Generalized Selection Combining for Cognitive Relay Networks

over Nakagami-m Fading

Deng, Y., Wang, L., Elkashlan, M., Kim, K. J., & Duong, T. Q. (2015). Generalized Selection Combining for
Cognitive Relay Networks over Nakagami-m Fading. IEEE Transactions on Signal Processing, 63(8), 1993-
2006. DOI: 10.1109/TSP.2015.2405497

Published in:
IEEE Transactions on Signal Processing

Document Version:
Peer reviewed version

Queen's University Belfast - Research Portal:

Link to publication record in Queen's University Belfast Research Portal

Publisher rights
(c) 2015 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other users, including reprinting/
republishing this material for advertising or promotional purposes, creating new collective works for resale or redistribution to servers or lists,
or reuse of any copyrighted components of this work in other works.

General rights
Copyright for the publications made accessible via the Queen's University Belfast Research Portal is retained by the author(s) and / or other
copyright owners and it is a condition of accessing these publications that users recognise and abide by the legal requirements associated
with these rights.

Take down policy

The Research Portal is Queen's institutional repository that provides access to Queen's research output. Every effort has been made to
ensure that content in the Research Portal does not infringe any person's rights, or applicable UK laws. If you discover content in the
Research Portal that you believe breaches copyright or violates any law, please contact openaccess@qub.ac.uk.

Download date:15. Feb. 2017

 Generalized Selection Combining for Cognitive
Relay Networks over Nakagami-m Fading
Yansha Deng, Student Member, IEEE, Lifeng Wang, Student Member, IEEE, Maged Elkashlan, Member, IEEE,
Kyeong Jin Kim, Senior Member, IEEE, and Trung Q. Duong, Senior Member, IEEE
Abstract—We consider transmit antenna selection with receive
generalized selection combining (TAS/GSC) for cognitive decode-
and-forward (DF) relaying in Nakagami-m fading channels. In
an effort to assess the performance, the probability density
function and the cumulative distribution function of the end-
to-end SNR are derived using the moment generating function,
from which new exact closed-form expressions for the outage
probability and the symbol error rate are derived. We then
derive a new closed-form expression for the ergodic capacity.
More importantly, by deriving the asymptotic expressions for
the outage probability and the symbol error rate, as well
as the high SNR approximations of the ergodic capacity, we
establish new design insights under the two distinct constraint
scenarios: 1) proportional interference power constraint, and 2)
fixed interference power constraint. Several pivotal conclusions
are reached. For the first scenario, the full diversity order of the
outage probability and the symbol error rate is achieved, and
the high SNR slope of the ergodic capacity is 1/2. For the second
scenario, the diversity order of the outage probability and the
symbol error rate is zero with error floors, and the high SNR
slope of the ergodic capacity is zero with capacity ceiling.
Index Terms—Cognitive relay network, generalized selection
combining, Nakagami-m fading.
I. I NTRODUCTION
The conflict between the stringent demand for high data
rate and data service on the one hand, and the unbalanced
spectrum occupation in time and geographic domains on the
other hand, has become a challenge for future wireless systems
[1]. To cope with this, cognitive radio, first coined by Mitola,
has rekindled increasing interest in the efficient use of radio
spectrum. In the underlay paradigm, the secondary users (SUs)
are allowed to access the spectrum allocated to primary users
(PUs) as long as the interference generated by the secondary
transmission is restricted below a certain threshold, namely,
interference temperature [2]. The constrained transmit power
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 pubs-permissions@ieee.org.
Manuscript received Jan.31, 2014; revised Sep. 22, 2014; accepted Jan. 26,
2015. The associate editor coordinating the review of this manuscript and
approving it for publication was Prof. Zhengdao Wang. This work of Y. Deng
was supported by China Scholarship Council. This paper has been presented
in part at IEEE International Conference on Communications (ICC), Sydney,
2014.
Y. Deng, L. Wang, and M. Elkashlan are with Queen Mary Uni-
versity of London, London E1 4NS, UK (email: {y.deng, lifeng.wang,
maged.elkashlan}@qmul.ac.uk).
K. J. Kim is with Mitsubishi Electric Research Laboratories (MERL), 201
Broadway, Cambridge, MA, USA (email: kkim@merl.com).
T. Q. Duong is with Queen’s University Belfast, Belfast BT7 1NN, UK
(email: trung.q.duong@qub.ac.uk).
at SU typically results in unstable transmission and restricted
coverage, which drives the demand for robust transmission
techniques suited for networks that are subject to power and
interference constraints [3]. Relaying is regarded as a cost-
effective approach for supporting high speed and long distance
networks [4, 5].
The majority of the studies on cognitive relay networks
have focused on single antenna protocols [6–8]. Multiple-
input multiple-output (MIMO) techniques, well-known for
their many benefits including enhanced reliability [9], spectral
efficiency [10], and co-channel interference suppression [11],
open up new dimensions for cognitive radio. For example, as
shown in [12], multi-antennas are utilized at SU to achieve
spatial multiplexing. In [13], the novel distributed antenna
selection is proposed in relaying system. In [14], the effect
of transmit antenna selection with receive maximal ratio com-
bining (TAS/MRC) on the the ergodic capacity was analyzed.
In [15], the outage performance of TAS/MRC and TAS/SC
are examined over Nakagami-m fading channel. It is shown
in [16] that the diversity order is independent of the number
of PUs and the selected number of receive antennas at SU.
Different from the aforementioned works, in this paper,
we consider cognitive relay networks from the viewpoint of
TAS/GSC as an effective design to enhance the reliability
of the secondary network and to mitigate interference to the
primary network. From a power perspective, cognitive spec-
trum sharing with network cooperation addresses fundamental
constraints on the transmit power at the SUs, while keeping
the interference temperature at the PUs to a minimum [17].
On the one hand, TAS is acknowledged as a core component
for uplink 4G long term evolution (LTE) and LTE Advanced
systems because of its low feedback requirement compared
with closed-loop transmit diversity [18]. On the other hand,
with the merits of low power demand and RF cost, GSC
offers a performance/implementation tradeoff between MRC
and selection combining (SC) for the secondary network
[19, 20]1 . Additionally, by excluding the antenna chains with
weak channel powers, GSC can be more robust to channel
estimation errors than MRC [21]. In [22], it is shown that
GSC outperforms MRC in a non-identically distributed noise
scenario.
The objective of this paper is to examine the impact
of TAS/GSC in underlay cognitive relay networks over
1 GSC is well applied to commercial wireless networks where the receiver
is subject to resource constraints, such as limited RF chains due to size and
complexity limitations [26].
Nakagami-m fading. The Nakagami-m fading environment is
considered due to its versatility in providing a good match
to various empirically obtained measurement data [23]. In
the secondary network, a single antenna which maximizes
the signal-to-noise ratio (SNR) is selected at the secondary
transmitter, while a subset of receive antennas with highest
SNRs are combined at the secondary receiver. For coverage
and reliability enhancement, a decode-and-forward (DF) relay
is used in the secondary network to assist the secondary
transmission. Note that the transmit powers at the secondary
source (S) and the secondary relay (R) are limited by two
constraints: 1) the peak interference constraint at the primary
receiver, and 2) the peak transmit power constraint at S and R.
It is also important to note that the performance of underlay
spectrum sharing is typically restricted due to these two strict
power constraints. With the help of TAS/GSC relaying, less
transmit power is required at S and R, which in turn reduces
the interference at the PU, allowing for high speed data
services over wide area coverage. The main contributions of
this paper are summarized as follows.
• We derive new exact closed-form expressions for the
cumulative distribution function (CDF) of the SNR with
TAS/GSC. Although the CDF expressions were presented
in [19, 24] with the aid of the trapezoidal rule, they are
not in closed-form and cannot be used to derive the CDF
of the SNR with TAS/GSC.
• We derive new exact closed-form expressions for the
outage probability and the symbol error rate (SER) to
accurately assess the joint impact of antenna configura-
tion and channel fading. We further derive the asymptotic
expressions for the outage probability and the SER under
the two cases: 1) proportional interference power con-
straint, and 2) fixed interference power constraint. We
confirm that the full diversity order is achieved for the
ND antennas, and PU with a single antenna. We assume that
the PU transmitter is located far away from the secondary
network. This assumption is typical in large scale networks
where the interference from the PU transmitter is negligible [6,
25, 26]. We also assume there is no direct link between S and
D due to long distance and deep fades [27]. Both the primary
channel and the secondary channel are assumed to undergo
quasi-static fading with independent and identically distributed
(i.i.d.) Nakagami-m distribution. We assume perfect channel
state information (CSI) between the secondary transmitter and
the PU can be obtained through direct feedback from the PU
[28], indirect feedback from a third party, and periodic sensing
of pilot signal from the PU [29]. In the secondary network, a
single transmit antenna among NS antennas which maximizes
the GSC output SNR at R is selected at S, while the LR
(1 ≤ LR ≤ NR ) strongest receive antennas are combined at R.
The signal transmitted by R is decoded and forwarded using a
single transmit antenna among NR antennas which maximizes
the GSC output SNR at D, and then combined at D with the
LD (1 ≤ LD ≤ ND ) strongest receive antennas. Let {g1ij }
denote the channel coefficients of the NS × NR channels from
S to R with i ∈ {1, . . . , NS }, j ∈ {1, . . . , NR }, and {g2jk }
denote the channel coefficients of the NR × ND channels from
R to D with k ∈ {1, . . . , ND }. Also, {h1i } denote the channel
coefficients of the NS × 1 channels from S to PU, and {h2j }
denote the channel coefficients of the NR × 1 channels from
R to PU. The channel coefficients follow the Nakagami-m
distribution with fading parameters mg1 , mg2 , mh1 , and mh2 ,
and average channel power gains Ωg1 , Ωg2 , Ωh1 , and Ωh2 . In
the following, k · k is the Euclidean norm, | · | is the absolute
value, and E[·] is the expectation.
The pilot symbol block Pi , (1 ≤ i ≤ NS ), are transmitted
from each transmit antenna at different time slots. Based on
these
 pilot symbols, R perfectly estimates CSI,
 then  arranges
g1ij 2 NR in descending order as g1i(1) 2 ≥ g1i(2) 2 ≥
 
proportional interference power constraint. For the fixed
j=1
interference power constraint, the diversity order is zero  2
with error floors in the high SNR regime.
• We derive an exact closed-form expression for the ergodic
capacity. Notably, this is the first closed-form expression
for cognitive relay networks with TAS/GSC in Nakagami-
m fading channels. More importantly, we obtain a tight
· · · ≥ g1i(NR )  ≥ 0 for the each transmit antenna i at S.
Note that before the transmission process, the selected number
of antenna chains LR and LD at the receivers are determined
by the limited number of radio frequency (RF) chains due
to size and complexity limitations. According to the rule of
GSC, the first LR (1 ≤ LR ≤ NR ) received signal power(s)
high SNR approximation of the ergodic capacity for the L R 
g1i(j) 2 . The selected
P 
two cases: 1) proportional interference power constraint, are comibined at R to obtain θi =
j=1
and 2) fixed interference power constraint. Interesting transmit antenna i∗ is determined at R by
conclusions are reached. On the one hand, the high
SNR slope is independent of the antenna configuration  
LR
and the fading parameters, but on the other hand, the  X 
high SNR power offset is fully described by the antenna i∗ = arg max θi = |g1i(j) |2 , (1)
1≤i≤NS 
j=1

configuration and the fading parameters in the primary
and secondary networks. The high SNR slope is 1/2
for the proportional interference power constraint, and is which maximizes the total received signal power. To this end,
equal to zero for the fixed interference power constraint. the index of the selected transmit antenna is sent back to S
through the feedback channel, so that only dlog2 (NS )e bits
II. S YSTEM AND C HANNEL D ESCRIPTION needs to be sent to S.As such, the selected channel vector is
We consider a dual-hop cognitive DF relay network con- denoted as g1i∗ θi∗ = g1i∗ (1) , · · · , g1i∗ (LR ) . Similarly, in the
sisting of S with NS antennas, R with NR antennas, D with second hop, the index of the selected transmit antenna at R is
determined by |SK | is the cardinality of the set SK , and SK denotes a set
( LD
) of (2mg1 + 1)-tuples satisfying the following condition
g2j(k) 2
X
j ∗ = arg max θj =
 
. (2) SK = nΦ
 Φ F F


k,0 , . . . , nk,mg1 −1 , nk,0 , . . . , nk,mg1

1≤j≤NR
k=1
mg1 −1 mg1
As
 such, we denote the  selected channel vector as g2j ∗ θj∗ =
X X
nΦ nF

k,i = LR − 1; k,j = NR − LR ,
g2j ∗ (1) , · · · , g2j ∗ (LD ) . i=0 j=0
According to underlay cognitive relay networks, the transmit
mg1 +LR −2 mg1 +NR −LR


k
powers at S and R are constrained as thereby |SK | = mg1 −1 mg1 , and SR =
F
 Q   Q  
 mg1 LP
R +bk
nρk ,0 , . . . , nρk ,mg1 LR +bFk  nρk ,n = nτ,k , k =

PS = min P, 2 and P R = min P, 2 ,
|h1i∗ | |h2j ∗ |   Φ n=0 F 
(3) 1, · · · , |SK |, with nΦ k,i , nk,i , nk,j , and nρk ,n ∈
Z+ . In (6), ~k , θk , and ηk are respectively given by
respectively, where P is the maximum transmit power con- mg1 L F
R +bk
Q n
straint at S and R, and Q is the peak interference power |SK | 
nτ,k  `k (n) ρk ,n 
Φ F n2 − 1 !
Y n=0
constraint at PU. ~k = ak ak ,
The instantaneous end-to-end SNR of the spectrum sharing L R n2 mg1 LR +bF
Q k
k=1
nρk ,n !
network with TAS/GSC and DF relaying is defined as γ = n=0
min {γ1 , γ2 }, where the instantaneous SNR of S → R link is (7)
2 |SK | mg1 LR +bF

2 kg1i∗ θi∗ k γ̄Q  X X k

γ1 = min kg1i∗ θi∗ k γ̄P , 2 (4) θk = µk (n)nρk ,n , (8)

|h1i∗ | k=1 n=0
and the instantaneous SNR of R → D link is and
g2j ∗ θ ∗ 2 γ̄Q 

 2 |SK | mg1 LR +bF
k
j
γ2 = min g2j ∗ θj∗ γ̄P , . (5)
X X
ηk = νk (n)nρk ,n , (9)

2
|h2j ∗ |
k=1 n=0
In (4) and (5), we define γ P = NP0 and γ Q = NQ0 , where where n2 , µk (n), νk (n), `k (n) aΦ F Φ F
k , ak , bk , bk , Υk1 , Υk2 ,
N0 is the noise power of an additive white Gaussian noise Υk3 , and Υk4 are defined in Appendix A.
(AWGN).
Proof. See Appendix A.
III. N EW S TATISTICAL P ROPERTIES 2
In this section, we derive new statistical properties of the B. Expressions for the CDF of g2j ∗ θj∗ in the Secondary
end-to-end SNR, which is a challenging problem due to the Channel
2
complex nature of TAS/GSC in Nakagami-m fading. Based The CDF of g2j ∗ θj∗ follow from (6) by interchang-
on these statistical characteristics, we present the exact and ing the parameters mg1 → mg2 , mh1 → mh2 , LR →
asymptotic outage probability, SER, and ergodic capacity. LD , NR → ND , NS → NR , SR → SD , |SK | →
|S | |S |
Without loss of generality, these new statistics can be easily |ST |, SR K → SD T , ~k → ~t , θk → θt , and
applied to other wireless networks with TAS/GSC. ηk → ηt , where ] ,
P PP P
··· ···
P
, SD =
Based on the expressions of γ1 and γ2 in (4) and (5), |S |
SD T
SD SD
1 t
SD |S |
SD T
2
respectively, we first derive the CDF of kg1i∗ θi∗ k in the  |S

 P T|
following lemmas. nτ,1 , . . . , nτ,|ST |  nτ,t = NR with {nτ,t } ∈ Z+ , |ST |
t=1
2
A. Expressions for CDF of kg1i∗ θi∗ k in the Secondary Chan- is the cardinality of the set ST , and ST denotes a set of
nel (2mg2 + 1)-tuples satisfying the following condition
2
Lemma 1. The expressions for the CDF of kg1i∗ θi∗ k are
 Φ
nt,0 , . . . , nΦ F F


ST = t,mg2 −1 , nt,0 , . . . , nt,mg2

derived as mg2 −1 mg2
  X X
LR NR NS nΦ nF

t,i = LD − 1; t,j = ND − LD ,

Fkg ∗ k2 (x) =
NS !
1i θi∗ mg1 − 1 ! LR i=0 j=0

mg2 +LD −2 mg2 +ND −LD

]
X


~k xθk e−ηk x , (6) thereby |ST | = mg2 −1 mg2 , and
F
|S |  m g2 LD
P t+b
SR K t


SD = nρt ,0 , . . . , nρt ,mg2 LD +bFt  nρt ,n = nτ,t ,

P
] PP P P   Φ  n=0
, ··· ··· SR =

where , t = 1, · · · , |ST |, with nΦ t,i , nt,i , nF
t,j , and nρt ,n ∈
|S | SR SR
1 k
SR |S |
SR K SR K Z+ . Note that ~t , θt , ηt follow from (7), (8), (9) by inter-

 |SP
K| changing the parameters µk (n) → µt (n), νk (n) → νt (n),
nτ,1 , . . . , nτ,|SK |  nτ,k = NS with {nτ,k } ∈ Z+ , `k (n) → `t (n) aΦ Φ Φ Φ Φ Φ F F
k=1 k → at , bk → bt , ck → ct , ak → at ,
bF F F F
k → bt , ck → ct , Υk1 → Υt1 , Υk2 → Υt2 , Υk3 → Υt3 , relaying in Nakagami-m fading is derived as
and Υk4 → Υt4 . Here, µt (n), νt (n), `t (n) aΦ Φ Φ F F
t , bt , ct , at , bt ,
F Pout (γth ) =
ct , Υt1 , Υt2 , Υt3 , and Υt4 follow from (A.15), (A.16), (A.14),
(A.6), (A.7), (A.9), (A.10), (A.17), (A.18), (A.19), and (A.20)  LR
  NS
NR

]
X
by interchanging the parameters mg1 → mg2 , mh1 → mh2 , 1− 1−
NS ! ~k Ξk (γth )
mg1 − 1 ! LR
Ωh1 → Ωh2 , LR → LD , NR → ND , NS → NR , SR → SD , S |
SR K
|
|S | |S |
|SK | → |ST |, SR K → SD T , nτ,k → nτ,t , nΦ k,i → nt,i ,
Φ
 N
ND
R

Φ Φ F F
nk,i → nt,i , nk,j → nt,j , and nρk ,n → nρt ,n .
 LD ]
X 
1−
NR ! ~t Ξt (γth ) . (13)
mg2 − 1 ! LD S | T|
C. Expressions for the CDF of γ1 SD

2 2
With the help of the CDF of kg1i∗ θi∗ k and |h1i∗ | , the Our new closed-form expression for the outage probability
closed-form CDF of γ1 is evaluated in the following lemma. is valid for an arbitrary number of antennas of the secondary
Lemma 2. The expression for the CDF of γ1 is represented network and arbitrary fading severity parameters in all the
as links.
 
LR NR
N S ]
X
Fγ1 (x) =
NS ! ~k Ξk (x), (10) B. Asymptotic Analysis
mg1 − 1 ! LR
|SK |
SR
1) Proportional Interference Power Constraint:
where We first examine the asymptotic behavior with the propor-
tional interference power constraint. As such, we assume that
Γ mh1 , m h1 Q


Ωh1 P
x
θk −ηk γ̄x both P and Q grow large in the high SNR regime. This applies
Ξk (x) = 1 −
e P +
Γ mh1 γ̄P to the scenario where the PU is able to tolerate a high amount
mh1 ηk x Q


of interference from S and R. With this in mind, we take into
mh1
mh1 x
θk Γ θk + mh1 , Ωh1 + γ̄Q P

θ +m . (11) account and study the effect of the so-called power scaling on
Ωh1 γ̄Q mh1 − 1 ! mh1 + ηk x k h1


Ωh1 γ̄Q the outage probability. Similar to [7, 25], we consider Q = µP ,
where µ is the power scaling factor and is a positive constant.
Proof. See Appendix B.
Theorem 2. When Q scales with P , the asymptotic outage
D. Expressions for the CDF of γ2 probability of cognitive spectrum sharing with TAS/GSC and
Similarly, the CDF of γ2 follows from (10) and (11) by DF relaying in Nakagami-m fading at high SNRs is derived
interchanging the parameters mg1 → mg2 , mh1 → mh2 , as
Ωh1 → Ωh2 , ηk → ηt , and θk → θt . Note that our expressions ∞ −Gd
+ o γ P −Gd ,


are valid for arbitrary fading severity parameters in all the Pout (γth ) = (Gc γ P ) (14)
links.
where the diversity order is
IV. O UTAGE P ROBABILITY
Gd = NR × min {mg1 NS , mg2 ND } (15)
In this section, we concentrate on the outage probability.
We derive a new closed-form expresssion for the exact outage and the SNR gain is
probability. In order to assess the performance at high SNRs,
we derive the asymptotic outage probabilities with the propor-
 ∆1 (LR )
 γth
 mg1 NS < mg2 ND
tional interference power constraint and the fixed interference ∆2 (LR )
Gc = γth mg1 NS > mg2 ND (16)
power constraint.  ∆1 (L R) ∆2 (LR )

γth + γth mg1 NS = mg2 ND ,
A. Exact Analysis
with
In DF relaying, the end-to-end outage probability is deter- Φ
Ωg1  K SK , NR , LR , mg1 , aΦ


Φ 
k , bk − m 1N
mined by the worst link between S → R and R → D links, ∆1 (LR ) = g1 R
mg1 (mg1 NR )!
which is given by [30]


Ξ mg1 , mh1 , Ωh1 , NS − m N1 N
Pout (γth ) = Pr (min(γ1 , γ2 ) ≤ γth ) Φ mh1 , Ωh1 + g1 R S , (17)
µmg1 NR NS
= Fγ1 (γth ) + Fγ2 (γth ) − Fγ1 (γth ) Fγ2 (γth ) . (12)
and

By substituting (10) and the CDF of γ2 into (12), the outage Ωg2  K STΦ , ND , LD , mg2 , aΦ Φ
t , bt − mg2 ND
1
∆2 (LR ) =
probability is finally derived in the following theorem. mg2 (mg2 ND )!


Theorem 1. The closed-form expression for the outage prob- 
Ξ mg2 , mh2 , Ωh2 , ND − m N1 N
Φ mh2 , Ωh2 + g2 R D . (18)

ability of spectrum sharing networks with TAS/GSC and DF µmg2 NR ND

In (17) and (18), we have
∞
N

Pout (γth )
Φ Φ Φ

L L
K S , N, L, mg , a , b = N −L

H1 Φ1 γγ̄th

mg1 NR NS
+ Ξ1 γγ̄th

mg1 NR NS

(mg − 1)!(mg !) 

 P Q

X (bΦ + mg (N − L + 1) − 1)!
 m N <m N

g1 S
g2 D


aΦ γth mg2 NR ND γth mg2 NR ND

, (19) 

bΦ +mg (N −L+1)
 H
 2 2 γ̄P
 Φ + Ξ 2 γ̄Q
S Φ (L)

j = mg1 NS > mg2 ND
mXh −1

m
−µ Ω h µm
Ωh
h
 H1 Φ1 + H2 Φ2 γth mg1 NR NS




Φ mh , Ωh = 1 − e h , (20) 
 γ̄P
j=0
j! 

+ H Ξ + H2 Ξ2 γγ̄th


mg1 NR NS
1 1


 Q

Γ(mg NR N + mh , µ m

Ωh ) mg1 NS = mg2 ND ,
h

Ξ mg , mh , Ωh , N = . (21) (25)
h mg NR N
(mh − 1)!( m
Ωh )
where
Proof. See Appendix C. mg1
mg1 NR Φ


 Ωg1 SK
K , NR , LR , mg1 , aΦ Φ
k , bk NS
H1 = (, 26)
(mg1 NR )!
Based on (15), we see that the diversity order is dominated mg2 mg2 ND

 ( Ωg2 ) K STΦ , ND , LD , mg2 , aΦ Φ
t , bt NR
by the fading severity parameter of the two hops and the total H2 = (27)
number of antennas at S, R, and D. Interestingly, it is inde- (mg2 ND )!
pendent of the fading severity parameters of the interference Φ1 = Φ (mh1 , Ωh1 ), Φ2 = Φ (mh2 , Ωh2 ), (28)
channel, and the selected number of antennas at R and D. The Ξ1 = Ξ (mg1 , mh1 , Ωh1 , NS ), (29)
negative impact of the peak interference power constraint is
reflected in the SNR gain. and

Corollary 1. The SNR gap between GSC and SC is derived Ξ2 = Ξ (mg2 , mh2 , Ωh2 , ND ). (30)
as Proof. The proof can be done in the same way as the proof
of Theorem 2.
− 10 log (T1 )

mg1 NS < mg2 ND
 mg110NR

Gc = − mg2 ND log (T2 ) mg1 NS > mg2 ND From (25), we see that the diversity order of the outage
probability tends to zero under the fixed interference power
 
 10 log ∆1 (LR )+∆2 (LR )

mg1 NS = mg2 ND ,
∆1 (1)+∆2 (1) constraint.
(22)
where V. S YMBOL E RROR RATE
N −1 In this section, we focus on the SER as another important
(mg1 !) R (mg1 − 1)! Φ Φ Φ performance evaluation metric. For most modulation schemes,
T1 = K(SK , NR , LR , mg1 , ak , bk ) (23)
NR (mg1 NR − 1)! the SER of a conventional wireless communication system can
be expressed as [32]
and r Z∞
a b e−bγ
N −1
(mg2 !) D (mg2 − 1)! Pe = √ Fγ (γ)dγ, (31)
T2 = K(STΦ , ND , LD , mg2 , aΦ , bΦ
). 2 π γ
t t 0
ND (mg2 ND − 1)!
(24) where a and b are modulation specific constants. For example,
a = 1, b = 1 for BPSK (binary phase shift keying), a =
2) Fixed Interference Power Constraint: 2(M − 1)/M , b = 3/(M 2 − 1) for M-PAM (M-ary pulse
2
Different from the proportional interference power con- amplitude modulation), and a = 2, b = sin (π/M ) for M-
straint which can tolerate an extremely high peak interference PSK (M-ary phase shift keying).
power constraint and may potentially violate and harm the A. Exact Analysis
PU transmission [6], in this subsection, we focus on a stricter
constraint where the peak interference power constraint is fixed Substituting (10) into (31), the SER of S → R link can
[31]. We present the asymptotic outage probability with the be derived by utilizing [33, eq.8.310.1], [33, eq.8.352.2], [33,
fixed interference power constraint in the following theorem. eq.9.211.4.8] and the polynomial expansion. Using the same
method, Pe2 , which is the SER of R → D link can be easily
Theorem 3. Under the fixed interference power constraint, the computed. Substituting the derived expressions of Pe1 and Pe2
asymptotic outage probability of cognitive spectrum sharing into
with TAS/GSC and DF relaying in Nakagami-m fading at high


SNRs is derived as Pe = 1 − 1 − Pe1 × 1 − Pe2 , (32)
yields the SER of cognitive relay networks with TAS/GSC and and ∆1 and ∆2 are given in (17) and (18), respectively.
DF relaying in the following theorem.
Based on (35), we find that the diversity order is inde-
Theorem 4. The closed-form expression for the SER of pendent of the modulation scheme and the peak interference
cognitive spectrum sharing with TAS/GSC and DF relaying power constraint Q. The fading severity parameters of each
in Nakagami-m fading is derived as hop and the antenna configuration have a direct impact on the
r   NS diversity order while the interference power constraint at PU
a b LR NR

Pe = 1 − 1 −
NS ! has a direct impact on the SNR gain.
2 π mg1 − 1 ! LR 2) Fixed Interference Power Constraint:
]X
Substituting (25) into (31), we derive the asymptotic SER
~k Π (mh1 , Ωh1 , θk , ηk )
under the fixed interference power constraint in the following
|S |
SR K theorem.
r   N
a b LD ND
R Theorem 6. Under the fixed interference power constraint, the
1−
NR !
2 π mg2 − 1 ! LD asymptotic SER of cognitive spectrum sharing with TAS/GSC
and DF relaying in Nakagami-m fading at high SNRs is
]
X
derived as
~t Π (mh2 , Ωh2 , θt , ηt ) , (33)
|ST | Pe∞
SD 
mg NR NS
mg NR NS

where 
 Θ1 Φ1 γ̄1P + Ξ1 γ̄1Q

 m N < mg2 ND ,
Γ mh , m hQ

mg2 Ng1R NSD



1


Ωh P
Γ θ + 2 1
θ


mg1 NR ND
1 1

Π (mh , Ωh , θ, η) = 1 −

 Θ2 Φ2 γ̄P + Ξ2 γ̄Q

1


Γ mh b + γ̄ηP

θ+ 2 γ̄P
= mg1 NS > mg2 ND ,
1
mh Q 1
θ+ 12 
1
mg NR NS

θ + mh − 1 !e− Ωh P Θ1 Φ1 + Θ2 Φ2 γ̄P

+


mh − 1 ! η


mg NR NS
+ Θ1 Ξ1 + Θ2 Ξ2 γ̄1Q




θ+mh −1

1 Q
m mh
m+ 12 12

1
X mg1 NS = mg2 ND ,
Γ θ+ γ̄Q
2 m! P Ωh (40)
m=0
1 3 ηQ
γ̄Q mh
 where
Ψ θ + , m − mh + ; b + . (34)
2 2 γ̄Q P ηΩh aΓ(mg1 NR NS + 21 )
Θ1 = √ H1 , (41)
B. Asymptotic Analysis 2 πbmg1 NR NS

aΓ(mg2 NR ND + 21 )
1) Proportional Interference Power Constraint: Θ2 = √ H2 , (42)
Substituting (14) into (31), together with the help of [33, eq. 2 πbmg2 NR ND
(3.310)], we derive the asymptotic SER under the proportional and H1 , H2 , Φ1 , Φ2 , Ξ1 , and Ξ2 are given by (26), (27), (28),
interference power constraint in the following theorem. (28), (29), (30), respectively.
Theorem 5. When Q is proportional to P , the asymptotic SER From (40), we find that the diversity order of the SER goes
of cognitive spectrum sharing with TAS/GSC and DF relaying to zero under the fixed interference power constraint.
in Nakagami-m fading at high SNRs is derived as VI. E RGODIC C APACITY
−G
Pe∞ = (Gc γ P ) d + o γ P −Gd ,


(35) The ergodic capacity is an important performance indicator
for cognitive underlay spectrum sharing. It is defined as the
where the diversity order is
maximum achievable long-term rate, where no delay limit
Gd = NR × min {mg1 NS , mg2 ND } (36) is taken into account. Under these assumptions, the ergodic
capacity is expressed as
and the SNR gain is

 Λ1 ∆1 mg1 NS < mg2 ND Z∞ Z∞
1 1 1 − Fγ (x)
Gc = Λ2 ∆2 mg1 NS > mg2 ND (37) Cerg = log2 (1 + x) fγ (x)dx = dx.

Λ 1 ∆ 1 + Λ 2 ∆2 mg1 NS = mg2 ND , 2 2 ln 2 1+x
0 0
where (43)
har 1 1
i− mg1 N1R NS To simplify (43), we define Fγ1 (x) = 1 + F̃γ1 (x) and
Λ1 = Γ mg1 NR NS + b, (38) Fγ2 (x) = 1 + F̃γ2 (x), and rewrite (43) as
2 π 2
Z∞
har 1 1 F̃γ1 (x) F̃γ2 (x)
1
i− mg2 N1R ND Cerg = dx, (44)
Λ2 = Γ mg2 NR ND + b, (39) 2 ln 2 1+x
2 π 2 0
 0 0
10 10
−1
Simulation −1 Simulation
10 10
Exact −2 Exact
−2
10 Asymptotic 10
Asymptotic

Outage Probability
Outage Probability

−3
10
−3 L R = LD = 1 10
−4
2 −4
10 3 10 L R = LD = 1
−5 µ= 2 −5 2
10 10
µ = 0.5 3
−6 −6
10 10
L R = LD = 1
−7
10
−7 2 10
3 −8
−8
10 10
−9 −9
10 10
0 5 10 15 20 25 30 0 5 10 15 20 25 30
γP (dB) γP (dB)
Fig. 1. Cognitive spectrum sharing with TAS/GSC and DF relaying: NS = 2, Fig. 2. Cognitive spectrum sharing with TAS/GSC and DF relaying: NS = 2,
NR = 3, ND = 3, mg1 = 1, mg2 = 2, mh1 = mh2 = 2, and γ Q = 2γ P . NR = 3, ND = 3, mg1 = 1, mg2 = 2, mh1 = mh2 = 2, and γ Q =
20 dB.

where 4 τ −l
  ν (η, l, k1 , k2 ) = Γ (τ ) (γ̄Q mh /ηΩh ) Ψ (τ , τ + 1 − l;
LR NR
NS ]
X
F̃γ1 (x) =
NS ! ~k sgn (ηk ) Ξk (x) (ηt + ηk ) γ̄Q mh /γ̄p ηΩh ) , (51)
mg1 − 1 ! LR S | |
SR K
(45) l−1
4 (γ̄Q mh /ηΩh − 1)
∂ (η, l) = θt +mh θk +mh
,
and (γ̄Q mh /ηt Ωh − 1) (γ̄Q mh /ηk Ωh − 1)
θ+mh −l+1 j+θ+mh −l−1

 
LD ND
NR ]
X
4 (−1)
F̃γ2 (x) =
NR ! ~t sgn (ηt ) Ξt (x). κ (θ, η, l, j) =
j−1
,
mg2 − 1 ! LD (γ̄Q mh /Ωh )
j+θ+mh −l
(1/ηt − 1/ηk )
j+θ+mh −l
S | |
SD T
(46) (52)

In the following, we assume mh1 = mh2 = mh and Ωh1 = with τ = θk + k1 + θt + k2 + 1.

Ωh2 = Ωh .
Our result can be applied and simplified to the special cases
A. Exact Analysis of TAS/MRC and TAS/SC in Nakagami-m fading channel, as
Substituting (45) and (46) into (44), and with the help of well as TAS/GSC in Rayleigh fading channels.
[33, eq.8.352.2], [33, eq.9.211.4.8], and the partial fraction
expression [33, eq.2.102], we obtain a general closed-form
expression for the ergodic capacity in the following theorem. B. High SNR Capacity analysis

Theorem 7. Our new closed-form expression for the ergodic

capacity of cognitive TAS/GSC relaying in Nakagami-m fading To examine the capacity performance in the high SNR
is given in (47) at the top of the next page. In (47), we have regime with γ P → ∞, we derive the high SNR approximation
defined the following terms o the ergodic capacity in closed-form. With the aid of the
Jensen’s inequality, a tight upper bound on the ergodic capacity
4
 Q mh

  1

θ
is given by [34]
∇ (θ) = 1 − Γ mh , Γ (mh ) , (48)
P Ωh γ̄P 1 1

θ+m h −1 Cerg = E [log2 (1 + γ)] ≤ log2 E (1 + γ) . (53)
4 θ + m h − 1 ! 1
θ − m hQ X 1 2 2
∆ (θ, η, j, k) =
e Ωh P
mh − 1 ! γ̄ Q j=0
j!
Thus, the tight high SNR approximation of the ergodic capac-
Q
j X j  
j mh
mh +j−k η
k ity is presented as [34, 35]
, (49)
P k Ωh γ̄Q ∞ 1 1
k=0
Cerg ≈ log E (1 + γ) ≈ log2 E (γ) . (54)
(50) 2 2 2
     "
1  LR NR NS ]
X  LD ND NR ]
X
Cerg =
NS ! ~k sgn (ηk )
NR ! ~t sgn (ηt ) ∇ (θk )
2 ln 2 mg1 − 1 ! LR mg2 − 1 ! LD
S | | S | |
SR K SD T
 θt +m
Ph 
ν(ηt ,l2 ,0,k2 )
ν (γ̄Q mh /Ωh , 1, 0, k2 ) − (γ̄Q mh /ηt Ωh −1)1−l2
l2 =1
∇ (θt ) ν (γ̄Q mh /Ωh , 1, 0, 0) + ∆ (θt , ηt , j2 , k2 ) ∇ (θk ) θt +mh
(mh /Ωh − ηt /γ̄Q )
 θk P
+mh 
ν(ηk ,l1 ,k1 ,0)
ν (γ̄Q mh /Ωh , 1, k1 , 0) − (γ̄Q mh /ηk Ωh −1)1−l1
l1 =1
+ θk +mh
∇ (θt ) ∆ (θk , ηk , j1 , k1 ) + ∆ (θk , ηk , j1 , k1 )
(mh /Ωh − ηk /γ̄Q )
 θkX
+mh 
γ̄Q
θk +mh γ̄Q
θt +mh
∆ (θt , ηt , j2 , k2 ) ∂ (η, 1) ν (γ̄Q mh /Ωh , 1, k1 , k2 ) + −∂ (ηk , l1 )+
ηk ηt
l1 =1
θtX
+mh
θtX
+mh 
κ θk , ηk , l1 , j 

θt +mh −j+1 ν (ηk , l1 , k1 , k2 ) + sgn (|ηk − ηt |) ν (ηt , l2 , k1 , k2 ) −∂ (ηt , l2 )
j=1 γ̄Q mh ηt Ωh − 1 l2 =1
θkX+mh i+θ +m −l  θk +θX t +2mh #
(−1) t h 2 (1 − sgn (|ηk − ηt |)) ν (ηk , l3 , k1 , k2 )
+ θk +mh −i+1
κ (θt , ηt , l2 , i) − θk +θt +2mh −m+1
. (47)
i=1 (γ̄Q mh /ηk Ωh − 1) l3 =θk +mh +1 (γ̄Q mh /ηk Ωh − 1)

Therefore, we can rewrite (54) as where

 
LR NR
NS ]
X
Υ=
NS ! ~k sgn (ηk )
Z∞ Z∞ mg1 − 1 ! LR
1  1   S | |
∞ SR K
Cerg ≈ log2 xfγ (x) dx = log2 (1 − Fγ (x)) dx
2 2 LD

ND
N R

]
X
0 0
NR ! ~t sgn (ηt )
Z∞ mg2 − 1 ! LD
1 S | |
SD T
= log F̃γ1 (x) F̃γ2 (x) dx. (55)
2 2 "
Xh −1
θt +m j2
0 Γ (θk + θt + 1) X
λ2 θk +θt +1
+λ ∆s (θt , j2 , k2 )
(ηk + ηt ) j2 =0 k2 =0

1) Proportional Interference Power Constraint: νs (ηt , θt , 0, k2 ) Xh −1

θk +m j1
X ∆s (θk , j1 , k1 )
θk +j2 +1
+λ θt +j1 +1
(58)
Based on (55), the high SNR approximation for the ergodic (1/µ) (1/µ)
j1 =0 k1 =0
capacity with the proportional interference power constraint is
written as Xh −1
θk +m j1
X Xh −1
θt +m j2
X
νs (ηk , θk , k1 , 0) + Ω
j1 =0 k1 =0 j2 =0 k2 =0
Z∞ #
∞ 1


x

 ∆s (θk , j1 , k1 ) ∆s (θt , j2 , k2 )
Cerg = log2 γ̄P + log2 F̃γ1 x F̃γ2 x d . θk +θt +1 j1 +j2 +1
. (59)
2 γ̄P (mh /Ωh ) (1/µ)
0
(56) In (58), we have defined
  θ +θ +j+1−θ
(θ + mh − 1)! −µ m
4 h j mh k t
∆s (θ, j, k) = e Ωh
,
Substituting (45) and (46) into (56), and with the help of [33, (mh − 1)!j! k Ωh
eq.8.352.2], [33, eq.9.211.4.8] and the binomial expansion, (60)
the high SNR approximation for ergodic capacity with the (θ+mh −l) θk +θt +2mh −l−1

4 (−1) θ+mh −l
proportional interference power constraint is derived in the κs (θ, η, l) = θ +θ +2mh −l k1 +k2 −l
, (61)
following theorem. (1/ηt − 1/ηk ) k t η
4 θ +θ +1
νs (η, ε, k1 , k2 ) = Γ (θk + k1 + θt + k2 + 1) (1/η) k t
Theorem 8. When Q is proportional to P , the high SNR 
approximation of the ergodic capacity is derived as Ψ θk + k1 + θt + k2 + 1, θk + k1 + θt +k2 + 2 − mh − ε;
(ηt + ηk ) mh µ 
, (62)
∞ 1 1 Ωh η
Cerg ≈ log (γ ) + log2 (Υ) , (57)
2 2 P 2 and
 0 0
10 10
Simulation Simulation
−2 −2
10 Exact 10 Exact
Asymptotic Asymptotic
Symbol Error Rate

Symbol Error Rate

−4 −4
10 10

−6 −6
10 10
QPSK L R = LD = 1
−8
10
−8
10 QPSK 2
BPSK L R = LD = 1 3
2 BPSK
−10 −10
10 L R = LD = 1 3 10
L R = LD = 1
−12
2 −12
2
10 3 10 3
0 5 10 15 20 25 30 0 5 10 15 20 25 30
γP (dB) γP (dB)
Fig. 3. Cognitive spectrum sharing with TAS/GSC and DF relaying: NS = 2, Fig. 4. Cognitive spectrum sharing with TAS/GSC and DF relaying: NS = 2,
NR = 3, ND = 3, mg1 = 1, mg2 = 2, and γ Q = 2γ P . NR = 3, ND = 3, mg1 = 1, mg2 = 2, and γ Q = 25 dB.

From (65), we see that the high SNR slope S∞ is indepen-

4 dent of the interference power constraint, the selected number
Ω = (1 − sgn (|ηk − ηt |)) νs (ηk , θk + θt + mh , k1 , k2 ) ,
θkX+mh of antennas at the receiver, and the primary network. We also
sgn (|ηk − ηt |) see that the high SNR power offset L∞ is independent of γ P
+ θ +m −k1 θt +m −k2 κs (θk , ηk , l1 )
ηk k h ηt h
from (66).
l1 =1
θtX
+mh 2) Fixed Interference Power Constraint:
θk +θt +2mh −l2 Substituting (45) and (46) into (55), we obtain the high
νs (ηk , l1 − mh , k1 , k2 ) + (−1)
l2 =1 SNR approximation of the ergodic capacity under the fixed
 interference power constraint.
κs (θt , ηt , l2 ) νs (ηt , l2 − mh , k1 , k2 ) . (63)
Theorem 9. When Q is fixed, the high SNR approximation
Note that similar as the asymptotic ergodic capacity, the of the ergodic capacity is given in (67) at the top of the next
tight high SNR approximations can well predict the perfor- page.
mance behaviours in the high SNR regime. Thus, we deduce From (67), we find that for the fixed interference power
the high SNR scaling law from the high SNR approximations constraint, the high SNR slope collapses to zero.
similar to the approach in [36] and [37]. Based on (57),
we characterize two key parameters determining the affine VII. N UMERICAL R ESULTS
approximation of the ergodic capacity in the high SNR regime, In this section, we present numerical results to verify our
namely the high SNR slope and the high SNR power offset new analytical derivations for cognitive TAS/GSC relaying in
[38]. The high SNR slope is also known as the degrees Nakagami-m fading channels. We set the threshold SNR as
of freedom or the multiplexing gain [39]. The high SNR γth = 5 dB. All the figures clearly show that the exact curves
power offset captures the joint effects of the fading model, are in precise agreement with the Monte Carlo simulations.
the number of antennas at each terminal, and the interference Importantly, the asymptotic lines accurately predict the exact
power constraint. We represent the high SNR approximation behaviour in the high SNR regime.
of the ergodic capacity as [38] Fig. 1 plots the outage probability with the proportional
∞
Cerg ≈ S∞ (log2 (γ P ) − L∞ ) , (64) interference constraint as we vary µ, LR and LD . The exact
and asymptotic curves are plotted by using (13) and (14),
where S∞ is the high SNR slope in bits/s/Hz/(3 dB) respectively. For the same µ, we observe that the outage
∞
Cerg 1 probability decreases with increasing LR and LD , due to
S∞ = lim = (65) an increase in the SNR gain (16). We also confirm that the
γ P →∞ log2 (γ P ) 2
and L∞ is the high SNR power offset in 3 dB units diversity order is independent of LR and LD as reflected
∞  by the parallel slope. Another observation is that the outage
 Cerg probability decreases with increasing µ, which is due to the
L∞ = lim log2 (γ P ) − = −log2 (Υ) . (66)
γ P →∞ S∞ relaxed interference power constraint at the PU receiver.
 "     
∞ 1  LR NR NS ]
X  LD ND  N R ]
X
Cerg = log2 NS ! ~k sgn (ηk ) NR ! ~t sgn (ηt ) λ2 γ̄P
2 (mg1 − 1)! LR (mg2 − 1)! LD
|S | |S |
SR K SD T

Γ (θk + θt + 1)
 θt +mXh −1 X j2
γ̄Q
θk +j2 Xh −1 X
θk +m j1
γ̄Q
θt +j1
θk +θt +1
+ γ̄ Q λ ∆ s (θ , j ,
t 2 2 sk )ν (η ,
t tθ , 0, k2 ) + λ
(ηk + ηt ) j2 =0
γ̄P j1 =0
γ̄P
k2 =0 k1 =0

Xh −1 X
θk +m j1 θt +mXh −1 X j2  #
∆s (θk , j1 , k1 ) ∆s (θt , j2 , k2 )
∆s (θk , j1 , k1 ) νs (ηk , θk , k1 , 0) + θk +θt +1 j +j
Ω . (67)
j1 =0 k1 =0 j2 =0 k2 =0 (mh /Ωh ) (P /Q) 1 2

6 6

5 5

Ergodic Capacity (bits/s/Hz)

Ergodic Capacity (bits/s/Hz)

4 4

3 3
L R = LD = 3 L R = LD = 3
2 2
2 1 2 1
Simulation Simulation
1 Exact 1 Exact
Asymptotic Asymptotic
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 35
γP (dB) γP (dB)
Fig. 5. Cognitive spectrum sharing with TAS/GSC and DF relaying: NS = 2, Fig. 6. Cognitive spectrum sharing with TAS/GSC and DF relaying: NS = 2,
NR = 3, ND = 3, mg1 = 1, mg2 = 2, mh1 = mh2 = 2, and γ Q = 2γ P . NR = 3, ND = 3, mg1 = 1, mg2 = 2, mh1 = mh2 = 2, and γ Q =
25 dB.

Fig. 2 examines the impact of the fixed interference pow- We see that the SER decreases as LR and LD increase, and
er constraint on the outage probability as varying LR and BPSK outperforms QPSK. Similar to Fig. 2, the SER becomes
LD . The exact and asymptotic curves are plotted using (13) saturated for γ P > 22 dB, which confirms that the diversity
and (25), respectively. Interestingly, the outage probability order goes to zero.
becomes saturated for γ P > 18 . dB. This
 is due
. to the fact Fig. 5 plots the exact ergodic capacity and its high SNR
Q
that when γ P → ∞, min P, |h1i∗ | ≈ Q
|h1i∗ | and
2 2
 .   .  approximation with the proportional interference power con-
min P, Q |h2j ∗ |2 ≈ Q |h2j ∗ |2 , as such the fixed peak straint from (47) and (57), respectively. We see that the high
interference power constraint becomes the dominant factor. By SNR approximations of the ergodic capacity are tight and well
setting LR = LD = 1 and LR = LD = 3, we also see that predict the behavior of the ergodic capacity at high SNRs.
TAS/MRC outperforms TAS/GSC, and TAS/GSC outperforms It is obvious that the ergodic capacity can be improved by
TAS/SC. increasing LR and LD . The parallel curves confirm that the
high SNR slope is independent of LR and LD .
Fig. 3 plots the exact and asymptotic SER with the pro-
Fig. 6 examines the impact of the fixed interference power
portional interference power constraint from (33) and (35),
constraint on the ergodic capacity. The exact ergodic ca-
respectively. The plot confirms that the diversity order is
pacity and its high SNR approximation are from (47) and
independent of the modulation scheme, LR , and LD . We see
(67), respectively. Interestingly, we find that the capacity
that the SER decreases as LR and LD increase. We also see
ceiling occurs for γ P > 30 dB.  This is due to the
fact
that BPSK outperforms QPSK, which is predicted from the
that when γP → ∞, min P, Q h1i∗ 2 ≈ Q h1i∗ 2 and

 
SNR gain (16).
min P, Q h2j ∗ 2 ≈ Q |h2j ∗ |2 . Once again, the fixed
 


Fig. 4 plots the exact and asymptotic SER with the fixed
interference power constraint from (33) and (40), respectively. interference power constraint becomes the dominant factor.
By setting LR = LD = 1 and LR = LD = 3, we see that Based on (A.3), and using the multinomial theorem [41],
L −1
TAS/MRC outperforms TAS/GSC and TAS/GSC outperforms we rewrite (Φ (s, x)) R as
TAS/SC.

X
LR −1 m g
m g LR −1 Φ
bk −ck x Φ

VIII. C ONCLUSIONS (Φ (s, x)) = aΦ

kx e
Ωg Φ
SK
We have taken into account the cognitive DF relay network mg1
bΦk −mg1


LR −1
with TAS/GSC over Nakagami-m fading. This framework is s+ , (A.5)
Ωg1
well suited for the reliability enhancement of the secondary
network and interference alleviation of the primary network. ( )
 mg1 −1
We derived new statistical properties of the end-to-end SNR. Φ
where SK nΦ Φ

P
nΦ
= k,0 , . . . , nk,mg1 −1  k,i = LR − 1

Based on these, we have derived closed-form expressions for i=0
the exact and asymptotic outage probability, symbol error rate,
n o
Φ + Φ Φ Φ
with nk,i ∈ Z , ak , bk , and ck are, respectively, given by
and ergodic capacity with the proportional and the fixed inter-
ference power constraints. Our results are valid for Nakagami-

mg1 −1 Φ mg1 −1
m fading and arbitrary number of antennas in the secondary LR − 1 ! Y 1
nk,i Φ X
network. Based on the relationship of the maximum transmit aΦ
k = m −1 , bk = nΦ
k,i i, (A.6)
g1 i!
nΦ i=0 i=0
Q
power constraint and peak interference power constraint, we k,i !
conclude that: 1) under the proportional interference power i=0
 
constraint, the diversity order is determined by the fading mg1
and cΦ
k = (LR − 1) s + . (A.7)
parameter and the antenna configuration of the secondary Ωg1
network, and the high SNR slope is 1/2; and 2) under the
fixed interference power constraint, the diversity order is zero Based on (A.2), we proceed to employ the multinomial
N −L
with error floor, and the high SNR slope is zero with capacity theorem to express (F (x)) R R as
ceiling.
NR −LR
X F F
bk −ck x
(F (x)) = aF
kx e , (A.8)
A PPENDIX A F
SK
A P ROOF OF L EMMA 1
( )
We first present the probability density function (PDF)
 m
Pg F
F
and CDF for the channel power gain of a single branch of where SK = nF F
k,0 , . . . , nk,mg1  nk,j = NR − LR
j=0
the secondary network channel with the Nakagami-m fading n o
as [40] with nF
k,j ∈ Z+
, aF
k , b F
k , and c F
k are, respectively, given by

xmg1 −1 mg1
mg1 − mg1

x
f (x) = e Ωg1 (A.1)
mg1 −1 F
NR − LR ! Y −1
nk,j+1 mg1
bFk


mg1 − 1 ! Ωg1
aF
k = mg1 , (A.9)
Q F j! Ωg1
and nk,j ! j=0
j=0
mg1

Γ mg1 , x Ωg1 mg1 −1 mg1
F (x) = 1 −
, (A.2) X mg1 X F
Γ mg1 bF
k = jnF F
k,j+1 , and ck = n . (A.10)
j=0
Ωg1 j=1 k,j
respectively. The marginal moment generating function (MGF)
of (A.1) is given by [19] Substituting (A.1), (A.5) and (A.8) into (A.4), and apply-
mg1 −1
m


− s+ Ω g1 x
ing [33, eq. (3.351.3)], ΦGSC (s) is derived as
i
mg1
mg1 X xe g1
Φ (s, x) = mg1
mg1 −i
. (A.3)
Ωg1 i=0 i! s + ΦGSC (s) =
Ωg1  
LR NR mg1
mg1 LR X X Φ F
As shown in [19, 24], the MGF expression for the channel
ak ak
mg1 − 1 ! LR Ωg1
power gain with GSC is expressed as SkΦ ∈SKΦ S F ∈S F
k K


Φ
Φ F

m g1

b −m g1 L R −1
ΦGSC (s) = bk + bk + mg1 − 1 ! s + Ωg1 k

  Z∞ . (A.11)
mg1
bΦ F
k +bk +mg1
NR L −1 N −L s + cΦ F
k + ck + Ωg1
LR e−sx f (x) (Φ (s, x)) R (F (x)) R R dx.
LR
0
(A.4) Let FGSC (x) denote the CDF of the channel power gain
of the secondary network with GSC. The Laplace transform
Here the MGF is defined as Φγ (s) = E [e−γs ]. of FGSC (x) is given by L [FGSC (x)] = ΦGSC (s) /s [20].
Therefore, the Laplace transform for FGSC (x) is are given by

mg1
sgn cF 1 X F
 
LR NR mg1
mg1 LR k

−n2
L [FGSC (x)] = Υk1 =−
nk,j + 1 , (A.17)
(mg1 − 1)! LR Ωg1 n − 1 ! LR j=1
Φ F


F bk + bk + mg1 − 1 ! X 1 m

n2 g1

1−n1 sgn cF
XX
aΦ a k
X
−(n2 −l+1)
k k
bΦ +bF +mg1 Υk2 = −1 nFk,n + 1
LR k k


Φ SF
SK K
n − 1 ! l=1 L R n=1


Φ
mg1
bk −mg1 LR −1
  mg1
s + Ωg1 l − n1 − 1 1 X F
n1 −l
. (A.12) n , (A.18)
bΦ +bF
k +mg1
l−1 LR n=1 k,n
cF mg1
k
s s + LR + Ωg1
k

mg1
sgn cFk 1 X F
−(n2 −n1 +1)
Υk3 =−
nk,j + 1 , (A.19)
Using the partial fraction expansion [33, eq. (2.102)], we n1 − 1 ! LR j=1
can rewrite (A.12) in an equivalent form. Then, taking the and
inverse Laplace transform of L [FGSC (x)], we obtain


mg1 LR −1 −bΦk
sgn cF
k
X
l+1
  X mg1X LR +bF
k Υk4 =
−1
LR NR n1 − 1 !
FGSC (x) =
aΦ F
k ak
l=1
mg1 − 1 ! LR S   mg1

K
n=0 n2 − n1 + l − 1 1 X F
−(n2 −n1 +l)
bΦ F
− n ,
k + b k + m g1 1 ! l−1 LR j=1 k,j
Φ F `k (n)xµk (n) e−νk (n)x ,
LR bk +bk +mg1 (A.20)
(A.13)
where n1 = n − mg1 (LR − 1) + bΦ k.
where the set SK has been defined in Lemma 1, `k (n), µk (n) 2
The CDF of kg1i∗ θi∗ k is given by Fkg ∗ k2 (x) =
1i θi∗
and νk (n) are, respectively, given by N
(FGSC (x)) S . Based on (A.13), and employing the multi-
2
nomial theorem, we can derive the CDF of kg1i∗ θi∗ k as (6).
`k (n)
 F mg1
−n2
mg1
µk (n)−bk 1 A PPENDIX B
nF
P
k,j + 1

Ωg1 LR

A P ROOF OF L EMMA 2

j=1



n=0


According to (4), the CDF of γ1 can be written as


mg1
µk (n)−bk
F
1−sgn(cF
k)


Υk1 + Υk2 −

= Ωg1 (n−1)!
 2 x  2 Q
Φ Fγ1 (x) =Pr g1i∗ θi∗ ≤ , h1i∗  ≤

 1 ≤ n ≤ mg1 L
R − 1 − bk

 γP P
 mg1
µk (n)−bk
F 1−sgn cF

Υk3 + Υk4 −
k

 2

 Ωg1 n−1 !  g1i∗ θi∗
x  2 Q
+ Pr ≤ , h1i∗  ≥ . (B.1)


mg1 (LR − 1) − bΦ F h1i∗ 2

k < n ≤ mg1 LR + bk , γQ P
 
(A.14)  2
The CDF of h1i∗  is expressed as

Γ mh1 , x m


h1
µk (n)
Ωh1
F ∗  x = 1 −
 
(B.2)
h1i

 0, n=0 Γ mh1

n − 1, 1 ≤ n ≤
mg1 (LR − 1) − bΦ

=
k By substituting (B.2) and (6) into (B.1), we derive the
n − sgn ck mg1 (LR − 1) − bΦ
F
k − 1,

 Φ F closed-form expression of Fγ1 (x) as (10).
mg1 (LR − 1) − bk < n ≤ mg1 LR + bk ,

(A.15) A PPENDIX C
A P ROOF OF T HEOREM 2
and
Based on (A.8), we consider transmission in the high SNR
νk (n) regime with γ̄P → ∞. Applying the Taylor series expansion
k
xj


 0, n=0 truncated to the kth order given by ex = k
P
j! + o x in

mg1
= Ωg1 , 1 ≤ n ≤ mg1 (LR − 1) − bΦ
k
j=0
(A.8), the asymptotic expression for (A.8) is written as
 cFk + mg1
mg1 (LR − 1) − bΦ F
k < n ≤ mg1 LR + bk ,

LR Ωg1 ,


(A.16) mg1
mg1 NR −LR mg1 NR −LR


NR −LR Ωg1 x
F x =
NR −LR . (C.1)
with n2 = bΦ F mg1 !
k + bk + mg1 . In (A.14), Υk1 , Υk2 , Υk3 , and Υk4
Substituting (A.1), (A.5) and (C.1) into (A.4) yields [8] C. K. Datsikas, N. C. Sagias, F. I. Lazarakis, and G. S. Tombras,

“Outage analysis of decode-and-forward relaying over Nakagami-m
Φ Φ Φ
mg1 NR K SK , NR , LR , mg1 , ak , bk

mg1 fading channels,” IEEE Signal Process. Lett., vol. 15, pp. 41–44, Jan.
ΦGSC s =( ) mg1 mg1 NR
. 2008.
Ωg1 (s + Ωg1 ) [9] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, “Space-time block
codes from orthogonal designs,” IEEE Trans. Inf. Theory, vol. 45, no. 5,
(C.2) pp. 1456–1467, Jul. 1999.
[10] H. Ding, J. Ge, D. B. da Costa, and Z. Jiang, “Link selection schemes for
Note that L [FGSC (x)] = ΦGSC (s) /s [20], we derive selection relaying systems with transmit beamforming: New and efficient
Φ proposals from a distributed concept,” IEEE Trans. Veh. Technol.,
, NR , LR , mg1 , aΦ Φ



L FGSC x =K SK k , bk vol. 61, no. 2, pp. 533–552, Feb. 2012.
mg1 NR mg1
r−1 [11] F. Rashid-Farrokhi, L. Tassiulas, and K. J. R. Liu, “Joint optimal power
1 X Ωg1
control and beamforming in wireless networks using antenna arrays,”
− mg1
r . (C.3) IEEE Trans. Commun., vol. 46, no. 10, pp. 1313–1324, Oct. 1998.
s r=1
s+ Ω [12] R. Zhang and Y.-C. Liang, “Exploiting multi-antennas for opportunistic
g1
spectrum sharing in cognitive radio networks,” IEEE J. Sel. Topics
Taking the inverse Laplace transform of (C.3), we obtain Signal. Process., 2008.
[13] H. Ding, J. Ge, D. B. da Costa, and T. Tsiftsis, “A novel distributed
Φ
, NR , LR , mg1 , aΦ Φ



FGSC x =K SK k , bk
antenna selection scheme for fixed-gain amplify-and-forward relaying
systems,” IEEE Trans. Veh. Technol., vol. 61, no. 6, pp. 2836–2842, Jul.
mg1 NR mg1 r−1


X Ωg1
m
− g1 x

2012.
1−
xr−1 e Ωg1 . (C.4) [14] V. Blagojevic and P. Ivanis, “Ergodic capacity for TAS/MRC spectrum
r=1
r−1 ! sharing cognitive radio,” IEEE Commun. Lett., vol. 16, no. 3, pp. 321–
323, Mar. 2012.
Again, employing the Taylor series expansion truncated to the [15] P. L. Yeoh, M. Elkashlan, T. Q. Duong, N. Yang, and D. B. da Costa,
k “Transmit antenna selection for interference management in cognitive
xj


kth order given by ex = k
P
j! + o x in (C.4), (C.4) can relay networks,” IEEE Trans. Veh. Technol., 2014.
j=0 [16] Y. Deng, M. Elkashlan, P. Yeoh, N. Yang, and R. Mallik, “Cognitive
be rewritten as MIMO relay networks with generalized selection combining,” IEEE
Trans. Wireless Commun., vol. PP, no. 99, pp. 1–1, May. 2014.
mg1 x
mg1 NR Φ [17] K. Letaief and W. Zhang, “Cooperative communications for cognitive
, NR , LR , mg1 , aΦ Φ



FGSC x = K SK k , bk .
Ωg1 radio networks,” Proc. IEEE, vol. 97, no. 5, pp. 878–893, May. 2009.
(C.5) [18] N. B. Mehta, S. Kashyap, and A. F. Molisch, “Antenna selection in LTE:
From motivation to specification,” IEEE Commun. Mag., vol. 50, no. 10,
pp. 144–150, Oct. 2012.
Based on

(C.5), the asymptotic expression for the CDF of [19] A. Annamalai and C. Tellambura, “A new approach to performance eval-

g1i∗ θ ∗
2 is given by


i uation of generalized selection diversity receivers in wireless channels,”

in Proc. IEEE Veh. Technol. Conf. (VTC), Fall 2001, pp. 2309–2313.
F


2 x = [20] X. Cai and G. B. Giannakis, “Performance analysis of combined
g1i∗ θi∗
transmit selection diversity and receive generalized selection combining
h m x
mg1 NR
g1
iNS in Rayleigh fading channels,” IEEE Trans. Wireless Commun., vol. 3,
Φ
K SK , NR , LR , mg1 , aΦ Φ
k , bk . (C.6) no. 6, pp. 1980–1983, Nov. 2004.
Ωg1 [21] M.-S. Alouini and M. K. Simon, “An MGF-based performance analysis
of generalized selection combining over rayleigh fading channels,” IEEE
By substituting (C.6) into (B.1), the first non-zero order Trans. Commun., vol. 48, no. 3, pp. 401–415, Mar. 2000.
expansion of the CDF of γ1 is attained and yields the asymp- [22] I. Ahmed, A. Nasri, R. Schober, and R. K. Mallik, “Asymptotic
totic outage probability of cognitive relay network with the performance of generalized selection combining in generic noise and
fading,” IEEE Trans. Commun., vol. 60, no. 4, pp. 916–922, Apr. 2012.
proportional interference power constraint as (14). [23] A. Lodhi, F. Said, M. Dohler, and A. H. Aghvami, “Closed-form symbol
error probabilities of STBC and CDD MC-CDMA with frequency-
R EFERENCES correlated subcarriers over Nakagami-m fading channels,” IEEE Trans.
Veh. Technol., vol. 57, no. 2, pp. 962–973, Mar. 2008.
[1] P. Kolodzy and I. Avoidance, “Spectrum policy task force,” Federal [24] Y. Ma and S. Pasupathy, “Efficient performance evaluation for general-
Commun. Comm., Washington, DC, Rep. ET Docket, no. 02-135, 2002. ized selection combining on generalized fading channels,” IEEE Trans.
[2] A. Goldsmith, S. A. Jafar, I. Maric, and S. Srinivasa, “Breaking spectrum Wireless Commun., vol. 3, no. 1, pp. 29–34, Jan. 2004.
gridlock with cognitive radios: An information theoretic perspective,” [25] J. Lee, H. Wang, J. G. Andrews, and D. Hong, “Outage probability
Proc. IEEE, vol. 97, no. 5, pp. 894–914, May 2009. of cognitive relay networks with interference constraints,” IEEE Trans.
[3] H. A. Suraweera, P. J. Smith, and M. Shafi, “Capacity limits and perfor- Wireless Commun., vol. 10, no. 2, pp. 390–395, Feb. 2011.
mance analysis of cognitive radio with imperfect channel knowledge,” [26] H. Ding, J. Ge, D. B. da Costa, and Z. Jiang, “Asymptotic analysis of
IEEE Trans. Veh. Technol., vol. 59, no. 4, pp. 1811–1822, May 2010. cooperative diversity systems with relay selection in a spectrum-sharing
[4] K. J. Kim, T. Q. Duong, and X.-N. Tran, “Performance analysis of scenario,” IEEE Trans. Veh. Technol., vol. 60, no. 2, pp. 457–472, Feb.
cognitive spectrum-sharing single-carrier systems with relay selection,” 2011.
IEEE Trans. Signal Process., vol. 60, no. 12, pp. 6435–6449, Dec. 2012. [27] I. Krikidis, H. Suraweera, P. Smith, and C. Yuen, “Full-duplex relay
[5] T. Xu, J. Ge, and H. Ding, “An efficient distributed link selection scheme selection for amplify-and-forward cooperative networks,” IEEE Trans.
for AF-based cognitive selection relaying networks,” IEEE Commun. Wireless Commun., vol. 11, no. 12, pp. 4381–4393, Dec. 2012.
Lett., vol. 18, no. 2, pp. 253–256, Feb. 2014. [28] A. Ghasemi and E. S. Sousa, “Fundamental limits of spectrum-sharing
[6] C. Zhong, T. Ratnarajah, and K.-K. Wong, “Outage analysis of decode- in fading environments,” IEEE Trans. Wireless Commun., vol. 6, no. 2,
and-forward cognitive dual-hop systems with the interference constraint pp. 649–658, Feb. 2007.
in Nakagami-m fading channels,” IEEE Trans. Veh. Technol., vol. 60, [29] J. M. Peha, “Approaches to spectrum sharing,” IEEE Commun. Mag.,
no. 6, pp. 2875–2879, Jul. 2011. vol. 43, no. 2, pp. 10–12, Feb. 2005.
[7] T. Q. Duong, P. L. Yeoh, V. N. Q. Bao, M. Elkashlan, and N. Yang, [30] K. J. R. Liu, A. K. Sadek, and W. Su, et al., Cooperative Communica-
“Cognitive relay networks with multiple primary transceivers under tions and Networking. Cambridge University Press, 2009.
spectrum-sharing,” IEEE Signal Process. Lett., vol. 19, no. 11, pp. 741– [31] T. C. Clancy, “Formalizing the interference temperature model,” Wireless
744, Nov. 2012. Commun. Mobile Comput., vol. 7, no. 9, pp. 1077–1086, May 2007.
[32] M. K. Simon and M.-S. Alouini, Digital communication over fading
channels. New York: Wiley, 2005.
[33] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series and
Products, 7th ed. San Diego, C.A.: Academic Press, 2007.
[34] H. Shin and J. H. Lee, “Capacity of multiple-antenna fading channels:
spatial fading correlation, double scattering, and keyhole,” IEEE Trans.
Inf. Theory, vol. 49, no. 10, pp. 2636–2647, Oct. 2003.
[35] S. S. Ikki and S. Aissa, “Multihop wireless relaying systems in the
presence of cochannel interferences: Performance analysis and design
optimization,” IEEE Trans. Veh. Technol., vol. 61, no. 2, pp. 566–573,
Feb. 2012.
[36] V. Gopal, M. Matthaiou, and C. Zhong, “Performance analysis of dis-
tributed MIMO systems in Rayleigh/Inverse-Gaussian fading channels,”
in Proc. IEEE Global Commun. Conf. (GLOBECOM), Dec. 2012, pp.
2468–2474.
[37] M. Matthaiou, N. D. Chatzidiamantis, G. K. Karagiannidis, and J. A.
Nossek, “On the capacity of generalized-K fading MIMO channels,”
IEEE Trans. Signal Process., vol. 58, no. 11, pp. 5939–5944, Nov. 2010.
[38] S. Jin, M. R. McKay, C. Zhong, and K.-K. Wong, “Ergodic capacity
analysis of amplify-and-forward MIMO dual-hop systems,” IEEE Trans.
Inf. Theory, vol. 56, no. 5, pp. 2204–2224, May 2010.
[39] E. Bjornson, P. Zetterberg, M. Bengtsson, and B. Ottersten, “Capacity
limits and multiplexing gains of MIMO channels with transceiver
impairments,” IEEE Commun. Lett., vol. 17, no. 1, pp. 91–94, Jan. 2013.
[40] D. B. da Costa and S. Aissa, “Cooperative dual-hop relaying systems
with beamforming over Nakagami-m fading channels,” IEEE Trans.
Wireless Commun., vol. 8, no. 8, pp. 3950–3954, Aug. 2009.
[41] R. L. Graham, D. E. Knuth, and O. Patashnik, Concrete Mathematics.
New York: Addison-Wesley, 1989.
Yansha Deng (S’13) is currently working toward the
M.S. degree at Central South University, Changsha,
China. She is also currently working toward the
Ph.D. degree in electronic engineering at Queen
Mary University of London, London, U.K.
Her research interests include multiple-antenna
systems, cognitive radio, cooperative networks,
molecular communication, and physical layer secu-
rity.
Lifeng Wang (S’12) is working towards his Ph.D.
degree in Electronic Engineering at Queen Mary
University of London. Before that, he received the
M.S. degree in Electronic Engineering from the
University of Electronic Science and Technology of
China, in 2012.
His research interests include physical layer secu-
rity, massive MIMO, millimeter-wave communica-
tions and 5G HetNets.
Maged Elkashlan (M’06) received the Ph.D. de-
gree in Electrical Engineering from the University
of British Columbia, Canada, 2006. From 2006 to
2007, he was with the Laboratory for Advanced
Networking at University of British Columbia. From
2007 to 2011, he was with the Wireless and Net-
working Technologies Laboratory at Commonwealth
Scientific and Industrial Research Organization (C-
SIRO), Australia. During this time, he held an
adjunct appointment at University of Technology
Sydney, Australia. In 2011, he joined the School
of Electronic Engineering and Computer Science at Queen Mary University
of London, UK, as an Assistant Professor. He also holds visiting faculty
appointments at the University of New South Wales, Australia, and Beijing
University of Posts and Telecommunications, China. His research interests fall
into the broad areas of communication theory, wireless communications, and
statistical signal processing for distributed data processing, millimeter wave
communications, cognitive radio, and wireless security.
Kyeong Jin Kim (SM’11) received the M.S. degree
from the Korea Advanced Institute of Science and
Technology in 1991, and the M.S. and Ph.D. degrees
in electrical and computer engineering from the Uni-
versity of California at Santa Barbara, Santa Barbara,
CA, USA, in 2000. From 1991 to 1995, he was a
Research Engineer with the Video Research Center,
Daewoo Electronics, Ltd., Seoul, Korea. In 1997,
he joined the Data Transmission and Networking
Laboratory at the University of California at Santa
Barbara. After receiving his degrees, he joined the
Nokia Research Center and Nokia Inc., Dallas, TX, USA, as a Senior Research
Engineer, where he was an L1 Specialist from 2005 to 2009. From 2010
to 2011, he was an Invited Professor with Inha University, Incheon, Korea.
Since 2012, he has been a Senior Principal Research Staff with Mitsubishi
Electric Research Laboratories, Cambridge, MA, USA. His research has
been focused on the transceiver design, resource management, scheduling
in the cooperative wireless communications systems, cooperative spectrum
sharing systems, device-to-device communications, secrecy systems, and GPS
systems.
Trung Q. Duong (S’05, M’12, SM’13) received his
Ph.D. degree in Telecommunications Systems from
Blekinge Institute of Technology (BTH), Sweden
in 2012, and then continued working at BTH as a
project manager. Since 2013, he has joined Queen’s
University Belfast, UK as a Lecturer (Assistant
Professor). He held a visiting position at Polytechnic
Institute of New York University and Singapore
University of Technology and Design in 2009 and
2011, respectively. His current research interests
include cooperative communications, cognitive radio
networks, physical layer security, massive MIMO, cross-layer design, mm-
waves communications, and localization for radios and networks.