA
Upconvertor

C-band RF

ASI
SDI Router Encryption
(SEM)

HBO and Cinemax

SDI 1080i HD Content
from Origination Control Rooms
Figure 1

SYSTEM ARCHITECTURE HBO’s originated HD programming will

feed the MPEG-4 encoders (via SDI) which
The MPEG-4 encoding system comprises outputs an ASI stream to feed a Transport
one rack-unit encoders for each channel of Multiplexer (TMX). The TMX muxes the
video. The system is very scalable and space individual ASI streams together, and outputs a
efficient. The single channel encoders are easily single MPEG-4 ASI steam which feeds a
cascaded and muxed. An identical redundant SmartStream Encryptor Modulator (SEM). The
(back-up) system is configured for each SEM will encrypt the ASI stream and the TMX
transponder multiplex. HBO has decided to will feed ASI to a Newtec DVB-S2 satellite
encode the MPEG-4 channels initially at 8 modulator. The last link in the MPEG-4
Mbps and determined that this encode rate was compression chain is the modulator output to an
the “sweet spot” since quality is very important IF switcher. The IF switcher provides a 70
in addition to optimizing transponder MHz output which feeds HBO’s C-band
bandwidth as much as possible. It should be satellite upconvertors and transmitters. The
noted that 8 Mbps is about half the bit rate of switcher also provides IF monitoring points
typical broadcast and cable MPEG-2 HD prior to satellite uplink.
program feeds. With future encoder
improvements and MPEG-4 tool enhancements, HBO has chosen to use DVB-S2 satellite
it may be possible to encode HD lower with no modulation. DVB-S2 modulators increase
loss in quality. All the MPEG-4 HD services spectral efficiency by using new and advanced
have a single AC-3 English audio program high-level coding techniques.
encoded with the video. The audio is 2.0 stereo
or 5.1 surround depending on the original source As defined by the DVB organization’s DVB-
material. S2 fact sheet, DVB-S2 makes use of the most
current modulation and coding techniques to
deliver performance that comes close to the
Shannon limit, the theoretical maximum

2008 NCTA Technical Papers - page 109

information transfer rate in a channel for a given HBO utilizes Motorola’s Broadcast Network
noise level. DVB-S2 uses a very powerful FEC Controller (BNC) software as the interface to
scheme which is a key factor in allowing the control and monitor the modular system,
achievement of excellent performance in the including the TMX and SEM units. The BNC
presence of high levels of noise and software communicates with HBO’s current in-
interference. The FEC system is based on the house custom authorization
concatenation of BCH (Bose-Chaudhuri- system, so that no major changes or additional
Hocquengham) with LDPC (Low Density Parity software applications are needed to authorize
Check) inner coding. 1 the new MPEG-4 services. Because no major
modifications will be necessary, the HBO
The DVB-S2 modulation scheme is not yet authorization hotline will be able to easily
widely used in the U.S for satellite distribution authorize services and input pertinent affiliate
feeds. It is, however, quite popular in Europe data just as they currently do for HBO’s MPEG-
and Asia and has been proven to be successful 2 linear and HBO On Demand/Cinemax On
domestically with contribution feed users. Demand customers.
Extensive testing performed by HBO and
Motorola with different FEC (forward error
correction) rates led to very positive results. One of the most important of HBO’s
Using a FEC rate of 5/6 yields an available requirements was to have high quality satellite
transponder payload of 72 Mbps. Using more receivers available by early 2008 so that all of
aggressive FEC rates vs. required power, the HBO’s distributors could downlink and process
modulation could be traded off to yield more the new HD feeds without having to integrate
throughput, but 72 Mbps met HBO’s payload additional and/or complex gear to their existing
requirements. 72 Mbps is almost double the headend infrastructure. Working closely with
transponder payload that most providers HBO’s engineers, Motorola was able to develop
currently have with MPEG-2 compression and two receiver solutions to complete their AVC
common QPSK modulation. system.

HBO’s MPEG-4 HD/DVB-S2 Encoding and Transmission System

Reception Options

Decrypts single service and outputs native MPEG-4

and transcoded MPEG-2 ASI and GigE.
•Ideal for MPEG-2 (existing Cable) plants who can send the transcoded
Single-decrypt
MPEG-2 stream directly to ASI devices (TMX, CherryPicker, BMR, etc.)
Transcoder
(DSR-6050) and to QAM. One receiver is required per HBO/Cinemax service.
•For MPEG-4 plants (i.e., IPTV, Telcos) who want to send single service
streams directly to their system.
Multi-decrypt
MPEG-4 Multi-decrypt, native MPEG-4 outputs ASI and GigE. Decrypts up to 64 services.
(DSR-4410MD) •For MPEG-4 plants (i.e., IPTV, Telcos) who can send MPEG-4 stream
directly to their system. Only one receiver required per transponder multiplex.
Figure 2

The first is a native MPEG-4 multi-decrypt The receiver will decrypt all streams on one
receiver, the DSR-4410MD. This unit is of HBO’s MPEG-4 transponders and output
capable of decrypting up to 64 services using a MPEG-4 via ASI and/or Gigabit Ethernet. This
single one rack-unit box. will be beneficial to newer MPEG-4 plants who

2008 NCTA Technical Papers - page 110

will be able to send the native MPEG-4 1) 3.5m diameter or larger antenna (or
channels through their system without the need equivalent gain and noise temp)
for conversions or additional processing 2) C-Band Digital PLL LNB,
equipment. HBO has conducted extensive field Recommended Specifications:
testing of this receiver and DVB-S2 modulation a) Noise Temp 20°K or better
with several MSO headends and affiliate labs. b) Avg gain 60dB
The results have been excellent in all c) LO Stability ±12 kHz or better
parameters, including successful reception links d) Extremely low phase noise
with low power levels in a wide variety of
geographical areas and varying weather HBO provided a testing matrix spreadsheet
conditions. to all test sites. This matrix document allowed
the test sites to record their
Motorola will also make available an MPEG- findings/measurements during the various test
4 – to – MPEG-2 single channel transcoding phases. Weekly conference calls were held with
IRD, the DSR-6050, which will output MPEG-4 the test sites, and e-mail alerts were generated to
and/or MPEG-2 via ASI and/or Gigabit brief test sites on any new developments and/or
Ethernet. This unit will be ideal for procedures.
conventional cable distributors who have
extensive MPEG-2 plants and have numerous Phase I “Out of Box” Experience and Baseline
field deployed MPEG-2 set tops. This will Measurements. (two week duration)
allow existing HBO/Cinemax affiliates to be
able to immediately take advantage of the new The first phase of the field test consisted of
HD MPEG-4 channels. install, set-up and reporting on the “out of the
box” experience. Test sites were instructed to
The IRD will decode the HBO MPEG-4 stream, configure the receivers as per the Motorola
and re-encode it to MPEG-2 all within a single Operator Guide using satellite transmission and
rack-unit device. HBO will set the transcoded service parameters that were provided in HBO’s
MPEG-2 output bit rate to ensure appropriate test schedule. Test sites were instructed to call
quality. HBO’s authorization hotline (just as they would
in a real world scenario) once the receiver was
MPEG-4 FIELD TEST PLAN set up and had signal lock on the test satellite
coordinates. Test sites were asked to provide
HBO developed a four-phase test plan that anecdotal feedback to HBO in regard to ease of
encompassed approximately eight weeks. installation and configuration. Once confirmed
MPEG-4 receivers were sent to 18 test locations authorized for service, the test sites were asked
consisting of various cable and telco operators to record the following initial baseline
and vendor labs. The test locations were chosen measurements:
based on their ability to process and report on
the MPEG-4 signals. It was important to HBO 1) Satellite C+N/No at L-Band (if available)
to select a diverse group of test sites (both in 2) L-Band input signal level at IDR/MRD RF
location and technical infrastructure), as this Port
would represent several “real world” scenarios 3) Front Panel IRD LED indicators, any alarms
which would simulate conditions at HBO’s present
various distributors. HBO specified test site 4) Status Menu/OSD Diagnostics
technical requirements that included the a) Signal Strength
following: b) “Signal Quality”

2008 NCTA Technical Papers - page 111

 c) Eb/No (Energy per bit to noise power modulation without taking bandwidth into
spectral density ratio) account. 2
d) VCT (Virtual Channel Table number)
e) VCN (Virtual Channel Number) Phase III Decoder Output Interfaces (one week
f) Authorization Status duration)
g) BER (Bit Error Rate)
5) Initial ASI or GbE stream analysis (high In this phase of the testing, field sites were
level) asked to measure, observe and document the
various outputs of the decoders in the HBO test
Phase II Transmission Performance (one week schedule. This included taking empirical
duration) measurements and recording the responses of
terminal equipment (subject to test site
The second phase of the field test allowed capability) including adherence to the following
HBO to vary uplink parameters while the test standards as applicable:
sites recorded changes to signal lock and Eb/No
levels on their receivers. This testing helped  DVB ETSI TR 101-290
HBO to determine the optimal transmission  ETSI TR 102-034
parameters for best real-world link performance  IEEE 488.1-1987
with minimal sacrifice of data throughput, while  ITU-T H.264
maintaining appropriate downlink margins.  ATIS
HBO used an FEC modulation rate of 5/6.  SMPTE-292M (HD-SDI Output via
external decoder)
Specifically, HBO attenuated the uplink  SCTE 40, SCTE 20 and other applicable
transmitter power by -1dB increments every 10 standards
minutes until -6dB was reached. At that point
the power was brought back up in -1dB Phase IV Long Term Stability (four week
increments every 10 minutes until full power duration)
was restored. The total test duration was 80
minutes. The field test sites were asked to HBO transmitted continuous program
record the receiver’s Eb/No level at the 10 services to the field test equipment at the test
minute intervals, and also note if the receiver sites during this phase of testing. Test sites
lost signal lock at any time. The identical test were asked to perform qualitative observations
was repeated at the same time the next day to of the decoded services at the video and audio
ensure all test sites were able to observe the outputs of a set top box or other decoding
receivers. devices (e.g., Sencore/Motorola MRD-3187).
Test sites were asked to observe and record (at
Eb/No (or Eb/No) is defined as Energy per bit regular intervals) the picture and aural quality of
to noise power spectral density ratio, and is the signals. Test sites were requested to
considered the measure of signal to noise for a maintain reception of “existing” MPEG-2 feeds
digital satellite/communication or data system. from HBO and compare the feeds “under test”
It is normally measured at the input to the to the existing feeds. No visual differences
satellite receiver and is used as the basic should be observed in either the decoded
measure of how strong the signal is. It is a MPEG-4 or the transcoded MPEG-2 output.
normalized signal to noise ration (SNR) The following items would be monitored:
measure, also known as the "SNR per bit". It is
especially useful when comparing the bit error
rate (BER) performance of different digital

2008 NCTA Technical Papers - page 112

  Eb/No levels should be new technology. Several sites were not familiar
steady/consistent. Any variations should with the MPEG-4 receivers and how to set them
be documented. up and obtain transponder signal lock. HBO’s
 Receiver signal status should be engineering team and Motorola's IRD product
“locked.” Any variations should be team addressed these issues directly with the
documented. test sites. One example of this was whether to
 Integrity and any irregularities of ASI set the multi-decrypt MPEG-4 receivers to
output stream manual or automatic mode. The difference
 Decoded Picture quality (color, motion, being, the ACP (Access Control Processor)
detail, artifacts, disturbances, etc.) addresses would be manually assigned by the
 Decoded Aural quality (levels, end user or automatically populated when the
dynamics, response, distortion, anchor (or “in care of”) unit address number
separation) was authorized. HBO found that it’s easier to
 Lip Sync within one frame (30-mSec) have the end user set the receiver to automatic
 Closed Captioning (CEA-608 and CEA- mode. This works especially well when the end
708) user is receiving all the channels on the same
transponder multiplex.
Test Completion
Another issue (affecting the MPEG-4
At the end of Phase IV, HBO collected and receivers) discovered during Phase I testing was
compiled all spreadsheet data as well as all of a setup step not clearly explained in the operator
the field sites’ subjective comments and guide. It seems that after selecting
opinions. After several weeks of test data an ACP number, the menu curser must be
analysis and interpretation, HBO’s engineering moved to the corresponding program number
team made their recommendations to Motorola and entered by using the up/down arrows for
as to the success of the field-deployed receivers each ACP number that is used. If the program
performance. number is not entered, the receiver will not
authorize that ACP.
MPEG-4 FIELD TEST RESULTS
Despite these minor issues, it seemed to be
Phase I “Out of Box” Experience and Baseline fairly easy for the test sites to set up and have
Measurements their receivers authorized for the MPEG-4
signals. The following chart includes the
Out of the 18 MPEG-4 (DSR-4410MD) receiver’s measured signal quality and Eb/No
receivers delivered to the test sites, one receiver for the 11 field test sites that submitted Phase I
was DOA (would not power up) and had to be data.
replaced.

Phase I revealed to HBO that there were

some “growing pains” to be expected with this

2008 NCTA Technical Papers - page 113

 DSR-4410 Phase I Results Summary Data

Test Antenna Date L-Band Sig. Level Front Panel Front Panel Signal Eb/No
Site Size at LED LED Auth. Quality
RF Input Port Signal Lock

A 4.5m 10/29/07 not available OK OK 94 11.6

B 5m 11/1/07 -29.6 dBm OK OK 98 14.3

C 7m 10/30/07 -15.6 dBm OK OK 99 14.2

D 5m 10/22/07 -47.1 dBm OK OK 98 14.8

E 9m 10/25/07 -38.6 dBm OK OK 100 10.9

F 4.5m 10/16/07 not available OK OK 100 15.0

G 3.8m 10/24/07 -25.3 dBm OK OK 92 10.7

H 5m 10/25/07 -22.8 dBm OK OK 93 11.2

I 3.8m 10/24/07 -48.2 dBm OK OK 100 12.4

J 5m 10/19/07 -47.0 dBm OK OK 95 12.4

K 5m 11/5/07 -20.8 dBm OK OK 100 14.2

Avg.* 5.24m -32.8 dBm 97.2 12.9

Figure 3

*The calculated averages in figure 3 indicate power was attenuated by -6dB all but three sites
that with a standard (or common) still had signal lock with an average Eb/No of
headend antenna, excellent downlink 7.35. So, 75% of the downlink sites were able
performance is achieved. to keep signal lock on the transponder at -6dB
power attenuation. This proves the strength of
Phase II Transmission Performance the DVB-S2 modulation. Except in the very
rare instance of local weather being unusually
Phase II yielded excellent satellite link extreme and/or the HBO satellite link
results. As the following graphs (figures 4 and experiencing a major transmission problem, it
5) will illustrate, 12 test sites were able to would be extremely uncommon for real world
participate in the uplink attenuation phase. conditions to come close to -6dB of uplink
Considering the different geographical locations attenuation.
and different downlink antenna types and sizes,
the average Eb/No at full uplink power was
12.1. When the uplink

2008 NCTA Technical Papers - page 114

 Eb/No Readings at Normal Uplink Power Level

16.0

14.0

12.0

10.0

Eb/No 8.0

6.0

4.0
2.0

0.0
A B C D E F G H I J K L
Test Sites

Figure 4

Eb/No Readings at -6dB Attenuation

to Uplink Power Level
14.0
12.0
10.0
8.0
Eb/No 6.0
4.0
2.0
0.0
-2.0
A B C D E F G H I J K L
Test Sites

Figure 5

validates that nominal downlink sites can

As illustrated in figure 5, only the three test receive DVB-S2 signals with margin.
site locations indicated in yellow E, I and
J (Eb/No 2.0, 6.2 and -2.0) lost signal lock at- Phase III Decoder Output Interfaces
6dB uplink attenuation. The other nine sites
were able to maintain transponder signal lock. In this Phase, the field sites were asked to
Site I tested with a small antenna, and this observe the output quality of the MPEG-4
receivers. Since the test sites all had different

2008 NCTA Technical Papers - page 115

types of measurement equipment (i.e., stream signal lock and authorization status during the
analyzers, etc.), the findings in this phase are Phase IV testing.
somewhat subjective. This actually benefited
HBO in determining how the receiver's outputs Eb/No and signal quality readings remained
would perform based on non-consistent readings steady as observed for several weeks at different
and stream observations that would vary from times of day with varying local weather
site to site. These results again depict “real conditions. The observed overall bit rate was
world" results, as no two headends are exactly consistent with what the HBO uplink was
the same. broadcasting for the duration of the testing. All
HBO required signal attributes (including closed
Many of the test sites were not able to look at captions, parental controls/ratings and 5.1
the MPEG-4 output due to the constraints of audio) were observed to successfully pass after
having an all MPEG-2 (typical cable plant) decoding the MPEG-4 signal. No major MPEG
infrastructure, but several could and found the artifacts (including blocking or freezing) were
outputs to be without any issues or anomalies. observed during the long term testing. Audio
Some sites were able to utilize next-generation and video quality were said to be very good and
set top boxes that indicates MPEG-4 technology there were no instances of receiver failure or
is rapidly becoming available to cable providers. need to re-boot or power cycle the units

The GigE and ASI outputs produced solid Transcoding IRD Testing
MPEG-4 streams. At the time of this testing
HBO was broadcasting six HD channels with a Directly following the successful testing of
total bit rate of about 46 Mbps. One test site fed the MPEG-4 receiver, HBO (in coordination
the ASI output to an active Terayon with Motorola) shipped another receiver to all
CherryPicker and logged zero errors. Another the field test sites. This unit is a single-decrypt
site observed legal value video and normal MPEG-4 – to - MPEG-2 transcoding IRD
audio as they decoded the ASI stream using a (DSR-6050). This new receiver will enable the
Sencore/Motorola MRD-3187. end user to provide a transcoded MPEG-2 ASI
signal that can be easily put into service in
existing MPEG-2 cable plants without the need
The successful Phase III testing indicted the for any additional equipment. The receiver
MPEG-4 receiver would be easily deployable in decodes the MPEG-4 signal and then re-encodes
plants that had an MPEG-4 infrastructure, such it to MEPG-2. This IRD will output both the
as IPTV Telco or in plants that would want to native MPEG-4 and the transcoded MPEG-2.
employ an external decoding device such as the The HBO uplink (utilizing Motorola’s BNC
Sencore/Motorola MRD-3187, or MPEG-4 software) can control the transcoded output data
cable set tops as they become available. rate to insure the best possible quality for each
HBO/Cinemax service.

Phase IV Long Term Stability The test plan for the transcoding IRD was
abbreviated to three phases, with Phase II
As with Phase III, many test sites were (uplink attenuation) being eliminated. It was
unable to monitor actual audio and video output HBO’s feeling that there was no need to
of the MPEG-4 receiver, but those that were duplicate the RF test, since the MPEG-4
able to, provided feedback that indicted the receiver performed extremely well. The front
receiver was stable and ready for deployment. end (and RF) portion of the DSR-6050 is
Some test sites simply observed the Eb/No, identical to the DSR-4410MD, which is based

2008 NCTA Technical Papers - page 116

on Motorola’s popular DSR-4400MD (MPEG-  The DSR-4410MD performed well with
2) receiver. no major issues and should be
recommended to HBO distributors who
wish to receive the new HD services
The initial pre-production version of the with a native MPEG-4 output format.
DSR-6050 had some stability and operational  Stability and operational issues with the
“bugs” which caused the unit to sometimes not DSR-6050 transcoder have been
addressed, and updated (production)
enable the MPEG-2 ASI output and also cause firmware proves this receiver should be
the IRD to intermittently “freeze up”, requiring recommended to HBO cable distributors.
a power recycle and/or factory reset. The initial This will allow cable MSOs to take
units also did not pass digital closed captions or advantage of HBO’s additional HD
channels, without concern about having
5.1 Dolby surround audio. With feedback from to alter any existing cable plant
HBO and the field test sites, Motorola addressed infrastructure or incurring the expense of
these problems and issued a firmware upgrade. purchasing additional processing
New DSR-6050s (with the updated firmware equipment.
code) were sent to the field test sites for
FINAL THOUGHTS
evaluation.
As HD television is becoming more popular
Data from the field sites indicates that
and the related consumer equipment costs are
Motorola successfully corrected all stability and
dropping, consumers have a desire for more HD
operational issues with the 6050 transcoders.
content. HBO has led the charge to increase HD
Closed captions and Dolby 5.1 audio pass
offerings to its distributors and their subscribers.
through the transcoded ASI output without
Implementing the Motorola MPEG-4/DVB-S2
consequence. Many of the test sites reported the
solution will address the demand for more HD
overall (video/audio) quality of the transcoded
content by providing a high quality, scalable and
(MPEG-2) output is as good or in some opinions
cost effective method of providing more HD
superior to HBO’s native MPEG-2 signal. As
channels using HBO’s existing satellite
of this writing, Motorola expects to put the
transponder inventory.
DSR-6050 into full production and make it
available to cable distributors by April 2008.
In launching this new system, HBO
maintains its presence as a technology leader.
Field Test Conclusions
HBO was the first programmer to use satellite
 The RF transmission link (utilizing distribution for its cable affiliates and was the
DVB-S2 modulation) is extremely first programmer to use encryption and digital
stable. compression. HBO also launched the first
 The Forward Error Correction (FEC) satellite-delivered MPEG-2 HD channel and
Rate of 5/6 has been chosen and will
yield a transponder data payload of 72 now will be the first programmer to offer all of
Mbps. This FEC rate is sufficient for its channels in HD using advanced DVB-S2
HBO’s MPEG-4 data requirements and modulation and MPEG-4 compression.
is also quite suitable for successful
headend reception performance with an Contact: [email protected]
average size downlink antenna, and will
not cause any downlink reception issues.

2
NOTES http://en.wikipedia.org/wiki/EbNo

1
DVB Fact Sheet, August 2007,
http://www.dvb.org/technology/fact_sheets/

2008 NCTA Technical Papers - page 117

 THE BENEFITS AND CHALLENGES OF DEPLOYING LARGE REGIONAL VOD ASSET
LIBRARIES.

Michael W. Pasquinilli, Vice President of Engineering

Sunil Nakrani, Research Scientist
Jaya Devabhaktuni, Systems Architect
Concurrent Computer Corporation

Abstract high definition (HD) content. These storage

values could be three to four times larger for a
Many of the domestic video-on-demand system with all HD assets.)
(VOD) systems in service today are being
upgraded to ten thousand hours of storage. Today there is a one-to-one relationship
Much of this storage is for traditional “on- between the VOD system and the digital set-top
demand” assets. There is also an industry trend box control system. The expansion of digital
towards recording an increasing amount of subscriber penetration is causing the digital set-
broadcast content onto the VOD system. top box systems to fragment into several
mirrored digital video systems each with their
The cost of deploying these very large encode, own dedicated VOD system. This means that a
ingest and storage libraries into each VOD single cable system today with one VOD system
system may prevent the launch of these new and one digital set-top box controller may soon
services. This paper discusses the economical split into three or four mirrored systems as digital
benefits and technical challenges of introducing subscriber penetration increases. With the
regional asset libraries that can support multiple current VOD architecture, each of these new
VOD systems. The relationship between network digital video systems will require their own VOD
bandwidth and asset caching will also be library storage.
explored.
The economical challenge facing the MSOs of
OVERVIEW deploying very large VOD asset libraries in each
digital set-top box system is further aggravated
The current baseline storage level for VOD by the fragmentation of these single digital video
deployments is ten thousand hours for many of systems into multiple mirrored systems. So
the major domestic MSOs. Ten thousand hours whether the cable operator has a national
of content, including overhead, equates to footprint of many sites, or is a large single
approximately 24 terabytes of storage. Some of system operator, it is likely that with the current
the more aggressive broadcast models have up to VOD architectures in place today duplication of
three hundred channels of broadcast the VOD asset library will be necessary.
programming recorded into the VOD system and
retained for up to two weeks. If one hundred To address this challenge several engineers in
percent of this content were retained, the VOD the cable industry are working towards
system would require approximately 246 developing a shared or “regional” VOD asset
terabytes of storage. Even assuming that the library that can serve multiple VOD systems.
cable operator is able to secure contractual rights One of the major differences between the VOD
for only twenty percent of this content, the library serving a local VOD system and a
storage requirements are still in the 50 terabyte regional VOD library is that a local VOD library
range. (This does not take into consideration serves a closed network with dedicated network

2008 NCTA Technical Papers - page 206

resources. In this case the local VOD library THE COMPUTER MODEL
may “play” or “stream” the asset directly to the
subscriber. However, in a regional VOD library, Since there are no regional VOD libraries in
the network is likely shared with other data and service in today’s domestic cable market, it was
is several Ethernet switches away from the not possible to gather measured data from an
subscriber. Issues such as QoS (quality of actual regional library. As an alternative,
service), packet jitter, packet routing and trick engineers at Concurrent developed a computer
mode latency make it less reasonable to expect a model that would simulate the operation of a
remote VOD library to stream across very large regional VOD library. Real-world measured
distances. In this case the regional VOD library VOD asset usage data were used to exercise the
must copy the asset to the local VOD system for model.
play out.
The Variables
The design concept discussed in this paper
applies the hybrid VOD model in use today The model had the following adjustable
between the headend VOD library and hub VOD variables:
edge cache devices to a higher level in the VOD
architecture. Now instead of a local headend System Cache: The amount of memory
VOD library, the VOD library is a regional available at the local VOD system to cache
library and, instead of hub VOD cache, the entire content pulled from the regional VOD library.
VOD system served by the regional library
equates to the hub cache. Time To Live (TTL): The amount of time that
a downloaded library asset resides on the local
THE BENEFIT VOD system cache before deletion.

There is clearly an economical benefit in Cache Management: A Least Recently Used

capital savings if a single 100-300 terabyte VOD (LRU) methodology was used to manage the
asset library could serve multiple VOD systems. local VOD system cache.
Just the storage, ingest and streaming costs for
such a system would be well over a million Network Bandwidth: This was the bandwidth
dollars. This does not take into account the costs assigned to each asset being pulled from the
in encoders, MPEG grooming, control systems, regional VOD library. For the sake of simplicity,
powering, cooling and operations. a value of 3.75Mbps was used as a baseline
minimum data rate for each asset pull. (It was
The primary component that could undermine assumed that all assets were MPEG 2 standard
the economic benefit of a regional VOD library definition.) Multiples of this data rate were used
is the network cost to transport the video to each to simulate a “best effort” data rate model. (It is
VOD system. This is where the hybrid VOD more likely that a best effort methodology would
library architecture model comes into play. The be applied to this variable in an actual
primary benefit of a hybrid VOD library deployment.)
architecture is the savings in headend to hub
transport costs. The challenge is to prove that The model assumes that 100% of the VOD
this same savings could be applied to the assets would initially be delivered to the regional
regional VOD library. VOD library. It was also assumed that the first
copy of an asset delivered to the local VOD
system would be cached locally. Assets would

2008 NCTA Technical Papers - page 207

only be purged from the local system if the
allocated cache value was exceeded. The TTL of
an asset would be directly tied to the active usage
of that asset by subscribers. If it was necessary
to delete an asset from the local VOD system
cache, the least recently used asset would be
deleted to make room for the next requested
VOD asset. No cached assets would be deleted
that were in active use.

It was possible in the model to have denial of

service. If the local VOD cache was full of Figure 1. “Three VOD System Model”
assets being actively played by subscribers,
additional requests for regional VOD library The Results
plays were denied. Since the objective of the
model was to discover the network bandwidth The first run of the model assumed that each
necessary to support the regional VOD library, individual asset transfer rate from the regional
no restrictions were placed on the network library to the local cache was fixed at 3.75Mbps.
bandwidth between the regional VOD library and No cap was placed on the network bandwidth
each of the local VOD systems. between the regional library and the local VOD
system. The local VOD system cache value
In several runs of the model the value of the started at 250GB and was incremented by 250GB
local VOD system cache was changed. We were until no Denial of Service (DoS) was
looking for the amount of local VOD system encountered due to lack of local VOD system
cache that would allow for no denial of service cache storage. Results of the first run are shown
and have the largest impact on reducing the in Table 1 below.
network bandwidth between the regional VOD
library and the local VOD system. System Cache @ Peak BW Avg BW
0% DoS
For the purpose of this simulation, actual VOD A 2.25TB 1.67Gbps 0.61Gbps
asset usage data was collected from three large B 3.0TB 2.47Gbps 0.83Gbps
regionally co-located systems that have network C 1.0TB 0.90Gbps 0.32Gbps
connectivity via the MSO’s internet backbone. Table 1. “First Run At 3.75Mbps/Asset”
Figure 1 shows a simplistic block diagram of the
configuration. In the next run all other variables were kept
constant, but the asset transfer rate was increased
to 15Mbps per asset. Results of the second run
are shown in Table 2 below.

System Cache @ Peak BW Avg BW

0% DoS
A 2.25TB 3.7Gbps 0.645Gbps
B 3.0TB 4.4Gbps 0.870Gbps
C 1.0TB 1.4Gbps 0.335Gbps
Table 2. “Second Run At 15Mbps/Asset”

2008 NCTA Technical Papers - page 208

 manageable. The primary technical challenge
Analysis facing the deployment of regional VOD libraries
is in the fact that the VOD standards and
Although System C had the highest number of architectures in place today were not designed to
VOD streams during the data collection cycle, accommodate content not entirely under the
and was very near the stream value of System B, control of the local VOD back office controller.
it required the least amount of local VOD system
cache to store assets pulled from the regional The following section will outline some of the
library. This was due to the following: In major areas that are being addressed in order to
System B the top 20% of assets accounted for support a regional VOD asset library.
83% of the views. In System C the top 20% of
assets accounted for 88.6% of the views. Metadata Publishing

This shows that the usage patterns of the The most common way that an asset is
subscribers in the system can have a direct and received into a VOD system is via a satellite
measurable impact on both the local cache and “catcher.” These catchers pass over to the VOD
network bandwidth necessary to support a system the asset and the metadata file. The VOD
regional VOD library. This is shown to be true back office manages placement of the metadata
even when the two systems have identical into the back office database and the transfer of
content offerings and very similar overall stream the asset into the VOD server(s). There are back
usage. office checks in place today to make sure that the
metadata and assets are both accounted for in the
Another important item to note from this data VOD system.
is that a four times increase of the individual
asset data transfer rate resulted in a negligible With the regional library assets, the ingest
increase in the average data rate and a less than point for the asset is at the regional library.
double increase in the peak data rate. This points Therefore there must be another entity outside of
to the benefits of allowing a best effort transfer the local back office that is keeping track of the
of the library assets up to the ingest data rate of metadata for the assets stored on the regional
the local VOD system. Using the local caching library. This other system must then publish to
method resulted in a combined peak network the local back office the metadata of the assets
bandwidth of only 9.41Gbps. If no local caching stored in the regional library. This needs to be
were used and instead each asset was streamed done so that the local back office can manage the
directly from the regional library, the combined provisioning, rules and lifecycle of the asset once
peak network bandwidth required to support it is within local control. Most all of the VOD
these three VOD systems would be 102Gbps. set-top box navigators in use today pull their data
from the local back office. This being the case,
TECHNICAL CHALLENGES the local back office must have all the metadata
for both local assets and regional library assets
The technical challenges facing us today are within its database.
not in fielding these large VOD asset libraries.
We have demonstrated the ability to support The problem here is that the local back office
large storage systems with high ingest rates and does not currently have the means to accept and
very low encoded content throughput latency. act on metadata whose associated assets are not
We can see from the data above that both the locally ingested. (The Comcast NGOD set of
network bandwidth and cache storage values are specifications begins to address many of these

2008 NCTA Technical Papers - page 209

issues, but as a proprietary specification cannot Latency
be discussed in this paper.) A new method needs
to be adopted that allows for the regional library An ideal architecture would have the
asset metadata to be identified as representing subscriber oblivious to the fact that a VOD title
remote content and changes must be made to the was being served from a regional library versus a
back office to recognize and manage assets not local VOD server. But this behavior is not
stored locally. currently guaranteed.

Trick Modes With proper network provisioning and QoS,

the path from the regional library to the local
It is most often the case that VOD trick mode VOD system should not contribute to the latency.
files and/or indexes are created during asset However, the methods for how the local VOD
ingest. This trick mode creation usually happens system ingests and creates trick modes and
as the asset file moves from the catcher into the propagate content will be the most likely cause
VOD system. Each VOD pump vendor has their of latency. Some VOD systems may not be able
own methods of creating trick mode to play an asset until the entire asset is copied
files/indexes. into the VOD server. This restriction may also
apply to the creation of trick modes. Depending
Since it is envisioned that a regional VOD on the ingestion point into the system and the
library would support multiple VOD pump methods of content propagation, there could be
vendors, the trick mode creation would not queuing delays before the asset arrives at the
happen at the ingest point at the regional library, VOD server designated for play out.
but must instead happen at the time when the
asset is delivered from the library to the local Asset Lifecycle
VOD system. Depending on how this process is
done, it could result in commercially Today the lifecycle of a VOD asset is defined
unacceptable latency for the subscriber. (An by the metadata associated with the asset. The
alternative approach considered is requiring all VOD system will retain the asset for as long as is
VOD vendors to adopt a common trick mode specified by the metadata. Since both the
standard.) regional library and the local system would have
access to the metadata, both systems can
A second consideration related to trick modes continue to use this information. However, for
is the effect of the fast forward function. It could the purpose of managing local storage, a library
be possible that a subscriber tries to fast forward asset “copy” must be marked eligible for deletion
to the end of the asset faster than the file transfer prior to the asset expiration date.
speed for that asset. This is why I mentioned
earlier in this paper that a best effort If a local system allocated one thousand hours
methodology of file transfer from the library to of storage for caching library content, some
the local VOD system would be preferred over content would need to be deleted at times to
the fixed “play rate” of the title. Otherwise the make room for other requested library content.
regional library must somehow support the The regional library may contain many tens of
function of trick mode play out especially for the thousands of hours of content and not all content
fast forward function. pulled down to the local site could necessarily be
retained for the full metadata-specified viewing
period. Therefore assets pulled from the regional

2008 NCTA Technical Papers - page 210

library must be “allowed” to be deleted to of systems supported by the regional VOD
adequately manage the local cache. library.

Bookmarks, Active Rentals, Resume Viewing An alternative way of dealing with ad zones is
to insert the local advertisement at the point of
One of the major challenges of a regional play out. In our case this would be at the local
library is supporting the subscriber experience of VOD system’s VOD server. Since VOD
being able to view an asset multiple times within provides a dedicated session to each subscriber
the rental window (usually 24 hours). To on demand, there is an opportunity to target ads
understand the issue consider the following at the subscriber level or to at the very least keep
scenario: The subscriber watches all but the last the ad zones intact.
ten minutes of an obscure video that was pulled
down from the regional library. The subscriber CONCLUSION
returns some twenty hours later to resume
viewing the remainder of his bookmarked video. Based on the computer model and the
Since this was an obscure video it was likely measured VOD asset usage data from just three
purged from the local cache to make room for VOD systems it would seem that a LRU
more popular or recently requested assets. In this managed local VOD system cache of about 1,000
case the subscriber only wants to watch the last hours per system and total network bandwidth of
twenty minutes of the asset. Does the library at least 5Gbps (peak) per system would be a
download the entire asset again to the local cache good starting point to support a regional VOD
or a partial file? Since it may be too difficult to asset library. This does not take into
push bookmark information all the way back to consideration the impact of high definition
the regional library, and it may be too difficult to content on these variables. Also, this specific
manage file fragments within the local cache, it model does not include “Start Over” type content
is likely that the entire asset must be copied again usage.
to the local system.
It is important to reiterate that a change in
Ad Zones subscriber usage patterns can significantly
change the resources necessary to support a
The largest regional VOD libraries are likely regional library. This being the case, computer
to be those made up of broadcast content modeling that utilizes actual measured asset
supporting “Look Back” and other network DVR usage results will be more important in defining
types of services. Ideally the cable operator resource requirements than will anecdotal or
would like to record just one copy of a regional historical experiences.
broadcast channel. Ad zones make this difficult.
In our three system examples, each of these There are quite a few technical challenges to
systems may have had four or more local be addressed before a regional VOD library is
advertisement insertion zones. One method of commercially viable and transparent to the
dealing with ad zones is to record one copy of a subscriber. These challenges represent changes
broadcast for each zone into the VOD system. to the VOD back office, VOD system and VOD
These broadcast recordings would start to add up server. When all of these challenges have been
as they are multiplied by the number of channels met the VOD system will have evolved into a
that have local advertisement opportunities, video delivery platform that will satisfy all of the
times the number of ad zones, times the number on-demand and broadcast needs of the MSO at
the national, regional and local level.

2008 NCTA Technical Papers - page 211

 THE COX NATIONAL BACKBONE:
BUILDING A SCALABLE OPTICAL NETWORK
FOR FUTURE APPLICATIONS AND NETWORK EVOLUTION
Dan Estes, Cox Communications
Gaylord Hart, Infinera

Abstract from a variety of interexchange carriers. These

costs had steadily risen as bandwidth needs
Cox Communications has recently begun increased and as consolidation occurred among
building out a national DWDM optical the various carriers. The business case for this
backbone which will run over a combination of national backbone network was a classic
owned and leased dark fiber spanning over example of build versus buy with several
12,000 miles. While Cox could have leased unknowns thrown in for fun. For example, we
transport capacity from a national carrier for had to anticipate the market dynamics of the
this purpose, the build vs. leased capacity future lease alternatives while also forecasting
analysis showed the higher costs of leased the demand for future services that had not been
transport would not be economical in the future. clearly defined.
In the final analysis, the business case for
building and operating this network was based However, there were some very compelling
primarily on the rapid growth of Cox’s cable “knowns.” History had shown that our cable
modem data services alone. However, this modem and business Internet traffic had
network provides additional incremental doubled every twelve to eighteen months with
economic benefits by allowing cost savings corresponding complexity and cost increases.
elsewhere in the network (e.g., by building Speed increases for cable modem and business
consolidated national headends) and by services were common with a corresponding
enabling new revenue generating service increase in packets being delivered over the
opportunities not traditionally addressed by existing leased backbone. Backbone circuits
cable operators (e.g., a national footprint for were filling at a rapid pace with long delays in
commercial services or cell service backhaul). getting new links in service. To compensate for
the delay in adding capacity, we were having to
Two paramount design considerations for order new circuits at about a 65% fill point due
the network were total cost of ownership and to the long lead times that some interexchange
service reliability. Other important carriers had in their order fulfillment processes.
considerations were network scalability,
network flexibility, and the ability to rapidly It is interesting to note that while the costs of
turn up new bandwidth and services. To meet transporting a megabit for a mile had declined in
these requirements, Cox is implementing its the period from 2003 to 2007, the bandwidth
national backbone with digital ROADMs needs far outpaced this declining cost rate.
(reconfigurable optical add/drop multiplexers). Figure 1, below, shows the aggregate bandwidth
growth of Cox’s network for the last 18 months.
INTRODUCTION With demand continuing to increase, the
business case was relatively straight forward.
Cox Communications undertook this project
for two main reasons: costs and scalability. Another driving factor in the decision to
Since 2001, with the demise of the @Home build this network was operational simplicity.
consortium, Cox had leased intercity transport In the current mode of operating, Cox had to

2008 NCTA Technical Papers - page 155

coordinate multiple entities in the turn-up of a circuits. One example of this was a market
new circuit. First, we would contact the request to connect multiple 10 Gig circuits from
interexchange carrier about three to five months a data center in one region to a colo facility in
before a cross section was expected to exhaust another. Because we had to wait on Type 2
its existing bandwidth with a circuit order for a carrier circuits, we missed the opportunity to
new intercity link. This was followed up with serve this customer.
equipment orders to up to four vendors for
DWDM and SONET last mile connections from One side benefit that has emerged for Cox’s
the interexchange carrier’s POPs to the Cox national backbone that was not considered in the
regional data centers where the circuits original business case is rapid disaster recovery.
terminated. Circuit turn up, interexchange During the California fires in the fall of 2007,
carrier acceptance testing, and end-to-end several fiber cables were burned, and capacity
throughput verification would follow. The for Cox’s San Diego and Orange County
entire process would take months from markets was diminished. The question arose as
inception to completion. If Cox owned its own what would have been the impact had this
national network, especially one in which national fiber network been in place. The
circuits could be seamlessly engineered from a answer came back that we would have been able
remote, centralized center, then capacity to re-home several wavelengths with a few
additions could be enabled quickly and with far keystrokes that would have bypassed the
less complexity than the leased mode. damaged fiber. The modular design of the
DWDM equipment we are deploying in our
A third driving factor in the decision to build backbone is such that the minimum bandwidth
a national backbone was one of flexibility. We that is installed in any cross section is 100 Gb/s.
had experienced unanticipated demand where In many cross sections, this full capacity is not
we could not respond fast enough using leased assigned on day one, making available some

Figure 1 -- Cox Communications Aggregate Internet Bandwidth Growth

2008 NCTA Technical Papers - page 156

 Figure 2 -- Map of Cox's Backbone Network

wavelengths that can be used for protected example, we have a softswitch in Atlanta that
services or that can flexibly be assigned as controls the VoIP call setup for multiple
needed for unusual circumstances. While each markets across the country. Signaling traffic
potential disaster cannot possibly be anticipated, from these markets flows over the backbone to
it is good to note that some outages can be Atlanta for hundreds of thousands of calls each
recovered from rapidly with a flexible DWDM day. Network reliability is of the utmost
network design that can provide end-to-end, importance since these calls include 911 and
any-to-any connectivity between all nodes on other potentially life threatening emergency
the network. The network that emerged from calls. Long distance traffic also rides the
these design considerations is shown in Figure backbone. We have class 4 control points such
2, below. that any calls that originate in a Cox market and
terminate in another Cox market is transported
NETWORK APPLICATIONS over the backbone’s IP infrastructure and are
terminated on trunking gateways in the remote
Cox’s primary use of the national backbone endpoint. These trunking gateways not only
will be for Internet access for our cable modem connect these calls to Cox’s local telephony
customers. We have designed our routing tables network, they also connect to the local Public
so that any Internet traffic that can be passed Switched Telephony Network. The incremental
directly at peering centers egresses from our savings of providing these services over our
network at these locations. While we still have own facilities amounts to millions of dollars a
several locations around the country with Tier 1 year.
Internet transit ports, due to the costs of
delivering traffic to these portals, we easily A very interesting byproduct of having a
justify the costs of building our backbone into national backbone is the ability to provide
the national and regional peering centers. We national distribution of high quality video
have points of presence in Palo Alto, Dallas, content. One of the needs in a Hybrid Fiber
Atlanta, Ashburn, New York, Los Angeles, and Coax architecture is to maximize the use of the
Chicago. available spectrum on the coax plant. We have
begun to digitize channels and distribute as
Cox also uses the national backbone to many into a 6 MHz QAM channel as we can at
deliver voice signaling and bearer traffic. We a quality level our customers’ expect. Most
have several soft switches that are used for multisystem operators have settled on
regional control of the VoIP endpoints. For statistically multiplexed groups of 12 MPEG2

2008 NCTA Technical Papers - page 157

standard definition channels with a maximum of Gigabit per second services on a competitive
two MPEG2 high definition channels per QAM. basis. In partnerships with other carriers and
Our marketing groups continue to ask for more other Local Exchange Carriers, we can serve
channels and higher quality, especially given the any location in the United States.
marketing hype from the Direct Broadcast
Satellite industry about “hundreds” of high NETWORK DESIGN CONSIDERATIONS
definition channels. A technical solution to this
competitive threat was needed. While MPEG4 In the selection of the DWDM vendor and
is a possible solution, the millions of MPEG2 the dark fiber provider for this project, there
capable set top boxes that are in our customers’ were two overarching design considerations:
homes means that solution is still a way off in first, the total cost of ownership of the network,
realistic deployment scenarios. A solution that and second, the reliability of the end-to-end
could be deployed faster was needed. services. Each market was to be designed so
Technology is available today using specialized that no single point of failure exists in the
encoders and closed-loop statistical multiplexers network. This includes dual entrance facilities
to enhance the capacity of a 6 MHz QAM. The to each building, bypassing some locations with
cost of these encoders and multiplexers is manhole fiber splices instead of entering the
prohibitive to deploy in every headend across building and patching through a fiber cross
the country. With available bandwidth on a connect, redundant power, and back-up power at
national backbone, the economics of a couple of every location where we deployed electronics.
centralized digital encoder headends and Hut spacing and the quality of the interoffice
nationwide IP distribution of multiplexed HD fiber also made a big difference in the overall
content are very attractive. By centralizing the design and costs of the network. While closely
encoding to only a few locations, Cox will be spaced re-gen sites might seem like they would
able to provide hundreds of high definition increase reliability, the trade-off of increased
choices and several hundred standard definition electronics costs did not justify the extra
choices to all markets across the country. As benefits. What emerged was a balanced
well, the higher cost, closed loop encoders network that met reliability and throughput
would provide a quality that is indistinguishable needs while maintaining reasonable hut spacing
from the current generation of encoders in use in across the country. It is interesting to note the
most headends. The national backbone enables variability of hut spacing as shown in Figure 3,
the video feeds to be fixed routed using MPLS below.
so that the path from any one of the national
headends arrives at a receive site from two Selection of a dark fiber vendor was pretty
diverse paths at all times. limited in that we wanted to use a single
provider for as many cross sections as we could.
Cox commercial business services also This would build a good relationship with that
benefits from this national footprint. Inter- provider as well as simplify the operational
market services such as Ethernet Private Line, handoffs in fiber restoration and turn-up.
High Capacity point-to-point services, and Another criterion was the quality of the fiber
point-to-multipoint WAN services can now be itself. Significant penalties would accrue if the
offered to business customers throughout Cox’s overall fiber types were such that extra re-gens
service area. Cox was already providing some or extraordinary dispersion compensation would
lower capacity services using the leased be required. Figure 4, above, shows that the
transport capacity. With a larger capacity vast majority of fiber for the Cox national
network of its own, Cox can offer up to 10

2008 NCTA Technical Papers - page 158

backbone is Corning E-LEAF, the desired fiber limitations of all-optical ROADMs and permits
from a design standpoint. unlimited node counts and network sizes and
flexibly supports multi-degree mesh networks.
There were many available options for
optical networking equipment and technologies, Network flexibility was a major engineering
but after evaluating these options against Cox’s criteria for the Cox national backbone. Some
network requirements, Cox decided to build its DWDM solutions would force a custom design
network with digital ROADMs. Digital with Dispersion Compensation Modules
ROADMs perform an optical-electrical-optical engineered for the maximum point-to-point span
(OEO) conversion for every DWDM distance anticipated at the time of the initial
wavelength at every node. While this may engineering of the route. While a network
seem expensive, modern photonic integrated could be designed in this manner, it would lock
circuits (PICs) have reduced this cost the solution into a specific set of origination and
substantially. For reconfigurability, the OEO termination points. If there was an
conversion allows the use of integrated digital unanticipated need, the network would have to
electronic switches in the ROADM instead of be re-designed and possibly reconfigured. This
all-optical wavelength-only switches. is not a problem with a digital DWDM ROADM
that provides 3R regeneration of the optical
signal at every re-gen and terminal location.
Network re-routing and short-term bandwidth
needs can also be accommodated easily with the
add/drop nature of the digital network.

Number of Huts
Figure 3 -- Distribution of Hut Spacing
Distances

Digital ROADMs typically perform a 3R

(regenerate, reshape, retime) operation at every
node for every wavelength, and this Figure 4 -- Fiber Types in the Cox Backbone
significantly reduces the optical engineering
complexity for digital networks built with these
ROADMs. What used to be a large and Protection alternatives were hotly debated
complex engineering problem for deploying all- topics during the initial design phase of the
optical ROADMs, the necessity of calculating national backbone. Cox made the decision to
worst-case performance characteristics for a use layer 3 protocols as the main protection
number of optical parameters for every path for mechanism in the network. While we could
every wavelength in the network, is reduced to a build a network that was completely redundant,
simple span engineering solution for digital cost analysis showed that if each wavelength
networks. In a digital network, only individual were protected on a 1:1 basis that the equipment
optical spans between adjacent nodes must be costs would increase by 80%. This was
engineered to guarantee any-to-any connectivity primarily due to the fact the protected service
of every node in a network. Moreover, the would always require an equivalent and
digital ROADM’s 3R OEO architecture dedicated amount of transport capacity for
eliminates the cumulative optical impairment protection as the primary path. One of the

2008 NCTA Technical Papers - page 159

features of a digital ROADM, since it is not Client interfaces are only required when
based upon transponders for transport, is that it services are actually turned up. When
allows you to have one wavelength protect additional services are needed, individual client
multiple other wavelengths. In general, we have optical modules will be sent to the two
settled on a 9:1 protection scheme such that one endpoints a couple of days ahead of the need.
wavelength in each cross section is available for Technicians, either Cox personnel or
maintenance or backup bandwidth. In some SmartHands technicians in remote collocation
cross sections where we have through routes environments, will slot the daughter boards as
from one market to another that bypass the local instructed in a work package. The ROADM
add/drop function, we are setting aside will then signal through its data communications
additional protection wavelengths, again on a channel to the centralized provisioning center of
many-to-one protection scheme. While we are the presence of the new plug-ins. Using point-
incurring some increased costs, the overall and-click provisioning, electronic cross
protection costs are still much lower than 1:1 connects would then be made to pass the
protection, and in turn we have increased route incoming traffic to the appropriate output port,
survivability and provided flexible bandwidth wavelength, and timeslot. End-to-end routing of
configuration alternatives that provide the new services is accomplished remotely via a
operational benefits to the network. GMPLS control plane, and no truck rolls are
required to intermediate sites in the path, and no
OPERATIONAL CONSIDERATIONS additional optical layer engineering is required.
End-to-end connectivity can be established in a
When you think of the operational matter of days instead of the months it takes in a
implications of a national fiber optic network leased circuit environment.
there are some key concepts that come to mind
such as rapid capacity activation, detailed During a turn-up event, the ability to
performance monitoring, specific alarm generate a test signal from the client interface
notifications that are pertinent to the problem, itself would be useful and would enable remote
the ability to rapidly isolate trouble, and having confirmation of circuit continuity even if there
positive assurance that services are being was not a technician present with a portable test
delivered as expected. Let’s delve into these in set. This is very easily possible in a digital
more depth. world. One of the benefits of a digital ROADM
is that performance monitoring is provided for
Rapid capacity activation is enabled in this every wavelength at every optical to electrical
network through a combination of upfront conversion point. In a digital optical network,
engineering and the use of digital ROADMs. this occurs at every node. With a G.709 digital
Cox has taken the steps to position the chassis wrapper applied to each service path in the
and re-gen sites for one year’s forecasted digital ROADM, bit error rate (BER) testing
growth. Once the chassis and common may be carried out prior to service turn-up using
equipment such as network management cards an internally generated pseudo-random bit
and bandwidth multiplexers are installed, adding stream or on live services in real time.
bandwidth capacity on the digital ROADM is
relatively easy and is accomplished in 100 Gb Every service interface has a corresponding
increments by adding line cards. Rarely would set of performance statistics derived from the
any one cross section experience more growth G.709 overhead. Appropriate threshold
than that in one year. crossing alerts can then be set to notify the
Network Operations Center of any degradation

2008 NCTA Technical Papers - page 160

in services as well as the specific span in the flexibility tomorrow. Key elements of this
network where this degradation has occurred. requirement are scalability, rapid service turn-up
From the viewpoint of maintaining a national and cutover, non-disruptive upgrades and node
fiber optic network, this type of visibility is additions, guaranteed any-to-any connectivity
paramount to providing reliable services. between any two nodes in the network, and an
optical layer that requires little or no re-
Along with the performance monitoring on a engineering as the network evolves.
span by span basis, another attribute that is
necessary in a national optical network is alarm Digital ROADMs provide a reasonable
filtering and suppression. If there is a major solution to all these requirements, and this is
outage in the network, there could be hundreds possible because a digital ROADM, due to its
of alarms generated from loss of services in all OEO and digital switching architecture,
of the daughter circuits that ride over the segregates the optical layer from the service
system. An intelligent Network Management layer. This makes service delivery independent
System is required to filter those alarms, of optical layer engineering and brings unique
suppressing the non-pertinent ones. Consistent capabilities to the network. At the optical layer,
naming conventions and intelligence embedded network design is reduced to simple span-by-
in the alarm processing system provide span optical engineering, and once the initial
information that will assist in pointing the span has been engineered, adding additional
technicians to the correct source of the problem bandwidth is accomplished by adding line
and allow more rapid service restoration. modules with no additional optical layer
engineering required.
Finally, a wonderful feature in a digital
ROADM is the ability to loop back sub-rate In the initial network design, one simply
interfaces at any electrical conversion point in provides a pool of available bandwidth at the
the network. This loopback capability can be optical layer. This pool of bandwidth is then
applied at a re-gen site, add/drop location, or used as a resource to be allocated to services as
terminal location to quickly isolate intermittent they are turned up. Allocation is implemented
trouble. It will be used in Cox to sectionalize at the node through the integral digital switch,
the network between maintenance entities where and allocation occurs across the network under
we have contracted maintenance activity in the control of a GMPLS control plane. This
certain cross sections of the country to an allows services to be routed through the network
outside party. via any combination of nodes that have
available bandwidth along the path. Providing
NETWORK GROWTH AND EVOLUTION sufficient bandwidth is available at the optical
layer, it is possible to connect any two nodes in
If optical transport networks simply grew a digital network with any service.
linearly, then network engineering and
operations would be a straightforward matter. Treating optical layer bandwidth as an
As we all know, this is rarely the case: services allocatable resource has other benefits, as well,
grow faster than expected and not always where including bandwidth conservation. Because the
expected, new services are introduced, new digital switch in a digital ROADM mediates
nodes need to be added, and sometimes whole between the optical layer and the service layer,
cable systems are sold to or acquired from other it can aggregate and groom multiple sub-rate
MSOs. All this points to the necessity of services onto a single optical layer lambda. For
engineering a network today for maximum example, multiple GigEs can be groomed onto a

2008 NCTA Technical Papers - page 161

single 10G wavelength. And it can do this on a around the failed path. This provides a cost-
service by service, wavelength by wavelength, effective protection mechanism without the
and span by span basis, ensuring optimum usage need for dedicating 1:1 protection bandwidth
of available bandwidth throughout the network. that cannot then otherwise be used.
In a similar fashion, super-lambda services can
be created by allocating one high-bandwidth Finally, digital ROADMs permit multi-
service across multiple optical layer lambdas. degree mesh networks to be efficiently designed
For example, a 40G client service can be and turned up. In this case, the ROADM’s
allocated across four 10G lambdas for transport integral digital switch is simply used to switch
at the optical layer. This allows 40G services to service traffic onto the appropriate optical layer
be transported across a network designed for interface for the intended mesh path. No
10G services, and without any network re- complex optical layer engineering is required
engineering. because the additional degrees coming off a
node are still treated as single spans at the
Digital ROADMs, because of their inherent optical layer. This capability allows a national
ability to switch services over any available network to be built with integral regional
bandwidth in the optical network, provide networks and thus allows regional and national
significant network migration capabilities which traffic to be carried over a common network.
can be used at any time to reroute traffic The mesh nature of such a network provides
through the network for load balancing or more traffic routing and protection options than
bandwidth optimization or to perform traditional rings.
maintenance operations such as adding a new
node in the network. Digital ROADMs SUMMARY AND CONCLUSION
typically provide a bridge-and-roll function that
allows a second, parallel path to be created Cox’s construction of a national fiber optic
through the network between two end-nodes based network has positioned Cox to
carrying live traffic. Once this “bridge” is economically provide expanded services for the
created, traffic is “rolled” onto the second path residential and business communities that it
in under 50 ms, and the first path is then free for serves. It has also given Cox a competitive
re-use or can then have maintenance operations advantage in rapidly responding to changes in
performed on it without disrupting live traffic. offers from Incumbent Local Exchange Carriers
and Direct Broadcast Satellite companies. The
The digital ROADM’s ability to reroute scalability of the Cox national backbone and the
services through the network in real time can digital ROADM used to implement it will also
also be used to provide flexible service provide Cox a way to deliver business services
protection options that eliminate the need for that have not been offered by an MSO in prior
providing dedicated, redundant optical paths for years and is an excellent complement to the Cox
every service on the network. Using shared local “last-mile” network. Given Cox’s
protection, sufficient additional optical layer operational excellence, this network will
bandwidth is provided for service protection in provide for outstanding levels of service for
the initial network design, and this additional higher speed data offerings, expanded video
capacity can be allocated on the fly to support lineups, and reliable voice applications, both
alternate routing of any failed path. Should a wired and wireless, for years to come.
fiber cut or equipment failure occur, the
GMPLS control plane recognizes this and
immediately and automatically reroutes traffic

2008 NCTA Technical Papers - page 162

 TIMELY AND SECURE:
REAL-TIME PERFORMANCE CHALLENGES OF CONTENT SECURITY
Reza Rassool, Chief Engineer,
Widevine® Technologies Inc.

visual illusion relies on precise timing of the

Abstract delivery of each sample of content. It is this
most basic illusion that must be maintained to
Encryption, authentication, and key distribution ensure that the delivered content is received in
are the mainstays of digital rights management the condition intended, and achieves its potential
(DRM) and conditional access (CA) systems in value in maintaining the attention of its
modern entertainment networks. In the audience. Traditional over the air broadcast
traditional DVB CA security model1, entitlement networks delivered a consistent stream that did
control messages (ECM) and entitlement not suffer from the temporal distortions of
management messages (EMM) are inserted into modern packet-based networks.
an encrypted MPEG stream. These messages
are received in a timely manner by a subscriber User Perception of Timing
device, to enable it to access the stream data. In
more modern delivery networks, watermarking, Psychologists would identify audio
fingerprinting, and digital copy protection are video timing as a hygiene factor. Herzberg
additional processes that have been inserted suggests a model for human motivation2
into the pipeline to secure the business of on- wherein certain essential factors are considered
line entertainment. All these security processes pre-requisites. Only once these hygiene factors
introduce measurable temporal distortions in are satisfied can we be motivated by other
bandwidth, latency, and jitter to the smooth factors. Herzberg’s original work concerned the
flowing of entertainment content to subscribers. motivation of employees. Since then,
While basic real-time requirements stem from Herzberg’s work has been applied to consumer
linear broadcast applications, file-based motivation. Consumers are typically motivated
delivery imposes a new set of constraints that to visit restaurants based on the menu rather
challenge engineers to deliver content in a than the quality of service. A certain level of
secure and timely manner. File-based service is a pre-requisite; in the same way
distribution calls for security processing that viewers expect to enjoy a movie where each
scales, persists and is faster than real-time. This frame and sample is delivered on time.
paper quantifies the potential temporal
distortions in the DVB CA security model, It turns out that hygiene factors cannot
detailing the perceptible effects on channel motivate a consumer. But, their deficit can
change time, temporal jitter and latency. certainly demotivate, in the same way that poor
service would negatively affect the enjoyment
of the meal - no matter how good the menu. For
INTRODUCTION the audience of this paper it is especially
important to understand the hygiene factors of
Television and movie content is, by its the digital television business. One of these is
very nature, a temporal experience. Frames and audio video timing. Surprisingly few studies
samples are presented to us in rapid succession have explored the area of user perception of
to give the illusion that we are observing objects temporal distortion in audio and video.
in motion and listening to sound. The audio-

2008 NCTA Technical Papers - page 79

 Distortion by Frame rate changes an interesting reason for this. A 1 kHz tone with
a small amount of flutter (around 0.1%) can
The illusion of motion can be maintained sound fine in an echo-free environment, but in a
at quite a low frame rate. It is surprising that reverberant room constant fluctuations will
utility is found in video conferencing systems often be clearly heard. 4 These are the result of
operating at less than 10 fps. The content frame the current tone ‘beating’ with its echo. What is
rate should not be confused with the display heard is quite pronounced amplitude variation,
frame rate. Even though the old silent movies to which the ear is very sensitive.5
had 16 frames of content per second, the
projectors would display each frame three times Distortion by Jitter and Latency
resulting in a display frame rate of 48 fps. The
display frame rate is critical and is related to, While over-the-air broadcasts deliver
but not identical to, a physiological concept isochronous streams in real-time, modern
called the flicker fusion threshold or flicker networks burst packets of data that need to be
fusion rate. Light that is pulsating below this buffered in memory for variable lengths of time
rate is perceived by humans as flickering; light and need to be processed to differing extents
that is pulsating above this rate is perceived as depending upon the type of data in the packet.
being continuous. The exact rate varies
depending upon the person, their level of Jitter is caused when a processing
fatigue, the brightness of the light source, and element in the pipeline operates in bursts. The
the area of the retina that is excited. Few people result is that the time each data packet spends in
perceive flicker above 75 hertz for CRT the element, from input to output, is not
monitors. A flicker free display is a hygiene constant. Even though the long term flow rate
factor. The content rate and the display rate through the processing element may be constant,
must be controlled independently. The display the instantaneous rate fluctuates. Jitter causes a
rate must be tightly locked to a steady clock, problem for subsequent downstream processing
while the content must be delivered at a rate that elements. Either their buffers overrun due to
was intended to maintain the illusion of motion. receiving a burst of several packets, or their
It is well known that content played at the buffers run empty due to gaps between the
wrong content frame rate adversely affects the bursts of data packets. Jitter is resolved by larger
viewer experience. Even the most dramatic buffers or by ensuring that each processing
Lillian Gish movies seem comical when played element operates in a timely manner.
at 24 fps. But at the original 16 fps, the content Timestamping each packet on arrival and
delivers the intended impact. holding the processed packet, until a fixed
period after the timestamp, before outputting it,
Distortion by Audio frequency changes3 reduces jitter to the resolution of the timestamp
but introduces a fixed latency.
In many respects, audio timing needs to
be more stringent than video timing. In the Buffer overrun in an element often
range 1kHz to 8kHz, the human ear can detect a results in packet loss, while buffer under-run
pitch shift that results from a change of requires the processing element to deploy an
frequency as little as 0.2%. Wow and flutter is under-run strategy. An MPEG decoder, for
particularly audible on music with oboe or piano instance, would repeat the previous frame at the
solo. While wow is perceived clearly as pitch output if the input buffer under-runs. The viewer
variation, flutter can alter the sound of the music sees a freeze frame.
differently, making it sound ‘cracked’. There is

2008 NCTA Technical Papers - page 80

 Latency is a fact of life in transmission out of synch but their evaluation of content is
systems. Latency must be constant so that end- negatively affected.
to-end propagation delay can be used to
schedule live or real-time events. Provided that In 2003, an ATSC Implementation
all elementary streams undergo the same latency Subcommittee (IS)7 studied the issue of AV
then the content is delivered as intended. synchronization. They said that the overall
Viewers with both cable and satellite systems audio-video synchronization error is the
may notice the different end-to-end delays of algebraic sum of the individual synchronization
each service when they switch from one to errors encountered in the chain. While a given
another. Buffering, introduced to smooth out synchronization error may cause either a
jitter, adds to the end-to-end delay. positive or negative differential shift in audio
video timing, the video signal is typically
Distortion by AV Synchronization drift subjected to greater delay than the audio signal,
and the tendency is therefore toward video
Reeves and Voelker6 reported on a lagging behind audio.
Stanford University study. When audio precedes
video by 5 video fields, viewers evaluate people IS finds that under all operational
on television more negatively (e.g. less situations, at the inputs to the DTV encoding
interesting, more unpleasant, less influential, devices, the sound program should be tightly
more agitated, less successful). This difference synchronized to the video program. The sound
is not large, but it is statistically significant. program should never lead the video program by
Viewers can accurately distinguish between a more than 15 milliseconds, and should never lag
television segment that is in perfect synch, and the video program by more than 45
one that is 5 fields out of synch. Viewers cannot milliseconds. In MPEG-2 the end-to-end delay
accurately tell the same segments are 2.5 fields from an encoder’s signal input to a decoder’s
signal output is regarded as constant.

Figure 1 PCR in adaptation field of MTS header

This end-to-end delay is the sum of the Decoder (STD) model allows a maximum
delays from encoding, encoder buffering, decoder buffer delay of one second. Audio and
multiplexing, transmission, de-multiplexing, video presentation units that represent sound
decoder buffering, decoding, and presentation. and pictures that are to be presented
Presentation time stamps are required in the simultaneously may be separated in time within
MPEG bit stream at intervals not exceeding 700 the MPEG transport stream (MTS) by as much
milliseconds. The MPEG System Target as one second. In order to produce synchronized

2008 NCTA Technical Papers - page 81

output, IS finds that the receiver must recover inserts a reference clock so that the receiver may
the encoder’s System Time Clock (STC) and regenerate the STC on decoding. The receiver
use the Presentation Time Stamps (PTS) to places each unit of coded data in a buffer to
present the audio-video content to the viewer generate a delay, then decodes and presents the
with a tolerance of +/-15 milliseconds of the data unit when its time stamp matches the STC.
time indicated by PTS. This process corrects the temporal offset
between the video and audio streams caused by
MPEG TIMING MODEL multiplexing.

MPEG supports timing metadata that Timestamps, tables and their constraints
may be inserted at encoding of the elementary
streams and at packetizing of the MTS. These The MPEG-2 System Standard defines
timestamps are read by the decoder to ensure the two types of timestamp that are added during
real-time performance of the stream. An MPEG- encoding: Presentation Time Stamp (PTS),
2 encoder includes System Time Clock (STC) as indicating time of presentation, and Decoding
a reference time. Time Stamp (DTS), indicating decoding start
time. The multiplexed MPEG transport stream
The system adds an STC value to the (MTS) includes a Program Clock Reference
coded AV data as a time stamp for each unit of (PCR), a timestamp marked periodically in the
presented information, and then multiplexes the adaptation field of the MTS header.
resultant data. Next, the multiplexing system

Figure 2 PTS and DTS in PES header

The PCR allows the receiver to

regenerate a system time clock to match the Clock Recovery Schemes8
timing of the encoding process. The PTS and
DTS timestamps are sent in the PES headers. An ideal MPEG decoder would
The Program Map Table (PMT) associated PIDs implement a numerically-locked loop (NLL) to
with program(s). The Program Association regenerate the 27MHz system clock from
Table (PAT) associated program number with incoming PCR values. 27MHz was chosen
PMT. The Conditional Access Table (CAT) because that is the frequency used to drive the
associated PIDs with private streams. video display electronics. Since the DVB
ATSC and DVB tighten the MPEG-2 specification requires that PCR values are
constraints on timestamps and tables. inserted in the stream, at most, 40ms apart this
requires the NLL to operate between 25Hz and
the MTS packet frequency, 2500Hz (for a
3.75Mb/s stream).

2008 NCTA Technical Papers - page 82

 from transmission jitter, and this will create
188 byte packets are received, demuxed, phase noise in the loop. This is removed by the
and placed in the buffers according the PIDs. loop filter so that a large number of phase errors
Each frame of MPEG data in the PID buffer are averaged over time before affecting the
contains its own timestamp. Once decoded each VCXO. The 48bit counter is divided by 300 to
frame is stored in the display buffer tagged with produce a 33bit counter. The ‘decode’ module
its own PTS. retrieves MPEG data from the PID buffer when
the 33bit counter matches the DTS of the frame.
As shown in figure 3, the NLL9 contains The ‘display’ module similarly retrieves a
a 27MHz VCXO (voltage controlled crystal decoded frame from the display buffer when its
oscillator), a variable-frequency oscillator based PTS value matches the 33bit counter.
on a crystal which has a relatively small
frequency range. The VCXO drives a forty-eight A heavily damped loop will reject jitter
bit counter. The state of the counter is compared well, but will take a long time to lock. Lock-up
with the contents of the PCR and the difference time can be reduced when switching to a new
is used to modify the VCXO frequency. In program if the counter is jammed with the first
practice, the transport stream packets will suffer PCR value in the new program.

description MPEG-2 ATSC DVB

PTS 90 kHz clock Interval <0.7s Interval <0.7s Interval <0.7s
33bit counter Jitter Jitter <15ms Jitter <15ms
DTS 90 kHz clock Interval <0.7s Interval <0.7s Interval <0.7s
33bit counter Jitter <15ms Jitter <15ms
PCR 27 MHz clock Interval <0.1s Interval <0.1s Interval <40ms
48bit counter Jitter <4ms Jitter <4ms Jitter <0.5ms
PAT Lists PMT PID Interval not specified Interval <0.1s Interval <0.5s
PMT Lists prog. PIDs Interval not specified Interval <0.4s Interval <0.5s
10
Table 1 Timestamps, Tables and their constraints

In legacy receivers, the NLL module and 4 bytes of header. This results in 1316
could not be implemented in software by the bytes, plus the packet overhead – 8 bytes for the
main CPU so it was either implemented in UDP header, 20 bytes for the IP header, 14
hardware or was radically simplified. Both bytes for the Ethernet header. (Fig.4)
choices have led to issues in the performance of
legacy receivers. One simplification was to What temporal distortions result from packet
replace the NLL with a 48bit counter driven loss?
from the video electronics 27MHz clock. This
results in a clock that does not dynamically UDP is an unreliable transmission
adjust to reproduce the original timing of the mechanism. Packets can be lost. Loss of IP
encoder.11 packets may occur for multiple reasons —
bandwidth limitations, network congestion,
MPEG TS OVER UDP failed links, and transmission errors. Packet loss
A typical IP packet carrying MPEG-2 usually results in bursty behavior, commonly
video-streaming data consists of seven MTS related to periods of network congestion.
packets, each containing 184 bytes of payload Depending on the type of transport protocol

2008 NCTA Technical Papers - page 83

used for the video streaming, a packet loss will about 2.8ms of content. Assuming also a
have a different impact on the quality of the 192kb/s audio stream, there will be 18 MTS
perceived video. When UDP is used, the lost packets containing video, for each one
packets will directly affect the image, as the containing audio. This means that a single UDP
information cannot be recovered and the image packet has a high chance of containing no audio.
will simply be corrupt or unavailable. When
using TCP, a packet loss will generate a The loss of a silent UDP packet results
retransmission, which can produce a buffer in the video stream jumping forward in time by
underflow and, consequently, a possible frozen 2.8ms while the audio stream is undisturbed.
image. Now, a well behaved MPEG-2 decoder should
present each video or audio “frame” at the
The loss of one UDP packet results in scheduled time – when its PTS/DTS values
the loss of 7 MTS packets. In a 3.75Mb/s match the recovered system clock.
CableLabs stream, one UDP packet represents

PID Display buffer

buffer

Packet Demux Decode Display

PCR

Compare 48 bit
counter ÷300

Loop 27MHz 27MHz

filter VCXO clock
Figure 3 Clock regeneration with NLL

Legacy receivers that omitted the NLL another channel and then switching back. Try it
suffer loss of synchronization. The PCR value is at home!
looked at only when a program switch occurs
and thereafter the system clock runs locked to Variable MPEG processing delays
an internal reference such as the CPU clock or
video display clock. This means that packet loss The MPEG-2 specification states that
results, inexorably, in loss of AV video or audio elementary-stream access units
synchronization. After 15 lost UDP packets, the that do not contain B pictures are to be
drift is noticeable. In a network with just 0.01% transferred immediately from the main buffers
loss, the sync drift would be noticeable after one to the decoders at the time denoted by its PTS.
hour of continuously viewing the same channel. The STD then decodes and outputs the data in
A simple user controlled remedy is to reset the the main buffers when the STC matches the
counter. This is achieved by switching to PTS.

2008 NCTA Technical Papers - page 84

 the main buffer when the STC matches the DTS.
However, a video elementary stream that Thereafter pictures are held in a re-order buffer
includes B-pictures requires that I and P pictures until its PTS matches the STC.12
be decoded before decoding the B-pictures, and
it is for this reason that the decoding time and This means that packet loss can cause
presentation time of I or P pictures differ. In drastically different perceived distortions
particular, the specification states that I or P whether the lost packet contains I, B or P data.
picture data are to be transferred immediately Higher compression results in greater temporal
from the main buffer to the decoder at the time distortion. A lost UDP packet from a 700kb/s
denoted by DTS. The system decoder then H.264 stream represents 15ms!
decodes and outputs the I-picture or P-picture in

Figure 4 UDP packet contains seven MPEG transport stream packets

DVB-CA SECURITY MODEL

The DVB-CA security model comprises up about 1% of the stream bandwidth. The
a combination of scrambling and encryption to ECMs are transmitted in a separate PID that is
prevent unauthorized reception. Encryption is multiplexed in with the original stream. The
the process of protecting the secret keys that are original stream is already time-stamped. The
transmitted with a scrambled signal to enable injection of ECMs causes jittering in the PCR
the descrambler to work. values of the original transport stream. An
important feature of DVB-CA multiplexer is to
ECM perform PCR correction to compensate for this
jitter. ECMs of moderns CA systems now carry
The scrambler key, called the control more that just control words. Watermark
word (CW) must, of course, be sent to the metadata, extended copy control information,
receiver in encrypted form within an entitlement and other metadata would cause the ECM to
control message (ECM). The CW is valid for a grow beyond a single MTS packet.
particular crypto-period (CP) which is typically
10 seconds long. ECMs must be received and In the absence of PCR correction the
the CW extracted and decrypted in advance of packets could arrive in an untimely manner –
MTS packets in the associated crypto-period. If outside the 0.5ms jitter spec of the DVB
the ECM is not available for the associated standard.
crypto-period in time, then the content cannot be
decrypted and the subscriber will suffer service EMM
loss.
The CA subsystem in the receiver will
The ECMs are repeated every 0.1s to decrypt the control word only when authorized
ensure that the stream is still decryptable even to do so; that authority is sent to the receiver in
under severe packet loss. The ECM stream takes the form of an entitlement management

2008 NCTA Technical Papers - page 85

message. This layered approach is fundamental with the content. In IPTV networks the EMMs
to all proprietary CA systems in use today. In a can be sent out-of-band via a reliable
traditional DVB network the EMMs are communication channel.
transmitted in-band, in another PID multiplexed

Figure 5 DVB-CA model

basis. Unidirectional networks such as satellite

SECURITY TIME CHALLENGES need to handle EMMs with special care. The
carousel transmission of the EMMs has to strike
Security affects the timing of streams, by a balance between reliability and security.
introducing jitter and consuming transmission
bandwidth, through the insertion of ECM The service operator must ensure that all
packets. the subscribers receive the EMMs for the
services to which they are entitled, while also
Client-side CPU load being careful not to expose an EMM for too
long to hackers.
The client CPU is burdened with an
increasing load to support more sophisticated Blocking EMMs revocation, an obvious
security systems including processes that insert hack to thwart key rotation, would allow a
watermarks into the content, monitor content to subscriber to access a service long after the
generate fingerprints, and monitor the receiving subscription has expired.
device to ensure that no theft is occurring. This
occurs as the client device labors under a six- In bidirectional IPTV networks, with
fold load increase in the transition from standard reliable TCP/IP communication, the EMMs can
to high definition. It means that timing is be issued on a just-in-time schedule. EMM
becoming ever more critical in modern client acknowledgements can also become part of the
devices. business logic of the service.

Timely arrival of keys In the case of an impulse VOD purchase, the

EMM cannot be pre-staged on the subscriber’s
In the case of linear content the EMMs set top box. To reduce ‘VOD Purchase Delay,’
are typically sent to the client at the time of the latency between the client’s purchase
subscription to the channel or bouquet of request and the start of the movie, the timing of
channels. EMMs are revoked and re-issued to the security communications needs attention.
rotate entitlement keys -typically on a monthly

2008 NCTA Technical Papers - page 86

Figure 6 VOD sequence diagram showing unmitigated VOD Purchase Delay

V O D s e rv e r

C A s e rv e r M id d le w a re s e rv e r M id d le w a re c lie n t C A c lie n t

V O D p u rc h a s e

A d d E n title m e n t

E n title m e n t
VOD
E n title m e n t A C K
Purchase
A d d E n title m e n tR e s p o n s e Delay
V O D p la y
M o v ie

Ordinarily the delay would be several these more secure devices the client can receive
seconds. One method to mitigate the delay is to EMMs with richer rights expressions. A simple
leave a leader of the movie in the clear. The timed entitlement includes the start and end
leader duration is just longer than the maximum time. The client will only use the entitlement
VOD purchase delay. An even more secure after the start time and will purge it after the end
solution is to encrypt the leader with a key that time. This would allow a customer to download
is only issued to those clients that have a movie onto a mobile device and watch it on a
subscribed to the VOD service. This approach plane or boat, disconnected from the CA server.
means that the Middleware server can issue the
‘VOD play’ command to the VOD server as Secure time
soon as it receives the VOD purchase message.
The subsequent CA communication will then Manipulation of the client clock is a
occur in parallel with the playing of the leader. well-known hack to retain expired service.
The EMM for the movie will arrive at the STB Secure clocks are tamper proof and are
in time to decrypt the remaining duration of the protected against unauthorized changes. The
movie. clock can be set through a secure protocol each
time the client connects with the CA server.
Timed entitlement Time should be maintained independently of
local time-zone, using UTC/GMT, to avoid
Normally EMMs are issued and revoked errors caused by the CA client and CA server
by the CA server. In this case the time is operating in different time zones. The CA server
maintained by the server. In the advent of secure should itself obtain time from a trusted NTP
processors, secure memory, and secure clocks, source.
the CA client can be safely implemented to
operate with a higher degree of autonomy. In

2008 NCTA Technical Papers - page 87

Figure 7 Hybrid CDN

File-based transmission

Modern content delivery networks encryption or decryption required to secure a

(CDN) transmit linear and file-based content. file-based CDN.
These hybrid networks aggregate content and
distribute files to remote service operators that TVN Entertainment13, a VOD service operator,
each serve separate populations of subscribers. delivers 5000 hours of file-based content per
In a network, as in figure 7, the security month to affiliate operators around the country.
processing has the traditional real-time
requirements at the point of displaying the This network is both encrypted and
content on the client device. However the rest of watermarked. In 2007, benchmarking tests TVN
the network treats the content as files. Files are Entertainment showed that an off-the shelf
transmitted from the aggregator to the operators single rack unit server, running Widevine
as fast as the satellite transponder allows. In this Cypher® DRM software, can encrypt or decrypt
environment files may be secured by different one gigabyte of MPEG file in 1.5 minutes while
conditional access systems in different legs of a traditional real-time DVB scrambler takes 35
the pipeline. Each file may need to be encrypted minutes.
at the aggregator, decrypted at the operator and
then re-encrypted with the operators CA of CONCLUSION
choice. Then the file is served to the subscriber
where it is decrypted, decoded, and displayed in At IBC07, SMPTE and EBU14 jointly
real-time. At the operator, however, bulk crypto declared:
processing should happen as fast as possible as
files are pitched from the aggregator at 20 times The current methods of timing and
faster than real-time. As described earlier in the synchronization for television, audio and other
paper, legacy CA systems that have moving picture signals rely on standards that
implemented their stream parsing and crypto- have been in place for more than 30 years.
processing in hardware have built their systems While these standards have proven to be robust
around the real-time clocking requirements. solutions that have served the industry well, they
These real-time scramblers and descramblers are predicated on technologies that are
cannot be easily retooled for the task of bulk becoming increasingly inappropriate for the

2008 NCTA Technical Papers - page 88

digital age with, for example, networked content oversimplification of the timing logic or its
sharing or higher frame rate HDTV image implementation in hardware precludes
formats; they now impose unacceptable traditional CA crypto-processors from reaching
limitations for the future. the performance of modern software bulk
encryptor / decryptors. The evolving security
The time constraints on MPEG streams landscape is challenged with emerging
are onerous. The addition of security processing requirements for watermarking, fingerprinting
provides an extra timing challenge for control and copy protection. A software solution is best
logic of servers and clients. Legacy CA systems placed to rise to these challenges - both timely
suffer from a lack of flexibility in dealing with and securely.
the timing of a faster than real-time CDN. The

References Sound and Vision for Broadcast Operations,

Doc. IS-191, 26 June 2003.
1
ETSI (1997), TS 101 197-1 Digital Video 8
Tryfonas and Varma, Timestamping Schemes
Broadcasting (DVB) DVB SimulCrypt Part 1: for MPEG-2 Systems Layer and Their Effect on
Head-end architecture and synchronization. Receiver Clock Recovery, IEEE
TRANSACTIONS ON MULTIMEDIA, VOL.
2
Frederick Herzberg, 'The Motivation to Work' 1, NO. 3, SEPTEMBER 1999, Page 251
(1959), Work and the Nature of Man (1966),
The Managerial Choice (1982); and Herzberg 9
Watkinson (2001), The MPEG Handbook,
on Motivation (1983). Page 333, Focal Press
3
E. Alexandra Athos et al (2007), Dichotomy 10
Isnardi (1999), MPEG-2 Systems, Sarnoff
and perceptual distortions in absolute pitch Corporation, August 25, 1999
ability, PNAS, September 11, 2007, vol. 104,
no. 37,14799 11
SS. Bindra (2006), Studio Systems, July –
August 2006
4
Audition, by Pierre Buser and Michel Imbert,
English translation by R. H. Kay, MIT Press, 12
Yoshimura (2002), Technologies and Services
Cambridge MA, 1992 on Digital Broadcasting (5), Broadcast
Technology no.11, Summer 2002
5
CD Audio Demonstrations, by A. J. M.
Houtsma, T. D. Rossing, W. M. Wagenaars, Dom Stasi (2007), Broadband Business, CED
13

Philips 1126-061. Magazine, Sept 2007

6
Reeves and Voelker (1993), Effects of Audio- 14
EBU, SMPTE announce joint task force on
Video Asynchrony on Viewer’s Memory, time, synchronization, Broadcast Engineering,
Evaluation of Content and Detection Ability, Sep 15, 2007
Stanford University.
7
ATSC (2003), ATSC Implementation
Subcommittee Finding: Relative Timing of

2008 NCTA Technical Papers - page 89

 TRANSCODING AND STATISTICAL MULTIPLEXING OF MPEG4 (H.264)
BROADCAST VIDEO
John Hartung, Ph.D.
EGT

Santhana Krishnamachari, Ph.D.

EGT

Abstract across multiple service groups and nodes

locally, regionally, or nationally.
The bandwidth demand of HD content is
driving the use of more efficient video The new HD channels will be received in
compression such as MPEG4 (H.264) various encoding formats and bit rates from
encoding for satellite distribution, and satellite distribution and terrestrial
statistical multiplexing for MSO access broadcasters. MPEG2 format is typically
networks. Transcoding and statistical received at rates around 15 Mbps or higher,
multiplexing are usually implemented and satellite distributors are beginning to use
independently; however, in this paper we MPEG4 encoding at 8 Mbps for new
show that this is not the best approach. We programming. Although MPEG4 capable set
show that integrating the transcoding and top boxes are beginning to be deployed, the
statistical multiplexing operations will result large numbers of legacy MPEG2 set top boxes
in improved video quality, reduced require the conversion of all content into
operational complexity and lower cost. The MPEG2 format. This paper describes various
paper is organized into four sections: approaches for transcoding and statistical
Introduction, Transcoding, Statistical multiplexing with quantitative comparisons. A
Multiplexing, and Conclusion. novel approach that combines transcoding
with statistical multiplexing is shown to have
INTRODUCTION the best compression efficiency and quality.

MSOs are planning to increase the number of TRANSCODING

HD programs they offer from around 25 today
to more than 100 over the next couple of Overview
years. This increase is placing a tremendous
strain on the available access bandwidth in the In general, transcoding from MPEG4 to
MSO HFC networks. Three approaches are MPEG2 requires a full decode and re-encode
being taken to solve this bandwidth problem: because many of the tools available in the
analog channels are being converted to more MPEG4 standard are incompatible with
efficient digital transmission, switched digital MPEG2. Examples of these tools include
broadcast is being deployed, and HD content advanced prediction algorithms such as the
is being statistically multiplexed so that 3 or use of multiple reference frames and
more HD channels are carried in a QAM. The intraframe prediction, and filtering in the
statistical multiplexing approach has the prediction loop to reduce blocking artifacts. In
advantage of not requiring the additional some specific instances the MPEG2
network infrastructure and software needed to parameters can be determined or estimated
support switched broadcast or of turning off from the MPEG4 parameters leading to higher
existing analog services. In addition, the cost quality and lower complexity.
of statistical multiplexing can be shared

2008 NCTA Technical Papers - page 101

Independent Decode-Encode encode functions also prevents the original
MPEG4 encoding parameters from being re-
One approach to transcoding from MPEG4 to used as initial MPEG2 encoding parameter
MPEG2 is to fully decode the MPEG4 frames estimates to reduce complexity. Reuse of
and then re-encode with an MPEG2 encoder. these parameters is especially useful in
This can be implemented with an entirely motion estimation where initial estimates can
separate decoder and encoder; however, this be used to reduce the search complexity by
approach does not produce the highest limiting the search range.
possible MPEG2 encoding quality and is
computationally expensive. One reason that Integrated Transcoding
quality is compromised is frame coding types
are not preserved, and therefore high quality An alternative approach is to decode the
reference frames, such as I and P frames, are MPEG4 input, and at the same time pass the
not re-encoded with the same types in MPEG4 encoding parameters to the MPEG2
MPEG2. This results in lower quality encoding stage. This is shown in Figure 1. In
reference frames and propagation of coding addition to preserving frame types and
distortion when they are used for prediction of reducing encoding complexity, the MPEG4
P and B frames. Separation of decode and parameters are also used by the First Pass

MPEG4 Decoder

MPEG4 Decoding
VLC-1
Bitstream Loop

MPEG4 Encoding Parameters Decoded

Video
MPEG4/MPEG2
Parameter
Translator

First Pass Look-ahead

Encoder Buffer

Coding Modes and Complexity Complexity/ Feature

Approximate Motion Vectors Estimate Rate Extraction

Adaptive
Post/Pre-Filter

Rate Control Filtered

MPEG2
Video
Bitstream
MPEG2 Encoder

Figure 1

2008 NCTA Technical Papers - page 102

Encoder and Adaptive Post/Pre-Filter. The need to be preserved in order to retain image
First Pass Encoder determines the MPEG2 details. The original MPEG4 encoding
encoding modes and approximate prediction parameters are used to adaptively remove
residuals from both the decoded video and encoding artifacts by estimating the encoding
MPEG4 encoding parameters, where possible. distortion from prediction parameters and
A relative complexity is determined for each quantization step sizes. The MPEG2 Encoder
frame within a group of pictures (GOP), and allocates a coding rate to each frame based on
this in turn is used by the second pass MPEG2 its’ complexity and the bits available for all
encoder Rate Control to determine an optimal frames within the GOP. This ratio of
target encoding rate for each frame within the complexity /rate indicates the amount of pre-
Look-Ahead Buffer, thereby achieving the filtering needed to minimize artifacts in the
best overall quality. MPEG2 encoded frame. Optimal filtering and
reduced complexity result from the
Integration of the decoding and encoding availability of both MPEG4 and MPEG2
functions also enables advanced Adaptive encoding parameters along with feature
Post/Pre-Filtering of the decoded video. This extraction of the decoded video.
filtering serves two purposes: removal of
encoding artifacts from the decoded MPEG4 Comparison
bit stream, and filtering to reduce MPEG2
encoding artifacts. For both types of filtering Figure 2 shows a plot comparing the quality
feature extraction is used to identify areas of Independent Decode-Encode with
having characteristics that mask distortion due Integrated Transcoding for 1920 x 1080i HD
to the response of the human visual system video. The original MPEG4 video is encoded
(HVS). For example, distortion in textured at 10 Mbps. Peak signal to noise ratio (PSNR)
areas is difficult to perceive so those areas can is used as an objective measure of the
be highly filtered to reduce the required difference between the original and encoded
number of coding bits, while areas with edges video. A higher PSNR represents better

MPEG4 to MPEG2 Transcoding

44

42
PSNR (dB)

40 Integrated Transcoding
Independent Decode-Encode
38
Independent Decode-Encode requires
about 1 Mbps additional rate for equivalent PSNR
at 13 Mbps
36

34
5 7 9 11 13 15
Encoding Rate (Mbps)

Figure 2

2008 NCTA Technical Papers - page 103

quality with about a .5 dB change resulting in programs that can be carried by re-encoding
a perceived quality difference. It can be seen each input at a lower rate that varies as a
from the plot that Independent Decode- function of the channel’s complexity. The
Encode achieves, on average, about .5 dB individual rates are controlled in order to
lower PSNR than Integrated Transcoding. maintain the original video quality at an
This translates into about a 1 Mbps higher rate aggregate rate that allows additional channels
to achieve equivalent performance. The next to be carried within a QAM. For a 38.8 Mbps
section also shows that the average PSNR is QAM channel this corresponds to an average
not the whole story when it comes to encoding rate of below 13 Mbps for three or
statistical multiplexing. Integrated more HD channels.
Transcoding also results in lower frame to
frame PSNR variance and therefore a more The challenge to achieving multiplexing gain
uniform and lower rate to achieve a constant is to combine channels such that their
quality. instantaneous encoding rate remains close to
their average rate. For SD this requirement is
STATISTICAL MULTIPLEXING met because of the large number (>12) of
channels transmitted in a QAM. However,
Overview with only three HD channels transmitted in a
QAM, the channel characteristics must also be
HD channels are delivered to a head end at a considered. One approach is to combine two
constant bit rate using either MPEG4 or low complexity channels, such as progressive
MPEG2 encoding. The bit rate is chosen to movie content, with one high action channel,
produce good quality for the most difficult such as sports. A second factor that limits the
sequences, even though a lower rate would be number of channels that can be multiplexed is
sufficient most of the time. For MPEG2 HD the efficiency of transcoding and rate shaping.
content this rate is 15 Mbps, or higher, These translate directly into the rate required
allowing only two channels to be transmitted to encode individual channels at high quality.
within a 6 MHz QAM channel. Statistical
multiplexing increases the number of Two methods have been used to transcode and

Decode Encode RateShape StatMux

Mpeg-2 HD
Streams @ 19 mbps

3:1 MPEG2 HD
Stream (38 mbps)

Receiver-Transcoders Rateshaping Statistical

Multiplexer

Figure 3

2008 NCTA Technical Papers - page 104

rate shape content for statistical multiplexing where two channels require 14 Mbps, then the
as described below. These are transcoding third channel must be reduced by greater than
from MPEG4 to MPEG2 followed by rate 33% (15 mbps to 10 mbps) to meet the total
shaping, and MPEG4 decoding and MPEG2 rate of 38.8 Mbps. The performance of this
re-encoding with a closed loop statistical approach is fundamentally limited by the fact
multiplexer. A third approach, MPEG4 to that is uses two stages of MPEG processing,
MPEG2 transcoding integrated with closed transcoding followed by rate shaping. In the
loop statistical multiplexing, is shown to comparison section we show that the
produce the best quality. performance falls well below the two other
approaches.
Transcoding and Rate Shaping
Decoding and Closed Loop Encoding
In this architecture the MPEG4 input is first
transcoded to MPEG2 in the Receiver- A second approach begins by decoding the
Transcoders. For HD MPEG4 delivered at 8 input MPEG4 bitstreams using Receiver-
Mbps this first stage of transcoding produces Decoders as shown in Figure 4. This output is
an MPEG2 output of about 15 Mbps. The then re-encoded using MPEG2 encoders
MPEG2 programs are then statistically within a closed loop statistical multiplexer. A
multiplexed in a second stage of rate shaping single stage of re-encoding, decode followed
to form an MPTS meeting the QAM rate as by encode, introduces less distortion than the
shown in Figure 3. The second stage is previous method, however separation of the
usually implemented using a rate shaper that decoder and encoder prevents reuse of the
modifies the original MPEG2 encoding original MPEG4 encoding parameters. This in
parameters without performing a full decode turn leads to a lower PSNR for the target rates
and re-encode. This approach runs into determined by the statistical multiplexer. The
problems when the rate reduction for any Comparison section also shows that both this
channel exceeds about 15%; significant video approach and the previous one, introduce
quality reduction occurs under these greater variance in the frame to frame PSNR.
conditions. A rate reduction of significantly This shows up as artifacts in the statistically
greater than 15% is fairly common, and multiplexed video that degrade the video
occurs whenever two of the channels need a more than would be reflected in the average
bandwidth above 13 Mbps to achieve PSNR comparisons.
adequate quality. If we consider the case

Decode Encode StatMux

HD Uncompressed
Video

3:1 MPEG2 HD
Stream

Receiver-Decoders MPE2 Encoders Multiplexing

Figure 4

2008 NCTA Technical Papers - page 105

Closed Loop Transcoding one channel peaks. Lower target rates can be
chosen for the easier channels, allowing a
The third approach converts the MPEG4 higher rate to be allocated for the complex
output from a receiver directly to MPEG2 channel.
using an Integrated Transcoder, as shown in
Figure 5. This approach achieves the best Figure 7 shows the MPEG2 output frame
performance by transcoding directly to the PSNRs for 1920 x 1080i video transcoded
statistical multiplexing rate in a single stage of using the three approaches. The original
MPEG encoding. The rate feedback also MPEG4 video was encoded at 10 Mbps and
enables the integrated transcoder to adapt the the output is at 13 Mbps. The important
post/pre filters for the target rate, rather than consideration here is the variance of the
an intermediate rate. The perceptual quality is PSNR for each frame. A lower PSNR
also improved by the lower encoding rate indicates that the frame is more complex and
variance achieved in this implementation as would need to be coded at a higher bit rate to
shown in the next section. achieve equal quality. Rate shapers having a

Encode

StatMux
Decode

H.264 HD
Streams @ 9-12 mbps

3:1 MPEG2 HD
Stream (38 mbps)

Receiver Closed Loop Transcoder

Figure 5

high variance produce frequent artifacts

Comparisons because there is a higher probability that
individual target rates exceed the aggregate
Figures 6 and 7 show the single channel available rate. The plot shows that the Closed
PSNR performance for the three statistical Loop Transcoder achieves the lowest
multiplexing approaches described above. The variance, followed by the Decoder with
results are for 1920 x 1080i HD video Closed Loop Encoding and Transcoding and
originally encoded using MPEG4 at 10 Mbps. Rate Shaping implementations.
Integrated Transcoding achieves a 2.75 Mbps
advantage over Transcoding and Rate Shaping
at rates around 13 Mbps as shown in Figure 6. CONCLUSION
It achieves a 1 Mbps advantage over
Decoding and Closed Loop Encoding as Integrating transcoding and statistical
shown in a previous section. These gains multiplexing produces several benefits over
produce higher overall quality, but are competing approaches, the most important
particularly important when the complexity of being optimum compression efficiency and

2008 NCTA Technical Papers - page 106

 Statistical Multiplexing Performance

44

42

Closed Loop Transcoding

PSNR (dB)

40
Decoding and Closed Loop Encoding
Transcoding and Rate Shaping
38
Transcoding and Rate Shaping requires about
2.75 Mbps higher rate for equivalent PSNR
than Closed Loop Transcoding
36

34
5 7 9 11 13 15
Encoding Rate (Mbps)

Figure 6

Frame PSNR at 13 Mbps

48
47
46
45 Closed Loop Transcoding
PSNR (dB)

44
Decoding and Closed Loop
43
Encoding
42
Transcoding and Rate
41 Shaping
40
39
38
75
79
83
87
91
95
99
103
107
111
115
119
123
127

Frame

Figure 7

video quality. This architecture achieves these on transcoding from MPEG4 to MPEG2
benefits through the reuse of the original standards, similar gains are achieved when
encoding parameters, both for conversion constant bit rate MPEG2 content is
between input and output formats, and for statistically multiplexed into an MPEG2
encoding at the statistical multiplexing rate. output.
Integrating the two functions also enables a
single stage of encoding thereby avoiding the
distortion due to two generations of
processing. Although this paper has focused

2008 NCTA Technical Papers - page 107

 UNICAST VIDEO WITHOUT BREAKING THE BANK: ECONOMICS, STRATEGIES, AND
ARCHITECTURE

S.V. Vasudevan, Xiaomei Liu, and Robert Kidd

Cisco

Abstract
Based on the results of this analysis, a switched
Driven by competition and consumer demands, architecture will be presented for cable
linear video delivery is following a trajectory operators to smoothly migrate their networks to
from broadcast to multicast and ultimately to support unicast delivery mechanisms for linear
unicast. Traditionally, video delivery has been video services. The proposed architecture
broadcast only. Today, cable operators are accomplishes the strategies for cost-effective
deploying switched digital video (SDV), which unicast delivery and supports:
uses multicast technology to improve the  A flexible combination of multicast and
bandwidth efficiency of HFC networks. The next unicast delivery mechanisms
logical progression to unicast delivery is on the  Traditional ad insertion based on
horizon and is positioned to become tomorrow’s geographic ad zones or a new
video delivery mechanism. generation of targeted ad insertion based
on demographic profiles
Unicast delivery of linear content is an  Fast channel change and personalization
incremental extension of the multicasting of unicast content
approach used in SDV implementations. The
incremental investment in bandwidth resources
to support unicast delivery can be offset by the INTRODUCTION
contribution of preferentially valued advertising
opportunities, reduced subscriber churn, and the Switched Digital Video (SDV) is now a
ability to attract new subscribers through mainstream technology that is delivering on its
differentiated service offerings. promise to dramatically improve upon the
bandwidth efficiency of the traditional linear
This paper analyzes the unicast value broadcast model. Aggressive SDV
proposition, including cost, revenue potential oversubscription ratios (the ratio between the
and return on investment. SDV field trial number of SDV programs offered and the
viewership statistics will be reviewed, and used number of stream resources provisioned) have
to shed light on the cost sensitivities related to been observed and will continue to increase as
channel popularity and HD penetration. Best more niche and HD content is added to cable
case and worst case scenarios for HFC MSO service tiers. Switching technology has
bandwidth consumption will be explored and proven to be a powerful addition the cable
analyzed, along with the cost structures operator’s bandwidth management capability.
associated with each of them. Cost mitigation
and revenue improvement strategies will be Yet there remains unlocked potential within
explored, demonstrating how cable operators the SDV infrastructure. Current generation
can optimally combine unicast and multicast systems support open standards, allowing the
approaches in order to maximize overall return insertion of new technologies and applications.
on investment. The session and resource managers (SRMs) that

2008 NCTA Technical Papers - page 223

respond to subscriber channel-change requests Switched unicast refers to a delivery mechanism
provide a level of intelligence and network in which, regardless of the IP transport and
awareness that was previously unavailable. SDV routing mechanisms deployed, the stream on the
systems maintain a real-time accounting of HFC side of the plant is destined for a single
programs being viewed as well as the number of tuner within a single set top; i.e. set tops within a
subscribers viewing those programs. This service group no longer share MPEG streams; if
awareness of program usage comes concomitant N viewers within a service group request MTV, N
with knowledge of bandwidth allocation. This instances of MTV are switched into that service
knowledge is powerful, for even with existing group. Typically the transport mechanisms on
switched multicast, there remains significant the IP network will also be via IP unicast, but
room for improvement in bandwidth utilization. hybrid solutions can be envisioned in which, for
Furthermore, Switched Unicast, an advanced example, ad servers at the edge join multicast IP
form of SDV, is drawing increasing interest. content, insert a targeted ad, and then send the
This emerging SDV architecture offers exciting new stream via IP unicast to the edge QAM.
opportunities to introduce new revenue-
generating services, but it also has the potential The drivers for deploying switched unicast are
to overwhelm available access bandwidth. Only compelling. Switched unicast offers the
with the knowledge of user demand provided by opportunity to personalize video. Since each
advanced SRMs is this latest architectural tuner now receives an individual video stream,
challenge tractable. media processing techniques can be used to
modify the stream to suit the preferences of an
It is first helpful to clarify the requisite individual subscriber. These modifications may
terminology. Switched multicast refers to a include graphics overlays on the screen that are
video delivery architecture where an MPEG tailored to the subscriber’s preferences. For
program, typically in the form of a single example, a ticker that displays preferred stock
program transport stream (SPTS), is IP- quotes, sports scores, localized weather
encapsulated and transported on a distribution information, etc.
network via IP multicast. A system session and
resource manager (SRM), acting upon channel But perhaps the most compelling driver for
change requests from subscribers, may then switched unicast is in the ability to personalize
instruct an IP-attached edge QAM to join the advertising. The North American cable industry
multicast. The edge QAM rebuilds a multi- has a long history in spot advertising as a
program transport stream (MPTS), containing revenue source, and many systems perform a
content requested by multiple viewers, and limited level of localization by dividing a cable
modulates the content onto the HFC network. system into zones and offering spot insertion on
The key differentiator of this approach is that a zone by zone basis. However, the ability to
multiple set top boxes within a service group can transcend beyond zones and offer advertising on
share a stream that is active within that service a personalized basis offers MSOs the opportunity
group; if N viewers within a service group watch to charge a premium for these slots – indeed the
the program MTV, only one instance of MTV is success of Google in its ability to personalize
switched into that service group. Thus, the advertising in the online space has made all
content is delivered over the IP network using IP participants in the advertising delivery chain sit
multicast and over the HFC plant with a stream up and take notice. Furthermore, with
sharing, RF multicast mechanism resembling that personalized advertising, ads need not be
on the IP network. restricted to 30-second spots – they can
additionally take the form of graphics logo

2008 NCTA Technical Papers - page 224

overlays, with additional possibilities enabled To better understand this relationship, viewership
from the incremental ability to launch interactive information was extracted and analyzed as part
ads. of a SDV trial with a major North American
MSO. By aggregating and post-processing
This raises an interesting question which is the channel change information from server log files,
subject of this paper – switched unicast has the it is analytically straightforward to determine not
ability to increase MSO revenue through the only the relative difference in resource
incremental benefits of personalized advertising, requirements between multicast and unicast, but
yet to reap this benefit requires an investment in it is also possible to analytically modify the
delivery infrastructure to enable these “virtual” size of the service group to better
personalized capabilities. Is the investment worth appreciate the sensitivity between service group
it? size and resource utilization for both the
multicast and unicast scenarios.
This turns out to be a challenging question to
answer, as the answer depends on a number of The viewership study was conducted over
variables, including subscriber viewing patterns, several weeks and included channel-change data
service group sizing, equipment costs, and the for 247 broadcast video services comprising 228
anticipated premium to be expected from placing standard definition programs and 19 high
a personalized ad as opposed to a zoned ad. To definition programs delivered to 680 tuners in a
better understand this topic, we propose to study service group. The study included four major
the subject from 3 perspectives: analyzing data steps: (1) computation of the viewership long-
from a real SDV deployment to understand tail; (2) segmentation of the long-tail into
multicast and unicast resource sensitivity based popularity quintiles; (3) segmentation of the
on program popularity, service group size, and settops into “virtual service groups”, and (4)
other factors; developing a financial model of computation of unicast (total) and multicast
switched unicast, which allows ROI to be (unique) streams required to deliver cumulative
analyzed based on a number of factors; and quintiles of programming to a range of service
exploring alternate models of SDV that enable group sizes.
unicasting on an opportunistic basis.
The concept of a “virtual service group” warrants
some discussion. One important result desired
SUBSCRIBER VIEWING PATTERNS AND from the analysis was the expected variation in
SWITCHED UNICAST video stream resource requirements with service
group size. In most production systems, a target
service group size is established and the inside
It is difficult to make purely analytical and outside plants designed to that target size. If
predictions regarding SDV efficiency, since an analysis is performed only with the existing
efficiencies ultimately depend on subscriber service group structure, bandwidth requirements
viewing patterns, which in turn is driven by the can be predicted for only a very narrow range of
behavior of human beings. However, by service group sizes. Therefore, instead of using
analyzing the pattern of channel change the existing system’s service group structure, the
messages from SDV, it is not difficult to infer nodes were regrouped into sets of virtual service
what the expected system behavior would be if groups containing tuner counts that span the
the system were unicast instead of multicast. range of interest.

2008 NCTA Technical Papers - page 225

Figure 1 provides viewership results for one the viewership curve is segmented into five
week of data and illustrates the classic long-tail sections, each of which includes 20% of the
viewership phenomenon. The graph is generated offered programs. These quintiles are used to
by summing the number of seconds viewed for evaluate stream requirements as a function of
each broadcast video program and then ranking program popularity.
the channels in order of decreasing total
viewership [1]. Once the programs are ranked,

1.60E+08

1.40E+08

1.20E+08
Cumulative Viewership Minutes

1.00E+08

8.00E+07

6.00E+07

4.00E+07

2.00E+07

0.00E+00
1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171 181 191 201 211 221 231 241
Channel Rank
(1 = most popular program)

Top 20% 2nd 20% 3rd 20% 4th 20% Bottom 20%

Figure 1. Program Viewership Ranking

Figure 2 illustrates switched multicast stream particular grouping of content: the bottom curve
usage and displays the number of peak unique illustrates the peak unique streams required for
simultaneous streams required as a function of the least viewed 20% of content; the next-to-
service group size and popularity of content. The bottom curve illustrates peak unique streams
horizontal axis displays the number of tuners per required for the least viewed 40% of content; and
service group, and the vertical axis displays the so on up to the top curve that illustrates the peak
number of peak unique simultaneous streams. unique streams required for the entire broadcast
Five curves are included on the plot, each of video lineup.
which illustrates peak stream requirements for a

2008 NCTA Technical Papers - page 226

 180
170
160
150
140
130
120
110
Multicast Peak

100
90
80
70
60
50
40
30
20
10
0
300 400 500 600 700 800 900 1000 1100
Service Group Size (Tuners)

Bottom 20% (50) Bottom 40% (99) Bottom 60% (149) Bottom 80% (198) All (247)

Bottom 20% Fit Bottom 40% Fit Bottom 60% Fit Bottom 80% Fit All Fit

Figure 2. Switched Multicast Peak Stream Usage

to fall below the upper bound of this prediction

Each curve is generated from a regression interval 97.5% of the time.
analysis of the raw data points, also included on
the plots, that result from processing each virtual Figure 3 is the unicast equivalent of Figure 2. In
service group for a specific grouping of content. this case, the vertical axis displays the number of
Two-sided 95% prediction intervals (not shown peak total simultaneous streams as opposed to
in the figure) were also computed. Future the number of peak unique simulcast streams;
individual peak stream count results are expected otherwise, the data analysis is equivalent to that
described above for the multicast case.

2008 NCTA Technical Papers - page 227

 950
900
850
800
750
700
650
600
Unicast Peak

550
500
450
400
350
300
250
200
150
100
50
0
300 400 500 600 700 800 900 1000 1100
Service Group Size (Tuners)

Bottom 20% (50) Bottom 40% (99) Bottom 60% (149) Bottom 80% (198) All (247)

Bottom 20% Fit Bottom 40% Fit Bottom 60% Fit Bottom 80% Fit All Fit

Figure 3. Switched Unicast Peak Stream Usage

Figure 4 summarizes the stream count dynamics

between unicast, multicast, and basic broadcast
for the entire 247-program lineup.
700

600

500
Average Peak Streams

Streams Required
400 For Unicast Delivery

300

Streams Required
For Linear Delivery

200

Streams Required
100 For Switched
Multicast Delivery

0
0 100 200 300 400 500 600 700
Tuners

Figure 4. Stream Requirement Overview

2008 NCTA Technical Papers - page 228

As expected, the different delivery mechanisms than the number required for simple linear
require dramatically different HFC bandwidth broadcast.
allocations. The number of streams required to
deliver the lineup via traditional linear broadcast One key caveat should be raised at this point
is of course constant and therefore independent regarding unicast: viewership statistics are
of service group size. The number of streams nonstationary from a statistical standpoint, that
required to deliver the same lineup via switched is, they change with time, and this fact has
multicast is significantly less since settops may significant practical implications. As an
share streams within an HFC service group. The example, consider the case of a weather or news
number of streams required to deliver the lineup channel, a service with average viewership
using switched unicast is linearly proportional to sufficiently far down the long-tail to support its
service group size. This latter viewership curve inclusion into a switched tier. As time goes by, a
is similar to that for VOD except that the slope of hurricane or other major news event will
the unicast demand curve for broadcast content is inevitably emerge, and the popularity of this
much steeper than that experienced with VOD. previously moderately-viewed programming
skyrockets. If the channel is offered on a
For delivery of the most popular content to larger multicast tier, viewers that flock to the channel
service groups, unicast requires significantly share a single stream; however, if the channel is
more bandwidth than not only multicast SDV but offered on a unicast tier, viewers receive their
also simple linear broadcast. For example, in own stream, and the required edge bandwidth
order to offer the entire broadcast video lineup to mushrooms beyond that which may have been
today’s typical 500 tuner service group, the predicted based upon prior viewership studies. A
unicast model of Figure 4 requires approximately unicast tier is much more sensitive to the choice
417 peak streams, the multicast model requires of selected content than a multicast tier and is
approximately 120 peak streams, and the therefore less stable from a bandwidth planning
broadcast model requires 247 streams. Making perspective. In order to mitigate this risk, a key
the simplifying assumption of a 50/50 SD/HD potential feature of a unicast system would be the
split for the unicast streams, approximately 104 ability of the system to automatically promote
QAM carriers would be required to carry the and demote between the unicast and multicast
unicast content. tiers.

Clearly the capacity to unicast the entire lineup is Finally, in the above discussion stream counts
not available on a typical service group in are used as a proxy for bandwidth requirements
today’s hybrid digital/analog HFC systems; and the two are often used interchangeably.
however, there are a number of revenue- However, if different streams have different
enhancing opportunities that can be supported bandwidths, aggregate stream counts and their
today by the surgical insertion of unicast associated total bandwidths may not be directly
technologies. These surgical deployments can be proportional. For example, if viewership of HD
more fully developed as the industry continues services is consistently higher than that for SD
the inexorable push toward smaller service services, the most viewed 20% of content may
groups and increased digital delivery. Service require substantially more bandwidth than the
group sizes are trending towards a future where stream estimates alone would predict. In the
250 tuner service groups will be the norm, and system under consideration, 19 of the 247
only 209 streams will be required to unicast the broadcast video programs offered are HD. The
entire the 247 channel broadcast lineup, fewer percentages of these HD services in the 20%,
40%, 60%, 80%, and 100% quintiles of content

2008 NCTA Technical Papers - page 229

were 0%, 21%, 42%, 11%, and 26% SWITCHED UNICAST ARCHITECTURE
respectively. Thus the spread of HD content is AND ROI MODELING
slightly skewed towards the most viewed
groupings; however, the effect of this skew on It was previously mentioned that perhaps the
bandwidth requirements is muted given the incremental revenue generated from switched
relatively small number of HD programs offered unicast could fund the investment in the
at the time of the study. necessary delivery infrastructure. In reality, this
is a complicated problem with numerous factors
contributing to the analysis. Before going into
details of ROI analysis, we will start with some
basic assumptions of a switched unicast
architecture.

Content Campaign
Information Manager
Service
SDV
Session
Subscriber Ad Manager
Information Decision
Service System

Placement
Information Ad OnDemand Encryption Edge
Service Management Resource Resource Resource
System Manager Manager Manager
STB

SDV Streaming/ad Bulk SDV

QAM Client
Grooming Server Encrypt

Figure 5. Switched Unicast Architecture

A switched unicast system architecture builds component that constantly ingests live linear
upon the existing, widely deployed switched content and streams it out as requested. The
multicast architecture and represents an streaming/ad server also detects ad placement
evolutionary path. Existing components may be opportunities and splices ads. In the control
retained and augmented, minimizing the plane, the SDV session manager manages SDV
incremental investment. In the example switched channel changes and orchestrates SDV resource
unicast architecture shown in Figure 5, switched allocations through various resource managers.
video content flows from a SDV groomer The targeted advertisement control plane
(performing VBR-CBR rate clamping) to STBs components are shown and are defined by the
passing through streaming/ad servers, bulk SCTE-130 specification. An ad decision system
encryptors and QAMs. The key difference from makes an intelligent decision on which ads to
traditional switched digital video is the insert for an ad placement opportunity based on
introduction of a streaming/ad server in the data information from the functional components
path. The streaming/ad server is defined as a labeled content information service, subscriber

2008 NCTA Technical Papers - page 230

information service and placement opportunity The following parameters were considered on the
information service. The streaming/ad server “revenue side”, shown in Table 2.
detects placement opportunities, notifies the ad
decision system about the placement Revenue Modeling Parameters Default Value
opportunities via the ad management system and Ad revenue per subscriber $300
inserts ads based on decisions made by the ad MSO per sub ad revenue $60
Possible ad revenue share with 30%
decision system. programmers and networks
Targeted ad percentage 30%
An analytical model representing switched CPM improvement of targeted 2
unicast sizing and pricing was developed to ads
evaluate optimal dimensioning parameters for Targeted advertising operating 80%
margin
future deployments, and quantitatively examine
targeted advertising development opportunities
that could provide an attractive return for the Table 2. Revenue Modeling Parameters and
infrastructure investment. Default Values

The analytical model for switched unicast The strategy for the analysis was as follows:
considered the following parameters on the using the number of channels, the extracted
“expense side”, shown in Table 1. The graphs program popularity, and the bandwidth of the
that appear later are computed from the default programming, the total aggregate bandwidth of
values indicated in the tables. the expected unicast streams was calculated.
Based on this information, the optimal service
Cost Modeling Parameters Default Value group size was calculated. If this optimal value
Service group size 1000 HHP was less than the existing service group size, the
Subscriber penetration 60% model would factor in the price of node splits to
Digital penetration 60% compute the infrastructure costs to achieve the
Tuners per household 1.8 proper service group size. Once this value was
Total channel offered 249
known, industry-current figures for QAM,
HD channel percentage 20%
Channel popularity Extracted from trial
streaming server, and transport costs (normalized
log data to a $/Mbps factor) were used to calculate
SD channel bandwidth 3.75 Mbps expected investment costs.
HD channel bandwidth 15 Mbps
Spectrum available for SDV 30 RF channels To migrate to switched unicast, operators incur
QAM channel bandwidth 38.8 Mbps both a data plane cost and control plane
AVC STB penetration 20%
Node split cost $15,000
investment cost. The data plane costs largely
Transport cost $9/Mbps consist of capital equipment (QAMs, streaming
QAM cost $13/Mbps servers, etc.) The control plane costs largely
Streaming server cost $15/Mbps consist of software enhancements to existing
SDV server platforms to enable unicast
Table 1. Cost Modeling Parameters and Default signaling. Taking advantage of the fact that
Values control plane components such as the SDV
session managers and resource managers are
already essential ingredients of switched
multicast video services, the additional cost for
providing switched unicast control plane
infrastructure is not significant. For our analysis,

2008 NCTA Technical Papers - page 231

the cost of SCTE-130 targeted advertisement unicast service. The estimated cost of node
control plane components was factored in with splitting is $15,000 per split.
an advertising operating margin of 80%. Major
data plane expenditures came from spectrum, Although advanced video coding saves 50%
QAMs, streaming servers and other transport bandwidth when compared with MPEG-2 video,
costs. it is not currently widely used in the linear video
service because of the broadcast nature of the
Comprehensive analysis has been done service today. A lowest common denominator is
evaluating the cost and benefits of various chosen in terms of STB capabilities. In other
technologies that can squeeze more bandwidth words, as long as legacy MPEG-2-only STBs are
out of the cable plant [2]. Among 1 GHz node in the field, the video must be offered in the
upgrades, all digital conversions, node splits, MPEG-2 format. This situation is changed with
advanced video coding, 1024QAM etc, node the introduction of switched unicast. With
splits stand out as the attractive approach with an switched unicast, AVC coding can be used when
appealing cost to benefit ratio. In our cost delivering unicast video to newer AVC-capable
analysis, we assume the plant starts with a 1000 STBs, while the MPEG-2-version of the program
HHP service group size, which is the equivalent can be used when delivering to legacy MPEG-2-
of 500 tuners per service group with 60% only STBs. In the cost analysis, we assumed 20%
subscriber penetration and 60% digital of STBs as AVC capable.
penetration. Assuming a maximum 30 RF
channels are available for digital linear video The result of per subscriber cost analysis is
service in a service group, node splitting to shown in Figure 6.
smaller service groups will be used until there is
enough spectrum for providing the switched
Switched Unicast Cost per Subscriber w. 20% HD
$350

$300

$250

$200
US ($)

$150

$100

$50

$0
0 20 40 60 80 100 120

Cumulative Polularity Perce ntage (%) SDU cost

Figure 6. Switched Unicast costs per popularity quintile

2008 NCTA Technical Papers - page 232

As one would expect, switched unicast per Currently, in a typical cable network [4], per
subscriber costs increase as more popular content subscriber advertising revenue is $60. Since the
is offered as unicast. The biggest jump comes MSO’s share is just 20% of per subscriber ad
when the top 20% most popular channels are revenue, the total per subscriber ad revenue is
offered as unicast. calculated as $300. The ad revenue modeling
assumes that 30% of placement opportunities in
Targeted advertising is often touted as the next switched unicast are targeted and assumes that
revenue growth engine for cable MSOs. MSOs have a 30% placement opportunity split
Switched unicast is an enabling platform for with broadcasters and cable programmers for
targeted advertising. In 2006, worldwide cable non-local ads. Then the potential MSO per
TV advertisement spending totaled $24 billion, subscriber ad revenue increase can be derived.
with nearly $5 billion contributing to MSO However, the uneven distribution of unicast
revenues. With only a 20% share of the total viewers with regard to the channel popularity
spend, targeted advertising could be an effective complicates the calculation. The additional ad
vehicle to improve this number by providing a revenue of a node is calculated next as the
higher-value product. [3] potential per subscriber ad revenue times the
unicast viewers in the node. Lastly, this ad
When estimating the ad revenue potential of revenue of the node is averaged to compute the
switched unicast, we apply the advertisement switched unicast ad revenue per subscriber. The
CPM (cost per thousand impressions) projected ad revenue is shown in Figure 7.
improvement of 2x for targeted advertising.

Switched Unicast Ad Revenue per Subscriber

$60

$50

$40
US ($)

$30

$20

$10

$0
0 20 40 60 80 100 120
Cumulative Polularity Perce ntage (%) Ad Revenue

Figure 7. Switched Unicast potential revenue per popularity quintile

2008 NCTA Technical Papers - page 233

 SDU Return of Investment w. 20% HD
350%

300%

250%

200%
ROI (%)

150%

100%

50%

0%
0 20 40 60 80 100 120
ROI
Cumulative Polularity Percentage (%)

Figure 8. Switched Unicast ROI per popularity quintile

The return on investment results are illustrated in percentage of HD channels in the offering
Figure 8. The diagram reveals that the ROI increases, the cost for switched unicast increases
decreases as more popular channels are offered dramatically. In fact, if more than 30% of the 247
as unicast. In spite of the decrease, even when all channels are HD channels, bandwidth and
channels are offered as switched unicast spectrum requirement will push the service group
assuming 20% HD channels, with the submitted size to below 125 tuners if the majority of the
parameters the ROI can be demonstrated to be as STBs are legacy MPEG-2 only STBs. This can
good as 16%. be intuitively understood by the fact that an HD
program can consume 4 times the bandwidth of
One interesting fact to notice is the sensitivity to an SD program, but may not necessarily
HD channel percentage in the channel lineup. command a 4x premium for spot ad insertion.
Figure 9 clearly demonstrates that as the

SDU Cost per Subscriber

$350.00

$300.00

$250.00

$200.00
US ($)

$150.00

$100.00

$50.00

$0.00
20 40 60 80 100
Cumulative Popularity Percentage (%)

5% HD 10% HD 15% HD 20% HD 25% HD

Figure 9. Switched Unicast costs with varying percentage of HD programming

2008 NCTA Technical Papers - page 234

What is the best strategy for MSOs? may find creative ways of getting more revenue
from HD ad insertion.
The previous analysis provides insight that
supports a recommended switched unicast Third, even though the spectrum capacity can be
strategy. First, although there is little doubt that increased by node splitting to get an HFC plant
the ad revenue potential of switched unicast is unicast ready, a better way might be to reclaim
well worth the investment, switched unicast can analog channels first to free up some spectrum.
be offered gradually with the least popular In the cost analysis, we assume that there are 30
channels offered in unicast first. This approach is RF channels available for linear digital video
even more appealing if we consider that the least services since most of the RF spectrum is used
popular channels also have more local ad by the analog tier. With analog reclamation,
inventory accessible to MSOs. In a switched more spectrum will be available for linear digital
video architecture, a switched digital video video services and fewer node splits would be
session manager could implement a policy needed.
control to prioritize the less popular channels in
unicast first. HYBRID UNICAST-MULTICAST DELIVERY
SYSTEMS FOR MAXIMUM BANDWIDTH
Second, the number of HD channels offered in OPTIMIZATION
the switched unicast must be carefully evaluated
to maximize the ROI. On one hand, it is tempting
to offer more HD channels in switched unicast. It has been observed that an appropriate premium
On the other hand, HD channels consume much incentive for a targeted advertisement service can
more bandwidth than SD channels. For MPEG-2 justify the infrastructure investment. What
video, the HD bandwidth is roughly four times guidance does such a statement provide to an
the SD bandwidth. Unfortunately, in today’s MSO? Should an operator wait until a certain
advertisement arrangement, there is no revenue CPM threshold is crossed before deploying a
premium to insert ads in HD channels as opposed targeted advertising solution?
to SD channels. The proliferation of AVC STBs
does alleviate the problem by reducing the HD It is possible to consider another variation of
bandwidth need by half. Additionally, MSOs switched unicast. This variation can be easily
understood by examining Figure 10.

2008 NCTA Technical Papers - page 235

 Linear and Switched Bandwidth Allocations
For Least-Popular Programs
18

17

16 Linear Bandwidth Required

For 137 SDs, 12 HDs
15

14

13

12

11 Bandwidth Recovered via SDV

QAM Channels

10

7
Switched Bandwidth Allocated
6
For 137 SDs, 12 HDs
5 Switched Bandwidth Potentially Wasted
4

2
Switched Bandwidth Used
1

0
Time

Figure 10. Switched Multicast Bandwidth Utilization

Indeed, Figure 10 shows that SDV saves When bandwidth utilization is low, for example
bandwidth over its linear video equivalent, but in non-peak hours, the system can respond to
what of the spectrum capacity that is allocated channel requests by creating a personalized
but not used? This represents a lost opportunity unicast channel. If bandwidth resource utilization
for better bandwidth utilization. To be more reaches a (configurable) threshold, the SDV
demonstrative, one could argue that this too is server can respond to subsequent channel
wasted bandwidth. The counter-perspective is requests with tuning information for a non-
that the unused bandwidth provides a “buffer personalized multicast or “shared” stream. Since
zone” to protect against possible resource the multicast stream is shared by all of the
overflows. This is true, but the design of SDV subsequent users, this provides an effective
systems demands that enough QAM resource be “safety valve” to cap the stream usage when
provisioned to handle the peak SDV necessary, while offering maximum
consumption periods (which is typically during opportunities for personalization. As the
the evening/prime time hours). During non-peak subscriber population churns through channel
hours, the bandwidth utilization can be quite low. changes, the aggregate number of active unicast
channel streams will reduce by attrition, and the
A hybrid multicast-unicast switched digital system can vary the unicast/multicast stream mix
delivery system can provide the optimum blend through a natural feedback process to manage the
of content personalization and bandwidth bandwidth utilization and the personalization
utilization in every situation. In such a system, a opportunities to an optimal level.
SDV server monitors resource utilization, and
uses this indication to determine whether to What would a personalized vs. non-personalized
respond to a channel request with tuning channel look like? An example (one possible
parameters for a unicast or multicast stream. embodiment) is shown in Figure 11 [5].

2008 NCTA Technical Papers - page 236

 Figure 11. Screenshot of linear channel vs. personalized channel

Multicast/unicast stream mix is an example of levels of personalization with a modest

policy-driven resource management. In this incremental investment to current SDV systems.
example, allocation of edge resources is driven
by a policy that seeks to optimize the insertion of REFERENCES
unicast streams while observing rules on
maximum bandwidth limits. But the rules do not [1] “A Comparison of Edge Bandwidth
need to be this simple. Future evolutions of this Requirements For Unicast versus Multicast
stream selection function could use more deeply Delivery of Switched Digital Video Services”,
sophisticated decision-making algorithms, Robert Kidd and Michael Shannon.
weighing factors such as subscriber profile,
program type, advertising value, time of day, and [2] ”An Evaluation of Alternative Technologies
other factors to hyperoptimize the allocation of for Increasing Network Information Capacity,
shared resources. Ron Shani and David Large”, NCTA 2005

CONCLUSION [3] ”Irreconcilable Differences or a Match Made

in Heaven? The Future of Advanced TV
Switched Unicast is an extension of Switched Advertising”, Ben Hollin and John Morrow,
Digital Video that enables content SCTE ET 2008
personalization and targeted ad insertion. While
the main goal of most SDV deployments is [4] Comcast 2007 Financial Results
centered around cost-effective programming
expansion, switched unicast offers a revenue [5] Screenshot courtesy ICTV and Turner. Use of
generation opportunity through targeted this screenshot does not necessarily imply
advertising and interactive services. Analysis of endorsement by ICTV or Turner of Cisco and/or
actual subscriber channel-change log data can the concepts presented in this paper.
provide valuable insight to the viewing patterns
that might be expected in a switched unicast
environment. Hybrid multicast/unicast
implementations offer an opportunity to
incrementally explore the value proposition of
interactive programming by enabling fractional

2008 NCTA Technical Papers - page 237

 VARIABLE BIT RATE VIDEO SERVICES IN DOCSIS 3.0 NETWORKS
Xiaomei Liu, Cisco Systems
Alon Bernstein, Cisco Systems

Abstract millions of video assets streamed or downloaded

to PCs.
DOCSIS 3.0 and Modular CMTS promise to
provide ten times the bandwidth at one tenth of Cable operators have many tools to address
the cost, compared to existing CMTS the overwhelming bandwidth crunch problem.
technology. With these forward-looking trends, Some of these approaches in the MSO toolkit
it is becoming increasingly viable to consider a include: analog channel reclamation, switched
channel-bonded DOCSIS network as a fully- digital video, node splitting, plant upgrades to
converged network to transport video, voice and 1GHz, and MPEG-4 part 10 video coding.
data. A bonded, converged and asynchronous Although cable operators can drill for additional
data pipe, married with variable bit rate (VBR) bandwidth in HFC networks with major capital
video coding, can deliver the full potential of expenditure, there is work that can be done to
IPTV over cable. eliminate any bandwidth inefficiency in HFC
networks first.
This paper examines the technical and
economic implications of VBR over DOCSIS. It Starting with video sources, it is common
proposes an IP-level VBR network statmux to knowledge that variable bit rate (VBR)
deliver VBR video over channel-bonded encoding of video is significantly more efficient
DOCSIS and quantifies the efficiency of the than constant bit rate (CBR) encoding. In
network statmux with the results of lab tests. It MPEG video encoding, while the CBR video
provides insights into various architecture keeps the bitrate constant, the VBR video
issues related to VBR delivery. Finally, it attempts to keep the video quality constant. The
explores mechanisms to improve robustness and nature of MPEG video encoding allows
enhance the subscriber viewing experience. encoders to use fewer bits for simple scenes and
more bits for complicated and motion rich
INTRODUCTION scenes. With comparable video quality, VBR
can yield 40 percent or more bandwidth savings
Cable networks are experiencing an over CBR [1]. As a result of its coding
explosion in demand for increased bandwidth. A efficiency, VBR video is widely used in DVD
significant amount of bandwidth pressure comes and in broadcast video applications such as
from High Definition Television (HDTV) digital satellite and cable.
service expansion, which MSOs have used as a
strategic move to compete with satellite and The introduction of broadcast-oriented
telco video service providers. Today, the 100+ MPEG statmuxes paved the way for delivery of
HD channel service is on the horizon as more VBR streams in broadcast video. Since most
HD content is offered. Meanwhile, content transmission channels have fixed bandwidth,
personalization and targeted advertisement are MPEG statmuxes combine a number of VBR
gradually tranforming the video delivery vehicle streams into a single aggregated constant bitrate
from broadcast to unicast. Yet over-the-top channel. The statistical distribution of bitrate
video services and user-generated video content peaks and valleys allows the combined streams
simultaneously drive bandwidth demand with to use less bandwidth than what is needed if
each VBR stream is sent individually. At any

2008 NCTA Technical Papers - page 1

given point in time, if the bandwidth of a VBR operation introduces delays, typically on the
bundle exceeds the capacity of an MPEG order of 1 second. This long latency is more
transmission channel, the MPEG statmux noticeable and undesirable for on-demand and
applies requantization at the MPEG level to interactive video services.
reduce the instantaneous bitrate of video streams
to fit the transmission pipe. This action does 3) It is also no surprise that traditional
come at the expense of a non-zero impact to MPEG statmuxes have difficulties dealing with
video quality. encrypted content considering that the rate
reduction techniques involved need to analyze
Ironically, in advanced video services such and re-encode the stream content. For example,
as Switched Digital Video (SDV) and Video on pre-encrypted VOD content makes the
Demand (VOD) where the last mile bandwidth elementary MPEG stream inaccessible for
efficiency is needed the most, CBR instead of transrating.
VBR video is deployed universally today. This
is because SDV and VOD present a challenging The business and technical issues pointed
case for traditional broadcast oriented MPEG out above have forced cable operators to give up
statmuxes: VBR efficiencies and opt instead for CBR video
delivery in switched and on demand video
1) MPEG statmuxes are computationally services.
intensive and costly. MPEG statmuxes achieve
rate reduction by transrating selected MPEG These assumptions change as video over
frames. Transrating is an expensive operation as DOCSIS becomes a reality. Not only does
macroblocks in pictures are re-quantized and re- wideband DOCSIS provide an IP transport to
encoded at the MPEG level. Although the cost MPEG video, it also brings along a promising
of MPEG statmuxes is not a concern in a new way of VBR statistical multiplexing.
broadcast network as the per stream statmux
cost is shared among all the subscribers in the DOCSIS 3.0 AND NETWORK STATMUXES
network, quite the contrary is true in the SDV
and VOD world. In an increasingly unicast- DOCSIS 3.0 is perhaps the most anticipated
based video delivery network, the per stream technology of the year in cable industry.
statmux cost now becomes a per-subscriber DOCSIS 3.0 takes the DOCSIS beyond just an
cost, which makes current MPEG statmuxes IP transport for data and voice services. IP
economically impractical to deploy at the video over DOCSIS is rapidly gaining traction
network edge. with DOCSIS 3.0. In fact, some MSOs already
have started market trials and deployments of
It is already a complicated operation to IPTV over DOCSIS are in the planning stages.
statistically multiplex MPEG2 encoded VBR
streams; it is an even more daunting task to Partly because of DOCSIS 3.0 channel
perform statmux on MPEG4 encoded VBR bonding and DOCSIS 3.0 enhanced multicast,
streams because of the incrementally intensive IP video over DOCSIS becomes a feasible
video computations involved. technical possibility. Channel bonding, the most
important feature of DOCSIS 3.0, makes the
2) MPEG statmuxes apply extensive stream channel bandwidth a magnitude higher than
analysis in order to mitigate the video quality before by allowing CMTSs to bond multiple
degradation caused by the transrating operation. downstream and/or multiple upstream RF
The stream analysis as well as the transcoding carriers in order to deliver higher bandwidth to

2008 NCTA Technical Papers - page 2

the home. Today, eight channel bonded cable IP/DOCSIS 3.0 transport supports wideband
modems are already available in the market, transmission with bandwidth upper capacity
enabling downstream bandwidth rates of around limited only by cable modem technologies.
300Mbps. This increased bandwidth capacity is Downstream bandwidth speeds of 300Mbps are
essential to bandwidth-hungry applications such enabled by today’s eight-channel cable modems.
as standard definition (SD) and high definition It is only a matter of time before much higher
(HD) video. bandwidths are available as Moore’s Law keeps
bringing down the cable modem cost. MPEG
DOCSIS 3.0 enhanced multicast adds transport, on the other hand, which does not
source-specific multicast (SSM) and Internet support channel bonding, has a bandwidth
group management protocol (IGMPv3) support. limitation of a single QAM channel. In North
In addition, multicast sessions can also be America, the bandwidth of a single QAM
managed with quality of service (QoS) channel is capped at 38.8Mbps with QAM256
guarantees. Switched multicast in an modulation. The Law of Large Numbers implies
IP/DOCSIS transport increases bandwidth that statistical multiplexing efficiency improves
utilization efficiency in the same way as as the number of VBR streams in the
switched digital video does in an MPEG transmission channel increases. The
transport. proliferation of HDTV in households and the
increasing number of HD streams in the
Equally important in terms of their potential network make wideband transport much more
technological impacts on the industry are attractive for the purpose of statistical
modular CMTS (M-CMTS) and universal multiplexing. While the narrowband MPEG
quadrature amplitude modulation devices transport struggles to provide efficient statistical
(QAMs). The separation of the DOCSIS media multiplexing of HD streams without
access control (MAC) and physical layer compromising the video quality, the bonded
protocol (PHY) allows independent scaling of DOCSIS 3.0 transport can easily support
upstream and downstream bandwidth. statistical multiplexing of HD streams with good
Economics of scale will drive down the costs of statistical multiplexing gains and video quality.
universal QAMs, lower the overall solution cost
of M-CMTS and make DOCSIS economically IP networks also have built-in quality of
viable for IP video delivery. service (QoS) capabilities. Cable modem
termination systems (CMTSs) implement
IP Transport vs. MPEG Transport advanced DOCSIS QoS features such as
admission and policy control, priority queuing,
IP transport distinguishes itself from MPEG traffic policing, traffic shaping etc. These IP
transport in a number of ways. IP transport is network features are readily applicable for VBR
asynchronous packet-oriented transport. IP video delivery.
networks also introduce jitter. MPEG transport,
on the other hand, is synchronous transport. IP networks also are converged networks.
When MPEG transport streams are delivered They allow data, voice and video to be
over IP networks, receivers must remove simultaneously delivered through the network.
network jitter in order to recover the original Converged networks provide great bandwidth
video source clock. IP set-top boxes (STBs) savings just by enabling all
normally have dejittering buffers that can
tolerate 100ms of network jitter.

2008 NCTA Technical Papers - page 3

 VoD
Server
QoS
control Queue
Local Cable IP
IP Network
Encoder CMTS w. Modem STB
Network Statmux
Linear video
source
IP
Encap
Figure 1 The VBR Network Statmux Architecture

services to share a single bandwidth resource network edge. Figure 1 depicts a high level
pool. If the slightly different peak hours of these system diagram of network statmuxes.
different services are also considered, the
bandwidth savings are even greater. Field data In this architecture, all IP video sources
indicates that bandwidth savings of 20 to 30 transmit VBR streams. VOD servers store and
percent are achievable with a truly converged stream VOD content in single program transport
network. Even better is the fact that these stream (SPTS) VBR format directly. The coding
savings are realized on top of the VBR over efficiency improvement of VBR over CBR
CBR bandwidth savings. In a converged IP bodes well for VOD as it brings 40 percent or
network, if at a certain instant there is leftover more storage and streaming capacity savings to
bandwidth after all the VBR video traffic, then VOD servers. Local encoders encode real time
lower priority data traffic can consume the video and send out SPTSs with desired VBR
unused bandwidth. In other words, not a single mean rates and peak rates. Linear video sources
bit of bandwidth is wasted! MPEG transport from satellite are converted from multi-program
networks, however, are special-purpose transport stream (MPTS) to SPTS and IP
networks used for video delivery exclusively. In encapsulation is added at the same time.
an MPEG transport network, either due to
MPEG virtual buffer constraints or low The IP video streams then travel from the
instantaneous video bitrate, MPEG statmuxes video sources through the network and arrive at
must insert NULL packets to fill the MPEG the cable network edges where the last mile is
transmission channel. The NULL packet filled the DOCSIS path. The CMTS, be it modular or
bandwidth is simply wasted. integrated CMTS, is the starting point of the
DOCSIS path. Powered with QoS control and
VBR Network Statmuxes advanced queuing features, the CMTS is an
ideal candidate to implement VBR network
The characteristics of IP transport make it a statmuxing. At the CMTS, if there is no
perfect match for VBR statistical multiplexing. congestion, i.e. when the combined VBR stream
The essence of an IP-level VBR network bandwidth is less than the bandwidth limitation
statmux, or simply network statmux, is to avoid of the IP/DOCSIS pipe, all video streams pass
transrating at the MPEG-level in an attempt to through packet buffers inside the CMTS with
solve the bandwidth oversubscription problem. minimum delays. When congestion occurs, the
Instead, queuing and buffering are used at the bursty VBR streams will be queued up at the

2008 NCTA Technical Papers - page 4

CMTS temporarily. In the extreme case that the Besides the number of streams participating
CMTS buffer is full, packets will be dropped by in statistical multiplexing, another crucial factor
the CMTS, though the probability of such drop that affects the statistical multiplexing
to occur is controlled by an admission control efficiency is the VBR peak (bandwidth) to mean
function that limits the number of video flows (bandwidth) ratio. The higher the peak to mean
admitted to a single pipe. The CMTS queuing ratio, the less efficient the statistical
will be further discussed in more details. multiplexing is. Early research [2] has shown
that the MPEG-2 encoded broadcast quality
The IP video streams eventually terminate at VBR video has a typical peak to mean ratio
IP STBs. The IP STB can either be a standalone from 1.3 to 2.4. IP network statmux design
IP STB behind a cable modem, or a hybrid STB assumes video sources have peak to mean ratios
with an embedded cable modem. As a result of within the limit of 2.4. This peak to mean ratio
CMTS queuing, additional network jitter is should be enforced at the video encoder for best
introduced to video streams. This jitter is video quality. Although service providers can
absorbed by the IP STB as it dejitters and use rate clampers along the video transmission
buffers packets before video is sent to video path, the method of rate clamping at the network
decoders. In order for VBR network statmuxing is inferior to the encoder peak rate enforcement
to work, there must be a limitation of the jitter solution as a consequence of additional video
introduced by the CMTS queuing so that the processing cost and degraded video quality.
end-to-end network jitter is tolerable to the IP
STB. In today’s well managed service provider The buffering and queuing scheme brings
network, end-to-end network jitter is a significant advantages to IP network statmuxes
magnitude lower than the dejittering capability over traditional MPEG statmuxes. First,
of IP STBs. This leaves a big jitter budget for network statmuxes preserve original video
the CMTS to implement VBR statmuxing. For quality keeping the video unchanged at MPEG
instance, the CMTS introduced jitter can be level. Video is produced by video sources and
limited to 60ms. The jitter limitation is enforced consumed at STB receivers. No network
on the CMTS by restricting the CMTS queue components within the transmission path re-
buffer size. The buffer size is chosen so that encode the video content between video sources
packets will stay in the buffer for less than the and STBs. For the same reason, pre-encrypted
jitter limit time. The CMTS queue can be content can now be easily multiplexed by
drained at the maximum bandwidth of the network statmuxes. Second, network statmuxes
bonded DOCSIS channel. introduce delays bounded by the network jitter
limit, which make them ideal for low latency
The VBR video traffic should be marked video services such as VOD and interactive TV.
with higher priority than data traffic either by Lastly, by avoiding the expensive transrating
differential services control point (DSCP) operation and by leveraging the built-in QoS
marking, or video flows should be explicitly capabilities of IP networks, network statmuxing
identified through flow specs. In a converged turns out to be a cost effective approach to VBR
IP/DOCSIS pipe, any underutilized bandwidth video delivery.
resources after the VBR video traffic can be
consumed by lower priority data traffic. As a summary, Table 1 below highlights the
key differences between network statmuxes and
traditional MPEG statmuxes.

2008 NCTA Technical Papers - page 5

 Network Statmux Traditional MPEG Statmux
Statmux Efficiency More efficient w. wideband Limited by the narrowband MPEG
DOCSIS QAM
Bandwidth Overflow Buffer and delay Transrate
Video Quality As good as original stream Quality degradation from transrating
Latency Less than 100ms, e.g. 60ms 0.5-1 second
Pre-encryption Transparent Have difficulty
Cost Buffering and QoS already built Additional system components
into IP transport Deep packet MPEG level transrating
No deep packet processing
Bandwidth Utilization 100% with converged network Null packet filling, suboptimal

Table 1. Key Differences - Network and Traditional MPEG Statmuxes

IRT/RTE

DCM Delay Emulator 7600 Quality Analyzer STB

CMTS QAM CM PC

Figure 2 VBR over DOCSIS System Test Diagram

THE EFFICIENCY OF NETWORK different sizes are used to smooth out the VBR
STATMUXES traffic.

Exactly how efficient are network statmuxes? Figure 2 presents the system diagram of VBR
Perhaps nothing answers the question better than over DOCSIS testing. In this experiment, video
lab results from a proof of concept project. The sources are obtained from satellite feeds. Video
basic design is to deliver VBR video streams into streams are converted from MPTS to VBR SPTS
a controlled IP/DOCSIS channel with embedded and IP encapsulation is added by a Digital
QoS features. Streams are added to the channel Content Manager (DCM). The VBR video
one by one until the video quality is affected by streams then pass through a network delay
packet drops. Based on the maximum number of emulator before they reach the DOCSIS CMTS.
VBR streams supported and the transmission In the DOCSIS path, a pre-DOCSIS 3.0 modular
channel bandwidth capacity, VBR statmux CMTS based wideband solution is used. In this
efficiency and bandwidth utilization solution, the CMTS, the EQAM and the cable
improvement over CBR are calculated. To modem together form the DOCSIS last mile.
further study the effect of QoS control on VBR VBR video streams are terminated at the IP STB.
statmux efficiency, network buffers with

2008 NCTA Technical Papers - page 6

A video quality analyzer is added to the path to Unfortunately, IP packet drops will cause
monitor video impairments. A PC is used as a considerable video quality impairments. To
receiver of best effort data traffic. ensure that video quality degradation due to IP
transport network is negligible from a
subscribers point of view, most carriers allow the
Video Source transport network to introduce at most one
visible degradation in video quality every two
Live SD-only satellite feeds are used as the hours. This criteria translates to 1E-6 maximum
VBR sources. In the video industry, SD CBR packet drop probability. Packet drops are
MPEG2 video streams are encoded at a nominal detected from QoS counters either in the CMTS
rate of 3.75Mbps. The video quality associated or in the video quality analyzer. In the test, VBR
with 3.75 Mbps CBR coding is well accepted as streams are inserted to the DOCSIS path one by
the standard for broadcast video. Assuming 40 one while packet drops are monitored. A VBR
percent VBR coding efficiency improvement stream can only be added if this addition will not
over CBR, the VBR stream with equivalent cause any packet drops for a two hour period.
video quality should have an average bitrate Besides, the probability of packet drop is
around 2.25Mbps. Two HITS satellite feeds measured over long term tests and is close to 5E-
selected for this experiment have just the right 7, which is better than the well accepted 1E-6
video characteristics. HITS1 and HITS9 each criteria.
comes as an MPTS bundle from the satellite at
27Mbps. Each bundle has 12 video programs. DOCSIS 3.0 Channel and QoS Options
The average bitrate is 2.25Mbps and the peak to
mean ratio is 2 to 2.4. Different DOCSIS 3.0 bonding group sizes
are used to investigate the statmuxing efficiency
Since these VBR streams originate from with regard to channel bandwidth. To make it
MPTS bundles, they are not good sources due to simple, the bandwidth increment is 38.8Mbps –
the correlated bitrate peaks and valleys. To create the bandwidth equivalent of a single QAM.
independent VBR streams with uncorrelated
bitrate peaks and valleys, the network emulator is The CMTS wideband default class queue is
used to insert different delays to individual assigned to the VBR streams. Since all VBR
streams. For example, the first stream of the video streams are delivered through multicast in
MPTS bundle is delayed 300ms, the second the test, the classifier at the CMTS classifies all
stream is delayed 600ms, the third stream is multicast video traffic to the default class queue.
delayed 900ms and so on until the last stream is The default class queue size is adjusted to reflect
reached. The emulated delays combined with the the maximum jitter introduced by the CMTS.
live feeds generate the desired VBR sources. Best-effort IP data traffic is mixed with video
traffic to drive the IP bandwidth utilization to
Video Quality Measure 100%. While the default class is used to transmit
video traffic, the best effort class queue is used to
The IP video delivery quality can be transmit unicast traffic to PC.
measured with both packet drops and network
jitter. As discussed earlier in the paper, the buffer Results and Discussions
sizes in the CMTS are chosen so that the
maximum jitter introduced is bounded. The
CMTS-introduced jitter will be removed by IP To facilitate discussions, two quantitative
STBs without affecting video quality. measures are defined. VBR statmux efficiency is
defined as

2008 NCTA Technical Papers - page 7

 Figure 3. VBR Network Statmux Efficiency in
Efficiency (%) = (#VBR streams * VBR average DOCSIS
rate) / channel capacity
Bandwidth Utilization Improvement vs. CBR
where #VBR streams is defined as the maximum
number of VBR streams that can fit in a 70 57.5%
transmission channel without causing packet
60
drops for a two hour window. The theoretical

Number of Streams
50
limitation of VBR statmux efficiency is 100% if
40 50%
one considers CBR as a special case of VBR
30
with a peak-to-mean ratio of one. 40%
20

The bandwidth utilization improvement is 10

also derived by comparing VBR statmuxing with 0

1 2 4
a CBR based solution. The bandwidth utilization Channel Bandwidth in QAMs CBR VBR
improvement is defined as
Figure 4. VBR Network Statmux Bandwidth
Improvement (%) = (#VBR streams - #CBR Utilization Improvement
streams) / #CBR streams
Several conclusions related to network
where the #CBR streams is defined as the statmux efficiency can be readily drawn from
maximum number of CBR streams with these results. First, the network statmux
equivalent video quality that can fit in the same efficiency improves when the transmission
channel. channel bandwidth increases. With 60ms QoS
buffers, the statmux efficiency is 81.2 percent for
The experiment results are displayed in a DOCSIS channel of 38.8Mbps and the
Figure 3 and Figure 4. Figure 3 highlights the efficiency rapidly reaches 91.3 percent when the
VBR network statmux efficiency vs. DOCSIS DOCSIS channel bandwidth is 232.8Mbps with
channel bandwidth. Figure 4 shows the VBR four bonded QAM channels. Quite contrary to
network statmux bandwidth utilization our original assumptions, VBR network
improvement over CBR vs. DOCSIS channel statmuxes are efficient for SD-only content even
bandwidth. when used with unbonded DOCSIS channels.

VBR Network Statmux Efficiency in DOCSIS Path Next, network statmuxes dramatically
improve the bandwidth utilization over the CBR
100.0%
solution. While the unbonded DOCSIS channel
95.0% delivers 40 percent more streams if VBR and
90.0% network statmuxing are utilized, the four channel
Efficiency (%)

85.0%
bonded DOCSIS can boast a 57.5 percent
enhancement. This improvement is superior to
80.0%
what traditional MPEG statmuxes can achieve in
75.0% the MPEG transport path. Because network
70.0%
statmuxes only require queuing and buffering
1 2 4 instead of the heavy MPEG level processing
Channel Bandwidth in QAMs DOCSIS required by traditional MPEG statmuxes, the
same or more bandwidth savings are achieved
with only a fraction of the cost.

2008 NCTA Technical Papers - page 8

 Finally, bandwidth utilization should not be
In addition, as the QoS buffer size increases, confused with VBR statistical multiplexing
the network statmux efficiency further improves. efficiency. Bandwidth utilization of a
This aspect of the testing was implemented on a transmission channel can reach 100 percent as
simulated Gigabit Ethernet (GE) path with QoS long as the lower priority traffic can be mixed
buffer control. To demonstrate the QoS buffer with video data and no single bit of bandwidth is
effect, the buffer size is represented in terms of wasted. There is no doubt that a converged IP
the maximum jitter introduced by the buffer. In pipe holds the promise to fully utilize the
Figure 5, when the bandwidth is two QAM transmission channel. To prove the point, the PC
equivalent, the statmux efficiency is 75.4 percent behind the cable modem pulls big files from the
with a 4ms buffer and the efficiency is 86.9 CMTS through the bonded DOCSIS channel
percent with a 60ms buffer. However, as the when the VBR video bandwidth utilization is as
channel bandwidth increases, the buffer size high as 90 percent. No packet drops are detected
introduced improvement is reduced. For instance, during the process due to the QoS features of the
if the channel bandwidth is 38.8 Mbps and the CMTS and the priority treatment of VBR video
buffer size is increased from 4ms to 60ms, the traffic.
statmux efficiency improves 11.5 percent. In
contrast, when the channel bandwidth is
232.8Mbps, increasing the buffer size from 4ms RELIABLE VIDEO DELIVERY
to 60ms only yields about 4 percent improvement
in statmux efficiency. Video streams are particularly sensitive to
packet drops. Because of the high level of
VBR Network Statmux Efficiency in Simulated Path
compression used in video delivery, even a
single packet drop could result in significant
100.0%
video artifacts. There are three main sources of
95.0% packet drops in IP networks:
Efficiency (%)

90.0%

85.0%
- Because the core of the network is usually
rich with bandwidth, packet drops at the core are
80.0%
usually not related to congestion. Instead, load
75.0% balancing actions, route changes and/or
70.0% temporary equipment failures could cause packet
1 2 4 6 drops.
Channel Bandwidth in QAMs 4ms delay
60ms delay
- The edge is relatively bandwidth-poor. It is
the pipe to the subscriber which is at risk of
Figure 5. The Buffer Size Effect being congested and as a result, it is where
packets are most likely to be dropped.
This is not unexpected since the statmux
efficiency is already very high when the channel
- Media errors: though technically packets
bandwidth reaches a high threshold, thus, room
might be dropped on the Ethernet part of the
for additional improvement is limited. Note that
network, most drops occur on the HFC network
the statmux efficiency of an IP/DOCSIS path is
itself due to RF issues.
slightly worse than that of the simulated IP/GE
path as a result of additional DOCSIS overhead
To minimize and possibly eliminate the video
at layer one and layer two.
artifacts caused by packet drops, we propose a

2008 NCTA Technical Papers - page 9

three-tier approach which addresses all major with a backhaul link of 1Gbps, and two video
sources of packet drops: flows, each one with a traffic peak of 6mbps.
Furthermore, we can assume that admission
- Admission control: the role of admission control limits the CMTS to accept only two
control is to make sure that the network, or in the flows for this case (though more could possibly
context of this paper, the CMTS specifically, can have been admitted). Both flows drain onto a
deliver content reliably. single output that is capable of 10mbps. We
assume that the targets of these flows are two
- Scheduling: while admission control makes cable modems as depicted in Figure 6.
sure that we deliver content reliability, it is the
scheduler that does the actual work of delivering
the content in real time in a reliable way.
6 Mbps CM1
- Error repair: Error repair was designed to
1 gbps 10
help in cases where packets are dropped because Mbps
of media errors and/or network flaps, however, it
could be used to help in cases where both 6 Mbps CM2
admission control and scheduling could still not
guarantee packet delivery.

ADMISSION CONTROL AND SCHEDULING

Figure 6. Queuing example

With CBR services, admission control is Since the video flows are 6Mbps each there
trivial. If a CBR flow requires X mbps, and the is no congestion risk on the backhaul. However,
bandwidth of the channel is 10X Mbps, then 10 when they get to the cable interface the worst
flows can be admitted. The CMTS would track case aggregate rate they can reach is 12Mbps
the number of flows that the cable segment has to while the cable interface in this example can
carry, and reject any request to activate an 11th support only 10Mbps. The tools the CMTS can
flow. use are queuing and scheduling:

When it comes to VBR the picture is more - queuing will buffer up the packets in a
complex. As explained previously in this paper, “queue” until the 10 Mbps channel is
VBRs can be oversubscribed because it is available again to send them.
“reasonable” to assume that not all flows send at
- scheduling will decide which queue to
peak traffic rate all at the same time. But what
service and in what order
does “reasonable” mean? There is a probability
that enough traffic peaks occur at the same time The CMTS can use the DOCSIS tools to
and in such an event the channel will not have define the queuing/scheduling structure needed
enough capacity to carry all the traffic. In such to deliver these flows reliably:
an event the CMTS scheduling queuing and
admission control disciplines will help to - A classifier that will uniquely identify the
minimize (or eliminate) packet drops in the event video flow. For example, the combination of
of traffic congestion. a destination IP address of the client device,
and a destination user datagram protocol
To illustrate how a scheduler works, we can (UDP) port are a good way to identify a
start with a simple example: assume a CMTS

2008 NCTA Technical Papers - page 10

 packet stream that belongs to a single video other services such as data. The priority
flow. This approach can also be used for parameter, along with the “guaranteed minimum
multicast flows. The detailed discussion of rate” parameter, gives an assurance that even if
how this classifier is created is outside the data services in a given channel are congested to
scope of this paper. the point where packets are dropped, there is still
enough bandwidth dedicated to high priority
- A service level agreement that defines how video flows.
to queue the flow. For example, in our case
it’s a flow that has a 6 Mbps peak rate. However, even with proper admission control
and scheduling, packet drops could still occur.
The CMTS manages queuing by dedicating a Facilitated with extremely low packet drop rate,
queue to each one of the video flows and by error retransmission and IP-based packet-level
controlling queue scheduling. forward error correction (FEC) are promising
cost-effective solutions for the packet drop
Naturally all the queuing/scheduling can do problem. Both technologies are well understood
is to mitigate the cases where the aggregate and have been applied to IP video applications to
traffic bursts are above 10mbps. If the bursts are protect video streams from common impairments
too long then packets will experience an of IP networks, such as packet loss.
unacceptable delay in the CMTS queues (and
eventually will get dropped as the CMTS queues MPEG over UDP/IP is widely used to
have limited size). By putting a limit on the transmit real time video traffic through IP
number of flow admitted by admission control, networks. With UDP transmission, packet drops
the MSO can control the tradeoff of how many are not reliably detected due to the lack of a
flows can be admitted to a channel vs. what the sequence number at the UDP layer and due to the
packet drop probability would be. limited capabilities of the MPEG transport
stream level continuity counter. By adding a real
An additional tool that can reduce the risk of time protocol RTP above UDP, packet drops are
packet drops (at the expense of having less video easily identified through a 16-bit sequence
flows committed) is the use of “guaranteed number at the RTP layer. Error retransmission
minimum rate”. This parameter in DOCSIS and FEC leverage the RTP encapsulation of the
defines the rate that a scheduler MUST deliver video stream and repair dropped packets at the
even in a case of congestion. In a way, one can network edge. An example architecture for error
view CBR as a case where the the peak rate retransmission and FEC and is shown in Figure
equals the committed rate. Based on this, the 7.
smaller the difference between the committed
rate and the peak rate, the smaller the risk of
packet drops.

Another tool in the DOCSIS toolkit is

“priority”. This parameter is critical in an
environment where we have mixed video with

2008 NCTA Technical Papers - page 11

 Retransmissions
Repair-S
IPSTB

Video Repair-C
Source CMTS CM

Video Stream

FEC Stream

Figure 7. Error Repair Architecture

In the above error repair architecture (Figure networks, but also demonstrates the tremendous
7), an error repair client is located at an IP STB. value and potentials of wideband DOCSIS in
When FEC is applied, the video source sends out video delivery. DOCSIS CMTSs, with their
an FEC stream along with the video stream. In built-in advanced QoS capabilities play an
this scheme, periodically selected media packets important role in VBR network statmuxing.
are used to generate FEC packets. The error
repair client is responsible for detecting packet The trend in tomorrow’s video delivery is
loss and recovering the lost packets utilizing the more HD content and more advance coded video
additional FEC stream. When error content. VBR network statmuxes respond to this
retransmission is utilized, a repair server at the trend by leveraging the channel bonding
network edge caches video content. The error capability of DOCSIS 3.0 and generate
repair client utilizes standard based RTP/RTCP unprecedented multiplexing efficiency as the
toolkit defined by IETF to request retransmission wide channel is promising to get wider. By
of lost packets. The same toolkit can also be used avoiding deep packet processing, VBR network
to accelerate channel changes in IPTV. statmuxes scale easily to future video coding
technologies.

CONCLUSION
REFERENCES
The ever increasing demand for bandwidth
requires efficient HFC bandwidth utilization.
DOCSIS 3.0 and IP video are shifting the VBR 1. Si Jun Huang, “Principle, Applications of
video delivery to a new paradigm. IP level VBR Variable Bit Rate Coding for Digital Video
network statmuxes overcome the shortcomings Broadcasting, w. Statistical Multiplexing
of traditional MPEG statmuxes and provide the Extension”, NAB 1999
least expensive, low latency and best video
quality approach to reap the benefits of VBR 2. Daniel P. Heyman and T.V. Lakshman,
video. “Source Models for VBR Broadcast Video
Traffic”, IEEE Transactions on Networking, Feb.
The proof of concept work introduced in this 1996
paper not only proves the feasibility of this VBR
network statmuxing in today’s DOCSIS 3.0

2008 NCTA Technical Papers - page 12

 VIDEO LAYER QUALITY OF SERVICE:
UNPRECEDENTED CONTROL AND
THE BEST VIDEO QUALITY AT ANY GIVEN BIT RATE

Ron Gutman,
Marc Tayer

Imagine Communications, Inc.

Abstract
INTRODUCTION
Spurred by DirecTV’s 2007 declaration that
it will be the world’s first television service The North American market for video
provider to reach 100 HD channels, cable subscribers is becoming increasingly
operators are moving rapidly to create competitive and fragmented, with cable, DBS,
additional bandwidth not only to carry dozens telco and Internet service providers all
more linear HD channels, but also to provide jockeying to gain a bigger piece of a growing
hundreds, and eventually thousands, of HD- pie. After a long gestation period, the HDTV
VOD titles. The video quality bar is market is finally hitting its stride. The most
simultaneously rising due to the mass consumer successful service providers will offer libraries
adoption of large HDTV displays and the of virtually unlimited content delivered
growing popularity of Blu-ray. conveniently and with the highest possible video
quality.
This paper discusses the fundamental and
elusive paradox of how to cost-effectively An important emerging element of this
increase bandwidth efficiency without infrastructure is Video Layer QoS, defined as
sacrificing video quality. Leveraging a concept the establishment of video quality levels at the
from IP networking, Video Layer Quality of content origination or delivery site, combined
Service (Video Layer QoS) involves creating with the process of sustaining these levels all the
minimum and maximum video quality values at way to the consumer viewing environment in
the service level, while adding the technical the most bandwidth efficient manner possible.
dimension of true Variable Bit Rate (VBR)
constant quality video coding. A Video Layer QoS solution provides:

Similarly, Video Layer QoS allows the 1. Excellent MPEG-2 video quality at 3:1
optimization of bandwidth efficiency while HD and 15:1 SD (per 256 QAM
guaranteeing the quality of service in a channel) on an end-to-end basis, from
sustainable manner throughout the various content origination all the way to the set-
switching, multiplexing and splicing stages of top box.
video communications networks. The paper also 2. Consistency (equalization) of a service
discusses the human visual perceptual system as provider’s video quality across the
well as related video processing and delivery Digital Broadcast, SDV, VOD and
aspects for a variety of digital video services. Network PVR (e.g., Start Over)
categories.

2008 NCTA Technical Papers - page 90

 3. The ability to assign different quality field grass or basketball courts; lack of detail,
levels to different classes of assets (e.g., softness or the absence of a “pop” effect in a
HD-VOD PPV), or even individual complex or colorful image; or tiling around
assets (e.g., the Super Bowl), at the logos or scrolling text areas. In many cases, the
discretion and under the control of the discerning viewer may need to wait for a period
content provider or operator. of high activity in the video stream in order to
4. Sustenance of the pre-calibrated video see artifacts, such as rapid motion, panning,
quality levels in a cost effective, non- scene changes, fades or flashes. This instability
disruptive and backward compatible and unpredictability of quality over time can be
manner throughout the various quite annoying, and is highly correlated to
multiplexing stages, including local and consumer complaints regarding delivered video
addressable ad insertion. quality.

THE HUMAN VISUAL SYSTEM AND The following images show the same picture
PERCEIVED VIDEO QUALITY compressed with three different methodologies.
In the first image, a “Compression by Quality”
A logical place to begin a discussion of technique is used, in which an objective video
video quality is the area of human visual quality quality measurement system is involved to
perception. The visual and perceptual system minimize compression artifacts relative to the
can not merely be construed in the context of source. In the second image, a typical MPEG-2
resolution, frame rate and bit rate since these encoder is used. In the third image, pre-
factors alone do not explain the phenomenon in processing and high frequency pixel filtering
which two streams with equivalent parameter techniques are used in addition to the traditional
settings can appear very differently to the compression methods.
human eye. The two streams may look quite
similar most of the time, but the majority of The first image appears noticeably sharper
subjects in a typical focus group will still select and cleaner than the other two, showing neither
one sequence over the other. the blocking artifacts of the second image nor
the blurriness or loss in detail of the third image.
When standing close to two identical screens All other things being equal, such an
positioned side-by-side, a trained set of eyes can improvement in perceived quality can be made
begin observing the traditional compression possible if the video quality has been
artifacts. To mention a few notorious examples, exhaustively analyzed and measured as part of
many of us have observed blockiness at facial the video processing and multiplexing solution.
edges, in sky-dominated backgrounds, and
during scene changes; random noise on football

2008 NCTA Technical Papers - page 91

Figure 1 – Compression by Quality

Figure 2 – Typical Compression

2008 NCTA Technical Papers - page 92

Figure 3 – Compression by Pre-Processing and High Frequency Pixel Filtering

Figure 4 – Human Visual System (HVS)

2008 NCTA Technical Papers - page 93

 In this manner, it is finally possible to these complex information streams while the
guarantee the Video Layer QoS at any second, viewer relaxes comfortably on his or her couch
at any frame and even at any pixel. But before watching television.
jumping more deeply into video quality
measurement, we must first briefly discuss the Each layer of visual processing or
anatomical and psycho-visual features of the compression removes unnecessary levels of
human visual system (HVS). informational redundancy, forming the video
signal into data that are essential for human
Visual stimuli, in the form of light and interpretation, i.e., entertainment or viewing
images coming from a TV screen, are focused satisfaction.
by the optical components of the eye, including
the cornea, pupil, lens and eye fluids. These The more redundant information that can be
stimuli are then translated into electrical signals removed during the video compression process,
by the neurons and photoreceptor cells in the the fewer bits are required to be delivered over a
retina, before being received and organized by communications network or stored on a storage
the lateral geniculate nucleus (LGN) in the medium. The ubiquitous audiovisual coding
brain. The LGN then transmits these signals to standards of MPEG-2, and increasingly MPEG-
the primary visual cortex which tunes and 4 AVC (H.264), are designed specifically for
processes them into spatial and temporal this purpose. However, with TV screens
frequencies, orientations and motion. Higher becoming increasingly larger, compression-
levels of visual processing, cognition and related
memory associations subconsciously analyze

Human Visual System (HVS) Feature Compression & Quality Measurement

Guidance
Eye optics modeled by a low-pass point Caveat: making tradeoffs against picture
spread function (PSF) sharpness is a risky area
Non-uniform retinal sampling “Compress the perimeter more” trick has
some utility but is not working as well as in
previous video quality tests
Luminance masking Potentially ripe area; extreme darks and
lights can be compressed more
Spatial frequency, temporal frequency and Another ripe area, but needs adaptation to
contrast sensitivity functions specific MPEG-2 and MPEG-4 AVC
compression impairments
Masking and facilitation Some image components do a good job of
masking the visibility of others. Very
difficult area to model and compute.
Neural pooling (cognition layer) Perceptible distortion is more annoying in
some areas of the scene (human faces, text,
sea or sky background) than in others.

Table 1 – Compression Tricks Relative to the Human Visual System (HVS)

2008 NCTA Technical Papers - page 94

artifacts which may previously have been VBR to CBR conversion (or vice versa) and
relegated to the category of acceptable or splicing stages.
imperceptible marginal noise are now distinctly
observable and in many cases even annoying to Step 1: Select or devise a video quality
ordinary consumers. measurement technique

Table 1 shows some known characteristics There are several subjective video quality
of the human visual system, and then comments testing methods that are accepted by industry
on their potential effectiveness with respect to professionals, such as the Double Stimulus
video quality measurement and image Impairment Scale, the Double Stimulus
compression. Continuous Quality Scale, and other methods
described in ITU-R BT.500-11. In contrast,
VIDEO PROCESSING AND objective video quality measurement methods,
MULTIPLEXING BY QUALITY by attempting to correlate as closely as possible
to subjective test results, are very elusive by
In this section we describe a method that definition. For this reason, through the history
allows “closing the loop” with respect to of digital video, subjective video methods have
objective video quality measurement systems, been heavily relied upon, with objective video
significantly increasing the signal quality (and quality method serving more as a sanity
bit rate efficiency), and providing Video Layer
QoS through the re-processing, re-multiplexing,

Objective Video Quality Computational Correlation with

Measurement Method Complexity Subjective Test Results
PSNR Simple Poor , even imperceptible
Peak-Signal-to-Noise-Ratio pixel errors contribute
Most common. Based on negatively to the measured
mean squared error (MSE). result
MPQM Complex Mixed. Certain parameters
Moving Pictures Quality are incorporated.
Metric
VQM Very Complex Good, measures
Video Quality Metric perceptible impairments
ANSI T1.801.03-2003 such as blurring, jerkiness
and distortion.
SSIM Complex Fair, uses a structural
Structural Similarity Index distortion measure instead
[3] of error.
ICE-Q™ Very Complex Excellent. Accounts for
Interchangeable Compressed numerous visual
Elements-Quality impairments; designed and
optimized specifically for
MPEG-2 and MPEG-4
AVC (H.264)
Table 2 – Objective Video Quality Measurement Systems

2008 NCTA Technical Papers - page 95

 objective video quality measurement system as
check or rationale for adding certain the arbiter, the resultant reconstructed signal is
compression tools to a standard. In other words, essentially guaranteed to be a constant quality
subjective video quality testing has been the signal at levels or grades which are known in
litmus test up until now. advance. This signal is Variable Bit Rate (VBR)
by definition since the activity and complexity
All objective video quality measurement vary over time. High complexity scenes will
methods use some form of HVS modeling. The automatically be processed at higher bit rates
more successful methods are backed by than low complexity scenes, with both types of
correlation with subjective video quality test scenes being coded at the same measured
results and have endured long periods of tuning. quality level, hence the notion constant quality.
They are also very computationally intensive, in
effect representing a form of artificial In great contrast to today’s encoding or
intelligence. Table 2 shows a summary of the rate-shaping methods, the video processor is
known objective video quality measurement configurable to a pre-calibrated quality level
methods and their correlation rather than a maximum, minimum, or average
bit rate. A recommended method to guide this
process is to use a mathematical scale, such as 1
to subjective tests results based on personal to 100, rather than more crude or subjective
experience as well as available information: groupings such as “good,” “bad,” or “average.”

Note that all of these objective video Step 3: Calibration

quality measurement methods involve the
comparison of two signals. For example, they During this stage, all of the available signals
may involve an uncompressed source vs. a or video assets need to be processed (i.e.,
compressed/decompressed signal, or a satellite- intelligently compressed using an effective
received compressed signal vs. a re-encoded objective video quality measurement system),
signal. This remark becomes more relevant and using the video processor from Step 2,
important in subsequent stages of a signal path, employed at various selected quality grades.
in which the stream is re-multiplexed, The system should be calibrated in such a way
potentially multiple times, before arriving at the that the service provider is reasonably
consumer’s set-top box. comfortable with the constant quality
experience at any grade. If this is not the case,
then the previous steps should be repeated. One
Step 2: Video Processing or Encoding can define a minimum of two quality grades in a
similar fashion to the QoS utilized in IP
For this step, a video processing device is networks as follows:
required capable of “closing the loop” with the
selected objective video quality measurement 1. QGa – target average video quality
method. It requires processing of every frame grade, for example “96”
and every macroblock of every frame, as part of 2. QGb – Guaranteed or minimum allowed
the selection of a constant video quality video quality grade, for example “90”
requirement, level or “grade.” Once the
decisions are made, by iteratively comparing the In some deployments, QGa can be defined as
re-processed options to the source, using the “just noticeable difference” (JND), which means

2008 NCTA Technical Papers - page 96

even expert viewers (i.e., “golden eyes”), cannot Step 5: Lineup
see substantial differences from the source.
Then, QGb can be defined as the quality level or Determining the digital service combination
grade at which, the vast majority of the time, per multiplex contains a goal of providing QGa
ordinary viewers can’t discern differences from quality on average and never less than QGb. The
the source. following equations can help optimize the
multiplex lineup using the bit rate measurement
A good practice for delivering the signals, statistics first. For example in 3:1 HD within a
including packing density, suggests a target of 256 QAM channel at 38.8Mbps:
no more than 1% of the time the video stream
will contain QGb. BQ)c <= 38.8Mbps

Step 4: Statistics
In order to guarantee the quality it is
possible to simulate the statmux by repeating
Process all of the target channels or video
this calculation for every second in the database
assets at QGa and QGb and gather statistics for
at least 24 hours, or preferably for one week.
Measure the respective bit rates per second and Ba(t), Bb(t))c <= 38.8Mbps
create two vectors, one for each quality grade.
Select Bb(t) only when needed and by
Ba(t) – bit rate measured per second at QGa measuring Bb(t) usage at less than 1%.
Bb(t) – bit rate measured per second QGb
Because of the natural statistical behavior
Per channel, calculate your global (time of constant quality signals, it is advisable to
tested) average bit rate at QGa and your global have the largest number of signals per mux as
maximum bit rate at QGb. possible.

BAa = Average (Ba(t))

BMb = Maximum (Bb(t)) Step 6: Statistical Multiplexing

BQ = Maximum (BAa, BMb) – defined as the Using the lineup as defined in Step 5, it is
channel effective bit rate for lineup allocation, now time to actively statistical multiplex the
utilized statically for digital broadcast and streams. The encoders should be able to encode
dynamically for VOD, SDV and Internet video. at multiple quality grades in real time and the
statmux should choose the highest quality grade
It is also possible to correlate the bit rate possible under the maximum channel bit rate
statistics to time of day or type of program. constraint. The grades are expected to extend to
Interestingly enough, the quality requirements the entire range between QGb (“90”) or even
during prime time are generally higher than lower, through QGa (“96”), and up to “100.”
average. In other words, the average bit rate of The proportion of null packets should be very
QGa and the maximum bit rate of QGb are close to 0% at any grade under “100.”
higher in prime time; therefore, BQ should be
calculated during this time window. The statmux device should report the
eventual quality grades utilized in the stream.
Some of the channels may change their content
type over time. HD channels currently using

2008 NCTA Technical Papers - page 97

upconverted SD content will use an increasing model calculation, but when converting the
proportion of native HD content over time. signal into CBR the percentage of time at which
Certain movie channels may rarely show it is running under quality grade QGa might be
concerts or sports events that are more difficult significantly higher than 1%. It is possible to
to compress, while other channels may alternate iteratively and heuristically determine the
between movies, sports and concerts. It is optimal CBR rate for QGa and QGb. Note that
important to monitor the average and this bit rate is significantly higher than BQ,
instantaneous video quality grade for every which is the effective bit rate in VBR. Since the
mux, including the percentage of time the CBR rates are generally expected to be
system is running at a grade under QGa 3.75Mbps for SD and 15Mbps HD, it is possible
(expected to be less than 1%) and the percentage to calculate, in advance, the average quality and
of time the system is running at a grade under percentage of time at which streams are running
QGb (expected to be less than 0.1%). at quality grades under QGa and QGb.

A service provider may also choose to

completely skip Steps 4 and 5 and base the THE BOTTOM LINE RESULT:
lineup selections entirely on quality statistics VIDEO LAYER QoS
rather than on the bit rate statistics. In this case,
the process involves adding or subtracting one Video Layer QoS provides an unprecedented
SD channel at a time to output muxes that are level of control for a system operator or content
over or under the video quality requirement, provider, all the way from content origination to
respectively. the set-top box. Assuming the IP and MPEG-2
transport layers are intact, this capability opens
Although the statmux uses the entire mux up new possibilities for ensuring video quality,
bit rate to provide the highest quality, it is not available with previous digital or analog
possible to assess the effective available bit rate delivery solutions. Technically, Video Layer
according to the desired calibrated thresholds. If QoS means maintaining the pre-determined
a certain mux consistently has average grades quality requirements (QGa and QGb) through the
above the QGa, there may be some available communications delivery network, including
bandwidth for other services. The available bit sustainability through the various re-
rate can be computed by monitoring BAa and multiplexing, splicing, encryption, edge
BMb in real-time even when the statmux is statistical multiplexing, and VBR to CBR
selecting other quality grades. conversion for services such as Start Over and
SDV.
Guaranteed available mux bit rate = 38.8Mbps -
BAa, BMb)c In order to take advantage of this capability,
the statmux device from Step 6 needs to convey
the following information per service:
Average available bit rate (opportunistic
data) = 38.8 Mbps – Average ( Ba(t), 1. QG(t) – instantaneous quality per frame
Bb(t))c) 2. QGa – target average quality grade, for
example “96”
The statmux device can also calculate in 3. QGb – Minimum allowed quality grade,
advance what it would take to convert any of the for example “90”
streams to a CBR. In this case, the minimum 4. QCBR – target CBR rate in a multi-rate
CBR rate would be BMb under the CBR buffer CBR switched environment, the rate that

2008 NCTA Technical Papers - page 98

 will support QGa on average and QGb no and nPVR applications) is relatively
more than 1% of the time. straightforward, with the quality levels being
5. BQ – channel effective bit rate for VBR calculated in advance at the content origination
lineup allocation in real time site as discussed in Step 6.

The importance of sending QG(t) In some cases, due to content complexity

indications is crucial for maintaining ultimate and also the inherent nature of CBR, the
video quality. As noted above, objective video selected CBR rates may need to be higher than
quality measurement techniques compare two the standard SD and HD rates of 3.75 Mbps and
signals and it will be impossible to compare the 15Mbps, respectively. With respect to VOD,
target to the original stream at a receive site at since VOD assets are originally encoded in
the terminal of the network. Given the CBR, it is possible to insert the QG(t)
instantaneous quality per frame QG(t), it information into the stored stream for
becomes possible to keep the quality within the downstream edge statistical multiplexing.
target range, where it requires re-multiplexing,
by repeating Steps 4-6. EDGE STATMUX

AD INSERTION A state-of-the-art edge statistical

multiplexer can increase, by up to 50%, the
There are two main approaches for ensuring number of streams per QAM channel without
video quality of advertisements during ad quality degradation for VOD and SDV
insertion. The first approach involves pursuing applications.
the highest quality possible for the ad, even at
the expense of the underlying digital services In order to simultaneously maintain the
not containing ads at the same time. In this Video Layer QoS and optimize the bandwidth
approach, during the splicing period (the ad efficiency, it is important to also involve the
avail), the other streams are constrained to being Edge or Session Resource Managers
multiplexed at QGa and not higher. (ERM/SRM). A brute force method involves
simply allocating 15 SD “blocks” of 3.75 Mbps
The second approach involves equalizing each per QAM channel (or 3 HD “blocks” of 15
the ad quality to the underlying stream quality to Mbps each), i.e., tricking the system into
the extent possible. In this case, the ad is thinking each QAM channel has available up to
multiplexed at QGa and not higher, or at the 56.25 Mbps.
eventual average quality grade of the primary
stream. In any case, the ads should be processed A more intelligent design can allocate the
and stored on the ad server at the maximum service bandwidth according to each service’s
possible bit rate and quality level, providing effective bit rate (BQ) as suggested in Step 4,
downstream flexibility. A third approach is to and then load balancing the quality across the
provide the advertiser with QGa and QGb on a switched QAM channels, thereby guaranteeing
per asset or group of assets basis. Video Layer QoS. This method ensures the best
quality at any given bit rate for edge and
switched applications including VOD, SDV,
CBR FOR VOD AND SDV nPVR, Switched Unicast and addressable ad
insertion. The effective video quality will be
Applying Video Layer QoS to the significantly higher than today’s capped quality
conversion of VBR signals to CBR (for SDV at 3.75 Mbps. SDTV CBR and overall network

2008 NCTA Technical Papers - page 99

efficiency will be 50% better allowing 15 SD References:
VBR streams per edge QAM channel. 1. Objective Video Quality Assessment, by
Zhou Wang, Hamid R. Sheikh and Alan
INTERNET VIDEO C. Bovik , Department of Electrical and
Computer Engineering , The University
Using this approach in conjunction with the of Texas at Austin
standard IP QoS mechanism ensures constant 2. Survey of Objective Video Quality
video quality and Quality of Experience (QoE) Measurements, by Yubing Wang , EMC
for video services over the Internet and to Corporation Hopkinton, MA 01748, USA
mobile device. The IP QoS guaranteed bit rate 3. Z. Wang, A. C. Bovik, H. R. Sheikh and
should be set to BQ and the maximum required E. P. Simoncelli, "Image quality
bit rate for the service should be set to QCBR. assessment: From error visibility to
In some preliminary assessment, it is shown that structural similarity," IEEE
this approach not only provides the best video Transactions on Image Processing, vol.
quality at any bit rate, but it also consumes 25% 13, no. 4, pp. 600-612, Apr. 2004
less bandwidth and storage.

CONCLUSION NCTA, The Cable Show

May 18-20, 2008
The cable industry is in the midst of a New Orleans
dramatic transformation toward an increasingly
competitive and complex environment. Multiple Ron Gutman, CTO and Co-Founder
categories of digital television services will co- [email protected]
exist on a unified platform, including digital
broadcast, VOD, SDV, nPVR, and Internet Marc Tayer, SVP Marketing &
video, each of which will encompass standard Business Development
definition and high definition signals. [email protected]

This evolving comprehensive suite of

services and architectures must be presented in a Imagine Communications, Inc.
transparent and convenient manner to 2053 San Elijo Ave
consumers, who now have multiple choices for Cardiff-by-the-Sea, CA 92007
their service provider. In this new environment, (760) 230-0110
a key consideration and a competitive
differentiator is the ability to provide true Video
Layer QoS, combining control and optimal
video quality across all categories with the
utmost in bandwidth efficiency.

2008 NCTA Technical Papers - page 100

 WHAT TECHNOLOGY WILL WIN IN THE BATTLE TO DELIVER BROADBAND VIDEO
TO CUSTOMER DEVICES?

By Dave Lively, Cisco Systems

Marty Roberts, thePlatform

A media management and publishing system

Abstract can give the cable operator or programmer
Today, with consumers increasing their more control over their delivery options. Traffic
consumption of broadband video and with cable can by dynamically directed to files on different
operators and programmers continuing their CDNs without consumers experiencing any
entry into the online video space, the need to quality impacts. Policies may be applied to
understand content delivery options is media to automate the management and storage
paramount. Cable operators already have the of old or unpopular media files. As decisions to
network capacity for delivering the content. The switch to a new content delivery option arise, a
question is one of where to store the content and media management solution can ensure the
stream it from. The first issue is whether to transition is easy for production staff and
build an infrastructure using generic web seamless for viewers.
streaming and download servers, or to build a
content delivery network (CDN) to handle the This paper will look at the impacts on the
job. Cable programmers often have network for both downloading content and
relationships with commercial CDNs but they streaming content, as well as using CDN
may not be efficiently leveraging their internal technology versus P2P technology to actually
digital storage and streaming servers. deliver the content (whether it’s being streamed
“live,” or downloaded for future viewing).
Peer-to-peer (P2P) also presents another Media management systems may be applied to
option. Cable operators can build their own provide additional control over delivery
application that leverages P2P protocols. P2P policies. Virtualization of content, storage, and
eliminates the cost of storage and Gigabit applications can also be leveraged by cable
Ethernet ports required when building a CDN operators and programmers for delivery of
by pushing that cost to the individual users (the content and even web-based applications in the
service provider is essentially co-opting their future.
users’ PCs for the storage and streaming). But,
this method incurs additional costs for more INTRODUCTION
upstream bandwidth, and potentially dealing
with network congestion. Plus, what's the More and more, consumers are looking to the
incentive for users to "donate" a portion of their Internet to get their content. And while user-
bandwidth and computing and storage generated / contributed content sites like
resources on their PC? Hybrid models also YouTube.com continue to dominate the online
exist, allowing operators to potentially leverage video market, the Internet is rapidly becoming a
the best aspects of all technologies. viable means to distribute premium studio
content as well.

2008 NCTA Technical Papers - page 13

 Top U.S. Online Video Properties by Videos Viewed Jan. 2008i

Total U.S. – Home/Work/University Locations

Videos Share (%) of

Property (000) Videos

Total Internet 9,814,010 100.0%
Google Sites 3,363,335 34.3%
Fox Interactive Media 584,132 6.0%
Yahoo! Sites 315,001 3.2%
Microsoft Sites 199,288 2.0%
Viacom Digital 197,737 2.0%
AOL LLC 118,033 1.2%
Disney Online 95,041 1.0%
Time Warner - Excl. AOL 85,467 0.9%
ESPN 81,402 0.8%
ABC.COM 49,017 0.5%
Figure 1: Top U.S. Online Video Properties by Videos Viewed Jan. 2008

Top U.S. Online Video Properties by Unique Viewers Jan 2008ii

Total U.S. – Home/Work/University Locations

Unique Viewers Average Minutes
Property (000) per Viewer
Total Internet 139,521 206.3
Google Sites 80,056 109.9
Fox Interactive Media 53,913 11.7
Yahoo! Sites 36,362 18.0
AOL LLC 21,859 7.4
Viacom Digital 21,690 33.0
Microsoft Sites 20,842 30.0
Time Warner - Excl. AOL 13,914 18.2
Disney Online 13,005 8.9
ESPN 8,798 15.9
Apple Inc. 8,743 21.2
Figure 2: Top U.S. Online Video Properties by Unique Viewers Jan. 2008

Major studios and content owners are

looking to the web as a new outlet and method Today, “Internet Video” is largely a
to monetize their content. With more computer-only phenomenon. But we are rapidly
consumers turning to the Internet as a source for approaching a time when consumers will have a
video, the increasing load will force both the choice on how the will receive their “television
content owners and service providers to and movie” content, from whom, and on what
examine new technologies to handle the device. Digital Media Adapters (DMAs) and
distribution effectively while maintaining a high Digital Media Servers (DMSs) are available
quality and reliable consumer experience. from major consumer electronics vendors today,

2008 NCTA Technical Papers - page 14

with new models that drive consumer trends of those titles has to be intuitive and fast. And
being announced every month. When the personal connection between a person and
consumers have a choice of getting premium “their shows” requires that the service is always
content through their cable operators, as well as available.
a variety of traditional and online competitors,
the providers with the most accommodating At the same time, the growing popularity of
overall solution will come out on top. Given digital video recorders (DVRs) and portable
their incumbent position with both television media players (such as the iPod) is causing
and broadband Internet service, cable operators consumers to demand a much more personalized
are in a prime position to be that provider. and portable video experience. They want to
view content highly relevant to them, on their
CONSUMER NEEDS terms. They want it on whatever device they
happen to be using at that particular moment,
Hundreds of television channels, thousands and they want the content to be available all the
of websites, millions of videos, all a few clicks time.
away. This is an era where there are very few
barriers to making content available to Engagement
consumers, leading to an unprecedented amount
of entertainment options from a rapidly growing According to a 2007 viewer study, nearly 60
number of sources. This has great promise, percent of adult consumers surveyed stated that
because of the high likelihood that something they watch online videoiii. More businesses are
very tailored to every consumer’s taste is recognizing that broadband video can help them
available out there somewhere. target audiences, generate real revenues, and
gain creative control of the user experience.
This simple consumer proposition places a Broadband video supports brands in a way that
heavy burden on content providers. It requires a breathes vitality into and can extend the life of a
balance between providing a breadth of content media or service provider business. It adds
choices and enabling easy discovery of that flavor, perspective, and additional information
content by the consumer. Breadth is to existing pages, increasing audience
increasingly measured in the hundreds of engagement. Figure 1 shows audience
thousands or even millions of titles. Discovery engagement for a top destination site on the web
.

2008 NCTA Technical Papers - page 15

 50.00%
45.00%
40.00%
35.00%
30.00%
25.00%
20.00%
15.00%
10.00%
5.00%
0.00%
Bounce <1 1-5 5-10 10-30 30-60 > 60
minute minutes minutes mintues minutes minutes

Engaged Users Disengaged Users

Figure 3: Sample Engagement Data for an Online Video Site

Relevancy recommendations that give consumers what

they want - the convenience of low-touch
Programming has a valuable place in the programming with the benefit of highly
entertainment universe. On TV, a network personalized, relevant entertainment. This
orders its shows to naturally lead the viewer approach allows everyone to win--consumers
from one to the next. This provides convenience get a better experience and content owners
and a sense of flow that can make the increase viewership, and ultimately, service
experience more enjoyable with less work. The providers can improve the experience and value
challenge in front of us is how to apply the of their service. But the consumer experience
notion of programming to a much broader set of doesn’t stop with content sourced from the
content while presenting a very personalized network. Increasingly, consumers are
and relevant experience. Clearly, the traditional contributing content as well. It might be highly
model of people making editorial decisions that personalized content meant for family and
result in a broadcast schedule begins to break friends such as video from a family event, or it
down as the model moves increasingly towards could be quickly captured video destined for a
a 1:1 engagement with consumers and includes larger audience. Consumers want the user-
a much larger body of content. Technology will generated content experience to be seamlessly
have to play a bigger part in determining what is integrated with the television and movie
presented to a consumer if the promise of programming experience. This increasingly
personalized programming is to be realized. The complex mash-up of personal content, friend
payoff for figuring this out is a better consumer and family content, professional programming
experience, driven by the presentation of more from multiple content owners, and new
relevant content. independent content optimized for web
distribution places even more demands on the
Layering better-informed programming cable operators and content owners looking to
with technology designed to improve and distribute content.
measure the effectiveness of the content guide,
the results are very smart, proactive

2008 NCTA Technical Papers - page 16

 CONTENT OWNER NEEDS Traditionally content providers have relied
on one or even two CDN suppliers under multi-
At the same time that consumer year contracts, which limits their ability to take
requirements are increasing, content owners are advantage of recent changes in the CDN market.
asking for more. Media companies have an But, with over a dozen global CDN suppliers
unprecedented ability to reach their audience now vying for business in the U.S. alone, there’s
through multiple outlets. This includes cable, no longer any reason for content providers to
Video on Demand, Digital Video Recording, tolerate onerous contractual terms with built-in
DVD, the Internet, and mobile devices. cost escalators and other unnecessary cost
Aggregators that have the ability to distribute burdens. Fortunately, the costs of implementing
content across platforms while maintaining a dynamic control over CDN services are
consistent, high quality experience will have an miniscule compared to the potential savings. By
advantage. turning to highly automated operations support
tools and services, content distributors of all
Branding and Cross-Platform Promotion types and sizes can achieve cost breaks across
the CDN domain, avoiding over-priced services
In a world that isn’t necessarily anchored by and remaining flexible enough to take advantage
a channel on the TV, providing branding of advances in CDN technology wherever they
opportunities around the content becomes occur.
critical. Associating the show with the provider
allows content owners to leverage their brand Unnecessary Cost Drivers
investments and connection to an audience. As
soon as a content owner feels comfortable with CDN providers typically set terms that
their brand attribution in one medium, cross- require customers to pay minimum monthly fees
platform promotion becomes a requirement. for storage and distribution, covering all traffic
Leading a consumer from promotional clips on volume up to a certain limit. After exceeding
their mobile phone to the full show on the web that limit, the charges are then based on
or a video on demand (VOD) system and finally incremental storage and distribution volume,
to a linear channel to find other, similar shows which can add up quickly. There are ways to
will become the norm. shift that traffic to cheaper distribution channels,
and there are typically no contractual barriers to
Content owners’ final requirement is to doing so. What can be termed “success-based”
decrease their distribution costs. When each cost escalators also come into play with delivery
incremental audience member costs the content of advertising with content over CDNs. As users
provider more, aggregators that can reduce this access more ad content in the service stream,
expense become very attractive partners. content providers have to pay more to CDNs,
sometimes upwards of $7 per thousand views,
Distribution Costs which translates to a large share of the $14-$25iv
cost per thousand (CPM) rate content suppliers
Service and content providers concerned typically charge advertisers. Some media
over costs of distribution have recourse to new companies are reducing CDN costs by
management capabilities that can significantly leveraging management tools that dynamically
lower the expense of doing business with execute their CDN requirements across multiple
content delivery network (CDN) suppliers. providers. This is often done by coordinating
use of CDN options with in-house storage
capabilities.

2008 NCTA Technical Papers - page 17

 This allows content providers to avoid
Cost-Cutting Strategies paying inflated rates for files already duplicated
and stored in the network that are on the
There are many ways media management downside of their usage curves. By moving such
infrastructure can be used to help bring down files out of the high-cost CDN storage
CDN costs. Some approaches coordinate environment, the content provider pays storage
external CDN resources with sizeable in-house rates that square with current usage trends.
infrastructures. Other approaches rely more
heavily on external support. In all cases, the The need to ensure that quality parameters on
content provider must be in a position to each content stream meet end user requirements
manage CDN fulfillment in a streamlined, also has an impact on the CDN selection
dashboard-based environment with mechanisms process. As content providers enable flows that
in place to seamlessly switch content flows from support full screen viewing, the file storage and
one CDN to another. distribution volumes escalate. The content
provider must be responsive to situations where
Companies utilizing their own digital media high bandwidth requests are pushing traffic
storage systems need to actively manage files by volume over a particular CDN cap faster than
pulling them from the CDN when those files are expected.
not in use. This is especially relevant for those
companies that have large content libraries or Along with sophisticated operations
already have invested in storage devices. functionalities, successful management of CDN
services requires a savvy approach to
Some firms may also want to direct contracting services. While long-term contracts
consumer video requests in non-peak hours to might appear to offer aggressive rates, CDN
in-house distribution servers, only using CDNs providers know that by locking customers in
for peak traffic. Policies that govern how traffic they will make out very well as traffic and
is allocated between internal resources and storage demand exceeds caps. Even if a content
outside CDNs can be very simple, for example provider has recourse to capabilities discussed
setting a weighted control on the flow of user above that can use multi-CDN access to avoid
requests for content where 60 percent may be over-cap costs, it’s prudent to use those
directed to the internal servers and 40 percent to management capabilities to facilitate working in
CDNs. In some cases, the content provider an environment where one-year contracts are
might build a system to monitor flows and available.
manage traffic so that, if traffic hits a certain
threshold, it is switched over to outside CDNs. Short-term contracts will allow a broadband
video business to capitalize on lower bandwidth
Achieving such capabilities requires highly costs and the latest delivery technologies. In
sophisticated management software and fact, broadband video providers with low
infrastructure controls capable of switching volumes can gain leverage in their negotiations
traffic across multiple CDNs without by looking at very short-term contracts of just a
disruptions to end users. Whether or not a few months duration. The company can then
company has internal storage and delivery bargain for reduced fees in exchange for
resources, it will want to use CDN switching agreeing to a longer-term contract.
resources to direct traffic to multiple CDNs to
ensure contractual caps are never surpassed. The key for content providers is to configure
their systems to optimize traffic distribution,

2008 NCTA Technical Papers - page 18

taking into account such factors as the extent to Going halfway is not a winning proposition.
which in-house resources can be leveraged and Content providers have to commit to getting the
how the provider can work within existing CDN content to all of the online destinations (i.e.
contract terms to improve efficiency. syndication) that make sense: that's distribution.
It will entail headaches. It will involve dealing
Implementing architecture that supports ever- with multiple video formats, different policies,
changing dynamics of the business, including and different advertising models. To get content
expansions of fields covered by the system, out there and to monetize it is a lot of work.
extensions of metadata categories and
application rules, types of security to be applied This strategy involves thinking approaching
to various content categories and ongoing online video holistically. ABC, CBS, NBC, Fox
variations in end user pricing and access rules is have all built their own Websites to host their
also a critical component. television content, but they've also invested in
distribution.
Selecting Content Libraries
Previously, broadcasters have seen the
In addition to making premium content Internet as supplemental to the TV business; it's
available online, television networks and cable something they've used as a promotional tool.
programmers have mined their stockpiles of Online, broadcasters can go out and find an
content, knowing that supporting video and audience rather than rely on an audience to find
audio clips make newer offerings more them. Beyond promotional deals, it's that long
interesting to their audience. Very often tail that allows broadcasters to connect with an
businesses have a lot of content in their archives audience.
that could pull in a very large online viewership,
either by itself, or when used as supporting Here's an example: If a consumer watches
material for related content. Heroes for the first time on TV and really wants
to catch up on what she missed, she can go
There are several major content sources online and watch back episodes. In essence that
from which companies can harvest media that library of long-tail content becomes a
paves the way for broadband video business destination. This is a nice driver for the market:
success. The best way to maximize content is to Broadcasters can find out where there is demand
get it into a digital library and add value with and do some real-time determinations of what
commerce and advertising solutions that use IP kind of content is popular.
based-communication and web-based
presentation. Broadcasters, cable channels and operators
need to create a destination to catch all of these
Multiple Outlets/Distribution viewers. But they also have to be willing to
follow the consumer. If a consumer can't get
As big television networks and broadcasters what they want through one outlet, then they
move their content online, one big challenge will go elsewhere because it doesn't end with
that presents itself is that unless you have having just one destination to catch these kinds
incredible brand awareness, exclusively of viewers. Brands have to reach different kinds
focusing on building your own destination site of people across the Internet, with different
is only one small part of building a successful preferences on where they want to go to get
online strategy. their content and how they want to consume it.

2008 NCTA Technical Papers - page 19

 If a media company cuts the right Streaming
distribution deals, its shows can land in the
places where consumers are landing natively. Today’s protocols for streaming video
They can reach customers in a way that's not depend on a fairly reliable transport network.
quite the same as in the operator environment The video is sent at a constant rate, and any
where walled gardens are the norm. This, in variance in delay (jitter) in the stream is
fact, flies in the face of a walled garden. compensated for by buffering at the client. This
results in delayed startup of streams while
As more network TV content moves online, buffering, as well as delays when switching
there will be increased pressure to open up the from one stream to another (which requires
walled garden a little bit and allow content to buffering of the new stream).
flow to more than one place. Ultimately, that
helps create the seamless experience the Streaming via traditional IP streaming
consumer is after. protocols presents a problem for video delivery
across the “generic” uncontrolled bandwidth of
Success will mean interoperability across the Internet, as it is subject to congestion and
platforms and distribution of content far and choke points across the network. The solution
wide. Standards are now emerging that will ease for this is to stream the video from a source as
interpretability between different platforms and close to the subscriber as possible. By
enable seamless and even greater multi-platform eliminating as many potential points of
delivery of content. congestion from the network as possible the
video can often be streamed at a higher rate with
better quality for the subscriber. A second
CONTENT DELIVERY METHODS improvement for delivering video via streaming
protocols is to apply Quality of Service (QoS) to
There are two primary methods for getting the stream, giving the video packets higher
content to consumers: the consumer downloads priority in the event of network congestion.
the content for either immediate or later However, this is typically only viable on a
viewing, or streams it for viewing “live”. single, controlled network, not across the
Streaming is generally defined as content being Internet in general. In fact, this is typically the
delivered to the subscriber “just in time” for way that cable operators transport both
viewing, typically without the ability for the broadcast and on demand video across their
user to “record” or keep a copy of the content. regional networks.
Downloading is generally defined as
transferring the content to the subscriber’s Streaming protocols typically limit the
device and then viewed locally. However, the bandwidth to the actual stream rate. On one
boundaries between these two methods are hand, this can prevent the client from taking
blurring as new technologies come on to the advantage of “pre-buffering” the video when
market. Protocols typically used for extra bandwidth is available. For example, by
downloading can be used to simulate a enabling fast start capabilities native to some
streaming experience, and vice versa. streaming protocols. On the other hand,
streaming protocols are very efficient in that
they only send packets if the client is viewing
them. For example, if a user starts watching a
10-minute video but decides to stop one minute

2008 NCTA Technical Papers - page 20

in, only the first minute of the video is utilizing burst transfers of the video at faster
transferred. than stream rate for quicker delivery. Some
methods also provide the efficiency of
Downloading streaming protocols by only transferring just the
video that is being watched in addition to a
Downloading content gives the user more small buffer.
control over when and where they want to view
the video, as once it’s downloaded to the client Peer to Peer (P2P)
device, no network connectivity is needed to
view the video. This allows for a very high P2P protocols can be used to enable
quality video experience regardless of available download type services or streaming type
network bandwidth or connectivity, assuming services. Where both traditional downloading
the subscriber has downloaded the video in and streaming protocols work get their content
advance. Digital Rights Management (DRM) in a fairly linear fashion (start at the beginning,
can be used to limit transfer of the content, the keep going until the end) from a single source,
number of times the content can be viewed, the P2P clients can get content simultaneously from
viewing window, etc. Downloading video via a large number of sources, and do so in a non-
broadband is most often done using the HTTP linear fashion (getting the last part of the file
protocol, but can be done via proprietary first for example). However, while P2P clients
protocols with dedicated clients as well. can source the data from multiple locations, it
Dedicated video clients also provide more still must all traverse the same broadband access
capabilities for the subscriber by managing link. Thus, P2P will have the same broadband
where videos are stored on the device, download bandwidth issues as traditional
preventing screen captures of the content, streaming and downloading protocols will. P2P
providing a common navigation engine, or protocols have routing metrics to determine the
allowing users to set up subscriptions to nearest, highest bandwidth, most reliable
download multiple videos. sources to source content from, increasing the
subscriber’s utilization of the access network.
Progressive download is the ability to start However, because there are not commonly
viewing a downloaded video as soon as there is accepted standards for P2P networking
enough video “in the buffer” to continually play protocols, they are not integrated into popular
the asset given the transfer speeds that are seen browsers, and typically require custom clients.
during the initial buffering. Progressively
downloaded content allows the subscriber to P2P protocols have a great advantage for
start viewing the video as soon as possible while content owners in that they don’t require any
simultaneously saving the video for later streaming or downloading infrastructure.
viewing. Within the progressive download Content owners need only to seed the content
model, downloading can simulate a streaming once into the P2P network. As more clients
experience, but using downloading protocols download that content from the initial peers,
instead of streaming protocols. they in turn make that content available to other
peers. Thus, the content owner’s individual
Whole asset downloads, and proprietary subscribers are using their own storage,
methods of downloading small “chunks” of the streaming resources (their PC) and broadband
video are common methods utilized today. network connection bandwidth to distribute the
These methods provide the benefit of using content for them, all at no cost to the content
download-type protocol architectures while owner. Since P2P protocols work best when

2008 NCTA Technical Papers - page 21

there are a large number of peers in the network could publish their content to a select number of
with the content a client is asking to download, service providers they peer with at a content
it tends to work best for the most popular level, who would in turn distribute that content
content, as that is likely to reside on the greatest as needed either to other service providers for
number of peers. delivery to their customers, or direct to
consumers who are using service providers that
P2P clients have downsides as well. For don’t have content peering relationships.
long tail or niche content, the number of peers
with the content might be very low, resulting in
slow download performance and a poor Content Delivery Networks
experience for the recipient. Since most P2P
networks rely on individual users who may or To help scale the delivery of content across
may not be online at any given time, niche the Internet, CDN providers have built
content may not even be available to requestors infrastructures that help virtualize that content
who wish to view it. Most P2P networks are across the network. Conceptually, CDNs work
also not managed by any central source or by ingesting the content from a source (such as
entity, and thus may end up being less reliable a content owner’s website) into a network of
overall as they rely on the individual users to intelligent caches distributed throughout the
each do their part. network. As subscribers request to view
content, a copy of that content is stored in a
P2P relies on individual users to source the cache closest to that subscriber. When the
content over their asymmetric broadband second subscriber requests to view that same
connections. While download speeds can be piece of content, the request is redirected to the
very fast, the upload bandwidth is typically a local cache, with no need to go back to the
much lower speed, and more subject to original source. As local caches “fill up” with
congestion. A P2P user can easily consume a content requested from subscribers, the least
large percentage of both the upstream and popular content is purged to make room for
downstream available bandwidth of a given more popular content. CDN caches can be
network segment, depending on the popularity placed in a tiered hierarchy, allowing for content
of content they host, and the volume of content population to match the interests of its local
they seek to obtain. The effects of popular P2P subscribers. This demand-based method keeps
applications that are used to distribute the most popular content closest to the users
copyrighted content is the best example requesting it, resulting in the best performance
recognized by operators today. for those subscribers. Niche content, however,
is always available from the source. CDN
The use of P2P protocols isn’t limited just to caches simultaneously serve the content to the
clients trying to use the Internet to distribute requestor while the content is populating the
content. Service providers could leverage P2P CDN cache’s local storage repository.
for distribution of content within their network,
as well as potentially leveraging P2P protocols But as described in the previous section, the
to “peer” with other service providers for more popular the content is, the more it is
content, in a similar way as they peer for downloaded and the more the content owner is
Internet packet transport today. Instead of charged for distribution. This is in stark
publishing their content to 3rd party CDN contrast to P2P, which performs best (and least
providers (described in the next section), or to expensively) for the most popular content.
client-based P2P networks, content owners However, also in contrast to P2P, CDNs are

2008 NCTA Technical Papers - page 22

managed networks and can provide service level Role of the Cable Operator in Online Video
guarantees to content owners, and deliver all
content regardless of how popular it is, from Today, when it comes to content destined for
within the operator’s own network. The most the television, the cable operator acts as the
popular content is more widely distributed primary, and often only, aggregator and
throughout the network, and thus typically distributor of content from multiple content
resides close to the subscriber, minimizing the owners to the operator’s subscribers.
number of “hops” or links through the network Consumers navigate linear broadcast content via
the content must traverse to reach the a program guide, with a menu-based navigation
subscriber. The closer to the subscriber, the portal for accessing popular movies, TV shows,
better - proximity minimizes the potential for and niche content on demand. But when it
congestion and enhances the overall customer comes to broadband video, many cable
experience. operators don’t have a presence at all. Today’s
business models are still developing around
Hybrid Delivery Options what and how consumers will pay for online
video. But as the comScore numbers in figure 1
Some companies are starting to look at show, consumers aren’t waiting for their cable
hybrid P2P and CDN delivery options. A client company to figure out. Cable operators have
would first look to see if the content is available the opportunity to play the same aggregation
from peers, and if not, start sourcing that content and distribution role for online content as they
from a CDN provider. Thus, the most popular do for traditional television content, and in some
and expensive content traditionally provided by cases, controlled delivery of the content may
a CDN would be delivered via the “free” P2P almost pay for itself.
network instead.

Online Video Trends (Jul 07-Jan 08)

(in MMs)

12,000

10,000

8,000

Total Videos Viewed

6,000
Videos Viewed - Google Sites

4,000

2,000

0
Jul. 2007 Sept. 2007 Nov. 2007 Dec. 2007* Jan. 2008

*Single heaviest month for online video consumption since comScore initiated its tracking service

Figure 4: Online Video Trending July 2007-Jan 2008

2008 NCTA Technical Papers - page 23

 for building a CDN than providers which need
Impact of Over the Top Video Delivery on to lease those facilities. In addition, this
Cable Operators infrastructure is already being used for the
transport of other services that are largely
Subscribers watching online video “over the funding its build-out; such as traditional cable
top (OTT),” or using the cable operators TV, voice services, business services, and even
bandwidth for video not distributed by the cable OTT online video itself.
operator, is effecting the operator both
technically and financially. The most obvious is By leveraging CDN technology within an
the substantial traffic growth existing infrastructure, cable operators can
Cable operators are already transporting all of cache popular content at the edge and eliminate
the online video that their subscribers are duplication of bandwidth across the network to
watching – they’re just not getting any revenue help alleviate some of the costs already
from it aside from the revenue they are incurred. Because the operator controls the
receiving for providing broadband Internet infrastructure, cable operators can also leverage
connectivity. For popular content, the cable QoS capabilities to prioritize video traffic in the
operator is probably transporting the same exact event of network congestion, providing a better
content countless times across their backbones. experience for consumers. With fewer potential
In addition to the hit that cable operator congestion points and the ability to prioritize
backbones are taking from a bandwidth video streams, cable operators can delivery
perspective, operators are also facing a service higher quality, higher bit-rate video to the
substitution challenge. This is happening today consumer, further enhancing the experience vs.
in the voice market with OTT providers such as OTT delivery, which contributes to continued
Vonage and Skype. Once premium video is subscriber loyalty and brand awareness.
available online, and viewable on the television
set via retail DMAs, subscribers can also start Cross-Platform Service Capability
shifting their content spend from the cable
operator to other providers. A subscriber shift is Beyond the advantages of more efficient
starting to happen today through such devices delivery of content, cable operators can add
and services as Apple TV and iTunes, Vudu, significant value in the services they can bundle
and Microsoft Xbox LIVE Marketplace. And as together and offer consumers. As cable
services continue to shift, so too will the networks transition to all-digital with set top
advertising revenue associated with those boxes, digital video recorders have become
services. commonplace in nearly every cable household.
A common platform establishes a foundation for
Cable Operators as Online Content Distribution cable operators to provide the same content to
Partners multiple screens for the consumer. By
partnering with the content owners, cable
By becoming an active participant in the operators can provide customers who are
distribution of online video, cable operators can subscribers to premium tiers access to that same
accomplish multiple goals while providing more content online. Consumers could access that
advantages to both the content owner and content through a portal that both maintains
consumer at lower cost than traditional CDN branding for the content owner and gives the
providers. Cable operators already own their consumer a single destination for all things
regional network and broadband infrastructure, video-related.
and have fundamentally lower cost structures

2008 NCTA Technical Papers - page 24

 across multiple content owners, premium and
As cable operators begin to deliver content free content, and personal content the cable
across platforms with a single infrastructure, operator can become the primary video
they can also start working with advertisers on experience provider for their consumers.
cross-platform advertising capabilities. Beyond a single portal for video content, the
Advertising campaigns can span from television cable operator has the opportunity to integrate
content to long-form online advertising. management of all video, content, and
Targeted banner and bumper ads can communication services for the subscriber into a
accompany online video, and different ads can single experience, allowing the subscriber to
be shown each time the consumer watches the access any service regardless of the device
content. By integrating online video with they’re using.
communication services such as email and
VoIP, cable operators can enhance the video
experience by becoming involved in the NEED FOR MEDIA MANAGEMENT AND
subscriber’s social experience. PUBLISHING

Capitalizing on the Online Portal Once a cable operator/content provider has

their distribution strategy worked out, the next
Opportunities for additional advertising or step is figuring out the media management.
click-through revenue rise once mainstream, Everyone in the value chain benefits from a
premium content is presented along with niche, system that can provide the control and
long tail content and user-generated content. automation necessary as content is provided
The goal for cable operators is to become both across platforms. Figure 5 provides an overall
the distribution network and the jump off point sense of the data flow involved in publishing
for all things video and content related. By online video.
integrating all of a consumer’s content needs,

2008 NCTA Technical Papers - page 25

 Figure 5: Online Video Workflow

Applying Business Policies/Control Examples include restricting media so it

can be accessed only on certain dates or
Control leads to the application of business in certain geographies.
policies for each piece of content. A media
management system must be able to reflect the o End-user restrictions. Control user
carriage deal including geographic restrictions, access via an integration with existing
air dates, end-user restrictions, digital rights “single sign-on” authentication
management, pricing models and advertising application. These controls enable the
policies. These set parameters that control when system to perform all the response tasks
content is available, who can access it, when it (prompting the user for ID and password
expires, and what delivery methods are allowed. if necessary), and either generating a
As these policies are applied to a video, they license directly, or using a management
must be enforced in every medium including system to pass the license information
VOD, the web and mobile. and grant access.

Common types of broadband video controls o Pricing policies. Help monetize video
include: content by supporting pricing schemes
such as free trial periods, pay-per-view,
o Content restrictions. Scope content and pay-per-download.
usage to reflect business requirements.

2008 NCTA Technical Papers - page 26

 o Advertising policies. Ensure a tight determining the best value: quality of service,
integration with advertising campaign the number of sites the content is uploaded to,
management systems to target and track and what reports and alerts are offered.
ads.
Applying policies that govern how traffic is
Automation allocated between internal resources and outside
CDNs, or directing traffic to multiple CDNs to
The only solution to complex content ensure that contractual caps are never surpassed
distribution strategies is automation. A media are a couple ways to reduce CND costs. Another
management system should gracefully meet approach is to move older files out of high-cost
various metadata schema and video format CDN storage to an in-house storage as the
requirements. The heavy-lifting tasks of audience moves on to newer content.
transcoding, file transfers and encryption are
best accomplished programmatically, Determining Formats
eliminating the need for personnel to manually
start each chore. The next step after content ingest is
formatting content. The formats needed depend
Dynamic Entitlement on where the media is being sent, as different
media have disparate technical requirements.
The final mission of a media management This often means creating more compressed
system is to enable the monetization of content. versions of a video for viewing within a browser
Entitlements should be dynamic, allowing or for faster download for viewers without
operators to combine consumer segments with broadband connections. Some examples include
pricing models and content restrictions. media companies that want viewers to be able to
Advertising needs to merge the emotional watch video as it downloads or content
connection of the television with the targeting of providers who wish to support both Windows
the web. An open approach to trying new and Mac players.
advertising campaign management systems and
relentless testing will result in consumers In addition to selecting the appropriate
receiving more relevant ads that are actually formats for a content or service provider’s own
appreciated for their educational value. site(s), there are also additional formats required
for syndicating media. For example, when
Choosing a Content Delivery Network syndicating to a mobile carrier, files must be
provides in formats that work on their devices.
If content owners have a large library of Considerations in selecting format can be
video they may choose to select a CDN to host impacted by the target audience, the video/audio
their files once they are published. The selection quality required, file security and the content
process isn’t one that can be addressed briefly. being posting to live or on-demand media.
There are a lot of factors to consider in

2008 NCTA Technical Papers - page 27

 Figure 6: Common Broadband Video Formats

Most videos files have at least two types of

file formats, the container and the codec. The
video file container holds data like audio and
video, which have been compressed using
codecs. Codecs compress files so they take up
less storage space on a computer and can more
quickly be transmitted across the Internet.
Figure 6 lists some common container formats
for broadband video. Note that there may be
exceptions— some formats will play on certain
platforms only if a special plug-in is installed, or
some videos will play on a device only if the
correct codecs are in the container file.

2008 NCTA Technical Papers - page 28

 SUMMARY But managing all of this content from so
More, more, more: More content choices, many different sources to so many different
more devices to view that content on, and more destinations requires a sophisticated system for
ways to find the content. Consumers only have media management to help automate and scale
so much time during the day to view video the process. Additionally, it provides
content, and want that time to be spent viewing mechanism for extending content and the
relevant content, not figuring out what to watch, consumer experience to broadband with the
where to find it, and how to get it on the device flexibility to assign specific policies and
they want. Content owners want more ways to business rules applicable to operators and
monetize their content. They want to sell it programmers, and most importantly,
direct to the consumer when they can, but also accomplishes this without relinquishing control
leverage the most popular distribution outlets to over specific media objects. As the market
make sure their content is easily accessible to continues to evolve and grow, the management
consumers. of online media will continue to grow in
complexity with sheer volume and types of
This means streaming the content live to content available. As new formats emerge and
consumers, allowing them to download it for the delivery mechanism, type of content and
either immediate or later viewing, and giving business rules change, media management
them the flexibility to move that content around systems need to serve as an extension of the
and share the experience with friends and programmer/operator product teams and
family. Cable operators are one of the primary dependably deliver video that meets the
distribution outlets for television content today, expectations of viewers accustomed to high
but certainly not the only one. The same quality video over the TV.
television show is available initially via
broadcast, and then online via the programmer’s
website, online through retailers such as iTunes
and Amazon.com, on demand through cable REFERENCES
operator’s VoD platform, and ultimately on
DVD and syndication as well. i
comScore Video Metrix, Press Release, March
2008
ii
comScore Video Metrix, Press Release, March
The Internet gives both content owners and
2008
consumers more avenues for receiving and iii
Online Video: A New Local Advertising
viewing that content. Cable operators have the Paradigm, The Kelsey Group, Inc., 2007
opportunity to continue to play their current TV iv
Based on today’s average CDN industry pricing
distribution role for online video as well. models, internal company resources
Multiple partnership and engagement models
exist, ranging from acting as a wholesale CDN
provider up to and including being the
aggregated retail storefront for content. And by
integrating the online video experience with the
current bundle of digital video, voice, and
broadband data cable operators can further
enhance their value proposition for consumers.

2008 NCTA Technical Papers - page 29

ISBN 0‐940272‐01‐6; 0‐940272‐08‐3; 0‐940272‐10‐5; 0‐940272‐11‐3; 0‐940272‐12‐1; 0‐940272‐14‐8; 0‐
940272‐15‐6; 0‐940272‐16‐4; 0‐940272‐18‐0; 0‐940272‐19‐9; 0‐940272‐20‐2; 0‐940272‐21‐0; 0‐940272‐
22‐9; 0‐940272‐23‐7; 0‐940272‐24‐5; 0‐940272‐25‐3; 0‐940272‐26‐1; 0‐940272‐27‐X; 0‐940272‐28‐8; 0‐
940272‐29‐6; 0‐940272‐32‐6; 0‐940272‐33‐4; 0‐940272‐34‐2; 0‐940272‐35‐0; 0‐940272‐36‐9; 0‐940272‐
37‐7; 0‐940272‐38‐5; 0‐940272‐39‐3; 0‐940272‐40‐7; 0‐940272‐41‐5; 0‐940272‐42‐3; 0‐940272‐43‐1; 0‐
940272‐44‐X; 0‐940272‐45‐8; 0‐940272‐46‐6; 0‐940272‐47‐4; 0‐940272‐48‐2; 0‐940272‐49‐0; 0‐940272‐
50‐4; 0‐940272‐51‐2; 0‐940272‐52‐0; 0‐940272‐53‐9; 0‐940272‐54‐7 

© 2015 National Cable and Telecommunications Association. All Rights Reserved.