You are on page 1of 4

IEEE COMMUNICATIONS LETTERS, VOL. 19, NO.

3, MARCH 2015

491

Cooperative AF Relaying With Beamforming and Limited


Feedback in Cognitive Radio Networks
Ali Afana, Telex M. N. Ngatched, and Octavia A. Dobre

AbstractThis letter studies cooperative beamforming for dualhop amplify and forward cognitive relay networks comprising a
pair of secondary source-destination and multiple antenna relay
in the presence of a pair of primary transmitter-receiver. The aim
of this work is to improve the secondary system performance with
limited feedback from the primary receiver while using a mean
value-based power allocation strategy. We derive closed-form expressions for the outage probability and average error probability
over Rayleigh fading channels. Numerical and simulation results
demonstrate the impact of the limited feedback on the secondary
system performance and show that an acceptable performance is
achieved when compared with the case of perfect channel state
information.
Index TermsBeamforming, cognitive radio, cooperative relaying, limited feedback.

I. I NTRODUCTION

PECTRUM sharing in cognitive radio (CR) relay networks has been proposed as an emerging technology to
improve the spectrum efficiency and enhance the reliability of
the secondary users (SUs) while adhering to the interference
limitations to primary users (PUs) [1]. In addition, employing
beamforming in cooperative CR relay systems has been proven
to be effective in achieving reliable transmission for the secondary systems and guaranteeing quality of service (QoS) for
the primary systems [2]. In the literature, most of the beamforming tools assumed that the SUs have prior instantaneous
channel state information (CSI) of the interference channels to
the PUs, which is impractical due to the heavy overhead.
Recently, few works considered the limited feedback in CR
systems [3]. In [4], cognitive beamforming with finite-rate feedback was considered in a multi-antenna downlink CR system,
where the outage probability of the SU link was analyzed. In
[5], the same problem was addressed using a different approach,
whereby the interference to the primary receiver (PU-RX)
was analyzed considering quantization errors due to limited
feedback.
In this paper, we consider a dual-hop amplify-and-forward
(AF) underlay CR relay system with limited feedback from a
PU-RX. In particular, we address the availability of partial CSI
via a finite rate feedback from the PU-RX to the secondary
relay, where the PU-RX quantizes its vector channel direction

Manuscript received November 27, 2014; accepted January 13, 2015. Date
of publication January 19, 2015; date of current version March 6, 2015.
This work was supported by the Research and Development Corporation of
Newfoundland and Labrador (RDC) and Natural Sciences and Engineering
Research Council of Canada (NSERC). The associate editor coordinating the
review of this paper and approving it for publication was C. Zhong.
The authors are with the Faculty of Engineering and Applied Science,
Memorial University, St. Johns, NL, Canada (e-mail: askafana@mun.ca;
odobre@mun.ca; tngatched@grenfell.mun.ca).
Digital Object Identifier 10.1109/LCOMM.2015.2393865

Fig. 1. Spectrum-sharing system with dual-hop relaying.

information (CDI) to one of 2B codebooks, for some positive


integer B, and conveys back the corresponding index. The
secondary relay then uses this quantized CDI to apply the zero
forcing beamforming (ZFB). Due to the quantization error of
the CDI, the relay cannot transmit with zero interference and its
transmit power should be adjusted to meet the PU interference
constraint. Meanwhile, the secondary source adjusts its transmit
power using the mean value (MV)-based power allocation
strategy [6], where only the MV of the interference channel gain
is used instead of the perfect interference CSI.
In this regard, we derive an upper bound for the secondary
relay transmit power that meets the PU interference constraint.
We also derive the statistics of the end-to-end (E2E) signal to
noise ratio (SNR) when using the MV-power allocation and
employing the ZFB with limited feedback. To investigate the
effect of the quantized CDI and MV-power allocation on the
secondary system performance, we derive closed-form expressions for its outage probability and average error probability.
The results show that the performance of the proposed system
with limited feedback is acceptable when compared with the
perfect CSI scheme.
II. S YSTEM AND C HANNEL M ODELS
A. System Model
We consider a dual-hop relaying system that is composed of
a secondary source S , a secondary destination D and an AF
secondary relay R of N antennas, sharing the same spectrum
band with a primary system consisting of a pair of a primary
transmitter (PU-TX) and receiver (PU-RX) as shown in Fig. 1.
All nodes, except the relay, are equipped with single antenna. It
is assumed that there is a direct link between S and D . The relay
node operates in the half-duplex mode and uses a transmission
protocol of two time-slots, TSk , k = 1, 2.
In TS1 , based on the MV-power allocation method, which
returns the MV of the interference channel gain (instead of
instantaneous CSI) from S to the PU-RX, S adjusts its transmit power under a predefined threshold Q and broadcasts its

1089-7798 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.

492

IEEE COMMUNICATIONS LETTERS, VOL. 19, NO. 3, MARCH 2015

message to the relay. In TS2 , using only quantized CDI, ZFB is


applied to mitigate the interference from the N antenna relay
to the PU-RX so that the relay is always able to transmit with
limited interference to the PU-RX.
All channels are assumed to be independent Rayleigh flat
fading, such that the power gains |hs,d |2 , |hs,r |2 , |hr,d |2 , | fs,p |2 ,
and | fr,p |2 , are exponential distributed random variables with
parameters s,d , s,r , r,d , s,p , and r,p , respectively. Let the
ZFB vector wTzf = [w1 , w2 , . . . , wN ] be used to direct the signal
to D . Furthermore, let hTr,d = [hr1 ,d , . . . , hrN ,d ] be the channel
vector between R and D , and fr,p = [ fr1 ,p , . . . , frN ,p ] be the
channel vector between R and PU-RX. As mentioned above,
we consider the case when only quantized CDI, i.e., limited
CSI is available at R . The primary and secondary receivers
feedback the quantized CDI to R just before the beginning
of TS2 , and this is based on quantization codebooks. All the
codebooks are designed in advance and stored at the relay
and receivers. Let Fr,p = {fr,p1 , fr,p2 , . . . , fr,p2B } be a codebook
of size 2B at the primary receiver, which is a collection of
unit norm vectors. Then, the optimal quantization codeword
selection criterion can be expressed as [7]

2


i = arg max fH
(1)
r,p j fr,p  ,
1 j2B

where fr,p =

fr,p
||fr,p ||

is the channel direction vector. Specifically,


index i is conveyed by the PU-RX and fr,pi is recovered at
R and used in lieu of the instantaneous CSI of the interference channel. Similarly, let Hr,d = {h r,d1 , h r,d2 , . . . , h r,d2C } be
a quantization codebook of size 2C at the secondary receiver,
which is also a collection of unit norm vectors. The same
process of codeword selection, index conveyance and CDI
recovery is used, where h r,di is recovered at R as a replacement
of the instantaneous CSI of the relaying channels between the
secondary relays antennas and D .
B. Transmission Protocol
In TS1 , S transmits its signal to D and R . The received signal
at D and R is given respectively as

(2)
ysd = Ps hs,d xs + n1 ,
(3)
ysr = Ps hs,r xs + nr ,
where Ps is the transmit power of S , xs is the information symbol of S , and n1 , nr denote the additive white Gaussian noise
(AWGN) value and vector at D and relays antennas, respectively. We assume that the transmitted symbols are equiprobable
with unit energy.
As R adopts AF relaying, it needs to weight the signals in
TS1 and to forward them to D . Hence, the received signal at D
from the relaying link is given by1

zf)hs,r xs +Ar hH
zf)nr + nD , (4)
yD = Ps Ar hH
r,d Diag(w
r,d Diag(w
zf ) denotes a diagonal matrix whose diagonal
where Diag(w
zf and Ar is the
elements are the elements of the vector w
1 In the system model, we consider that the interference from the primary
transmitter on the secondary receiver is represented as AWGN, when the
primary transmitters signal is generated by random Gaussian codebooks [1]
[5].

normalization constant designed to ensure that the total transmit


power at the relay is constrained and it is given by [2]
  
 1
H
2
H
zf
,
(5)
Ar = Pr w
zf Ps hs,r hs,r + I w
where Pr is the maximum available power at R and 2 is
the variance of all noises. Then, by using maximum ratio
combining, the total received SNR at D is given as

2

2  H Diag(w

2
P
A
)h
h
s r
zf s,r 
r,d
Ps |hs,d |
e
eq =
+
.
(6)

2
2

 2
2

A2r hH

w

zf
r,d

C. MV-Power Allocation Method


In the underlay approach of this model, S can utilize the PUs
spectrum as long as the interference it generates to the PU-RX
remains below the interference threshold Q. In general, the PURX is assumed to know the interference channel gain (or its
estimate) [6]. Therefore, it can calculate the MV of this gain.
Then, the estimated arithmetic MV (a constant) is fed back to
S . Consequently, the adoption of the MV-power allocation can
considerably reduce the feedback burden [6] when compared to
the scheme which requires instantaneous CSI feedback on every
symbol unit or block of symbols. Hence, Ps is constrained as
Ps = min( E(| fQ |2 ) , Pmax ), where E() is the expectation value of
s,p
the coefficient and Pmax is the maximum available power at S .
III. S UB -O PTIMAL ZFB W EIGHTS D ESIGN
Our objective is to maximize the received SNR at D to enhance the performance of the secondary system while limiting
the interference to the PU-RX. In the case of partial CSI, the
ZFB weight vector in the presence of a PU-RX is proved to
be [2]
1



zf = T h r,d T h r,d 
,
(7)
w
where T = I fr,p (fr,p fr,p )1 fr,p is the projection idempotent
matrix with rank (N 1). According to the ZFB principles,
zf is chosen to lie in the orthogonal space of fr,p such that
w
zf | = 0 and |h H
zf | is maximized. By substituting (5) and
|fH
r,p w
r,d w
(7) into (6), and after simple manipulations, the equivalent SNR
at D can be written in the general form of eeq = sd + rd as

2
Ps
hs,r 2 Pr2 T h r,d 
Ps
e
2
2
.
(8)
eq = 2 |hsd | +


2
Ps
Pr  2

+1
2 hs,r  + 2 T hr,d

IV. P ERFORMANCE A NALYSIS


A. End-to-End SNR Statistics
Due to the quantization errors inherent to partial CSI, R
cannot transmit with zero interference. In this case, R should
reduce the transmit power to meet the interference constraint.
Since the instantaneous interference CDI cannot be exactly
estimated because of the partial CSI, we consider the average

AFANA et al.: COOPERATIVE AF RELAYING WITH BEAMFORMING AND LIMITED FEEDBACK IN CR NETWORKS

is the maximum average threshinterference constraint [5]. If Q


old, the interference constraint is given by

2

zf  = E Pr ||fr,p ||2 |fr,p w


zf |2 Q.
(9)
I = E Pr fH
r,p w
Using ZFB based on limited CDI from the PU, the relationship
between the residual I, Pr , and the feedback amount B is given
in the following Proposition.
Proposition 1: Given Pr and B, the residual I at the PU-RX
B
Pr N N1
2
.
after ZFB is tightly upper bounded by N1
Proof: Using the notion of random vector quantization
(RVQ) [7], the relationship between the true and quantized
channel direction vectors can be written as

fr,p = 1 efr,p + efo ,


(10)
where fr,p is the quantization codeword based on (1), e = 1
2
|fH
r,p fr,p | is the magnitude of the quantization error, and fo is
the unit norm vector isotropically distributed in the nullspace
of fr,p . Note that fo is independent of e. The interference term
is derived as follows
H

2 
2
H 2
fr,p w
zf  =  1 e fH

zf  = e fH
zf  , (11)
+
efo w
w
zf
r,p
ow
zf is in the null space
where (11) follows from the fact that w

of fr,p , i.e., |fH


|
=
0.
Due
to
the
independence
between the
w
zf
r,p
channel magnitude and direction, I reduces to

2



zf  .
(12)
I = E Pr ||fr,p ||2 E[e]E fH
ow
Hence, to obtain I, we need to compute the properties of the
random variables (RVs) in the three expectation terms. Since
||fr,p ||2 is chi-square distributed with 2N degrees of freedom,
we have E[||fr,p ||2 ] = N. For an arbitrary quantization code 2
word, fr,p , (1 |fH
r,p fr,p | ) is beta distributed as (N 1, 1)[7].
H
Since e = 1 |fr,p fr,p |2 is the minimum of 2B independent
(N 1, 1) RVs, its expectation is tightly upper bounded as
B
zf are i.i.d. isotropic vectors [4],
E[e] < 2 N1 . Since fo and w
zf |2 ]
[5], [7] in the (N 1) dimensional null space of fr,p , E[|fH
ow
1
is (1, N 2) distributed [7], with an expectation of N1
.
B

Pr N N1
As a result, we have I < N1
2
. According to the above
and considering maximum power
Proposition, assuming Q = Q
Pr ), where
constraint, Pr is upper bounded as Pr min(,
max
B
(N1)Q 2 N1 and Pr is the maximum available power at R .
=
max
N
Before analyzing the system performance, we need to know
the CDF and PDF of the SNR at D . We first start with the
SNR from the relaying link, which can be shown to be upper
bounded by

rd

zf |2
Ps ||hs,r ||2 Pr |hH
r,d w

zf |2 + 2
Ps ||hs,r ||2 + Pr |hH
r,d w
2
zf |2
Ps |hs,r || Pr ||hr,d ||2 |h H
r,d w

2
2 H |2 + 2
Ps ||hH
zf
s,r || + Pr ||hr,d || |h w
r,d

Ps |hs,r ||2 Pr ||hr,d ||2 (1, N 1)


=
Ps |hs,r ||2 + Pr ||hr,d ||2 (1, N 1) + 2
1 22N 2 22

=
,
1 22N + 2 22 + 1

22N

22

(13)

493


1 = min( 1/Qs,p , Pmax ) 2 , and 2= Pr2 . The right hand-side terms
of (13) and (14) are derived as analyzed in the above proof.
To this end, the received SNR eeq can be tightly upper
bounded by
up

eeq sd + min(1 , 2 ) = sd + rd .
  

(15)

eq

Proposition 2: The CDF and PDF of the equivalent SNR eq


at D are given respectively by

1  N
x
1 (N, 1 x)e D N
Feq (x) = N
(N,

x)
2
2
2 (N)
x

1 N1 1 k (k , 1 x) e

1 k=0
k!k2

2 k (k , 2 x)
, (16)
k!k2

and


( )
ex/D 1 N1 2 k
(k , 2 x)
feq (x) =

D
2 k=0 k!1


1
+ N
N (N, 2 x) ,
1 (N) 2
min(

(17)

,Pmax )

1/s,p
where 1 = 11 + 12 , 2 = 11 + 12 1D , D =
, k =
2
k + 1 and (., .) is the lower incomplete Gamma function.
up
Proof: Using order statistics, the CDF of rd can be shown
to be given by

Fup (x) = F1 (x) + F2 (x) F1 (x)F2 (x),

(18)

rd

( x )k

1
2 . By
where F1 (x) = 1 e 1 N1
k=0 k! and F2 (x) = 1 e
subsituiting the CDFs F1 (x) and F2 (x) into (18), the CDF of
up
rd is derived as

Fup (x) =
rd

N
1 N1
1 (N, 1 x)
+
(k , 1 x).
N
2 k=0
1 (N)

(19)
up

By differentiating (19) with respect to x, the PDF of rd is


derived as


x
xN1 +
1 x
x
2
1
2
up
f (x) = N e
+ e N,
.
(20)
rd
2
1
1
We note that the Laplace transform of the PDF of eq in (15)
can be presented as
 1
 



1
L feq (x) = L fup (x)
s+
.
(21)
D
rd
D
First, the CDF of eq can be obtained by using the
differential property of the Laplace transform, utilizing
[9, Eq. (1.1.1.14)] to calculate the inverse Laplace transform
and solving the resultant integral by using [8]. Second, for the
PDF of eq , we utilize the integral property of the Laplace
transform, [9, Eq. (1.1.1.13)] and solve the resultant integral
by using [8]. In the next section, we use Feq (x) and feq (x)
to derive closed-form expressions for the outage and error
probabilities performance.

(14)

and 2 = 2
are chi-squared
where 1 = 1
RVs with 2N and 2 degrees of freedom, respectively,

B. Outage Probability
An outage event occurs when the total received SNR falls below a certain threshold th and is expressed as Pout =Pr(eq <th ).

494

IEEE COMMUNICATIONS LETTERS, VOL. 19, NO. 3, MARCH 2015

Fig. 2. Outage probability vs. Q (dB) for different values of B and s,p .

Fig. 3.

By using (16), the outage probability of the secondary system


can be written as
Pout = Feq (th ).

(22)

C. Average Error Probability


This probability could be evaluated using the following
identity

a b ebx
Feq (x)dx,
Pe =
(23)
x
2 0
where the values of (a,b) depend on the modulation scheme and
Feq (x) is given in (16). Using [8, Eq. (6.455.2)], a closed-form
expression for the average error probability at D is obtained as

(N + 12 )
a b
Pe =
(N)
2 (N)N N(1 + b)N+ 21
1

(k + 12 )
1 N1
+
(k ) , (24)

1
2 k=0
k!k (1 + b)k + 2 k
1

1
where () =2 F1 (1, + 12 ; + 1; 1+b
) 2 F1 (1, + 12 ; + 1;
2
1 +b ) and 2 F1 (., .; .; .) is the Gauss hypergeometric function
defined in [8, Eq. 9.111].

V. N UMERICAL R ESULTS
In this section, we investigate the derived results for the
system performance through numerical examples and simular
= 10 dBm. Since
tions. We use th = 0 dB and Pmax = Pmax
we focus on the primary feedback, we set C to 25 bits for the
R -D channels. We consider binary phase shift keying (BPSK)
modulation scheme; however, the derived expressions are valid
for higher order modulation schemes.
Fig. 2 shows the outage performance versus Q for N =
4, for different values of B and s,p . As observed from the
figure, as the value of Q increases, the outage performance
improves substantially. Additionally, as B increases, the system performance improves. Furthermore, as s,p increases, the
outage performance improves. This is due to the fact that the
interference channel to the PU-RX becomes weaker, allowing
the SU to transmit more power.
Fig. 3 presents the error performance versus s,p for N = 3,
4 and different values of B. Similar conclusions to the outage

Average error probability vs. s,p for different values of B and N.

performance hold. Furthermore, as s,p increases, the error


performance saturates due to the sources transmit power
limitation. Additionally, as the number of antennas increases,
the performance improves due to the combination of diversity
and beamforming. We compare the results of N = 4 and
B = 15 for C = 25 and C , with the perfect CSI of [2].
The figure shows that there is a gap between the perfect and
limited feedback CSI due to the partial CSI availability and the
quantization residual error. However, this gap is acceptable,
given the practicality of the scheme. It is worth noting that
simulation results, for the equivalent SNR eq , confirm the
theoretical findings in both Figs. 2 and 3.
VI. C ONCLUSION
We investigated the impact of the limited quantized feedback
from the PU-RX to the secondary R on the secondary system
performance, while the MV-based power allocation method was
used at S . Using this feedback, we employed ZFB to limit the
interference inflicted on the PU-RX. The results showed that
the secondary system performance is acceptable using limited
feedback when compared with the perfect CSI case.
R EFERENCES
[1] A. Goldsmith, S. A. Jafar, I. Maric, and S. Srinivasa, Breaking spectrum gridlock with cognitive radios: An information theoretic perspective,
Proc. IEEE, vol. 97, no. 5, pp. 894914, May 2009.
[2] A. Afana, V. Asghari, A. Ghrayeb, and S. Affes, On the performance
of cooperative relaying spectrum-sharing systems with collaborative distributed beamforming, IEEE Trans. Commun., vol. 62, no. 3, pp. 857871,
Mar. 2014.
[3] Y. Huang et al., Cognitive MIMO relaying networks with primary
users interference and outdated channel state information, IEEE Trans.
Commun., vol. 62, no. 12, pp. 42414254, Dec. 2014.
[4] K. Huang and R. Zhang, Cooperative feedback for multiantenna cognitive
radio networks, IEEE Trans. Signal Process., vol. 59, no. 2, pp. 747758,
Feb. 2011.
[5] J. Noh and S. J. Oh, Beamforming in a multi-user cognitive radio system
with partial channel state information, IEEE Trans. Wireless Commun.,
vol. 12, no. 2, pp. 616625, Feb. 2013.
[6] K. Tourki, F. A. Khan, K. A. Qaraqe, Y. Hong-Chuan, and M.-S. Alouini,
Exact performance analysis of MIMO cognitive radio systems using
transmit antenna selection, IEEE J. Sel. Areas Commun., vol. 32, no. 3,
pp. 425438, Mar. 2014.
[7] N. Jindal, MIMO broadcast channels with finite-rate feedback, IEEE
Trans. Inf. Theory, vol. 52, no. 11, pp. 50455060, Nov. 2006.
[8] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Products,
7th ed. Amsterdam, The Netherlands: Elsevier, 2007.
[9] A. P. Prudnikov, Y. A. Brychkov, and O. I. Marichev, Integrals and Series.
New York, NY, USA: Gordon and Breach, 1992, vol. 3 and 5, Inverse
Laplace Transforms.

You might also like