MIMO Wireless Channels: Capacity and Performance Prediction
MIMO Wireless Channels: Capacity and Performance Prediction
Performance Prediction
Abstract
1. INTRODUCTION
Here we suggest a simple classification of MIMO channel and devise a MIMO channel
model whose generality encompasses some important practical cases. Unlike the
channel model used in [3, 15], our model suggests that the impact of spatial fading
correlation and channel rank are decoupled although not fully independent, which
allows for example to describe MIMO channels with uncorrelated spatial fading at the
transmitter and the receiver
but reduced channel rank (and hence low capacity). This situation typically occurs
when the distance between transmitter and receiver is large. Furthermore,our model
allows description of MIMO channels with scattering at both the transmitter and the
receiver. We use the new model to describe the capacity behavior as a function of the
wavelength, the scattering radii at the transmitter and the receiver, the distance
between TX and RX arrays, antenna beamwidths, and antenna spacing. Our model
suggests that full MIMO capacity gain can be achieved for very realistic values of
scattering radii, antenna spacing and range. It shows, in contrast to usual intuition,
that large antenna spacing has only limited impact on capacity under fairly general
conditions. Another case described by the model is the “pin-hole" channel where
spatial fading is uncorrelated and yet the channel has low rank and hence low
capacity.We show that this situation typically occurs for very large distances between
transmitter and receiver. In the 1 * 1 case (i.e. one TX and one RX antenna), the
pinhole channel yields capacities worse than the traditional Rayleigh fading channel.
Our results are validated by comparing with a ray tracing-based channel simulation.
We find a good match between the two models over a wide range of situations.
CLASSIFICATION
We briefly review the capacity formula for MIMO channels and present a classification
of MIMO channels. We restrict our discussion to the frequency-flat fading case and we
assume that the transmitter has no channel knowledge whereas the receiver has
perfect channel knowledge.
Where H is the M * N channel matrix, IM denotes the identity matrix of size M; and _ is
the average signal to-noise ratio (SNR) at each receiver branch. The elements of H
are complex Gaussian with zero mean and unit variance, i.e., [H]m;n _ CN(0; 1) for m
=1; 2; :::;M; n = 1;2; :::;N. Note that since H is random C will be random as well.
Assuming a piece-wise constant fading model and coding over many independent
fading intervals2, EHfCg can be interpreted as the Shannon capacity of the random
MIMO channel [5].
1The superscript _ stands for Hermitian transpose.
2EH stands for the expectation over all channel realizations.
Note that the low rank models (ULR, CLR, 1_1 LR)above do not use the traditional
normal distribution for the entries of H but instead the product of two Gaussian
variables. This type of distribution is shown later to occur in important practical
situations. In the 1*1 case,The LR model has worsened fading statistics. This is due to
the intuitive fact that a double Rayleigh channel will fade \twice as often" as a
standard Rayleigh channel [4].
Figure 1: Propagation scenario for SIMO fading correlation. Each scatterer transmits a
plane-wave signal to a linear array several distributions can be considered for the
DOAs, including uniform, Gaussian, Laplacian etc. [10, 11]. The addition of different
plane-waves causes space selective fading at the RX antennas. It is well known that
the resulting fading correlation is governed by the angle spread, the antenna spacing
and the wavelength. The RX array response vector h can now be modeled as
(2)
where R_θr,dr is the M * M correlation matrix. For uniformly distributed DOAs, we find
[10, 12]
where S (odd) is the number of scatterers with corresponding DOAs θr. For “large"
values of the angle spread and/or antenna spacing, Rθr,dr will converge to the identity
matrix, which gives uncorrelated fading. For “small" values of θr,dr, the correlation
matrix becomes rank deficient (eventually rank one) causing(fully) correlated fading.
For the sake of simplicity,we furthermore assume the mean DOA to be orthogonal to
the array (bore-sight). Note that the model provided in (2) can readily be applied to
an array of TX antennas with corresponding antenna spacing and signal departure
angle spread.
The propagation path between the two arrays is obstructed on both sides of the link
by a set of significant near field scatterers (such as buildings and large objects)
referred to as TX or RX scatterers. Scatterers are modeled as omni-directional ideal
reflectors. The extent of the scatterers from the horizontal axis is denoted as Dt and
Dr, respectively. When omni-directional antennas are used Dt and Dr correspond to the
TX and RX scattering radius, respectively. On the RX side, the signal reflected by the
scatterers onto the antennas impinge on the array with an angular spread denoted
by θr, where θr is function of the position of the array with respect to the scatterers.
Similarly on the TX side we defined an angular spread θ t. The scatterers are
assumed to be located su_ciently far from the antennas for the planewave
assumption to hold. We furthermore assume that Dt;Dr _R (local scattering condition).
For uncorrelated TX antennas, the S_N channel matrix describing the propagation
between the N TX antennas and the S scatterers Y = [y1; y2; :::; yN] simply writes
where Gt = [g1; g2; :::; gN] is an S _N i.i.d. Rayleighfading matrix. However, there is
generally correlation between the TX antennas because of finite angle spread and
insufficient antenna spacing. Therefore, a more appropriate model becomes
impinge on the RX array. Denoting the distance between the s-th scatterer and the
m-th RX antenna as ds;m, the vector of received signals from the n-th TX antenna can
be written as
Collecting all RX and TX antennas according to Z =[z1; z2; :::; zN], we obtain
where Φ is the M * S matrix in (7). The problem with the expression in (8) is the
explicit use of deterministic phase shifts in the matrix Φ which makes the model
inconvenient.The simple equivalence result below allows us to get rid of this
inconvenience and obtain a new and entirely stochastic MIMO model.
After proper power normalization3 and replacing Y by (6), we obtain the following new
MIMO model
In (9), the spatial fading correlation between the TX antennas, and therefore the TX
spacing through The rank of the MIMO channel is primarily controlled through
. The model in (9) shows that it is well possible to have uncorrelated fading at
both sides,and yet have a rank deficient MIMO channel with reduced capacity. Such a
channel is dubbed a “pin-hole" because scattering (fading) energy travels through a
very thin air pipe, preventing the rank to build up. In practice, this occurs when the
product Dt,Dr is small compared to the range R, making θS small, and causing the rank
of to drop. Note that Dt, Dr play a role analogous to dt, dr in the green field case,
as shown
in [4].Eq. (9) suggests that in the scattering case the rank behavior of the MIMO
channel is mainly governed by the scattering radii and by the range. Scatterers can
be viewed as virtual antenna arrays with very large spacing and aperture. Unlike the
usual intuition, the physical antenna spacing has limited impact on the capacity
provided antennas remain uncorrelated, which occurs at λ/2 spacing for reasonably
high local angle spread/antenna beamwidth. Note that if scattering is absent at one
end of the link, the relevant parameter on that particular end driving the MIMO rank
becomes the antenna spacing. When either the TX or the RX antennas are fully
correlated due to small local angle spread, the rank of the MIMO channel also drops.
In this situation, both the diversity and multiplexing gains vanish, preserving only the
RX array gain. Note that there is no TX array gain since we assumed that the channel
is unknown in the transmitter. From the remarks above it follows that antenna
correlation causes rank loss but the converse is not true. The new model contains not
one but the product of two random Rayleigh distributed matrices. This is in contrast
with the traditional Rayleigh MIMO model of [2, 8]. Depending on the rank of .
of two independent Gaussians. In the high rank region, becomes the identity
matrix. Using the central limit theorem, the product GrGt approaches a single
Rayleigh distributed matrix,which justifies the traditional model in that particular
case. In the low rank (i.e. rank one) region, is the all one matrix. The MIMO
and low rank models can still be defined, according to the rank taken by .
The high rank model is the traditional Rayleigh channel. The low rank model has
double Rayleigh distribution. Note that the model does not suggest the existence of a
“correlated high rank" MIMO channel, which corresponds also to intuition.
4. MONTE CARLO SIMULATIONS
The capacity distribution predicted by the proposed stochastic MIMO model for
various values of the key parameters is compared to that achieved by an actual ray
tracing channel with the same parameters. The ray tracing model follows the
scenario depicted in Fig. 2. In all examples we used S = 20 TX and RX scatterers
which are randomly distributed uniformly around a line perpendicular to the x-axis.
We found that the final capacity results are insensitive to the particular distribution of
the scatterers as long as Dt,Dr and the angular spreads remain fixed. We used M = N
= 3 and placed the scatterers at a distance Rt from the TX array and Rr from the RX
array. We use Rr = Rr = Dt =Dr in all simulations in order to maintain a high local angle
spread and hence low antenna correlation. The frequency was set to 2GHz and the
SNR was 10 dB. To introduce random fading we use small random perturbations of
the TX and RX antenna array positions. We plot the capacity distribution (model and
ray tracing) for three separate sets of control parameters, covering the region
between the UHR and the ULR models. The curves obtained are shown in Fig. 3.
Fig. 4 illustrates the impact of the rank of on the capacity in the 1 * 1 case.
The proposed channel model predicts the capacity distribution up to one
Fig .3 Capacity c.d.f. obtained with MIMO model for three sets of parameters. From
left to right. Set 1: Dt = Dr = 30m; R = 1000km: Set 2: Dt = Dr = 50m;R = 50km: Set
3: Dt = Dr = 100m; R = 5km:
Figure 4: Capacity c.d.f. obtained for the 1 * 1 model. We use two sets of parameters:
from left to right. Set 1: Dt = Dr = 30m; R = 1000km. Set 2: Dt = Dr =100m; R =
5km:bps/Hz in all cases and becomes almost exact as we approach UHR and ULR
regions.
Finally, we look at the capacity (rank) build-up as function of the scattering radius.Fig.
5 is a plot of average capacity for varying Dt =Dr with R fixed at 10 km. The high
capacity region is quickly attained, even for a very large range. Existing
measurements suggest practical scattering radiuses of around 100 meters [11].
5. CONCLUSION
Figure 5: Mean capacity as a function of Dt = Dr: The range R is fixed to 10km. The
capacity builds up quickly as the scattering radius increases. outdoors for very
reasonable values of scattering radius, almost regardless of how large the antenna
spacing is. We pointed out the
existence of the pin-hole channels which can occur for very large values of the range
R.
REFERENCES
[1] A. J. Paulraj and T. Kailath, \Increasing capacity inwireless broadcast systems using distributed
transmission/directional reception," U. S. Patent, no. 5,345,599,reless systems employing OFDM-based
spatial multiplexing
[2] H. B¨olcskei, D. Gesbert, A. Paulraj, \On the capacity of IEEE Transaction on Communications, revised
Sept.
2000.
[3] D. Gesbert, H. B¨olcskei, D. Gore, A. Paulraj, \Outdoor MIMO wireless channels: Models and performance
prediction"
IEEE Transaction on Communications, submitted July 2000.
[5] J. Bach Andersen, \Array gain and capacity for know random to appear in the IEEE Journal on Selected
Areas in Com-
munications, 2000.