0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2323295, IEEE Transactions on Vehicular Technology
3
P
1
P
N
Space-
time
coding
Beam
Forming
U
. . .
Rx1Tx1
TxN
Space-
time
decoding
. . .
Channel
estimation
. . .
Feedback path
RxK
Power
control
Output
data
Input
data
. . .
Fig. 1. System diagram.
to exploit the available CSI. A complex orthogonal STBC
(OSTBC), which is represented by an N × T transmission
matrix D, is used to encode L input symbols into an N-
dimensional vector sequence of T time slots. The matrix D
is a linear combination of L symbols satisfying the complex
orthogonality: DD
H
= ε(|d
1
|
2
+ ··· + |d
L
|
2
)I
N
, where
{d
l
}
l=1,...,L
are the L input symbols, and ε is a constant
which depends on the STBC transmission matrix [4]. Hence,
the transmission rate of the STBC is r = L/T .
In this paper, we assume that CSI is perfectly known at
the receiver, and the transmitter has the knowledge of R
t
and
receives R
r
and imperfect (time-delayed) mean feedback from
the receiver. The channel matrix H
w
in H can be related to
its τ time-delayed version
ˆ
H
f
at the transmitter as [34]
H
w
= ρ
ˆ
H
f
+ E
w
=
ˆ
H
w
+ E
w
, (1)
where E
w
is a K × N channel error matrix independent of
ˆ
H
f
. The entries of the K ×N matrix
ˆ
H
f
are i.i.d. zero mean
complex Gaussian r.v.s of unit variance, while the entries of
E
w
are i.i.d. complex Gaussian r.v.s with zero mean and
variance σ
2
e
= 1 − ρ
2
, where ρ is the correlation coefficient
given by ρ = J
0
(2πf
d
τ), J
0
(·) is the zero-order Bessel
function of the first kind [37], and f
d
is the maximum Doppler
frequency [38]. The correlation between the entries of
ˆ
H
f
and the entries of H
w
is given by the correlation coefficient
ρ. The entries of
ˆ
H
w
= ρ
ˆ
H
f
∼ CN(0, ρ
2
). Based on the
Kronecker relationship, the channel matrix H is expressed as
H = R
1/2
r
(
ˆ
H
w
+ E
w
)R
1/2
t
=
ˆ
H + E, where K × N matrix
ˆ
H = R
1/2
r
ˆ
H
w
R
1/2
t
is the channel mean feedback [7, 26], and
E = R
1/2
r
E
w
R
1/2
t
. The correlation matrix of H conditioned
on
ˆ
H is given by
e
R
h
= E{H
H
H|
ˆ
H} =
ˆ
H
H
ˆ
H + E{E
H
E} (2)
where
e
R
h
is a N × N matrix. Utilizing E = R
1/2
r
E
w
R
1/2
t
,
E{E
H
E} in (2) can be expressed as
E{E
H
E} = E{(R
1/2
t
)
H
E
H
w
(R
1/2
r
)
H
R
1/2
r
E
w
R
1/2
t
}
= R
1/2
t
E{E
H
w
R
r
E
w
}R
1/2
t
= σ
2
e
P
K
k=1
ζ
r,k
R
t
(3)
where {ζ
r,k
, k = 1, . . . , K} are the eigenvalues of R
r
. The
last equality in (3) can be obtained from the following Lemma.
Lemma: If E
w
is a complex Gaussian random matrix
whose entries are i.i.d. with zero-mean and variance σ
2
e
,
then E{E
H
w
R
r
E
w
} = σ
2
e
P
K
k=1
ζ
r,k
I
N
, where {ζ
r,k
} are the
eigenvalues of R
r
.
Proof: Let R
r
= V
r
Λ
r
V
H
r
be the eigenvalue decompo-
sition (EVD) of R
r
, where Λ
r
= diag{ζ
r,1
, ..., ζ
r,K
}, then
we have:
E
H
w
R
r
E
w
= E
H
w
V
r
Λ
r
V
H
r
E
w
=
e
E
H
w
Λ
r
e
E
w
=
K
X
k=1
ζ
r,k
e
w,k
e
H
w,k
(4)
where the K × N matrix
e
E
w
= V
H
r
E
w
, and {e
w,k
} are the
N ×1 column vectors of
e
E
H
w
. Since V
r
is a unitary matrix,
˜
E
w
and E
w
have the same statistical distribution. Thus the entries
of
˜
E
w
are also i.i.d. complex Gaussian r.v.s with zero mean
and variance σ
2
e
= 1 − ρ
2
. Hence,
E{E
H
w
R
r
E
w
} =
X
K
k=1
ζ
r,k
E{e
w,k
e
H
w,k
} = σ
2
e
X
K
k=1
ζ
r,k
I
N
.
(5)
Substituting (3) into (2) gives
˜
R
h
=
ˆ
H
H
ˆ
H + σ
2
e
X
K
k=1
ζ
r,k
R
t
(6)
Using EVD,
˜
R
h
can be expressed as
˜
R
h
= V
h
Λ
h
V
H
h
(7)
where the N × N unitary matrix V
h
contains the N eigen-
vectors of
˜
R
h
, Λ
h
= diag{a
1
, ..., a
N
} is the diagonal matrix
containing the nonnegative eigenvalues {a
n
} sorted in a de-
creasing order.
With beamforming U and power control P, the transmitted
signal matrix can be expressed as
X = UPD (8)
where the N × N unitary matrix U (whose elements are
u
ij
, i, j = 1, . . . , N) is set equal to the orthogonal eigenvector
matrix V
h
of
˜
R
h
[26], P = diag(
√
P
1
,
√
P
2
, ...,
√
P
N
) is
a diagonal N × N matrix, where {P
n
, n = 1, . . . , N} is the
power control to the N eigen-beams with the following power
constraint.
X
N
n=1
P
n
= 1, P
n
≥ 0. (9)
Thus, the received signal matrix can be written as
Y = HX + Z = HUPD + Z (10)
where Y is a K × T received signal matrix, Z is a K × T
noise matrix with i.i.d. entries modeled as complex Gaussian
r.v.s with zero-mean and variance σ
2
n
. The equivalent received
signal vector after the space-time decoder of the receiver can
be expressed as [4, 5]
y = ε||HUP||
2
F
d + ˜z (11)
where y is an L ×1 received signal vector, d = [d
1
, . . . , d
L
]
T
is the L × 1 input data symbol vector. Each symbol in the
transmission matrix has an average power T
¯
S/(εL), where
¯
S is the average transmitted power radiated from N transmit
antennas, ||HUP||
2
F
=
P
N
n=1
P
n
P
K
k=1
|
P
N
i=1
h
ki
u
in
|
2
, and
˜z is an L ×1 noise vector whose elements are i.i.d. complex
Gaussian r.v.s with zero-mean and variance ε||HUP||
2
F
σ
2
n
.