Anda di halaman 1dari 5

KEEE494: 2nd Semester 2009 Week 13

MIMO System Capacity


1 MIMO Channel Capacity for Equal Power Allocation
The capacity indicates the maximum possible data rate without error theoretically. In this class, we derive the
capacity of the MIMO systems.
Model: Assume x
1
, , x
N
T
are independent and zero-mean such as
E[x
i
x

j
] = 0, for i = j
and the total power of the transmitted signal is P and the equal power is allocated to each transmit antenna such
as
E[|x
i
|
2
] =
P
N
T
for all i {1, , N
T
}
Now consider the covariance matrix of x given as
R
xx
= E[xx
H
]
Figure 1: N
R
N
T
MIMO channel.
Then,
R
xx
=
_

_
E[|x
1
|
2
] E[x
1
x

2
] E[x
1
x

N
T
]
E[x

1
x
]
2
E[|x
2
|
2
] E[x
2
x

N
T
]
.
.
.
E[x
N
T
x

1
] E[x
N
T
x

2
] E[|x
N
T
|
2
]
_

_
=
_

_
P
N
T
0 0
0
P
N
T
0
.
.
.
0 0
P
N
T
_

_
=
P
N
T
I
N
T
1
where I
N
T
is the N
T
N
T
identity matrix. Hence, total power signal power is P = trace(R
xx
).
Let us assume g
ij
, which is the channel gain from jth transmit antenna to ith receive antenna, are all known at
the receiver but unknown at the transmitter. Also let us assume that the normalized channel gain such that
N
T

j=1
|g
ij
|
2
= N
T
(or on average
N
T

j=1
E[|g
ij
|
2
] = N
T
).
Signals
Received signals:
r = Gx + n
Covariance matrix of the AWGN:
R
nn
=
2
I
N
R
Average SNR at each receive antenna

i
=
P

2
=
(

N
T
j=1
|g
ij
|
2
)
P
N
T

2
=
P

2
where we assume that g
i,j
is known (deterministic) normalized as

N
T
j=1
|g
ij
|
2
= 1. For random case of
g
ij
,
i
is the random variable. For example, for the Rayleigh fading case, that is, g
ij
is complex Gaussian,

N
T
j=1
|g
ij
|
2
is the central chi-square random variable with 2N
T
degree of freedom.
MIMO system channel capacity derivation
The channel capacity is dened as the maximum possible transmission data rate such that the probability
of error is arbitrary small.
Singular Value Decomposition (SVD) theorem: Let G be the channel N
R
N
T
matrix. Then, it can be
written as
G = UDV
H
where
D : N
R
N
T
non-negative diagonal matrix
U : N
R
N
T
unitary matrix
V : N
T
N
T
unitary matrix
(1) Note that the diagonal entries of D are the non-negative square roots of eigenvalues of matrix GG
H
.
Denote as the eigenvalues of GG
H
, that is,
(GG
H
)y = y
where y is the N
R
1 eigenvector. Then,

is called singular values of G.


(2) The column of V are eigenvectors of GG
H
.
(3) The columns of V are eigenvectors of G
H
G.
2
Now the received signal can be rewritten as
r = UDV
H
x + n.
Let us introduce the following transformation:
r

= U
H
r
x

= V
H
x
n

= U
H
n
Then, we can rewrite the received signal as
r

= Dx

+ n

The number of nonzero eigenvalues of matrix GG


H
is equal to the rank of matrix G, denoted by t, that is, t is
the rank of G.
For the N
R
N
T
matrix G, the rank is at most
t = min(N
R
, N
T
)
which means at most t of its singular values are nonzero. Let us denote the singular values of G by

i
for
i = 1, 2, , t. Then, we can consider the following two cases:
(1) If N
R
< N
T
and t = N
R
= min(N
T
, N
R
), we have
r

i
=
_

i
x

i
+n

i
, i = 1, 2, , t
r

i
= n

i
, i = t + 1, , N
R
which is illustrated in Fig. 2.
Figure 2: SVD MIMO channels for t = N
R
= min(N
T
, N
R
).
(2) If N
R
> N
T
, we have t = min(N
T
, N
R
) which is shown in Fig. 3.
Finally, using the Shannon capacity formula, the average channel capacity C is
C = W
t

i=1
log
2
_
1 +
P
r
i

2
_
3
Figure 3: SVD MIMO channels for t = N
T
= min(N
T
, N
R
).
where P
r
i
=

i
P

2
is the received signal power in the ith sub-channel. Hence, we can rewrite the channel capacity
as
C = W
t

i=1
log
2
_
1 +

i
P

2
_
= W log
2
t

i=1
_
1 +

i
P
N
T

2
_
2 MIMO Channel Capacity for Adaptive Power Control
The channel capacity of the MIMO systems was derived as
C = W
t

i=1
_
1 +

i
P
N
T

2
_
Let us assume that the channels are known at the transmitter. Then we can allocate the power to each transmit
antenna to maximize the channel capacity. In this case, the channel can be rewritten as
C = W
t

i=1
_
1 +

i
P
i

2
_
where P
i
is the transmit power to the ith transmit antenna. Now we want to maximize C by P
i
.
Solution: Water-lling principle can be applied in deriving the capacity under the following power con-
straint:
N
T

i=1
P
i
= P, i = 1, 2, , N
T
where P
i
is the power allocated to antenna i and P is the total power, which shall be kept constant. Then
consider the normalized capacity given as
C
W
=
N
T

i=1
log
2
_
1 +
P
i

2
_
4
Following the method of Lagrangian multipliers, let us introduce the function
Z =
N
T

i=1
log
2
_
1 +
P
i

2
_
+
_
P
N
T

i=1
P
i
_
where is the Lagrangian multipliers,
i
is the ith channel matrix singular value, and
2
is the noise
variance.
The unknown transmit powers P
i
are determined by setting the partial derivatives of Z to zero such as
Z
P
i
= 0

Z
P
i
=
1
ln 2

i
/
2
1 +P
i

i
/
2
= 0
Thus we obtain for P
i
as
1 +
P
i

2
=

i
/
2
ln 2
P
i
=

2

i
_

2
ln 2
1
_
=
1
ln 2


2

i
Since P
i
0, we nally have
P
i
=
_
1
ln 2


2

i
_
+
where
[x]
+
=
_
x, if x > 0
0, if x 0
Now we consider the singular value decomposition of channel matrix G. Then, the received power at
sub-channel i in the equivalent MIMO channel model is given by
P
r
i
=
i
_
1
ln 2


2

i
_
+
=
_

i
ln 2

2
_
+
The MIMO channel capacity is then
C = W
t

i=1
log
2
_
1 +
P
r
i

2
_
= W
t

i=1
log
2
_
1 +
1

2
(
i

2
)
+
_
5

Anda mungkin juga menyukai