Sie sind auf Seite 1von 6

2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 2: MAC and Cross-Layer Design

Power Allocation for Device-to-Device Communication


Underlaying Massive MIMO Multicasting Networks
Belal S. Amin* , Yahia R. Ramadan* , Ahmed S. Ibrahim* , and Mahmoud H. Ismail*
*
Department of Electronics and Electrical Communications Engineering, Cairo University, Giza 12316, Egypt
Department of Electrical Engineering, The American University of Sharjah, PO Box 26666, Sharjah, UAE

E-mail: {belal.s.amin, yahia.r.ramadan, asibrahim, mhismail}@ieee.org

AbstractDevice-to-device communication (D2D) is an emerg- Many interference management techniques were proposed
ing technology aiming at enhancing the performance of next in the literature to solve the issue of mutual interference
generation wireless communication systems. D2D communication between D2D devices and CUEs. In most of those techniques,
enables two mobile stations to communicate directly without
traversing the Base Station (BS). Smart techniques should be D2D pairs are not admitted to operate on the frequency re-
applied to manage the interference between the BS and the sources in which they will violate required Quality of Service
D2D communicating devices. In this paper, we investigate the (QoS) constraints on the CUEs, e.g., [3]. Other algorithms
problem of BS precoder design and Power Allocation (PA) to were proposed based on the locations of the users, prohibiting
deploy D2D communication in a single-cell network. We assume any CUE and D2D pair to operate in the same area [4].
that the BS uses a very large number of antennas, referred to as
Massive Multiple-Input-Multiple-Output (MIMO). We propose Other researchers satisfy these QoS constraints while aiming
solution algorithms to maximize the sum of the achievable data at optimizing some objective function such as maximizing the
rates of the D2D pairs, or their minimum, while maintaining total sum rate of the network [5], or minimizing the total power
some Quality of Service (QoS) constraints on the Cellular User consumption [6]. The common characteristic of all these trials
Equipments (CUEs) which communicate with the BS. Simulation is that they involve mode selection, i.e., D2D pairs may not
results show that the proposed solutions are superior to the
conventional power allocation schemes. operate on some resources at all.
Index TermsDevice-to-device communication, Massive The problem of deploying D2D communication in a MIMO
MIMO, Multicasting. network has gained less interest in the literature. In [7], the
authors added the advantage of precoder design to maximize
I. I NTRODUCTION the sum rate of the network, but also with binary Power
Allocation (PA). In [8], the effect of different MIMO precoding
Recently, the data rate demand of mobile users has dras- algorithms on the performance of the network with D2D
tically grown. New techniques have been proposed for new communications was evaluated.
wireless technologies to meet this demand. One of these tech- In this paper, we investigate a scenario that deploys both
niques is Device-to-device communication (D2D), a scenario D2D communications and Massive MIMO. A BS with large
in which two mobile nodes communicate directly without number of antennas multicasts a single data stream to the
traversing the Base Station (BS) or the core network [1]. CUEs. In such multicasting scenario, we may enable a mobile
Advantages of D2D communication include increasing the station to send the data directly to another mobile station
capacity of the network, enhancing the spectral efficiency, in its proximity. We target the problem of optimizing the
and extending the battery life of the mobile stations. Another precoder design at the BS and the PA of the BS and the D2D
technique is Massive MIMO where the central BS is equipped transmitters, in a way that will increase the achievable sum
with large number of antennas. Massive MIMO improves rates of the D2D pairs subject to preserving required QoS
the energy efficiency of the network due to its ability to constraints on the CUEs. We formulate a MaxSum optimiza-
concentrate all the energy into the direction of the intended tion problem and provide two possible solutions for it. We also
receivers [2]. introduce other solutions for the problem of maximizing the
D2D devices can communicate either through the same minimum achievable rates of the D2D pairs while maintaining
frequency resources used by the network (inband D2D) or the same QoS constraints. We provide simulations to compare
using other frequency bands (outband D2D). Inband D2D the performance of the proposed solutions.
is divided to underlay D2D, in which devices are admitted The paper is organized as follows: in Section II, we in-
communication using the same frequency resources used by troduce the system model as well as the parameters and
the cellular user equipments (CUEs), or overlay D2D using notations of the paper. In Section III, we formulate the
dedicated frequency resources that CUEs are not allowed MaxSum optimization problem. In Section IV, we propose two
to use. Since inband underlay D2D reuses the frequency possible solutions to it. In section V, we introduce the MaxMin
resources, it increases the spectral efficiency of the network, optimization problem and propose its solution. In Section VI,
but raises the problem of interference between the D2D simulation results are presented and discussed before the paper
communicating devices and the CUEs [1]. is finally concluded in Section VII.

978-1-4799-8406-0/15/$31.00 2015 IEEE 1219


2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 2: MAC and Cross-Layer Design

II. S YSTEM M ODEL


We consider a downlink cellular network with a BS
equipped with N antennas, K CUEs, and D D2D pairs as
shown in Fig. 1. Each of the CUEs and the D2D devices
is equipped with a single omnidirectional antenna. We will
consider a scenario where the number of antennas N is much
larger than K (N K) to make use of the benefits of
Massive MIMO. In a multicasting setting, the BS transmits
the same data symbol sB to all the CUEs. On the other hand,
the dth D2D transmitter sends a symbol sd to its corresponding
receiver. We assume that all the data symbols have an average
unit power, i.e., E{|sB |2 } = E{|sd |2 } = 1. The received
signals at the k th CUE and the dth D2D receiver are given by

p D p
X
ykCUE = BS PBS hH
k ws B + d PD gd,k sd + nk , (1) Fig. 1. Single cell multicasting network with Massive MIMO, K = 4 cellular
d=1 users (red terminals), D = 3 D2D pairs (green terminals). Dashed lines
represent interference.
and
D p
sources is independent. Thus, the SINR of the dth D2D pair
p X
ydD2D = d PD d,d sd + d PD d ,d sd
d =1
is calculated as
d 6=d
d PD |d,d |2
D2D = PD . (4)
p
+ BS PBS fdH wsB + zd , (2) d
d =1 d PD |d ,d |2 +BS PBS |fdH w|2 +No
d 6=d
respectively, where
III. M AX S UM P ROBLEM F ORMULATION
PBS and PD are the maximum powers of the BS and a
D2D transmitter, respectively, We want to jointly determine the optimal values of the
BS and d are power adjustment factors for the BS and PA factors 1 , 2 , . . . , D , BS and the precoding vector w
the dth D2D transmitter, where BS , d [0, 1] d in a way that maximizes the sum rate of the D2D pairs
{1, . . . , D}, while preserving QoS constraints on the CUEs links. We note
w is an N 1 precoding vector at the BS, that the large number of antennas at the BS insures good
nk and zd are additive white Gaussian noise C(0, No ) communication links, with very high SNRs, between the BS
at the k th CUE and the dth D2D receiver, respectively. and the CUEs [9]. Let
RkCUE = E log 1 + CUE
 
The channel gain between any two nodes is modeled as a k (5)
Rayleigh fading channel with path loss in the form h where
is the path loss between the two nodes and h is the small and
RdD2D = E log 1 + D2D
 
scale complex Gaussian channel coefficient C(0, 1). In (1), d (6)
hk is the N 1 channel gain vector from the BS to the be the average achievable rates of the k CUE link, and dth
th
k th CUE, and gd,k is the channel gain from the dth D2D D2D pair link, respectively, where E{.} is the expectation
transmitter to the k th CUE. In (2), l,m is the channel gain operator. The optimization problem can be formulated as
from the lth D2D transmitter to the mth D2D receiver, and fd
is N 1 channel vector from the BS to the dth D2D pair. We D
assume that all the channels gains are available at the BS. max
X
RdD2D , (7a)
From (1), the Signal to Interference plus Noise Ratio (SINR) 1 ,2 ,...,D ,BS ,w
d=1
at the k th CUE is
subject to kCUE th , k {1, ..., K}, (7b)
BS PBS |hH
k w|
2
0 i 1, i {1, 2, ..., D, BS}, (7c)
CUE
k = PD . (3)
2
d=1 d PD |gd,k | +No |wH w|= 1, (7d)
In (2), the second term represents the interference from the where th is the minimum required SINR at the CUEs for
other D2D transmitters, and the third one represents the satisfactory performance. Constraint (7b) represents the QoS
interference from the BS. The total interference power can constraint, while (7c) represents the power constraint, and
be calculated as the sum of the powers of these interference (7d) represents the fact that the BS precoder should have
sources. This is due to the fact that the interference from these normalized unit power.

1220
2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 2: MAC and Cross-Layer Design

(D+1)
IV. M AX S UM P ROBLEM S OLUTION where fk = ak eD+1 th bk , and  denotes component-wise
inequality. The optimization problem (8a)(8c) is nonconvex,
The optimization problem (7a)(7d) is nonlinear, and
difference of convex (DC) programming problem [12].
not convex, and hence, not directly solvable. We propose
to suboptimally decouple the problem by first choosing a To solve the DC program, we propose an iterative solution.
suitable precoder w and then optimizing the PA factors By approximating the logarithm log cTd Id + No with its
1 , 2 , . . . , D , BS as will be explained in the sequel. first order Taylor expansion, (8a) is approximated to
D
X D
 X
A. Precoder Design = argmax log cTd + No log cTd Id o + No

d=1 d=1
Beamforming Precoder (BF): In [10], it was shown that the
D
optimum precoder for a multicasting massive MIMO system, X IdT cd
without the existence of D2D users, P is a weighted sum of ( o ) , (9)
K d=1
cTd Id o + No
the channels of the CUEs, i.e., w = k=1 k hk where the
weights k act as power distribution factors. For simplicity, we which is a concave function in . o is an initial value for the
will consider equal power distribution among the users, i.e., PA vector . Using this initial value, the optimization problem
k = 1/K. In this case, the precoder coefficients are given by with (9) as its objective function is concave, and can be solved
PK using interior point method to get the optimum value . Then,
k=1 hk we set o = and solve the optimization problem again.
wBF = q P ,
K
N k=1 k
(h) The new optimum value is then used as an initial value for
another iteration, and so on. Eventually, the optimum value
(h)
where k is the path loss between the BS and the k th CUE. will converge to the optimum value of the original function
Due to the law of large the fact that fd and wBF (8a) as will be shown later. Interior point methods solve the
 Hnumbers, and problem in typically
2 a.s.
are independent, E |fd wBF | 1, where a.s. denotes n 10 to 100 steps, with each step requiring
o
3
N an order of max (D + 1) , (D + 1)2 (K + 2D), F opera-
asymptotic convergence [11]. Hence, the interference from the
BS on the D2D users in (4) is, on average, constant. tions, where F is the cost of evaluating the first and second
derivatives of (9) and (8b) [13]. Thus, if we denote the number
Zero Forcing (ZF): The second variant is the ZF precoder
of iterations of the algorithm as NOptSum , and the number of
where the BS will transmit its data in the null space of
steps of the interior point method as NIPM , the complexity of
the channels between the BS and the D2D receivers, thus
the OptSum solution is
cancelling out the interference made by the BS on them. In
other words, w is chosen such that O NOptSum .NIPM . max D2 K + 2D3 , F .
 

fdH wZF = 0, d {1, 2, . . . , D}, MaxSum problem solution 2 (Heuristic): Due to the high
H complexity of the previous solution, we propose a suboptimal
and |wZF wZF |= 1.
heuristic that tries to maximize the sum rate of the D2D
B. Power Allocation pairs while maintaining the QoS constraints on the CUEs. The
details of the heuristic are shown in Algorithm 1. The D2D
MaxSum PA problem solution 1 (OptSum): After choosing transmitters are initialized to operate with full power. CUEs
a specific precoder w, let are chosen one by one in a descending order according to the
, [1 , 2 , . . . , D , BS ]T , interference power on them. Once a CUE is chosen, if its SINR
is below the SINR threshold, powers of the D2D pairs are
ak , PBS |hH
k w| ,
2
adjusted such that this CUE operates on the SINR threshold.
bk , [PD |g1,k |2 , PD |g2,k |2 , . . . , PD |gD,k |2 , 0]T , In the power adjustment phase, two factors are taken into
cd , [PD |1,d |2 , PD |2,d |2 , . . . , PD |D,d |2 , consideration: a) a D2D pair with high channel gain between
PBS |fdH w|2 ]T , its transmitter and the CUE is preferred to have lower power
because it induces higher interference on the CUE, this is
(m) expressed by the parameter Id , and b) a D2D pair with higher
el be a column vector of m zeros with the lth element set to
1, and Id be the identity matrix with the dth diagonal element channel gain between its transmitter and receiver is preferred
set to zero. The optimization problem (7a)(7d) can now be to retain its high power, due to the possibility to achieve higher
rewritten in matrix form as rate, this is expressed by the parameter Cd . According to those
two parameters, each D2D pair is given a score Wd which is
D
directly proportional to Id and inversely proportional to Cd .
cTd + No
 
The higher the score Wd of a certain D2D pair is, the lower its
X
= argmax log , (8a)
d=1
cTd Id + No power gets. A weighting factor [0, 1] defines the weights
of the parameters Id and Cd in the calculation of Wd . If = 1,
subject to fkT th No 0, k {1, ..., K}, (8b)
pairs with lower possible rates are the ones that contribute to
0   1, (8c) the power adjustment, which is preferable to maintain higher

1221
2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 2: MAC and Cross-Layer Design

sum rate, but this may violate the constraint that new  0. If
= 0, the possible rates of the D2D pairs are not taken into D
RdD2D

max min , (10a)
consideration and the pairs adjust their powers according to 1 ,2 ,...,D ,BS ,w d d=1
their interference on the CUE. A simple quantized search for a subject to kCUE th , k {1, ..., K}, (10b)
suitable choice of is used. The
 algorithm has a computational
complexity of order O KD2 . 0 i 1, i {1, 2, ..., D, BS},
(10c)
Algorithm 1 D2D power allocation algorithm |wH w|= 1. (10d)
1: Step 1: Initialize: = 11(D+1)
After choosing a specific precoder w, BF or ZF, maximiza-
2: Step 2: Calculate the interference on CUEs
tion of the minimum D2D rate corresponds to the maximiza-
D
X tion of the minimum D2D SINR. Hence, the optimization
Ik = d PD |gd,k |2 , k {1, . . . , K} problem can be modeled as
d=1
D
3: Sort the CUEs in a descending order according to the cTd + No

interference power on them. = argmax min , (11a)
d cTd Id + No d=1
4: Step 3:
5: for k = 1 to K do subject to F th No  0, (11b)
6: if kCUE < th then 0   1, (11c)
7: Step 4: Calculate the wanted drop in interference on where F is a K (D + 1) matrix with fdT
as its rows.
the k th CUE as Ikdrop = Ik Ith where The optimization problem (11a)(11c) is a Generalized Lin-
PBS |hH 2 ear Fractional Program (GLFP) that can be solved using
k w|
Ith = No the famous bisection algorithm for quasiconvex optimization
th
problems [13]. Since the bisection algorithm halves the search
8: Step 5: Calculate the share of each D2D pair in the space in each iteration, the total number of iterations is
interference drop log2 (o/) where o is the accuracy of the initial solution,
9: for d = 1 to D do and is the required final accuracy. We terminate the algorithm
10: - Calculate the share of the dth D2D transmitter whenever < 104 or 30 iterations have passed, whichever is
2
in the current interference as Id = d PD |gd,k | /Ik earlier. Each iteration is a Linear Programming (LP) feasibility
11: - Calculate an indicator for the strength of the dth problem that can be solved in polynomial time.
2 P
pair channel as Cd = |d,d | / Dd=1 |d,d |
2

12: end for VI. S IMULATION R ESULTS


13: for = 1 : 0.1 : 0 do In this section, we use Monte-Carlo simulations for a
14: - Calculate the shares of the D2D transmitters of downlink single-cell network to compare the performance
the interference drop as of the different PA and precoding schemes. The simulation
parameters are shown in Table I. We provide the simulation re-
Wd = (1 )Id + (1 Cd )
sults of our proposed algorithms, comparing their performance
15: - Calculate the new PA factors as with a conventional QoS-constrained TDMA scheme. In the
Wd Idrop TDMA scheme, for each channel instance, e.g., coherence
new
d = d PD time, only one D2D pair is chosen for operation in a round-
PD |gd,k |2 d=1 Wd
robin fashion. The power adjustment factor for the operating
16: if new has an all positive values then D2D transmitter is calculated such that it does not violate the
17: = new QoS constraint on all the CUEs.
18: break; Fig. 2 shows the convergence of the iterative algorithm
19: end if used in the OptSum solution for the MaxSum optimization
20: end for problem. The rate achieved by the optimum PA obtained from
21: end if the convex optimization solver after each iteration is plotted.
22: end for It is shown that after each iteration of the algorithm explained
in Section IV, the objective sum rate is increased. The value of
the global maximum of the sum rate obtained by the complex
V. M AX M IN P ROBLEM F ORMULATION AND S OLUTION exhaustive search algorithm is plotted for reference. As shown,
Maximizing the sum rate of the D2D pairs does not guar- the algorithm converges to the global optimum value obtained
antee fairness between the pairs. Another objective function by exhaustive search.
that guarantees some kind of fairness is to maximize the To study the performance of the different PA algorithms
minimum rate of the D2D pairs. The optimization problem on the sum rate of the D2D pairs, Fig. 3 shows this sum
is then formulated as rate as a function of the number of D2D pairs D. The

1222
2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 2: MAC and Cross-Layer Design

50
TABLE I ZF, OptSum
45
S IMULATION PARAMETERS ZF, TDMA

Sum rate of D2D pairs (bits/s/Hz)


40 ZF, Heuristic
Parameter Value ZF, MaxMin
35
Cell radius (R) 200 m
30
Number of base station antennas (N ) 100
Number of CUEs (K) 5 25
Base station maximum power (PBS ) 30 dBm 20
Mobile terminal maximum power (PD ) 13 dBm
Path loss exponent () 3 15

Noise power 107 mW 10


Users distribution inside the cell uniform
5
Average inter-D2D distance 12 m 1 2 3 4 5 6 7 8 9 10
Number of D2D pairs (D)
Threshold SINR for CUEs (th )  20.96 dB
Number of iterations of OptSum NOptSum 8
Fig. 3. Effect of number of D2D pairs on the achievable sum of D2D rates
for different PA techniques.

24 14
ROptSum algorithm ZF, OptSum
23
ZF, TDMA
Exhaustive Search 12
22 ZF, Heuristic

Minimum D2D rate (bits/s/Hz)


ZF, MaxMin
D2D sum rate (bits/s/Hz)

21 10
20
8
19

18
6
17

16 4

15
2
14

13 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Iteration number Number of D2D pairs (D)

Fig. 2. Convergence of the iterative algorithm used in the OptSum solution Fig. 4. Effect of number of D2D pairs on the minimum of D2D rates for
with uniform random initialization. D = 3. different PA techniques.

precoder is chosen as ZF. It can be seen that among the and has a maximum value of 1 and minimum value of 1/D.
PA schemes, OptSum algorithm outperforms the MaxMin and Table II compares between the different PA schemes in terms
TDMA algorithms. All the schemes have an increasing sum of the D2D performance. It can be seen that the MaxMin
rate with increasing D, except for the TDMA scheme, due algorithm is the best in terms of fairness.
to the fact that, regardless of the number of the D2D pairs, To study the effect of the different precoding techniques on
only one pair is active at a time. The PA heuristic has inferior the performance of the network, Fig. 5 shows the Cumulative
performance to the OptSum Algorithm, but better than the Distribution Function (CDF) of the D2D achievable rates
MaxMin algorithm in terms of the sum rate. under different PA schemes with using BF and ZF precoders.
As for the minimum of the rates of the D2D pairs, Fig. It is clear that the ZF precoder is better than the BF precoder
4 shows the performance of the different PA algorithms as a due to its ability to suppress interference on the D2D receivers.
function of the number of D2D pairs D. MaxMin algorithm The figure also shows that the CDF of the D2D rates under
outperforms OptSum and TDMA algorithms. Once D is MaxMin PA exhibit higher minimum rate and lower maximum
greater than 1, the TDMA algorithm achieves a zero minimum rate than those achieved by OptSum PA. The PA heuristic
since D 1 pairs are always turned off in every channel algorithm acts as a tradeoff between the two schemes, by
instance. It is also seen that the heuristic algorithm achieves achieving lower minimum rate than MaxMin and lower sum
the same performance as both OptSum and MaxMin for a
single D2D pair in the cell.
To have an indication about the fairness of the PA schemes, TABLE II
Jains Fairness Index (JFI) is used, where C OMPARISON BETWEEN THE PA SCHEMES . D = 3.

P 2 OptSum Heuristic MaxMin


D Sum rate (bits/s/Hz) 20.723 16.98 15.78
d=1 R d
JF I = P  Min rate (bits/s/Hz) 2.4 3.81 5.25
D 2 JFI 0.66 0.77 0.98
D. R
d=1 d

1223
2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 2: MAC and Cross-Layer Design

1
of the sum rate of the D2D pairs, but is the most complex.
0.9
OptSum A suboptimal, but far less complex, heuristic was proposed.
Heuristic
0.8
MaxMin was shown to have a higher fairness index than all
MaxMin
0.7
the other PA schemes and is recommended when fairness is
0.6
MaxMin an objective.
CDF

0.5
BF precoder Heuristic R EFERENCES
0.4

0.3
OptSum [1] A. Asadi, Q. Wang, and V. Mancuso, A survey on device-to-device
communication in cellular networks, IEEE Communications Surveys
0.2
Tutorials, vol. PP, no. 99, pp. 11, 2014.
0.1
ZF precoder
[2] E. Larsson, O. Edfors, F. Tufvesson, and T. Marzetta, Massive MIMO
0 for next generation wireless systems, IEEE Communications Magazine,
5 0 5 10 15 20 vol. 52, no. 2, pp. 186195, February 2014.
D2D rate (bits/s/Hz) [3] R. Zhang, X. Cheng, L. Yang, and B. Jiao, Interference-aware graph
based resource sharing for device-to-device communications underlaying
cellular networks, in IEEE Wireless Communications and Networking
Fig. 5. CDF of the D2D achievable rates under different PA and precoding Conference (WCNC), April 2013, pp. 140145.
schemes. D = 3. [4] H. Min, J. Lee, S. Park, and D. Hong, Capacity enhancement using an
interference limited area for device-to-device uplink underlaying cellular
networks, IEEE Transactions on Wireless Communications, vol. 10,
1
no. 12, pp. 39954000, December 2011.
0.9 [5] D. Feng, L. Lu, Y. Yuan-Wu, G. Li, G. Feng, and S. Li, Device-to-
0.8
device communications underlaying cellular networks, IEEE Transac-
OptSum tions on Communications, vol. 61, no. 8, pp. 35413551, August 2013.
0.7
[6] X. Xiao, X. Tao, and J. Lu, A QoS-aware power optimization scheme
0.6 in OFDMA systems with integrated device-to-device (D2D) communi-
cations, in IEEE Vehicular Technology Conference (VTC Fall), Sept
CDF

0.5
MaxMin 2011, pp. 15.
0.4 [7] L. Wei, R. Hu, T. He, and Y. Qian, Device-to-device (D2D) communi-
0.3 cations underlaying MU-MIMO cellular networks, in IEEE Globecom
0.2
Heuristic Workshops (GC Wkshps), Dec 2013, pp. 49024907.
[8] P. Janis, V. Koivunen, C. Ribeiro, K. Doppler, and K. Hugl,
0.1 Interference-avoiding MIMO schemes for device-to-device radio un-
0 derlaying cellular networks, in IEEE 20th International Symposium on
20 30 40 50 60 70 80 90
CUE SINR (dB)
Personal, Indoor and Mobile Radio Communications, Sept 2009, pp.
23852389.
[9] T. Marzetta, Noncooperative cellular wireless with unlimited numbers
Fig. 6. CDF of the SINR of the CUEs under different PA and precoding of base station antennas, IEEE Transactions on Wireless Communica-
techniques. Solid lines represent ZF precoder. Dashed lines represent BF tions, vol. 9, no. 11, pp. 35903600, November 2010.
precoder. D = 3. [10] Z. Xiang, M. Tao, and X. Wang, Massive MIMO multicasting in
noncooperative cellular networks, IEEE Journal on Selected Areas in
Communications, vol. 32, no. 6, pp. 11801193, June 2014.
rate than OptSum. [11] J. Hoydis, Massive MIMO and HetNets: Benefits and challenges,
Newcom Summer School on Interference Management for Tomorrows
As for the performance of the CUEs, Fig. 6 shows the effect Wireless Networks, 2013.
of the different PA and precoding schemes on the CDF of the [12] R. Horst and N. Thoai, DC programming: Overview, Journal of
CUEs SINRs. All the algorithms have similar performance in Optimization Theory and Applications, vol. 103, no. 1, pp. 143, 1999.
[13] S. Boyd and L. Vandenberghe, Convex Optimization. New York, NY,
terms of preserving the QoS constraint. BF precoder performs USA: Cambridge University Press, 2004.
better than ZF precoder, since it has the ability to direct the
available power at the BS to the CUEs. For the PA heuristic,
one of the K CUEs in each channel instance operates on the
SINR threshold. This CUE is the one that is most affected by
the operation of the D2D pairs. For K = 5, it is shown in the
figure that 1/K = 20% of all the CUEs have an SINR equal
to the SINR threshold.
VII. C ONCLUSION
In this paper, we investigated many precoding and PA
schemes for D2D communication underlaying a downlink
single-cell multicasting network. We proposed solution algo-
rithms for the problems of maximizing the sum rates of the
D2D pairs and their minimum rate. It was shown that the ZF
precoder has much higher performance than BF precoder in
terms of the rates achieved by the D2D pairs, but slightly
worse in terms of the SINRs of the CUEs. It was shown that
MaxSum PA is better than the other PA schemes in terms

1224

Das könnte Ihnen auch gefallen