You are on page 1of 12

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2487220, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXXX 2015

Joint Admission Control, Mode Selection and


Power Allocation in D2D Communication Systems
Muhammad Azam, Mushtaq Ahmad, Muhammad Naeem, Member, IEEE, Muhammad Iqbal, Ahmed Shaharyar
Khwaja, Member, IEEE, Alagan Anpalagan, Senior Member, IEEE and Saad Qaiser, Senior Member, IEEE

AbstractDevice to device (D2D) communications can help


in achieving the higher data rate targets in emerging wireless
networks. The use of D2D communication imposes certain
challenges such as interference with the cellular and D2D users.
A well designed joint admission control, network mode selection,
and power allocation technique in a cellular network with
D2D capability can improve overall throughput. The proposed
technique jointly maximizes the total throughput and number
of admitted users in cellular networks under quality of service
and interference constraints. The joint admission control, mode
selection and power allocation problem falls into a class of
mixed integer non-linear constraint optimization problems which
are generally NP-hard. Due to the combinatorial nature of the
problem, its optimal solution needs exhaustive search of integer
variables whose complexity increases exponentially with the number of user pairs. In this paper, we invoke outer approximation
approach (OAA) based linearization technique to solve the joint
admission control, mode selection and power allocation problem
(JACMSPA). The proposed method gives guaranteed -optimal
solution with reasonable computational complexity. Simulation
results verify the effectiveness of the proposed approach method.
Index TermsDevice-to-Device communication, admission
control, mode selection.

I. I NTRODUCTION
Demand for higher data rates in future cellular networks is
being witnessed globally. Data hungry applications on mobile
devices such as multimedia downloading, video streaming, online gaming and large file sharing among users are forcing
cellular operators to adopt new technologies. These technologies allow the operators to satisfy customers demand of
enhanced data rates and increase their revenues. The policy
formulating entities and regulators around the globe are also
Manuscript received November 07, 2014; revised April 06, 2015 and July
04, 2015; accepted September 12, 2015.
Copyright (c) 2015 IEEE. Personal use of this material is permitted.
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to pubspermissions@ieee.org.
M. Azam, is with Foundation University Institute of Engineering and
Management Sciences, Islamabad, Pakistan.
M. Ahmad and M. Iqbal are with Department of Electrical Engineering,
COMSATS Institute of Information Technology, Wah Campus, Pakistan.
M. Naeem is with the Department of Electrical Engineering, COMSATS
Institute of Institute of Information Technology, Wah Campus, Pakistan
and with the Department of Electrical and Computer Engineering, Ryerson
University, ON, Canada. (email:mnaeem@rnet.ryerson.ca)
A. S. Khwaja and A. Anpalagan are with the Department of Electrical and Computer Engineering, Ryerson University, ON, Canada.
(email:alagan@ee.ryerson.ca)
S. Qaiser is with School of Electrical Engineering and Computer Science
(SEECS), NUST, Pakistan
This work was supported in part by NSERC Discovery Grants.

following this trend. Rightly realizing the importance of higher


data rates, the 3rd generation partnership project (3GPP) has
stressed the need to increase the bandwidth requirement of
IMT-Advanced1 systems up to 100 MHz by the incorporation
of new technological components [1]. The IMT-Advanced
systems promise to improve local area services through efficient utilization of scarce radio resources. Long term evolution
(LTE) technologies are designed to provide high data rates [2]
[4]. Ref. [1] presents the concept of D2D communication in
long term evolution advanced (LTE-A) to meet higher data
rate demands.
D2D communication in advanced cellular networks is a
very promising technique wherein user equipments share the
same radio resources used by cellular users to enhance the
throughput of the cellular systems. According to [5], different techniques are possible for the controlled management
of resources by the evolved node-B (eNB). Non-orthogonal
frequency sharing method is more suitable to enhance spectral
efficiency. A lot of work has been done through wireless local
area networks and wireless personal area networks. Although
technologies such as bluetooth and ultra wide band provide
higher data rates, they require manual peering and have no
control over interference. However, in D2D communication,
due to the controlled assignment of radio resources by the base
station (BS) or eNB in the licensed band, problems such as
manual peering and interference among users are alleviated
to a greater extent [1]. According to [5], benefits of D2D
communication include: reduction in end-to-end latency due
to reduced number of hops, higher bit rates due to proximity
of users, low power consumption thereby enhancing battery
life of devices, wireless cellular network evolving toward
the advanced and intelligent architectures to achieve better
network capacity, coverage and quality of service.
The use of D2D communication imposes certain challenges
such as interference with cellular and D2D users while reusing
the same radio resources with the cellular users. To achieve
higher data rate of future wireless networks while simultaneously satisfying the quality of service with power constrained
is a challenging task. A well designed joint admission control,
mode selection2 and power allocation (JACMSPA) technique
in cellular network with D2D capability can improve overall
throughput of cellular network. Various techniques have been
1 International Mobile Telecommunications-Advanced (IMT-Advanced) are
requirements issued by the the International Telecommunication Union (ITU)
for future generation mobiles phone and Internet access service.
2 Mode selection determines whether the user pair will communicate directly
in a point to point fashion or communicate with the help of cellular eNB.

0018-9545 (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/TVT.2015.2487220, IEEE Transactions on Vehicular Technology
2

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXXX 2015

TABLE I
C OMPARISON O F D IFFERENT R EFERENCES . AC = A DMISSION C ONTROL , MS = M ODE S ELECTION , PA = P OWER A LLOCATION
Ref

Admission
Control

[7]
[10]
[11]

Mode Selection
X
X
X

Power Allocation

[12]

[14]

[15]

[16]

JACMSPA

Algorithm

Remarks

Heuristic
Heuristic
Greedy

Mode selection without the knowledge of perfect CSI


Graph theory based resource allocation
Radio resource allocation based on greedy algorithm
and successive interference cancellation in Device-toDevice (D2D) communication
System throughput maximization for cellular network
with D2D capability.

Traditional convex optimization


techniques
Greedy scheme

[17]

[18]

[19]

[20]

[22]

Adaptive
Approximation
Algorithms
Robust
optimization
algorithm
Linear Programming
Convex Approximation
Heuristic

Greedy

Heuristic

Evolutionary Algorithm

[23]
[25]

[28]

[30]
[32]
[34]
[36]

X
X
X

X
X
X

Heuristic
Heuristic
Heuristic

suggested in the literature to enhance data rates of future


cellular networks. Currently, local services prefer to reuse
the spectrum to increase the system throughput. Unlike the
traditional cellular network, D2D can establish a direct link in
user equipments by sharing the same resource blocks with
other cellular users, thereby improving spectral efficiency.
However, cellular users experience interference from the D2D
transmitters. To control the interference level efficiently, eNB
will be in charge of assigning the resource blocks to the
D2D devices. Literature review and some design challenges
related to resource allocation and spectrum sharing between
cellular and D2D users to enhance the efficiency of network
is discussed in next section.
II. L ITERATURE R EVIEW
In [11], the authors propose a greedy algorithm and successive interference cancelation (SIC), which allows the coexistence of three pairs of D2D user equipments and cellular
users in a channel. SIC approach maximizes the performance
of D2D links and reduces the interference from D2D users
to cellular users. Fractional frequency reuse (FFR) is used
to reduce the interference with co-channel cells [12]. They

Empirical D2D results for underlay WINNNER II


project
Power optimization for orthogonal amd non-orthogonal
resource sharing modes between cellular and D2D communication.
Best-effort Successive Interference Cancelation algorithm, canceling interfering signals.
This admission control is without mode selection.

Utility Maximization and Admission Control for a


MIMO Cognitive Radio Network.
This admission control is without mode selection.
This admission control is without mode selection.
Distributed power control and set based admission control
Channel assignment and mode selection in OFDMA
based cellular network.
Subcarrier allocation for D2D multicast system with
OFDMA scheme.
GA based resource allocation in D2D network.
Coverage analysis and calculation of erotic capacity in
cellular D2D network.
A bipartite matching based resource allocation.
Resource allocation as a coalition formation game.
Resource allocation in cooperative two-way cellular
D2D network.

suggest that if user equipments are in outer cell region,


then D2D and cellular user equipments experience a tolerable interference. In [23], the authors investigate power and
channel allocation using a greedy method. In [13], the authors
suggest to improve the spectrum utilization and optimize the
performance by maximizing the weighted sum rate of D2D
and cellular users. Several efficient algorithms were analyzed
on the basis of complexity and overall performance. To optimize different parameters of future cellular networks, power
control is considered in [14], [22] whereas [15] examines both
admission and power control in D2D and cellular networks.
A distance dependent mode selection is investigated in [24].
In OFDMA based systems, in order to guarantee quality of
service (QoS), the authors in [25] propose a resource allocation
scheme while considering signal to interference plus noise
ratio (SINR) threshold value for cellular users. The authors
divides the scheme into two steps: In the first step, subcarriers
are assigned under the premise of ensuring the minimum data
rate of the D2D multicast groups and the second step involves
QoS of cellular users.
Literature review and some design challenges related to spectrum sharing between cellular and D2D users to enhance the

0018-9545 (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/TVT.2015.2487220, IEEE Transactions on Vehicular Technology
SHELL et. al:M. NAEEM

Fig. 1. A cellular system with D2D capability

energy efficiency are discussed in [26]. In [27], the authors


propose a new spectrum sharing protocol. It allows the D2D
user equipments to communicate bi-directionally with one
another while supporting two-way communications between
the BS and cellular users. In [28], the authors investigated
the resource reusing mechanism in more than one D2D pair
allowing to share the multiple resource blocks to obtain
higher sum-rates. Mode selection is also performed based
on evolutionary algorithm. In [29], the authors presented
distributed implementation. The sum-rate of D2D system is
maximized by maximizing the transmission rate using game
theory. An interference management strategy is proposed in
[30] to increase the overall capacity of cellular users and D2D
system. A conventional technique is used that limits maximum
transmit power of D2D transmitter, so that it does not generate
harmful interference from D2D user to cellular users. In D2D
communication, the decision to assign a part of resource to
downlink or uplink is very important. It should be done on
priority to enable the reuse of resources for higher system
capacity. Ref. [31] proposes interference control in downlink
mode to limit the BS interference to D2D user equipments
by selecting cellular users that are closer to BS. To improve
the QoS of both D2D and cellular users, a three-step scheme
is proposed in [32], which involves admission control and
subsequent power allocation for each selected D2D pair and its
potential cellular user partners. D2D needs intelligent resource
sharing as suggested in [33] to optimize the spectrum utilization. Ref. [34] investigates resource sharing issue to optimize
the system performance in D2D communication in cellular
networks from a cooperative and distributive perspective. In
this scheme, utility function is maximized for each user and
provides incentive to cooperate with other users to form a
strong group to increase the likelihood to win its preferred
spectrum resource.
A. Contributions
Based on the literature review and a closed look at Table I
reveal that a joint user, mode selection with power allocation is

TABLE II
N OTATIONS
Symbol
K
c
d
R
Ckul
Ckdl
Ckc
Ckd
max
PeNB
max
Pc,k
min
Pd,k
pdk
pul
k
pdl
k
xk
Ckmin
hk
fk
gk
c
d

do
d
k
h

U
US
UT
pc
pm
ps

Definition
Set of user pairs
Cellular mode
D2D mode
Radius of D2D transmission
Capacity of the kth cellular user in uplink
Capacity of the kth cellular user in downlink
Capacity of the kth cellular user pair-i.e.,
1
min(Ckul , Ckdl )
2
Capacity of the kth D2D users
Maximum power of eNB
Maximum power of the kth user in cellular mode
Minimum power threshold of the kth user in D2D mode
outside radius R
Power of the kth in D2D mode
Power of the kth in uplink cellular mode
Power of the kth in downlink cellular mode
Binary indicator for D2D or cellular mode
Minimum rate requirement of the kth user
Channel gain between the kth user and eNB link
(uplink)
Channel gain between the kth receiver and eNB link
(downlink)
Channel gain between the kth D2D user pair
Set of selected users in cellular mode
Set of selected users in D2D Mode
Set of all selected Users-i.e.,c d
Reference distance for the antenna far field
Distance between secondary transmitter and receiver
Rayleigh random variable associated with the kth SU
Log normal shadowing
A utility function to maximize the joint user admission
and throughput
A utility function to maximize the admitted users
A utility function to maximize the throughput
Probability of cross over
Probability of mutation
Probability of selection

an open area of research. There are some papers in the literature that claim joint admission control and power allocation but
they always try to solve the user selection and power allocation
separately. Since the user selection and power allocation are
not separable and in addition to joint user selection and power

0018-9545 (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/TVT.2015.2487220, IEEE Transactions on Vehicular Technology
4

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXXX 2015

allocation, we are also jointly determining the mode selection.


In [17][21], authors proposed different joint admission and
power control schemes for wireless systems. All these work
do not include the mode selection in their formulation. As
per the best knowledge of the authors, the literature review
reveals that there is no throughput maximization scheme in
cellular network with D2D capability that jointly controls
the admission of the users, mode selection (whether to be
in cellular or D2D mode) and power allocation under QoS
and interference constraint. The existing work focuses on
individual aspects as shown in the self explanatory Table
I. The closest possible existing work is done in [28]. The
authors apply genetic algorithm (GA) for admission control
and mode selection. The main difference between [28] and our
formulation is joint admission control and mode selection. In
[28], authors proposed separate admission control and mode
selection where as in this paper, we propose a joint admission
control and mode selection scheme. For a fair comparison,
in this paper, we also compare the results of GA with our
proposed algorithm. The scope of the proposed work fills
the gap and tries to maximize the throughput considering
JACMSPA mechanism. The main contribution of this paper
are summarized as follows:
1) We formulate a constrained JACMSPA optimization problem that maximizes the overall throughput of the future
cellular networks by jointly controlling admission of
the users, satisfying mode selection and power control
constraints of the users and base station.
2) The JACMSPA is a class of mixed integer non-linear constraint optimization problems, which are generally NPcomplete. Due to the combinatorial nature of JACMSPA,
the optimal solution needs exhaustive search of integer
variables whose complexity increases exponentially with
the number of user pairs. In this paper, we apply a
linearization technique that uses outer approximation approach (OAA) to solve this NP-complete JACMSPA. The
proposed OAA gives guaranteed -optimal results with
finite convergence. In -optimal solution, for any > 0,
the -optimal algorithm gurantee the solution within of
the optimal. We also compare the OAA algorithm with
genetic algorithm. The results show that performance of
OAA is much better than GA.
3) The -optimal solution is analysed in detail with the help
of simulation results.
Throughout this paper, we use A, a, and a to represent
matrix, vector and an element of a vector respectively. This
paper is organized as follows: The system model and problem
statement are described in section III. In section IV, we
present JACMSPA solution technique using OAA method. The
simulation results are analyzed in section V and finally we
conclude this paper in section VI.
III. S YSTEM M ODEL AND P ROBLEM F ORMULATION
We consider a cellular network having D2D communication
capability as shown in Fig. 1. There are two possible modes
of communication, (a) cellular mode and (b) D2D mode. We
assume that the users selected for D2D communication use

non-orthogonal sharing mode. In this mode, the D2D users


will re-utilize cellular network resources. Let K be the set of
user pairs that want to communicate with one another. We
dl
d
denote by pul
k , pk and pk , the kth users transmitted power in
uplink (to eNB), the kth users transmitted power in downlink
(to the kth receiver) and the kth user power in D2D mode.
The channel gain between the kth user pair is gk . We denote
hk and fk as the channel gains between transmitting user-eNB
link (uplink) and eNB- receiving user (downlink) respectively.
o

Let the antenna gain be Go and = 10 10 be the log normal


shadowing, where o is zero mean Gaussian random variable
with standard deviation [37]. The channel hk is modeled as
[39]:
 
do

,
(1)
hk = hk Go
d
where do and d are the antenna far field reference distance and
distance between the receiver and transmitter, respectively. The
k is the Rayleigh
path loss exponent is denoted by and h
random variable. The channel capacity of the kth user is
k hk
defined as Ck = log(1 + pN
). A summary of symbol
0
notations is shown in Table II. In cellular mode, the eNB will
act like a relay and the communication between the cellular
user pair needs two time slots. The possible rate
 of the kth
user in cellular mode is Ckc = 12 min Ckul , Ckdl . The rate for
pd g

k k
). For
kth pair in D2D communication is Ckd = log(1 + N
0
mode selection, we define a binary mode selection indicator
as:

1, Cellular mode
xk =
0, D2D mode.

To meet the QoS of the kth user pair, the pair must satisfy
its minimum rate Ckmin . For any power constraint, wireless
network satisfying every users rate requirement is not always
possible3 . Traditional admission control schemes generally
select the users that can give higher aggregate throughput. In
this paper, we propose a framework for joint admission control
and mode selection that not only maximizes the throughput
but also maximizes the number of admitted users under the
minimum rate and power constraints. Let be the set of
admitted users. One admitted user can only be served either
in cellular or D2D mode and the set of admitted users is
the union of admitted cellular and D2D users. Mathematically,
this is written as:
= c d
c d =

(2)

We introduce a utility function that maximizes admitted users


and throughput as:
X


dl
U , x, pd , pul , pdl = US ()
UT xk , pdk , pul
(3)
k , pk
k


||
dl
and UT xk , pdk , pul
= xk Ckc +
where US = |K|
k , pk

(1 xk ) Ckd . The utility function U , x, pd , pul , pdl ensures that if cellular mode is selected for any admitted user,
3 There are a number of reasons for this-e.g., channel may be bad, battery
is low, increase in power may increase interference to others etc.

0018-9545 (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/TVT.2015.2487220, IEEE Transactions on Vehicular Technology
SHELL et. al:M. NAEEM

then the terms related to D2D mode should be zero and


vice versa. Mathematically, we can write the joint admission
control, mode selection and power allocation constrained optimization problem as:

max
U , x, pd , pul , pdl
,x,pd ,pul ,pdl

subject to


min
dl
, k
C1 : UT xk , pdk , pul
k , pk Ck
min
C2 : pdk (1 xk )R Pd,k
, k
max
C3 : pul
k xk Pc,k , k
X
max
C4 :
xk pdl
k PeN B

(4)

C5 : = c d

C6 : c d =
C7 : xk {0, 1}, k K

Theorem 1. JACMSPA for D2D cellular network is NPcomplete.


Proof : The proof is given in the Appendix A.
Due to NP-complete nature of JACMSPA, determining optimal solution in polynomial time is not possible. An exhaustive
search algorithm for (5) would enumerate all the users and
mode selection options, which increases its computational
load exponentially with the number of users. The structure of
optimization problem in (5) is very interesting. With known
discrete variables, the objective function of (5) is a concave
function in power, and all the constraints are either linear or
convex. By exploiting this special structure, in the next section,
we will present a branch and bound based outer approximation
algorithm (OAA) to solve (5).
IV. P ROPOSED A PPROACH

TO A

S OLUTION

As mentioned in the previous section, the coupling of intedl

0
,
k

K
C8 : pdk 0, pul

0,
p
ger
domain, with the continuous domain and non-linearities in
k
k
the problem make the class of problem mentioned in (5) very
The objective function in (4) is a max-min problem. By
challenging. As the integer variables (user admission and mode
introducing a new variable tk , k K, we can rewrite an
selection in our case) increase, the complexity analysis results
equivalent maximization problem as
tend towards NP-completeness. Despite all these challenges,

d
ul
dl
the optimization problem in (5) has a very special structure. By
max
Ut t, , x, p , p , p
t,,x,pd ,pul ,pdl
exploiting this special structure, in this section we will present
an OAA method to solve (5). The OAA solves (5) in a finite
subject to
sequence of alternately nonlinear programming sub-problems
(5)
C1-C8 of (4)
(by fixing the discrete variables and x) and relaxation of a
C9 : Ckul tk , k c
mixed integer linear program based master problem (MILPC10 : Ckdl tk , k c
MP). The solution of the sub-problem provides a point that
will generate supporting hyperplanes of the objective and
where
constraint functions. The OAA method adds one linearization
X


d
ul
dl
d ul dl
for each constraint and the objective function for every subUt t, , x, p , p , p
= US ()
UT,t t, xk , pk , pk , pk
problem. These linearizations of the problem are collected in
k
(6) a MILP-MP. The solution of the master program determines a
and
new set of discrete variables that will be used for next iteration

[40].
d
dl
(7)
UT,t t, xk , pdk , pul
k , pk = xk tk + (1 xk ) Ck

The utility function in the above optimization problem maximizes the admitted users and total throughput. Constraint
C1 is the minimum rate constraint of each user. If any user
can not meet the rate constraint condition, then that user is
not selected for transmission. Constraints C2, C3 and C4
are power constraints for D2D, cellular uplink and cellular
downlink mode respectively. Constraint C2 ensures that in
case of D2D communication, the power experienced by any
other cellular or D2D user beyond radius R should be less
min
than a threshold power Pd,k
. Constraint C3 is the uplink
power constraint and C4 is the downlink sum power constraint.
Constraints C2, C3 and C4 also ensure that the respective
power should be zero if respective mode is not selected.
Constraints C5 and C6 ensure that D2D and cellular users
are mutually exclusive.
The formulation in (5) is a non-convex mixed integer nonlinear programming problem. To prove the hardness of (5), we
can reduce the multiple-choice multiple-dimensional knapsack
problem to the JACMSPA optimization problem.

A. Algorithm description

Let us denote U as the set of constraints C1 to C10 in (5),


P = {t, pd , pul , pdl } and = x. We can easily prove
that (5) satisfies the following propositions:
Proposition 1. P is compact, nonempty, convex set and the
objective function Ut and U are convex in P for fixed values
of .
Proposition 2. Ut and U are once continually differentiable.
Proposition 3. A constraint qualification holds at the solution
of each nonlinear continuous sub-problem obtained by fixing
the values of .
Proposition 4. The nonlinear programming problem obtained
by fixing can be solved exactly.
These propositions make the problem in (5) a special class
of problems that can be solved using OAA method [38]. The
convexity of (5) ensures that the linearization of constraint and
objective function produces outer approximation. Although

0018-9545 (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/TVT.2015.2487220, IEEE Transactions on Vehicular Technology
6

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXXX 2015

Algorithm 1 : OAA
1: j 1
2: Initialize j
3: 103
4: Convergence F ALSE
5: while Convergence
== F ALSE do

(
arg min
Ut j , P
j
P
6:
P

subject to U j , P 0;
7:
UpperBound
Ut j , P
arg min

,P,

subject to

j
j

,
P


t
8:
( , P , )

j

Ut j ,P j PP

U j , P j


j

0
U j , P j PP
0
9:
LowerBound
10:
if UpperBound LowerBound then
11:
Convergence T RU E
12:
else
13:
j j+1
14:
j
15:
end if
16: end while

as:

min min Ut j , P

P

subject to:
U j , P 0;

We can also write (9) as

min ()

(10)

where

() = min Ut j , P
P

subject to
U j , P 0

(11)

The problem (10) is the projection of (5) on space. Since


a constraint qualification holds at the solution of every primal
problem (8) for every j , the projection problem has the same
solution as the problem below:



 P Pj
j
j
j
j
min min Ut , P Ut , P

P
j
subject to
(12)


j


P P
0
U j , P j U j , P j
j

By introducing a new variable , we can rewrite an equivalent minimization problem as:

min

(8)

Solution of this problem will give the P j that will be used


for master problem. Primal problem gives the upper bound
and the master problem will give the lower bound. The master
problem is derived with the help of primal solution, i.e., P j
and is based upon linearization (outer approximation) of the
non-linear objective Ut and constraints U around the primal
solution P j [41], [42]. The solution of the master problem
provides the information for the next set of integer variables
j+1 . As the iteration proceeds, these two bounds come close
to each other. The algorithm will terminate when the difference
between the two bounds is less than . The master problem is
derived in two steps: In the first step, we need projection of
(5) onto the integer space-. We can rewrite the problem (5)

,P,

subject to


 P Pj

Ut j , P j Ut j , P j
j



 P Pj
0
U j , P j U j , P j
j

(13)

10

OAA
ESA

10
Sum Rate (b/s)

the algorithm proposed in the later section is also applicable


to non-convex objectives, it can give local optimal solution
instead of global optimal solution. Proposition 3 is useful
as many nonlinear programming solvers use Kuhn-Tucker
conditions, which require a constraint qualification to hold.
The OAA uses sequence of non-increasing upper and nondecreasing lower bounds for mixed integer problems that
satisfy the propositions 1 to 4. The OAA converges in a finite
number of iterations with -convergence capability [40]. The
sequence of upper and lower bounds are obtained by solving
primal and master problem, respectively. The primal problem
is obtained by fixing variables. At the jth iteration of OAA,
let the values of integer variable be j . We can write the
primal problem as:

min Ut j , P
P

subject to:
U j , P 0.

(9)

7.09

10
6

10

7.08

10

6.01 6.02 6.03


Number of users

10

5
Number of users

Fig. 2. Performance of OAA with exhaustive search algorithm (ESA) for


different number of users

This is the master problem used to generate lower bound.


Under the propositions 1, 2 and 3, (13) is equivalent to (5).
Problem (13) is now a mixed integer linear programming
problem and can be solved using iterative framework. A
pseudo code for OAA is given in Algorithm 1.

0018-9545 (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/TVT.2015.2487220, IEEE Transactions on Vehicular Technology
SHELL et. al:M. NAEEM

10

OAA Rate Threshold = 100kbps


OAA Rate Threshold = 200kbps
GA Rate Threshold = 100kbps
GA Rate Threshold = 200kbps

OAA Rate Threshold = 1Mbps


OAA Rate Threshold = 2Mbps
GA Rate Threshold = 1Mbps
GA Rate Threshold = 2Mbps

Sum Rate (b/s)

Sum Rate (b/s)

10

10

6
7
Number of users

max =
(a) Performance of OAA for different number of users with PeNB
max = 0.5W .
4W, Pc,k

35

10

10

10

15

GA
OAA

Rate Req
= 25kb/s
Spectral Efficiency(b/s/Hz)

25

20

15

10

30

OAA K = 4
GA K = 4
OAA K = 8
GA K = 8

12

10
Spectral Efficiency(b/s/Hz)

25

max =
(b) Performance of OAA for different number of users with PeNB
max = 2W .
20W, Pc,k

30

Rate Req
= 50kb/s

Rate Req
= 100kb/s
Rate Req
= 150kb/s

Rate Req
= 200kb/s

20
Number of users

10
15
Number of Users

20

(c) Performance of OAA and GA for different number of users

(d) Performance of OAA and GA for different data rates

Fig. 3. Comparison of OAA and GA

B. Discussion on Algorithm optimality and convergence


If the problem holds all four prepositions and the discrete
variables () are finite, then the Algorithm 1 terminates in
a finite number of steps at an -optimal solution [40]. As
mention earlier, in an -optimal solution, for any > 0, optimal algorithms guarantee the solution within of optimal
solution. Lower values of mean high degree of accuracy.
The main reason for -optimal solution of OAA is its branch
and bound like architecture. In branch and bound procedure,
any combination of discrete variable ( in our case, which
is the union of users and mode selection variable) will never
be used twice. The optimality
of P in (13) implies that

point in (13). If
is greater than Ut j , P j for any feasible

the value of is less than Ut j , P j , it means that the
master problem has no feasible solution for the choice of
discrete variables j . If there is no feasible solution for any
particular j in (13), then the algorithm excludes that value
of j from any subsequent master problems. It means that
the algorithm is finitely converging. The optimality of the

algorithm follows from the convexity of the objective and


constraints for any fixed values of . It is proven in the
mixed integer programming literature that the convergence
rate of OAA is linear [42]. A detailed convergence proof of a
general OAA algorithm is given in [38]. An exhaustive search
algorithm (ESA) for (5) would enumerate all the user and
mode selection options, which increases its computational load
exponentially with the number of users. In this paper, we have
compared the results of the proposed OAA with ESA and GA.
The simulation results show that the performance of OAA is
almost same as optimal solution obtained and ESA and better
than GA. One more advantage of OAA is its guaranteed optimal results. The problem with the GA is that it cannot
guarantee any optimal or -optimal solution. GA can give good
result but there is no proven convergence criterion for GA
where as OAA has proven convergence to -optimal solution.

0018-9545 (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/TVT.2015.2487220, IEEE Transactions on Vehicular Technology
8

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXXX 2015

TABLE III
S IMULATION PARAMETERS

10

Parameter
max
PeNB
max
Pc,k
min
Pd,k
Ckmin
do
eNB coverage
d
o
Go
ps
pm
pc

UT

10

Rate of each user (b/s)

10

10

10

10

10

Value
{2,4}Watts
{0.5,0.75}Watts
100mW
{100,200}kbs
20
1000m
Uniformly distributed distance
10dB
50
0.90
0.10
0.50

10

3
4
Users Index

V. S IMULATION R ESULTS

max = 4W, P max = 0.5W and


(a) Comparison of U and UT with PeNB
c,k
min
= 100kbps. Sum-rate for U and UT is same.
Ck

10

U
UT
6

10

Rate of each user (b/s)

10

10

10

10

10

10

5
6
Users Index

10

max = 2W, P max = 0.5W and


(b) Comparison of U and UT with PeNB
c,k
min
= 200kbps. Sum-rate for U = 1.68Mbps and UT = 4.33Mbps
Ck

10

U
UT
6

10

Rate of each user (b/s)

10

10

10

10

10

10

5
6
Users Index

10

max = 4W, P max = 0.5W and


(c) Comparison of U and UT with PeNB
c,k
min
= 200kbps. Sum-rate for U = 2.574Mbps and UT = 3.2Mbps
Ck

Fig. 4. Comparison of U and UT

In this section, we show simulation results to demonstrate


the performance of the proposed OAA scheme. The results
show the effect of number of users on the total throughput
and also analyze the effectiveness of joint admission control
and mode selection utility. The system parameters used in the
simulation are shown in Table III. We use Basic Open Source
Nonlinear Mixed Integer Programming (BONMIN) software
for OAA. We compare the results of OAA with standard
continuous GA [43].
In the simulation, the eNB maximum coverage is set to 1000
max
meters. The maximum eNB power PeN
B is set to {2, 4} Watts.
The coverage distance for D2D is 20 meters. In cellular mode
max
Pc,k
= {0.5, 0.75} Watts. In all the simulation results, do =
20 meters, Go = 50 and o = 10dB. We assume that distance
d is greater than do . For convergence, is set to 0.00001. In all
simulations, for GA, probability of cross over, pc , probability
of mutation, pm and probability of selection, ps are set to 0.9,
0.1 and 0.5 respectively. Elitist selection and uniform cross
over rules with global best value adaptation are used to get
high quality GA solutions. We apply GA for discrete variables
(user and mode selection), and for each realization of discrete
variables a conventional convex optimization algorithm is used
to get the optimal power allocation. In the simulation results,
U represents a utility function that jointly maximizes the
throughput and users admission considering mode selection
where as UT represents a utility function that only maximizes
the throughput considering mode selection.
In Fig. 2, we compare results of optimal solution obtained
by ESA and OAA for different number of users. For ESA
approach, we need to 22K enumeration of discrete variables4
and for each realization of discrete variable, there is a need
to solve one non-linear convex optimization problem. The
computational complexity of ESA increases exponentially.
Thats makes ESA unsuitable for such kind of problems. The
result in Fig. 2 shows that the performance of OAA is almost
similar to the optimal ESA algorithm. To get one result using
ESA required a lot of time, for brevity, we only present one
result of ESA in this paper.
Figs. 3(a), 3(b), 3(c) and 3(d) investigate the performance of
OAA and GA for joint user and throughput maximization on
4K

variables for admission control and K variables for mode selection

0018-9545 (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/TVT.2015.2487220, IEEE Transactions on Vehicular Technology
SHELL et. al:M. NAEEM

10

10

K=5
K = 10
K = 15
K = 20

U
UT
6

10

10
Sum Rate (b/s)

Rate of each user (b/s)

10

10

10

10

10

10

10

10

15
Users Index

20

25

30

10
100

200

300

400
500
600
700
Coverage Distance (m)

800

900

1000

max = 4W, P max = 0.75W and


(a) Comparison of U and UT with PeNB
c,k
Ckmin = 100kbps. Sum-rate for U = 7.46Mbps and UT = 19.6Mbps

Fig. 6. Coverage distance versus Sum throughput comparison for different


number of users.
7

10

U
UT
6

10

Rate of each user (b/s)

10

10

10

10

10

10

10

15
Users Index

20

25

30

max = 4W, P max = 0.5W and


(b) Comparison of U and UT with PeNB
c,k
Ckmin = 200kbps. Sum-rate for U = 6.18Mbps and UT = 6.38Mbps

10

U
7

UT

10

Rate of each user (b/s)

10

10

10

10

10

10

10

10
12
Users Index

14

16

18

20

max = 20W, P max = 2W and


(c) Comparison of U and UT with PeNB
c,k
min
= 1M ps. Sum-rate for U = 123Mbps and UT = 110.41Mbps
Ck

Fig. 5. Comparison of U and UT

the cellular network. Fig. 3(a) shows a plotnof sum-rate versus


o
max
max
=
the number of users with the parameters PeN
,
P
B
c,k
{4, 0.5} and {20, 2} watts respectively. The results present the
comparison of different rate thresholds for a power constraint
cellular network with D2D capability using OAA and GA.
As there is an increase in the rate requirement, for same
constraints, the sum-rate decreases. This is because less number of users are admitted due to high rate requirement and
stringent power constraint in D2D mode. In Fig. 3(c), we
present the comparison results of OAA and GA in terms of
spectral efficiency and in Fig. 3(d), we show the OAA and
GA results for different data rates. All the results highlight the
effectiveness of OAA over GA. This is because of -optimal
nature of OAA algorithm. Due to stochastic nature of GA,
there is no guarantee that the GA will give optimal solution.
Figs. 4(a), 4(b) and 4(c) investigate the effect of joint user
and throughput maximization (in log scale) on the cellular
network. These figures compare the throughput obtained by
users using U and UT utilities. The straight horizontal line
shows the rate requirement. In Fig. 4(a) with parameters
max
max
PeN
= 0.75 watts and Ckmin = 200kbps, the
B = 2W, Pc,k
performance both utilities is almost same. Sum-rate for U and
UT is same. In the figure, fifth user is not selected due to bad
channel conditions.
max
For Figs. 4(b) and 4(c), we set the parameters as PeN
B =
max
min
max
2W, Pc,k
= 0.5W, Ck
= 100kbps and PeN B =
max
4W, Pc,k
= 0.5W, Ckmin = 200kbps, respectively. In Fig.
4(b), we can see that six users are admitted with U utility
while only two users are selected for UT utility. Sum-rate for
U = 1.68 Mbps and UT = 4.33 Mbps. Similarly for Fig. 4(c)
four users are admitted with U utility while three users are
selected for UT utility. Sum-rate for U = 2.574 Mbps and UT
= 3.2 Mbps. We can see that although sum-rate for UT is high,
it is at the cost of low number of admitted users.
Figs. 5(a), 5(b) and 5(c) compare the throughput obtained
by users using U and UT utilities for 30 users with param-

0018-9545 (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/TVT.2015.2487220, IEEE Transactions on Vehicular Technology
10

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXXX 2015

max
max
eters as PeN
= 0.75W, Ckmin = 100kbps,
B = 4W, Pc,k
max
max
max
PeN B = 4W, Pc,k = 0.5W, Ckmin = 200kbps and PeN
B =
max
min
20W, Pc,k = 2W, Ck = 1Mbps, respectively. Fig. 6 shows
the effect of eNB coverage area the throughput. In Fig. 5(a),
21 users are admitted with U utility while only 18 users are
admitted with UT utility. With the increase in rate requirement
for Fig. 5(b), 17 users are admitted with U utility while only
13 users are admitted with UT utility. We can conclude from
the results that for cellular system with D2D capability, joint
admission control and mode selection add more fairness as
compared to sum-rate maximization.

VI. C ONCLUSION
In this paper, we presented a computationally viable solution
for solving the joint admission control, mode selection and
power allocation problem in D2D communication. We made
use of the special structure of this problem to propose a
solution based on branch and bound outer approximation
approach. The proposed OAA method gives guaranteed optimal results and has a reasonable computational complexity.
We verify the effectiveness of the proposed approach method
by simulations that demonstrated the effect of number of users
on the total throughput and also analyzed the effectiveness of
joint admission control and mode selection utility.
R EFERENCES
[1] K. Doppler, M. Rinne, C. Wijting, C. B. Ribeiro and K. Hugl, Deviceto-device communication as an underlay to LTE-advanced networks,
IEEE Communications Magazine, vol. 47, no. 12, pp. 42-49, Dec. 2009.
[2] N. Dan, B. Li, B. Lan and C. JunRen, Resource allocation over cooperation for cross-cell D2D communication underlaying LTE network,
In Proc. of IEEE TENCON 2013, Oct. 2013.
[3] Y.-S. Chen and J.-S. Hong, Intelligent Device-to-Device Communication in the Internet of Things, IEEE Systems Journal, vol. 7, no. 1, pp.
77-91, March 2013.
[4] S.-H. Lu, L.-C. Wang, T.-T. Chiang and C.-H.Chou, Cooperative Hierarchical Cellular Systems in LTE-A Networks, IEEE Systems Journal,
2014.
[5] K. Akkarajitsakul, P. Phunchongharn, E. Hossain and V. K. Bhargava,
Mode selection for energy-efficient D2D communications in LTEadvanced networks: A coalitional game approach, In Proc. of IEEE
Communication Systems (ICCS), Nov. 2012.
[6] O. Bello and S. Zeadally, Intelligent Device-to-Device Communication
in the Internet of Things, IEEE Systems Journal, 2014.
[7] J. Gao, X. Liao, J. Deng and P. Ren, A Mode Shifting Resource Allocation Scheme for Device-to-Device Underlaying Cellular Network, in
Proc. AASRI on Parallel and Distributed Computing and Systems, vol.
5, pp. 40-47, 2013.
[8] B. Peng, C. Hu, T. Peng and W. Wang, Optimal Resource Allocation
for Multi-D2D Links Underlying OFDMA-Based Communications, In
Proc. of Wireless Communications, Networking and Mobile Computing
(WiCOM), Sept. 2012.
[9] L. Bao Le, Fair Resource Allocation for Device-to-Device Communications in Wireless Cellular Networks, In Proc. of IEEE Global
Communications Conference (GLOBECOM), Dec. 2012.
[10] H. Zhang, T. Wang, L. Song and Z. Han, Graph-based resource
allocation for D2D communications underlaying cellular networks, In
proc. of Communications in China - Workshops (CIC/ICCC), pp. 187192, Aug. 2013.
[11] Y. Tao, J. Sun and S. Shao, Radio resource allocation based on greedy
algorithm and successive interference cancellation in Device-to-Device
(D2D) communication, In Proc. of IET International on Information
and Communications Technologies (IETICT), April 2013.
[12] B. Wang, L. Chen, X. Chen, X. Zhang and D. Yang, Resource Allocation Optimization for Device-to-Device Communication Underlaying
Cellular Networks, In Proc. of IEEE on Vehicular Technology , May
2011.

[13] J. Wang, D. Zhu, H. Zhang, C. Zhao, C. F. James Li and M. Lei


Resource optimization for cellular network assisted multichannel D2D
communication, ELSEVIER Signal-Processing, vol. 100, pp. 23-31, July
2014.
[14] C.H. Yu, K. Doppler, C. B. Ribeiro and O. Tirkkonen, Resource
Sharing Optimization for Device-to-Device Communication Underlaying
Cellular Networks, IEEE Trans. on Wireless Communications, vol. 10,
no. 8, pp. 2752-2763, August 2011.
[15] C.H. Yu, O. Tirkkonen, K. Doppler and C. Ribeiro, Power Optimization
of Device-to-Device Communication Underlaying Cellular Communication, In Proc. IEEE ICC, June 2009.
[16] C.H. Yu and O. Tirkkonen, Device-to-Device underlay cellular network
based on rate splitting, In Proc. of IEEE Wireless Communications and
Networking Conference (WCNC), April 2012.
[17] E. Matskani, N.D. Sidiropoulos, Z.-Q. Luo and L. Tassiulas, Efficient
Batch and Adaptive Approximation Algorithms for Joint Multicast
Beamforming and Admission Control, IEEE Transactions on Signal
Processing, vol. 57, no. 12, pp. 4882 - 4894 Dec. 2009.
[18] H. Du and T. Ratnarajah, Robust Utility Maximization and Admission
Control for a MIMO Cognitive Radio Network, IEEE Transactions on
Vehicular Technology, vol. 62, no. 4, pp. 1707 - 1718 , May 2013.
[19] Y.-F. Liu, Y.-H. Dai and Z.-Q. Luo , Joint Power and Admission
Control via Linear Programming Deflation, IEEE Transactions on
Signal Processing, vol. 61, no. 6, pp. 1327 - 1338, March, 2013.
[20] X. Gong, S.A. Vorobyov, and C. Tellambura, C. , Joint Bandwidth
and Power Allocation With Admission Control in Wireless Multi-User
Networks With and Without Relaying, IEEE Transactions on Signal
Processing, vol. 59, no. 4, pp. 1801 - 1813, April 2011.
[21] E. Matskani, N.D. Sidiropoulos, Z.-Q. Luo and L. Tassiulas, Convex
approximation techniques for joint multiuser downlink beamforming and
admission control, IEEE Transactions on Wireless Communications,
vol.7, no.7, pp.2682,2693, July 2008.
[22] P. Liu, C. Hu, T. Peng, R. Qian and W. Wang, Admission and power
control for Device-to-Device links with quality of service protection in
spectrum sharing hybrid network, In Proc. of IEEE Personal Indoor
and Mobile Radio Communications, Sept. 2012.
[23] Y. Cheng, Y. Gu and X. Lin, Power and Channel Allocation for
Device-to-device Enabled Cellular Networks, Journal of Computational
Information Systems, vol. 10, no. 2, pp. 463-472 January 2014.
[24] S. Xiang, Tao Peng, Ziyang Liu and Wenbo Wang, A distancedependent mode selection algorithm in heterogeneous D2D and IMTAdvanced network, In Proc. of IEEE Globecom Workshops (GC Wkshps), Dec. 2012.
[25] P. Bo, H. Chunjing, P. Tao, Y. Yang and W. Wenbo, A resource
allocation scheme for D2D multicast with QoS protection in OFDMAbased systems, In Proc. of IEEE Personal Indoor and Mobile Radio
Communications, Sept. 2013.
[26] G. Fodor, E. Dahlman, G. Mildh, S. Parkvall, N. Reider, G. Miklos
and Z. Turanyi, Design aspects of network assisted device-to-device
communications, IEEE Communications Magazine, vol. 50, no. 3, pp.
170-177, March 2012.
[27] P. Yiyang and L.Y.Chang, Resource allocation for device-to-device
communication overlaying two-way cellular networks, In Proc. of IEEE
Wireless Communications and Networking Conference (WCNC), April
2013.
[28] H. Pang, P. Wang, X. Wang, F. Liu and N. N. Van, Joint Mode Selection
and Resource Allocation Using Evolutionary Algorithm for Deviceto-Device Communication Underlaying Cellular Networks, Journal of
communications, vol. 8, no. 11, Nov. 2013.
[29] W. Si, Z. Xiaoyue, L.Zhesheng, Z. Xin and Y. Dacheng, Distributed
resource management for Device-to-Device (D2D) communication underlay cellular networks, In Proc.of IEEE Personal Indoor and Mobile
Radio Communications (PIMRC), Sept. 2013.
[30] 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 Trans. Wireless Communications, vol. 10, no.
12, pp. 3995-4000, Dec. 2011.
[31] R. Chen, X. Liao, S. Zhu and Z. Liang, Capacity analysis of Deviceto-Device resource reusing modes for cellular networks, in Proc. IEEE
Communication, Networks and Satellite (ComNetSat), pp. 64-68, July
2012.
[32] D. Feng, Lu Lu, Yi Yuan-Wu, Li G.Y., Gang Feng and Shaoqian Li,
Device-to-Device Communications Underlaying Cellular Networks,
IEEE Trans. Communications, vol. 61, no. 8, pp. 3541-3551, August
2013.

0018-9545 (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/TVT.2015.2487220, IEEE Transactions on Vehicular Technology
SHELL et. al:M. NAEEM

11

[33] J. Wang, D. Zhu, C. Zhao, J.C.F. Li and M. Lei, Resource Sharing of


Underlaying Device-to-Device and Uplink Cellular Communications,
IEEE Communications Letters, vol. 17, no. 6, pp. 1148-1151, June 2013.
[34] R. Zhang, L. Song, Z. Han, X. Cheng and B. Jiao, Distributed resource
allocation for device-to-device communications underlaying cellular
networks, In Proc. of IEEE Communications (ICC), June 2013.
[35] H. S. Chae, J. Gu, C. Bum-Gon and M. Chung, Radio resource allocation scheme for device-to-device communication in cellular networks
using fractional frequency reuse, In Proc. of IEEE on Communications
(APCC), Oct. 2011.
[36] Y. Pei and Y.C. Liang, Resource Allocation for Device-to-Device
Communications Overlaying Two-Way Cellular Networks, IEEE Trans.
Wireless Communications, vol. 12, no. 7, pp. 3611-3621, July 2013.
[37] M. F. Hanif and P. J. Smith, On the Statistics of Cognitive Radio Capacity in Shadowing and Fast Fading Environments, IEEE Transactions
on Wireless Communications, vol. 9, no. 2, pp. 844-852, Feb. 2010.
[38] M.A. Duran and I.E. Grossmann, An Outer-Approximation Algorithm
for a Class of Mixed-Integer Non-linear Programming, Mathematical
Programming, vol. 36, pp.307-339, 1986.
[39] A. Goldsmith, Wireless Communications, Cambridge University Press,
2005.
[40] R. Fletcher and S. Leyffer, Solving mixed integer nonlinear programs
by outer approximation, Mathematical Programming, vol. 66, pp.327349, 1994.
[41] C. A. Floudas and P.M. Pardalos, Encyclopedia of Optimization, 2nd
ed., Springer, 2008.
[42] C. A. Floudas, Nonlinear and Mixed-Integer Optimization: Fundamentals and Applications, Oxford University Press, 1995
[43] A.E. Eiben and J.E. Smith, Introduction to Evolutionary Computing,
Springer Verlag, 2003
[44] H. Kellerer, U. Pferschy and David Pisinger, Knapsack Problems,
Springer, 2004.

A PPENDIX A
P ROOF OF T HEOREM 1
We prove that even with known uplink and downlink powers
the joint admission control, network mode selection, and
power allocation is a NP-complete problem. One example
of known power is equal power distribution among selected
users. We first show that the JACMSPA is equivalent to the
0-1 multiple-choice multiple-dimensional knapsack problem
(MCMDKP). We will start by describing the input, output formal description of decision problem associated with
MCMDKP and JACMSPA.
Problem 1. The MCMDKP problem is to select the items
xj,q in disjoint classes to maximize the total profit such that
an item can only be selected by at most one class subject to
satisfaction of W resource constraints.

Instance: The number of users K, rate threshold Rk , k =


1, 2, , K; channel gains hk , fk , gk of the kth user, and
arbitrary users power.
Output: An selection of users and their respective mode.
Decision problem associated with JACMSPA: The decision
problem associated with the JACMSPA is to determine, for a
given throughput C, whether it is possible to select multiple
users multiple mode ( cellular or D2D) such that the rate of
each selected users with any specific mode is more than R.
Lemma 1. The JACMSPA problem is polynomial time verifiable.
Proof. We can easily observe that if we are given a set of
selected users which represents the mapping between D2D
and cellular mode, we can verify in polynomial time that the
total throughput of the selected users is at least C and the rate
of each selected user is more than R. Since the dimension
of the matrix for users and mode selection is 2 K, so its
length is polynomial in the size of the input. We can write
an algorithm that can verify the result in O(2K) iterations,
which shows polynomial time verification.
Lemma 2. MCMDKP is polynomial time reducible to
JACMSPA.
Proof. We can do reduction by simple equivalence. We make
the K users as disjoint classes, the items as mode. The Rk
rate of selected users is the same as capacity of knapsack
with W dimensions. We make the channels hk , fk , gk as item
profit. For any given instance, J, Q, W , fj,q of MCMDKP,
if there is a matrix X with entries of selected items xj,q
that maximizes the total profit such that an item can only
be selected by at most one class subject to satisfaction of W
resource constraints, then there will be a user-mode selection
matrix that maximizes the the total data rate (sum-capacity)
of the system such that a user can only operate in one mode
subject to rate constraint. This reduction by simple equivalence
means every yes-instance of MCMDKP implies yes-instance
of JACMSPA in polynomial time.
Lemma 1 and Lemma 2 implies the NP-completeness of
joint admission control, mode selection and power allocation
problem.

Instance: J disjoint classes,Q items and W resource constraints (capacity of knapsack with W dimensions). The qth
item of class j has profit fj,q and weight wj,q [44].
Output: A selection of items X.
Decision problem associated with MCMDKP: The
MCMDKP decision problem is to determine, for a given
profit F , whether it is possible to load the multidimensional
knapsack so as to keep the total weight in each dimension no
greater than W , while making the total profit at least equal to
F.
Problem 2. The JACMSPA problem is to select the disjoint
subsets of users that are either using D2D or eNB for
communication such that: 1) the total data rate (sum-capacity)
of the system is maximized and 2) Data rate of each selected
user must be more than or equal to a predefined threshold.

Muhammad Azam received the B.Sc. degree in


Electronics Engineering from University of Engineering and Technology Taxila, Pakistan in 2012,
and M.Sc. degree in Electrical Engineering from
COMSATS Institute of Information Technology Pakistan, in 2015. Currently, he is a Lab Engineer
in Electrical Engineering Department, Foundation
University Islamabad (Rawalpindi Campus) His research interests include Wireless communications,
Resource Allocation in 5G Networks, Optimization
in Energy-Communication Systems, Smart Grid, and
related topics.

0018-9545 (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/TVT.2015.2487220, IEEE Transactions on Vehicular Technology
12

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXXX 2015

Mushtaq Ahmad received the BSc , MS in Electrical Engineering and Masters of Telecommunication
Management from University of Engineering and
Technology Lahore, University of Michigan Ann Arbor, USA and Institut National Des Telecom France
in 1990, 1994 and 1999 respectively. Possessing
double Masters degree both in the field of Electrical Engineering and Telecom Management from
renowned universities of the world , the author has
diverse working experience with Incumbent operator
PTCL, regulator PTA, policy formulation experience
in the capacity of Member Telecom at Ministry of Information Technology and
experience in the capacity of Chief Executive Officer of reputable public and
private institutions. The author is currently working as a Principal Engineer
at Comsats Institute of Information Technology Wah, Pakistan and pursuing
his PhD degree from same institution . His main research areas are Resource
Allocation in next generation networks.

Muhammad Naeem (S09-M12) received the BS


(2000) and MS (2005) degrees in Electrical Engineering from the University of Engineering and
Technology, Taxila, Pakistan. He received his PhD
degree (2011) from Simon Fraser University, BC,
Canada. From 2012 to 2013, he was a Postdoctoral
Research Associate with WINCORE Lab. at Ryerson University, Toronto, ON, Canada. Since August
2013, he has been an assistant professor with the
Department of Electrical Engineering, COMSATS
Institute of IT, Wah Campus, Pakistan and Research
Associate with WINCORE Lab. at Ryerson University. From 2000 to 2005, he
was a senior design engineer at Comcept (pvt) Ltd. At the design department
of Comcept (pvt) Ltd, Dr. Naeem participated in the design and development
of smart card based GSM and CDMA pay phones. Dr. Naeem is also a Microsoft Certified Solution Developer (MCSD). His research interests include
optimization of wireless communication systems, non-convex optimization,
resource allocation in cognitive radio networks and approximation algorithms
for mixed integer programming in communication systems. Dr. Naeem has
been the recipient of NSERC CGS scholarship.

Muhammad Iqbal was born on October 15, 1976 in


Multan. He got B.Sc. Electrical Engineering degree
in 1999 from University of Engineering and Technology, Lahore. After completing B.Sc. Electrical
Engineering, he served in the state owned telecommunication company for more than seven years.
In 2007 he completed his MS Telecommunication
Engineering from the University of Engineering and
Technology, Peshawar. After completing PhD by
July, 2011 from Beijing University of Posts and
Telecommunications, P.R. China, he rejoined COMSATS and till this date working as Assistant Professor, Electrical Engineering
Department, CIIT, Wah Campus. His research interests include signal and
information processing, wireless communication, smart grid and applied
optimization.

Ahmed Shaharyar Khwaja received the Ph.D. and


M.Sc. degrees in signal processing and telecommunications from the University of Rennes 1, Rennes,
France, and B. Sc. degree in electronic engineering
from Ghulam Ishaq Khan Institute of Engineering
Science and Technology, Topi, Pakistan. Currently,
he is a postdoctoral research fellow with WINCORE
Lab. at Ryerson University, Toronto, ON, Canada.
His research interests include compressed sensing,
remote sensing and optimization problems in wireless communication systems.

Alagan Anpalagan received the B.A.Sc. M.A.Sc.


and Ph.D. degrees in Electrical Engineering from
the University of Toronto, Canada. He joined the
Department of Electrical and Computer Engineering
at Ryerson University in 2001 and was promoted
to Full Professor in 2010. He served the department as Graduate Program Director (2004-09) and
the Interim Electrical Engineering Program Director
(2009-10). During his sabbatical (2010-11), he was
a Visiting Professor at Asian Institute of Technology
and Visiting Researcher at Kyoto University. Dr.
Anpalagans industrial experience includes working at Bell Mobility, Nortel
Networks and IBM Canada. Dr. Anpalagan directs a research group working
on radio resource management (RRM) and radio access & networking (RAN)
areas within the WINCORE Lab. His current research interests include
cognitive radio resource allocation and management, wireless cross layer
design and optimization, cooperative communication, M2M communication,
small cell networks, and green communications technologies.
Dr. Anpalagan serves as Associate Editor for the IEEE Communications
Surveys & Tutorials (2012-), IEEE Communications Letters (2010-13) and
Springer Wireless Personal Communications (2009-), and past Editor for
EURASIP Journal of Wireless Communications and Networking (2004-2009).
He also served as Guest Editor for two EURASIP SI in Radio Resource
Management in 3G+ Systems (2006) and Fairness in Radio Resource Management for Wireless Networks (2008) and, MONET SI on Green Cognitive and
Cooperative Communication and Networking (2012),. He co-authored of three
edited books, Design and Deployment of Small Cell Networks, Cambridge
University Press (2014), Routing in Opportunistic Networks, Springer (2013),
Handbook on Green Information and Communication Systems, Academic
Press (2012).
Dr. Anpalagan served as TPC Co-Chair of: IEEE WPMC12 Wireless
Networks, IEEE PIMRC11 Cognitive Radio and Spectrum Management,
IEEE IWCMC11 Workshop on Cooperative and Cognitive Networks, IEEE
CCECE04/08 and WirelessCom05 Symposium on Radio Resource Management. He served as IEEE Canada Central Area Chair (2013-14), IEEE Toronto
Section Chair (2006-07), ComSoc Toronto Chapter Chair (2004-05), IEEE
Canada Professional Activities Committee Chair (2009-11). He is the recipient
of the Deans Teaching Award (2011), Faculty Scholastic, Research and
Creativity Award (2010), Faculty Service Award (2010) at Ryerson University.
Dr. Anpalagan also completed a course on Project Management for Scientist
and Engineers at the University of Oxford CPD Center. He is a registered
Professional Engineer in the province of Ontario, Canada.

Saad Qaisar received his Masters and Ph.D. degrees in Electrical Engineering from Michigan State
University, USA in 2005 and 2009, respectively,
under supervision of Dr. Hayder Radha (Fellow,
IEEE). He is currently serving as an Assistant Professor at School of Electrical Engineering Computer
Science (SEECS), National University of Sciences
Technology (NUST), Pakistan. He is the lead researcher and founding director of CoNNekT Lab:
Research Laboratory of Communications, Networks
and Multimedia at National University of Sciences
Technology (NUST), Pakistan. As of September 2011, he is the Principal
Investigator or Joint Principal Investigator of multiple research projects
spanning cyber physical systems, applications of wireless sensor networks,
network virtualization, communication and network protocol design, wireless
and video communication, internet measurements analysis, multimedia coding
and communication. He has published over 50 papers at reputed international
venues. He is also serving as the chair for Mobile-Computing, Sensing and
Actuation for Cyber Physical Systems (MSA4CPS) workshop in conjunction
with International Conference on Computational Science and Its Applications
(ICCSA). He is the lead researcher on a joint internet performance measurements study in Pakistan with Georgia Institute of Technology, current head of
a core working group for establishment of Pakistan IXP, technical consultant
to United Nations Food and Agriculture Organization (UNFAO), lead architect
for Pakistan Laboratory Information Management System project and actively
engaged with projects funded by King Abdullah City of Science Technology
(KACST).

0018-9545 (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.

You might also like