5.

1 Sompling Theorem 261

SAMPLINGAND g() |G)

5 ANALOGTODIGITAL
CONVERSION (a) (b)
-2TB
0
A

2rB

ö,, (0)

Lowpass |G)
filter

A
s briefiydiscussed in Chapter 1, analog signals can be digitized through sampling and aantit
This analog-to-digital (A/D) conversion sets the foundation of modern digital commui
tems. In the A/D converter, the sampling rate must be large enough to permit the analog sienal t
AANAA 0
2rB

be reconstructed from the samples with sufficient accuracy. The sampling theorem, which is the hais (d) (e)
determining the proper (lossless) sampling rate for a given signal, has played a huge role in signal procesinp
communication theory, and A/D circuit design. Figure 5.1 Sampled signal and its Fourier spectra.

5.1 SAMPLING THEOREM Because the impulse train ör, (1) is a periodicsignal of period Ts. it can be expressed as an exponential
Fourier series, already found in Example 2.16 as
We first show that a signal g(t) whose spectrum is band-limited to B Hz,
that is,
27
Gð) = 0, ST, (1) =T T
= 2rf (5.2)
for |f| > B n=-o

can be reconstructed exactly (without any error) from its discrete time Therefore. g) = g)oT, (1)
samples taken uniformly at a ate o
R samples per second. The condition is that R > 2B. In other
perfect signal recovery is f, = 2B Hz.
words,. the minimum sampling frequency 10 1
cht(s1 (A)in Specha
by (5.3)
To prove the sampling theorem, consider a T n=-o
B Hz (Fig. 5.1b)." For signal g() (Fig. 5.la) whose spectrum is band bm
convenience,
rate of f, Hz means that we take f, spectra are shown functions of f as well as of o. Sampling g) at2
uniform per second. This uniform sampling can be accomplished
by multiplying g() by an impulsetrain ör, (t)samples
10 nnd G), the Fourier transform of o(). we take the Fourier transfornm of the summation in Eq. (5.3).
every T, seconds, where T, 1/fs. This of Fig. 5.Jc,
consisting of unit impulses repeating pele ne requency-shifting property, the transform of the nth term is shifted by nfs. Therefore.
sampled
signal consists of impulses spaced everyresults in the sampled signal g() shown in Fig.
T, seconds 5.1d, Thelocated
(the sampling interval). The nth impulse, at
t= nT,, has a strength g(nT,) which is the value of g(t) at = nT,. Thus, the relationship betweenthe
sampled signal g() and the original analog signal g(t) is G) = T Gf -nf) (5.4)
S n-00

(51) This means that

|3() =g()8r, () =8nT,)8(t nT)| Period f, = 1/T, theHz,spectrum
as
Gf) consists of G), scaled by a constant 1/Is, repeating periodically with
n
After untform shownthat in Fig. 5.le.
Can g(t) be sampling generates a set of signal samples (g(kT,)), the vital question becomes:
80), reconstructed
equivalently in the
from g() without any loss or distortion? If we are to reconstruct g() trom
* The spectrum
G) in Fig. 5.lb is shown as real, for Fig. 5.1, perfect frequency domain we should be able to recover Gð) from G(). Graphically from
260
convenience. Our arguments are valid for complex OU7 recovery. ispossible if there is no overlap among the replicas in G(). Figure 5.le clearly
 CONVERSION
262 SAMPLING AND ANALOGTO-DIGITAL 5.1 Sampling Theoten 26:
shows that this requires JdealReconstruction
analog signal fromThe uniform samples, the idcal interpolation filter transfer function found i
its
Torecoverthe impulse response of this filter, the inverse Fourier transform of Hf). i
shown in Fig. 5.2a.
is
Ea.(5.7)
Also, the sampling interval T, = 1/f. Therefore, h(t) = 2BT, sinc(2r Bt)
Loco
T < Nyquist sampling rate, that is, 2BT, = 1, then
2B Assuming the use of
(56, \h()= sinc (2r B)
Thus, as long as the sampling frequcncy f, is greater than twice the signal bandwidth B (in hertz),
consist of nonoverlapping repetitions of G). When this is true, Fig. 5.le shows that g(0)) can shown in Fig. 5.2b. Observe the very interestingfact that h(t) = 0 at all Nyquist sarnpling instz
from its samples g) by passing the sampled signal g() through an idealIJowpass filter of be
reçoverBeH.A This h(t) is
The minimum sampling ratc f =2B required to recover g) from its samples g() is
for g(), and the coresponding sampling interval T, = 1/2B is called the Nyquist interval for bandwiNyquidthst
called the (t=:n/2B) except =0. When the sampled signal g()is applied at the input of this filter, the output is
being an impulse, generates a asinc pulse of height equal to the strength of the sarmpl
Each samplein g(), The process isidentical to that shown in Fig. 5.6, except that h() is asinc pulse ins
signal gr).* the lowpas,rale shown in Fig. 5.2c.
We need to stress one important point regarding the possibility of f, = 2B and a pulse. Addition of the sinc pulses generated by all the samples results in g(t). The kth sa
of arectangular kT.), He
lowpass signals. For ageneral signal spectrum, we have proved that the sampling rate fs >particular
2B.
class a einnut g() is the impulse g(kT)8( kT); the filter output of this impulse is g(kT)h(t -
now be expressed as a sum.
the spectrumG) has no impulse (or its derivatives) at the highest frequency B, then the overlap However.
is n ii the flter output to g(), which is g), can
as long as the sampling rate is greater than or equal to the Nyquist rate, that is,
g) =)&(kT,)h(t - kT;)
k
f> 2B
If, on the other hand, G ) contains an impulse at the highest frequency tB, then the equality must be -Xs(kT;):sinc (2
k
B(t - kT)I
removed or else overlap will occur. such case, the sampling rate f must be greater than 2B Hz. A wel.
known example is a sinusoid g(t) = sin 2r B(t - to). This signal is band-limited to B Hz, but allits samples =) g(KT,) sinc (2r Bt - k)
are zero when uniformly taken at a rate fs = 2B (starting at t = lo), and g() cannot be recovered from its
Nyquist samples. Thus, for sinusoids, the condition of f, > 2B must be satisfied.
Equation (5.10) is the interpolation formula, which yields values of g(t) between samples as a v
sum of all the sample values.
5.1.1 Signal Reconstruction from Uniform Samples
The process of reconstructing a continuous time signal g) from its samples is also known as interpolation h(t)
Hð)
In Fig. 5.1, we used a constructive proof to show that a signal g(t) band-limited to B Hz can be
reconstrucied
(interpolated) exactly from its samples. This means not only that uniform sampling at above the Nyquist ralt T
preserves all the signal information, but also that simply passing the sampled signal through an ideal lowps
filter of bandwidth B Hz will reconstruct the original message. As seen from Eq. (5.3), the sampled SB -1
2B
Contains a component (1/T;)g), and to recover g(t) [or G)l, the sampled
signal R
f
(a)
\B) =B(nT,)8(1 nT:) (b)

must be sent through an ideal lowpass filter of bandwidthB Hz and gain T,. Such an ideal filter response has
the transfer function Sampled signal Reconstructed signal
g) g(t)
(5.1)
H) =T,n()-T,n() T,= 2B

* The theorem stated here (and proved subsequently) applies to lowpass signals. A bandpass signal whose spectrum exists over a
takenatabove
frequency band fe - B/2 < |fl <fe+ B/2 has a bandwidth B Hz. (c)
the Nyquist frequency 2B. The sampling theoremis generally moreSuchacomplex in such
signal is alsocase. It usesdetermined
unigquely
by samples
two interlaced uniformsanmpling
trains, each at half the overall sampling rate Rs >
B. See, for example, Refs. Iand 2. Figure 5.2 ldeal interpolation.
 ANALOGTO-DIGITAL CONVERSION
SAMPLING AND
264
5.1 Sampling Theorem 265
Example 5.1
signals Seting f =0
Show that for ideally sampled
X(2)X,(-)dà
|ig(0)dt =T, lskT)|
k=-0
right-hand side
Changing variable in the
Solution
Using Eq. (5.10a) = Xnf)X,(-Ddf
k=m,n
IJsing Fourier transform of X*(1) for

J) = sinc (27B(t - kT))

where
Using Table 2.1 (pair 18) and time-shifting property of Fourier transform -/ -0

() = sinc (2TB(t - kT,) ’ -j2xRT, 1 e-2T(m-n)fT,

4B2
-B
We rewrite g) as B
1

4B?
cos 27(m- n)fT, df + sin 2r (m -n)fT, df
n m -B
-00
We now evaluate
For m#n, both the integrals result in zero.
For m=n, second integral results in zero and value of the first integral is 2B.
Am() n()dt = w())dt
Hence,
where
w) = Xm (t)x, (t)
We note that the Fourier transform of w(t) is DR Tor m=n
In) A,()dt =
for m # n

w()e 2rf dt = Omy = T,omn

= Xn(f) * X,f) = W) where Smn is called Kronecker delta and is given by

1 for m = n
Because multiplication of signal in time domain is for m n
Hence, eauivalent to convolution in frequency o and T, is the sampling period.
Therefore [ [gí0 dt becomes
Wf)= X,(a)X,f - )dà
or
Is{0) d=T,snl,) s(mT,)8na
n m

w()e infdt = Xm()x,()e-j2rtdt =T, ) Lg(kT,))', n=m=k

k=-o
-0

Xm(2)X,f - )da
 266
The Solution
The lower the A values This sarmples
Eq. Solution where
We
Find
spectrum signal
Example5.3
ratesignal (5.10b use
Exampl
a
e5.2
g(0)signal the the signal
and 4 are SAMPLING
spectrum higher MHz.having = is survives.
interpolation
zero.
sarmpling
of shown gí)
the Show
cut-off spectrum and l that
only AND
sampled is Figure g(nT;) in Thus. interval is
visualizedfrequencies that Fig. one
formula band-l
g(0) imited
g(t) G(f) 5.3 =S.3. term 7, ANALOG-TO-DIGITAL
G) signal 0n is =
can 1
98 as Signal Observe the
is be = and to
96 (1- 0).
(coTesponding(5.10b) to Nyquist
f)((- = recovered reconstructed B
-96 MHz Nothat
-1 g(tT) Hz
and other this s)= construct interval and
Gð 0
by CONVERSION
|G) 98 where
passing from signal is sinc to = whose
- MHz, the k
f,) () fronm for g(t27,)
96 fo the (27 =g()
respectively. the = Nyquist satisfhes only 0) g0). samples
nf,) Br)
98 sampled 96 signal in its that =
= MHz samples these the g(t3T,) are
G) MHz f, samples, is,
that l summation T,
signal and conditions. has ==0
=
*1O) B in
= Example Sincp 1/2R
through a
bandwidth
MHz 2 onall
5.2. the but
a
bandpassideally is ight-hand
side o one
samg)eandHz. B
of
he
filtersampled Nyqur
wit u
Or
Noting
that conditiontheso
GivenSolutionscheme signalClearly G)
signalg() A Example5.4
is impulses, Her f,
g() for multiplied the
visualized =
such bandpass
exists ).MHz. 4
-..

G). ’recovery. that by -100

-98 as
with
g(t)
for a -100 G) We
constant filter
The -Tst<T.
and innote
sampled with
-96 -98 frequency G)
hence f,,
f 1 -96
T,) n=-0 thlower
e -96 is
spectrum, uniquely
Gð) isIts a
-2
shaded and
8(t S( G,) spectrum domaibandpass
n.
S(t higher -4 -84
- nT,) = part
- nT,) G) G.(f). 02
nT) Gð) of 0 spectrum.
n=-0O0
cut-off 4 sA
A
t’ ’ the
’ fromthe is 8
n= m=-0 n=-X
) obtained is recovered
sampled spectrum.
frequencies 4

Sf Gð)
96 96
S(f- nf.) as at (A is
frequencies 96 obtained
96
nf. MHz MHz f.
) sampled 10098
100
and usingSampling
Theorem5.1
nfs, MHz S
spectrum, 98
n MHz convolution
= MHz f,
0,
Obtain
#2.
#1,
. wil
and
propose produce of
a
train
a
the 267
of
 268
reconstructionas we
1niepotaion realizabie lently
Practical Since Hence. R)
canbe shifted
of We
Therefore versions where
of We g(i). observe
Eq.
may For established T,
(5.10). seen gít) g(t SAMPUNG
=
apply
practicalsignal
recovered the - that
from Signal
pulse the For ’Gf. required nI) Using
reconstruction in ofthe
implementation,
reconstruction
p() practical the Sec. must g)time AND
infnitely
Reconstruction it from condition around frequency
to 2.12 further not ANALOG
analyze g-(1)
functiong.()
ure application that overlap
systems long implies by is = convolution
the G.f)
5.4
the pulse this
nature ideal muliplying with nT, TO-DIGITAL
accuracy that n
al ) reconstructionp() from of
(Interpolation) coresponding
[(n preced1ng =
of
signal lowpass property.
pli) G() +)7, 0.tl.£2,... I,) n=-1
as the the
of
shown it CONVERSION
the reconstruction
uniform can
sinc filter with T,> 2T - g(i
reconstructed
nT,)p(t reconstruction
in also an 1 T]> s
pulse is to
Fig. signal noncausal be ideal (n- the
[nT, with wi
5.4. p(t) recovered 8.()
time
nT) -
nT,) samples. (e.g., l)T)sampled
time
However, must
+
signal. gate =
pulse
a and
T] spreador g.()
be CD by of succeedingspectrum
Let easy unrealizable. the height
player), used nT,
uswe above
denote must to in ±T. G,f)
generate. we the scheme. g(t-
first and uniqueFor
-(nt1)T)versiarecoven is
the need ideal This width a
new usene o summadon
ro reconstructia
to can 7
signal cay implemea be
equiz d
u

summation This e This

Such thmeans (5.Eq4on
.) In (5.1tEq
)hat To
pulses. practicalshould The distortionless
Additionally, the
Let equalizer relationship Denotefrequency
the
ilters frequency determine
Asus be
signal the except are that
shown now its
inverse filter equalizer
transferoftencenter the relation
reconstruction
consider clearly domain,
in for reconstructed n,
Fig. Ef) the referred })
of and the to
5.6, a P)must illustrates
lowpass = the
very reconstruction
requires
that hltered
to =)&nT)pt-
relationship p()
storage systemwithin be as original
channel
8(nT)
orData
function *
simple signal g()
Figure lowpass term that
E) G() equalizers.
= by
the PO. analog
utilizing E(f)P) g)
interpolating signal EÇ)P) with the
GNbetween
5.5 Samples p() = equalizer as
in n ET). To
using =
bandwidth fully nT) signal
Practical generator
such nature = PO-af
Pulse
=0, the
= thatis, 0, Distortionless recover pulse
pulse an to T,, n
reconstruction pt) = g(t),
must
n
signal equalizer. stop p()
of |fl fI>f- remove * we
generator Ball g0),
consists n
can
nstruction. Hz. < saT,)8(0nt) -
Equalizer frequency B reconstruction
requiresthat further - see
Ef) Figure all n)
B the of and from
that
multiple
filtering the
generates 5.5
content
shifted the
&) original Theorem5.1
demonstrates propertiesSampling
replicas of
replicas
above g()
short analog
f- Gf becomes of of
(zero-order the G) signal convolution
diagram B -
Hz, nf) necessary.shifted
(5.14a) (5.13b) can (5.13a)
hold) and (5.14b) in
of the rely and 269
a it to
 chosentrain, an such
that add As 270
Hence, become This This
analog another It The
very Of as means For is result. a is
the important transfer a SAMPUNG
course, long unrealizable.
equalizer gate
small, that passband time the
as
pulse
this the delay equalizer function
to the
filter
is for of AND
yield rectangular equalizer
Equivalently, gain
a to us unit Fiqure
requirement
the to the to frequency of
recover of ascertain filter height 5.6 ANALOG-TO-DIGITAL
following Ef) Ef) reconstruction p)
=T;Ef) reconstruction Ef) T
= =FlexibleE(f)
this to response Pf) with Simple
for the does be that
equalizer original requires Tp)(7fsin P() is n
pulse
rectangular a not well the T,/P) the interpolation
sin (7fTp)Sin such
zf need defined, equalizer should =1, Fourier duration
pulse 1/B I,< that
p) passband analog that CONVERSION
to sinc by
econstruction width achieve <B
I#0
f1 it Ifl B< Il satisfy R(1)
signal is e passband
< (transtorm
zf (t T,. means
p T Imperative -nls The
response: is afio (1/T,|f|< B T,)
shorter infinite reconstruction of
g(t) (1/T, e of - simple
response - 0.51p
1fl sB
-il,Il(¢/7p)
pulse from than
gain.
for B) rectangular
the 1/B, us
generator. to is B)
Otherwise
equalizerthe realizable. shifted
reconstruction
nonideal it choose will
may first pulsee
by
In be short a generate
practice, First 0.5T.
possible pulwisdeth t of
., to all,
(517) cam puls design would (5.16) we
ca
l,

theorempliwansg There The ratexactly

e with to towicvcles, or This flier vithout
Realizability the gapany Tf straight-Iine linear example This
realize beth ws sampled a 51.2 struction.
it
increases, a zero a A
>means is sional iangle
Treachery is smaller
from even gradual 47nB). unrealizable as
practical ca n is means
another beyond shown is interpolation gi Such
its that signal Pratical
sampled segment s pulse improve ven that
the time this This
samples, cut-off insolution we between of ascases very
fundamental recovered filter. the Fig. a
proved of delay. yields can ing() A(/2T,),instead
Aliasing characteristic
first 5.7b. recover Reconstruction
practice: through
to successiVe the
at (Prob. MATLAB on little
make
even This The cycle
G),
Issues the
on signal th is
only We Nyquist 5.zero0-order-hold the
the
practical if
shows of problem the it an
1-9).results the of distortion
f he
tadvantage consisting
can exercise design
a assumption approaches G)
(dotted signal can ideal in staircase
sampling now cycles, in
be rate remains
difficulty
that (Fig. isclosely lowpass Filters Signal an in of
recover to g(1) f, the
it in area of
sample as interpolation
Gð) that th e rate isthis5.76). repetitions from shown = interpolation. filter Sec
inpossible in approximated filter 2B equalizer
s9 when
(b (a the in desired isAccording case Fig. Gð) Sampling
t Gð) higher the its by
reconstructing
signal inHz, very
B B using
is 5.7b). from signal samples (dotted Fig.
closely.
signal of the in
more than inthat
which The either short
gt) G) 5.7a.
spectrum
practice the toBut G) at area the
is the required the even with awith only
rat e in To first-order-hold
linearunnecessaryrectangular
[or
and successive
band-limited. a Nyquist Paley-Wiener infinite with recover Fig .
signal to in
from higher a G) interpolator,
recover this finite infinite5.7a). Reconstruction
filter G]
rate. consists
g() pulses
f from case, band than time sample or
a can time As Sampling
Theorem5.1
All However, criterion, the th edelay. from very
its band-limited be by gap seen filter,
practical of are
samples. better filterusing Nyquist delay g), topswhose simple.
between in
repetitions used
as gain
it a inSec. we are which
impulse
approximated the lowpass
is in
rate the 2.12, connected An
signals The signalimpossible successive
is need signal
sampling response.
(f resultsillustrative
required of
sam > such to response
are g(1) filter recon-
2B pass G)
271
a by ina
 Interterence
littlesimplified
signal*Figure reconstructed
limits sampling
antialiasingcutoff the half.inthe this Defectors Contend the the
Thus, quency
the
frequencies in a 2. L 272
rnore nents f/2) spectrum la ppi n g Si gnal s .
you I causes: information frequency
spectra. folding double you defectingenemy Thefolding Fig component The The
the pled unavoi
outputsignal
d abl e,
bandwi dth . tiVimcee-lçannotimited.
involved. picture. 58b the f/2 We action got were the components of-f. Note
from f 5.8 b reappearance versa be
Fortunately, shows army wi th problem frequencies los bet w een ), whi c h SAMPLUNG
signal Hz,
g). followjeopardy wind
before sid e in ma y theat is
cycles In fltering frequency Eliminated: sabotage platoon
frequency above and of R(). about asand (but time-limited th
signal ofthe the of be the Dot
reality, that asSuch exactly commander
the has but als o
regardlil eussstrated the are a at
other from g(t) the of e frequency viewedspectra tail G If
therepeat signal is,
shown
is defection,
of lost remains inthreconstructed crossat Gf. ing spectnecessari r um ly they AND
than
practical
all all
to Gaa suppression f/2 fighting asaliasing f,/2. folding
Fig. bel
in ow of
the the performed
fs/2
the betrayal twocaused an Fi g . this of samplbut ed and of ca n and
the cycles ininite ) in same 5.8b f2. as
if
= of
The platoons effective nominally is as 5. 8 c. f:/2)+Í tailinverted G() Gaf spectitral the Fig.in be are
imnediate spectra Hz. when begins, you the by depicted frequency th e beyond is Gn band-lANALOG-
imited TO-DIGITAL of
number
overlap
Fig. V2T
procedure. and the
analogous or signal. sampling t5.im8b.e-limited,
This before betrayed
would c. InThis l o st (Fig.signal no simultaneously C hni te
also
the 5.8d. of sabotage, Antialiasing
higher Hz. you defectors
tofighting loyal Such frequency loisngercycles. consists
eighbors. and ofantialiasingway, This We nonproductive |f | 5.8c),
mustrepeating sampling. incapacitate. lose to
in f,/2 the tailThus,showS tail
Fig. process inversion, or passed ratBecause
e. duration
interact
frequencies should The army, aliasing is folded >/2 of
simultaneously.
decay is force. to that Hz, the folding possible,
we a and their 5.8c. f/2=1/27 which is Because
When with cycles, lose called potential solution only of but up through Samplioverl ng appiofng are or
at the
Filter destroys components of back Hz e
thnon-band-
non-time-limited CONVERSION
suchhigherevery
schemeFigure eliminate one solution will army. In an
aliasing,
these known as, of wi dth.
only only the can (the by activity. have
addition, army oback onto a event at
quencies. another defectors thatwhatever mpersonate_, version
theoretically.recover antothe infinite
-limited,
ption the the
antialiasingis
5.8e The th e very as onto ideal higher
a Cycl
neighboring components
(suppress) be
these defecting)"
partly to when Hz, the overlapping es as time- A We
cycle used. to during orequencieS
integrity spectral not bandwidth
COmponents shows
beyond accomplished the
use army components itselfwhich spectrum lowpass of and can
because frequency are loval a only is of-limited
This An problem
rectifies th e means,
problem Gf) rate iGU) demonstrate
is certain
is non-band-limited),
tified,
liasing
not the
filter. al actual in folding
are at shown
results spectral
overlap.
cycles the platoon. platoon of the reduces
infinitetheof antialiasing double a calledthe distorted filter tails, repealing in signal
sampled the platoon the we folding this
in would fighting, reappear component above in
an Figure5.al8dso by defecting frequency losing or of G)
This to
jeopardy. aliasing, cut-off
butcase, Fi g. cannot(Prob.
ignificant an Js/2 as g(t)
spectrum
signal be neutralize has frequency. eliminale does
nonot every,
the filter ideal is the (aliased) all frequency,
foldingTe a spectral
ovetg the 5.8a;
width obVious. result exactly longer 5.1-\0) be
an d platool a secretly components th e reappear of frequency fL Clearly,
of folding army band-imitea
First, hey
al This lowpass from bg) efin cutparts ial of (he Hz(
utations abmecoonuYnt of es ential y ycomponents the components shown is lower
as For two rom has have all \hal
detecor. wi l defect
it ed sampitA
practical isa frequei shows tha tsolhe ution
have has lower as
frenneinstaz sepr comrla
the prace a
somewh belon fe sat como infh siy
band and the iler a beyond los es t il i los to i

spectrum
Pecrum.gure

(solid) (d) 5.8

Sampling (no
when Aliasing
distortion
Reconstructed
spectrum
tialiasing
schemeeffect. Reconstructed
spectrum
of Reconstructionfilter
filter
lower Reconstruction Reconstruction
filter
using (a) g)
filterSpectrum frequencies)
antialiasing
is Antialiasing
filter G,f) HÍ)
Used. of Hð) H(f)
Hadf) -w2 -w2
a
practical
filter.
(e) folded
back Lost
tail is
Baa) 0 0
Sampledsignal (b) (a)
(e) (d) (c)
Gf) Gð)
g().
Gadl)
signal (b) Sampler
wJ2 w2
Spectrum distorts
Foldedtail
spectrum frequencies
lower

frequencies
higherofLost frequencies
higherofLost Losttail
of Baat) f f
(dotted)
sampled tail tail
results Sample
signal results
spectrum Sample
signal Sampling
Theorem5.1
and spectrum
g(h. in in
the loss loss
reconstructed (c)
Reconsructed
f
->f

signal
signal
273
 ANALOG-TO-DIGITALCONVERSION
SAMPUNG AND 275
274 5.1 Sampling Theorem
corTuptin9
components now cannot reappear, the components
f/2 Hz. These suppressed Clearly, use of an antialiasing filter results in
for |f| <f/2. Thus, although we lost the spectrum the
frequency. beyond f/2 H7ol 0rsiequeni Gf) - Gð- n28)
distortionreconsis tcutructiend
trum the folding
below Gaf) = G)
intact. The effective aliasing di
remains
for all the frequençies below f/2again that the antialiasing operation must m be performed Gð)
elimination of folding. We stress
sampled. to reduce noise. Noise, generally, has a wideband
An antialiasing filter also helps itself will cause the noise
before watlfhe owsiignngal
antialiasing, the aliasing phenomenon
to appear in the signal band.
components outside the
Antialiasingsuppresses the entiree noise spectrum beyond
spectdersuirm,ed asngdnawibbao 2B -B.
B 2B

The antialiasing filter. being anideal filter, is

ynrealizable. In practice we use a
leaves a sharply attenuated residual spectrum beyond the tolding
frequency f./D steep cut-of frequency iÍ/2.l er, (a)
Aliasing Aliasing

whid Equivalent lowpass signal G, (f)

Sampling Forces Non-Band-Limited Signals toof Appear Band-Limited
Figure 5.8b shows the spectrum of a signal R))
consists overlapping cycles of
G(f).
may also view the spectrum in Fig. This
samples of g(). However, we
Rt) are sub-Nyquist 5.8b means tha
overlap.
Gaf) (Fig. 5.8c). repeating periodically every f Hz without actually The
to f,/2 Hz. Hence, these (sub-Nvquist) samples of g(t) are the Nyquist spectrum Gaf)asisthe specnum
In conclusion, sampling a non-band-limited signal g() at a rate f Hz makes the
Nyquist samples of some signal gat), band-limited to f,/2 Hz. In other words,
samples for g
samples appearSignalto be
sampling makes
band-imitel (b) -B

limited signal appear to be a band-limited signal ga() with bandwidth fs/2 Hz. A Sanon-band,the ees 5.9 ll Nor-band.limited signal spectrum and its sampled spectrum G(h. (b)Equivalent lowpass signal spectrum
similar 2B.
Gah constructed from uniform samples of g(h) at sampling ate
if g(t) is band -limitedbut sampled at asub-Nyquist rate. conclusion aples:
5.1.3 Maximym Information Rate: Two Pieces of Rnling of aband-limited signal ga(t). Thus, through Fig. 5.9, we demonstrate that sampling g() and 8al!)
Information per Second per Hertz athe rate of 2B Hz will generate the same independent information sequence (gn):
Aknowledge of the maximum rate at which informaíon can be
BHz is of fundanental importance in digital commyfication. We transmitted over a channel of
bandwidth
now derive one of the basic relatiomcki
&n =gnl,) =ga(nT,), T,=
2B
(5.18)
in communication, which states that a maximum of 2B
independent pieces
transmitted. error free, over a noiseless channel /of bandwidth BHz. The of information per second com b
Also, from the sampling theorem, a lowpass signal ga(t) with bandwidth B can be reconstructed from its
theorem. result follows from the samnlina uniform samples [Eq. (5.10)]
First, the sampling theorem shows that alowpass signal of
samples uniformly taken at the rate of 2B samples per second. bandwidth B Hz can be fully recovered from &al) )&n sinc (2r Bt k)
of independent data at the rate of 2B Hz can come from Conversely, we need to show that any sequenct
B. Moreover. we can construct this uniform samples of a lowpass signal with bandidth n

lowpass signal from the independent data sequence.

Suppose a sequence of independent data samples is denoted as (&n). Its rate is 2B samples per second Assuming no noise, this signal can b¹ transmitted over a distortionless channel of bandwidth B Hz. error
Then there always exists a (not nece[sarily free. At the receiver, the data sequence (&a)can be recovered from the Nyquist samples of the distortionless
band-limited) signal g) such that
channeloutput ga() as the desired information data.
1 This theoretical rate of conmunication assumes a noise-free channel. In practice, channel noise is
&n = gnT,), T,= unavoidable, and consequently this rate will cause some detection errors. In Chapter 13, we shall present
2B
in Fig. 5.9a we illustrate again the the Shannon capacity which determines the theoretical error-free communication raé in the presence of
effect of sampling the noise.
f,= 2B Hz. Because of aljásing. the ideal
sampled signal non-band-limited signal g) at Sanpung
R() =)gnT,)8(( nT,) 5.1.4 Nonideal Practical Sampling Analysis
=&a(nT,)8(( nT,) Inus far, we have mainly focused on ideal uniform sampling that can use an ideal impulse sampling pulse
Tain to precisely extract the signal value g(kT) at the precise instant of t = kT,. In practice, no physical
where g. (t) is the device can carry out such a task. Consequently. we need to consider the more practical
words, sub-Nyquistaliased lowpass signal other implementation of
sampling of asignal glt)whose samples ga(nT,) equal to the samples of g(nT,). InNvquis
generates samples obtained by
Sampling. This analysis is important tothe better understanding of erors that typically occur during practical
AD conversion and their effects on signal
that can be equally well
l reconstruction.
 ANALOG-TO-DIGITAL CONVERSION 5.1 Sampling Theorem 277
276 SAMPUNG AND

Practical samplers take each signal sample over a short time interval I, around g()
takes ashort snapshot of duration T, from the
every I, seconds, the sampling device still photographs of asprinter
pled. This is just like taking asequence of
still picture by averaging
during an
the
signal g)
100-metovererthe being sran
Much like aregular camera that generates a picture scene
the practical sampler would
the window Tp. that is.
generate a sample value at t= KT, by averaging the values of Oymgie
SignalwindgloNhT i
8ik7,) =T, J-T,/2 g(kT, + n) di
(a)
(619,
Depending on the actual device, this averaging may be weighted by a
device-dependent
q) such that
averaging Tuncoo, q()

ikE)=T,J-,2
q)g(kT, + ) dt
(5.19 (b)

Thus, we have used the camera analogy to establhsh that practical samplers in fact
of the form generate sampled
R() = 81 (kT,)8( kT,)
(520
We will now show the relationship between the
analog signal g() in the frequency domain.
practically sampled signal g(t) and the original Joue.
We will use Fig. 5.10 to illustrate the
weighting. This means that
relationship between g(t) and g() for the special case of unitorm
(c)
q)) = s 0.57, Filter impulse
0 |>0.5T, response h,(
As shown inFig. 5.10, g1{t) can be
Snapshots equivalently obtained by first using natural gating" to generate the signal
t= kT, () =8 (kT) ÔI -KT,)
Averaging
g(0) = g) 4,)
where
(5.21) filterg()
4, () =
n=-0
(d)
Figure 5.10b illustrates the snapshot signal (). We
can then define an averaging filter
with impulse l Figure 5.10 llustration of practical sampling.

2 2 AS lustrated in Fig. 5.10c, the practical sampler generates a

sampled signal () by sampling the averaging
elsewhere er output g1(kT;). Thus, we have used Fig. 5.10c to
or transfer function establish the equivalent process of taking snapshots,
ng and sampling in generating practical samples of g(t). Now we can
Haf) = sinc (f T,) Claionships to analyze the distortion examine the frequency domain
In the generated by practical samplers.
Sendingthe naturally gated that folowing analysis, we will consider a general weighting function g(t) whose only constraint is
snapshot signal g() into the Signal
averaging filter generates the output
8i) = ha) * g(0) q(t) = 0, (-0.5T,. 0.5T,)
278 SAMPLING AND ANALOGTO-DIGITAL CONVERSION
To begin, note that g, (t) is penodic.
Therefore, its Fouier series can be written a.
5.1 Sampling Theorem 279
4, () = e.emos lowpass signal G) with bandwidth B Hz, applying an ideal lowpass (interpolation) filter will
Forthe distorted signal
ceneratea
where
qt)e ns dt Fof)Gf)
Tius, the averaging filter output signal is
T,J0ST, in which
(5.26a)

gi) = ha() * g()q,, ())

Fo) = -er sinc (z(f +nf)T] (5.26b)

= halt)* ) Qng)henwst he seen from Eqs. (5.25) and (.26) that the practically
sampled signal already contains a known
distortion Fof).
in the (S.2), erenver. the use of a practical reconstruction pulse p(t) as in Eq. (5.12) will
frequency domain, we have distortion. Let us reconstruct g() by using the practical samples to generate
generate additional

Gi)= HO) }) =81(nT)p(t - nT,)

Qn Gf- nf) n

Then from Ea. (5.13) we obtain the relationship

between the spectra of the reconstruction and the original
message G) as
= sinc (7fT,) Q, Gð - nf,)
n=-0 (5.23)
Because ð) = POEIGð + nf) (5.27)
n

Since G( f) has bandwidth B Hz, we will need to design a new

equalizer with transfer function E(f) such
|)XkT;)(t KT,) that the reconstruction is distortionless within the
bandwidth B. that is.
we can appiy the sampling
theorem to show that f| < B
E)PO)Fof) =Flexible B<|f| <fi- B (5.28)
)= Gif+ mf,) Ifl>f-B
This single equalizer can be designed to compensate for two
sources of distortion: nonideal sampling
FoC) and nonideal effect in Pf). The equalizer design is made practically possible effect in
=7 sinc (2rf +2m2rf.) Tp f+ mfs nf,)
reconstruction
both distortions are known in advance. because

5.1.5 Some Applications of the Sampling Theorem

(5.24)
-ef) Lne sampling theorem is very important in signal analysis,
The last equality came processing, and transmission because it allowS us
Plaee a continuous time signal by a discrete sequence of numbers. Processing
We can define from the change of the a continuous time signal
frequency responses sumation index = m-n. nerefore equivalent to processing a discrete sequence of numbers. This leads us direcly into the area of
digital filtering. In the field of communication, the ransmission of a continuous time message reduces to
RNnssion of a sequence of numbers. This opens doors to many new techniques of
Ff) =e sinc ((f +(n + e)mf) nuous communicating
are used to time signals by pulse trains. Thbe continuous time signal gí) is sampled, and sample values
This definition allows us to
Tp] modify certain parameters of a periodic pulse train. We may vary the amplitudes (Fig. 5. 1lb).
g S.1lc), or positions (Fig. 5.11d) of the pulses in proportion to the. sample values of
conveniently write &). the_signal
or Accordingly,we can have pulse amplitude modulation (PAM), pulse width modulation (PWM),
pulse position modulation (PPM). The most important form of pulse modulation today is pulse code
)=F (f)Gi(f +f) (5.25) modulation (PCM),), introduced in Sec. 1.2. In all these cases, instead of transmitting g), we transmit the
corrreconstruct
and esponding pulse-modulated signal. At the receiver, we read the information of the pulse-modulated signal
tthe analog signal g().
 ANALOG-TO-DIGITAL CONVERSION 281
280
SAMPUNG AND 5.2Pulse Code Modulation

CODE MODULATION
5.2PULSE the most useful and widely used of
all the pulse modulations mentioned.
modulation (PCM) is analog signal into a digital signal (A/D
Pulse code 13, PCM basically is atool for
converting an
in Fig. 5. can take on any value over a continuous
(a)
As
shown signal is characterized by an amplitude that
conversion). An analog
can take on an infinite number of values. On the other hand, digital signal amplitude
meansthat it signal by means
range. This An analog signal can be converted into a digital
on onlya finite number of values. one of the closest permissible numbers (or
can take
quantizing, that is, rounding off its value to
samplingand
in Fig. 5.14. The amplitudes of the analog signalNext, m(t) lie in the range(-mp, m,),
of
quantized Ilevels). as shownsubintervals, each of magnitude Av = 2m,/L. each sample amplitude is
q enartitioned into L sample falls (see Fig. 5.14 for L = 16).
value of the subinterval in which the
Pulse locations approximated bythe midpoint the signal is digitized, with quantized
are the same but now approximated to one of the L numbers. Thus,
their widths Each sample is one of the L values. Such a signal is known as an L-ary digital signa.
change samples taking on any a binary digital signal (a signal that can take on only two values)
is very
From practical viewpoint, an L-ary signal into a
engineering. We can convert
Aacirahle because of its simplicity,economy, and ease of This code.
for the case of L= 16was shown in Fig. 1.5.
(c) binary signal by using pulse coding. Such a codingdigits from 0 to 15, is known as the natural binary code
formed by binary representation ¡f the 16 decimal to be
code will be discussed later. Each of the 16 levels
Pulse widths are
the same but NBC). Other possible ways of assigning a binary is now converted to a (binary)
analog signal m(t)
transmitted is assigned one binary code of four digits. The
their locations
convenience. This contraction of binary digit" to bit" has
change digital signal. A binary digit is called a bit for
used throughout the book.
become an industry standard abbreviation and is transmit this binary data, we need to assign
(d Thus, each sample in this example is encoded by four bits. To
distinct pulse shape to each of the two bits. One possible way is to assign a negative pulse to a binary 0
a
is now transmitted by a group of four binary
Figure 5.11 Pulse-modulated signals. (o) The unmodulated signal. (b) The PAM signal. (c) The PWM
(PDM) signal. (d The and a positive pulse to a binary 1(Fig. 1.5)so that each sample
PPM signa. pulses (pulse code). The resulting signal is a binary signal.
8,(1)

Bit
LPF Sampler Ouantizer encoder
8,(0) 1 01 1

Figure 5.13 PCM system diogram.

levels
quantization
Allowed Quantized samples of m)
m)
Figure 5.12 Time division multiplexing of two
signals.
One advantage of using pulse modulation is that it permits the simultaneous transmission of several
signals on atime-sharing basis--time pulse-modulated signal
occupies only a part of the channel time,division
we canmultiplexing (TDM).pulse-modulated
transmit several Because signals on the same
channel by interweaving them. Figure 5.12 shows the TDM of two PAM signals. Inthis manner wecan
muliplex several signals on the same channel by
Another method of transmitting reducing pulse
several3. Inbaseband widths. divisionmultiplex
ing (FDM), briefly discussed in Chapter FDM, signals
various signals are multiplexed by sharingthe channel
bandwidth. The spectrum of each message is shifted to a simultaneously
is frequency
signal.The
2m

information of various signals is specifc band not occupied anyother way,TDM

by
and FDM are duals of each other.located in
nonoverlapping frequency bands of the channel. In a -m,

Figure 5.14 Quantization of a sampled analog signal.

 ANALOG-TO-DIGITAL CONVERSION 5.2 Pulse Code Modulation 283
282 SAMPUNG AND

about 15 kHz. However. for specch, subjective tests

The audio signal bandwidth is messages can be extremely reliable without deterioration.
Analog messages
all the componentsabove 3400 Hz are suppressed3
articulation (intelligibility) is not affected if
show
rather than high fidelity the components hat
Reproduction with digitalfilms, for example, lose quality at each successive stage of reproduction and

Since `igphae
communication is intelligibility 8. photocopies and cost.
objective in telephone
filter. The resulting signal is then sampled at arate of 8000 suchas transported physicallyfrom one distant place to another, often at relatively high
Hz are eliminated bv a lowpass be performance or capacity
(8 kHz). This rate is intentionaly kept higher
filters can be applied for signal reconstruction.
than the Nyquist sampling rate of 6.8
kH,
Each sample is finally quantized into 256 s.soamplethsat seconA
per
must

9 The
ofdigital hardware continues to
ccost the same time
halve every two or three years, while
period. And there is no end in sight yet to this breathtaking and relentless
to encode each sample (28 256).
which requires a group of eight binary pulses second. Thus, a levels (Lre=56,
alizae doubles over
digitalItechnology. As aresult, digital technologies today dominate in any given
exponential progressin storage technologies.
requires 8 x 8000= 64,000 binary pulses per
The compact disc (CD) is an application of PCM. This
is a high-fidelity
situation telephone signa area of communication or

signal bandwidth to be 20 kHz. Although the Nyquist sampling rate is only 40 kHz, the
rate of 44.1 kHz is used for the reason mentioned earlier. The signal is quantized into a rather
requiring
he
actual sa audn AHistorical Note
mathematical concepts for describing
(L= 65,536) of quantization levels, each of which is represented by 16 bits to largesanumbe
reduce the quantizing eo, mplne, ent Indian writer Pingala applied what turns out to be advanced
so presentedthe first known description of a binary numeral system, possibly as early
prosody, and in doing BCE.4 Others, like R. Hall in Mathematics of Poetry place him later, circa 200 BCE.
The binary-coded samples (1.4 million bit/s) are then recorded on the CD.
eighth century
asthe Leibniz (1646-1716) was the first mathematician in the Westto work out systematically
Gottfried Wilhelm this discovery,
inary representation (using ls and Os) for any number. He felt a spiritual significance innothingness.
3.2.1Advantages ofDigital Communication blieving that 1, representing unity, was clearly a symbol for God, while 0 represented
He
proves that God
neoned that if all numbers can be represented merely by the use of 1 and 0, this surely
Here are some of the advantages of digital communication over analog communication nothing!
Created the universe out of
1. Digital communication, whichcan withstand channel noise and distortion much better than:
as the noise and the distortion are within limits, is more analog
rugged than analog communication. With as longt
messages, on the other hand. any distortion or noise, no matter how small, will
distort the receivedIsSignal 5.2.2 Quantizing
2. The greatest advantage of digital communication over
analog communication, however. is the viak.
of regenerative repeaters in the former. In an analog As mentioned earlier, digital signals come froma variety of sources. Some sources such as computers are
communication system, a message signal becomes:
progressivelyweaker as it travels along the channel, whereas inherently digital. Some sources are analog, but are converted into digital form by a variety of techniques
the
distorion grow progressively stronger. Ultimately the signal is cumulative channel noise and the siond such as PCM and delta modulation (DM), which will now be analyzed. The rest of this section provides
Amplification offers litde help because it enhances the signal and overwhelmed by noise and distortion quantitative discussion of PCM and its various aspects, such as quantizing, encoding, synchronizing. the
Consequently. the distance over which an analog message can be the noise byis the same proportion required transmission bandwidth. and SNR.
transmission power. For digital communication, a long transmission path transmitted limited by the initial For quantization, we limit the amplitude of the message signal m(t) to the range (-my, m,), as shown
noise and interferences. The trick, however. is to set up may also lead to overwhelming in Fig. 5.14. Note that m, is not necessarily the peak amplitude of mt). The amplitudesof m(i) beyond ±m,
distances short enough to be able to detect signal pulses repeaterthestations along the transmission path a are simply chopped off. Thus,m, is not a parameter of the signal m(t): rather, it is the limit of the quantizer.
accumulate sufficiently. At each repeater station the pulsesbefore noise and distortion have achance to
are detected, and new, clean pulses are trans The amplitude range (-mp, m,) is divided into L uniformly spaced intervals, each of width Av = 2m,/L.
mitted to the next repeater station, which, in turn, Asample value is approximated by the midpoint of the interval in which it lies (Fig. 5.14). The quantized
are within limits (which is possible duplicates the same process. If the
rectly. This way digital messages canbecause of the closely spaced repeaters), pulses noise and distortol samples are coded and transmitted as binary pulses. At the receiver some pulses may be detected incorectly.
be can be detecled Hence, there are two sources ofeorin his scheme: quantizationere, and pulse detection error, In almost
most significant error in PCM comes fromtransmitted over longer distances with greater reliability. I all practical schemes, the pulse detection error is quite smallcompared to the quantization error and can be
increasing the number of quantizing Jevels, quantizing. This eror can be
the price of which is paid in reduced as much as desieu ghored. In the present analysis, therefore, we shall assume that the error in the received signal is caused
transmission medium (channel) an increased bandwian o exclusively by quantization.
3. Digital hardware lf m(kT,) is the kth sample of the signal m(), and if m(kT,) is the corresponding quantized sample, then
implementation is flexible and permits the Irom the interpolation formula in Eq. (5.10),
large-scale integrated circuits.
4. Digital signals can be
use of
microprocessors, digital switching,a
5. It is easier and
coded to yield
extremely low error rates and high fidelity as well as for privacy.
more efficient to m() = ) m(kT,) sinc (2rBi - kI)
6. Digital multiplex several digital signals.
7. Digital communication
k
is inherently more
signal storage is relatively easy efficient than analog in exchanging SNR for bandwidth. and
information from a distant electronic database. and inexpensive. It also has the ability to search and select
îm) =m(kT,) sinc (2rB: kr)
k
*
Components below 300 Hz may also be where m(t)
The error inpulse detection can be madesuppressed without affecting the is the signal reconstructed from
negligible. articulation. Teconstructed signal is g) =no- m). Thus,quantized samples. The distortion component g() in the
284 SAMPUNG AND ANALOG-TO-DIGITAL CONVERSION
q) =)im(kT) mkT,) sinc
(2r Bi - kn) 5.2 Pulse Code
Modulation 285
=) qkT, sinc (27 Br - k) error is equally likely to lie anywhere in the range (-Av/2,
Assuming that the Av/2), the mean square
quantizing error q' is given by*
where gikT,)is the quantization error in the kth
sample. The
acts as noise, known as
quantization noise. To calculate thesignal q()or is the undesired Av
power, the mean
have
square signalyalu,e anofd, hence. (Avy?
cI/2
qt), we 12 (5.32)
()=
im- o
J-T/2
a m
3L2 (5.33)
1 I/2
-lim -akT,) sinc (2rBi - Dorause a) is the mean square value or power of the quantization noise, we shall denote it by N.

We can show that (see (5.29a)

that is.
Prob. 2.14-4) the signals sinc (27 Bt - mI) and sinc (2 Bt Ng = ')= ";
3L2
-
nT) are
orthogonal. Assuming that the pulse detection error at the receiver is negligible, the
reconstructed signal mít) at the
receiver output is
sinc (27 Bt - mI) Sinc (2r Bt - m#n
nI) dt = -
1
m(1) = m() + q()
(5.29%) The desired signal at the output is m(t), and the
Because of this result, the integrals of the (quantization) noise is q(t). Since the power of the message
and we obtain cross-product terms on the right-hand side of Eg. (5.29a) signal m(t) is m²(), then
vanish.
S, = m')
(t) = lim
T 7kT)
T/2 k sine² (2rB kr) dt N, = Ng = 3L
m

= lim and
T’x
J-T/2
sinc (2Bt kr) dt S
= 3L? m'()
N, (5.34)
From the m;
orhogonality relationship (5.29b), it follows that n this equation, m, is the peak
amplitude value that a quantizer can accept, and is therefore a
e quantizer. This means S, /N. the SNR. is a
parameter ot
5.18 with u=0). linear function of the message signal power m() (see Fig.
qt)= lim 1
T’o 2BT fkT) (5.30)
k
J.2.3 Principle of Progressive Taxation: Nonuniform
Because the sarnpling rate is 2B, the
the right-hand side of total number of
Quantization
au nat So/No, the SNR, is an indication of the quality of the
The quantum levels areEq. (5.30) represents the average,samples over the
the mean of averaging
interval T is 2BT. Hence.
the square of the quantization error.
received signal. Ideally we would like to
the subinterval (of separated by Av= 2m,/L.
height Av) in which the Since a midpointof
ed constant SNR (the same quality) for all values of
the message signal power m(). Unfortunately, the
the
quantization error lies in the range sample falls, thesample value is aapproximated bythetAv/2.
maximum quantization error is
Thus,
Those who are
(-Av/2, Av/2), where quantization errorfamiliar
q
with the theory of probability can derive this result directly by noting
iss l/(2mp/L) = L/2m, that the probability density of the
over the range \ql s mp/L and is zero elsewhere. Hence,
fmp /L m;
2mpL (531)
-mp/L -mp/L 3L?
 CONVERSION
286 SAMPLING AND ANALOGTO DIGITAL 5.2 Pulse Code Modulation 287

m(), which varies from speaker to speaker increments Am into larger

SNR Isdirectly proportional to the signal power output signal y. The compressor maps input signal
of the ditferent by vertical axis is the
40 dB (a power ratio of 10*). The signal power can also vary because Am
signals, and vice versa for large input signals. Hence, a given interval
in Eq. (5.34) can vary widely, depending on the lengtspeaker of as
hs theand much a The
increments Ay for
small input
circuits. This indicates that the SNR of steps (or smaller step size) when mis small. The quantization noise is lower
of the circuit. Even for the same speaker. the quality of the
the person speaks softly. Statistically, it is found that
received signal will deteriorate
smaller amplitudes predominate
in ma rkedcloymnecinwhen
he lengh containsa
largernumber
signàlIpower. An approximately logarithmíc compression characteristic
SImalterinput
yields a quanti-
proportional to the signal power m-), thus making the SNR practically independent
amplitudes are much less frequent. This means the SNR willbe low most of the ti
The root of this d1fficulty lies in the fact that the quantizing steps are of
speech and
uniform value larget zation
noise nearly
power over a large dynamic range (see later Fig. 5.18). This approach of equaliz-
input signal similar to the use of progressive income tax to equalize incomes. The loud talkers
The quantization noise N = (Ar/12 (Eq. (5.32)]is directly proportional to the ofthe SNR
The problem can be solved by usingsmaller steps for smalleramplitudes (nonuniformsquare of the step
in Fig. 5.l5a. The same resultis obtained by first compressing signal samples and quantizing,
quantization. The input-output characteristics of a compressor are shown in
i
the
then
Fig. 5.15b. Theusing a
v=2mL.
asashownsize
unifonm
ing
and
the

signals.
appears
stronger signals are
penalized with higher noise steps Av to compensate the soft talkers and weaker

several choices, two compression laws have been accepted as desirable

Among:
standards by the ITU-T®:
of the world and on
in Europe and the rest
is the normalized input signal (i.e. the input signal amplitude mdivided by the signal peak horizovanltuael axim,s used in North America and Japan, and the A-law used
the -law routes. Both the u-law and the A-law curves have odd symmetry about the vertical axis. The
international
given by
-law (for pasitive amplitudes) is
u m 5.35a)
Quantization m(1) 0 <
levels mp mp

is
The A-law (for positive amplitudes)
m
0<
Mp Mp A
y= (5.35b)
Am
|ha(l+ln:m, mp
These characteristics are shown in Fig. 5.16.
The compression parameter u (or A) determines the degree of compression. To obtain a nearly constant
(a) S,/N, over adynamic range of input signal power 40 dB, u should be greater than 100.Early North American
m, channel banks and other digital terminals used a value of =100, which yielded the best results for 7-bit
(128-level) encoding. An optimum value of u= 255 has been used for all North American 8-bit (256-level)

u1000 A =1000

u=l00
0.8 0.8 A=
87.6
| =J0

0.6 0.6

Uniform Ay 0.4
0.4

0.24 0.2

Am m (0 0.2 0.4 0. 0.8 0 0.2 0.4 0.6 0.8

mp
Nonuniform
(b) (a)
(b)

Figure 5.15 Nonuniform quantization. Figure 5.16 (a) u-law characteristic. (b) Alaw characteristic.
 258
essage The cOmpandr
preSSngDt comperbie
changes.
problem
Ss gta
output ompressOn
Generally aedthe 2t Theterinals
signal chaeTsO
signal cacgedsamples ANOGTOGTAL
AMEVGAD
CONVERSION
S\R s tesuits
usei the
Figure poweris compandar Figure
íor segalRang. an and
the
5.18 the the compiemeto haseariierhe CopreT
output m)tm S.17
sbon czses been
Rotio n
Figure
S\R mut be
in of N S ime standardizei value
o u [n+)} is of S17 zohcon
signci 20 30 40 Fg =
beniwih but of
5.18. 255 tretorei
hat to of
o ts desnbes isu compressorond
ontiztion Relative 2nd sample of
a ot byDOW
u incTease the he
20 = the ther taimost
compressoT.
Signal values. ITU-T.°
noise 0(uniforn increases use
does onginal exçpander channel
#=
255 of extinct. PCM
in power Because compressor
PCM not
40 Ihevalues
m-(t), L=256 quantization) its for
with arise compressor and For nonuniform
m-(t) bandwidth.
neither Butin the
and dB here. at nonlineanty
and the Expander
without the expander A-law,
60 It receiver
as happens time Quonhvat.
ression. a a
function expander
scale toge the by value
along
that using of
with: A
of when an =
mi) eApe 815
a
( (56

men EncoderThePCMThe omatsueams cOnpressor oOicompressed

t nodulated compressor
or Such DetworkAmerican earlier15-segmented
tomarginally D2
characteristic each third Tyo Compandor
iode A
logarithmic The
er,multiplexed similar channel Anquadrants
In a logarithmic diode matched is
Samples. ge combination
also into d Saving. the must altermative of
standard and piecewise inferior bankstandards. and the
has "au" at (u work is diodes compressor
desired
Which compression kilobits both
64 InThe =
the
approximation madesaturation
that (ignoring athe
binary PAM files 255) approximations in approach third
A-law expander
audio end-to-end oflinear
Though terms is to in form
makes differing used variable
fit parallel
output can cases,converts
digits. per
compressor with file a incan Piecewise
Figure
5.19
n be
sampling
second operate
devices, a of in (Fig. finite
is the be
quential Is options format
the against trueconjunction SNR. to Output
A decoded with real1zed
applied 14-bitcharacteristics use resistor
number first
variety ITU-T to 5.19)
that (kbit/s). at (A usedusing other =255 a uthe The a opposite quadrant:
signed digital piecewise to current).
arisons at e in th e =G.711 by is V= by
ofthe useither
th e
true with the
piecewise of
added
rate 87.6) networkcompressor points polarityY a
encodersinput u-law Since linear Sun,same terminal semiconductor
of standard is =255
u teighth
he
format. converts Unix, inferior linear in In charocteristic
compressor
linear
the of and the 8000 compression Tl (usuallypractice,
parallel.
to PCM elements linear
carrier bit provide
enerate is A-law. A-law and characteristic
device
working approximation (L
available.7.
encoder, It Hz, through to
should samples
13-bit = Coder
input
Java, either approximation system. 256) seven) adjustable By the -) diode
and G.77 a that
exhibiting
an the signed either law. with approximate
adjusting
n-bit which be to the ofuse with on because
10 encoder
u-law audio th e and The th e
We noted logarithmic the the u=255 a to 5.2
ewordquantizes linear characteristics the segmented is uthe resistorS
shall -law in piecewise th e has = ideal
various the Pulse
that are thus
PCM "au" true almostlogarithmic
method
255 characteristic
characteristics. VH
discuss mutually expander are Code
The and
the creates &-bit or characteristic law
Microsoft samples the fles linear approximation resistors.placed characteristic
sample encodes universally is Modulation
here compatible, fromsamples, A-law.° canobtained of widely
characteristics.
approximation. willchoice in in
digit-at-athe to be
is cach audio be the the
npared WAV logan1thmc pulse-code
leading The used series
when superior in replaced in resulting first of
sample audioaudtosignals
today's Norh is such
-law th e only thein with and 289
to A a
 wIth 290
C=8xa For Therefore, RTheSolution encodedacceptable
bandwidth 8.3. This (Sec.second.
minimum each
Because For a proportionalvalue.
5.2.4proportional dependingsample upperreference
to located.
minimum Example
The sígnal A 5.5
a
Frombinary The Nyquist= quantized we is 5.1.3), quantized binary Decodingis
The in
is orThe voltageSAMPUNG
6000 shall the Because a This lower
encoding
quantization binary m(1) theoretical of sequence
Transmission whetherthon
e the voltages
8000 Eq. from error PCM.
ansmission coding, see we 28 to kth process
obtained
sampling
(14)
x required band-limited quantized
2the digit the upper half
(5.37),
= Eq. signal.
samples in that require samplesasample AND
64,000 the unit of we + inverse
is involves
theof are
L = to for
minimum 0+0+2*+0+2+2+0=
n continues conveniently
or
must
(5.31). step
8000 rate transmit Ifsample bandwidth is, assign
binary applied by
we practical a a
bandwidth bitUs. 24are minimum per thus, allowed
therange.ANALOG-TO-DIGITAL
a
need be is Hz. is suchbinary to ofsample loweranswering
Av, second, Bandwidth
digits a sampleencoding.
to until combination
Because power a Ry the amplitude kHz 3 transmission encoded distinct
n and =2xmultiplexedsignalscoded. reasons (1 a the is generated
is channel Hz) can resistor
valuFore. inhalf
BT = 2 the we L=2" successive
=we log, of
maximum 3000 (the
are Find sampled require
can into group
be The Inlast the of
2. we s upper the of
C/2 256 mptime-division-multiplexed, bandwidth of arranged 2"R. thibinary by
can Hence, maximumthe may BT transmit n first reterence
CONVERSION
=
= signal. at bandwidth abits. of and a
transmit
32
6000 minimum a use = total n exaniple, The case, lowertherange. code
orquestions, bank
=the 0.5
quantization binary
digits digit
100"2L=200
kHz. 8 rate nB Because in
a a
bits next Hz quantization 33% transmission required Hzmaximum of 2"
the 150. currents each in theIndigit resistors voltages R,
of
up (samples B 2nBis, lon=
g, L binary
code a ofthe
perhigher bandwidth
distinct Output This the i second 1
Hz, signal a
beginning or
sample. higher to bit/s, (bits)
eachto completes in ncode half
2 error given of digits proportional
bit/s value transmit patterns, all has is0
per error) bandwidth two that the of
step,
eof is the generated, 2R,
second). determine of than by m(t) SNR word been
is with
per We AAv/2. a pieces resistors another
channel 0.5% isthe the band-limited the applied subinterval ....
2'2R.R.
z that isaL
hertrequire 2nB 10010110 generated. whether to
Nyquist PCM of D/A
The the higher of the depending a2
of of information
pieces conversion. are 10 digit
to actual minimum required the signal. Lquantization a 20,25
bandwidth, power than resistor
added.
in 11 the
peak rate. of to will which or sample
Salip to The this In information HzB give 1s
on
ampintuae Secs. sumyThe ihas
transmit of2 wheher
oigenerated
traisus
transiu maximum minimum. nerrequireS a is
ces levels. curem diferen beeN in
we a isL 8.2 (5.38) (5.37
require total of256. = mip and per a he he

output the Superior Dandwidth

doCan OuiputWhere to trade From
bandwidth Substitution where
SNR.Compare
Example
5.6 this, rewrite From 1.536/20.768 =The
Eal be ofEq.
controlled
But SNR a
to = SNR(5.40) multiplexedsignalhas
the m() FMincreases
Eq.yields
of
Expo(n5.e3n7)tEqi.,al
it (a 10 (5.39) for
case requires
or 6
logi0 Eq. Increase
of bandwidth a we L2:= MHz
bandwidth
of PM. by dB large (5.38)
only c. as observe
L= transmissionincrease). This C= 22", of
a benefit
64
doubling from into and transmission
shows Vo/ dB Do isthat of
with B Thus, attractive Eq. the the
32 in the [In total a
the
= kHz (5.39)
4 of
bandwidth. that log1o10
= terms SNR (1 3 ()m output
the if + Output of
case kHz increasing
to we (a bandwidth. CM
bandwidth 36 10logio =
+ ofand
increases yields u) SNR
of is increase 6n) SNR.comes ]' =
L=
transmitted kHz No S. [compressed SNR
We |uncompressed N, So in 24
(an c(=dB
n This c(2)28r/B = Eq. x
256 shall
to n by exponentiallyclose =c(2)2n 64,000
increase from (5.34)
from quadruple 1 relationship
using see (increasing to
the 8 the case, =
later of to 10logjo case, or
a 9, Eq. 1.536
point the that only the upper with in
binary is Eq. in (5.36)
of
SNR. 12.5%). SNR one
frequency
c+ clearly
theoretical the Eq. Mbits,
bit (5.36)] 5.2
view
companded 2n transmission (5.34)] can
In
quadruples, in Pulse
of this This log102 seen which
the be
transmission limit.
by Code
respect, andshowscodeword) expressed
phase using requires
PCM transmissionbutthe bandwidth A Modulation
PCM that thesmall
with modulation quadruples as
bandwidth in decibel a
PCM, increase minimum
strikingly is B.
= (5.41) (5.40)
scale This (5.39)
and 100. also SNR the in 291
of
 Hence. Soluton
Fot 292
For
instance,Thus, Cxpressed
instance, or We Logarithmic
defined Such
Comments the The
the certain a SNR L= =
Although use is difference t4, SAMPUNG
100-watt cable asthe 256,
cable the for n=
100-watt in
"100 same 10
case
logo L n
units atenuation units on =
between = f.
watt" the the with 64.
power unit 8. the AND
power ofdecibel signal Logarithmic
(power
ratio). and the
dBW may to frequency The transinission
logarithmic the transmission ANALOG
is (loss) express
power fomer two
is20 as be is
considered a
dBW. measure 11.76 is power
is variables requires SNRs
attenuated Thus, scalesUnits N,hbandwidth
S
TO-DIGITAL
12
is bandwidth 10loga=
PaBm Similarly, PaBw dBloBi010 gain =
of G= an or just No So (a
= to
power or
SNRSNRs. are dB. =a+6n
= be x= about t
10 10
byloss very which N S, is
CONVERSION
power a a is is 27.49dB
=(101))²[n 36) nB
log logo powerratios, factor over 10 convenient x A 32 dB
1 dB, 33% 3 =
T00 W OB10 logarithmic isa = kHz, 24
measured 100 of a where
ratio it =-11.76 certain more ratio 39.49 kHz.
=20 is 15, N S =-8.51I
dBm 50 often
of the when bandwidth of dB
with dBW 100 transmission unit l6.
dB cable
used for a Thus,
respect with variable
as
gain the compared the
mW
to1
respect a
medium.is power
measure has SNR
to ratio a tofor
l-watt of For
large the L=
power is
poWer instance, the
dynamic latter
256
is powel, decibel
is
dBm. is 16
if (dB Tanoe imes
over
For

world,
Inthe2.048
different except gaer schemes
everal passing nenpulses, edum. Dinaryapplied The AT1 MHz. 1.544 of 4 the thAision-multiplexing is orimarily
at reauiring talk was pied instal which were hulky. Historical of NotA e
an the schematic kHz Bellsignal because digital between become PCM
Circuit,ue Ine
asdemultiplexed sampler Time System' s give no t by
BecauTELEPHONY:
se DIGITAL5.3
earlier is Coincidentalhappens
y, la rg e, but
United Mbitus level O the codeword* At to now poor a
designed very other new so they and fthe of
parts e commutators
the Regenerative the an d bandwidth
version, Of samples the outputDivision rather
adopted
(in Dellsignals T1 of used
high-frequency
regenerate pairs attractive. underground
services heavy SYSTEMS
of
States receiver., input a TI to were its
unavailability
each the contrast system,shown (see T1 carrier
to systems
closelywith offor that bea iunreliable,
mplementation.
in System DS2, availablefarore
through (i.e., represents of channels at poor
elsewhere transmit several several
the (water, devith l
cate
world, and North Multiplexing theaboutsmal
Stmple in distributedin repeaters Fig. ancarrier
the system. new audio Ancables
Japan introduced to DS3,called Fig. Fig. encoder existing
5.20b). of
America decoder attempt the
SWiandtches
Was with T1, a system performance."
24 pulses times on voice gas,
and 3.5a) lowpass
5.20
time-division-multiplexed a voiCe because tended suitable
encoded wibefore
th digital spaced time-division-multiplexed Vacuum
appropriatebefore The that
thechannels sewer, system time
and
DS4, opened this are to is before larger was
by the
and 24 the
filter converts
eách regenerative
A spaced same range
space th
consumese to invention
between
devices,
elapsed
years
more
thanth20
e s
Japan.
channels the signal purpose. not approximatelysignal, quantizes shown pair etc.). dissipatedswitching tubes,
seven as made transistor overheat.
ITU-T ITU-Tdescribed TI
periodically thancable (04
through was
interfaces carrier mechanical intheof ofthe The
bits., level each the now in pulses available PCM TI
IN
Because wiresrepeaters, that was kHz) Moreover, onoverloaded, used 5.3
An
standardization, and 24
binary each Fig . power
little a
standardized Sampling channel. converted required limited a PCM lot
additional system next,(DS1), 1 channels). that get unacceptable FDMfor amplitude
and was before Digital
in Mbit/s). 1.544 6000 sample 5.20a, under of
international
of by but
pulses used too digging invented, the was
the narTOW PAM PCM repeaters
spaced suffered heat.
bit
in in is
Sec. isare distorted scaletoparticularly the the Telephony:
the used The
feet to
and All to modulatiop. and just
was widespread its done Systems
high-speed intoapartdigital Signal. 24 telephonetransmit
approximately up
streets is
waiting invention
added the 30-channelUnited 5.4 pulses desired eneodes can
severelyat
30-channelThe further channels high increase streets a CARRIER
connections. two
by samplesdetect The and
signals-offered nearly
for electronicof form, work indemand in having PCM
adoption States, for
signal1ng. standards 2 audio multiplexed onenoisy. frequencies. from and of
toelectronic it signals Unfortunately, many large ideal the
PCM multiplex
us (decoding). the is
into are satisfactorily t he the in
dozens duration. audio invention
for vacuum T1
signal pulses sent sampled This 6000 noise. causing citiescities.was Itswitch. transistor,
continue system of system
gates over eight wi th capacity telephone Carrier
the signal isfeet
switching is and PAM a the Furthermore,
Ironically,
the dislocations many was tubes
Tlvariations of into The(such reconstructed The binary
the total the of
is
with retransmit in of apart, on
solution. already th e
Systems
carrier
to used progressIvely 1.544 transmission Signal history by were
usedbein as samples pulsesasequence. a bandwidth
bandwidth noisy cables
frequency- easyservice
notto as
a a
circuits. clean transistor, switches not
system all rate Mbitsbridge is PCM
over were now of lines This cross were occu-
of byare new the up had only 293
 seven onlrequired
y: 128 levels
=quantizing
L
lossS pattern),a (framingbit pattern code given th1ollow
e not doessequence this If
nal.
signaing. for used was eighth
bit bits.
Theintormation TI, Verston
of earlier the InSynchronization
receiving
ter the logic
of the examined
by frameis each from bit first the formed
by sequence The
four the using channel
by
signal1ng
fora four-state provide possible
to channel
isit, speech
patterns possible descri lor formedin be
unlikely
to is
that pattern special forms:a frame, each beginning
of the atonesignal.bits,framing
gnaling available
for superframe
are per bits two Since length. frames
in 12 ishere bed Superframe each
the thatsuperframe.
Note thewithin frames 12th and 6th thelocations
of delermination
the of Suncee Sequence so
thata chosen are biFraming
ts frame. per bits 193
of
of ueadded framing
is bit total
ot each makesa This frame.
Therefore. frame each where of
alsol2 butbefore,
repeatin
s whiboundaries
ch patastern, specialframeidentification
formaof bitframi
s ng the allows
Theframe. thus
allIOIT100.
pattern
each ofbit fiTherst occupy
the and before inning correctiya a information
bits separatea begins. sure necessary
beto is it
To
js. 125 akes frame each second, samples
per 8000
sampus Because
the information
bits. 192 8= 24. x has frame is
kbitts
as transmitted
8at are biframing
ts The this. otcare take developed
to Superframe,
was thecalled cole Tron sample) (corresponding Each frame. called
a channels is 24
ructure, framing new frames.A signaling the arframes
e which receiver, thidentify,
e at waSry
to one to codeword containing
one segment
seque multiplexed
ain channels
are
5.2A1. Fig. shown in
frame,it sixth every channels
in superimposed
the on signaling
is When stream. bit the located
in De 24the of
each samples
ofcorresponding
to codewords Binary
channels the thatidentified
so boundaries
be frame that onlrequired
y format framing seven-bit older e
Signaling Synchronizing
and
5.21. Fig. shown
in format
is frame The bit/s. 1333 8000/6= of
rate aoccur
at signal each tor bit3aing
s
penalty.
The acceptable an
considered
beto framesis six ofoutimpairing
one suffered
bydegradation DNK
slight signaling.
The robbed-bit called derived
is channel
so signaling the encoding,
and bit calle7d S system. corrier T1
5.20 Figure
chnique This framing
bit. Iandinformation
bits 192 are there frames, remaining the all Inbit. aming
1and bits,
signaling information
24bits, 168 =24 therefore,
x 7ha_ frame, sixth Every diSIgniticant
git. b)
leastcorresponding
the to error possible havea wil signal voice each sample of sixth every that means This
ation. transmit
this signal
to sample
aof sixth every significant
ofbit) least (theinformation
bit one Use
formation,
we this for slots time extra create than ratherrequired, channel
is signaling low-speed rather a
only Sinavailable.
ce longer noisbieighth
t theprovided
by channel signaling version,"
the earlier the in
used output Coder
bits seven the instead
transmission
of for used now are bit[ eight alSince
l effectively. channels the use to
other
eachcommunicate
with toable be
mustswitches systems,
the switching telephone between signals transmit 24Ch.
used
to system
are thisdeveloped
by channels Whenon-hook/off-hook
signals. telephone as
well pulses,as 3Ch.
dialingcorrespondingto signaling
bits transmit toneed webits,framing information
and addition
to n 1Ch. 3Ch.
reframe. to
case)possible worst the (in ms50 about and detect toms6to04 6Ch.
about takes framing
Itbit. actually
the whether
is itdetermine examined
to positionis next the adetnd ected,
format. signaling system Tl
5.21 Figure 7Ch. 4Ch,
2Ch. ICh.
Ch
bits bits bits 2Ch.
Information Information Information
I9, 13, 1,7,
S|6|7|8|1|2|314|5|6|7|8|
121314|5|6|7|8] FU|2|3|4| except
frames All (a)
24
-Ch. -Ch.2 T -Ch.I o23 LPF oo
bit o
Framing
LPFl 240 LPF
Decoder processor
Digital
2 LPF
no.Frame Channel
6
7
medium
IransmiSSIon
bit
Signaling, bit Framing
bits bits bits processor
Information Information Information 7,1, Digital Coder
19. 13,
nos. Frame
21314|5|6|7|8|2|345|6|7|8|
5|6|718 2134|
24
-Ch. -Ch.2 -Ch.I Channcl
295 Systems Carrier TiPCMTelephony:
in Digital 5.3 CONVERSION TO-DiGITAL ANALOG SAMPUNG
AND 294
 of channels
ationship evident a and one (known tributaries,
digitaltransmitted channelsarechannels withmessage systems signaling
suchnaling
of formats. repeatsdetection much data iraming. for falthough,
or 2 traming eachbandw he of network of 296
of tapplieations.
multiplexing
the words bit The aMULTIPLEXING nals, DIGITAL signaling.
Several 5.4 transmitted.
the the
bit-by-bit superframe
extended
SONET-formatted
rate The at 256 The For The more channel todata a The
(ESF)Advances in IwO
incoming North facsimile.
To this framing require SAMPUNG
that a low-bit-rae
networks various
(or
in 2 8channel
as in
of are TJ time asbegin Is Therefore. Conference
(with time is kbi/s can channel.The signal1ng
being
24attractive kbit/s with
exists the (Fig.3R carrier, not American byte basis may over sample) frames be
consists from with, case a idth channel only
channel be eight slots called kbitus the digital
minimum identical, (known or The a wi l
transferred
reasons, framingfoundtrunsport overhead
frames 6., AND
among the wordconsiderhigh-frequency as of and superframe requirement. two-state bits:
5.22c,d). discussed bit signals although bits
per will
significantly
on to
of various digital diverse TDM common in
service for ANALOGTO-DIGITAL
bit three signals, frame. channel is
Chapterinformationhighly electronics inand in
a 00),
interleaving). as rates each).European be including much (framing)
these length
rates, the transmitted a 12, DSI signaling. O1,
hierarchy bit the of (time as can the Each out
channels
Similar
high-bit-rate involves insignals less providers reliable cvche format, 18,
fornmat, and
10,11.
rates. or case
various a channel from of13. Ansignal
and, of Sec. digitized division be bit The Postal decrease he and
24.
the digit medium. rate frame
remaining
tvulnerable the redundancy ESF
of multiplexed, the onemor capacity most three-
Thehence, results 5. 3 , touses indevelopment ESF thansignal is was to
multiplex B, nserng interleaving) Figure tributaries all is interofice
most chánnels perform coding
case be tributaries multiplexing). 2.048 has and the frames24 Although
C, uses bit fomat thecheckingapplicatons Sixteen-state introduced and
CONVERSION
this can
channel
and multuplexed. voice Because Telegraph importance
30
applications. to performane of
of eight-bit interleaving 5.22b 16 theory four-state
Mbit/s, óunterfeiting older (CRC-6) check the
pletely typeframe be D byies needsignal or bits 8x that of caries the
of is the signaling intelligent ESF in most
attained
identical in
shows with combined, per = carTy superframe provided
of allocated as not Administration
corresponding 240 require
signaling lengthduring framing have
scheme must word
succession Byte shown (PCM)., medium
The of the signal signalingswitch-
(except theidentical be frame received as
by information robbed-bit th e repetitive the made
nnels
nous is beinterleaving. interleaving, signals (CCIS) than only and task. it
bit interleaving in the a to network-switching
messages format.
the by is -(0-swilch
combining
proportionately
lowest same. is are error is 1970
is a rate at Fig. computer t h e divided carries
thus techniques
practical multiple from the bit time-shared form used tw0-state unneceSsaryA
from systems.
signaling, patterns CRC-6
theby supertrame tonew
R lowest 5.22a, rates. to (CEPT) pattern More detection possible take
and the one 32for bits, or three
into signaling appliations
of output, various voice
corresponding
frame data pattern
distant signaling usedare
only the of one channels Whenrattehes used Multiplexing or high-bit-rate 001011 discussions advantage
level),
in words, by has and The associated channel, and to
more
slots. on channels, signalsnodes,
when of
channel building 1-by-word atelemery incoming various standardized universal use
specval
certan Iin
synchronization all terminal channels: and kbils 2 structure,
isbits called in
different formed word- SOmetimes in the
some
bit to where eight tothe with a, the the of he
be on and the full
ca incomingsignal, only separate patern telephong
the to deployment bi ts funetion nSF make use eighth
reducea
scussedcormmon (channel
i chanle th e 8
of bit s by dae 30 4 kbws
30 speech a of ESF of
kbiws used
a r e four PCM thsat th e bi
A) voice and each sie.
flater.airsliymple lengtFourhs. -channeinmul!comitipnlgexed. DSI
multiple signal taken bits. bahassis is s
Its

Figure|
interleaving
5.22
Channel Channel
B
channel Time Channel
Channel A
division
having B
I A BL BUU
multiplexing
different of

bit
rate;
digital
(d) (a)
signals:(a) (b) A
alternate (c A,A,A;A_ B,C,
(d) A D,
BA,CiA, A,
scheme
digit B, B,
A, B, C,
interleaving;
for (b) B,B, D,
B, D,A4BA_C, A,
(c). A,C,A, CC,C,C, B,C;
D,A Multiplexing
Digital5.4
D,
A4B, A,D, D, B,C,
word(or D,D,D4
As D,
Cy
byte) A,D,
interleaving:

297
 298
Solution S6:
S5: S1. Example
The 5.7 bits requires
slot thoutput
e
Frame Similarly
signals, S' Instructures.Label Design to in
is the A A S2. output a At
Hence, then e channel.
thframe, the
SAMPLING
for
first the a kHz 4kHz 4 S3.
two-level data receiving
S' for the
multiplexed level diagram analoganalog S4: of receiving
is second-levelsampling
S6. following
each bits. and For
digital multiplexing These with AND
analog) kHzPCM(4 showing signal signal with this
terminal,
analog) kHz (4 S6
Sç
multiplexing system
rate with sources each purpose,
multiplexing data S,:4
kb/s S:kb/4s producing kb/s 4 bits ANALOG-TODIGITAL
PCM 8-bit 4-bit kb/sSa
4S:kb/4 s producing
isPCM are bit to the
rate 8 sources appropriate digital is
b kb/s64
k scheme required partwithin the
uniquely incoming
kb/s32 is
samples/s.
kb/s16 64streams S1, 8-bit/'s4-bit'
amplesample output
scheme kb/s a
ofreceiving
S2, bit in This is
the slot.
synchronize
produced which to so-called
overheadbits. digit
(8 S3, rates be CONVERSION
bi For terninal
can t and time-division strean
PCM S5, and the PCM PCM
250 us be S4 accomplished byin
coding by first
b coding). scan stream stream
S5 (rotation)
Scan
are time must must
and ’ multiplexed.
level
(rotation)
is S'= withthe be be
-Scan 4-bit S6. multiplexes multiplexed.
16 abldivided
e
b Since kb/s
62.5us (rotation) PCM
So 4000/s
rate = rates. framing
adding
correctly
identify to
beginning distributedand
112 = both It
giving is sources
rate kb/s depicted Also
=
S5
16,000/s digital and show of
S1, cach
as synchronizalm and
S6 S2, appropriate lhe o
multiplexed frame,
data are S3, Thisbil.
cach

rate kHz 4 and wIth

32 SA
frame only cach
kbis analog

uyantage sOmeare
Conipcated o efficiency
Ihe furtherIhis ontprovide
lounlikely, bits,
channels Mo. Each of Feure Note Bits BiFrame
transmit are I total athen CA. four Format
Si5.g4nal.1 that
Frame perioBidt Frame
presence this always subframe b'are period interval for interval
ou ne ofadd (12 Fo, channels 5.23 = S, -
guaranteeDy Ol ofidle. that patern, the is 48 placement bl,
x data CA, illustrates
additional identifies mai1152/1176 n 1. is62.5
takingswitching the majority This signal
of has b2, = =F
Thus, 6 a
second us.
inactivity, Mo M framing
the x bits CA, each b3,
=8.928 T6000 0
bits 4 six
MË
[48]M,
(4 8]
MË
account that
M
means of
information demultiplexer the Mo, overhead = from and
overhead at a
of and
bits
S=
operations, cases, M, wil ~
1152 typicala =62.5
|4(48]8]CA th e the four Fo rate notthein b4 us. 250
(48] at Mi also pattern, 98%. each FË. in
numberof us.
system
any notprovides subframes,
are data In of four us,
Cp [48]Cc [48]CB
the have frame
about all The channel). bit between bits: format,
1.544 successive
frames
Figure statistics and given all which bits CA b
is searches overhead 0s for
(48) of incoming a and followed and for
transmission underutilized.
also time, verification bit pattern example Mbit/s. that 62.5us
5.23 each the We these of S,
Fo Fo Fo Fo ofrather stuffing, M, 6
of 101010...The multiplexer for x begin the is |b
the channels corresponding the andF bits 4 overhead by The
DMI/2 (48] [48] [48] [48] signal the DMI/2
careful leastat as of wi th 24
=the
with main unique.
channels We
discussed the 0111 repeatand
Cp Co C CA areoverhead next subframe
ultiplexer sources, system one can, are genuine subscript,Q
uses all overhead multiplexer.
[48] [48]Cc (48| [48] formed
pattérnby 48 bits frame b 8.928us
demanded therefore,channel. active receiver to to 1s.multiplexed
later. a are 1
(multiframe) thereafter.
format. Cp (48]Cg CA planning. it FoFjFoFË synchronize line The bits bit 48 (first
is all are
possible This Fmaking Mo,
interleaved line We
(48) [48] accept the could in digitsalways
will Fig. followed
obviously In time:sequence. lock bits, in have Multiplexing
Digital5.4
F. F F FË not 5.2 3. on are a Fig .
to any more òverhead total
0 and consists here
exceed some onto the
periodic and b
(48] [48] [48| [48] ensure situation
random
traffic by data 5.23)
It so bit-by-bit
involves input those 1176 on. 48bits of
anthe transmit The this isbiframe.
possible, ts has
channels wrong 010101 bits/frame. multiplexed four
acceptably number with Thus, from
bits C MoM1 After overhead
much interleaving
subframes.
data, sequence. arealthough subscript there the
avail more to
locking . .
low take and used M,M1. and The are data four bits 299
 For are
probability 300
nominal
through Signal
isaccommadate From
stuffing.
require wil required accommodate
pulse tiplexed and when rate.oscillate channel vacant Because cause If the tiplexer.
receiver. 5.4.2 a the In the condition
sentHence, the 1000 incoming other known
In
Negative (3)Three the We This cable
through now the tend DS1 slots the cable km
preceding TDMA SAMPLING
we
overhead
rate overhead signal
positive/negative
to i0 channels alwaysshows banks atsignals Jf he
tpulses to This types lasts as of
accommodate lag variants exactly with theextra Asynchronous
temperature coaxial data time this
needmay of appear be is only a
overhead al pulse such or cabie of systems
the (Fig. wil need that in no in 2difficult can data division occurring. AND
at bits. become of xdiscussion.
cable
multuplexer bits. the stuffing a 5.24).
are an even the other the data. pulsestransit
a temperature 10 be fraction
positive tributary
The slightly The asynchronous. e
thsituation. storage equipment. and
bits. the terminal
samedigitalNorth
cannot increases
to m/ s
carrying , even stored employed ANALOG-TO-DIGITAL
information all pulse in These multiple-access
telegraphy,
positive/negative receiver availableAt pulse svnchronously
frequency, arive ofMultiplex
ULUnstufiedsignal incoming it we
pulse ispulses.slower is a some stuffing. stuffing Obviously. (known American slots be takes when and
stuffing
equal complement accommodateddrops, sooner. by assumed Channels 2 transmitted
second, afor
rate knows stage. need transmissiontelephony.structures
Information to about tributaries at scheme asleading the
1°F. 1/200 xthalel
10°
th e than rate a In an network ta h e
thus tsynchronization which
at th e the this
muluplexed rat e second channelspulses
of (TDMA)
some the
exceeding positive elastic he propagation
producing
nominal pulse stuffed-pulse those of system
exist: method to stufed of and late. leads to CONVERSION
stuffed-pulse
positive another the of
at arereceived in per this
times stuffingabout of store) Although the delays
Output stuffed
digits
Transmitted
signal Input rate the will
their pulse (1)
often second. are designsystems.
systems,
of transit Bit performance
acceptable for type
and tributaries, maximum that positive with
muluplexer, a nominally
signal required any pulse decide stufing. anand cause temporary velocity between are
signal is elasticpulse generated pulses Stuffing parameters have
negative left-outa position positions of the dummy Assuming timeand
unimportant.
after
to combination stuffing. the the of
multiplexer that oscillators been
to and incoming rate. thepulse asynchronicity
stuffing data
store wil atall
moothing including stuffing
accommodate and digits willincreasethey the the
pulse isthismultiplexer stuffing, are bydrop, increase developcd
satellitefor
eliminates The
thustransmitted Hence, and must 1thesame are
lag (alsorarely crystal (pulse million incoming
Position
ofandits are and nominal Hence, chosen
at the time has data pulse in rate.
the th e (2) known
received in
quitestuffing). the bethe by
others. multiplexed become so oscillators temporarily pulses communicationspeech
all first slots that
through that time negative rate stuffing the rateabout example, consider channels
For in So
inconmisuljon network. stable, multiplexer propagation overload that
SCnekAll two pulse.
in is as of
this the great the slots
higher justification at pulses 0. 01% wil
will anySystemsand
information S is overhead tributary
sigla pulse synchronous a they stored be and
muuy enoug in work individual in the
condit overload
transmit ed od the thanstuthng wil wil rersi transit.
Thi speed
even h. t mul-
o cau mul that to not in

Signalofrate consisting
a eventually rate 4 Japan.and
data-rate 3 the bit inthis wedpis
that and alloCa
44.736 6.312 rate There kbit/s. Two a sA3 F,the e line so The
at of for e
a channels.
The major
telecommunications nresent bit. last irst tocono. nvey
per
Mbit/s.
rate Mbit/s. 1.544 oftransmitted are Plesiochronous For
t48-bit 1.The Cdigits
DSOkbit/s642
24 four multiplexed classes the informat
No. ionframe.
139.264 Mbit/s. channels the in
intormationstuffinginsertion
Channels Finally, Seven orders, It digital second
signal 1.544
DS1Mbit/s 1 multiplexes
over
sequence bit
This Fig.
Mbit/s. DS2 Four of of is about
of is5.23
igure three 64
or signal, multiplexers
a hierarchy subframe,associated
in
indicated any
voice-grade (Fig. are
There DS3
signals DS1kbit/slevels, stusuffi
that stuffed ffingcient used
5.25 bank or
signals called
channels the
switch
digitalChannel 5.25). (almost subframe.
signal6.312
DS2Mbit/s 3 signals are each. of developed with by in to to
is
multiplexing/The are stuffed pulse
North alsó
multiplexed are channel. "digital The accommodate
the using channel transmit
The ofused
aare
multiplexed immediate Osin
erican lower output rates North by
Synchronous) bit For
signal 1/DM
2 multiplexed signal in fo r any A, stuffing
44.736
DS3Mbit/s The practice. the wil the
rate ofAmerican all one bits
digital
by of Bel l thebefirst the expected
sécond level up
following Cg
multiplexing DM2/3 a bythis first to System subframe,channelsubframe
thethree information.
by a 9600 The convey
erarchy multiplexer
DMI/2 level class 0" secondbit
DM
2/3 a (DS0) hierarhy
DM3/4NA
multiplexer and Cs. variations
is
is of bit/s first Digital informationIf
(AT&T 1/1CDM multiplexer the category the denoted
hierarchy, multiplexers ininto oúrrently a Only
is T1 is followingstuffed
the FË beenhabist
a the inone
System). 139.264
Mbit/s 3/4NADM implemented 5.4
multiplexer multiplexer to DSl a the
Northsignal is Hierarchy by
stuffing
about instuffed
Mbius
3.152 known yield to bit Fsetting input Digital
signal
DS1C (digital used included
yield is wil bit,
a American of
at for stuffed,the
as toDS3 a data immediately that signalMultiplexing
bit
a
level or in andbit,so all
the yield DS2 channel much combining North thein is, the per
signal the rate. input
digital hierarchy, rate ANSI Cs
a signalsignal1) higher of America
stuffed three
first channelin B,
DS4NA at The
channel
data a bank, on.
follow
rate at
uplow stan such bits 301
a at bit is to bit
 On.SO 302
Consider
the e
inabl139.264 offour into has 24plesi. gnal DSSUst
O aem1
inues.(DPCM)transmit. In PCM 8.448 channels The
successive the the to
lower adopted. anbeenEuropeand In all 64
In In general.
5.5 interface E-l of signal(DDS).
Nyquist other analog is world, the inputs SAMPUNG
a DIFFERENTIAL is not Mbit/s.Mbit/s. form
level
linescarrier incoming of kbit's for
simple onethese Many Figure Fig. to which
words.
sarnplesamples.messages very a with and an kbi tU s
64transmission
a
such 5.26 atThis many of TI AND
different
ideas 5.26 an E-3 hierarchy.
2.048 appropriate
channels
scheme: E2 34368
Mbit/sE3 139.264
Mbit/sE4 one provides
the scheme. eficient Mbit/s
2.048 EI demonstrates E-5 other each. multiplexer need
values. exploit Single-user
line ANALOG-TO-DIGITAL
Piesiochronous 8.448 worldwide
Propersample we another line line MbiUs
ideas PDH with
one
instead can system parts Some format standards
ofcan through
Mbit/s x4 with higher
level the
Thus, the have y4 across
exploitation values make data (30based onof of
characteristics because PULSE the data channels). the themDM1/2 the
if of are a been digital relative throughput
the
not
network. for
transmittingm good Y30 throughput world,
(k] kbit/s64 44.736
Mbit/sT3 three line
muliplexing may restricted
be be
multiplexing
not proposed hierarchy multiplexer
transmitted.
independent,
is of guess it
CODE Canada
USAand CONVERSION
generates relationship progressively, isanother be
the this X different of shown
about of 24 1.5 44
redundancy
kth the the to (PDH) T1 T2 of
34.368
improve 565.148 hierarchy,
sample,sample a So telephone in 30 MbitUs need The
only digital
source according Mbit/s
1.544 Mbit/s
6.312 systems
sampleMODUATION T-Carrier and
and many Mbit/s, not case to
values,leads the Mbit/s. generatinganFig. digitized signals
instead generally value
signals. the recommended 5.26. theby digital beof
bits to points(North
encoding ITU-T 32.064
Mbit/sJ3 97.728
Mbit/sJ4 an channels DSI the
to and J1 J2 Starting with
we from Japan Because E-4
fromthe appropriate signals of
signals highervoice
encoding
of a: there Differentia Recommendation American,
Japanese, of
mitting
difference
transmitthe requires their rates
line E-2
knowBedge efficiengy
iya pat of of channels
levels
different with line obtained as
common 64
great so loW
signal pulse throughputdataof withbase kbit/s TU similar. diga
exan.For alone. Any
is
bandwipách
ofdth to networks as
m deal interface. a (E0 byt 2.4
[k|. . datlevel as
with code A/D formmal,
an
multipleitg
of a
standar
othetand throughra channels
wefewer conversion. must of
al ation E1
transmilbetwe nredundanybsample vmodul be

formula approximately,
[m(kT,) and Tuture Equation m(t inldescribing
ike, Before Spirits DPCM.DPCM), tPnstnuct
he lie revious a the the dimk ferencethe
Trom foPastas e signal ahelp uninitiated, difference
+ us. this who We
(prediknowl
ction edgebandwiquantdtrhe)ducii,zatnigotnhe siatmerplateively
Glendower:
hotspur: approXimationSoWs m(kT, s1gnal T,) pursuit. quantization 1,
Shakespeare, m(t), What can
e (5.42a) as canfuture of where is ksample at we
e
bur Why,
m(kT, ± Taylor, superior wil the error) improve we
of values. the can
Samples the that value which summon
is Not the between samples can d[ k]
will socan I higher T) as be values, several
Henry we - shown shows more, prediction
DPCM, estimate d Thus,
interval receireconst
ver. ruct
they can call has quite close [k] upon increase =
we
order incan T)]/T,, =m att
Maclaurin, tothe
at
cone l,the IV, Eg. the and = the [k] m
or use, [kt in +that derivativesunlike so! help to the previous noise, Av
spirits
so Part find T, . m(t we successive
naive this Now.
derivatives of m then m the -
when can
[k] m the (5.42b) we1],
Eq. In from + Wefromseems a receiver [k}m|k]. peak=
from 1, Shakespeáre's shall sample |k ], the [k]. mm
you any Act ~better crude a and fact,
obtain (5.42b). T,) can prediction generate samplescheme SNR, which m,/L, (k-
a,
do vasty IlI,
m
[k a
of
the and amplitude difference Thus,
ajm improves soknowledgeeven ~ likebriefly and 1|.
call Scene in + m(t) m(t) = all alsospirit valuesamples. iteratively. or for
deep. [k wi l prediction 1] on. Let mystérious their byfor is
for the if orders summon m values.
At given a from At
them? I: be = Settingt us we + + world. diseuss Wiener described |k] estimating the
1|+ series, as 2mm[k]+ denote T,m(t) spirits, taken
isdifference th e given
a given m,
the of T,m(t) Consequently, by between
by knowledge
we [k] know at If ofreceiver.
prediction. of the t. our
adding receiver If L the
aym weadd the
Electrical stuff. the as th is SNR,
Av/12. (or
the m[k- - =the justsignal +m) Using our
[k- require spirits approach theinthe prediction (predicting) transmitted successive
more (k kT, kth spirits (prdiction n),
the smallforT, fit previou_ the estimate
also, we knowing
2] + samplein and th e This this of 5.5
+Thus, moreterms l)th 1] first engineers of only preceding this receíved can the
Eq. its + Taylor come
Taylor,
derivative,
(5.42b). derivatives
for to scheme, is weisreducemeans
th e
reduces difference Differential
values
samples
+ samples in sample in of 3 T, signal sample worth
error) determine m [k] d
aym general, m() -m() + series when psychics, paragraph,
the Maclaurin, appear d k].value
[k] that the is and
[k series from and by called.* prediction value, known d its then (or n
in f or [k ] to for reduced isseveral Pulse
mwe at
to
salt, the the
transmission quantization
of generally [k], d
N] wethe the
on recognizing [k]. instant can this wizards, wi l we th e a Code
can past. be that
which as estimate
th e tw o Consider,
signal, Wiener, the estimate
transmit given we
express
that still
hopelessly is, the
predicted be kth previous
right-hand The is. t,
(estimation). considerably. can Modulation
previous predict we mediums, differential mis even sample interval n much
larger that m we and [k] a
the (kT,) can special î bandwidth). (or reconstruct
[k]. [k] m the
= smaller transmission smaller sample
i(kT,) this
can for the
example. outclassed m (estimated) from m
prediction thsamples. e side. a Thus. difference
from [k]
(3.43)
=predict (5.42b) (5.42a) express like and To [k Av,
Because
num m
value case PCM
-
To [k], th an wethe thus than [k|value m 303
~ to the the 1]. of
)4
205

fhie ie the yuati ol an Nilh ode prehenIge Nwld ieul

And he ftel tlet elin je elnple tinne dely. Predictor

We aYe itlined beiE a VEry cinple prmedore fot r i o deeigo nt
dietionewfiiente t, in Fu ( 44) Afe detemined ton the clatitical fur best etiehon,
The pwedi ntesctibed in q (5 44y ik called alincon pedi to Iis orslation
basertnlly a between valous
Inpt
elsy jine ), whee ihe tsp gaine ATe set equal tuthe tAvetAnl
peditioneellicienss, W8 show in y 3)
reictor
InutnkDelay Delay Delay Delany (b)
ny
Figure 5.28 DPCM systorn (a) ransmitler; (b) recoiver

where k|is the quantization ertor. Tbe predictor output , k] isfed back to its input so that the predictor
input m, 1k] is

Output mk| m(k| +qlkl (5.47)

Figure 5.27 Ironsversol iler (topped deloy line) used os o lineor This shows that m, kl is a quantizcd version of mk|. The predictor input is indeed m, |k|, as assumed. The
predicor.
quantized signal d, |k|is now transmitted over the channel. The receiver shown in Fig. 5.28h is identical to
lhe shaded portion of the transmitter. The inputs in both cases are also the same, namely, d, |k|. Thereforc.
Analysis of DPCM e predictor output must be in, k (ihe sane as the predictor output at the transmitter). Hence, thereceiver
As tnentioncd carlict, in DPCM we transmit not put(which is the predictor input)is also the same, viz., m, Ik| m|k| + q[k], as found in Eq. (5.47).
m|k and its the present sample m lkL. but d|k (the
predicted value n difference e s shows that we are able to receive the desircd signal m|k] plus the quantization noise q [k<. This is the
the rcceived d|k]is added to |k). At the receiver, we gencrate in k| fron the past sample valuesto which quantization noise associatcd with the difference signal d|kl, which is generally much smaller than n (k].
the recesver, instead of the past generate mk. There is,
however, one associatcd with this scheme. AI
samples m|k-|. m|k-- 2,.... difficulty
as well as l |k|. we have their quantized
e received samples m,Ikj are decoded and passed through a lowpass filter for D/A conversion.
versions m,|-). m, |k- 2}).... Hence, we
cstimate of the quantized sample m, Ik1. in terms canot determine éMk We can detcrmine only mgths
of the quantized saunples , |k - ).
SNR Improvement
To determinc the Improvement in DPCM over PCM, let
will incrcase he error in M, |Kinle m, and d, be the peak amplitudes of m(t) and d).
reconstruction. In such a case, a estinate
better strategy iss to determine m,Ik, the ... respect
factorivelyd,|mp:
. If we Because
of m, Jk] (instead of m(k), at the use the same value of L in both cases, the quantization step Av in DPCM is reduced by
The difference d|k]l transnitter also from the quantized samples mql |k-2). the the quantization noise power is (Av)/12, the quantization noise in DPCM is
and fromthe reccived =m|k] - m, k] is now
d |k], we can transmitted via PCM. At he recciver, we can generate m,|k reduced by the factor (p/d,)', and the SNR is increased by the same factor. Moreover, the signal power is
reconstruct Wem,shall
Figure 5.28a shows a DPCM ransmitter. [k. sOon show that the predictor input is mIkJ. Naturaly, proportional to its value squared (assuming other statistical properties invariant). Therefore, Gp (SNR
its output is im, 1k]. the mprovement due topeak
prediction) is at least
predicted value of m, k. The difference
(545) Pm
d|k]= m G= P
(k}-n, \k]
is quantized to yield where Pm and Pa are the powers of m() and d(i), respectively. In terms of decibel units, this means that the
(546)
SNR increases by 10 log1o(Pm/Pa) dB. Therefore, Eg. (5.41) applies to DPCM also with avalue of athat
is higher by
d,[k] = d{k} + qlk] 10 logio(Pm/P) dB. In Example 6.28, a second-order predictor processor for speech signaBs
 images mav 306
For Solution
Fr transmitted erorAmpitude
if (or eror orzation the Adaptivetive Ihe Hence, and the Ther Sarnpling (aCalculate
TransnisSIon Example5.8 biperts heanaByis
the the smali Av 5.6 PCM. signal
It is quantizer
number compression TransmissionmiaimumL the DPCM, telephone
sampie.
Thus, szed.SAMPING
quantizedlargest size. is too error high
important
epending smali. ADAPTIVE A =2number rate 8-bit of Alternately. thfoer For
For DPCM satisties let 8-bit/sample
the dr) using
is of positive bit =f. a asthS
negative too 32 the codingtransmissionranges signal 25
samplesexanple, Therefore, quantization the at
ratio= of rate AND
large (ADPCM) bit step = dBCSE.
to on encoder. quantization = 2 mi)
Av+0.25 rate integer Bimplies from in
whether
note because
B
size x such the ANALOG
oscillate value when on =8x 4kHz is
that DIFFERENTIAL =
be -0.1Vto 10 sane SNR
), it levelsFigure can the such
Ar. the bit PCM. the cases as
itthe the thewould 40>10" bdx Av channel levels = rate systems
indicates Av further Step 8xnumber Sk range SNR.
improvement TODIGTAL
nearquantized
quantized
pred1ction be IS is5.L29 condition.
i0 that Suppose PCM in The
zero. toofixed. = = size samples/s
10
betier illustrates improve Av-Ar Ar= B, 2 L 0.1 same -1\ thvoced
usingshort-terme
that 16. = of V.
big = quantization hit
then
prediction When
= 25.6 == Ai=
Ar
64 DPCMand ln
tprediction
he error for or 8x signal to DPCMcan ratefor CONVERSION is
the kb/s DPCM
the
prediction the the x 32 0.2 10which = 1
V. found
r
foquantizer athe
fixedefficiency 100 D Av. 256
rediction quantization basic PCM 6.25mV
= 2 isThe
quantizing
error error x transmitted
Av DPCM
could he to
= 5= gives levels. henceand maximum speechand
often spectra
quantization
configuration X =7.8125mV is be
error
samples da cannot 20% 40kb/s x 10
error of allowed operate at 5.6
[k] =5. = L,
is isstep DPCM is dB.
large the frequency
using the in
is vary canlarge cover Hence 25.6 2
small be
adaptiveto
Av = bit to lower ln
incorporating:
stepencoding
is by
of but rate be
differential 32 practice.
and closethto a or he
ADPCM. 5-bit 256. inI
and
egoodsmall.
be Av L' within or
than
Avneeds should compresSion even the
A" indicator necessary gnalrang
siapplied, coding spectra
that the
25% PCM
signal 24
decrease.
tolargest
needs grow. to For be kbi/s. for SNR
is 2. 1(DPC) of
So pracca PCMby
improremex of
of that either used and ralio.PCHthatin is lon-actmtY
the in kHz. 4
Aris the rmuet.
an DPCM where to4
larg quant adap- h
Converseposly itivpre veadiluectied
which tour Osadvantages, 80
disappeared.
Tanzania,
Nigeria,
Mali, auvely
NOTAS, Irequency e asmechanism
the DECT used istioncodec, respect of attes etor nsech loss.no For
DULATION Sample
DELTA3.7 million Xiaolingtong Personal ADPCM important
NTT PHS
widely(IMA) ADPCM iThevely. bits, 2
for atuisescoder ne
results
times deployment mitedcaled is
residential the most 16,
examplCompar
e, eaddaptive
at
corelation Laboratory 0s now usedMicrosoft 3 the quantizer that FiADPCM
gur5.e29
the in PHS band,Handy-phone a to ADPCM, speech bits, 24. an it
2006.Surprisingly, in the common eighth-order and
encoder
in the mobile in same iswith both
Nyquista phones used China. in use main the 4 32, decoder very and
rediction
Small
error "poor and codec. bits, and the
used network
in
in DECT ADPCM also
sampling DPCM,
has the
and mainly over
networks, business ADPCM and 40 common modulator
rate) in (NTT-Personal. man's Japan receiver
Inand
PHS System known In predictor. (called many
China base addition kbit/s.
5 rate. ADPCM encoder
theDPCM 100 (digital codec
bits, mk|
Honduras, hascellular" insystem
in countriescordless
practical ADPCM in
baseband
signal. (or DECT as speech codec) reconstruction
This quantized
the and
This is alone, seen stations1989,Japan, practice adjust
enhancedsupported the respectively, They For can uses
further similar PHS), to eas1ly further the
thereDDI-Pocket, PHSChina, offersphone DVI encoders the correspond different
applications. for an
thatcan resurgence
where abecause are worldwide. for receiver adaptive
exploited also ADPCM. ITU-T speech represents an
is tocommunications. cordless cordless by
was lowits low-power, Taiwan,
much marketed
a or compress
8-bit can
of cellular WAVE use quality have predictor orderNth Adaptive
quantizer
be an its G.726 equivalently. signal apply quantizer
deltain inand Another 32 to
encoded
increases explosive limited simpler telecommunications) voice, The The PCM
a access
cost
markets
oflikeshort-range and players, kbi/s. four the
ASTEL) network, as specification,
32 levels, 2:1
elsewhere themajor fax, different
samplesITU-T sequence
bandwidth encodednumber same controlled
modulation
(DM) range to kbit/s to channelTo
using the deployment
expansion Personal Designed In
and quantization G.726 the
correlation
between
adjacent user data, practice, standardat algorithm
only
inimplementoperating
China,Japanfacilities. and ITU-T the to of same only
bit 8
in Access ofand
Interactive these specifies kHz.' or bebits
by
onebi(L
t of poor Asia. for sizes
Vietnam,
Taiwan. in multimedia the system, G.726 there G.726
storage needed
the Modulation
Delta5.7
subscribers, and and in short-range
32 levels to
oversamplingby hardware costs roaming 1995 The
Originally include for The encoder
cellular
deploy.
Unlike the Systemkbit/s ADPCM are four reduction with for
but service which
Multimedia ofeach G.726 specifies into samples.
Thus,2). = 1880 muitiple different
ability. ADPCM
(PAS) the 4, athe output
reaching has communications. signal
a
developed
is by to
useitself
OKI speech
8, ADPCM 4-bit Av
Bangladesh,since often 1930 and asspeech 16. an identically. bits.
(typically offsetthe PHS an is variations
ADPCM ADPCM waveform. Hence,
samples, codec Associa andsample ADPCM virtually
DM
is nearly nearly pejo MHz known accesswidelycodec
first 32, pre 307
is the
 earlier.
quantized
samples.integrator
output becauseof byThis Proetng Substiuon DRVnFrg N
s ea
the
impulses,
Figure shows N
he strengths
is DM.
m,the ThusDM albD
first-order 531 that tteratvely this we
{k].
feedhack then DMzaaN
bi
shows which of the quation
the
the rmiver wia R
predictor acumulator
input in sr
a whenporion this d
practical into ae
(demodulator) maner, traing d is a
passedimpulses Eq ft
is of eyeàtr etng DCM
ementation the (S4S)
replaced through (receiver) and (muiatr
fr ant
modulator (sum yields the hat
assuming hih n
of just is prtr, hasetand a us
by
lowpass a the may (
a (which areas s al,
low-cost of bean zero
deltathe tilter accumulator
realized s signal. th wo
under
is initial ransnuttr
n
identical
shown earier, kvels
egrator yields er
the by condition, (demiulator)
modulator an in (
(adder). Fig. is
impulses).
the integrator
to just =
and )
circuit desired the 5.30,
If that a the tor
and
demodulator). the fnom ar mee
time quantzation
(such signal We is,
may output m, identicaldelay
emodulator. whih
as [0]
also its =0.e of This
an we I,
The output to stategy a ot
wntcan (the
inegratRCAs replace those |Rm
is sang
(
discusw denxtInunulatht ith wanthseut e
tN

roducesa COnstant
signalof amplifer
at
binary demodulated
modulator
(Fig. The e,
rate a Cmoreme
signal
of
f signal in
train the
samples (L feedback eta
of =) d()=Sgnal puse
narrow amplitude modulahion:
per hat m() m(r) 5.3la) rains:
path. Start-up
pulses
second,needed is is (e)
-m,() E, consists
compared modulation (a)
and Let and
d,R] Ito tu
us lntegrat
where if (b) anpliet
d) is see of
generate applied delta
(to f, with how a errors ntegrator
Ampltier
imulate negative, is is comparator demodulators
tvpically a to the thìs
1-bit feedback
a (d) (b)
pulses) comparator. delta
comprator
nnuchDPCM. the and Sampier
moulator (e) WLow-ass
signal a message
sampler fiter
positivewithhigher
a The (which d)ll
conparator works. signal
than
output is in
Nyquist rate.
the positive.serves the versus
overad
-Sioe
pulse output is dretpath Moulation
Det5.t7a
-E. as inegrotr
the a
Thus,predicted
is
when m comparator
sampBed evtput
sampler
thusThe the and
signal)i)
ditferenn Is
a an sigal
m)
and a by integrator
a vutyat
sapler
isa 309