Vous êtes sur la page 1sur 6

ON THE CAPACITY OF THE MIMO CHANNEL

- A TUTORIAL INTRODUCTION -

Bengt Holter

Norwegian University of Science and Technology


Department of Telecommunications
O.S.Bragstads plass 2B, N-7491 Trondheim, Norway
bholter@tele.ntnu.no

ABSTRACT the signals that originated from different transmit an-


tennas.
Multiple-input multiple-output (MIMO) systems
are today regarded as one of the most promising re-
search areas of wireless communications. This is due 2. CHANNEL CAPACITY
to the fact that a MIMO channel can offer a sig-
At the input of a communication system, discrete
nificant capacity gain over a traditional single-input
source symbols are mapped into a sequence of chan-
single-output (SISO) channel. In this paper, a tutor-
nel symbols. The channel symbols are then transmit-
ial introduction on the channel capacity of a MIMO
ted/conveyed through a wireless channel that by na-
channel will be given.
ture is random. In addition, random noise is added to
the channel symbols. In general, it is possible that two
1. INTRODUCTION different input sequences may give rise to the same
output sequence, causing different input sequences to
The increase in spectral efficiency offered by MIMO be confusable at the output. To avoid this situation,
systems is based on the utilization of space (or an- a non-confusable subset of input sequences must be
tenna) diversity at both the transmitter and the re- chosen so that with a high probability, there is only
ceiver. Due to the utilization of space diversity, MIMO one input sequence causing a particular output. It is
systems are also referred to as multiple-element an- then possible to reconstruct all the input sequences at
tenna systems (MEAs). With a MIMO system, the the output with negligible probability of error.
data stream from a single user is demultiplexed into A measure of how much information that can be
nT separate sub-streams. The number nT equals the transmitted and received with a negligible probability
number of transmit antennas. Each sub-stream is of error is called the channel capacity. To determine
then encoded into channel symbols. It is common this measure of channel potential, assume that a chan-
to impose the same data rate on all transmitters, but nel encoder receives a source symbol every Ts second.
adaptive modulation rate can also be utilized on each With an optimal source code, the average code length
of the sub-streams [1]. The signals are received by nR of all source symbols is equal to the entropy rate of the
receive antennas. source. If S represents the set of all source symbols
With this transmission scheme, there is a linear in- and the entropy rate of the source is written as H(S),
crease in spectral efficiency compared to a logarithmic the channel encoder will receive on average H(S) Ts in-
increase in more traditional systems utilizing receive- formation bits per second.1 Assume that a channel
diversity or no diversity. The high spectral efficien- codeword leaves the channel encoder every Tc second.
cies attained by a MIMO system are enabled by the In order to be able to transmit all the information
fact that in a rich scattering environment, the sig- from the source, there must be
nals from each individual transmitter appear highly
uncorrelated at each of the receive antennas. When H(S)Tc
R= (1)
the signals are conveyed through uncorrelated chan- Ts
nels between the transmitter and receiver, the signals 1 The entropy rate is a function of the statistical distribution
corresponding to each of the individual transmit an- of the source S. If the source S represent a discrete memoryless
random variable, the entropy rate of the source is equal to the
tennas have attained different spatial signatures. The source entropy, and is defined as H(S) = E[− logb (pS )]. It
receiver can use these differences in spatial signature is common to use b = 2 and the entropy is then expressed in
to simultaneously and at the same frequency separate information bits per source symbol.
information bits per channel symbol. The number R where x is the (nT ×1) transmit vector, y is the (nR ×
is called the information rate of the channel encoder. 1) receive vector, H is the (nR × nT ) channel matrix,
The maximum information rate that can be used caus- and n is the (nR × 1) additive white Gaussian noise
ing negligible probability of errors at the output is (AWGN) vector at a given instant in time. Through-
called the capacity of the channel. By transmitting out the paper, it is assumed that the channel matrix
information with rate R, the channel is used every is random and that the receiver has perfect channel
Tc seconds. The channel capacity is then measured knowledge. It is also assumed that the channel is
in bits per channel use. Assuming that the channel memoryless, i.e., for each use of the channel an inde-
has bandwidth W , the input and output can be rep- pendent realization of H is drawn. This means that
1
resented by samples taken Ts = 2W seconds apart. the capacity can be computed as the maximum of the
With a band-limited channel, the capacity is mea- mutual information as defined in (2). The results are
sured in information bits per second. It is common also valid when H is generated by an ergodic process
to represent the channel capacity within a unit band- because as long as the receiver observes the H process,
with of the channel. The channel capacity is then only the first order statistics are needed to determine
measured in bits/s/Hz. the channel capacity [3].
It is desirable to design transmission schemes that A general entry of the channel matrix is denoted
exploit the channel capacity as much as possible. Rep- by {hij }. This represents the complex gain of the
resenting the input and output of a memoryless wire- channel between the jth transmitter and the ith re-
less channel with the random variables X and Y re- ceiver. With a MIMO system consisting of nT trans-
spectively, the channel capacity is defined as [2] mit antennas and nR receive antennas, the channel
matrix is written as
C = max I(X; Y ), (2)  
p(x)
h11 · · · h1nT
 h21 · · · h2nT 
where I(X; Y ) represents the mutual information be-  
H= . .. .. , (5)
tween X and Y . Eq.(2) states that the mutual in-  .. . . 
formation is maximized with respect to all possible hnR 1 · · · hnR nT
transmitter statistical distributions p(x). Mutual in-
formation is a measure of the amount of information where
that one random variable contains about another vari-
able. The mutual information between X and Y can hij = α + β (6)
also be written as  β
= α2 + β 2 · e− arctan α (7)
I(X; Y ) = H(Y ) − H(Y |X), (3) = |hij | · e φij
. (8)

where H(Y |X) represents the conditional entropy be- In a rich scattering environment with no line-of-sight
tween the random variables X and Y . The entropy (LOS), the channel gains |hij | are usually Rayleigh
of a random variable can be described as a measure distributed. If α and β are independent and normal
of the amount of information required on average to distributed random variables, then |hij | is a Rayleigh
describe the random variable. It can also described as distributed random variable.
a measure of the uncertainty of the random variable.
Due to (3), mutual information can be described as 4. SISO CHANNEL CAPACITY
the reduction in the uncertainty of one random vari-
able due to the knowledge of the other. Note that the The ergodic (mean) capacity of a random channel
mutual information between X and Y depends on the with nT = nR = 1 and an average transmit power
properties of the channel (through a channel matrix constraint PT can be expressed as [2]
H) and the properties of X (through the probability 
distribution of X). The channel matrix H used in
C = EH max I(X; Y ) , (9)
the representation of the input/output relations of a p(x):P ≤PT
MIMO channel is defined in the next section.
where P is the average power of a single channel code-
word transmitted over the channel and EH denotes
3. SYSTEM MODEL
the expectation over all channel realizations. Com-
It is common to represent the input/output relations pared to the definition in (2), the capacity of the
of a narrowband, single-user MIMO link by the com- channel is now defined as the maximum of the mu-
plex baseband vector notation tual information between the input and the output
over all statistical distributions on the input that sat-
y = Hx + n, (4) isfy the power constraint. If each channel symbol at
the transmitter is denoted by s, the average power where Φ = E{xx† }2 is the covariance matrix of the
constraint can be expressed as transmit signal vector x. The total transmit power

is limited to PT , irrespective of the number of trans-
P = E |s|2 ≤ PT . (10) mit antennas. By using (4) and the relationship be-
tween mutual information and entropy, (13) can be
Using (9), the ergodic (mean) capacity of a SISO sys- expanded as follows for a given H
tem (nT = nR = 1) with a random complex channel
gain h11 is given by [4] I(x; y) = h(y) − h(y|x) (14)
 h(y) − h(Hx + n|x)
C = EH log2 1 + ρ · |h11 |2 , (11) = (15)
= h(y) − h(n|x) (16)
where ρ is the average signal-to-noise (SNR) ratio at = h(y) − h(n), (17)
the receiver branch. If |h11 | is Rayleigh, |h11 |2 follows
a chi-squared distribution with two degrees of freedom where h(·) in this case denotes the differential entropy
[5]. Eq.(11) can then be written as [4] of a continuous random variable. It is assumed that
 the transmit vector x and the noise vector n are in-
C = EH log2 1 + ρ · χ22 , (12)
dependent.
where χ22 is a chi-square distributed random variable Eq. (17) is maximized when y is gaussian, since
with two degrees of freedom. Figure 1 shows the the normal distribution maximizes the entropy for a
given variance [2]. The differential entropy of a real
SISO capacity gaussian vector y ∈ Rn with zero mean and covari-
ance matrix K is equal to 12 log2 ((2πe)n det K). For
6
a complex gaussian vector y ∈ Cn , the differential
entropy is less than or equal to log2 det(πeK), with
5 equality if and only if y is a circularly symmetric com-
plex Gaussian with E{yy† } = K [3]. Assuming the
Capacity [bit/s/Hz]

4 optimal gaussian distribution for the transmit vector


x, the covariance matrix of the received complex vec-
3 tor y is given by
  

2 E yy† = E (Hx + n) (Hx + n) (18)
  
1
= E Hxx† H† + E nn† (19)
= HΦH† + Kn (20)
= Kd + Kn .
0
0 2 4 6 8 10
SNR [dB]
12 14 16 18 20 (21)

Figure 1: Ergodic capacity a Rayleigh fading SISO The superscript d and n denotes respectively the de-
channel (dotted line) compared to the Shannon ca- sired part and the noise part of (21). The maximum
pacity of a SISO channel (solid line). mutual information of a random MIMO channel is
then given by
Shannon capacity for a gaussian channel (solid line)
I = h(y) − h(n)
and the capacity of a Rayleigh fading channel (dot-

ted line) according to (12). The Rayleigh fading curve = log2 det πe Kd + Kn − log2 [det (πeKn )]

d 
presented in Figure 1 equals the result in [6] (capacity = log2 det K + Kn − log2 [det (Kn )]
of a Rayleigh fading channel with optimal power and    
−1
rate adaption at the transmitter under the assump- = log2 det Kd + Kn (Kn )
  
tion of perfect channel estimation and return channel −1
= log2 det Kd (Kn ) + InR
free of errors and delay).   
−1
= log2 det HΦH† (Kn ) + InR .
5. MIMO CHANNEL CAPACITY
When the transmitter has no knowledge about the
The capacity of a random MIMO channel with power channel, it is optimal to use a uniform power distri-
constraint PT can be expressed as bution [3]. The transmit covariance matrix is then
 given by Φ = PnTT InT . It is also common to assume
C = EH max I(x; y) , (13) 2 The superscript † denotes Hermitian transpose
p(x):tr(Φ)≤PT
uncorrelated noise in each receiver branch described The number of parallel subchannels is determined by
by the covariance matrix Kn = σ 2 InR . The ergodic the rank of the channel matrix. In general, the rank
(mean) capacity for a complex AWGN MIMO channel of the channel matrix is given by
can then be expressed as [3, 4]
rank(H) = k ≤ min{nT , nR }. (29)
   
PT
C = EH log2 det InR + 2 HH† . (22) Using (29) together with the fact that the determinant
σ nT of a unitary matrix is equal to 1, (26) and (28) can
be expressed respectively as
This can also be written as
 k  
     ρ
ρ C = EH λi
C = EH log2 det InR + HH† , (23) log2 1 +
nT
(30)
nT i=1
 k  
 ρ 2
where ρ = PσT2 is the average signal-to-noise (SNR) = EH log2 1 + σ . (31)
ratio at each receiver branch. By the law of large i=1
nT i
numbers, the term n1T HH† → InR as nT gets large
and nR is fixed. Thus the capacity in the limit of large In (30), λi are the eigenvalues of the diagonal matrix
nT is Λ and in and (31), σi2 are the squared singular values
C = EH {nR · log2 (1 + ρ)} . (24) of the diagonal matrix Σ. The maximum capacity of a
MIMO channel is reached in the unrealistic situation
Further analysis of the MIMO channel capacity given when each of the nT transmitted signals is received
in (23) is possible by diagonalizing the product matrix by the same set of nR antennas without interference.
HH† either by eigenvalue decomposition or singular It can also be described as if each transmitted sig-
value decomposition. By using eigenvalue decompo- nal where received by a separate set receive antennas,
sition, the matrix product is written as giving a total number of nT · nR receiving antennas.
With optimal combining at the receiver and re-
HH† = EΛE† , (25) ceive diversity only (nT = 1), the channel capacity
can be expressed as [4]
where E is the eigenvector matrix with orthonormal 
columns and Λ is a diagonal matrix with the eigenval- C = EH log2 1 + ρ · χ22nR , (32)
ues on the main diagonal. Using this notation, (23)
can be written as: where χ22nR is a chi-distributed random variable with
    2nR degrees of freedom. If there are nT transmit an-
ρ tennas and optimal combining between nR antennas
C = EH log2 det InR + EΛE† . (26)
nT at the receiver, the capacity can be written as3 [4]
  
The matrix product HH† can also be described by us- ρ 2
C = EH nT · log2 1 + χ2nR . (33)
ing singular value decomposition on the channel ma- nT
trix H written as Eq.(33) represent the upper bound of a Rayleigh fad-
ing MIMO channel. In Figure 2, the Shannon capacity
H = UΣV† , (27)
of a SISO channel is compared to the upper bound of
where U and V are unitary matrices of left and right (33) with nT = nR = 6. Even though this bound on
singular vectors respectively, and Σ is a diagonal ma- the MIMO channel represent a special case, Figure 2
trix with singular values on the main diagonal. clearly shows the potential of the MIMO technology.
All elements on the diagonal are zero except for
the first k elements. The number of non-zero singular When the channel is known at the transmitter,
values k equals the rank of the channel matrix. Us- the maximum capacity of a MIMO channel can be
ing (27) in (23), the MIMO channel capacity can be achieved by using the water-filling principle [2] on
written as the transmit covariance matrix. The capacity is then
    given by
ρ † †  k  
C = EH log2 det InR + UΣΣ U . (28)  ρ
nT C = EH log2 1 + i λi (34)
i=1
nT
After diagonalizing the product matrix HH† , the ca-  k  
 ρ 2
pacity formulas of the MIMO channel now includes = EH log2 1 + i σ , (35)
unitary and diagonal matrices only. It is then easier to i=1
nT i
see that the total capacity of a MIMO channel is made 3 Assuming the artificial case of no interference between the

up by the sum of parallel AWGN SISO subchannels. received nT signals.


Ergodic capacity of a MIMO fading channel
transmit antenna selection is to improve the capacity
70
by not using the transmit antennas that correspond
to the linearly dependent columns, but instead redis-
60 tributing the power among the other antennas. Since
the total number of parallel subchannels in the sums
of eq.(31) and (35) is equal to the rank of the channel
Capacity [bit/s/Hz]

50

matrix, the optimal choice is to distribute the trans-


40
mit power on a subset of k transmit antennas that
30
maximizes the channel capacity. It is shown in [9]
that the optimal choice of k transmit antennas that
20 maximizes the channel capacity results in a channel
matrix that is full rank. In [10], a computationally ef-
10 ficient, near-optimal search technique for the optimal
subset based on classical waterpouring is described.
0
−10 −5 0 5 10 15 20 25 30 35 40 In [11], the capacity of MIMO systems with receive
SNR [dB]
antenna selection is analyzed. With such a reduced-
Figure 2: The Shannon capacity of a SISO channel complexity MIMO scheme, a selection of the L best
(dotted line) compared to the ergodic capacity of a antennas of the available nR antennas at the reciever
Rayleigh fading MIMO channel (solid line) with nT = is used. This has the advantage that only L receiver
nR = 6. chains are required compared to nR in the full-complexity
scheme. In [11], it is demonstrated through Monte
Carlo simulations that for nT = 3 and nR = 8, the ca-
where i is a scalar, representing the portion of the pacity of the reduced-complexity scheme is 20bits/s/Hz
available transmit power going into the ith subchan- compared to 23bits/s/Hz of a full-complexity scheme.
nel. The powernconstraint at the transmitter can be
expressed as i=1 T
i ≤ nT .
7. OUTAGE CAPACITY
Clearly, with a reduced number of non-zero singu-
lar values in (31) and (35), the capacity of the MIMO In this paper, the ergodic (mean) capacity has been
channel will be reduced because of a rank deficient used as a measure for the spectral efficiency of the
channel matrix. This is the situation when the signals MIMO channel. The capacity under channel ergodic-
arriving at the receivers are correlated. Even though ity where in (9) and (13) defined as the average of the
a high channel rank is necessary to obtain high spec- maximal value of the mutual information between the
tral efficiency on a MIMO channel, low correlation is transmitted and the received signal, where the maxi-
not a guarantee of high capacity [7]. In [8], the ex- mization was carried out with respect to all possible
istence of pin-hole channels is demonstrated. Such transmitter statistical distributions. Another mea-
channels exhibit low fading correlation between the sure of channel capacity that is frequently used is out-
antennas at both the receiver and transmitter side, age capacity. With outage capacity, the channel ca-
but the channels still have poor rank properties and pacity is associated to an outage probability. Capac-
hence low capacity. ity is treated as a random variable which depends on
the channel instantaneous response and remains con-
6. ANTENNA SELECTION stant during the transmission of a finite-length coded
block of information. If the channel capacity falls be-
The MIMO channel capacity has so far been opti- low the outage capacity, there is no possibility that
mized based on the assumption that all transmit and the transmitted block of information can be decoded
receive antennas are used at the same time. Recently, with no errors, whichever coding scheme is employed.
several authors have presented papers on MIMO sys- The probability that the capacity is less than the out-
tems with either transmit or receive antenna selec- age capacity denoted by Coutage is q. This can be
tion. As seen earlier in this paper, the capacity of the expressed in mathematical terms by
MIMO channel is reduced with a rank deficient chan-
nel matrix. A rank deficient channel matrix means Prob {C ≤ Coutage } = q. (36)
that some columns in the channel matrix are linearly
dependent. When they are linearly dependent, they In this case, (36) represents an upper bound due to
can be expressed as a linear combination of the other fact that there is a finite probability q that the channel
columns in the matrix. The information within these capacity is less than the outage capacity. It can also
columns is then in some way redundant and is not con- be written as a lower bound, representing the case
tributing to the capacity of the channel. The idea of where there is a finite probability (1 − q) that the
channel capacity is higher than Coutage , i.e., [10] S. Sandhu, R. U. Nabar, D. A. Gore, A. Paulraj,
”Near-optimal selection of transmit antennas for
Prob {C > Coutage } = 1 − q. (37) a MIMO channel based on shannon capacity,”
Conference Record of the 34th Asilomar Confer-
8. SUMMARY ence on Signals, Systems and Computers, 1:567–
571, 2000.
In this paper, a tutorial introduction on the capac-
ity of the MIMO channel has been given. The use [11] A. F. Molisch, M. Z. Win, J. H. Winters, A.
of multiple antennas on both the transmitter and re- Paulraj, ”Capacity of MIMO systems with an-
ceiver side of a communication link have shown to tenna selection,” In Proc. IEEE International
greatly improve the spectral efficiency of both fixed Conference on Communications (ICC), 2:570–
and wireless systems. The are many research papers 574, 2001.
published on MIMO systems, reflecting the percep-
tion that MIMO technology is seen as one of the most
promising research areas of radio communication to-
day.

9. REFERENCES

[1] S. Catreux, P. F. Driessen, L. J. Greenstein, ”At-


tainable throughput of an interference-limited
multiple-input multiple-output (MIMO) cellular
system,” IEEE Transactions on Communica-
tions, 49(8):1307–1311, aug 2001.
[2] T. M. Cover, J. A. Thomas, Elements of Infor-
mation theory, John Wiley & Sons, Inc., 1991.
[3] I. Telatar, ”Capacity of multi-antenna gaussian
channels,” AT&T Technical Memorandum, jun
1995.
[4] G. J. Foschini, M. J. Gans, ”On limits of wireless
communications in a fading environment when
using multiple antennas,” Wireless Personal
Communications, 6:311–335, aug 1998.
[5] J. G. Proakis, Digital Communications,
McGraw-Hill, Inc., 1995.
[6] M-S. Alouini, A. Goldsmith, ”Capacity of nak-
agami multipath fading channels,” Proc. Vehic-
ular Technology Conference, 1:358–362, 1997.
[7] D. Chizhik, G. J. Foschini, R. A. Valenzuela,
”Capacities of multi-element transmit and re-
ceive antennas: Correlations and keyholes,” Elec-
tronic Letters, 36(13):1099–1100, jun 2000.
[8] D. Gesbert, H. Bölcskei, D. Gore, A. Paulraj,
”MIMO wireless channels: capacity and per-
formance prediction,” Global Telecommunica-
tions Conference (GLOBECOM ’00), 2:1083–
1088, 2000.
[9] D. A. Gore, R. U. Nabar, A. Paulraj, ”Select-
ing an optimal set of transmit antennas for a
low rank matrix channel,” Proc. ICASSP 2000,
49(8):1307–1311, aug 2000.

Vous aimerez peut-être aussi