You are on page 1of 4

MEASUREMENT AND MODELING OF THE MIMO WIRELESS CHANNEL

Jon W. Wallace 1 , Michael A. Jensen 2 , A. Lee Swindlehurst 3

Department of Electrical and Computer Engineering, Brigham Young University, 459 Clyde Building, Brigham Young University, Provo, UT 84602, USA. E-mail: wallacej@ee.byu.edu

2

As (1) above, but E-mail: jensen@ee.byu.edu As (1) above, but E-mail: swindle@ee.byu.edu

ABSTRACT This paper presents measured data at 2.45 GHz from a narrowband multiple-input multiple-output (MIMO) channel probe in an indoor ofce/laboratory environment. The measured transfer matrices are used in a study of the statistical behavior of the channel capacity. The effect of antenna element polarization and array size are highlighted. The data is also used to validate a simple statistically based physical propagation model. INTRODUCTION The increasing demand for capacity in wireless systems has motivated considerable research aimed at achieving higher throughput on a given bandwidth. One important recent discovery shows that in a multipath environment, the use of space-time coding with multiple antennas on both ends of the link can increase the capacity of the wireless channel [1]-[2]. Assessing the performance of these algorithms requires detailed understanding of multiple-input multiple-output (MIMO) channels as well as models that capture their complex spatial behavior. In this work, we discuss data from an experimental platform designed to measure the transfer matrix for indoor MIMO channels. The data is used to demonstrate the effect of polarization and array size on the achievable capacity for MIMO architectures. A propagation-based statistical model is shown to provide results that match closely with measured observations [3]-[4]. EXPERIMENTAL MEASUREMENTS The experimental platform, depicted in Fig. 1, uses a custom MIMO communications system operating at 2.45 GHz to directly measure the wireless MIMO channel transfer matrix H, where Hmn represents the transfer function between the nth transmitter and mth receiver antennas [5]. The transmitter creates N unique co-channel binary phase shift keyed signals using 1000-bit pseudorandom binary sequences at a chip rate of 12.5 kbps. Each signal is amplied to 0.5 W and fed into one of the N transmit antennas. The receiver downconverts, samples, and stores the signals from each of the M receive antennas for subsequent post-processing. A variety of different antenna arrays are used, including a 4-element dual-polarization patch array ( 2 spacing) and a 10-element linear monopole array ( 4 spacing). The raw data collected using the measurement platform is processed to obtain one estimate of the time-variant channel matrix for each complete codeword (80 ms intervals). Consistent transfer matrix normalization is accomplished by specifying an average single-input single-output (SISO) signal-to-noise ratio (SNR) and scaling all elements of each channel matrix accordingly. For the results shown below, a SISO SNR of 20 dB has been used. Past presentations of data obtained with the measurement platform discussed above have concentrated on the statistical properties of the transfer matrix (element magnitude and phase distributions, spatial and temporal correlations, etc.). In this work, we present representative results documenting the capacity of the measured channels and use these results to motivate our modeling efforts. Capacities are computed from the measured H matrices according to the water lling solution of the channel orthogonalized by the singular value decomposition (SVD) [2], [6]. Capacity Dependence on Polarization To assess the role of polarization in the performance of MIMO architectures, four transmit/receive channels were used to excite both feeds on two 2 separated dual-polarization patches. Appropriate submatrices of H were then used to compare the complimentary cumulative distribution functions (CCDFs) of capacity for three different 2 2 channels: (1) 2 elements with same polarization (V or H) but separated by 2 (SP), (2) 2 elements which have orthogonal polarization and are colocated (DP), and (3) 2 elements which have both orthogonal polarization and are separated by 2 (DPS). Fig. 2 shows the results of this study. The remaining curves represent capacities computed via 106 Monte Carlo trials

Microwave Generator N N

CHANNEL
M N M

Microwave Generator M

N BPSK Generator Transmit Array Receive Array

M PC Based Data Storage

Optional Synchronization (10 MHz)

Optional Synchronization (10 MHz)

Fig. 1: High level system diagram of the narrowband wireless MIMO measurement system with independent identically distributed (iid) complex Gaussian H matrix elements with unit variance. IID and DIAG indicate full and diagonal matrices respectively, with the diagonal version representing use of dual polarizations with no cross-coupling. As can be seen, the use of dual-polarizations signicantly increases the available channel capacity. The channel normalization applied in Fig. 2 ensures each 2 2 sub-channel has the same SISO SNR. Because of weak coupling between orthogonally polarized channels, this normalization is optimistic. A more realistic comparison normalizes the co-polarized matrix elements to achieve an average SISO SNR of 20 dB and scales the remaining elements accordingly to preserve the relative receive power levels. Fig. 3 depicts the CCDFs resulting from this normalization. In this case, the relative SNR decrease for dual-polarized matrices causes the simulated single polarization channel (sSP) to outperform the dual polarization case (sDP). A similar behavior is observed for the measured single polarization channel, although there is an offsetting degradation in capacity due to nite signal correlation. Thus, dual polarizations are a good choice for compact arrays of closely spaced elements. Capacity Dependence on Number of Antennas The dependence of capacity on the number of antennas is explored here using 2, 4, and 10 monopole transmit and receive antennas. To make a fair comparison, each array possesses the same total length (2 25). Fig. 4 shows the capacity CCDFs per number of transmit and receive antennas. Also, Monte Carlo simulations were performed to obtain capacity CCDFs for channel matrices having iid complex Gaussian elements with unit variance. These results indicate an excellent agreement between the measured 2 2 and ideal 2 2 (independent Gaussian) channel due to the very wide antenna separation (2 25). The measurements show that as the number of elements increases, the actual capacity per antenna drops due to higher correlation between adjacent elements (which is not modeled under the iid assumption). PHYSICAL CHANNEL MODEL While measured data such as that presented above provides a comprehensive description of the channel behavior, simple models are highly desirable for assessing behavior of spatial multiplexing systems. Here, we propose an extension of the Saleh-Valenzuela model [3] that includes angle-of-arrival (AOA) statistics [4] to model the channel. Angle-of-departure (AOD) statistics are assumed to follow the same distribution as AOA, which is reasonable for the indoor channel with the same basic conguration on transmit and receive. The model is referred to here as the SVA model.
sDP

P(cap > abscissa)

IID DP
KEY SP IID DP DPS DIAG

DPS
Single Pol Elements Ideal Gaussian IID Dual Pol Elements Dual Pol Separated Elements Ideal Gaussian IID Diagonal

sSP sDP

Capacity (bits/use)

Capacity (bits/use)

Fig. 2: CCDFs for 2 2 channels employing different polarization/separation schemes.

Fig. 3: CCDFs for 2 2 channels employing different polarization/separation schemes with realistic normalization.

q 6s

p Ds

i 6s

9 ( V 25 7Q I 2 75 2 0( G 0 F @9130`U" YTS8"R& I P864 31)& 9 ( V 2 S 7 2 754 H F G 0 F @91X0'WU" '5T8"QR& I IP8D" 2 1G %# E" D0 B0@5 1864 31)& C A 92 75 2 0( 2 75 2 0( 864 31)& 2 754 H F 8D" 2 1G %#

f 6s

b ca

#'$ & $ & $ %# ! " 

DIAG

q hf g

SP

P(cap > abscissa)

a eb

r hf g

p hf g

i hf g

 

 

  

sSP

P(cap > abscissa)

Capacity (bits/use/ch)

Fig. 4: Capacity CCDFs per number of antennas for transmit/receive arrays of increasing number of elements. The array length is 2.25 for all cases. The SVA model characterizes the channel by representing each multipath component in terms of its amplitude, arrival time, and AOA/AOD. Based upon experimental observations, these arrivals or rays arrive in clusters in both space and time. The directional channel impulse response arising from L clusters and K rays per cluster is
g l n l n e g l n l n e l m m l j h ig

h R T
f e o

1 L 1K 1 k T LK 0 k 0
k k l

T R k
o l

where T R are transmit and receive angles, k is the complex ray gain, T R are the mean transmit and receive cluster T arrival angles, and k R are the relative angles of transmit and receive for the kth ray in the th cluster. To simplify the model, average ray power in each cluster is constant so that k 0 2 , where denotes the 2 complex normal distribution. The cluster amplitude is Rayleigh distributed with E exp T . The arrival time distribution is a conditional exponential with a normalized unit arrival rate. Details concerning the model implementation can be found in [3]-[4]. The notation SVA is used in this paper to denote the SVA model with constant average ray power and unit cluster arrival rate, where is the standard deviation of ray AOA/AOD. The narrowband channel matrix is computed by integrating the antenna eld patterns against the directional impulse response in transmit and receive angles. Comparison of Model and Data 26 , 2. For simulation, Based upon high resolution AOA measurements [4], we begin with the parameters T R transmit and receive cluster arrival angles are assumed to be uniform on 0 2 . Gain patterns for the antenna obtained from moment method simulations are used to compute the transfer matrix. Fig. 5 compares probability density functions (PDFs) for capacity and pairwise joint PDFs for transfer matrix elements of measured data and Monte Carlo simulations of the SVA model for 4 4 data. In these and later simulations, 105 channels were realized (100 cluster congurations with 1000 channels each). PDFs are computed for each cluster conguration, and these 100 functions are in turn averaged to obtain the results shown in the gures. Apparent in the gure is the good t of both the capacity and pairwise amplitude PDFs. The discrepancy in phase is probably due to the fact that the uniform cluster AOA/AOD assumption is not strictly valid over the limited data set. Fig. 6 compares the PDFs for the measured and simulated 10 10 channel. While the parameters from [4] do not yield the desired t for capacity, the gure shows that adjustment in either the cluster decay rate or angular ray spread improves the agreement. Detailed AOA/AOD measurements at the 2.4 GHz carrier are required to further study the discrepancy. Agreement in the amplitude and phase PDFs is comparable to the 4 4 data and does not change signicantly with the parameter adjustments. Other Channel Models In many cases, it is assumed that underlying distribution on H is multivariate complex normal with covariance matrix obtained as the average covariance of the true distribution or R E hhH , where h is a stacked channel matrix. In other cases, the covariance matrix is assumed to be the elementwise square root of the power covariance matrix of the true h vector. In either case, our simulations have shown that such models tend to overpredict channel capacity due to the necessity of using multiple measurement scenarios to create the covariance matrices.
y x h h | h { o z ~ f } u vs g l h y w l w x n e u s q g w l cf e vtrl w g f e p ol

v %y u

B Y BY` D  6h` D B Y BY  6h B Y BY`"D  6hW y v xv u v v %w u w v %t u t

x u y x u w x u d x u

R k

(1)

Additionally, we have used a ray tracing algorithm to accurately model the channels under investigation. In such a case, we nd the ray statistics to match those in the model, but the associated capacity is always 20-50% too low. It is assumed that the limited number of rays modeled in the ray tracing approach accounts for this discrepancy, although more detailed studies are required. Finally, we have also extended the SVA model to incorporate dual-polarized antennas and account for depolarizing scattering. The results of this extension agree very well with measured capacity statistics, and therefore the model shows promise for assessing the performance of dual-polarized systems. CONCLUSIONS This paper has explored the ability of a simple statistical model to capture key features of the narrowband indoor MIMO wireless channel. Ultimately, a tradeoff exists between model complexity and accuracy. However, we have shown that even simple models (like the SVA model), which are based partially on channel physics, match capacity and pairwise PDFs of measured data quite well. REFERENCES [1] G. J. Foschini and M. J. Gans, On limits of wireless communications in a fading environment when using multiple antennas, Wireless Personal Communications, vol. 6, no. 3, pp. 311335, March 1998. [2] Gregory G. Rayleigh and John. M. Ciof, Spatio-temporal coding for wireless communication, IEEE Transactions on Communications, vol. 46, no. 3, pp. 357366, March 1998. [3] Adel A. M. Saleh and Reinaldo A. Valenzuela, A statistical model for indoor multipath propagation, IEEE Journal on Selected Areas of Communications, vol. SAC-5, pp. 12813, February 1987. [4] Q. Spencer, B. Jeffs, M. Jensen, and A. Swindlehurst, Modeling the statistical time and angle of arrival characteristics of an indoor multipath channel, IEEE J. Selected Areas Commun., vol. 18, no. 3, pp. 347360, Mar. 2000. [5] J. W. Wallace and M. A. Jensen, Characteristics of measured 4x4 and 10x10 MIMO wireless channel data at 2.4GHz, in 2001 IEEE Antennas and Propagation Intl. Symposium Digest (APS 2001), Boston, MA, July 2001, p. to appear. [6] Thomas M. Cover and Joy A. Thomas, Elements of Information Theory, John Wiley & Sons, 1991.

Fig. 5: Comparison of capacity PDFs and joint magnitude and phase PDFs for 4 4 measured data and SVA model simulations.

Fig. 6: Comparison of capacity PDFs and joint magnitude and phase PDFs for 10 10 measured data and SVA model simulations.

312'$90(T(F# !()SQU0V8 !67560 4 0 3 2 $ 0 T F !  Q !   1'U(# ()SR# "  6(E# !(D CB # !65A @ 0 F

ca `  7bW

3 2 $ 0 & $ 0 ! 0 5 1'1)('98 6760 4

e dGf7 WX c a ` W Y

 )  

Y WX

P  P

g c a ` W Y 6a db7 WX

3 2 $ 1'GB # 7() @ ! 5 A

3 2 $ 0 & $ !   1'1)('%# " 

   

e db7 WX c a ` W Y

P  P

"Y  8c "' 8 ' 8 c ca `WY  dbf7 WX


g c a ` W Y 6a db7 WX

8 X 8 8 X
I X H X

X X

X X X

c X 8@ c X X@ X X X X X X X X e@ "h c@) @@X 8 c X X  "1" X X  "1 X X X  "1"  "1" X X "1" X "1"D X e@ PD "h e@ `h @@X @@@  i P

6X XP @@`



X X X X X X X   

You might also like