See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/280576052

Sum-rate Maximizing in Downlink Massive MIMO Systems with Circuit Power

Consumption

Conference Paper · October 2015

DOI: 10.1109/WiMOB.2015.7347995

CITATIONS READS
0 90

2 authors:

Rami Hamdi Wessam Ajib

Hamad bin Khalifa University Université du Québec à Montréal
13 PUBLICATIONS 40 CITATIONS 172 PUBLICATIONS 1,095 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Energy-Aware Resource Allocation in Next Generation Wireless Networks: Application in Large-Scale MIMO Systems View project

Energy Preservation in Mobile Adhoc and Sensor Networks View project

All content following this page was uploaded by Rami Hamdi on 17 November 2016.

The user has requested enhancement of the downloaded file.

 2015 IEEE 11th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)

Sum-rate Maximizing in Downlink Massive MIMO Systems

with Circuit Power Consumption
Rami Hamdi 1,2 and Wessam Ajib 2
1 Department of Electrical Engineering, École de Technologie Supérieure
[email protected]
2 Department of Computer Science, Université du Québec à Montréal

[email protected]

Abstract—The downlink of a single cell base station with optimal power allocation in order to maximize the sum-
(BS) equipped with large-scale multiple-input multiple-output rate. The proposed precoder archives high performance near
(MIMO) system is investigated in this paper. As the number of Regularized Zero Forcing (RZF) and better than MRT with
antennas at the base station becomes large, the power consumed
at the RF chains cannot be anymore neglected. So, a circuit no matrix inversion. In [10], the authors investigate the uplink
power consumption model is introduced in this work. It involves of multi-cell massive MIMO system with MRC receiver.
that the maximal sum-rate is not obtained when activating all An efficient algorithm is proposed to compute jointly the
the available RF chains. Hence, the aim of this work is to optimal training duration and power allocation. In [11], the
find the optimal number of activated RF chains that maximizes number of RF chains is assumed to be limited. Hence,
the sum-rate. Computing the optimal number of activated RF
chains must be accompanied by an adequate antenna selection a joint antenna selection and user scheduling strategy is
strategy. First, we derive analytically the optimal number of introduced for downlink massive MIMO systems. As the
RF chains to be activated so that the average sum-rate is large number of antennas involves the growth of the Channel
maximized under received equal power. Then, we propose an State Information (CSI), the authors in [12] assume limited
efficient greedy algorithm to select the sub-optimal set of RF CSI at the BS and efficient antenna selection algorithm is
chains to be activated with regards to the system sum-rate. It
allows finding the balance between the power consumed at the proposed. In [13], the authors propose a polynomial time
RF chains and the transmitted power. The performance of the algorithm to select antennas with maximum SNR.
proposed algorithm is compared with the optimal performance In massive MIMO systems, as the number of antennas
given by brute force search (BFS) antenna selection. Simulations is getting larger, the power consumed by the RF chains
allow to compare the performance given by greedy, optimal and becomes not negligible [14], [15], [16], [17]. Consequently,
random antenna selection algorithms.
the maximal sum-rate is not obtained when activating all
Index Terms-Massive MIMO, number of RF chains, antenna the RF chains. Hence, the optimal number of RF chains to
selection, circuit power consumption.
be activated that maximizes the sum-rate must be derived.
In [16], the authors investigate the optimization of the power
I. I NTRODUCTION allocation and the number of transmit antennas based on
A major challenge of designing future wireless networks is the asymptotic approximation of the capacity. They extend
to support the huge increase in data traffic demand. Massive their work in [17] to multi-users and imperfect CSI case
MIMO system is one important technology that will be assuming MRT precoding. But, the optimization of the power
adopted in the fifth generation (5G) of cellular networks [1], allocated among users is still based on the average capacity
[2], [3], [4]. It is based on the installation of large number of over channel realizations. It will be more interesting if the
antennas at the Base Station (BS). The growth of the number optimization is based on the instantaneous throughput. As
of antennas involves quasi-orthogonality between the users’ results, the optimal number of transmit antennas should be
channels in consequence of the law of large numbers. Hence, derived jointly with antenna selection strategy depending of
a huge gain in spectral efficiency and energy efficiency can the channel gain coefficients. In this work, we derive first
be obtained while using low complexity transmit and receive analytically the optimal number of activated RF chains that
techniques such as Maximal Ratio Combiner (MRC), Max- maximizes the sum-rate averaged over channel realizations
imal Ratio Transmission (MRT) and Zero Forcing (ZF) [5], under the assumption of equal received power. Then, an
[6], [7]. efficient greedy algorithm is proposed to compute the close-
Few works have previously investigated the optimization to-optimal number of RF chains and to select the best subset
of resource allocation in massive MIMO systems. In [8], of antennas that maximizes the sum-rate. The proposed
the authors proposed an algorithm to compute the optimal algorithm allows to find the balance between the power
power allocation among users in downlink massive MIMO consumed at the RF chains and the output transmitted power.
systems with MRT precoder. However, a low complexity Monte-Carlo simulations allow to compare the performance
linear precoder is designed for downlink massive MIMO of the proposed algorithm with random and brute force search
systems in [9]. The beamforming vector is computed jointly (BFS) optimal antenna selection.

978-1-4673-7701-0/15/$31.00 ©2015 IEEE 437

 2015 IEEE 11th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)

In this paper, all power variables are assumed to be unitless The signal received by user k can be written as
since they are normalized by the average noise power. ⌊.⌋
denotes the floor function, ⌈.⌉ denotes the ceiling function, √ dk 2
−α

diag(p) is a diagonal matrix whose diagonal entries are yk = pk sk + nk . (3)

ση
the elements of the vector p, (x)+ denotes max(0, x), (.)H
represents the Hermitian matrix,CSN is a binomial coefficient The received SIN R for user k is expressed as
S!
defined as CSN = (N −S)!N ! and Tr{.} denotes the trace of
a square matrix. d−α
k pk
The rest of the paper is organized as follows. In Section II, SIN Rk = . (4)
σ2 η2
the system model is presented. In Section III, the problem is
formulated. In Section IV, the optimal number of RF chains Hence, the sum-rate is expressed as
that maximizes the average sum-rate is analytically calculated
under the assumption of equal received power between users. K
d−α
 
k pk
X
Then, an efficient greedy algorithm is proposed to compute R= log2 1 + 2 2 . (5)
the number of RF chains that maximizes the sum-rate and to σ η
k=1
select the best antennas in Section V. The computational com-
plexity of the proposed algorithm is calculated and compared Since the number of antennas is large in the considered
to the optimal BFS algorithm in Section VI. Numerical and system, we assume a non-negligible circuit power consump-
simulation results are shown and discussed in Section VII. tion [14], [15]. Let pc denotes the fixed power consumed
Finally, we conclude and discuss the main results in Section at each activated RF chain (Digital to Analog Converter
VIII. (DAC), mixer, frequency synthesizer, filter) and pmax the
maximal available power at the BS. Hence, the circuit power
consumption constraint can be expressed as
II. S YSTEM M ODEL
We consider the downlink of a single cell massive MIMO N
X
systems. The base station (BS) is equipped with a large pout + αn .pc ≤ pmax , (6)
number of antennas N serving K single-antenna users with n=1

N ≫ K. The channel gain is represented by complex matrix where αn is an antenna coefficient that is set to 1 if antenna
G = D 1/2 H. The small scale fading is defined by complex n is activated and to 0 otherwise and pout is the output
matrix H = [h1 , h2 , ..., hK ], where hk ∈ C1×N is the transmitted power given by
k th channel vector for user k, is assumed to be quasi-static
Gaussian independent and identically distributed (i. i. d.) slow K
X
fading channel. Considering only path loss, the large scale pout = pk . (7)
fading is defined by the vector d = [d−α −α −α
1 d2 ...dK ], where α
k=1
is the path loss exponent and dk the distance between the BS
It is to be noted that ⌊pmax /pc ⌋ represents the maximum
and user k, D is a diagonal matrix defined as D = diag(d).
number of RF chains that can be supported (assuming no
The noise is assumed to be additive Gaussian white (AWGN)
transmission power) by the system due to the circuit power
random variable with zero mean and unit variance. The vector
constraint.
p = [p1 p2 ...pK ] is assumed to be the distribution of power
among users. We consider that BS knows perfectly the CSI.
We consider the Zero Forcing (ZF) as beamforming strategy
because it cancels the inter-user interference and it achieves
high performance [5], [6], [7]. The beam forming matrix is
expressed as III. P ROBLEM F ORMULATION
H H (HH H )−1
W = , (1)
η
where η the normalization factor defined as η = The circuit power consumption involves that the maximal
p
Tr{(HH H )−1 }. achieved sum-rate is not obtained when activatingPNall RF
Hence, the received downlink vector of signals is expressed chains. Hence, the optimal number of RF chains n=1 αn
as that maximizes the sum-rate should be derived. Moreover,
the aim is to find the optimal P
balance between the power
N
D2
1
√ consumed by the RF chains n=1 αn .pc and the trans-
y= diag( p)s + n, (2) mitted power pout . The number of RF chains should be
ση
optimized jointly with adequate antenna selection strategy.
where s is the vector ofq
transmitted data symbol and n is the The transmitted power must be allocated optimally among
Tr{D}
vector of AWGN. σ = K is a normalization factor. users. Hence, the main problem can be formulated as

438
 2015 IEEE 11th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)

(
K
d−α ⌊φ⌋ if R(⌊φ⌋) > R(⌈φ⌉) or ⌊φ⌋ = ⌊pmax /pc ⌋
 
k pk ∗
X
maximize R= log2 1 + S =
p,αn :n=1..N 2 2
σ η (α1 , α2 , ..., αn ) ⌈φ⌉ otherwise
k=1
K N (13)
where
X X
subject to pk + αn .pc ≤ pmax , pmax + K.pc
k=1 n=1 φ= < pmax /pc . (14)
2pc
XN
αn ≥ K, Finally, the power allocated for each user k is given by
n=1
αn ∈ {0, 1}, n = 1..N. pmax − S ∗ .pc σ 2
pk = . (15)
(8) K d−α
k
It is to be highlighted that the term η 2 in the objective However, the number of RF chains that optimizes the
function depends on the fading coefficients of the selected instantaneous sum-rate (for one channels realization) should
antennas. So, an antenna selection strategy is necessary to be derived. It depends of the coefficients of the selected
find the optimal number of RF chains that maximizes the antennas. In consequence, the number of RF chains should
sum-rate. Consequently, the problem becomes combinatorial be optimized jointly with adequate antenna selection strategy
with exponential complexity growth in N . It is known that, which will be the focus of the next section.
after selecting the best antenna, water filling is the optimal
strategy for power allocation among users [19].
V. E FFICIENT G REEDY A LGORITHM

IV. E QUAL R ECEIVED P OWER The aim of this work is to find the optimal number of
RF chains and have an antenna selection that maximizes
In this section, equal received power between users is the sum-rate. The number of RF chains must be derived
assumed and the sum-rate averaged over channel realizations jointly with adequate antenna selection strategy. The antenna
is approximated. Then, we derive analytically the optimal selection is a combinatorial problem with exponential growth
number of RF chains that maximizes the sum-rate averaged in N . Moreover, the optimal antenna selection can be done
over channel realizations. The number of RF chains is given with high complexity brute-force search (BFS) algorithm.
by So, we propose a low complexity greedy iterative algorithm
that determines the number of RF chains maximizing the
N
X instantaneous sum-rate. At each iteration for a fixed number
S= αn . (9) of RF chains, the best antenna n∗ that maximizes the sum-
n=1 rate is determined among the set of non selected antennas Λ.
Note that since the sum-rate is averaged over the channel Since, the cost function η 2 is infinite for S < K as the ZF
realizations, the antenna selection is meaningless. Since equal Precoder does not exist, we choose the first K antennas by
received power between all users is assumed, the transmitted optimal BFS. The best antenna should minimize the term η 2
power allocated for user k can be expressed as
argmaxΛ R = argminΛ η 2 (α1 , α2 , ..., αN ) (16)
2
pmax − S.pc σ
pk = . (10) Hence, the best antenna at each iteration can be derived as
K d−α
k

The term η 2 is approximated in [5], [6], [7], [18], when n∗ = argminΛ η 2 (α1 , α2 , ..., αN ) (17)
K, S −→ ∞
As the selected antennas are given, the optimal strategy for
1 1 S power allocation among users is given by Water filling [19]
= −→ − 1. (11)
η2 Tr{(HH H )−1 } K +
σ2 η2

1
This approximation is validated by simulations in Section pk = − −α , (18)
ln(2)µ dk
VII. Hence, the sum-rate averaged over the channel realiza-
tions can be expressed as where µ is the Water level.
The convergence of the proposed algorithm is obtained
  when the instantaneous sum-rate starts decreasing (RS <
(pmax − S.pc )(S − K)
R = K. log2 1 + . (12) RS−1 ). The convergence point exists since the sum-rate is
K2
a concave function, the second order derivative is negative
d2 R
In consequence, the optimal number of RF chains that dS 2 < 0. Hence, the proposed algorithm allows to determine
maximizes the average sum-rate over channel realizations can the number of RF chains, the selected antennas, the power
be derived allocated among users and the achieved sum-rate.

439
 2015 IEEE 11th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)

Algorithm 1 Greedy Algorithm Hence, the optimal antenna selection can be obtained
1: Initialization with very high complexity. However, the proposed efficient
2: αn ← 0, n = 1 : N algorithm is polynomial time. The complexity order for both
3: Λ ← {n, n = 1 : N } algorithms are presented in Table I.
4: for S = K + 1 : ⌊pmax /pc ⌋ do
5: Select the best antenna n∗ from Λ that minimizes η 2 TABLE I
C OMPUTATIONAL C OMPLEXITY OF THE PROPOSED GREEDY AND BFS
6: αn∗ ← 1 ALGORITHMS .
7: Λ ← Λ \ {n∗ } +
PK  1 σ2 η2
= pmax − S.pc ⊲ find µ BFS Greedy
8: k=1 ln(2)µ − d−α
by bisection method
k
Complexity order CBF S Cgreedy
+
O(8.1016 ) O(6.105 )
 
2 2
9: 1
pk ← ln(2)µ − σd−α
η
, k = 1 : K, power alloca- N = 64, K = 10, ⌊pmax /pc ⌋ = 32

tion
k
N = 128, K = 10, ⌊pmax /pc ⌋ = 128 O(3.1030 ) O(4.107 )
10: Break If RS < RS−1
11: end for
VII. N UMERICAL R ESULTS
In this section, we present numerical results to show the
VI. C OMPLEXITY A NALYSIS performance of the proposed greedy algorithm compared to
In this section, the computational complexity order for random antenna selection and optimal BFS algorithm. Monte-
both optimal brute force antenna selection and the proposed Carlo simulations are done to show the optimized number of
greedy algorithm are calculated. For the brute-force search RF chains and the maximal achieved sum-rate.
optimal algorithm, the number of possible combinations of Simulation parameters are summarized in Table II. The BS
sets of the antennas is is equipped with 256 antennas serving 10 users.

⌊pmax /pc ⌋ TABLE II

X S YSTEM PARAMETERS .
CSN −S+1 (19)
S=1 Symbol Description Value
For the proposed greedy algorithm, the number of combi- K number of users 10
nations of sets of the antennas is given by N number of antennas 256
⌊pmax /pc ⌋
X Pc fixed power RF chains 0.05
N −S+1 (20) α path loss exponent 3.7
S=1 cell radius 500 m
For each selected set of antennas, the coefficient η 2 is
calculated. So, we need one matrix multiplication and one Fig. 1 shows the maximal achieved sum-rate in function
matrix inversion. The complexity order of the matrix multi- of the number of antennas N under the proposed greedy
plication is O(SK 2 ) and the complexity order for the matrix algorithm and optimal antenna selection. Due to the high
inversion is O(K 3 ). complexity of BFS algorithm, simulation results for optimal
Hence, the computational complexity order of the BFS antenna selection are presented for limited number of anten-
algorithm is given by nas N , three users K = 3 and pmax = 1. It is clear that
the achieved sum-rate increases as the number of antennas
increases since it offers more diversity to the transmitter. The
 
⌊pmax /pc ⌋
difference between the performance of the low complexity
X
CBF S = O  CSN −S+1 (SK 2 + K 3 ) (21)
S=1 proposed greedy algorithm and the optimal antenna selection
strategy is small.
and the computational complexity order of the greedy algo- In Fig. 2, we see the achieved average sum-rate in function
rithm is given by of the number of RF chains under the proposed greedy
  algorithm and random antenna selection for different values
⌊pmax /pc ⌋
X of total available power at the BS. First, the obtained results
Cgreedy = O  (N − S + 1)(SK 2 + K 3 ) confirm the hypothesis that the maximal achieved sum-rate
S=1 is not obtained when activating all RF chains. Then, the
= O(⌊pmax /pc ⌋K 2 (K(N + 1) performance of the proposed greedy algorithm is significantly
(22)
1 higher than the random antenna selection because the greedy
+ (N + 1 − K)(⌊pmax /pc ⌋ + 1) algorithm allows to select the best antenna at each iteration.
2
1 Fig. 3 shows the optimal number of RF chains to be
+ (⌊pmax /pc ⌋ + 1)(2⌊pmax /pc ⌋ + 1))) activated that maximizes the average sum-rate in function
6

440
 2015 IEEE 11th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)

3 110
Greedy
2.9
Random (simulation)
Maximal achieved sum-rate (bits/sec/Hz)

100
Greedy Random (analytic)
2.8 Optimal

Number of transmit RF chains

90
2.7

80
2.6

2.5 70

2.4
60
2.3
50
2.2

2.1 40

2 30
10 15 20 25 30 35 40 45 50 55 60 3 4 5 6 7 8 9 10
Number of antennas N pmax

Fig. 1. Maximal achieved sum-rate under optimal and greedy antenna Fig. 3. Number of activated transmit RF chains under random and greedy
selection (K = 3, pmax = 1). antenna selection.

22

20
Greedy, pmax = 3
Greedy, pmax = 6
antenna selection. So, there is more available power for
Greedy, pmax = 9 transmitting pout . Hence, more users can be served under the
18
Random, pmax = 3 proposed greedy algorithm and the maximal achieved sum-
Average sum-rate (bits/sec/Hz)

Random, pmax = 6
16
Random, pmax = 9 rate increases significantly.
14

16
12

10
14
8
Average sum-rate (bits/sec/Hz)

6
12
4

2
10

0
20 40 60 80 100 120 140 160 180
Number of transmit RF chains
8 Random, pmax = 8
Random, pmax = 6
Fig. 2. Average sum-rate under random and greedy antenna selection. Random, pmax = 4
6 Greedy, pmax = 8
Greedy, pmax = 6
Greedy, pmax = 4
of the maximal available power at the BS. We observe that 4
5 10 15 20 25 30
the analytic expression of the optimal number of RF chains Number of users K

obtained under equal received power and random antenna

selection fits with simulation results. Then, it is clear that Fig. 4. Impact of the number of user K on the system performance.
the optimized number of RF chains obtained by the proposed
greedy algorithm is less than the number of RF chains Finally, Fig. 5 shows the optimized number of RF chains in
obtained under random antenna selection due to the adequate function of the number of users K for different values of the
antenna selection strategy and the power is optimally dis- maximal available power at the BS. The number of RF chains
tributed among users. So, the proposed algorithm allows to increases when increasing the number of users K since we
use less number of transmitted RF chains and more power is need more RF chains for transmitting. Also, as the maximal
available at the base station for transmitting. Hence, it allows available power pmax increases, there is more flexibility to
to increase the achieved sum-rate. use more transmit RF chains. It is clear that the computed
Now, we investigate the impact of the number of users K number of RF chains under the greedy algorithm is less
on the system performance. In Fig. 4, we see the maximal than random antenna selection. In consequence, there is more
achieved sum-rate in function of the number of users under available power for transmitting and the system performance
the proposed greedy algorithm and random antenna selection. is ameliorated.
So, the optimal number of users that maximizes the sum-rate
can be observed for different values of that maximal available VIII. C ONCLUSION
power pmax at the BS. The proposed greedy algorithm allows The downlink massive MIMO system with non negligible
to transmit with less number of RF chains than random circuit power consumption is investigated in this paper.

441
 2015 IEEE 11th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)

100
[8] L. Zhao, Hu. Zhao, F. Hu, K. Zheng and J. Zhang, “Energy Efficient
Power Allocation Algorithm for Downlink Massive MIMO with MRT
90 Precoding,” in proc. of IEEE Vehicular Technology Conf. (VTC), pp. 1-5,
Sept. 2013.
Number of transmit RF chains

[9] S. Zarei, W. Gerstacker and R. Schober, “A Low-Complexity Linear

80
Precoding and Power Allocation Scheme for Downlink Massive MIMO
Systems,” in proc. of Asilomar Conf. on Signals, Systems and Computers,
70 pp 285-290, Nov. 2013.
[10] H. Q. Ngo, M. Matthaiou and E. G. Larsson, “Massive MIMO
60
with Optimal Power and Training Duration Allocation,” IEEE Wireless
Commun. Let., vol.3, issue: 6, pp. 605-608, Dec. 2014.
Random, pmax = 8 [11] M. Benmimoune, E. Driouch, W. Ajib and D. Massicotte, “Joint
50 Random, pmax = 6 Transmit Antenna Selection and User Scheduling for Massive MIMO
Random, pmax = 4 Systems,” in proc. of IEEE Wireless Commun. and Networking Conf.
Greedy, pmax = 8 (WCNC), March 2015.
40
Greedy, pmax = 6
[12] M. Al-Shuraifi and H. Al-Raweshidy, “Optimizing Antenna Selection
Greedy, pmax = 4
Using Limited CSI for Massive MIMO Systems,” in proc. of Int. Conf. on
30
5 10 15 20 25 30 Innovative Computing Technology (INTECH), pp. 180-184, Aug. 2014.
Number of users K [13] M. Gkizeli and G. N. Karystinos, “Maximum-SNR Antenna Selection
Among a Large Number of Transmit Antennas,” IEEE J. of Sel. Topics
Fig. 5. Number of activated transmit RF chains in function of the number in Signal Processing, vol. 8, no. 5, Oct. 2014.
of users K. [14] Y. Pei, T.-H. Pham and Y.-. Liang, “How Many RF Chains are Optimal
for Large-Scale MIMO Systems When Circuit Power is Considered?,”
in proc. of IEEE Global Commun. Conf. (GLOBECOM), pp. 3868-3873,
Dec. 2012.
Since this work considers the circuit power consumption, [15] H. Li, L. Song, D. Zhu and M. Lei, “Energy Efficiency of Large Scale
MIMO Systems with Transmit Antenna Selection,” in proc. of IEEE Int.
the maximal sum-rate is not obtained when activating all Conf. on Commun. (ICC), pp. 4641-4645, June 2013.
the available RF chains. For this reason, we focus in this [16] D. W. K. Ng and Robert Schobe, “Spectral Efficiency in Large-Scale
paper on optimizing the number of activated RF chains MIMO-OFDM Systems with Per-Antenna Power Cost,” in proc. of
Asilomar Conference on Signals, Systems, and Computers, pp. 289-294,
that maximizes the sum-rate. So, we derive analytically the 2012.
optimal number of activated RF chains that maximizes the [17] D. W. K. Ng, E. S. Lo and Robert Schobe, “Energy-Efficient Resource
average sum-rate under equal received power. However, the Allocation in OFDMA Systems with Large Numbers of Base Station
Antennas,” IEEE Trans. on Wireless Commun., vol. 11, issue 9, July
number of RF chains that maximizes the sum-rate should 2012.
be optimized jointly with adequate antenna selection. Hence, [18] B. Hochwald and S. Vishwanath, “Space-time multiple access: linear
an efficient greedy iterative algorithm that optimizes the growth in the sum rate,” in proc. of Annual Allerton Conf. Communica-
tions, Control and Computing, Oct. 2002.
number of activated RF chains is proposed. It allows to [19] E. Driouch and W. Ajib, “Efficient Scheduling Algorithms for Mul-
find the balance between the power consumed at the RF tiantenna CDMA Systems,” IEEE Trans. on Vehicular Technology, vol.
chains and the transmitted power. Simulations are done to 61, no. 2, Feb. 2012.
compare performance of the proposed algorithm to random
and optimal antenna selection. Numerical results show the
efficiency of the proposed algorithm.

R EFERENCES

[1] C. X. Wang, F. Haider, X. Gao, X. H. You, Y. Yang,D. Yuan, H. M.

Aggoune, H. Haas, S. Fletcher and E. Hepsaydir, “Cellular Architecture
and Key Technologies for 5G Wireless Communication Networks,” IEEE
Commun. Mag., vol. 52, no. 6, pp. 122-130, Feb. 2014.
[2] J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly, A. Lozano, A. C. K.
Soong, and J. C. Zhang, “What Will 5G Be?,” IEEE J. of Sel. Areas in
Commun., vol. 32, no. 6, June 2014.
[3] F. Boccardi, R. W. Heath Jr., A. Lozano, T. L. Marzetta and P. Popovski,
“Five Disruptive Technology Directions for 5G,” IEEE Commun. Mag.,
vol. 52, no. 2, pp. 74-80, Feb. 2014.
[4] E. G. Larsson, O. Edfors, F. Tufvesson and T. L. Marzetta, “Massive
MIMO for Next Generation Wireless Systems,” IEEE Commun. Mag.,
vol. 52, no. 2, pp. 186-195, Feb. 2014.
[5] 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 Processing Mag., vol. 30, no. 1,
Jan. 2013.
[6] H. Yang and T. L. Marzetta, “Performance of Conjugate and Zero-
Forcing Beamforming in Large-Scale Antenna Systems,” IEEE J. of Sel.
Areas in Commun., vol. 31, no. 2, Feb. 2013.
[7] H. Q. Ngo, E, G. Larsson and T. L. Marzetta, “Energy and Spectral
Efficiency of Very Large Multiuser MIMO Systems,” IEEE Trans. on
Commun., vol. 61, no. 4, Apr. 2013.

View publication stats

442