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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
Performance Analysis and Location Optimization
for Massive MIMO Systems with Circularly
Distributed Antennas
Ang Yang, Yindi Jing, Member, IEEE, Chengwen Xing, Member, IEEE,
Zesong Fei, Member, IEEE, Jingming Kuang
AbstractWe analyze the achievable rate of the uplink of a
single-cell multi-user distributed massive multiple-input-multiple-
output (MIMO) system. Each user is equipped with single an-
tenna and the base station (BS) is equipped with a large number
of distributed antennas. We derive an analytical expression for
the asymptotic ergodic achievable rate of the system under
zero-forcing (ZF) detector. In particular, we consider circular
antenna array, where the distributed BS antennas are located
evenly on a circle, and derive an analytical expression and
closed-form bounds for the achievable rate of an arbitrarily
located user. Subsequently, closed-form bounds on the average
achievable rate per user are obtained under the assumption
that the users are uniformly located. Based on the bounds, we
can understand the behavior of the system rate with respect
to different parameters and find the optimal location of the
circular BS antenna array that maximizes the average rate.
Numerical results are provided to assess our analytical results
and examine the impact of the number and the location of the
BS antennas, the transmit power, and the path-loss exponent
on system performance. Simulations on multi-cell networks are
also demonstrated. Our work shows that circularly distributed
massive MIMO system largely outperforms centralized massive
MIMO system.
Index TermsMassive MIMO, distributed MIMO, achievable
rate analysis, antenna location optimization.
I. I NTRODUCTION
With the demands of the wireless data services nowadays,
high spectrum efficiency or data rate is undoubtedly an im-
portant feature of future wireless systems [1], [2]. In order to
improve the data rate of wireless systems, various innovative
ideas have been proposed and investigated. The multiple-input-
multiple-output (MIMO) concept is one of the most successful
ones in recent years [3], [4].
Manuscript received August 5, 2014; revised December 28, 2014 and April
5, 2015; accepted May 26, 2015. This work was supported in part by the
Ericsson, the National Science and Engineering Research Council (NSERC)
of Canada, the National Natural Science Foundation of China under Grant No.
61371075, the National Natural Science Foundation of China under Grant No.
61421001, 863 project No.2014AA01A701, and 111 Project of China under
Grant B14010. The review of this paper was coordinated by Dr. Masoud
Ardakani.
A. Yang, C. Xing, Z. Fei, and J. Kuang are with the School of Information
and Electronics, Beijing Institute of Technology, Beijing, China. (Email:
taylorkingyang@163.com, chengwenxing@ieee.org, feizesong@bit.edu.cn,
JMKuang@bit.edu.cn).
Y. Jing is with the Department of Electrical and Computer Engineer-
ing, University of Alberta, Edmonton, AB, T6G 2V4 Canada (e-mail:
yindi@ualberta.ca).
*The corresponding author is Chengwen Xing.
1536-1276 (c) 2015 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.
1
Recently, distributed MIMO systems, or distributed antenna
systems, was proposed to further improve the data rate, in
which multiple transmit or receive antennas are distributively
located to reduce the physical transmission distance between
the transmitter and the receiver [5][15]. The distributive
antennas are assumed to be connected to the central unit
via high-bandwidth and low-delay backhaul such as optical
fiber channels. It has been proved that distributed MIMO
outperforms centralized MIMO in outage probability and
achievable rate. In [5], for a single-cell single-user distributed
MIMO system with arbitrary antenna topology, the authors
analyzed the outage performance as well as the diversity and
multiplexing gains. To maximize the cell averaged ergodic ca-
pacity, the authors in [9] proposed a squared distance criterion
for antenna location design and derived analytical expressions
for antenna locations under uniform user distribution. In [10],
different radio resource management schemes were compared
for multi-cell multi-user distributed antenna systems. For a
single-cell distributed MIMO systems with multiple uniformly
distributed users, upper bounds on the ergodic capacity of one
user and approximate expressions of the sum capacity of the
cell were derived in [11]. In [12], both single-cell and two-cell
distributed MIMO systems with uniformly distributed users
were considered. The cells are assumed to be circular and the
distributed base station (BS) antennas have circular layout.
The locations of the distributive antennas were optimized to
maximize lower bounds on the expected signal-to-noise ratio
(SNR) and signal-to-leakage ratio. The resource (including
power, subcarrier, and bit) allocation problems in single-cell
multi-user distributed antenna systems were investigated in
[14]. In [15], for multi-cell networks with multiple remote
antennas and one multi-antenna user in each cell, the input
covariance matrices for the users were jointly optimized to
maximize the achievable ergodic sum rate.
In the above literature of distributed MIMO, single user
or multiple users with orthogonal channels are assumed, so
there is no inter-user interference. However in current and
future wireless systems, it is expected to have multiple users
sharing the same time-frequency resource. In such systems, a
user will suffer from the interferences of other users in the
cell, which can largely degrade the system rate. To conquer
the interference problem, the concept of massive MIMO,
where the BS is equipped with a very large number of
antennas (usually of hundreds or higher) was proposed and
attracted increasing attention recently [2], [16][27]. With a
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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
large number of antennas at the BS, according to the law of
very long vectors, transmission channels for different users
are orthogonal to each other. User-interference diminishes and
very high data rate can be achieved with low complexity
signal processing. The ergodic achievable rates of the single-
cell multi-user massive MIMO uplink with linear detectors,
i.e., maximum ratio combining (MRC), zero-forcing (ZF),
minimum-mean-square-error (MMSE), have been derived in
[21]. The achievable rates of both the uplink and the downlink
of multi-cell multi-user massive MIMO systems with linear
precoders and detectors were analyzed in [22].
As the combination of the two promising concepts, massive
MIMO and distributed MIMO, distributed massive MIMO
systems are of great potential in fulfilling the increasing
demands of next generation communication systems [28][33].
In [28], for a single-cell network with circularly distributed
BS antennas and single-antenna users under an adaptive user
power control, the uplink capacity was analyzed when the
number of users and the number of BS antennas both approach
infinity with a fixed ratio. Capacity-maximizing antenna radius
was obtained analytically in closed-form when the path-loss
exponent is 4 and numerically for other path-loss exponent
values. The authors in [29] focused on an uplink massive
MIMO system consisting of multiple users and one BS with
several large-scale distributed antenna sets. The deterministic
equivalence of the ergodic sum rate was derived and an
iterative waterfilling algorithm was proposed for finding the
capacity-achieving input covariance matrices. [30] analyzed
the sum rates of distributed massive MIMO systems, in which
L multiple-antenna radio ports form a virtual transmitter and
jointly transmit data to a centralized multiple-antenna BS. The
capacity and spatial degrees of freedom of distributed massive
MIMO system were investigated in [31], where single-antenna
users cooperate in clusters to form distributed antenna arrays.
In [32], a simplistic matched filtering scheme and a subspace
projection filtering scheme were investigated in a fixed size
single-cell network, where the BS antennas are assumed uni-
formly and randomly located in the cell serving single-antenna
users. In [33], for networks where one large-scale distributed
BS with a grid antenna layout serving single-antenna users, the
energy efficiency maximization problem was formulated under
per-antenna transmit power and per-user rate constraints. Low
complexity channel-gain-based antenna selection method and
interference-based user clustering method were proposed to
improve the system energy efficiency.
In this paper, we consider the uplink of a single-cell multi-
user distributed massive MIMO system in which the BS,
equipped with a large number of distributed antennas, receives
information from multiple single-antenna users. We then
consider circular antenna layout, which has a good balance
among the tractability of theoretical analysis, practicality, and
performance. Circular layout has been investigated in [9], [12],
[28], [34][36] and is shown to be an appropriate candidate
for distributed antenna deployment. We analyze the asymptotic
achievable rate of the system under linear ZF detector for
arbitrary but known antenna locations and for circular antenna
layout. Based on the analysis, the location of the circularly
distributed BS antennas is optimized. Compared to [28], our
1536-1276 (c) 2015 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.
2
work differs in transceiver scheme, sum-rate calculation, and
assumptions. Thus our results and derivations are fundamen-
tally different. Compared to [29], [30], the topology of the
distributed BS antennas used in our work is different, as the
BS in [29] is the combination of several centralized BSs and
the BS in [30] has centralized antennas. Compared to [31], the
single-antenna users in this work are non-cooperative, while
the users in [31] cooperate with each other and form a virtual
transmitter with distributed antenna array. Thus our system
model and derivations of achievable rate are different from
those in [29][31]. Compared to [32], [33], we focus on the
derivations of achievable rate and the location optimization of
the distributed BS antennas, while [32] faces the problem of
interference control through the use of second-order channel
statistics and [33] works on the energy efficiency maximization
problem.
The major contributions of this paper are summarized as
follows.
We provide new analytical expression for the asymptotic
achievable rate of multi-user distributed massive MIMO
systems with arbitrary but known user location and an-
tenna deployment (see Proposition 1).
We consider a practical circular antenna layout, where
antennas of the BS are located evenly on a circle.
The asymptotic achievable rate for an arbitrarily located
user and two closed-form tight bounds are derived (see
Theorems 1 and 2).
Furthermore, for the circular antenna layout, tight closed-
form bounds on the average rate of the cell with uniform
user location are obtained (see Theorem 3). These results
can be used to predict the system performance and
understand its behavior with respect to the number of
BS antennas, the location of BS antennas, the cell size,
and user transmit power.
Based on the acquired tight bounds, we derive the opti-
mal radius of the distributed antennas for the maximum
average achievable rate (see Lemma 1), which guides the
fundamental and practical problem of antenna placement
for distributed MIMO systems.
Finally, numerical results are provided to assess our
analysis. It is shown that multi-user circularly distributed
massive MIMO is largely superior to centralized massive
MIMO in uplink achievable rate.
The remaining of the paper is organized as follows. The
distributed massive MIMO system model and asymptotic
achievable rate analysis for an arbitrary but known user
location and antenna deployment are presented in Section II.
Asymptotic achievable rate analysis of circularly distributed
massive MIMO is present in Section III. Location optimization
of circularly distributed BS antennas is provided in Section
IV. Simulation results are presented in Section V. Conclusions
are drawn in Section VI. Involved proofs are included in the
appendices.
Notation: Boldface lowercase letters denote vectors, while
T
boldface uppercase letters denote matrices. We use () , ()
H
and () to denote the transpose, conjugate, and conjugate
transpose of a matrix or a vector, respectively. For a matrix Z,
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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
3

tr(Z) is its trace. The symbol IM denotes the M M identity which is the M 1 channel vector between the kth user and
matrix, while 0M,N denotes the M N matrix whose entries all M BS antennas. Let
are zeros. The symbol E denotes the statistical expectation
operation. The symbol kkF denotes Frobenius norm of a G , g1 gK ,
matrix or a vector. The function log2 () is the base-2 logarithm which is the channel matrix between all K users and all M BS
and ln() is the natural logarithm. antennas. Perfect channel state information (CSI) is assumed
at the BS, that is, the BS knows G precisely.
II. S YSTEM MODEL AND A SYMPTOTIC ACHIEVABLE R ATE The uplink communication is studied, where the K users
A NALYSIS transmit their data in the same time-frequency resource to the
A. Multi-User Distributed Massive MIMO System Model BS. Let x be the K 1 signal vector containing the user
data, where its k-th entry xk is the information symbol of the
We consider a single-cell multi-user distributed massive kth user, normalized as E{|xi |2 } = 1. Symbol P denotes the
MIMO system. In this system, there is one BS equipped with average transmit power of each user, which implies that all
M antennas which are spatially distributed [32], [33]. The the users are assumed to have the same transmit power. The
number of antennas M is assumed to be large, e.g., a few M 1 vector of the received signals at the BS is
hundreds. This is different to the multi-user centralized MIMO
system [16][27], where the BS antennas are centralized y = P Gx + n, (3)
and spatially co-located. Compared with centralized MIMO
where n is the noise vector, whose entries are independent
systems, distributed MIMO systems provide macro-diversity
and identically distributed (i.i.d.) CSCG random variables with
and have enhanced network coverage and capacity, due to
zero-mean and unit-variance, that is, n CN (0M,1 , IM ).
their open and flexible infrastructure [6], [8], [29]. This is also
Assume that the transmit power of each user scales with M
different to systems with distributed BSs, in which several BSs
according to P = Eu /M , where Eu is a finite constant [21].
with multiple antennas cooperate with each other to jointly
Moreover, assume that Eu 1.
transmit or receive the information; and each BS has its own
ZF linear detector is used at the receiver, which has low-
power constraint [37], [38]. We assume that the distributed
complexity and good performance in massive MIMO systems
BS antennas are connected with high capacity backhaul (e.g.,
[21], [22]. ZF separates data streams from different users
optical fiber channels) and have ideal cooperation with each
other. There are K users, each equipped with single antenna. by Hmultiplying
1 H the received signal vector y with A ,
G G G [21]. From (3), we have
We assume that M K, which is necessary in achieving
1 H 1 H
high spectral efficiency in massive MIMO systems [16], [21]. r , Ay = GH G G y = P x + GH G G n.
Denote the channel coefficient between the mth antenna of
the BS and the kth user as gmk . We consider both the path-loss Focusing on the kth element of r, we have,
and small-scale fading as follows [33]:
p r k = P x k + aHk n, (4)
gmk = hmk mk , m = 1, 2, . . . , M, k = 1, 2, . . . , K, (1)
where ak is the kth column of A. Since n CN (0M,1 , IM ),
2
where hmk is the small-scale fading coefficient, which is it can be shown that aH
k n CN (0, kak kF ). The ZF detector
model as a random variable with zero-mean is chosen here for the simplicity of the presentation and our
and unit-variance.
Furthermore, we assume that E |hmk |8 < which is work can be straightforwardly extended to the MMSE linear
satisfied for most popular fading models. Without loss of detector.
generality, Rayleigh fading is adopted in the simulation, where
hmk follows circularly symmetric complex Gaussian (CSCG) B. Asymptotic Achievable Rate Analysis
distribution, i.e., hmk CN (0, 1). hmk s are assumed to be
In this subsection, we analyze the asymptotic achievable
mutually independent. mk models the path-loss, given as
v rate of the multi-user distributed massive MIMO uplink when
d0 M , for a general BS antenna deployment and user
mk = , (2)
dmk location. The number of users K is assumed to be an arbitrary
but finite number. Here we provide the following proposition.
where dmk is the distance between the mth antenna of the BS
and the kth user and d0 is the minimum distance between Proposition 1. The average achievable rate of the kth user
a user and a BS antenna, which exists due to physical in the single-cell multi-user distributed MIMO uplink has the
implementation [7]. Symbol v is the path-loss exponent with following asymptotic behavior:
typical values ranging from 2 to 6, i.e., 2 6. In a.s.
centralized MIMO, d1k = = dM k , thus 1k = = M k , Rk log2 1 + Eu k , as M , (5)
which indicates that the large-scale fading coefficients from a.s.
where denotes the almost sure convergence as M goes to
the kth user to all BS antennas are the same [16], [21]. We infinity and k is defined as
assume no channel shadowing but the derived results can be
M
straightforwardly extended to channels with shadowing. Let 1 X
T k , lim mk . (6)
M M
gk , g1k gM k , m=1

1536-1276 (c) 2015 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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
4

Proof: From (4), the ergodic achievable rate of the kth

user is
( !)
P Tm
Rk = E log2 1 + 2 User
kak k r
U Central processor of BS
1 1 O
ru
= E log2 1 + Eu h i . (7) Distributed antenna of BS
M (GH G)1
kk Backhaul
1/2
Notice that gk = Dk hk ,
where Dk , R
diag{1k , 2k , . . . , M k } denotes the propagation
1/2 1/2 1/2 1/2
matrix with Dk , diag{1k , 2k , . . . , M k }, and
T 1
hk , [h1k , h2k , . . . , hM k ] . The vector M hk has Fig. 1. Distributed massive MIMO system with circular BS antennas.
i.i.d. entries whose mean is zero and variance is 1/M . The
8
8th order moment of its kth entry equals E |h
mk | /M 4 ,
which is of order O(1/M 4 ) when E |hmk |8 < . And
which is the same as Eq. (13) in [21].
obviously entries of 1M hk are independent of Dk . Thus
we can use Lemma 4 and Lemma 5 in [40] to obtain the
following asymptotic results: III. A SYMPTOTIC A NALYSIS FOR THE ACHIEVABLE R ATE
OF C IRCULARLY D ISTRIBUTED A NTENNAS
M
1 2 1 X Theoretically speaking, BS antennas in a distributed massive
kgk k ik
M M i=1 MIMO system can take arbitrary locations and topology.
H The optimization of the antenna locations can be highly
1 1 1 a.s.
= hk Dk hk tr(Dk ) 0, (8) challenging, if not intractable, due to the large number of
M M M
H antennas and design parameters. On the other hand, arbitrary
1 H 1 1 a.s. antenna locations or optimal topology may have prohibitive
gk gi = hk Dk hi 0, i 6= k. (9)
M M M backhaul cost and installation cost. In real applications, it is
1
H more practical to consider manageable antenna topology.
Notice that K is finite and M G G is a K K matrix.
From (8) and (9), we have We consider circularly distributed BS antennas where all
( M M M
) antennas are on a circle centered at the cell enter. Circular
1 H 1 X 1 X 1 X antenna layout has been considered in [9], [12], [28], [34]
G G diag i1 , i2 , iK
M M i=1 M i=1 M i=1 [36] and shown to have good performance. It is chosen in this
a.s work due to its appropriate balance among the tractability of
0.
H theoretical analysis, practicality, and performance.
Since k > 0 and the matrix M 1
G G is non-singular, we First, circular antenna layout leads to tractable analysis, thus
have there is good potential of obtaining optimal and insightful
1 design solutions. With uniform angle spread, the circular struc-
1 H a.s
G G diag 1/1 , 1/2 , . . . , 1/K 0. ture is actually one dimensional, thus its design is equivalent to
M
1 1 the design of the radius only. For other designs such as the grid
Thus the (k, k)-th element of M (GH G) almost surely structure and arbitrary topology, the analysis can be prohibitive
converges to 1/ k , and the inverse of the (k, k)-th element of for insightful results. In addition, compared with grid layout,
1 1
H arbitrary topology, and poisson point process based layouts,
M (G G) almost surely converges to k , that is
circular layout can be more practical in the implementation
1 1 1 a.s
h i = h 1 i k . (10) and cost point of view. It needs less optical backhaul instal-
M (GH G)1 1 H
MG G lation, and the implementation of circular layout expects less
kk kk
confliction with existing power lines and other civil structures
By using (10) into (7), the proposition is proved.
underground and aboveground. Lastly, compared with other
simple and low-dimensional designs (e.g., centralized antennas
The result in Proposition 1 applies to massive MIMO
and line layout), circular layout is expected to have better
systems with arbitrary antenna deployment and user location.
performance. Compared with centralized antennas, it takes
It can also be proved to be applicable to the MMSE detector.
advantage of the macro-diversity via distributed antenna loca-
The multi-user centralized MIMO system considered in
tions; while compared with line layout, it has better symmetry
[16], [21][23] can be seen as a special case of our multi-
in a circular cell.
user distributed MIMO system with all the distributed antennas
In this section, for distributed massive MIMO systems with
located at the same place, i.e., mk = k . Thus, from (5), we
circular antenna layout, we first specify the system model,
can obtain the achievable rate result for centralized massive
then analyze the asymptotic achievable rate for an arbitrary
MIMO uplink as
user, and finally derive the average sum-rate per user assuming
a.s.
Rk,central log2 (1 + Eu k ) , as M , (11) random uniform user location.

1536-1276 (c) 2015 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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
5

The single-cell multi-user distributed massive MIMO sys- MIMO system with respect to the cell size and antenna
tem is shown in Figure 1. We consider a circular cell with location. Thus, in what follows, we derive lower and upper
radius R. Circular cell is widely used [12], [34], [39] and has bounds on the achievable rate in closed-form.
been shown to have similar performance to hexagonal cell.
But it enjoys more tractable analysis. The center of the cell is Theorem 2. Define
denoted as O. The distributed BS antennas are located evenly v2 1 !
on a smaller circle with radius r, whose center is the same B,1 ru2 + r2
R (ru ) , log2 1+ Eu dv0 v1 , (15)
as the center of the cell. We denote the location of the mth |ru2 r2 |
BS antenna as Tm . Thus, the length of the segment OTm is RB,2 (ru )
r. Notice that since the BS antennas are evenly located on the v2 1 !
23( 2 1) 2 v2 12
v
circle and the antenna number is assumed to be large, the angle ru2 + r2
, log2 1+ Eu dv0 v1 .
dimension of the antenna location (for example, the angle of |ru2 r2 | (v 1)
the segment OTm and the horizontal axis) has little effect on (16)
the system performance and only the radius of the antenna
circle matters. We denote the location of an arbitrary user as For the circularly distributed massive MIMO system with
U . Let ru be the distance between the user and the cell center, antenna radius r, the asymptotic achievable rate of a user
i.e., the length of the segment OU . at distance ru (where |r ru | d0 ) from the cell center,
denoted as Rasy (ru ), can be bounded as follows
A. Asymptotic Achievable Rate of an Arbitrary User for Cir- B,1
R (ru ) Rasy (ru ) RB,2 (ru ), if 2 4,
cularly Distributed Massive MIMO
RB,1 (ru ) Rasy (ru ) RB,2 (ru ), if 6 4,
The following theorem on the asymptotic achievable rate of
Rasy (ru ) = RB,1 (ru ) = RB,2 (ru ), if = 2 or 4.
an arbitrary user at distance ru for the circularly distributed (17)
massive MIMO system is obtained.
Proof: See Appendix B.
Theorem 1. For the circularly distributed massive MIMO Theorem 2 provides both lower and upper bounds on the
system with antenna radius r, the ergodic achievable rate of achievable rate. When v 4, RB,1 (ru ) is a lower bound
the user at distance ru (where |r ru | d0 ) from the cell and RB,2 (ru ) is an upper bound; when v 4, RB,1 (ru )
center has the following asymptotic behavior: is an upper bound and RB,2 (ru ) is a lower bound. When
a.s.
R(ru ) Rasy (ru ) , log2 (1 + Eu I0 ) , as M , (12) = 2, 4, RB,1 (ru ) and RB,2 (ru ) are the same and equal
the achievable rate of the user. Moreover, it is evident that
where the bounds in Theorem 2 are in closed-form. We note that the
2 ( 2 ) 2 ( v2 21 )
3 v 1
v r2 + ru2 coefficient in (16) only depends on , the
I0 (ru ) , dv0 r2 ru2 2 P v2 1 , (13) (v1)
|r2 ru2 | path-loss exponent and can be easily calculated offline.
with P () the Legendre function [42]. To justify the tightness of the two closed-form bounds, we
For = 2, 4, 6, closed-form expressions for the achievable analyze their difference as follows:
rate can be obtained as follows: B,1
R (ru ) RB,2 (ru )
R(ru )
v
2 2 1
v (ru +r )
2
a.s. 1 + Eu d0 |r2 r2 |v1
Rasy (ru )
= log2 u

3( 1)
v

log 1 + E d v 1
, if v = 2, (r
v
2 +r 2 ) 2 1 2 2 2 v1
( )

2

u 0 |r 2 r 2 | 1 + Eu dv0 |ru2 r2 |v1 (v1)
2 2


2
u
2
u

log2 1 + Eu dv0 r2 +r2u 3 , if v = 4,
23( 2 1) 2 v2 12 (a)
v
|rru |
, 2 +r 2 2 (14) log2 < 0.6,

3
r u 1 (v 1)

1 + Eu dv0 r r
2 2
u , if v = 6.

log 2 2 2 3
2|r ru |
where Inequality (a) in the above derivation is obtained by
Proof: See Appendix A. numerical calculations for 2 v 6. This shows that the two
In this theorem, to facilitate the presentation, we introduce bounds are close to each other with less than 0.6 bits/s/Hz
a new notation, Rasy for the asymptotic ergodic achiev- difference. The difference is negligible for massive MIMO
able systems when Eu = P M 1. Thus, either RB,1 (ru ) or
rate of a user when M . Note that Pa (b) =
F a, a + 1; 1; 1b 2 , where F (, ; ; ) is the Gauss hyper- RB,2 (ru ) can function as a tight closed-form approximation
geometric function [42]. Many softwares for scientific com- of the achievable rate. Our simulation results in Section V also
putations such as Matlab have this function. Thus the result justify the tightness of the bounds.
in (12-13) of Theorem 1 can be easily calculated. However, Since both bounds RB,1 (ru ) and RB,2 (ru ) increase in
due to the special function, for 6= 2, 4, 6, the achievable log2 Eu and Eu = P M , where P is the transmit power and
rate is not in closed-form and little insight can be obtained on M is the number of BS antennas, the user achievable rate is
the performance behavior of the circularly distributed massive proved to increase in log2 P and log2 M .

1536-1276 (c) 2015 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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
6

B. Asymptotic Average Achievable Rate of the Cell for Circu- By using (15) in Theorem 2,
larly Distributed Massive MIMO B,1
Rtemp
In the previous subsection, we have analyzed the asymptotic Z v2 1 !
rate of an arbitrarily located user in the cell. In this subsection, 2 ru2 + r2
= 2 ru log2 1+ Eu dv0 v1 d(ru )
we derive the asymptotic average rate of a user in the cell, R 4rd0 |ru2 r2 |
which indicates the average experience of user service. The Z 2 v !
2 2 1
users are assumed to be randomly and uniformly located [12], 2 v ru + r
2 ru log2 Eu d 0 v1 d(ru ), (24)
[41]. Since the minimum distance between a user and a BS R 4rd0 |ru2 r2 |
antenna is d0 , the users can only locate inside the circle with
where the approximation follows log(1 + x) log x for
radius r d0 or outside the circle with radius r + d0 , The area
2 2 x 1. We can use this approximation since Eu 1. From
of which is (r d0 ) + R2 (r + d0 ) = R2 4rd0 .
the definition of and by using change of variable for the
Recall that ru denotes the users distance to the cell center.
integration,
Define , [0, r d0 ] [r + d0 , R], which is the range for ru ,
i.e., ru . The probability density function of ru is [41] B,1
Rtemp
("Z Z #
2 (rd0 )2 R2
fru (x) = R2 4rd0 x x . (18) 1
0 otherwise 2 + log2 (Eu dv0 ) dt
R 4rd0 0 (r+d0 )2
" #
The angle of the users location (to the horizonal axis) is v Z (rd0 )2 Z R2
uniformly distributed on [0, 2). + 1 + log2 t + r2 dt
2 0 (r+d0 )2
The following theorem on the asymptotic average achiev- "Z # )
(rd0 ) 2 Z R2
able rate per user is proved.
(v 1) + log2 t r dt .
2

Theorem 3. Define 0 (r+d0 )2

(25)
RB,1 , log2 Eu + vlog2 d0 R
v r2

With the aid of ln x = x ln x x, we have
+ 1 1 + 2 log2 R2 + r2 v
2 R B,1 1
Rtemp log2 (Eu dv0 ) + 1 2 4rd
r2 2 R 0
(v 1) 1 2 log2 R2 r2 n
R 2 2 2 2
R + r log2 (R + r )
r2 v h i h i
(3v 4) 2 log2 r + log2 e, (19) 2 2
+ r2 + (r d0 ) log2 r2 + (r d0 )
R 2 h i h i
2 v2 21 v 2 2
R B,2
, R B,1
+ log2 +3 1 . (20) r2 + (r + d0 ) log2 r2 + (r + d0 )
(v 1) 2 o
2r2 log2 r R2 4rd0
When d0 r < R and Eu 1, for the circularly
distributed massive MIMO system with uniformly distributed v 1 h 2
2 R r2 log2 (R2 r2 )
users, the asymptotic average achievable rate per user of the R 4rd0

cell, denoted as Rasy , can be bounded as follows : 2rd0 d20 log2 2rd0 d20
B,1
R & Rasy & RB,2 , if 2 4, 2rd0 + d20 log2 2rd0 + d20
R B,1
. Rasy . RB,2 , if 6 4, (21) i
+ 2r2 ln r R2 4rd0 . (26)
Rasy RB,1 RB,2 , if = 2 or 4.
By using the condition d0 r < R and ignoring the lower
where . represents approximately smaller and & presents
order terms of d0 in (26), we can obtain with straightforward
approximately larger. B,1
calculations Rtemp RB,1 , which is defined in (19). Simi-
Proof: With uniformly distributed user location and the B,2
larly, we can show that Rtemp RB,2 . From (17) in Theorem
probability density function of the users distance to the cell 2 and the definitions in (23), the results in (21) are proved.
center in (18), the asymptotic average achievable rate per user The approximations (as in . and &) in (21) are based on
of the cell can be calculated as: Eu 1 (high SNR approximation) and d0 r, and are
Z
2 obtained by ignoring lower order terms of Eu and r/d0 in the
Rasy = 2 ru Rasy (ru )d(ru ), (22)
R 4rd0 achievable rate formulas.
It is evident that our derived bounds on the asymptotic
where Rasy (ru ) is the asymptotic achievable rate for a user
average achievable rate in Theorem 3 are in closed-form. Also,
at distance ru . Define
Z calculating the difference between the two bounds, we have
B,1 2
Rtemp , 2 ru RB,1 (ru )d(ru ), B,1 2 v2 21 v

R 4rd0 R R B,2
= log2 +3 1 < 0.6.
Z
B,2 2 (v 1) 2
Rtemp , 2 ru RB,2 (ru )d(ru ). (23)
R 4rd0 (27)

1536-1276 (c) 2015 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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
7

Thus, either bound can be used as a tight approximation of circular BS antenna array can be designed for the maximum
Rasy with the error being less than 0.6 bits/s/Hz. The error is average achievable rate for the distributed massive MIMO
negligible when Eu = P M 1. The tightness of the bounds system.
will also be justified by our simulation results in Section V. From (28) we have ropt /R = (t0 + 1)1/2 . Thus, the ratio
of the antenna radius and the cell radius depends on the path-
IV. L OCATION O PTIMIZATION OF THE C IRCULAR BS loss exponent v only and is independent of the transmit power
A NTENNA A RRAY P (or Eu ) and the BS antennas size M . This is very appealing
In the previous section, the ergodic achievable rate of an in wireless network designs and implementation. For example,
arbitrarily located user and the average achievable rate per the improvement of the hardware in distributed MIMO, such
user of the cell for uniformly located users are derived when as increasing the number of the distributed antennas, will not
M . We can see from the results that the radius of affect the optimal location of the distributive antennas. For a
the distributed BS antennas r largely affects the achievable given v value, the radius of the circular BS antennas array
rate. In this section, we turn to derive the optimal radius of should increases linearly in the cell radius R.
the circularly distributed BS antenna array to maximize the Moreover, for any v [2, 6], we can show that ropt /R
average achievable rate of the cell, which is one of the most [0.7, 0.78] via straightforward numerical calculations. On one
important measures of wireless system performance. hand, the optimal ropt /R value is far away from the centralized
In Theorem 3, both the upper and lower bounds, RB,1 and massive MIMO case, where r/R = 0. On the other hand,
B,2
R , are derived for the average achievable rate per user. The for different v values within the practical range (v [2, 6]),
bounds are in closed-form and shown to be close to each other. ropt /R has small change. Actually, for any v [2, 6], our
Thus, in the radius optimization, we aim at maximizing RB,1 . simulation shows that setting the radius of the circular antenna
The same result can be obtained if RB,2 is used since the array as r = 0.75R induces less than 5% loss in the average
difference RB,2 RB,1 only depends on v and is independent achievable rate compared to the optimal radius. This result is
of r, the radius of the circular antenna array. useful in further simplifying the practical design of circularly
distributed massive MIMO systems. Also, it can be shown that
Lemma 1. The radius of the circular BS antenna array for ropt /R is an increasing function of v. Thus, as the path-loss is
the distributed massive MIMO system that maximizes RB,1 is: more severe, antennas should be installed further away from
s the cell center for the maximum average rate.
R2
ropt = , (28)
t0 + 1
V. N UMERICAL RESULTS
where t0 is the solution of the following equation: In this section, we present numerical results to show the
x
2
3+ v2
+ 2x
2
2+ v2
1 = 0. (29) performance of the distributed massive MIMO system with
circular BS antenna array and justify the accuracy of our
B,1
Proof: The derivative of R defined in (19) with respect theoretical results. The impacts of different parameters, such
to r can be calculated to be: as the number and the location of the distributed BS antennas,
dRB1 the user transmit power, and the path-loss exponent, on the
dr achievable rate are also investigated. We also simulate multi-
2 2
rlog2 e R R cell systems to understand the effects of inter-cell interference
= (v 2) ln + 1 +(2v 2)ln 1 . and pilot contamination. In the simulations, the cell radius is
R2 r2 r2
(30) set as R = 1000 meters. For each cell, there is a massive BS
with M circularly distributed antennas located on a circle of
By making dRB1 /dr zero, we have radius r and K users. The users have the same transmit power,
v
2 v/21 2 which is set to be rmid Eu /M where rmid = R/2 = 500
R v1
R meters. So, if a user is located 500 meters away from a BS
+ 1 1 = 1. (31)
r2 r2 antenna, the average received SNR of the antenna from the
v
After replacing R2 /r2 1 with t and rearranging the expres- user is P = Eu /M . The normalization with rmid in the
sion in (31), we obtain (29). transmit power does not affect the behavior of the simulation
Next we show that the solution of (31), denoted as curves but only affects the position of the curves on the Eu -
ropt , is the maximum of RB1 . From (30), we can see axis. The minimum distance between a BS antenna and a user
1 B is d0 = 1. The small-scale channel fading hmk is generated
that rlog 2e dR 1
is a decreasing function of r. Thus

R2
1 B
dr
1 B as CSCG with zero-mean and unit-variance, thus Rayleigh
rlog2 e dR 1
> 0 when r < r and rlog2 e dR 1 fading.
R 2 dr opt R 2 dr <
0 when r > ropt . Notice that rlog R2
2e
> 0. Thus we have
dR /dr > 0 when r < ropt and dRB1 /dr < 0 when
B1 A. Achievable Rate of an Arbitrarily Located User
r > ropt . This ends the proof. We first study the simulated achievable rate of an arbitrarily
The equation in (29) only depends on the path-loss exponent located user and compare with the derived asymptotic ana-
v and can be easily solved offline by software such as Matlab. lytical result in (12-13), as well as the closed-form bounds
By using the result in Lemma 1, the radius of the distributed RB,1 (ru ) in (15) and RB,2 (ru ) in (16). We set r = 500

1536-1276 (c) 2015 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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
8

18
24 Simulation Simulation, distributed MIMO

Average ergodic achievable rate of the cell

Analytical result in (12-13) Simulation, centralized MIMO
16
The ergodic achievable rate of one user

22 v=5.5 Bound RB,1(ru ) in (15) Bound RB,1 in (19)

20 Bound R B,2
(ru ) in (16) 14 Bound RB,2 in (20)
v=4.5
18
12
16 Eu = 30dB
10
14

12 8
Eu = 22dB
v= 3.5
10
v=2.5 6
8 2 3
0 200 400 600 800 1000 10 10
The distance between the user and the cell center ru Number of BS antennas (M)

Fig. 3. The average ergodic achievable rate of the cell for different M where
Fig. 2. Comparison of the analytical expressions and bounds of the ergodic K = 9, and v = 3.6. For circularly distributed MIMO, r = 760 meters; and
achievable rate of circularly distributed massive MIMO with simulation, where for centralized MIMO, r = 0.
M = 150, K = 9, r = 500 meters, and Eu = 31.8dB.

14

meters, M = 150 and Eu = 31.8dB. With this setting, we Average ergodic achievable rate of the cell
have P = 10dB. Figure 2 shows the achievable rate for 12

different user location ru (the distance of the user to the cell

center) and path-loss exponent v. 10

We can see from the figure that the derived analytical ex-
pression in (12-13) accurately predict the simulated achievable 8

rate. The derived closed-form bounds in (15) and (16) are very
close to the simulated and numerically calculated achievable 6

rates. The figure also shows that RB,1 (ru ) is a lower bound
Simulation, distributed MIMO
and RB,2 (ru ) is an upper bound when v 4, while RB,1 (ru ) 4
Simulation, centralized MIMO
is an upper bound and RB,2 (ru ) is a lower bound when v 4,
which confirm our results in (17). Furthermore, Figure 2 shows 2 Bound RB,1 in (19)
that the achievable rate is higher for smaller distance between Bound RB,2 in (20)
the user distance ru and the radius of the circular BS antenna 0
10 15 20 25 30
array r. For either ru > r or ru < r, the achievable rate is a Eu (dB)
concave function of ru .
Fig. 4. The average ergodic achievable rate of the cell for different Eu
where M = 100, K = 9, and v = 3.6. For circularly distributed MIMO,
B. Average Achievable Rate of the Cell r = 760 meters; and for centralized MIMO, r = 0.
In Figure 3 and Figure 4, we show the average achievable
rate per user of the cell, and compare with the derived
bounds RB,1 in (19) and RB,2 in (20), as well as the rate value. For example, increasing Eu from 14dB to 18dB brings
of centralized massive MIMO. We set r = 760 meters for the an achievable rate advantage of about 17% at M = 100.
circularly distributed MIMO and r = 0 for centralized massive Moreover, the figures show that circularly distributed MIMO
MIMO. We assume a practical urban scenario with the path- system largely outperforms the centralized one. For example,
loss exponent v = 3.6 [43]. The user locations are randomly at M = 100 and Eu = 30dB, it achieves about 39% higher
generated to be uniformly distributed in the cell. rate.
It can be seen from both figures that the simulated achiev-
able rates and the derived closed-form bounds have tight C. Impact of the Radius of the Circular BS Antenna Array
match, especially for large values of M and Eu . When M in- Next, we show the impact of the radius of the circular BS
creases, the average achievable rate of the cell keeps a constant antenna array, r, on the average achievable rate of the cell.
value for a fixed Eu , since the user transmit power P decreases Notice that the r = 0 case corresponds to the centralized
with M for a fixed Eu . This shows that in the circularly massive MIMO system, where the BS antennas are located
distributed massive MIMO system, in achieving a given rate at the center of the cell.
level, the required user transmission power decreases with M . Figure 5 plots the simulated average rate of the cell and
The average achievable rate increases with Eu for a fixed M the derived bounds with respect to different r for three cases:

1536-1276 (c) 2015 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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
9

15 12
14
Average ergodic achievable rate of the cell

Average ergodic achievable rate of Cell 1

M =150, Eu =31.8dB
13 10

12
11 8

10
6
9 M =100, Eu =30dB
8
4
7 Simulation
M =100, Eu =25dB Singlecell, perfect CSI
6 Bound RB,1 in (19)
2 Multicell, perfect CSI
5 Bound RB,2 in (20) Multicell, pilot contamination
4
0 200 400 600 800 1000 0
Radius of the distributed antennas (r) 0 200 400 600 800 1000
Radius of the distributed antennas (r)
Fig. 5. The average ergodic achievable rate of the cell for different r, where
K = 9 and v = 3.6. Fig. 7. The average achievable rates of single- and 7-cell systems for different
r, where Eu = 26.8dB, M = 150, K = 9, and v = 3.6.

13
Optimal r in this paper
Optimal r in [9] for all Eu values. The average rate gain brought by the optimal
Average ergodic achievable rate of the cell

12 Optimal r in [12] radius in this paper is about 5% for Eu = 14dB and v = 5.4.

11 D. Simulation on Circular BS Antenna Array in Multi-Cell

v=5.4
10
9 contamination, in this subsection, we consider a 7-cell network
8
7
10
Fig. 6. The average ergodic achievable rate of the cell for different Eu ,
where M = 100, K = 9 and v = 3.6, 5.4.
1) M = 100, Eu = 25dB, 2) M = 100, Eu = 30dB, and 3)
M = 150, Eu = 31.8dB. We set v = 3.6. We can see that
the radius of the distributed BS antenna array has significant
influence on the average rate of the cell and proper antenna
location results in significant improvement in the average rate
to the centralized case. For example, increasing r from 0 to
750 meters boosts up the average rate by about 30% when
M = 150, Eu = 31.8dB. The figure also verifies that the
optimal r for different M and Eu remains the same, which
is about 750 meters. This conforms with our result in Lemma
1 that the optimal r is irrelevant to the values of M and Eu ,
but only depends on v.
In addition, in Figure 6, we compare the average achievable
rate of circularly distributed massive MIMO with the optimal
antenna radius derived in Lemma 1 of this paper and those
derived in [9], [12]. We set M = 100, K = 9, and v =
3.6, 5.4. We can see that our solution achieves the highest rate
Systems
v=3.6
In reality, cellular systems have multiple cells. To un-
derstand the effects of inter-cell user-interference and pilot-
where (without loss of generality) Cell 1 is the cell of interest
and the other 6 cells embrace around Cell 1. The cell radius
is R = 1000 meters. The distance between the center of
Cell 1 to another cell center is 2R. Each BS has M = 150
15 20 25 30
circularly distributed antennas with radius r, and each cell
Eu (dB) has K = 9 users. The path-loss exponent is set as v = 3.6.
The channel model is the same as the single-cell network. For
the simulation of the pilot-contamination effect [16], the CSI
training phase needs to be considered. We set the power each
user spends for training to be the same as the power it spends
for data-transmission.
The simulated average ergodic achievable rate per user
for Cell 1 is shown in Figures 7 and 8, where Figure 7
is for different r values and Eu = 26.8 dB (so the user
transmit power is 5dB) and Figure 8 is for different M values
and r = 750 meters for the circularly distributed massive
MIMO system. We compare three cases: 1) single-cell network
with perfect CSI; 2) multi-cell network with perfect CSI
(thus only the effect of multi-cell interference is considered);
and 3) multi-cell network with pilot-contamination (thus both
effects of inter-cell interference and pilot contamination during
training are considered). In Figure 8, we further draw the
performance of centralized massive MIMO (where r = 0)
with multi-cell and pilot contamination for comparison. Note
that in Figure 7, the performance of the centralized massive
MIMO is shown by the rate-values at r = 0.
We can see from the figures that inter-cell interference
degrades the system performance (especially when r is close to

1536-1276 (c) 2015 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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
10

tightness of the bounds were rigorously justified. From these

Distributed MIMO, singlecell, perfect CSI results, behavior of the system achievable rate with respect
16 Distributed MIMO, multicell, perfect CSI to different parameters such as the size of the BS antenna
Distributed MIMO, multicell, pilot contamination array, the location of the BS antenna array, the path-loss
Average achievable rate of Cell 1

14
Centralized MIMO, multicell, pilot contamination exponent, and the transmit power can be understood. We also
12 derived tight closed-form bounds on the average achievable
10
rate per user assuming that users are randomly located in
Eu = 30dB the cell, from which the optimal radius of the distributed BS
8 antenna array that maximizes the average rate was derived.
Numerical results were illustrated to justify our analytical
6
results. Our work has shown that circularly distributed massive
4 MIMO largely outperforms centralized massive MIMO. Our
derived results can assist infrastructure providers in solving
2 Eu = 10dB
the fundamental problems of performance measurement and
0 1 antennas placement for distributed massive MIMO systems in
10 10
2
10
3
practice.
Number of BS antennas (M)
A PPENDIX A
Fig. 8. The average achievable rates of single- and 7-cell systems for different P ROOF OF T HEOREM 3
M , where K = 9, r = 750 and v = 3.6.
As shown in Figure 1, we denote the angle of the segments
OTm (where Tm is the location of the mth BS antenna and
R, the cell radius) but the performance degradation diminishes O is the cell center) and OU (where U is the location of the
as the number of BS antennas M increases (as is shown in user) as m . The distance between TM and U , denoted as
Figure 8). Pilot contamination brings further rate reduction Dm , can be expressed as
especially when r is small. Compared with centralized massive q
2
MIMO (where r = 0) with multi-cell and pilot contamination, Dm = r2 sin2 m + (r cos m ru ) . (32)
circularly distributed massive MIMO has significantly better Without loss of generality, we assume that 1 = 0 and the
performance when the radius is properly designed (e.g., r evenly circularly distributed BS antennas are labeled
m1such that
750 meters). For example, when M = 100 and Eu = 30dB, m = m1 2 for 1 m bM/2c and =
M m M 1 2
it achieves about 76% higher rate. Therefore, our conclusion for dM/2e m M . Let , 2/M .
that circularly distributed massive MIMO largely outperforms From Equations (2) and (32),
centralized massive MIMO is also valid in multi-cell networks, M
even though with the circular BS antenna layout, the BS 1 X
mk
antennas of adjacent cells are much closer to each other. This is M m=1
due to the large-scale of the BS antenna array, which mitigates M
user-interference. dv0 X 1
= v
In addition, from Figure 7, we can see that the achievable M m=1 Dm
rates for all three cases (for single-cell network and 7-cell
dv0 X h 2 2 i v2
M
2
network) have similar behaviour with respect to the antenna = r sin m + (r cos m ru )
radius r. The optimal radii for the three cases are 0.75R, M m=1
0.7R, and 0.7R, respectively. If we use the optimal radius dv0 1 X h 2 2 i v2
M
2
for the single-cell network, which is 0.75R, in the 7-cell = r sin m +(r cos m ru ) (33)
M m=1
network, the induced rate reduction due to the ignorance of
v Z h i v2
inter-cell interference and pilot contamination is less than 2% M d0 2
r2 sin2 + (r cos ru ) d. (34)
(by comparing the rates at r = 750 and r = 700). Thus, our 2
results for single-cell systems can still be valuable for multi- Employing [44, 2.5.16.38], we have, from (34),
cell systems. 2
M
1 X M v 2 v
2 2 v r + ru2
mk d0 r ru P 2 1 .
VI. C ONCLUSIONS M m=1 |r2 ru2 |
In this paper, we have considered the uplink of a single-cell (35)
multi-user distributed massive MIMO system, where the BS By using (35) in (5), Equation (12) can be proved. Equa-
equipped with a large number of distributed antennas receives tion (14) can be subsequently obtained by using [42, 8.912].
information from multiple users each equipped with single To further illuminate the asymptotic result in (34), we
antenna. The ZF detection is used at the BS. For circularly dis- investigate the difference between (34) and (33). To help the
tributed BS antennas, we derived analytical expressions of the presentation, we use the following notation:
achievable rate of an arbitrarily located user and two closed- h i v2
2
form expressions that bound the rate from both sides. The f () , r2 sin2 + (r cos ru ) .

1536-1276 (c) 2015 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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
11

For [0, ], we can obtain via straightforward calculations where

that ! v2 1
h i v2 1 z2 1
f 2 g (v, z, ) = 1 + cos
= v r2 sin2 + (r cos ru ) rru sin 0. z
! v2 1
(36) z2 1
+ 1 cos .
This shows that f decreases with when [0, ]. For the z
simplicity of presentation, we assume that M is even. The
Next, we look for bounds for g (v, z, ). We derive the
proof for odd M is similar. The difference between (34) and
derivative of g (v, z, ) with respect to z as follows:
(33), denoted by Idif f , can be bounded as follows:
! v2 2
v
1 z2 1
1 Xh 2 2 i v2
M
2 g (v, z, ) = 2 1+ cos
Idif f , r sin m + (r cos m ru ) z z2 z2 1 z
M m=1
Z ! v2 2
1 z2 1
f () d 1 cos cos . (39)
2 z
M
1 X 2 h i v2
2
=2 r2 sin2 m +(r cos m ru ) Since z 1, we have z 2 1/z [0, 1]. Thus for [0, 2 ],
M m=1
Z
1 z g (v, z, ) 0 when v 4 .
f () d (37)
0 z g (v, z, ) 0 when v 4
M
M
Z M
m
We can subsequently bound g (v, z, ) as follows:
1 X 2
m1 2 1 X 2 2
= fk 2 f () d
m=1 M M m=1 m1 M 2
g(v, , ) g (v, z, ) g(v, 1, ) when 2 v 4
.
g(v, 1, ) g (v, z, ) g(v, , ) when 6 v 4
m 2
M
1 X2
m1 (40)
f 2 f 2
m=1 M M M Define
v Z 2
2 2 z 2 1
= [f (0) f ()] = [(r ru )v (r + ru )v ] B1 , g(v, 1, )d
M M 0
M v Z 2 h

i
0. z 2 1 v
1 v
1
= (1 + 0) 2 + (1 0) 2 d
0
In obtaining (37), we use the symmetry in f (). This analysis v
= z 2 1 ,
shows that the difference between (34) and (33) scales as v Z 2
1/M . For large but finite number of antennas, (34) is a tight z 2 1
B2 , g(v, , )d
approximation of (33). 0
v Z 2 h i
z 2 1 v
1 v
1
= (1 + cos ) 2 + (1 cos ) 2 d
0
A PPENDIX B v Z
P ROOF OF T HEOREM 4 z 2 1 v
1
= (1 + cos ) 2 d
0
r 2 +r 2 Z
Define z , |r2 ru2 | . Notice that z 1 always. Using [42, z 2 1 v 1 v2
v
u
8.882.1], we have = 2 2 cos d
0 2
2 v Z
r + ru2 z 2 1 2 v
P v2 1 = P v2 1 (z) = 2 2 cosv2 (t) dt
|r2 ru2 | 0
Z p v2 1
23( 2 1) 2 v2 12 v 1
v
1
= z + z 2 1 cos d = z2 . (41)
0 (v 1)
v Z ! v2 1
z 2 1 z2 1 The last step is obtain by using [42, 3.621.1] and [42, 8.384.1].
= 1+ cos d
0 z From (38) and (40), P v2 1 (z) can be bounded as:
v Z 2 ! v2 1
z 2 1 z2 1 B1 P v2 1 (z) B2 when 2 v 4
= 1+ cos d . (42)
0 z B1 P v2 1 (z) B2 when 6 v 4
v Z ! v2 1 By applying (42) in (13) and (12), the first two lines of (17)
z 2 1 z2 1
+ 1+ cos d can be obtained. For the special cases of = 2, 4,with the

2
z aid of [42, 8.338.2], (1/2) = , and (3/2) = /2, we
v Z 2 3
2 2 v3 2 ( v2 12 )
z 2 1 have = 1. The two bounds are equal. Thus the
= g (v, z, ) d, (38) (v1)
0 last line of (17) is proved.

1536-1276 (c) 2015 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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
R EFERENCES
[1] A. Osseiran, F. Boccardi, V. Braun, K. Kusume, P. Marsch, M. Maternia,
O. Queseth, M. Schellmann, H. Schotten, H. Taoka, H. Tullberg,
M. A. Uusitalo, B. Timus, and M. Fallgren, Scenarios for 5G mobile
and wireless communications: The vision of the METIS project, IEEE
Commun. Mag., vol. 52, no. 5, pp. 26-35, May 2014.
[2] V. Jungnickel, K. Manolakis, W. Zirwas, B. Panzner, V. Braun, M. Los-
sow, M. Sternad, R. Apelfrojd, and T. Svensson, The role of small
cells, coordinated multipoint, and massive MIMO in 5G, IEEE Com-
mun. Mag., vol. 52, no. 5, pp. 44-51, May 2014.
[3] E. Telatar, Capacity of multi-antenna Gaussian channels, EURASIP
Trans. Telecommun., vol. 10, no. 6, pp. 585-596, Nov. 1999.
[4] A. Sibille, C. Oestges, and A. Zanella. MIMO: from Theory to Imple-
mentation. Academic Press, 2010.
[5] W. Roh and A. Paulraj, Outage performance of the distributed antenna
systems in a composite fading channel, in Proc. IEEE Veh. Tech-
nol. Conf. (VTC) Fall, vol. 3, Sept. 2002.
[6] W. Roh, High performance distributed antenna cellular networks,
Ph.D. Thesis, Stanford Univ., 2003.
[7] H. Liu, Y. Zhang and J. Luo, Distributed Antenna Systems: Open Ar-
chitecture for Future Wireless Communications. Auerbach Publication,
2007.
[8] H. Zhu, S. Karachontzitis, and D. Toumpakaris, Low complexity
resource allocation and its application to distributed antenna systems,
IEEE Wireless Commun. Mag., vol. 17, no. 3, pp. 44-50, Jun. 2010.
[9] X. Wang, P. Zhu, and M. Chen, Antenna location design for generalized
distributed antenna systems, IEEE Commun. Lett., vol. 13, pp. 315-317,
May 2009.
[10] T. Wu and P. Hosein, Radio resource management strategies for dis-
tributed antenna systems, in Proc. IEEE Wireless Commun. Networking
Conf. (WCNC), Apr. 2010.
[11] L. Dai, A comparative study on uplink sum capacity with co-Located
and distributed antennas, IEEE J. Sel. Areas Commun., vol. 29, no. 6,
pp. 1200-1213, Jun. 2011.
[12] E. Park, S.-R. Lee, and I. Lee, Antenna placement optimization for
distributed antenna systems, IEEE Trans. Wireless Commun., vol. 11,
no. 7, pp. 2468-2477, Jul. 2012.
[13] J. Wang, H. Zhu, and N. J. Gomes, Distributed antenna systems
for mobile communications in high speed trains, IEEE J. Sel. Areas
Commun., vol. 30, no. 4, pp. 675-683, May 2012.
[14] H. Zhu and J. Wang, Radio resource allocation in multiuser distributed
antenna systems, IEEE J. Sel. Areas Commun., vol. 31, no. 10,
pp. 2058-2066, Oct. 2013.
[15] W. Feng, Y. Wang, N. Ge, J. Lu, and J. Zhang, Virtual MIMO in multi-
cell distributed antenna systems: Coordinated transmissions with large-
scale CSIT, IEEE J. Sel. Areas Commun., vol. 31, no. 10, pp. 2067-
2081, Oct. 2013.
[16] T. L. Marzetta, Noncooperative cellular wireless with unlimited num-
bers of base station antennas, IEEE Trans. Wireless Commun., vol. 9,
no. 11, pp. 3590-3600, Nov. 2010.
[17] C. Artigue, P. Loubaton, On the precoder design of flat fading MIMO
systems equipped with MMSE receivers: A large system approach,
IEEE Trans. Inform. Theory, vol. 57, no. 7, pp. 4138-4155, Jul. 2011.
[18] A. Pitarokoilis, S. K. Mohammed, and E. G. Larsson, On the optimality
of single-carrier transmission in large-scale antenna systems, IEEE
Wireless Commun. Lett., vol. 1, no. 4, pp. 276-279, Aug. 2012.
[19] H. Huh, G. Caire, H. C. Papadopoulos, and S. A. Ramprashad, Achiev-
ing massive MIMO spectral efficiency with a not-so-large number of
antennas, IEEE Trans. Wireless Commun., vol. 11, no. 9, pp. 3226-
3239, Sep. 2012.
[20] F. Rusek, D. Persson, B. K. Lau, E. G. Larsson, T. L. Marzetta,
O. Edfors, and F. Tufvesson, Scaling up MIMO: Opportunities and
challenges with very large arrays, IEEE Signal Process. Mag., vol. 30,
no. 1, pp. 40-46, Jan. 2013.
[21] H. Q. Ngo, E. G. Larsson, and T. L. Marzetta, Energy and spectral effi-
ciency of very large multiuser MIMO systems, IEEE Trans. Commun.,
vol. 61, no. 4, pp. 1436-1449, Apr. 2013.
[22] J. Hoydis, S. ten Brink, and M. Debbah, Massive MIMO in UL/DL
cellular systems: How many antennas do we need? IEEE J. Sel. Areas
Commun., vol. 31, no. 2, pp. 160-171, Feb. 2013.
[23] H. Yang and T. L. Marzetta, Performance of conjugate and zero-
forcing beamforming in large-scale antenna systems, IEEE J. Sel. Areas
Commun., vol. 31, no. 2, pp. 172-179, Feb. 2013.
[24] F. Fernandes, A. Ashikhmin, and T. L. Marzetta, Inter-cell interference
in noncooperative TDD large scale antenna systems, IEEE J. Sel. Areas
Commun., vol. 31, no. 2, pp. 192-201, Feb. 2013.
1536-1276 (c) 2015 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.
12
[25] R. Aggarwal, C. E. Koksal, and P. Schniter, On the design of large
scale wireless systems, IEEE J. Sel. Areas Commun., vol. 31, no. 2,
pp. 215-225, Feb. 2013.
[26] H. Q. Ngo, E. G. Larsson, and T. L. Marzetta, The multicell mul-
tiuser MIMO uplink with very large antenna arrays and a finite-
dimensional channel, IEEE Trans. Commun., vol. 61, no. 6, pp. 2350-
2361, Jun. 2013.
[27] T. Datta, N. A. Kumar, A. Chockalingam, and B. S. Rajan, A novel
Monte Carlo sampling based receiver for large-scale uplink multiuser
MIMO systems, IEEE Trans. Veh. Technol., vol. 62, no. 7, pp. 3019-
3038, Sep. 2013.
[28] J. Gan, Y. Li, L. Xiao, S. Zhou, and J. Wang, On sum rate and
power consumption of multi-user distributed antenna system with cir-
cular antenna layout, EURASIP J. Wireless Commun. and Networking,
vol. 2007, no. 2, Jan. 2007.
[29] J. Zhang, C.-K. Wen, S. Jin, X. Gao, and K.-K. Wong, On capacity
of large-scale MIMO multiple access channels with distributed sets
of correlated antennas, IEEE J. Sel. Areas Commun., vol. 31, no. 2,
pp. 133-148, Feb. 2013.
[30] M. Matthaiou, C. Zhong, M. R. McKay, and T. Ratnarajah, Sum
rate analysis of ZF receivers in distributed MIMO systems, IEEE
J. Sel. Areas Commun., vol. 31, no. 2, pp. 180-191, Feb. 2013.
[31] A. Ozgur, O. Leveque, and D. Tse, Spatial degrees of freedom of
large distributed MIMO systems and wireless ad hoc networks, IEEE
J. Sel. Areas Commun., vol. 31, no. 2, pp. 202-214, Feb. 2013.
[32] H. Yin, D. Gesbert, L. Cottatellucci, Dealing with interference in
distributed large-scale MIMO systems: A statistical approach, to appear
on IEEE J. Sel. Topics Signal Process., 2014. [Online]. Available:
http://arxiv.org/abs/1310.6674.
[33] J. Joung, Y. K. Chia, and S. Sun, Energy-efficient, large-scale
distributed-antenna system (L-DAS) for multiple users, to appear
on IEEE J. Sel. Topics Signal Process., 2014. [Online]. Available:
http://arxiv.org/abs/1312.1870.
[34] W. Feng, X. Xu, S. Zhou, J. Wang, and M. Xia, Sum rate character-
ization of distributed antenna systems with circular antenna layout, in
Proc. IEEE Veh. Technol. Conf. (VTC) Spring, Apr. 2009.
[35] S. Firouzabadi and A. Goldsmith, Optimal placement of distributed
antennas in cellular systems, in Proc. IEEE Signal Process. Advances
in Wireless Commun. (SPAWC), Jun. 2011.
[36] K. Huang, J. Chen, and V. K. N. Lau, Communications using ubiquitous
antennas: Free-space propagation, submitted to IEEE International
Conf. Commun. (ICC), Jun. 2015.
[37] A. Tolli, M. Codreanu, and M. Juntti, Cooperative MIMO-OFDM
cellular system with soft handover between distributed base station
antennas, IEEE Trans. Wireless Commun., vol. 7, no. 4, pp. 1428-1440,
Apr. 2008.
[38] Y. Hadisusanto, L. Thiele, and V. Jungnickel, Distributed base sta-
tion cooperation via block-diagonalization and dual-decomposition, in
Proc. Globecom, 2008.
[39] L. Han, Y. Tang, S. Shao, and T. Wu, On the design of antenna location
for OSTBC with distributed transmit antennas in a circular cell, in
Proc. IEEE International Conf. Commun. (ICC), May 2010.
[40] S. Wagner, R. Couillet, M. Debbah, and D. T. M. Slock, Large system
analysis of linear precoding in correlated MISO broadcast channels
under limited feedback, IEEE Trans. Inform. Theory, vol. 58, no. 7,
pp. 4509-4537, Jul. 2012.
[41] W. Choi and J. Y. Kim, Forward-link capacity of a DS/CDMA system
with mixed multirate sources, IEEE Trans. Veh. Technol., vol. 50, no. 3,
pp. 737-749, May 2001.
[42] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
Products, 7th Ed. Academic Press, New York, 2007.
[43] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.
Cambridge University Press, 2005.
[44] A. P. Prudnikov, Y. A. Brychkov, and O. I. Marichev, Integrals and
Series, Volume 1: Elementary Functions. Gordon and Breach Science
Publishers, 1986.
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/TWC.2015.2441057, IEEE Transactions on Wireless Communications
13

Ang Yang received the B.S. degree in Electronic
Engineering from Beijing Institute of Technology
(BIT), Beijing, China in 2009. He is currently
working toward the Ph.D. degree with the Research
Institute of Communication Technology (RICT) of
BIT. Since October 2012, he has been with the
Electrical and Computer Engineering Department,
University of Alberta, Canada, where he is a visiting
Ph.D. student under the supervision of Prof. Yindi
Jing. His research interests are in the general area
of multivariate statistics, MIMO systems, network
coding, and cooperative communications.
Jingming Kuang received the B.S. and M.S. degree
in Information and Signal Processing at Beijing
Institute of Technology (BIT), Beijing, China, in
1966 and 1981, respectively, and Ph.D. degree in
Electrical Engineering from Technical University
of Berlin, Berlin, Germany, in 1988. He has been
a Professor with the School of Information and
Electronics, BIT since 1989. He is the founder of
BIT-Ericsson Research Center of Digital Communi-
cations, which was built in 1999. From Aug. 1993
to Aug. 2007, he has been the vice-president and
the president of BIT. His research interests include theory and techniques of
wireless communication and digital signal processing.

Yindi Jing received the B.Eng. and M.Eng. degrees

in automatic control from the University of Science
and Technology of China, Hefei, China, in 1996
and 1999, respectively. She received the M.Sc. de-
gree and the Ph.D. in electrical engineering from
California Institute of Technology, Pasadena, CA, in
2000 and 2004, respectively. From October 2004 to
August 2005, she was a postdoctoral scholar at the
Department of Electrical Engineering of California
Institute of Technology. Since February 2006 to
June 2008, she was a postdoctoral scholar at the
Department of Electrical Engineering and Computer Science of the University
of California, Irvine. In 2008, she joined the Electrical and Computer
Engineering Department of the University of Alberta, where she is currently
an Associate Professor.
She serves as an Associate Editor for the IEEE Transactions on Wireless
Communications and a member of the IEEE Signal Processing Society
Signal Processing for Communications and Networking (SPCOM) Technical
Committee. She also served as a track co-chair for Transmission Technologies
and Communication Theory, IEEE Vehicular Technology Conference (VTC)
2013-Fall. Her research interests are in wireless communications, focusing on
massive MIMO, cooperative relay networks, training and channel estimation,
distributed beamforming, and the analysis on network fundamental limits.

Chengwen Xing received the B.Eng. degree from

Xidian University, Xian, China, in 2005 and the
Ph.D. degree from the University of Hong Kong,
Hong Kong, in 2010. Since September 2010, he
has been with the School of Information and Elec-
tronics, Beijing Institute of Technology, Beijing,
China, where he is currently an Associate Professor.
From September 2012 to December 2012, he was
a visiting scholar at the University of Macau. His
current research interests include statistical signal
processing, convex optimization, multivariate statis-
tics, combinatorial optimization, massive MIMO systems and high frequency
band communication systems.
Dr. Xing is currently serving as an Associate Editor for IEEE Transactions
on Vehicular Technology, KSII Transactions on Internet and Information
Systems, Transactions on Emerging Telecommunications Technologies and
China Communications.

Zesong Fei received the Ph.D. degree in Electronic

Engineering in 2004 from Beijing Institute of Tech-
nology (BIT). He is now a Professor in BIT and cur-
rently with the Research Institute of Communication
Technology (RICT) of BIT, where he is involved in
the design of the next generation high-speed wire-
less communication. His research interests include
mobile communication, channel coding and modula-
tion, cognitive radio and cooperative networking. He
was the chief investigator of a China national Natural
Science Fund project. He is the senior member of
Chinese Institute of Electronics and China Institute of Communications.

1536-1276 (c) 2015 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.