EmpirIcaIAnaIysisol BroadbandZ^Z MIMO

ChanneIsInOutdoor-Indoor 5cenarIos
O. ernndez, M. DomIngo, andH. F. orreS
Departamento de Ingenieria de Comunicaciones, Laboratorios de I+D de Telecomunicaciones
Universidad de Cantabria
Plaza de la Ciencia sIn. Santander, Spain
E-mail: [email protected];[email protected];[email protected]
Abstract
This paper presents the experimental characterization of a wideband Z^Z MIMO channel in typical indoor-outdoor scenarios in
the . GHz band. The transmitter array emulates a base station positioned on the terrace of a building, and the receiver
array, situated inside a nearby building, represents the end-user terminal. In a previous work, the authors analyzed the
advantages attainable with a Z^Z MIMO channel compared to a SISO channel for the case of narrowband wireless systems in
these types of environments without line-of-sight (NLOS). The new results for the wideband channel show that as in the
narrowband case, the capacity degradation due to penetration losses can be practically compensated for by the gain of a Z^Z
MIMO channel. Furthermore, the frequency diversity attainable in the case of wideband systems appears as a new
contribution, reinforcing the conclusion obtained for the narrowband case.
This paper also presents a stochastic Z^Z MIMO wideband channel-modeling method that represents a novel simplification
with respect to other wideband MIMO models. The resulting models are accurate, but also easy to implement and
computationally efficient. Using this method, outdoor-indoor channel models are obtained from a measurements campaign on
real MIMO channels in different environments. The empirical data and the mOdels presented in the paper will be very useful in
the deployment of future and present broadband fixed wireless-access systems (BFWAS).
Keywords: Electromagnetic propagation; land mobile radio propagation factors; land mobile radio diversity systems; MIMO
systems; communication channels; channel models; WiMAX; broadband fixed wireless-access systems; wireless LAN
1. Introducti on

t is widel y accepted that MIMO (multipl e-input multipl e-output)

systems - which use multipl e antennas at the receiver and at the
transmitter ends of the radio channel
-
represent one of the most
suitabl e technologi es for providing the ever-increasing spectral
efci ency expected for the new generation of wirel ess systems. For
the case of wideband systems, the use of MIMO combined with
OFDM (orthogonal-fequency multipl e access) is the most prom
ising strategy [I]. In fact, the new wirel ess local-area networks
designed to ofer metropolitan coverage (WMAN, wireless metro
politan-area networks) consider MIMO as a fndamental element.
The wirel ess metropolitan-area networks are evolving towards
networks capabl e of providing fxed as wel l as mobile services. In
this sense, the most rel evant standard is IEEE. 802. 1 6e, commer
ci al l y known as WiMAX, which adopted a MIMO-OFDM physi
cal l ayer to provide broadband fxed and mobi l e access [2].
In this context, the outdoor-indoor scenarios attain great rele
vance. The possibility of deploying indoor customer-instal l abl e
end-user terminals is focusing the attention of operators, due to the
cost savings that can be obtained, as well as the obvious advan
tages associ ated with terminal mobility [3 ]. However, the range of
coverage for indoor terminal s is lower than for outdoor terminals,
with a reduction of between 65% and 75% in the 3. 5 GHz band
lLLLAntennas and Prpagatin Magazine, Vol. 52, No.6, December 2010
[3 ]. This reduction of coverage range creates problems for
deployment, mainly due to the l ack of symmetry between indoor
and outdoor coverage. In a previous paper [4], the authors studied
the capacity reduction of SISO (singl e-input singl e-output) na
rowband wirel ess systems in outdoor-to-indoor environments
without line-of-sight (NLOS). It was shown that this degradation
can be practically compensated for by the gain of a 2x2 MIMO
channel. For wirel ess operators, WiMAX is a highly adequate and
interesting technology for deploying metropolitan-area wirel ess
networks. The l atest versions of the standard already contemplate
the possibility of using two antennas in transmission and/or recep
tion. However, for the correct deployment of these it is necessary
to know the radio channel and to dispose of the channel model s,
and to establish the expected advantages of using one or more
antennas. This paper i s one of the frst publications to provide thi s
type of information on these scenarios.
Despite the importance of the outdoor-indoor channel , there
are few published measurements. Among the most rel evant are
those carried out by the Uni versity of Lund in the 5. 2 GHz band
[5 ], and the measurements presented in [6, 7] at 2 GHz. These pre
vious experimental results on outdoor-indoor MIMO channels
-
apart fom being carried out in other fequency bands
-
focused on
other aspects of the MIMO channel: more-theoretical aspects,
mostly designed to contrast the di ferent existing MIMO channel
lSSN1045-9Z43Z010/$Z5Z010lEEE 55
moaels. This paper focuses on the 3 . 5 GHz band in which WiMAX
systems operate, a band for which the authors are unaware of the
existence of any other outdoor-indoor MIMO measurement cam
paign.
As previous work, the authors developed an experimental
setup based on the use of general-purpose instrumentation for
wideband channel characterization. As a second step, single-input
single-output (SISO) wideband channel models, using the tapped
delay-line (TDL) approach for outdoor-indoor environments in the
context of wireless metropolitan-area network systems, were
developed [8]. Now, in thi s paper we present the experimental
characterization and modeling of wide band 2x2 MIMO channels in
typical indoor-outdoor scenarios. A 2 x2 MIMO channel measure
ment system is presented, which was developed using general-pur
pose equipment that can be found in any reasonably equipped
microwave laboratory. Thi s i s in sharp contrast with the measure
ment systems devel oped ad hoc by large laboratories or companies,
and which are stil l very expensive. The design of the measure
ments campaign and its subsequent implementation were of a
clearly practical nature (or engineering perspective), which, with
out compromising its scienti fc rigor, focused on veri ting and
highlighting the benefts that a relatively simple MIMO system
(only 2x2 antennas) might have on wireless metropolitan-area
network systems. This ofers data of practical use, such as the
expected penetration losses in bui ldings in the 3 . 5 GHz band, and
the realistic capacity values avai lable in the outdoor-indoor 2x2
MIMO channel .
These new results showed that also i n the wideband case, the
capacity degradation due to penetration losses can be practically
compensated for by the gain of a 2x2 MIMO channel. Further
more, the fequency diversity attainable in the case of wideband
systems appears as a new contribution, reinforcing the conclusion
obtained for the narowband case. It was experimental ly verifed
that in the case of broadband systems, the diversity gain in the fe
quency domain improves the outage-capacity levels of the system
with respect to narowband systems. This phenomenon was theo
retically known in ideal channels. However, the channels analyzed
were real channels, which combine spatial and frequency correla
tion. There are very few experimental results on this gain in real
channels in the literature.
Finally, the paper presents a stochastic 2x2 MIMO wideband
channel-modeling method that uses tapped delay lines (TDL),
based on [8], and a simple spatial-correlation matrix. The modeling
method i s compared with measured data, obtaining good results.
Although the number of measured scenarios was not big enough to
obtain a complete model for general outdoor-indoor scenarios, this
preliminary model is of great value as a realistic reference, due to
the lack of information about outdoor-indoor channel performance.
Tn fact, the model has been used for simulating di ferent space
time processing methods over reali stic channels [9]. At present, the
authors are developing new measurement campaigns in order to
generalize the model .
The rest of thi s paper i s organized as follows. Section 2
describes the wideband measurement system and the scenarios in
which the measurements were made. In Section 3, the experimen
tal results are presented, as well as the analysis of the major MIMO
channel characteristics. These show that capacity degradation is
practically compensated for by the gain of the MIMO system. Sec
tion 4 presents the MIMO channel-modeling method fom a gen
eral point of view. Section 5 describes the application of the gen
eral modeling method to the outdoor-indoor MIMO case based on
tem and the study of its capacity. Finally, some conclusions are
drawn in Section 6.
Z.Scenarios and Measurement System
Z. 1 Scenarios
In order to perform the characterization and modeling of
MIMO broadband fxed wireless-access channels i n outdoor
indoor environments, several measurement campaigns were carried
out in the 3 . 5 GHz fequency band, with a bandwidth of 250 MHz.
The measurements were made between two buildings of the Uni
versity of Cantabria, separated by a distance of 120 m (Figure 1).
The transmitter array was positioned as a base station on the ter
race of a building, while the receiver, acting as the end-user termi
nal, was placed inside the other building, in three scenarios with
di fferent characteristics.
The frst scenario was an assembly room: an open area with
few obstacles around the receiver array. The second scenario was
an electronics laboratory, equipped with computers, workbenches,
measuring equipment, and cupboards. These two scenarios were at
a height that was around 6 m lower than the transmitter. The third
scenario was a typical ofce foor, at a height of 15 m bel ow the
transmitter. The infuence of the l ine of sight between the trans
mitter and the receiver was considered in the channel characteriza
tion, placing the receiver array in three di fferent local areas for
each measurement scenario. The frst measurement area was
located close to a window, opposite the transmitter, i n a quasi -line
of-sight (QLOS) situation. In thi s area, the line of sight was
obstrcted only by a window. On moving away from this window
and going into the room, the visual line of sight between the
transmitter and the receiver was lost. In these conditions of non
line-of-sight (NLOS), measurements were made in two local areas:
NLOS, in which the receiver was already indoors in the building,
but only one of the inner walls obstructed the direct sight of the
the measured data, detailing the efects of correlation on the sys- Figure 1. The measurement scenarios.
56 lLLLAntennas and Prpagation Magazine, Vol. 52, No.6, December 2010
AN _____________
Signal Generador
Transmitter
Agilent ECG4433B
Reference
|U%HZ
OdBm
Amplifer
Sector Antennas,
G=15dBi
Main
Controller
Omnidirectional
Antennas, 0 dBi
Receiver
Agilent
PNA E8362A
Figure 2. The wideband channel-measurement system.
transmitter; and HNLOS (hard-NLOS), in which the receiver was
also located in an indoors zone of the building, but with several
partitions between the transmitter and the receiver. The foor plans
of the three scenarios and the different areas measured were pre
sented in detail in [4].
Z.Z Measurement System
A wideband measurement system, based on an Agilent
ECG4433B signal generator and an Agilent PNA E8362A network
analyzer, was used. The system performed a 400-tone frequency
sweep in the band of interest: 3. 385-3. 635 GHz. The two ends of
the measurement system were synchronized in fequency and
phase thanks to an exteral reference of 10 MHz, generated by
rubidium oscil l ators and discipl ined by GPS receivers. Figure 2
shows a schematic diagram of the measurement system used. The
transmitter array was made up of two sector antennas with a gain
of ! dBi , separated 1 0, and oriented towards the measurement
scenario. Meanwhile, the receiver array was composed of two bi
conic omnidirectional broadband antennas, spaced 1 . 5, apart,
with a gain of approxi mately 0 dBi for the 3. 5 GHz band.
The signal generator of the transmitter stage performed the
fequency sweep with a typical power of 0 dBm. The generator
output was ampli fed by means of a broadband ampl ifer with a
gain of 30 dB, to compensate for the propagation losses. For the
measurement of each MIMO channel, two sweeps were made
sequentially, one with each transmitting antenna. The selection of
the transmitting antenna was made through an h switch.
At reception, a Ietwork analyzer performed the fequency
sweep in parallel with the transmitter, capturing the 400 fequency
tones. Taking advantage of the two analyzer inputs, the reception
was performed simultaneously by both receiver antennas. The
measurements were made in the absence of moving persons to
lLLLAntennas and Prpagation Magazine, Vol. 52, No.6, December 2010
maintain the time-invariant

cha
nn
e
f
behavi
o
r during the time it
took for one snapshot, which was 30 seconds.
To obtain the transfer fnction of each of the MIMO sub
channels (H {o),H_o),H{o), and H{o)), the MIMO
channel was measured in 32 locations for each local area.
The confguration and control of the measurement system
was performed centrally from a personal computer, with an ad
hoc-developed sofware tool . The fequency sweep was synchro
ni zed in both stages from thi s PC, and the corresponding transmit
ter antenna was selected and the received data were stored and
processed.
J. Capacity of Measured MIMO Channel s
From the MIMO channel measurements, the channel capacity
was calculated in the three local areas of the di fferent scenarios.
The MIMO channel capacity depends on the values of the channel
matrix, and on the signal-to-noise ratio (SNR) in the receiver. In
order to evaluate the infuence of the correlation on the perform
ance of the MIMO channel , a comparison of the capacity in each
of the three l ocal areas was made for each scenario. Thi s compari
son was performed for two conditions of SNR: one for a fxed
typical value of 20 dB, and another for the site-specifc SNR val
ues measured in each local area. When comparing the channel
capacity in diferent scenarios at a fxed SNR value, the efect of
the correlation on capacity was identifed. In the second case, using
the measured SNR values, the combined effect of the correlation
and the measured signal-to-noise ratio was analyzed.
J.1 Capaci ty for Fixed SNR
First, the cumulative probability distribution fnction (CDF)
of the broadband channel capacity measured for a constant SNR of
57
Z dB is presented. The capacity results obtained with a common
SNR all owed the analysis of the capacity degradati on due to the
infuence of the spatial correlati on when passing fom line-of-sight
to non-line-of-sight zones.
Figures 3 -5 present the capacity statistics for the three l ocal
areas, for each scenario. These fgures also include the cumulative
probabil ity distribution fncti on of the capacity obtained fom two
theoretical wideband channel models, used as a reference. These
two models presented uncorrelated Rayleigh fading in the spatial
domain and a di fferent degree of correlati on in the frequency
domain: one totally fat in frequency ( fat fading), and another
totally selective in frequency (fequency selective). Jointly and
inseparably, the measured channels underwent the efects of the
spatial and frequency correlation, al ways showing l ower capacities
than those of the reference channels. It is worth noting that for a
constant SNR, when the situati on went fom line-of-sight to non
line-of-sight and then to the more obstructed hard non-Iine-of
sight, the capacity curves were cl oser to the theoretical curves.
This was due to scattering richness of the more-obstructed situa-
o
o
oz
jo c
c
V o
,
a
o
(
2
Q o)
.

....._.......

..
_
......-
_
-*-

!
'
* ..

i
.
. .......
_
1- ......

........|.
oz - -- -
o+
Figure . The MIMO channel-capacity cumulative probability
distribution function in the assembly-room scenario in line-of
sight, non-Iine-of-sight, and hard non-Iine-of-sight areas,
obtained with a 20 dB SNR.
o

"""*_
o
I
"-" .--
oz
i
.
_oc
c
o
,
a
c
o(
Q o)
oz
o+
o
(
c
'
. .. ....
i
.. . +++++
--i--
I
o=+o=-J ++= + . .:..++==+& . +.
+..=+

-++
-lO
-NlO
. -HNlO
'

+
i

-
frm. Si
| f|atfig
+o +z +
(
+c
~MH
Figure 4. The MIMO channel-capacity cumulative probability
distribution function in the laboratory scenario in line-of-sight,
non-Iine-of-sight, and hard non-Iine-of-sight areas, obtained
with a 20 dB SNR.
58
o
o
oz

]o c
.
.
V o
,
o z
o+

._....

........._..
| |
i I
--

_
++.=.++==~++
o++a.+.++..+.a~.
+ a. .=

`
..,.. ++++-+-*_
|
.,...

......

.......
i
.

.
~~
-lO
-NlO
-HNlO
~fr S

-
g
+o+ z ++c
~MH
Figure 5. The MIMO channel-capacity cumulative probability
distribution function in the offce-foor scenario in line-of-sight,
non-Iine-of-sight, and hard non-Iine-of-sight areas, obtained
with a 20 dB SNR.
Table 1. 10% outage and mean capacity with a 20 dB SNR.
1 0%
Mean
Diference
Outage
Capacity
with Upper
Caci! Limit
;. LOS 7.3 8. 2 3 . 1

0
NLOS 8.0 9.5 1 . 8
0

^
HNLOS 8.3 9. 9 1 .4
LOS 5. 4 7. 4 3.9
0
Z
NLOS 7.9 9.0 2.3
0
.
o
HNLOS 9. 4 1 0. 1 1 .2 =
..
LOS 7. 1 3 8. 2 3. 1
0
0
l
NLOS 7.8 9. 2 2. 1
<

HNLOS 8.7 9. 6 1 . 7

Freq. Selective
1 1 .2 1 1 . 3 0
Fading
Flat Fading 8.7 1 1 . 3 0
dons, which provided less spatial correlati on that improved mean
capacity, and higher fequency correlati on (or delay spread) that
improved outage capacity .
The results obtained showed how the presence of line-of
sight leads to a signifcant l oss in capacity with respect to the ideal
upper limits. The direct ray acts as a predominant component over
the rest of the scattering, increasing the statistical correlati on
between the sub-channel s. In the three scenarios, the average
capacity presented a decrease of about 3 bpslHz below the average
capacity of the ideal upper limit (Table 1 ).
On passing on to non-line-of-sight situations, the set of scat
tering components tended to present more-uni form amplitudes, and
also a more-uni form distribution of angles-of-arrival at the receiver
of between 0 and 21 . Thanks to thi s richness of multi path compo
nents, the sub-channels presented a greater independence, leading
to a signifcant improvement in the capacity obtained in the non-
/LLLAntennas and Prpagaton Magazine, Vol. 52, No.6, December 2010
line-of-sight environments in comparison with the line-of-sight
areas. In these areas, the capacity was very close to the upper ref
erence limits, with an average capacity of between ! and 2 bpslHz
below the theoretical limit.
An inherent characteristic of a broadband channel i s the
effect of the frequency diversity, refected in an increased slope in
the cumulative probability distribution fnction capacity curve,
giving rise to a gain in the outage capacity. Comparing the capacity
values for the broadband measured channels with the fat fe
quency reference channel, it was observed that the 10% outage
capacity was close to or even surpassed the corresponding outage
capacity of MIMO channels with no spatial correlation, but with a
high frequency correlation (fat frequency). Since the fequency
diversity appeared because of a greater time dispersion associated,
in tum, with a greater scatering richness, it could be concluded
that the most-obstructed situations offered the twofold advantage
of presenting less correlation in the spatial and fequency domains.
J.ZCapacity for Measured SNR
Secondly, the capacity of the evaluated MIMO channels i s
presented, calculated using the SNR values measured in each area.
In combination, these curves showed the infuence on capacity of
the spatial correlation and the real SNR. Figures 6-8 present the
cumulative probability distribution fnction of the capacity in the
scenarios studied. For each scenario, these graphs compare the
capacity of the MIMO channel with the capacity of the SISO
channels i n each of the local areas. Thi s comparison showed that
the use of multiple antennas in transmission and reception did
indeed lead to a signi fcant i mprovement with respect to the SISO
channel .
In the previous section, Figures 3 -5 showed how the presence
of line-of-sight increased the spatial correlation, providing less
capacity in these areas than in non-Iine-of-sight situations. How
ever, since the propagation in line-of-sight areas undergoes less
attenuation, the SNR at reception in these areas was greater than
that obtaintd in the non-line-of-sight areas: in the measured sce
nario, the SNR in line-of-sight areas was 15 dB higher (Table 2).
From the point of view of capacity, i t could be observed in Fig-
c
)
c
n
c
,
c
:
c

a c)
c z
c1
c
,

Figure 6. The MIMO channel-capacity cumulative probabilit

distribution function in the assembly-room scenario in line-of
sight, non-Iine-of-sight, and hard non-Iine-of-sight areas,
obtained with the measured SNR.
IEEE Antennas and Prpagation Magazine, Vol. 52, No.6, December 2010
c .
ca
_c z
cc
n
c
,
a
=c

a c )
cz
c !
!
i

~IH
Figure 7. The MIMO channel-capacity cumulative probability
distribution function in the laboratory scenario in line-of-sight,
non-Iine-of-sight, and hard non-Iine-of-sight areas, obtained
with the measured SNR.
c
c
cz
a

: c
c)
cz
c1

,
==

~MH
Figure 8. The MIMO channel-capacity cumulative probability
distribution function in the ofce-foor scenario in line-of-sight,
non-Iine-of-sight, and hard non-Iine-of-sight areas, obtained
with the measured SNR.
ures 6-8 that although the corelation between sub-channels was
lower for non-Iine-of-sight situations, the reduction in the SNR in
these areas led to signifcantly lower capacity values. In the line
of-sight areas, the MIMO channel thus provided a capacity of
between 1 8 and 23 bits/s/Hz, while on passing over to the non-Iine
of-sight and hard non-Iine-of-sight areas, the capacity in these sce
narios was reduced by between 5 and 9 bits/slHz.
Comparing the capacity of the MIMO channels with that
obtained for SISO channels, it could be observed how the capacity
gain provided by the MIMO systems in the non-Iine-of-sight areas
al lowed a simi lar capacity as for that obtained by the SISO system
i n line-of-sight situations to be reached. In terms of the signal -to
noise ratio, the MIMO channel in non-line-of-sight situations pro
vided a gain that compensated for the fall in SNR caused by the
greater propagation losses in the non-Iine-of-sight areas.
Table 2 shows how in the evaluated scenarios the
implementation of MIMO in non-line-of-sight areas provided a
59
Table 2. The mean capacity in SISO and MIMO channels obtained with the measured SNR.
SISO MIMO
Channel Channel
Capacity Capacity
(bpslHz) (bpslHz)
. LOS 15. 3 23. 3
.
0
NLOS 10.3 16. 2
< 0

^ HNLOS 8. 1 12. 6

LOS 14. 8 20. 5

NLOS 1 0. 1 15. 1
I

HNLOS 7. 8 12. 1
~
LOS 12.5 18. 2
< ..

NLOS 6. 6 9
o
HNLOS 7. 1 10.3
gain in the channel capacity that was equivalent to improving the
SNR in a traditional SISO channel by 15 dB.
These results enable the implementation of communication
systems in indoor-outdoor environments without the need for
deploying the client-end antenna outdoors. Even if the end-user
terminal is located inside the bui lding in non-line-of-sight situa
tions, its capacity will not be degraded, thanks to the gain provided
by the MIMO system. This gain wi ll compensate for the propaga
tion losses due to penetration in the bui lding, providing a capacity
equivalent to that obtained by the SISO system in line-of-sight
areas.
4. MIMO Channel Modeling
The development of communication systems that use multi
ple antennas in transmission and reception require MIMO channel
models to design the modulation and encoding stages. Several
wideband models have been proposed based on the tapped-delay
line approach (TDL) or multi -taps, assuming that there is a fnite
number of relevant scatterer clusters, and that each of the paths
emerging fom the same scatterer cluster undergo the same delay.
Several authors have proposed di ferent ways to model the taps.
Oestges [10] and Patzold [11 ] developed a model where the scat
terers were distributed in ellipses with di ferent sizes. Molisch [12]
and B6Iskei's [13] models were based on taps defned fom the
geometrical characteristics of di ferent scatterers or clusters of
scatterers. Kermoal [14] and Kai Yu [1 5] presented stochastic
wideband models, where each tap was defned by a matrix of cir
cularly symmetric Gaussian elements and an associated spatial
correlation matrix.
This section presents a stochastic modeling method for
MIMO 2x2 broadband models based on a SISO channel model
with delay l ines (TDL) and a correlation matrix. The SISO channel
model reproduces in the time and fequency domains the four sub
channels that make up the MIMO channel . As the receiver anten
nas are in the same local area, the four sub-channels share the same
fading statistics, so that it is possible to describe them jointly by
means of one tapped-delay-line model. In tum, the correlation
matrix describes the stati stical relationship between the sub-chan
nels: more specifcally, between the frst taps of these sub-chan
nels.
60
Capacity SNR
Improvement with
SNR
Reduction
Regard to SISO
(dB)
with Regard
Channel with LOS to LOS Area
(bpslHz) (dB)
8 46. 2
.
0. 9 3 1. 1 15. 1
-
2. 7 24. 4 2 1. 8
5 . 7 44. 6

0. 3 30.5 14. 1
-
2. 7 23. 3 2 1. 3
5. 7 37. 5

-3 . 5 1 9. 7 1 7. 8
-2. 2 2 1. 3 16. 2
In [8], the authors described in detail the SISO channel-mod
eling method and presented the corresponding SISO channel mod
els obtained in the outdoor-indoor scenarios described in Section 2.
The models presented in [8] were delay-cluster-based channel
models, which represented each of the equivalent propagation
paths by means of a delay line with a delay, -,,and a complex
gain, ,,with k=1, ... , L, and with L being the total number of
taps. The channel can be defned as
L
(-)=_e(-

-,). (1)
k=1
The complex gain of each tap is a random variable the amplitude
of which follows a Rician distribution. Each ,can be modeled
using the mean value of its modulus, ,=t,,and the x,
parameter of the Rician distribution that follows , .
The tapped-delay-line modeling of a SISO channel is charac
terized by the number of taps (L), their mean values ,,,,the
Rician x, parameter, and the delay ,-,, of each of these taps.
These parameters were obtained empirically, using the measured
broadband channel realizations.
To confgure the taps, a number of L samples of each chan
nel response were grouped together to form a realization of a tap.
The statistics of each tap in a local area were obtained considering
all the realizations of the tap. Tap statistics in each local area were
therefore obtained considering 64 tap values. The criterion for
choosing the number of taps (L) and, consequently, the number L,
was to maintain a tradeof between the simpli city of the tapped
delay-line model and the corresponding bandwidth of the channel
response in the fequency domain. Given that the separation
between the taps of these models was greater than the inverse of
the bandwidth of the measured channel, the fequency response
was bandlimited and, thus, the model had a smaller bandwidth than
the measured channels. Following thi s criterion, each tap was
obtained by grouping together fve samples in each tap, L=J
The separation between taps was approximately -=5/250 MHz,
so the bandwidth of the model was l imited to -
.
=50 MHz.
The MIMO channel modeling was completed with the
incorporation of additional information into the tapped-delay-line
lLLLAntennas and Pagatn Magazine, Vol. 52, No. 6, December 2010
SISO chimneI mooeL Thi s info
r
atio
n
, given
b
y the corr
e
latio
n
matrices, refected the statistical dependence among the sub-chan
nels. Each correlation-matrix coefcient indicated the statistical
dependence between two taps of two di ferent sub-channel s. It was
assumed that the channels were wide-sense stationary (WSS), and
that the components fom the di ferent delay clusters were uncor
related (uncorrelated scattering). The correlation thus depended
only on the relative separation between antennas, and there was
only correlation between taps with the same delay, while taps with
di fferent delays were uncorrelated:
i fr .
(2)
i fr; s'
where m,pE[I,2,3, . . . , MR] and n,qE[I,2,3, ... , MTx] ' In Equa
tion (2), hrepresents the response of the kth tap of the sub-chan
nel between the jth transmitter and the ith receiver, with
j 1,
..
. , MTx and i 1, . . . , MR, and with MTx and MR being the
number of transmitters and receivers, respectively. The parameters
,
,and

,represent the average value and the variance of the

kth tap, respectively. As all of the h
i
sub-channels are generated
with the same tapped-delay-line SISO model,
,
,and

,are
independent of i andj.
A MIMO channel with sub-channels that are modeled with 1
taps also requires P correlation matrices. The kh correlation
matrix, .,describes the statistical relationship between the kth
taps of the various sub-channels. Each of the coefcients i s
obtained following Equation (2). For a 2x2 MIMO channel, the
correlation matrix is presented in Equation (3). Taking advantage
of the symmetry properties of correlation [1 6], the correlation
matrix of a 2x2 MIMO channel can be described by means of six
di fferent correlation coefcients:
[ 1
i,i
.
/
z,i
i
/
i i,i
z,i
/
ii,i
zz,i
[
/
z,i
a
i,i

/
:z,i
i:,i
/
zz,i
:i,i
/
z,i
i:,i
J
iz,i
z,i
zz,i
/z,i
/
iz,i
:i,i
/
iz,i
z,i
/
zz,i
z,i
/
zz,i j
i:,i
/
zz,i
z,i
/
zz,i
z,i
1
(3)
When the taps are Rayleigh-fading, the Mtap of the MIMO chan
nel, H, can be expressed as in Equation (4), where Hk is a
matrix of complex Gaussian-random-variable elements with zero
mean and unit variance, and with ,being the sigma parameter of
the Rayleigh distribution. The vec H) fnction groups the col -
umns of the H matrix i n a single-column-wise vector. The .
matrix is obtained fom the Cholesky decomposition of .,
.,..,
(4)
lLLLAntenas and Prpagation Magazine, Vol. 52, No.6, December 2010
For Rician-fading taps, the channel matrix with the appropriate
correlation between sub-channels is obtained by adding a determi
nistic component and applying the correlation only to the random
par of the tap:
(5)
In Equation (5), .,represents the deterministic component ampli
tude of the M tap, where ,.,,.,,and .

is the identity
matrix of order two.
The spatial correlation between the di ferent taps leads to a
level of correlation between the various tones in the fequency
domain. The correlation coefcient, r, between the two sub-chan-
nels H
mn
, and H __ ,,) is obtained for each tone fol l owing
Equation (6). The expectation was computed over the number of
samples (64) obtained in the evaluated local area.
with m,pE[I,2,3, . .. , MR] and n,qE[I,2,3, . . . , MTx] '
Developing Equation (6) as i n the Appendix, a compact
expression is obtained as a fnction of the parameters that defne
the channel model :
(7)
If the amplitudes of all the taps undergo Rayleigh fading, the cor
relation coefcient can be expressed as a fnction of the mean
amplitude of each tap:
(8)
From Equation (7), it can be observed that the spatial correlation in
the proposed model i s independent of fequency. In thi s model , all
of the tones thus present the same spatial correlation. It is also
observed fom Equation (8) that each correlation coefcient, p ,
M
is proportional to the average value of the correlation coefcients
applied to the taps, weighted by the square mean amplitude of the
taps. This means that the channel model ing only requires the cor
relation to be applied to the taps with the greatest amplitude, given
that these make the most-signifcant contribution to the correlation.
Since in general the frst tap is the one with the greatest power, the
present modeling method only needs to apply the correlation to the
frst tap, and thus only one single correlation matrix is required.
This claim was supported experimentally by comparing the capac
ity results obtained by means of simulations using one or several
correlation matrixes in the MIMO channel modeling in line-of-
61
sight, non-line-of-sight, and hard non-line-of-Sight si tuations. The
mean relative error committed in the ergodic and 1 0% outage
capaci ty when modeling with a single correlation matrix with
respect to the modeling with the set of correlation matrices was 3 %
(with SNR 20 dB).
The proposed MIMO channel modeling method i s valid for
si tuations with di ferent amplitude distributions (Rayleigh or Rice).
Moreover, it is less complex than the other models presented in the
li terature. This is because it does not require a great number of cor
relation matrices, as in the case of stochastic models [14, 15 ], or
knowledge of the angular characteristics of the scatterer clusters, as
is the case of geometrical models [12, 13, 17]. It also enables the
exi sting tapped-delay-line SISO models to be used by means of the
appropriate correlation model, in order to obtain MIMO channel
models for these scenarios.
. Outdoor-Indoor MIMO Channel Model s
Once the modeling method has been outlined, MIMO chan
nels models are presented and described by means of the COrre-
sponding tapped-delay-line SISO model (Table 3) and i ts correla
tion matrix (Table 4). In order to demonstrate the reliability of the
models, the capaci ty of the simulated MIMO channels was com
pared with that of the measured channel .
.1 MIMO Channel Model s
MIMO channel models combined the tapped-delay-line SISO
models with a single correlation matrix. All of the MIMO system
sub-channels were generated according to the same tapped-delay
l ine SISO channel model, and then the correlation matrix was
applied only to te frst tap of the MIMO channel obtained.
To obtain the SISO channel model for a specifc area, the
measurements fom scenarios that showed a similar behavior in the
time and fequency domains were grouped together. In the line-of
sight areas, the analysis of the measurements showed that the three
scenarios had a similar behavior in both domains. However, in the
non-line-of-sight and hard-non-line-of-sight situations, the scenar
ios showed more-dissimilar behaviors. They were therefore
grouped into laboratory and ofce scenarios on the one hand, char-
Table . The tapped-delay-line channel models for diferent areas in both types of scenarios.
r
:
r

62
TAP 1 2 4 5 6 7
<' VCE 2 0 0 0
- - -
\:
MEAN(dBl Q 0 0
-
20. 5
-
22. 5 -26. 3
- - -
Q-
` DELAY (ns) 3 27 49 64
- - -
\:
VICE 4 0 0 0 0 0 0
Q
MEAN(dBl 0. 0
-
12. 0 -17. 6
-
20. 5
-
20. 0
-
24. 3
-
30. 3 Q

DELAY (ns) 5 27 47 68 87 106 126

" VICF
Q
2 0 1 0 0
- -
'
MEAN (dB) 0. 0
-
11. 2
-
15. 8 -21.6
-
27. 8
- -
0
(
DELAY (ns) 5 27 47 66 86
- -
\:
VICc 4 0 1 0 0 0 0
Q
Mean (dB) 0. 0
-
9. 8
-
13. 2
-
15. 6
-
18. 6
-
22. 9
-
29. 7 Q

Delay (ns) 5 28 48 67 86 107 126

" VICc
Q
3 0 0 0 0
- -
Mean (dB)
t
0. 0
-
8. 9
-
11. 9
-
16. 8 -21. 8 0
- -
(
Delay (ns) 5 28 47 67 86
- -
Table 4. The correlation matrices (module) for the MIMO channel models
for diferent areas and scenarios.
Open Closed Scenario
1 0. 913 0. 956 0. 874
r
0. 913 I 0. 884 0. 953

= 0. 956 0. 884 I 0. 917

0. 874 0. 953 0. 917 1
Open Scenario Closed Scenario
1 0. 726 0. 841 0. 644 0. 707 0. 926 0. 716
r
0. 726 I 0. 695 0. 833 0. 707 I 0. 686 0. 923

,
~
0. 841 0. 695 I 0. 742 0. 926 0. 686 I 0. 7 19

0. 644 0. 833 0. 743 I 0. 7 16 0. 923 0. 7 19

r
I 0. 737 0. 842 0. 706 I 0. 701 0. 839 0. 673
C
0. 737 I 0. 694 0. 862 0. 701 I 0. 702 0. 841
~
0. 842 0. 694 I 0. 667 0. 839 0. 702 I 0. 712

0. 706 0. 862 0. 667 I 0. 673 0. 841 0. 7 12

8
-
-
-
0
-
3 1. 7
408
-
-
-
0
-
27. 3
764
-
-
-
/cLLAntennas and Prpagation Magazine, Vol. 52, No.6, December 2010
acterized as scenarios with a large number of obstacles near the
receiver (closed scenario), and on the other hand as the assembly
room scenario, which was of a more-open nature (open scenario).
Table 3 shows the SISO channel models for outdoor-indoor sce
narios in line-of-sight, non-line-of-sight, and hard-non-line-of
sight situations. These models presented minor di ferences from
those presented by the authors in [8], due to slight changes intro
duced in the processing of the measurements.
From the MIMO channel point of view, the channel model
was obtained by grouping together the set of realizations of sce
narios that showed a similar behavi or in the time, frequency, and
spatial domains. The spatial -domain behavior of the channels
maintained the same divi si on of open and cl osed scenarios for non
line-of-sight and hard-non-line-of-sight as that shown by the SISO
channels. Table 4 presents the correlati on matrices obtained for
every area of each type of scenario. The correlati on coefcients
refected the statistical relationships among the various sub-chan
nels at the frst tap. These values were high, because the frst tap
showed a Rice-type statistic (see Table 3, Tap 1), due to the pres
ence of a dominant component that increased the degree of corre
lation.
b.Z Capacity in Model ed MIMO Channel s
Using MIMO channel models, i t was possible to reproduce
the channel behavi or in scenarios with conditi ons similar to those
studied. In order to assess how the model ft the measured channel,
the capacity of the modeled channels was compared with that
obtained from the measured channels. Given the random nature of
the channel, the cumulative probability distribution fnctions of
the two capacities were compared. Figures 9- 1 1 show the cumula
tive probability distributi on fnctions of the capacity in line-of
sight, non-line-of-sight, and hard-non-line-of-sight situations for
vari ous scenari os. The capacity curves obtained were compared
with the two theoretical upper limits described and used in Sec
ti on 3 . 1.
The measured and simulated channels underwent the efects
of the spatial and fequency correlati on j ointly and inseparably,
i
I
o --

c
++..

.~.+.~...

=.
c1 ---
-
I
cs - --|-----

.
'

'
g ! I
U I
-
c

c
`

c:
c: ~_
-_.--
c:
0ap
Figure V. The cumulative probability distribution function of
the MIMO channel capacity measured and modeled in line-of
sight areas ( SNR 20 dB).
/LbLAntennas and Prpagation Magazine, Vol. 52, No.6, December 2010
MEBSU|EC
0.9
MOCEEC
0.8
|B |BCID

|IE SEECIVE
( 0.7

J 0.6
c
"10.5
O
_ 0
.
4
0.3
0.2
6 8 10 12 14 16
LBBCIy 0SmZ
Figure lOa. The cumulative probability distribution function of
the MIMO channel capacity measured and modeled in non
line-of-sight areas ( SNR 20 dB), for the open scenario.
0.9
_0.8
c
0.7

. 0.6
c
"10.5
O
_ 0.4
0.3
0.2
MEBSUIEC
MODEEC
0.1
|B |BCIR
|E SEECIVE
4---- - 6 8 10 12 14 16
LBBCIy 0OSMZ
Figure lOb. The cumulative probability distribution function
of the MIMO channel capacity measured and modeled in non
line-of-sight areas ( SNR 20 dB), for the closed scenario.
al ways showing l ower capacities than those of the reference chan
nels. A compari son of the capacity curves showed that the model
ing using the proposed method led to a si gnifcant improvement
with respect to the theoretical models, which did not consider the
spatial correlation.
It could be observed fom the results obtained that the capaci
ties of the modeled channels for each type of area and scenario ft
the capacities obtained with the channel measurements with little
error, not only for two paricular values (the mean and 10% outage
capacities)but also for the probabi lity distributi ons of the capacity
values. In some appl ications, it i s important to know not onl y the
capacity of the channel, but also the channel eigenvalues, which
show details about the structure of the channel. As one example,
Figure 12 shows the distribution of the measured and modeled
channel -matrix eigenvalues in two di fferent scenarios. It could be
seen that the modeled ei genvalues, including the weaker eigenval
ues, ft the structure of the measured channel.
63
s - : .
LBBC|y 0SMZ
s
Figure l l a. The cumulative probability distribution function of
the MIMO channel capacity measured and modeled in hard
non-Iine-of-sight areas ( SNR 20dB) for the open scenario.
!
= +:o.o+-+ + ._..
c

..! .=..=..... . ...... .... ... .= . .o o.. .. o---

! |

|
!
c

= ==oo=o o==oo=,oooooo=oooooooooo ,ooooooooo=oooo l =. =

| i
|
i
ooooo=o==eo=o
o
)
.. oooo oooooooo=oo =ooooooo=oo= o=+ o= i ==o

==o=o=oooo +o = o =
'
n
I i !
c

)
eooo=ooooooooooooo ======= = .

o === o=o ooo

=o==
r
e
c

z
.o =o- ++.+-.+. _ +o=o o=,o+=!a=u ... .. t-ooo ooooooooo
'
eoooo=ooo -- J.
Mea
s
ur
e
! , | i
2

o1 __ oooooooooogoooooooo -A1 Moel

-A
2
Moel
o c z + c c z +
A (d8j
Figure 12a. The cumulative probability distribution function of
the 2x2 MIMO channel eigenvalues measured and modeled for
the line-of-sight scenario.
6. Conclusi ons
Knowledge of the radio channel i s a crucial factor in the suc
cessfl implementation and confguration of the various stages of
communication systems
-
modulation, encoding, or equalization -
and for assessing the fnal performance characteristics of the sys
tem, i . e. , capacity or bit-error rate. This paper presented the
experimental characterization of a wideband 2x2 MIMO channel
in typical indoor-outdoor scenarios in the 3. 5 GHz band. A general
modeling method suitable for SISO channels, extending to MIMO
channels, and intended for point-to-point communications in mixed
scenarios was also presented.
In previous work, the authors analyzed the advantages attain
able by establishing a 2x2 MIMO channel over a SISO channel for
the case of narrowband wireless systems in outdoor-to-indoor envi
ronments without line-of-sight (NLOS). The new results for the
wideband case showed that the capacity degradation due to pene-
64

MBSd|EC
: .
MOCl C
: -
|B |BC|R
:

|E

J : s
r
VI
:

O
_ :
: :
: .
s - : .
LBBC|y 0SmZ
s
Figure 1 1 b. The cumulative probability distribution function
of the MIMO channel capacity measured and modeled in hard
non-Iine-of-sight areas ( SNR 20 dB) for the closed scenario.
c
c
c
)
7cc

cs
0 4
a
c)
c z
c+
c
c
==. +...+===.+_ =..
'
t .. . ..
.. o..
I
-== . .. ==.

ooo oooo ooo oa=e==o=a a=a=+ o =oaa ==o==o=a a=o=aoa= oo=a= o.=.=. . . .. . . .
i |
i .
.. .... . ... - -.j--- =. ..... .. _ . ..-.-..
| ,I ! !
|
| . ' ' '
o .+.+o. .o... ..+=. =o + + ".++ " =1==+.++=
r
..
=====+o+++=. .+. . . .+o,. +. . .=+++= .+= +.++=.o+o .+o=. .+..++++.+ . + .. +.=+.=====.. .==.o .==.
.
...
: ; !

! !
i
.t. =.== =a - ==w . . .=.=-=... = .. =.. . .o.=+= ,=
t I !
=. . ==oo=oe==oo oo=oooooooo ooo ooo oo oo o oooooooeoo

oo ..=.. .+. . ..
. o=oo o ! 1-- ... te ... . ...
i i I .-A2
Mea
s
ure

=.
.
=.ooa_o oa oao o + a = oogo=oo a==o
'
e

e+e oe oe o.. o. . _
I I I ! 2
+ c
A (d8j
c i z ++
Figure 12b. The cumulative probability distribution function
of the 2x2 MIMO channel eigenvalues measured and modeled
for the non-Iine-of-sight scenario.
tration losses can also be practically compensated for by the gain
of a 2x2 MIMO channel in the wideband case. Furthermore, the
fequency diversity attainable in the case of wideband systems
constitutes a new contribution, reinforcing the conclusion obtained
for the narrowband case.
The modeling method proposed leads to a correct ftting of
the channel both in the time and the fequency domain, requires a
low number of parameters, and is computationally efcient. These
qualities make the model suitable for applications.
The models proposed were described using a small number
of parameters (the number of taps, their relative amplitUdes, and
their statistics), obtained experimentally by measurement cam
paigns on 250 MHz channels performed in the scenarios under
study. Although the number of measurements available at thi s
moment does not permit the establishment of a complete model for
the whole range of outdoor-indoor scenarios, it can be concluded
that the modeling method proposed allows realizations of SISO
/LLLAntennas and Prpagaton Magazine, Vol. 52, No.6, December 2010
and MIMO channels to be -obtained simply and -efciently by
means of a small number of parameters. Moreover, taking into
account the lack of data about wideband MIMO channels in thi s
fequency band and for these scenarios, the channel models pre
sented for mixed scenarios are of special interest for the develop
ments currently taking place in broadband fxed wireless access
(BFWA).
7. Acknowl edgment
Thi s work was supported by the Spanish Ministry of Science
and Innovation (project number TEC2008-02730ITEC).
. Appendi x:
Devel opment of Equati on (7)
The correlation coefcient, r, between two sub-channels,
H mn
,) and H p
q
,,) , for each tone i s obtained according to
Equation (9), with p, m E [I, 2, 3, . . . , MR] and
q, n E [1, 2, 3, . . , MTx ] :
Equation (9) is developed to express the correlation coefcient, r,
as a fnction of the parameters that defne the model : the variance
in the amplitude of each tap,
(
j , the delay, 'j , and the spatial
correlation coefcients,
P
r
(
i) , applied to each tap.
Given that the channel in the time domain and the channel in
the fequency domain are related through the Fourier transform,
I
H(J)
=
'h
k
e-j
2
1jrk ,
k=|
E
H
mnH;q - IHmn lp
2 2
(Hmn ( H;
( 1 0)
( 1 1 )
the numerator o f the correlation coefcient i s obtained as a fnc
tion of the parameters of the tapped-delay-line model:

I I
E

HmnH;q , - IHmn lp =
'E-,-, ] . .:.
k=| k=|
( 1 2)
Given the defnition of the correlation coefcient,
/LLLAntennas and Ppagation Magazine, Vol. 52, No.6, December 2010
hf _ pq
( ) P
h
' ;" - Pmn
k
( 1 3)
the sum of the product of the spatial correlation coefcients i s
obtained through the variance in the amplitude of the taps:
I
E

HmnH;q , - IHmn l
p
=
'
p
(
k),k
'
( 1 4)
k=|
Each of the variances in the denominator of Equation (9) i s devel
oped by expressing them as a fnction of the variance and the
delay of each tap:
(kmn =
E
I
Hm
i, - IHm
Thus, the denominator is expressed according to
I I I
2 2
'
2
'
2
'
2
(H (
H'
= (h r (h s = (h k
.
mn
p
" 7
r=| s=| k=|
( 1 5)
( 1 6)
Finally, combining Equations ( 1 4) and ( 1 6), the correlation coef
cient can be expressed as the average of the correlation coefcients
applied to the taps weighted by the variance of each tap:
( 1 7)
9. References
1 . S. Ahmadi, "An Overview of Next-Generation Mobile WiMAX
Technology," IEEE Communications Magaine, June 2009, pp.
84-99.
2. IEEE, "IEEE 802. 1 6e-2005 : IEEE Standard for Local and Met
ropolitan Area Networks Part 1 6: Air Interface for Fixed and
Mobile Broadband Wireless Access Systems, Amendment 2:
Physical and Medium Access Control Layers for Combined Fixed
and Mobile Operation in Licensed Bands and Corrigendum 1
(Technical Report)," 2005.
3. WiMAX Forum, "WiMAX Deployments with Seif-Installable
Indoor Terminals, " (white paper), June 2005, available at
http://ww.wimaxforum. org/technology/downloads/Deployments_
with_Indoor_Terminals_Revl _2_2_.pdf.
4. Richard Jaramillo, Oscar Ferandez, Rafael P. Torres, "Empiri
cal Analysis of 2x2 MIMO Channel in Outdoor-Indoor Scenarios
for FBWA Applications," IEEE Antennas and Propagation Maga
zine, 48, 6, December 2006, pp. 57-69.
65
5. S. Wyne, A. F. Molisch, P. Almers, G. Eriksson, 1. Karedal, and
F. Tufesson, "Outdoor-to-Indoor Ofce MIMO Measurements
and Analysis at 5. 2 GHz," LLL Transactions on Vehicular Tech
nolog, 57, 3, April 2003, pp. 1 3 74- 1 3 86.
6. 1. Medbo, F. Harrysson, H. Asplund, and 1. E. Berg, "Measure
ments and Analysis of a MIMO Macrocell Outdoor-Indoor Sce
nario at 1 947 MHz," Proceedings of 59th IEEE Vehicular Tech
nology Conference, Milan, 2004.
!. H. T. Nguyen, J. B. Andersen, and G. Pedersen, "Characteriza
tion of the Indoor/Outdoor to Indoor MIMO Radio Channel at
2. 1 40 GHz," COST 273 Meeting, Gothenburg, Sweden, 2004.
8. Oscar Ferandez, Richard Jaramillo E., Marta Domingo, Luis
Valle, and Rafael P. Torres, "Characterization and Modeling of
BFWA Channels in Outdoor-Indoor Environments," 1LLL Anten
nas and Wireless Propagation Letters, 6, 2007, pp. 236-239.
. Juan Jose Garcia Ferandez, Matilde Sanchez-Ferandez, Ana
Garcia Armada, Rafael P. Torres, Marta Domingo, and Oscar
Ferandez, "Analysis of Beamforming and Spatial Multiplexing
Strategies in WMAN Outdoor-Indoor Scenarios," Proceedings of
IEEE 69th Vehicular Technology Conference (VTC) Spring, April
2009.
H. C. Oestges, V. Erceg and A. 1. Paulraj , "A Physical Scattering
Model for MIMO Macrocellular Broadband Wireless Channels, "
1LLL Joural on Selected Areas in Communications, 25, 5, June
2003, pp. 72 1 -729.
I I . M. Pltzold and B. O. Hogstad, "A Wideband MIMO Channel
Model Derived fom the Geometric Elliptical Scattering Model,"
Proceedings of 3rd Interational Symposium on Wireless Commu
nication System, ISWCS' 06, Valencia, Spain, September 2006, pp.
1 38- 1 43.
1 2. A. F. Molisch, M. Steinbauer, M. Toeltsch, . Bonek, and R. S.
Thoma, "Capacity of MIMO Systems Based on Measured Wireless
Channels," LLL Joural on Selected Areas in Communications,
O, 3, April 2002, pp. 561 -569.
1 3 . H. Bolcskei, D. Gesbert, and A. 1. Paulraj , "On the Capacity of
OFDM-Based Spatial Multiplexing Systems," LLL Transactions
on Communications, 50, February 2002, pp. 225-234.
1 4. K. 1. Pedersen, J. B. Andersen, J. P. Kermoal and P. Mogensen,
"A Stochastic Multiple Input Multiple-Output Radio Channel
Model for Evaluation of Space-Time Coding Algorithms," Pro
ceedings of IEEE Vehicular Technology Conference, Fall 2000, 2,
pp. 893-897.
1 5. K. Yu, M. Bengtsson, B. Otersten, D. McNamara, P. Karlsson
and M. Beach, "A Wideband Model for NLOS Indoor MIMO
Channels," IEEE Vehicular Technology Conference VTC 2002,
pp. 370-374.
1 6. O. Ferandez, M. Domingo, and R. P. Torres, "Empirical
Analysis of the Correlation of MIMO Channels in Indoor Scenar
ios at 2 GHz," LL Proceedings Communications, 152, I , February
2005, pp. 82 -88.
1 7. H. Bolcskei, M. Borgmann, and A. 1. Paulraj , "Performance of
Space-Frequency Coded MIMO-OFDM," 1LLL Journal on
Selected Areas in Communications, 21, 3, April 2003, pp. 427-439.
lLLLAntenas and Prpagation Magazine, Vol. 52, No. 6, December 2010