Multiuser MIMO Downlink: Multiplexing Gain

Without Free Channel Information

Umer Salim
Mobile Communications Department
Eurecom Institute
06560 Sophia Antipolis, France
[email protected]
Dirk Slock
Mobile Communications Department
Eurecom Institute
06560 Sophia Antipolis, France
[email protected]
AbstractA multiuser MIMO system is considered with no
initial assumption of channel state information (CSI) at any of
the receivers or the BS transmitter. For a system working under
frequency-division duplexing (FDD) mode, simple practically
realizable transmission strategy is proposed which provides
necessary channel state information to both sides with minimal
resource utilization for downlink (DL) data transmission. For
the given transmission strategy, high SNR degrees of freedom
(DOF) for the DL channel are specied. If the users have data
to transmit in the uplink (UL) direction, with no extra burden
of training, full multiplexing gain of the multiple-access channel
(MAC) can be achieved. Thus the given strategy makes the system
fully scalable for the data transmission in both directions.
I. INTRODUCTION
In multiple-antenna downlink channels, capacity or achiev-
able data rates can be excessively increased just by adding
multiple antennas at the transmitting end. Thus if a base station
(BS) has M transmit antennas and the number of users in the
system is K with K M, this downlink channel can support
data rates M times larger than a single antenna BS, although
all users may have single antenna each in both cases [1], [2],
[3]. So under favorable conditions, the sum capacity of the
downlink channel is comparable to the capacity of a point-
to-point MIMO channel having the same number of transmit
and receive antennas. Apart from this sum capacity aspect,
there are two advantages of this broadcast channel. It requires
mobile users to have a single antenna each so user terminals
are quite inexpensive and simple. The second advantage is
that point-to-point MIMO links are plagued by line-of-sight
channel conditions where channel matrices are of reduced
rank and they lose their multiplexing abilities. In a multiuser
channel, naturally users are far apart so the assumption of
independent channel for each user holds very well and the
channel matrix is of full-rank with probability one and is much
well-conditioned as compared to the channel matrix of a point-
to-point MIMO link [4].
But these promising advantages of multiuser MIMO dont
come for free. To realize these high throughputs, BS has
to transmit to multiple users over the same bandwidth. Or-
thogonal transmission schemes such as time-division multiple
access (TDMA), frequency-division multiple access (FDMA)
and code-division multiple access (CDMA) are highly sub-
optimal as effectively BS will be transmitting to a single user
over a particular resource. The other price to pay to achieve
these high data rates is that BS must know the forward channel
to all users [1]. This point is in sharp contrast to point-to-point
MIMO. In point-to-point MIMO, channel state information at
the transmitter (CSIT) only affects the power offset of the
capacity. The slope of the capacity versus SNR curve, normally
termed as the multiplexing gain or the degrees of freedom
(DOF), remains unaffected by CSIT [5], [3].
We dont impose any initial assumption of channel knowl-
edge on either side. But we dont prevent any side (transmitter
and receiver) to learn/feedback the channel and subsequently
use this information for precoding/decoding of data. Most
of the initial results on the information theoretic capacity
analysis of the broadcast channel came with the assumption
of perfect channel state information at the transmitter (CSIT),
and each user knows its own channel (CSIR). Inherently all
channels are non-coherent and the users (receivers) need to
estimate the channels implicitly (data driven) or explicitly by
some kind of training (pilots transmission) to get CSIR. In
frequency-division duplex (FDD) mode of operation, downlink
(forward) channels are normally different from the uplink (re-
verse) channels. So the users need to feedback their estimated
forward channel information on the reverse link. On the other
hand, the acquisition of CSIT gets facilitated when the system
operates under time-division duplex (TDD) mode. In this case,
reciprocity implies that the forward channel matrix is the
transpose of the reverse channel matrix [6]. As most of the
current wireless systems use FDD mode, so we focus on FDD
system in this contribution.
Notation: E denotes statistical expectation. Lowercase letters
represent scalars, boldface lowercase letters represent vectors,
and boldface uppercase letters denote matrices. A

denotes
the Hermitian of matrix A.
II. SYSTEM MODEL
The system we consider consists of one BS having M
transmitting antennas and K single-antenna user terminals. In
the downlink, the signal received by k-th user can be expressed
as
y
k
= h

k
x +n
k
, k = 1, 2, . . . , K (1)
where h
1
,h
2
, . . . , h
K
are the channel vectors of users 1
through user K with h
k
C
M1
(C
M1
denotes the M-
dimensional complex space), x C
M1
denotes the M-
dimensional signal transmitted by the BS and n
1
, n
2
, . . . , n
K
are independent complex Gaussian additive noise terms with
zero mean and unit variances. We denote the concatenation of
the channels by H

= [h
1
h
2
h
K
], so H is the K M
forward channel matrix with k-th row equal to the channel
of the k-th user (h

k
). Similarly G is the M K reverse
channel matrix. Due to frequency division duplexing, we take
them completely independent of each another. The BS has an
average power constraint of P for the DL input and each user
has the same power constraint of P
u
.
The channel is assumed to be block fading having coherence
length of T symbol intervals where fading remains the same,
with independent fading from one block to the next [7]. The
entries of the forward channel matrix H and the reverse
channel matrix G are independent and identically distributed
(i.i.d.) complex Gaussian with zero mean and unit variance.
Initially all receivers and the BS transmitter are oblivious of
the channel realization in each block.
In normal downlink scenarios, the number of users (K) will
be more than the number of BS transmit antennas (M). It
is well-known that with perfect CSIT and CSIR, downlink
channel with M transmit antennas and K single antenna
users with K M achieves the multiplexing gain (DOF)
of M [8] i.e., the dominant term of the sum-capacity of this
downlink channel is M log(P). Extra number of users does not
contribute to increasing the multiplexing gain of this system
although denite power gain can be achieved by scheduling
over the users. In this contribution, our main point of concern
is the multiplexing gain or the DOF of non-coherent downlink
channel so we focus our attention on the case with K = M.
III. TRANSMISSION STRATEGY AND ANALYSIS
For a downlink channel having a transmitter equipped with
M transmit antennas and M single antenna receivers with
perfect CSIT and CSIR, the rst order term of the sum
capacity is M log(P) [8]. If we compare this to the capacity of
the same downlink channel with only CSIR available where
the dominant term of the sum capacity is only log(P), the
difference in the sum capacity forcefully dictates that the
transmitter must be fed back the DL channel information.
For our block fading channel with coherence length of T
symbol intervals, we divide this interval in four phases, 1)
Initial UL and DL training, 2) Uplink Feedback, 3) Final DL
training and 4) coherent data transmission. The rst phase is
Fig. 1. Coherence interval Divided in Training and Data Phases
the initial UL and DL training phase where users transmit
pilots on the UL frequency so that BS estimates the UL
channels and BS transmits pilots on the DL frequency so
that each user estimates its corresponding DL channel. But
to multiplex data streams of multiple users, BS requires the
information of DL channel which is fed back in the second
feedback interval. Based upon this channel information, BS
may choose some transmission strategy which could be a
simple linear beamforming strategy like zero forcing (ZF),
some non-linear strategies like vector perturbation or the
optimal dirty paper coding (DPC). Thus the third phase is the
downlink training phase where the BS transmits pilots so that
users estimate their corresponding effective channels. When
this third phase ends, both sides of the downlink channel have
necessary channel state information albeit imperfect. Thus
starting from a downlink channel with no CSIT and no CSIR,
reaching up to the forth data phase, we have a downlink
channel with imperfect CSIT and CSIR and hence in this
data phase, BS may choose good transmission strategies and
users can decode data coherently. The data rates obtained and
their scaling with SNR show that these training phases are
benecial.
Below we give a detailed analysis of the transmission phases
mentioned above.
A. Initial UL and DL Training Phase
In this training phase, users transmit pilot signals which are
known at the BS. As there are M users, hence the length of this
training interval is T
1
M. For this uplink training, the use
of orthogonal training sequences by all users is very attractive
because in that case all users can transmit simultaneously to
the BS with their full power without interfering with each
other. Thus pilot signal matrix (combined from all users) is

T
1
A where A is a MT
1
unitary matrix hence AA

= I
M
where I
M
denotes a M M identity matrix. If Y
u
denotes
the MT
1
matrix of the received signal by M antennas of the
BS in this training interval of length T
1
, the system equation
for this uplink training phase becomes
Y
u
=
_
P
u
T
1
GA+Z
u
(2)
where Z
u
is a M T
1
matrix having i.i.d. zero mean unit
variance complex Gaussian noise entries. As pilot signal
matrix A is known at the BS, it can formulate an MMSE
estimate of the uplink channel matrix G which is given by

G =

P
u
T
1
P
u
T
1
+ 1
Y
u
A

(3)
And the estimation error corresponding to each UL channel
entry is given by

2
1UL
= E[|G
ij


G
ij
|
2
] =
1
P
u
T
1
+ 1
(4)
Similarly on the DL frequency, BS transmits pilot signals
which are known to all users. Thus they are capable of esti-
mating the DL channel realization. Each user has to estimate
M channel coefcients which link M antennas of the BS to its
single antenna. The channel vector for user k can be expressed
as h
k
=

h
k
+

h
k
where

h
k
is the estimated channel at k-
th user and

h
k
is the estimation error vector both with i.i.d.
Gaussian entries. The estimation error variance for any DL
channel entry denoted by
2
1DL
is given by

2
1DL
= E[|H
ij


H
ij
|
2
] =
1
T
1
P/M + 1
(5)
The estimation error variance for each UL (DL) channel
entry goes inversely proportional to the training length T
1
and
the power constraint of the user terminals P
u
(the BS power
constraint P).
B. Uplink Feedback Phase
In the initial training phase, BS has estimated the UL
channel matrix G hence on the UL frequency band, users can
transmit simultaneously and BS can decode the data of this
MAC channel based upon its estimate

G. We adopt analog
feedback strategy at the users side hence k-th user just feeds
back M coefcients of

h
k
assuming them to be perfect. As
each user comes with a single antenna so the transmission of
M complex coefcients will require this interval length to be
at least M hence T
2
M.
y
FB
=
_
P
u
Gf +z
FB
(6)
Here y
FB
represents the M-dimensional vector received by
BS for one symbol interval during this feedback phase, f
denotes the data sent by M users which actually consists of
the DL channel coefcients estimated by the users in the rst
training interval. BS decodes f based upon its knowledge of

G
considering that estimate as perfect. For the k-th user channel
h
k
, the BS estimate is denoted by

h
k
and

h
k
denotes the
corresponding estimation error. The variance of each entry of

h
k
is
2
FB
. This estimation error gets augmented (as compared
to
2
1DL
) with the fact that k-th user transmitted

h
k
instead
of h
k
and the BS had access to

G in place of G.
C. BS Transmission Strategy: ZF Precoding
It is known that the dirty paper coding (DPC) is the capacity
achieving transmission scheme for MIMO broadcast channel
and achieves the full capacity region [9] but this scheme is
complex and its implementation is quite tedious. So a lot of
research has been carried out to analyze the performance of
simpler linear precoding schemes. Zero forcing precoding, one
of the simplest linear precoding strategy, has been shown to
behave quite optimally at asymptotically high values of SNR
and achieves the full DOF of a coherent downlink channel [8].
It means that the rst order term of the sum capacity of the
downlink channel remains same whether one employs DPC
or ZF precoding at the BS. In this contribution we are mainly
interested in analyzing the DOF obtainable with some simple
transmission scheme hence BS uses ZF precoding based upon
the knowledge of the DL channel matrix

H obtained through
explicit feedback from users.
In ZF precoding, unit norm beamforming vector for user
k (denoted as v
k
), is selected such that it is orthogonal to
the channel vectors of all other users. Hence with perfect
CSIT, each user will receive only the beam directed to it
and no multi-user interference will be experienced. For the
case in hand, where the BS has imperfect estimate of the
channel matrix, there will be some residual interference. If we
represent ZF beamforming matrix by

V = [ v
1
v
2
v
M
], the
transmitted signal x becomes x =

Vu and the signal received
by user k (1) can be expressed as
y
k
= h

Vu +n
k
= h

k
v
k
u
k
+

j=k
h

k
v
j
u
j
+n
k
(7)
Due to imperfect MMSE estimation at the BS and the choice
of ZF beamforming unit vectors, we have
h

k
v
j
=

h

k
v
j
+

h

k
v
j
=

h

k
v
j
(8)
hence the received signal at k-th user becomes
y
k
= h

k
v
k
u
k
+

j=k

k
v
j
u
j
+n
k
= h
k,k
u
k
+

j=k
h
k,j
u
j
+n
k
(9)
h
k,k
is the effective scalar channel for user k and h
k,j
are
the coefcients which arise due to imperfect ZF beamforming
as BS had no access to perfect channel realizations. Although
users estimated their corresponding channel vectors in the rst
training phase to feed them back later, but they know nothing
about their effective scalar channels.
D. Final DL Training Phase
We assume a very simple downlink training strategy. If the
BS had the perfect knowledge of the forward channels to
all users, due to ZF beamforming vectors each user would
only receive the signal from the beam directed to it and no
interference from any other beam would be observed. Here BS
estimates the users channels and therefore channel estimates
and the corresponding ZF beamforming vectors are imperfect
so each user receives some unwanted signal contribution from
the beam directed to any other user. But this interference
is of the order of the channel noise so for this DL training
phase, BS activates all beams simultaneously for T
3
symbols
times. In each symbol interval, every user receives through
its effective scalar channel, the Gaussian noise of the channel
and the interference due to imperfect channel estimates and
ZF beamforming vectors.
y
k
= h
k,k
u
k
+

j=k
h
k,j
u
j
+n
k
(10)
Based upon this received signal and the known pilots, k-th user
can form the MMSE estimate of the effective scalar channel
h
k,k
which is given by

h
k,k
=
E[h
k,k
y

k
]
E[y
k
y

k
]
y
k
=
_
PT2
M
PT2
M
+
PT2
M
(M 1)
2
FB
+ 1
y
k
(11)
As v
k
is a unit vector independent of h
k
, so effective scalar
channel h
k,k
= h

k
v
k
is zero mean complex Gaussian with
unit variance. As a result, MMSE estimate

h
k,k
and the
estimation error

h
k,k
both are complex Gaussian
h
k,k
=

h
k,k
+

h
k,k
(12)

h
k,k
CN
_
0,
PT
3
M
PT
3
M
+
PT
3
M
(M1)
2
FB
+1
_

h
k,k
CN
_
0,
PT
3
M
(M1)
2
FB
+1
PT
3
M
+
PT
3
M
(M1)
2
FB
+1
_
The estimation error variance in estimating this scalar effec-
tive channel is inversely proportional to the downlink power
constraint. When this phase of downlink training ends, both
the BS and all of the users have estimates for the channel
and coherent transmission with imperfect CSIT and CSIR is
possible.
The length of this nal DL training phase T
3
is independent
of the number of transmit antennas M at the BS and the
number of users.
E. Coherent Data Phase
We adopt the strategy of independent data transmission to
all users from the BS with power equally divided among them.
So k-th user destined signal, u
k
is Gaussian i.i.d. i.e, u
k

CN(0, P/M). The intuition is that in case of perfect CSIT
and CSIR, Gaussian signals are the optimal ones.
After the training and the feedback phases described earlier,
both the BS and all users have imperfect channel estimates.
So with ZF beamforming employed, the signal y
k
received by
user k (9) may be expressed as
y
k
=

h
k,k
u
k
+

h
k,k
u
k
+

j=k
h
k,j
u
j
+n
k
(13)
The above equation differs a lot from (9) as there user k
was unaware of its scalar channel h
k,k
but (13) effectively
represents a point-to-point coherent channel with channel

h
k,k
known at user k, although there is Gaussian noise, some
interference coming from the ZF beamforming vectors of other
users and the noise due to imperfect estimation of the effective
channel at users side.
F. Lower Bound of the Achievable Rate
We are interested in calculating the achievable sum rate
of this DL channel or its lower bound which could at least
point to the number of DOF achievable. If we denote the rate
obtained by k-th user as R
k
, then it is the mutual information
between u
k
and y
k
with channel

h
k,k
known
R
k
= I(u
k
; y
k
) (14)
In this case, the problem is that we cannot simply use the
expression for the mutual information of known scalar channel
because of the presence of interference terms whose distribu-
tions are unknown. If we combine the noise, the interference
and the estimation error contribution in y
k
(eq. (13)) in an
effective additive noise w
k
, then
w
k
=

h
k,k
u
k
+

j=k
h
k,j
u
j
+n
k
(15)
Due to the use of MMSE estimation in the downlink
training, we remark that the signal is uncorrelated with the
noise and all interfering terms.
E[u
k
(

h
k,k
u
k
+

j=k
h
k,j
u
j
+n
k
)

] = 0 (16)
The above expectation is zero because of the property of
uncorrelated MMSE estimation error, the use of independent
signals for different users and that the noise is independent
of everything else. Now once we have shown that all additive
noise terms are uncorrelated with the desired signal, we can
invoke Theorem 1 from [10] which states that the worst case
uncorrelated noise has the zero mean Gaussian distribution.
So we can replace the effective scalar additive noise w
k
of
unknown distribution with a noise of the same second moment
but having Gaussian distribution, it will give a lower bound
to the rate R
k
of k-th user but we can instantly write the
expression for the mutual information as
R
k
E

h
k,k
log
_
1 +
|

h
k,k
|
2
E|u
k
|
2
E[w
k
w

k
|

h
k,k
]
_
= E

h
k,k
log
_
1 +
P
M
|

h
k,k
|
2
E[w
k
w

k
|

h
k,k
]
_
(17)
IV. HIGH SNR DOF OF THE SUM RATE
The rate for k-th user given in eq. (17) can further be lower
bounded as
R
k
E

h
k,k
log
_
P
M
|

h
k,k
|
2
E[w
k
w

k
|

h
k,k
]
_
= E

h
k,k
log
_
P
M
|

h
k,k
|
2
_
E

h
k,k
log
_
E[w
k
w

k
|

h
k,k
]
_
E

h
k,k
log
_
P
M
|

h
k,k
|
2
_
log
_
E[w
k
w

k
]
_
(18)
where the last inequality follows from the Jensens inequality.
With this, we only need to compute the 2nd moment of w
k
,
denoted as
2
w
= E[w
k
w

k
].
As all of the users are symmetrically distributed, so the sum
rate of this downlink channel is given by
R
sum
=
T T
1
T
2
T
3
T
MR
k
(19)
where we have incorporated the DOF loss in the sum rate due
to one feedback and two training phases.
If we increase the rst training phase duration T
1
, it
improves the quality of the

Gat the BS and of

Hat users but it
gives only a gain in SNR offset which is logarithmic in nature
but the coefcient (T T
1
T
2
T
3
) reduces the DOF of
the sum rate linearly with increase in T
1
so the optimal length
of the rst training phase should be the minimum possible at
high SNR, hence T
1
= M.
For the second feedback phase in the UL direction, minimal
length is M as each single antenna user needs at least
M intervals to feedback M channel coefcients. About the
third training phase in the downlink direction of length T
3
,
reasoning is not very different. With the increase in this
training interval, users are better able to estimate their effective
scalar channels which gives SNR gain, logarithmic in nature
but increase in T
3
directly hits DOF due to the coefcient
(T T
1
T
2
T
3
) in front of the logarithm. So to exploit the
maximum number of DOF at high SNR, the optimal (minimal)
value of T
3
comes out to be 1. Hence adopting these values,
the sum rate becomes
R
sum

T 2M 1
T
M
_
E
g
k,k
log
_
P
M
| g
k,k
|
2
_
log(
2
w
)
_
(20)
For very large values of P (the BS power constraint) and if
power constraints of users are of the same order as that of
P, it can be shown that DOF are maximized with minimal
training and feedback phases. So for limiting value of P, the
lower bound to the multiplexing gain of the sum rate becomes
lim
P
R
sum
log(P)

T (2M + 1)
T
M (21)
If we compare this multiplexing gain to the multiplexing gain
of the same downlink channel with full CSIR and no CSIT
where DOF is only 1, we see that even for very practical values
of the block coherence interval T in mobile environments, this
lower bound M[1(2M+1)/T] is comparatively much larger
and to make the BS learn the channel pays off very well.
An upper bound to the sum rate of non-coherent downlink
channel can be obtained by letting all the users cooperate
so that we get an M M MIMO system. For this the
multiplexing gain has been given to be M(1 M/T) in [11].
But the problem is inherently different. In broadcast channel,
the transmitter must know the DL channel. Due to FDD, the
users rst need to estimate DL channel to later feed it back. So
it does not seem probable that DL multiplexing gains higher
than given by our scheme can be achieved for this system.
V. UL DATA TRANSMISSION
As the system in hand has two different frequency bands,
out of which one is dedicated for UL transmission. Strictly
speaking as soon as the rst phase ends and BS obtains the
estimate of the UL channel matrix G, this information is
sufcient to use the UL channel as MIMO MAC and full DOF
can be obtained during the rest of the coherence interval (T
T
1
) [12]. But the users have to provide DL channel information
H to the BS to use DL frequency resource properly which
requires 2nd feedback interval. So after the second feedback
phase, UL frequency band can be used as MIMO MAC which
will give us multiplexing gain of M(1 2M/T).
Remark 1: To obtain full multiplexing gain at high DL
SNR, the UL power (users power constraints) should be of
the same order as of the DL SNR. As users receive some
interference due to imperfect channel estimates at the BS so if
the DL SNR goes on increasing for a xed UL power (which
means a xed estimation error at the BS), the signal power
and the interference power both will increase with DL SNR
and the system will become interference limited causing the
collapse of the DOF.
Remark 2: The channels of concern in this paper are fast
fading channels which arise for fast moving mobile users
e.g. for user speeds of 100Km/h, carrier frequency of 2GHz
and coherence BW of 100KHz, coherence time will be about
100 symbol intervals [13]. So even for BSs having 8 or
16 antennas, training/feedback interval minimization becomes
really necessary.
VI. CONCLUDING REMARKS
We studied the capacity of a multiuser MIMO downlink
channel without any initial assumption of channel knowledge.
We gave a complete transmission strategy, through which the
BS and all users acquire necessary channel information. Sum
rate analysis shows the achievability of signicant multiplex-
ing gain with this scheme in both UL and DL directions.
ACKNOWLEDGMENTS
Eurecoms research is partially supported by its industrial
members: BMW Group Research & Technology, Bouygues
Telecom, Cisco, Hitachi, ORANGE, SFR, Sharp, STMicro-
electronics, Swisscom, Thales. The research work leading to
this paper has also been partially supported by the European
Commission under the ICT research network of excellence
NewCom++ of the 7th Framework programme and by the
French ANR project APOGEE.
REFERENCES
[1] G. Caire and S. Shamai (Shitz), On the achievable throughput of a
multiantenna gaussian broadcast channel, IEEE Trans. on Information
Theory, vol. 49, pp. 16911706, July 2003.
[2] N. Jindal and A. Goldsmith, Dirty paper coding versus tdma for mimo
broadcast channels, IEEE Trans. on Information Theory, vol. 51, pp.
17831794, May 2005.
[3] N. Jindal, Mimo broadcast channels with nite rate feedback, IEEE
Trans. on Information Theory, vol. 52, pp. 50455060, November 2006.
[4] D. Gesbert, M. Kountouris, J. R. W. Heath, C. B. Chae, and T. Salzer,
From single user to multiuser communications: Shifting the mimo
paradigm, IEEE Sig. Proc. Magazine, 2007.
[5] I. E. Telatar, Capacity of multi-antenna gaussian channels, European
Transactions on Telecommunications, pp. 585595, November 1999.
[6] T. L. Marzetta, How much training is required for multiuser mimo?, in
Proc. Asilomar Conference on Signals, Systems and Computers, Pacic
Grove, CA, USA, November 2006, pp. 359363.
[7] T. Marzetta and B. Hochwald, Capacity of a mobile multiple-
antenna communications link in rayleigh at fading, IEEE Trans. on
Information Theory, vol. 45, pp. 139157, January 1999.
[8] N. Jindal, A high snr analysis of mimo broadcast channels, in Proc.
IEEE Int. Symp. Information Theory, Adelaide, Australia, 2005, pp.
23102314.
[9] H. Weingarten, Y. Steinberg, and S. Shamai, The capacity region of the
gaussian multiple-input multiple-output broadcast channel, IEEE Trans.
on Information Theory, vol. 52, pp. 39363964, September 2006.
[10] B. Hassibi and B. M. Hochwald, How much training is needed in
multiple-antenna wireless links?, IEEE Transactions on Information
Theory, vol. 49, pp. 20582080, April 2003.
[11] L. Zheng and D. N. C. Tse, Communication on the grassmann
manifold: A geometric approach to the noncoherent multiple-antenna
channel, IEEE Trans. on Information Theory, vol. 48, pp. 359383,
February 2002.
[12] S. Murugesan, E. Uysal-Biyikoglu, and P. Schniter, Optimization
of training and scheduling in the non-coherent mimo multiple-access
channel, EEE Journal on Selected Areas in Communication, vol. 25,
pp. 14461456, September 2007.
[13] D. Tse and P. Viswanath, Fundamentals of Wireless Communications,
Cambridge, U.K. Cambridge Univ. Press, 2005.