This paper addresses the problem of resource allocation in amplifyandforward (AF) relayed OFDM based cognitive radio networks (CRNs). The purpose of resource allocation is to maximize the overall throughput, while satisfying the constraints on the individual power and the interference induced to the primary users (PUs). Additionally, different from the conventional resource allocation problem, the rateguarantee constraints of the subcarriers are considered. We formulate the problem as a mixed integer programming task and adopt the dual decomposition technique to obtain an asymptotically optimal power allocation, subcarrier pairing and relay selection. Moreover, we further design a suboptimal algorithm that sacrifices little on performance but could significantly reduce computational complexity. Numerical simulation results confirm the optimality of the proposed algorithms and demonstrate the impact of the different constraints.
1. Introduction
T
he cognitive radio (CR) technology has been proposed to improve the spectrum utilization and provide adaptability for wireless transmission on licensed spectrums
[1]
. The CR performance and the spectrum utilization can be further improved by adopting cooperative communications and orthogonal frequency division multiplexing (OFDM) technology
[2
,
3]
.
With the internal flexibility of OFDM in power loading across subcarriers, a lot of works have been already done on the resource allocation in noncognitive relay systems
[4

6]
. However, resource allocation in OFDMbased cooperative CRNs is more complex than that in a conventional OFDM system because more constraints must be considered to protect the performance of primary users (PUs). Thus, many existing resource allocation algorithms are not suitable for OFDMbased cooperative CRNs. Resource allocation for OFDMbased cooperative CRNs have attracted much attention recently. The authors in
[7]
proposed a resource allocation algorithm in cognitive wireless networks based on game theory. The problem of relay selection and optimal resource allocation for twoway relaying CRN was investigated in
[8]
. The authors in
[9]
proposed a joint subcarrier matching and power loading algorithm in relayaided CRN. A problem of throughput maximization in a multicarrier underlay CRN with constrained transmission power and interference threshold had been studied in
[10]
. An extension of
[10]
that employed an amplifyandforward (AF) relay to aid transmission could be found in
[11]
. The authors in
[12]
proposed a resource allocation algorithm for multiuser OFDMbased CRNs, in which the proportional rate constraints were considered. However, the works in
[7

12]
considered limited practical constraints.
The motivation for this paper is twofold. Firstly, the rate provided by allocated subcarriers may be too low for practical usage. Therefore, it is important to guarantee that the rate of each subcarrier is not below a certain threshold. Secondly, the resource allocation problems are generally NPHard, so the designation of lowcomplexity and highperformance algorithm is a great challenge. The main contributions of this paper are the following aspects. 1) The joint resources (powers, subcarriers, relays) allocation problem is investigated in AF relayed OFDMbased CRNs. 2) Besides the power and interference constraints considered in traditional schemes, the rateguarantee constraints are also considered to adapt to practical usage. 3) An asymptotically optimal resource allocation algorithm that adopts dual decomposition technique is proposed. 4) A lowcomplexity heuristic algorithm with little performance degradation is designed to solve the problem.
The remaining of this paper is organized as follows. Section 2 introduces the system model while the problem is formulated and the optimal scheme is presented in section 3. Section 4 gives a suboptimal algorithm, of which the computer simulation and numerical analysis are provided in section 5. Finally, we draw conclusions in section 6.
2. System Model
The system model of the OFDMbased cooperative CRN is shown in
Fig.1
, where the cognitive users share radio spectrum with a primary system. The available spectrum bandwidth of CRN is divided into
N
subcarriers, and the bandwidth of each subcarrier is Δ
f
(Δ
f
=
B
/
N
). We assume that the source transmits signals to destination through
M
relays and there is no direct link between them. Moreover, the relays operate in halfduplex mode with AFprotocol, where AFprotocol is divided into two time slots. In the first time slot, one relay is selected to receive the signals from the source on the
j^{th}
subcarrier. In the second time slot, the selected relay amplifies the signals and forwards them to the destination via subcarrier
k
. The
j^{th}
subcarrier in the source should be paired with only one subcarrier
k
in the destination, which may not be the same as
j
. It is assumed that perfect instantaneous fading gains are available at the transceivers of both CRN and PU.
System model of OFDMbased cooperative cognitive radio network
In OFDMbased CRN, the interference introduced to the PU on the
i^{th}
subcarrier is described as
[13]
where
d_{i}
represents the spectral distance between the
i^{th}
subcarrier and the PU band.
G_{i}
denotes the square of the channel gain between the
i^{th}
subcarrier and the PU band.
P_{i}
is the total transmission power emitted by the
i^{th}
subcarrier, and
T_{s}
is the symbol duration. Let
ρ_{i}
denote the interference factor of the
i^{th}
subcarrier to the PU band. Similarly, the interference power introduced by PU signal with power spectrum density
ϑ
(
e^{jw}
) into the band of the
i^{th}
subcarrier is
[13]
J_{i}
can be modeled as additive white Gaussian noise (AWGN) by applying the law of large number or by assuming that the primary and cognitive system are using an independent and random Gaussian codewords
[14
,
15]
.
From the source to the
m^{th}
relay, denote the channel coefficient over the
j^{th}
subcarrier by
and and the power allocation by
and the transmitted symbol by
. The signal received at the
m^{th}
relay is
is the summation of AWGN and the interference introduced by the PU signal into the
j^{th}
subcarrier, where
. If the
m^{th}
relay is chosen to amplify and forward the signal on the
k^{th}
subcarrier, the received signal at the receiver will be
where
is the power that the relay allocates to transmit the received signal on the
k^{th}
subcarrier.
is the channel gain between the
m^{th}
relay and the destination on the
k^{th}
is the square of the
j^{th}
(
k^{th}
) subcarrier fading gain over source to
R_{m}
(
R_{m}
to destination).
is the summation of AWGN and the interference introduced by the PU signal into the
k^{th}
subcarrier, where
.
The transmission rate over subcarrier pair (
j
,
m
,
k
) at high signaltonoise radio can be approximated as (5), such an approximated is reasonable as discussed in
[16]
.
3. Problem Formulation and Optimal Solution
The optimization objective is to maximize the overall throughput of CRN by optimizing the power allocation, subcarrier pairing and relays assignment, while satisfying multiple practical constraints. Accordingly, the optimal problem is formulated as
where
N
denotes the total number of subcarriers while
I_{th}
is the interference threshold prescribed by PU.
P_{S}
and
P_{Rm}
are the available power budget in the source and the
m^{th}
relay respectively.
ρ_{j}
and
ρ_{m,k}
are the
j^{th}
(
k^{th}
) subcarrier interference factor to the PU band from the source and the
m^{th}
relay respectively.
R_{th}
is the rateguarantee threshold.
ϕ_{(j,k)}
= 1 if the
j^{th}
subcarrier from the source is paired with the
k^{th}
to the destination, and zero otherwise. Additionally,
φ
_{(j, m, k )}
is the relay assignment indicator which equals to one if the pair (
j
,
k
) is assigned to the
m^{th}
relay and zero otherwise.
Finding the optimal variables in (6) is a mixed binary integer programming problem. It is difficult to find the global optimal solution. However, under the time sharing condition the duality gap is asymptotically zero for sufficiently large
N
[17]
. Since the time sharing condition is readily satisfied in our case, we can solve the dual problem of the original problem to obtain an asymptotically optimal solution. To make the analysis more clearly and without loss of generality, the noise variance is assumed to be constant for all the subcarriers and users, i.e.
. The dual problem associated with the primal problem (6) can be written as
where
τ
and
are the dual variables associated with the power constraint at the source and the different relays respectively. The dual variables
υ
and
are related to the interference constraints during the first and second time slots respectively. Moreover,
η
_{(j, m, k )}
is the dual variable associated with the rateguarantee constraint. The dual function
f
(
τ
,
υ
,
,
,
η
_{(j, m, k )}
) is defined as follows
where the Lagrangian function
L
is defined as
The dual function in (8) can be rewritten as (10)
where
Thus the problem (10) is decomposed into three subproblems.
1) Subproblem 1:
Power allocation scheme. We assume that (
j
,
m
,
k
) is a valid subcarrier pair, and set the different dual variables with the initial values. Therefore, the optimal power allocation can be determined by solving the following problem for every (
j
,
m
,
k
) pair,
Solving (12) for the optimal power, we can obtain the optimal power allocation
and
.
2) Subproblem 2:
Relay assighnment scheme. The power variables can be eliminated by substituting the optimal power allocation found by (12) to (10). Correspondingly, we can solve the following problem for every (
j
,
k
) to get the best relay,
Therefore, the optimal relay assignment strategy is achieved by allocating the (
j
,
k
) pair to the relay which maximizes the function
. If
, then set
φ
_{(j, m, k )}
= 1 and zero otherwise. By performing this allocation, the best relay is determined for every possible subcarrier pair.
3) Subproblem 3:
Subcarriers pairing scheme. The optimal subcarriers pair can be obtained by the following problem after the powers and relay allocation are determined for every subcarrier pair,
In order to get the solution of (14), the Hungarian method is used.
Once the optimal solutions, i.e.
, are obtained, substituting them into (9) and then the result back into (8), we can get the optimal dual function for the given values of the dual variables. The subgradient method can be used to solve the dual problem with guaranteed convergence. For any initial values
τ
^{0}
,
υ
^{0}
,
, the dual variables at the (
i
+ 1)
^{th}
iteration can be updated as
where
δ^{i}
is the step size that can be updated according to the nonsummable diminishing step size policy
[18]
. With the updated values of the dual variables, the optimal power allocation and subcarrier matching are evaluated again. The iterations are repeated until convergence is reached.
4. Suboptimal Algorithm
In order to decrease the computational complexity without sacrificing performance, we propose a suboptimal heuristic algorithm for the aforementioned optimization problem by which the different resources are allocated jointly with lower computational complexity than that of the optimal solution. The proposed algorithm takes into consideration the different channel qualities, the rateguarantee constraint, the available power budgets, the interference introduced to the PU and the limitation introduced by using AFprotocol.
The suboptimal algorithm addresses the optimization problem in two steps. Firstly, subcarrier pairing and relay selection scheme is proposed with initial power values. Then, the optimal power allocation scheme is used to improve the system performance. We commence the description of the proposed schemes by defining the sets S1 and
S
2 to include all the nonassigned subcarriers in the source and the destination sides respectively. Moreover, define the set
C
to contain all the relays in the network.
 4.1 Proposed Subcarrier Pairing and Relay Selection Scheme
In the proposed subcarrier pairing and relay selection scheme, we are going to use the harmonic mean criterion to select the best relay for each pair of subcarriers. In the source side, assume that the available source power is distributed uniformly over the subcarriers, i.e.
P_{j}^{uni}
=
P_{S}
/
N
, and also assume that the interference introduced to the PUs by every subcarrier is equal; from (1), the maximum allowable power that can be allocated to the
j^{th}
subcarrier is
P_{j}
^{max}
=
I_{th}
/ (
Nρ_{j}
). Therefore, the allocated power to the
j^{th}
subcarrier in the source side is
= min(
P_{S}
/
N
,
I_{th}
/ (
Nρ_{j}
)). To complete the algorithm, we define the harmonic mean criterion as
For every subcarrier in the source side, we search over all the relays and all the nonassigned subcarriers in the destination sides to find the subcarrierrelay pair with the largest
H
. Moreover, the selected relay are updated while the rateguarantee constraint is considered. This trend continues until the set S1 becomes empty. The assigning procedures of a particular subcarrier
j
∈
S
1 are as follows

1)For every relaym∈Cand subcarrierk∈S2 , evaluateandwhere S2 means the cardinality of the setS2. Therefore, the allocated power to theksubcarrier in themthrelay is= min(PRm/S2,Ith/ (Nρm,k)).

2)Find the best relaym*and the subcarrier pairk*satisfying. Compute the communication rateR(j,m*,k*)for the selected link (j,m*,k*).

Case1: IfR(j,m*,k*)

Case2: IfR(j,m*,k*)≥Rth, then setϕ(j,k*)= 1 ,φ(j,m*,k*) = 1 and.

Additionally, update the power budget of themthrelay as.

Remove the subcarrierjandk*from the setsS1 andS2 respectively.

3)This trend continues until all the subacarriers in the source are processed, i.eS1 = Ø.
 4.2 Proposed Power Allocation Scheme
The proposed subcarrier pairing and relay selection scheme determines the best paring link (
j
,
m^{*}_{j}
,
k^{*}_{j}
) for every
.
m^{*}_{j}
and
k^{*}_{j}
are the best relay and paring subcarrier at destination for subcarrier
j
at source. Consequently, the initial optimization problem (6) can be redescribed as
The Lagrangian of (17) is defined as
where
λ
,
μ
,
are Lagrangian variables. The optimal solution can be obtained from the KKT conditions as follows,
where
,
and
.
In order to determine the optimal power values, we divide the objective function of the (17) into two subproblems equivalently and apply an alternating optimization method to work out the
and
efficiently.
1)subproblem 1:
Optimal power allocation for
. As shown in (19),
is regarded as a function of
, where [
x
]
^{+}
= max(
x
, 0). We can adjust
λ
and
μ
to satisfy
and
, so we can obtain the
with (19) by setting
to an initial value.
The power allocation scheme at the source is described in
Table 1
.
The power allocation at the source
The power allocation at the source
2)subproblem 2:
Optimal power allocation for
. The power allocation scheme at the relay is the same with algorithm 1 except that
λ
,
μ
,
,
P_{S}
,
ρ_{j}
, (19),
are replaces by
.
The proposed power allocation scheme is an iteratively algorithm. The convergence of the proposed scheme is guaranteed. Rewrite the objective function in (17) as follows,
where
.
Denote the global optimal point as
f
^{*}
, we can get the following theorem,
Theorem 1:
Proof:
Start with an arbitrary initial point P
_{S,Rm}
, for
n
≥ 1, we can get
Denote
, from(22), can get
Equation (23) demonstates the sequence
f^{n}
is nondecreasing. So, it must converge to
f
^{*}
because
f
is bounded from above. The above proof shows that the alternating optimization scheme for power allocation converges, and the converged value is the optimal solution.
For the optimal solution derived in the section 3,
MN
^{2}
function evaluations are performed to find the power allocation in every iteration. Afterwards,
M
function evaluations are performed for every possible subcarrier pair where there are
N
^{2}
2Ndifferent subcarrier pairs. By including the computational complexity of the Hungarian methord, the optimal algorithm has a complexity of
O
(
T
(
MN
^{2}
) +
N
^{3}
) where
T
is the number of iterations required to converge. For the proposed suboptimal algorithm, every subcarrier in the source side requires no more than (
M
+
MN
) function evaluations to be paired and assigned to the relay in the subcarrier pairing and relay selection scheme, and the complexity of the power allocation scheme is
T
. Therfore, the complexity of the proposed suboptimal algorithm, is
O
(
MN
^{2}
+
T
), where
means the cardinality of the set
(
≤
N
). The proposed suboptimal algorithm achieves much lower computational complexity.
5. Simulation results
The simulations are performed under the scenario given in
Fig.1
. The channel gains are outcomes of independent Rayleigh distributed random variables with mean equal to 1. All the results have been averaged over 10000 iterations. In the simulations,
Optimal
and
Suboptimal
schemes apply the dual decomposition technique presented in Sec.3 and the proposed method presented in Sec.4 respectively. Furthermore,
RRA
and
RSA
refer to the method by which the relays and the subcarriers are assigned and matched randomly respectively, while
IFPA
allocates power inversely proportional to the interference level
[19]
. All the parameters in the simulations are described in
Table 2
.
Simulation parameters
The comparisons for our proposed algorithms and the other algorithms are shown in
Fig.2
,
Fig.3
and
Fig.4
. It can be found that our proposed algorithms, the
Optimal
and the
Suboptimal
perform better than the others. This is because that the power allocation, subcarrier pairing and relay assignment are performed jointly in our proposed algorithms, while the others take only some of them into consideration. It is worth noticing that the gap between the
Optimal
and the
Suboptimal
is small, suggesting that the suboptimal algorithm provides a good approximation to the optimal. From
Fig.2
we can see that the overall throughput grows with the increase of power budgets, and all the algorithms obtain close solutions when power budgets are large. From
Fig.3
we can observe that the overall throughput grows with the increase of interference threshold, and all the algorithms have a near performance in the low interference threshold region. The same interpretation can be applied on
Fig.4
in which the overall throughput decreases as rateguarantee threshold grows for all the algorithms.
Overall throughput vs. available power budgets with P_{S} = P_{Rm} .
Overall throughput vs. allowed interference threshold.
Overall throughput vs. rateguarantee threshold.
Fig. 5
illustrates the overall throughput of the suboptimal schemes vs. the available power budgets in source and relay under different number of relays. In the same power budget, the overall throughput increases with the number of relays increases. Moreover, the gain of overall throughput decreases with the number of relays increases.
Overall throughput vs. the available power budgets under different number of relays.
In
Fig.6
, depicts the overall throughput of the suboptimal schemes vs. the available power budgets in source and relay under different number of subcarriers. It can be noted that the overall throughput grows with the number of subcarriers, and the gain of overall throughput increases with the increase of power budgets.
Overall throughput vs. the available power budgets under different number of subcarriers.
6. Conclusion
In this paper, we have investigated the resource allocation problem in OFDMbased cooperative CRN. To maximize the overall throughput under the consideration of multiple practical limitations, the joint subcarrier pairing, best relay selection and power allocation scheme has been proposed by using the dual decomposition technique. Due to the high computational complexity of the optimal scheme, a heuristic suboptimal algorithm is presented. In the first step, the subcarrier pairing and relay selection scheme is proposed to satisfy the rateguarantee constraint, and remove the integer constraints from the problem. In the second step, the power allocation scheme is considered to improve the system performance. The suboptimal algorithm shows to perform almost equally well as the optimal scheme with a much lower complexity. Moreover, the performance of the proposed algorithms outperform the others algorithms,
RSA
,
RRA
and
IFPA
.
BIO
Xuezhou Yang was born in Nanchong, China. He is currently working towards the Ph.D. degree in National Key Laboratory of Science and Technology on Communications at University of Electronic Science and Technology of China, Chengdu, China. His research interests span the broad area of wireless communication and networks, cooperative communication system. Recently, he has been working on the cooperative spectrum sensing in cognitive radio network, and resource allocation for OFDMbased cognitive radio networks.
Wei Tang received his B.S., M.S. and Ph.D. degrees in Communication and information system from University of Electronic Science and Technology of China, Chengdu, China. He currently works in National Key Laboratory of Science and Technology on Communications at University of Electronic Science and Technology. His research interest includes wireless communication and networks, cooperative communication system.
Wei Guo received his B.S. and M.S. degrees from University of Electronic Science and Technology of China, Chengdu, China. He currently works in National Key Laboratory of Science and Technology on Communications at University of Electronic Science and Technology of China as a professor. His research interest includes wireless communication and networks, Satellite and space communications technology, software systems and medical systems and telecare/telehealth networks.
Miltola J.
,
Maguire G. Q.
1999
“Cognitive radio: Making software radios more personal”
IEEE Peronal Commun
Article (CrossRef Link)
6
(4)
13 
18
DOI : 10.1109/98.788210
Jia J.
,
Zhang J.
,
Zhang Q.
2009
“Cooperative relay for cognitive radio networks”
in Proc. of 2009 IEEE Inforcom
Article (CrossRef Link)
2304 
2312
Feng Daquan
,
Jiang Chenzi
,
Lim Gubong
,
Cimini L.J.
,
Feng Gang
,
Li G.Y
2013
“A survey of energyefficient wireless communications”
IEEE Communications Surveys and Tutorials
Article (CrossRef Link)
15
(1)
167 
178
DOI : 10.1109/SURV.2012.020212.00049
Yaru Fu.
,
Qi Zhu.
2013
“A joint resource allocation scheme for relay enhanced multicell orthogonal frequency division multiple networks”
KSII Transactions on Internet and Information Systems
Article (CrossRef Link)
7
(2)
288 
307
DOI : 10.3837/tiis.2013.02.007
Chang TainSao
,
Feng KaiTen
,
Lin JiaShi
,
Wang LiChun
2013
“Green resource allocation Schemes for relayenhanced MIMOOFDM networks”
IEEE Tran on VT
Article (CrossRef Link)
62
(9)
4539 
4554
Linshu Lv.
,
Qi Zhu.
2013
“Joint relay selection and resource allocation for cooperative OFDMA network”
KSII Transactions on Internet and Information Systems
Article (CrossRef Link)
6
(11)
3008 
3025
AbdulGhafoor Omar B
,
Ismail Mahamod
,
Nordin Rosdiadee
2013
“Resource allocation in spectrum sharing adhoc cognitive radio networks based on game theory: an overview”
KSII Transactions on Internet and Information Systems
Article (CrossRef Link)
7
(12)
2957 
2986
DOI : 10.3837/tiis.2013.12.001
Alsharoa Ahmad
,
Bader Faouzi
,
Alouini MohamedSlim
2013
“Relay selection and resource allocation for twoway DFAF cognitive radio networks”
IEEE Wireless Communication Letters
Article (CrossRef Link)
2
(4)
427 
430
DOI : 10.1109/WCL.2013.051513.130211
Ge Mengyao
,
Wang Shaowei
2012
“Fast optimal resource allocation is possible for multiuser OFDMbased cognitive radio networks with heterogeneous services”
IEEE Trans. Wireless Communication
Article (CrossRef Link)
11
(4)
1500 
1509
DOI : 10.1109/TWC.2012.021512.111233
Shaat Musbah
,
Bader Faouzi
2012
“Asymptotically optimal resource allocation in OFDMbased cognitive networks with multiple relays”
IEEE Trans. Wireless Communication
Article (CrossRef Link)
11
(3)
892 
897
DOI : 10.1109/TWC.2012.011012.110880
Ahmad Guftaar
,
Sidhu Sardar
,
Gao Feifei
,
Wang Wei
2013
“Resource allocation in relayaided OFDM cognitive radio networks, ”
IEEE Tran on VT
Article (CrossRef Link)
62
(8)
3700 
3710
Wang Shaowei
,
Huang Fangjiang
,
Yuan Mindi
,
Du Sidan
2012
“Resource allocation for multiuser cognitive OFDM networks with proportional rate constraints”
International Journal of Communication Systems
Article (CrossRef Link)
25
(2)
254 
269
DOI : 10.1002/dac.1272
Weiss T.
,
Hillenbrand J.
2004
“Mutual interference in OFDMbased spectrum pooling systems”
in Proc. of 2004 IEEE Vehicular Technology ConferenceSpring
vol.4, Article (CrossRef Link)
1873 
187
Bansal G.
,
Hossain M. J.
,
Bhargava V. K.
2008
“Optimal and suboptimal power allocation schemes for OFDMbased cognitive radio systems”
IEEE Trans. Wireless Communicatio.
Article (CrossRef Link)
7
(11)
4710 
4718
DOI : 10.1109/TWC.2008.07091
Cove T. M.
,
Thomas J. A.
2006
Elements of Information Theory
2nd edition
John Wiley and Sons
Hasna M.
,
Alouini M.S.
2004
“Harmonic mean and endtoend performance of transmission systems with relays”
IEEE Trans on Communications
Article (CrossRef Link)
52
(1)
130 
135
DOI : 10.1109/TCOMM.2003.822185
Yu W.
,
Lui R.
2006
“Dual methods for nonconvex spectrum optimization of multicarrier systems”
IEEE Trans. IEEE Trans on Communications
Article (CrossRef Link)
54
(7)
1310 
1322
DOI : 10.1109/TCOMM.2006.877962
Boyd S.
,
Mutapcic A.
2006
Subgradient methods
Standford University
notes for EE364
Almalfouh S. M.
,
Stuber G. L.
2011
“Interferenceaware radio resource allocation in OFDMAbased cognitive radio networks”
IEEE Trans. on VT
Article (CrossRef Link)
60
(4)
1699 
1713