1390 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 60, NO.

5, MAY 2012
Uncoordinated Beamforming for
Cognitive Networks
Simon Yiu, Chan-Byoung Chae, Senior Member, IEEE,
Kai Yang, Member, IEEE, and Doru Calin, Senior Member, IEEE
AbstractIn this paper, we propose jointly-optimized beam-
forming algorithms for cognitive networks to maximize the
achievable rates, where primary and cognitive users share the
same spectrum and are equipped with multiple antennas. We
consider the transmission of a single information stream in
both primary and secondary links. No coordination is required
between the primary and cognitive users and the interference
cancellation is done at the cognitive user. Specically, the beam-
forming vectors of the cognitive link are designed to maximize
the achievable rate under the condition that the interference both
at the primary and cognitive receivers is completely nullied.
Furthermore, it is proved that the achievable rate of a general
N
C
t
2 (N
C
t
transmit antennas at the cognitive transmitter and
2 receive antennas at the cognitive receiver) cognitive multiple-
input multiple-output (MIMO) link employing the optimal pro-
posed beamformers (which completely nullies the interference to
and from the primary link while maximizing its own achievable
rate) is the same as the rate of an interference-free (N
C
t
1) 1
multiple-input single-output (MISO) link employing an optimal
maximal ratio transmission beamforming vector. The sum rate
performance of the proposed algorithms is evaluated by Monte
Carlo simulations. The impact of the number of transmit and
receive antennas on the proposed algorithm is also discussed.
Index TermsBeamforming, MIMO, cognitive network, fading
channel, interference cancellation.
I. INTRODUCTION
T
HE opportunistic use of the wireless spectrum has been a
hot research topic in the wireless communications com-
munity in recent years due to the intense competition for the
use of spectrum at frequencies below 3 GHz [1]. In particular,
cognitive networks have received much attention [2]. Related
works on capacity region have been studied extensively, for
example, in [3][6]. A cognitive network consists of a number
of traditional wireless service subscribers and the so-called
cognitive users. The traditional wireless service subscribers
have the legacy priority access to the spectrum and are usually
called the primary users in the literature. On the other hand,
cognitive users, which are also known as the secondary users,
are allowed to access the spectrum only if communication
Paper approved by D. J. Love, the Editor for MIMO and Adaptive Tech-
niques of the IEEE Communications Society. Manuscript received January 3,
2011; revised August 19, 2011.
S. Yiu, K. Yang, and D. Calin are with Bell Labs, Alcatel-Lucent, Murray
Hill, NJ, 07974, USA (e-mail: {simon.yiu, kai.yang, doru.calin}@alcatel-
lucent.com).
C.-B. Chae is with the School of Integrated Technology, Yonsei University,
Korea (e-mail: cbchae@yonsei.ac.kr).
Parts of this paper were presented at the 6th International ICST Conference
on Cognitive Radio Oriented Wireless Networks and Communications, Osaka,
Japan.
Digital Object Identier 10.1109/TCOMM.2012.030712.110008
does not create signicant interference to the licensed primary
users. This can be achieved in several ways as discussed in
[7] and references therein. For example, the cognitive user
can transmit concurrently with the primary users under an
enforced spectral mask. Another strategy, commonly referred
to as spectrum sensing, is to have the cognitive users monitor
the spectrum and access it when an unused slot is detected.
Since the primary user network should not be required to
change their infrastructure and topology for spectrum sharing
with the secondary users, the secondary users should be
able to detect the presence of primary users in the spec-
trum independently and efciently. Spectrum sensing [8] is
the fundamental problem that many researchers attempt to
address in the literature [9][11]. The problem is essentially
a detection where the goal is to nd an optimal decision
threshold. The design of the threshold creates an interesting
trade-off between the probability of miss detection and the
probability of false alarm. A low probability of miss detection
can be achieved by using a low threshold at an expense of
high probability of false alarm. Likewise, one can design a
system with a low probability of false alarm by utilizing a high
threshold at an expense of high probability of miss detection.
A high probability of false alarm is not desirable because
the cognitive user remains silent even when the spectrum is
not being used by the primary user. On the other hand, the
cognitive user constantly creates interference to the primary
user in a system with a high probability of miss detection. In
this paper, we address the interference problem in cognitive
networks from another perspective. There is no spectrum
detection (sensing) involved and both primary and secondary
users can transmit simultaneously through the same spectrum
without interfering each others. In [12], the authors proposed
opportunistic spectrum sharing algorithms by exploiting multi-
antennas but they mostly focused on removing interference
only from a secondary transmitter to primary receivers. Linear
vector precoding for downlink cognitive systems is considered
in [13]. In particular, an optimal interference-free precod-
ing scheme was proposed which completely removes the
interference to the other system. However, multiple antennas
are considered only at the transmitter side and as a result,
beamforming vectors can be used only at the transmitters.
The authors in [14] also consider multiple antennas at the
transmitter side and show that in a device with K+N antennas
can completely nullify N 1 interferers while achieving
a diversity gain of K + 1. Consequently, a network with
N such devices is essentially interference-free and therefore
0090-6778/12$31.00 c 2012 IEEE
YIU et al.: UNCOORDINATED BEAMFORMING FOR COGNITIVE NETWORKS 1391
the capacity of the network can be increased. Unlike prior
works, in this paper, we propose uncoordinated beamforming
techniques in cognitive networks to completely remove the
cross interference between the primary and secondary users
using multiple antennas at both the transmitters and receivers.
The proposed algorithms also maximize the achievable rates of
both links through uncoordinated beamforming. Beamforming
is a well-studied subject in the signal processing literature. It
can be used for either directed transmission or reception of
energy in the presence of noise and interference [15]. In recent
years, beamforming also plays an important role in multiple
antenna system. For example, beamforming exploits channel
knowledge at the transmitter to maximize the signal-to-noise
ratio (SNR) at the receiver by transmitting in the direction of
the eigenvector corresponding to the largest eigenvalue of the
channel [16]. This beamforming concept has been extensively
studied for the MIMO broadcast channel [17] and multi-cell
environments [18].
In this paper, we consider a cognitive network that consists
of a single primary and secondary user. Each user consists
of a transmitter and a receiver. Both transmitters and re-
ceivers are equipped with multiple antennas and beamforming
transmit/receive vectors. In particular, we claim the following
contributions:
We propose three designs for the beamforming vectors.
The beamforming vectors are designed such that the
interference caused by the cognitive transmitter to the
primary receiver and the interference caused by the
primary transmitter to the cognitive receiver is completely
nullied while maximizing the rate of both the primary
and secondary links. Simulation results show the effec-
tiveness of the proposed design.
By using multiple antennas at the secondary user, the
proposed designs do not require knowledge of the cogni-
tive communication link at the primary user. In fact, the
secondary user is invisible to the primary user and the
performance of the primary link is not affected by the
secondary user at all. This is in contrary to prior work
in the literature which requires coordination between the
primary and secondary users [19][22].
We prove that the maximum achievable rate of an N
C
t
2
(N
C
t
denotes the number of transmit antennas at the
cognitive transmitter) cognitive multiple-input multiple-
output (MIMO) system under a zero interference con-
dition is the same as the maximum achievable rate of
an interference-free (N
C
t
1) 1 multiple-input single-
output (MISO) system which employs an optimal trans-
mit beamforming vector.
Organization: This paper is organized as follows. In Section
II, we introduce the system model of the proposed trans-
mission scheme and formulate the beamforming optimization
problem. In Section III, we propose three beamforming solu-
tions which cancel the interference both at the primary and
secondary receiver while maximizing the spectral efciency
of both the primary and secondary links. Finally, we present
some simulation and numerical results in Section IV, and draw
some conclusions in Section V.
Notation: In this paper, bold upper case and lower case
.
.
.
u
f
v
t
W
H
G
D
.
.
.
.
.
.
.
.
.
Fig. 1. The network model under consideration. The primary user is equipped
with N
P
t
and N
P
r
antennas at the transmitter and receiver, respectively. The
cognitive user is equipped with N
C
t
and N
C
r
antennas at the transmitter and
receiver, respectively.
letters denote channel matrices
1
and beamforming vectors,
respectively. []

, E{}, j

1, C, Null(),
max
(), and
1
N
denote Hermitian transposition, statistical expectation, the
imaginary unit, the set of complex number, the null space of a
vector, the largest eigenvalue of a square matrix, and an N1
vector with all ones, respectively.
II. NETWORK AND CHANNEL MODELS
Consider a cognitive network with a single primary user
and a single cognitive (secondary) user as depicted in Fig. 1.
Each user consists of a transmitter and a receiver. The primary
transmitter and receiver are equipped with N
P
t
and N
P
r
antennas, respectively. Likewise, the secondary transmitter and
receiver are equipped with N
C
t
and N
C
r
antennas, respectively.
Unless stated otherwise, N
C
t
2, and N
C
r
2 is assumed.
Furthermore, we assume that the antennas are uncorrelated and
the channel is frequency non-selective which can be easily
achieved by using multiple-input multiple-output orthogonal
frequency division multiplexing (MIMO-OFDM) [23]. Note
that, however, our solution is not directly related to the channel
model. Once channel information is known, the cognitive
transmitter and receiver can compute the transmit/receive
beamforming vectors using the proposed algorithms. The
MIMO channel between the primary transmitter and receiver
is denoted by W whereas the one between the secondary
transmitter and receiver is denoted by H. The interference
channel from the primary transmitter to the secondary receiver
is denoted by D and the interference channel from the
secondary transmitter to the primary receiver is denoted by
G. We model the individual channel elements in W, H, D,
and G, as independent and identically distributed (i.i.d.) zero-
mean complex Gaussian random variables with unit variance
(Rayleigh fading). The primary transmitter employs a beam-
forming vector u for the transmission of its data symbol x
P
.
Note that we consider the transmission of a single stream
of information in the primary link. Of course this is not
optimal from the information theory perspective. However,
we note that this assumption is not restrictive. The results
presented in this paper is also valid for the case of spatial
1
We still use bold upper case letters for vector channels.
1392 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 60, NO. 5, MAY 2012
{v
opt
, f
opt
, t
opt
, u
opt
} = argmax
v,f,t,u
{log
2
(1 + SINR
P
) + log
2
(1 + SINR
C
)}
subject to
_
v

Gf = 0 and t

Du = 0
u

u = f

f = v

v = t

t = 1
. (1)
{v
opt
, f
opt
, t
opt
, u
opt
} = argmax
v,f,t,u
{log
2
(1 + SINR
P
) + log
2
(1 + SINR
C
)}
subject to
_
f Null(v

G) and t Null(Du)
u

u = f

f = v

v = t

t = 1
. (2)
multiplexing.
2
We consider single stream of information only
for ease of conveying the main idea. At the cognitive link,
the transmitter employs a beamforming vector f for the
transmission of its data symbol x
C
. x
P
and x
C
are assumed
to be complex zero-mean unit variance random variables.
Furthermore, let v and t be the receive combining vector for
the primary and secondary receiver, respectively. Finally, we
impose a unit energy constraint on all beamforming vectors,
i.e., u

u = f

f = v

v = t

t = 1.
Let P
P
and P
C
be the transmit power at the primary
and secondary transmitter, respectively, the received signal
at the primary receiver and the secondary receiver are given
respectively by
r
P
=
_
P
P
v

Wux
P
+
_
P
P
v

Gfx
C
+v

n
P
(3)
and
r
C
=
_
P
C
t

Hfx
C
+
_
P
C
t

Dux
P
+t

n
C
. (4)
The elements in the noise vectors n
P
and n
C
are modeled
as i.i.d. zero-mean complex Gaussian random variables with
variance
2
P
and
2
C
, respectively. The resulting signal-to-
interference-plus-noise ratio (SINR) of the primary and cog-
nitive links are given by
SINR
P
=
P
P
v

uu

Wv
P
P
v

Gff

v +v

v
2
P
(5)
and
SINR
C
=
P
C
t

ff

Ht
P
C
t

Duu

t +t

t
2
C
, (6)
respectively. It is obvious from (5) and (6) that in order to
achieve zero interference, the beamforming vectors v, f, t,
and u have to be designed such that v

Gf = 0 and t

Du =
0. In addition to guarantee zero interference, our goal is also
to maximize the sum rate. For a single stream transmission,
the sum rate is given by [24]
R
S
= log
2
(1 + SINR
P
) + log
2
(1 + SINR
C
) . (7)
Therefore, the design problem can be mathematically formu-
lated as (1) given at the top of this page.
In the next section, we present three solutions to the
above optimization problem assuming that the primary user
has completely no knowledge of the secondary user while
achieving zero interference at both receivers.
2
In this case, the beamforming vectors become precoding matrices.
III. BEAMFORMING VECTOR DESIGNS
Considering the rst constraint of (1), it is obvious that
zero interference can be achieved by appropriately designing
v or f and t or u. We assume that the cognitive user is
transparent to the primary user since the performance of the
primary user should not be affected by the secondary link. To
achieve zero interference caused to the primary receiver, the
secondary transmitter can beamform in the null space of v

G.
Likewise, at the cognitive receiver the receive beamforming
vector t can be designed such that it is in the null space of
Du in order to avoid the interference caused by the primary
transmitter. Note that v

G is an 1 N
C
t
vector and the
dimension of its null space is N
C
t
1. Similarly, the dimension
of Du is N
C
r
1 and the dimension of its null space is
N
C
r
1. The optimization problem in (1) now becomes (2)
given at the top of this page. Having achieved zero interference
both at the primary and secondary receivers, the remaining
question is how to maximize the sum rate in (7). The rst
constraint in (2) obviously suggests that f depends on v and
t depends on u. The sum rate optimal solution requires nding
v and u such that R
S
is maximized and the optimal solution
requires the knowledge of W, H, D, and G, i.e., the global
channel state information (CSI). This is not reasonable from
the primary users point of view because the primary user
should not be required to know the existence of the secondary
user. Therefore, it is reasonable for the primary user to simply
optimize v and u to maximize its own rate assuming no
interference from the secondary transmitter. After obtaining
v and u, the secondary user can choose f and t (which are
functions of v and u, respectively, cf. the rst constraint of
(2)) to maximize its own rate.
The rate of the primary user can be maximized by appro-
priately designing v and u. Since no interference is created
at the primary user and the only constraint for the beam-
forming vectors v and u is the energy constraint, standard
approaches in existing literature can be used to design v
and u to maximize the rate of the N
P
t
N
P
r
interference-
free MIMO link [25], [26]. Since we restrict ourselves to the
transmission of a single stream of information, the spectral
efciency can be maximized by maximizing the SINR due
to the monotonic property of the logarithm function. It is
well known that the SINR maximizing receive beamformer
for a point-to-point link is the maximal ratio combining
beamformer. In this case, the receive beamforming vector is
YIU et al.: UNCOORDINATED BEAMFORMING FOR COGNITIVE NETWORKS 1393
{f
opt
, t
opt
} = argmax
f, t
_
P
C
t

Hff

t
t

t
2
C
_
(8)
subject to
_
f Null(v

opt
G) and t Null(Du
opt
)
f

f = t

t = 1
. (9)
simply v
opt
= Wu/

Wu. With this design and the

zero interference condition, (3) becomes
r
P
=

P
P
u

Wu

Wu
x
P
+
u

Wu
n
P
(10)
and the corresponding instantaneous SINR is given by
SINR
P
=
P
P
u

Wu

2
P
. (11)
Obviously, the spectral efciency of the primary link log
2
(1+
SINR
P
) can be maximized by beamforming in the direction
of the eigenvector corresponding to the largest eigenvalue of
W

W. We denote the optimal transmit beamforming vector

as u
opt
.
Using again the monotonic property of the logarithm func-
tion, the spectral efciency of the cognitive link can be
maximized by maximizing SINR
C
. To maximize the SINR of
the cognitive communication link, the design of f and t is not
as exible as the one for v and u. This is because the feasible
value of f and t is now constrained by the zero interference
requirement. Specically, the optimal beamformers can be
obtained by solving the optimization problem (8) and (9)
shown at the top of this page.
We present three proposals for the design of f and t in the
next three subsections.
A. Method 1: Discrete Search
Let F and T be the set of basis vectors which spans the
null space of v

opt
G and Du
opt
, respectively. Note that the
cardinality of F and T are N
C
t
1 and N
C
r
1, respectively.
The instantaneous SINR of the cognitive link given by
SINR
C
=
P
C
t

Hff

t
t

t
2
C
(12)
can be maximized by performing an exhaustive search in F
and T , i.e.,
{f
discrete
, t
discrete
} = argmax
fF, tT
_
P
C
t

Hff

t
t

t
2
C
_
.
(13)
Note that for N
C
t
= N
C
r
= 2, there is only one vector in the
set F and T . In general, (N
C
t
1)(N
C
r
1) computations are
required to obtain the best beamformers f
discrete
and t
discrete
.
Although zero interference can always be guaranteed at both
receivers by selecting the beamformer pairs f, t as in (13),
the obtained solution is not optimal in the sense of maximum
sum rate because the search in (13) is not carried out over the
entire null space of v

opt
G and Du
opt
.
B. Method 2: Gradient Algorithm
Since any vector in the null space of v

opt
G and Du
opt
satises the zero interference condition, there could be po-
tentially other vectors in those spaces which yield a higher
SINR
C
than f
discrete
and t
discrete
. Suppose the columns of

G and

D contain the basis vectors of the null space of v

opt
G
and Du
opt
, respectively. The optimal beamformers are in the
form of
f =

Ga

a
and t =

Db

b
, (14)
where a C
(N
C
t
1)1
and b C
(N
C
r
1)1
. The constrained
optimization problem in (8) can now be formulated as an
unconstrained one whose goal is to nd a C
(N
C
t
1)1
and b C
(N
C
r
1)1
such that the objective function in (8)
is maximized, i.e.,
{a
opt
, b
opt
} = argmax
a, b
_
P
C
b

Gaa

Db
b

ba

a
2
C
_
.
(15)
Unfortunately, there is no closed-form solution to (15) and
we have to resort to numerical methods to solve the problem.
In particular, a simple gradient (steepest accent) algorithm
can be used to obtain a
opt
and b
opt
and we denote the
resulting solutions as a
grad
and b
grad
. Suppose f(a[i], b[i])
is the objection function in (15), the gradient algorithm is
given by
_
a[i + 1]
b[i + 1]
_
=
_
a[i]
b[i]
_
+
_
f(a[i], b[i])/a[i]

f(a[i], b[i])/b[i]

_
, (16)
where i is the discrete iteration index and is the adaptation
step size. Furthermore, the two gradients in (16) are given
respectively by (17) and (18) shown at the top of next page in
which K is an irrelevant constant. The time index i is dropped
in the two gradients for ease of presentation. In the next
section, some guidelines in choosing the adaptation constant
and the initial values a[1] and b[1] are provided.
C. Method 3: Optimal Solution for N
C
r
= 2
As mentioned in the last section, there is no closed-form
solution to (15). However, if we x the number of receive
antennas of the cognitive user to two, the joint optimization in
(15) becomes a single (vector) variable optimization problem
and a closed-form solution is feasible. We rst rewrite (8) as
{f
opt
, t
opt
} = argmax
f, t
_
P
C
f

tt

Hf
t

t
2
C
_
, (19)
assuming the same constraints as in (9). Suppose N
C
r
= 2,
the null space of Du
opt
is one dimensional. Assume that
the null space of Du
opt
is spanned by t
0
and therefore, the
1394 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 60, NO. 5, MAY 2012
f(a[i], b[i])
a[i]

= K[((b

ba

a))(

Dbb

Ga) (a

Dbb

Ga)(b

ba)] (17)
f(a[i], b[i])
b[i]

= K[(b

ba

a)(

Gaa

Db) (b

Gaa

Db)(a

ab)], (18)
receive beamforming vector at the cognitive receiver is given
by t
opt
= t
0
. Recall that the optimal beamformers are in the
form of (14) and let

h =

G

t
0
, (20)
the optimization problem in (19) becomes
{a
opt
} = argmax
a
_
P
C
a

h

a
a

a
2
C
_
. (21)
The argument in (21) (which is essentially the SINR of the
cognitive user) is known as the generalized Rayleigh quotient
and by invoking the Rayleighs principle [27], it can be
bounded by
P
C

min
(

2
C
SINR
C
=
P
C
a

h

a
a

a
2
C

P
C

max
(

2
C
.
(22)
Therefore, SINR
C
can be maximized by choosing a
opt
to be
the eigenvector corresponding to the maximum eigenvalues
of

h

. It is interesting to note that although there is no

constraint on a C
(N
C
t
1)1
, the optimal solution a
opt
is
always a unit vector and consequently, we can obtain the op-
timal transmit beamforming vector directly by f
opt
=

Ga
opt
without the normalization in (14).
D. Discussion
In this subsection, we show that the rate of an N
C
t
2
MIMO link employing f
opt
and t
opt
obtained by Method 3
(which completely nullies the interference to and from the
primary link while maximizing its own rate) is the same as the
rate of an interference-free (N
C
t
1) 1 MISO system which
employs an optimal maximal ratio transmission beamforming
vector given by a
opt
. First, consider

h in (20). Since t
0
and the
columns in

G are all unit vectors,

h is an (N
C
t
1)1 vector
whose elements are complex Gaussian random variables with
mean zero and unit variance. Second, recall that the argument
in (21) with a
opt
represents the optimal maximum SINR
value of the N
C
t
2 cognitive link. Obviously, this is also
the optimal maximum SINR of a (N
C
t
1) 1 interference-
free communications link where a
opt
is used as the optimal
beamformer. Since the SINR of both links are the same, the
rate of both links are also the same by the Shannon formula.
It is also interesting to note that the optimal beamforming
vectors f
opt
and a
opt
of both systems are related by (14).
The above observation is intuitively appealing because
for the N
C
t
2 link, one antenna both at the transmitter
and receiver are used to avoid the interference to and from
the primary link. Therefore, there are only N
C
t
1 and 1
antennas remained at the transmitter and receiver, respectively,
to maximize its own SINR via beamforming.
For N
C
r
> 2, a closed-form solution to f
opt
and t
opt
requires solving a joint optimization problem and is not
feasible. However, we argue that for N
C
r
> 2, beamforming
should not be used anyway at high SINR because in the high
SINR regime, it is more attractive to use the multiple antennas
for spatial multiplexing purpose for rate maximization [28].
Nevertheless, beamforming is still attractive for N
C
r
> 2 at
the low SINR regime. In such case, the beamforming vectors
can be obtained numerically by using the gradient algorithm
proposed in Section III-B. Finally, it is worth to mention that
the beamformers obtained by all methods are independent
of neither the instantaneous value nor the variance of the
elements in the noise vector.
Finally, it is noted that the proposed schemes do not require
any joint optimization between the primary and secondary net-
work. However, knowledge of Du
opt
and v

opt
G is required
at the secondary network. The secondary receiver can obtain
that information by monitoring the primary link. For example,
when the primary transmitter sends a training sequence to the
primary receiver for channel estimation, the secondary receiver
can also monitor the sequence and estimate the effective
channel Du
opt
. Also we note that the secondary receiver can
estimate the effective channel Du
opt
directly and therefore, it
is not required for the secondary receiver to estimate the actual
channel D and the primary transmit beamforming vector
u
opt
separately. The secondary transmitter can estimate the
effective channel v

opt
G in a similar manner.
IV. NUMERICAL AND SIMULATION RESULTS
In this section, we present some simulation results. Since
the performance of the primary link is independent of the
secondary link, without loss of generality we assume N
P
t
=
N
P
r
= 1 for all results shown in this section for simplicity.
We rst consider an example of the gradient algorithm pre-
sented in Section III-B. For the initialization of the gradient
algorithm,
a[1] =
1
N
C
t
1
_
N
C
t
1
and b[1] =
1
N
C
r
1
_
N
C
r
1
(24)
are assumed. Furthermore, = 0.05 is used for the adapta-
tion size. These parameters are selected based on extensive
simulations to ensure rapid convergence of the algorithm.
Based on our observation, the initial values a[1] and b[1] have
little effect on the convergence behaviour of the algorithm.
Therefore, we chose the all one vectors for convenient. We
note that another convenient choice would be the vectors
obtained by Method 1. However, in that case, an exhaustive
search has to be performed before running the numerical
method. With the chosen parameters, it was found that the
algorithm generally converges within 100 iterations. As an
example, we consider one realization of the channels shown
in (23) assuming N
C
t
= 3 and N
C
r
= 2, The output of the
algorithm at i = 500 is given by (25). The corresponding
convergence behavior is shown in Fig. 2. Furthermore, with
YIU et al.: UNCOORDINATED BEAMFORMING FOR COGNITIVE NETWORKS 1395
G =
_
0.5140 + j 0.5567, 0.2146 j 0.6282, 0.2078 + j 0.8111

D =
_
0.5072 + j 0.3457, 1.1528 + j 0.7316

H =
_
0.7758 j 0.9689, 0.5724 + j 1.2102, 2.0819 + j 0.0723
1.0171 + j 0.1707, 0.2299 j 0.2257, 0.5338 j 0.2212
_
(23)
a[500] = [0.3984 j 0.2189, 0.6436 + j 0.2190]

and b[500] = [0.7071] (25)

G =

0.6340 j 0.4856, 0.0356 + j0.1627

0.2021 j 0.1605, 0.7960 + j 0.2071
0.5438, 0.5438

and

D =
_
0.9116 j 0.0299
0.4100
_
(26)
f
grad
=
_
0.1073 + j 0.2877, 0.6264 + j 0.1824, 0.6929

t
grad
=
_
0.9116 + j 0.0299, 0.4100

(27)
0 100 200 300 400 500
3.3
3.4
3.5
3.6
3.7
3.8
3.9
4
4.1
4.2
i
f
(
a
[
i
]
,
b
[
i
]
)
Fig. 2. f(a[i], b[i]) vs. iteration index i for the channels shown in (23). =
0.05 was assumed and (24) was used for the initialization of the algorithm.
the channel realizations in (23),

G and

D are given by (26).
Combining (26) with the results in (25), f
grad
and t
grad
are given by (27). For comparison, the optimal beamfomers
obtained by using Method 3 are given by (28) shown at
the bottom of this page. A quick calculation conrms that
Gf
grad
= t

grad
D = Gf
opt
= t

opt
D = 0.
The sum rate R
S
in (7) as a function of the average
SNR E{SINR
P
} = E{SINR
C
} for N
C
t
= 3 and N
C
r
= 2
is shown in Fig. 3. Note that since N
P
t
= N
P
r
= 1, W is
simply a scaler and u
opt
= 1 and v
opt
= W/

W can
be used for the primary link. The three methods presented
in Section III are used to generate the beamformers. For
0 5 10 15 20
1
2
3
4
5
6
7
8
9
10
SNR [dB]
S
u
m

R
a
t
e

[
b
p
s
/
H
z
]


1x1
Method 1
Method 2
Method 3
2x1
3x2
Fig. 3. Sum rate comparisons vs. SNR for a network with N
C
t
= 3.
comparison, the sum rate results of a network with an 1 1
primary link and a 11, 32, and 21 secondary link system
assuming that the two links do not interfere with each others
are also shown. In these scenarios, the beamformers f and t
are designed only to maximize the rate of the secondary link
assuming single stream without any interference nullication
constraint. This essentially gives us an idea of the price has
to be paid for interference nullication. As expected, the sum
rate performance of beamformers obtained from Method 2 and
Method 3 are identical and are better than the performance
of the ones obtained from Method 1. In general, Method 2
and Method 3 are 1.4 dB better than Method 1. Note also
that the results obtained for Method 2 and Method 3 coincide
f
opt
=
_
0.0089 + j 0.3070, 0.6518 j 0.0290, 0.6559 j 0.2232

t
opt
=
_
0.9116 + j 0.0299, 0.4100

(28)
1396 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 60, NO. 5, MAY 2012
0 5 10 15 20
0
2
4
6
8
10
12
SNR [dB]
S
u
m

R
a
t
e

[
b
p
s
/
H
z
]


1x1
Method 1
Method 2
Method 3
9x1
10x2
Fig. 4. Sum rate comparisons vs. SNR for a network with N
C
t
= 10.
with the 2 1 curve. This is in accordance with Theorem 1
presented in Section III. Finally, the gap between the 3 2
curve and Method 3 curve is about 4.5 dB and it essentially
represents the price that has to be paid in order to avoid cross
interference between the primary and secondary links. In the
last example shown in Fig. 4, we consider a system with
N
C
t
= 10 assuming the rest of the parameters unchanged as
in Fig. 3. Similar observation as in Fig. 3 can be made. In this
case, Method 2 is about 5 dB better than Method 1. The curve
for Method 2 and Method 3 coincides with the 9 1 curve as
expected. Finally, it is noted that the gap between the Method
3 and 10 2 curves reduces to about 2 dB. This is due to
the diminishing of return phenomenon observed generally in
MIMO system as the number of antennas increases.
V. CONCLUSION
In this paper, we considered interference cancellation and
rate maximization via uncoordinated beamforming in a cogni-
tive network which consists of a single primary and secondary
user. The secondary cognitive user was allowed to transmit
concurrently with the primary licensed user. The beamforming
vectors of the cognitive user were designed such that the
interference is completely nullied both at the primary and
secondary receivers while maximizing the rate of the cognitive
link. Since no interference is created at the primary receiver,
traditional approaches can be used to design the beamforming
vectors or precoding matrices of the primary user. Three
approaches were proposed for the design of the beamform-
ing vectors of the cognitive link. The optimal beamforming
solution for the special case of N
C
r
= 2 was also derived.
For N
C
r
> 2, we resorted to numerical methods to solve the
optimization problem. For an N
C
t
2 cognitive link, it was
found that if the optimal beamformers (the ones that nullify the
interference at both the primary and secondary receivers while
maximizing the achievable rate) are used at the cognitive link,
its rate is identical to an interference-free (N
C
t
1)1 system
employing a maximal ratio transmission beamformer. Finally,
it is noted that we motivate the uncoordinated beamforming
and rate maximization concept in a cognitive network. How-
ever, the results can also be applied to practical systems, e.g.,
small cell deployment in a macro network.
REFERENCES
[1] United States frequency allocations: the radio spectrum, U.S. De-
partment of Commerce, National Telecommunications and Information
Administration, Ofce of Spectrum Management, Oct. 2003.
[2] J. Mitola, Cognitive radio, Ph.D. dissertation, Royal Institute of
Technology (KTH), 2000.
[3] S. A. Jafar and S. Shamai, Degrees of freedom region of the MIMO
X channel, IEEE Trans. Inf. Theory, vol. 54, no. 1, pp. 151170, Jan.
2002.
[4] S. Sridharan and S. Vishwanath, On the capacity of a class of MIMO
cognitive radios, IEEE J. Sel. Topics Signal Process., vol. 2, no. 1, pp.
103117, Feb. 2008.
[5] N. Devroye, P. Mitran, and V. Tarokh, Achievable rates in cognitive
radio channels, IEEE Trans. Inf. Theory, vol. 52, no. 5, pp. 18131827,
May 2006.
[6] E. G. Larsson and E. A. Jorswieck, Competition versus cooperation on
the MISO interference channel, IEEE J. Sel. Areas Commun., vol. 26,
no. 7, pp. 10591069, Sep. 2008.
[7] S. Srinivasa and S. A. Jafar, The throughput potential of cognitive
radio: a theoretical perspective, IEEE Commun. Mag., vol. 45, no. 5,
pp. 7379, May 2007.
[8] T. Yucek and H. Arslan, A survey of spectrum sensing algorithms for
cognitive radio applications, IEEE Commun. Surveys Tuts., vol. 11,
no. 1, pp. 116130, Apr. 2009.
[9] Z. Quan, S. Cui, A. H. Sayed, and H. V. Poor, Optimal multiband
joint detection for spectrum sensing in cognitive radio networks, IEEE
Trans. Signal Process., vol. 57, no. 3, pp. 11281140, Mar. 2009.
[10] G. Ganesan and Y. Li, Cooperative spectrum sensing in cognitive radio
networks, in Proc. 2005 IEEE Dynamic Spectrum Access Netw., pp.
137143.
[11] S. M. Mishra, A. Sahai, and R. W. Brodersen, Cooperative sensing
among cognitive radios, in Proc. 2006 IEEE International Conf.
Commun., pp. 16581663.
[12] R. Zhang and Y.-C. Liang, Exploiting multi-antennas for opportunistic
spectrum sharing in cognitive radio networks, IEEE J. Sel. Topics
Signal Process., vol. 2, no. 1, pp. 88102, 2008.
[13] J. Zhou and J. Thompson, Linear precoding for the downlink of
mulitple input single output coexisting wireless systems, IET Commun.,
vol. 2, pp. 742752, July 2008.
[14] J. H. Winters, J. Salz, and R. D. Gitlin, The impact of antenna
diversity on the capacity of wireless communication systems, IEEE
Trans. Commun., vol. 42, no. 2, pp. 17401751, Feb. 1994.
[15] B. D. V. Veen and K. M. Buckley, Beamforming: a versatile approach
to spatial ltering, IEEE ASSP Mag., pp. 424, 1988.
[16] G. Jongren, M. Skoglund, and B. Ottersten, Combining beamforming
and orthogonal space-time block coding, IEEE Trans. Inf. Theory,
vol. 48, no. 3, pp. 611627, Mar. 2002.
[17] C.-B. Chae, D. Mazzarese, N. Jindal, and R. W. Heath, Jr., Coordinated
beamforming with limited feedback in the MIMO broadcast channel,
IEEE J. Sel. Areas Commun., vol. 26, no. 8, pp. 15051515, Oct. 2008.
[18] C.-B. Chae, S. Kim, and R. W. Heath, Jr., Network coordinated
beamforming for cell-boundary users: linear and non-linear approaches,
IEEE J. Sel. Topics Signal Process., vol. 3, no. 6, pp. 10941105, 2009.
[19] X. Jing and D. Raychaudhuri, A spectrum etiquette protocol for
efcient coordination of radio devices in unlicensed bands, in Proc.
2003 IEEE Personal, Indoor Mobile Radio Commun., pp. 172176.
[20] M. Buddhikot, P. Kolodzy, S. Miller, K. Ryan, and J. Evans, Dimsum-
net: new directions in wireless networking using coordinated dynamic
spectrum access, in Proc. 2005 IEEE International Symp. World
Wireless Mobile Multimedia Netw., pp. 7885.
[21] J. Perez-Romero, O. Sallent, R. Agusti, and L. Giupponi, A novel on-
demand cognitive pilot channel enabling dynamic spectrum allocation,
in Proc. 2007 IEEE Dynamic Spectrum Access Netw., pp. 4654.
[22] S. Panichpapiboon and J. M. Peha, Providing secondary access to
licensed spectrum through coordination, Wireless Netw., vol. 14, no. 3,
pp. 295307, June 2008.
[23] H. Bolcskei, MIMO-OFDM wireless systems: basics, perspectives, and
challenges, IEEE Wireless Commun. Mag., vol. 13, no. 4, pp. 3137,
Aug. 2006.
[24] J. G. Proakis, Digital Communications, 4th edition. McGraw-Hill, 2000.
YIU et al.: UNCOORDINATED BEAMFORMING FOR COGNITIVE NETWORKS 1397
[25] G. J. Foschini, Layered space-time architecture for wireless commu-
nication in a fading environment when using multi-element antennas,
Bell Labs Technical J., vol. 1, no. 2, pp. 4159, 1996.
[26] G. J. Foschini, G. D. Golden, R. A. Valenzuela, and P. W. Wolniansky,
Simplied processing for high spectral efciency wireless communi-
cation employing multi-element arrays, IEEE J. Sel. Areas Commun.,
vol. 17, no. 11, pp. 18411852, Nov. 1999.
[27] K. Washizu, On the bounds of eigenvalues, Quarterly J. Mechanics
Applied Mathematics, vol. 8, no. 3, pp. 311325, 1955.
[28] J. G. Andrews, W. Choi, and R. W. Heath, Jr., Overcoming interfer-
ence in spatial multiplexing MIMO cellular networks, IEEE Wireless
Commun. Mag., vol. 14, no. 6, pp. 95105, Dec. 2007.
Simon Yiu is a Member of Technical Staff at Bell
Laboratories, Alcatel-Lucent, Murray Hill, NJ, USA
. Prior to joining Bell Labs, he was a Postdoctoral
Fellow at the School of Engineering and Applied
Sciences at Harvard University, Cambridge, MA,
USA . He was also a lecturer while at Harvard Uni-
versity where he received a Certicate of Distinction
in Teaching. He received the Ph.D. degree in Electri-
cal and Computer Engineering from The University
of British Columbia, BC, Canada in 2007. Dr. Yiu
is a recipient of the Bell Labs Teamwork Award
and the NSERC Postdoctoral Fellowship and Canadian Graduate Fellowship.
His research interest spans the broad area of wireless communications and
cognitive networks.
Chan-Byoung Chae (S06-M09-SM12) is an As-
sistant Professor in the School of Integrated Tech-
nology, College of Engineering, Yonsei Univer-
sity, Korea. He was a Member of Technical Staff
(Research Scientist) at Bell Laboratories, Alcatel-
Lucent, Murray Hill, NJ, USA from June 2009
to Feb 2011. Before joining Bell Laboratories, he
was with the School of Engineering and Applied
Sciences at Harvard University, Cambridge, MA,
USA as a Post-Doctoral Research Fellow. He re-
ceived the Ph.D. degree in Electrical and Computer
Engineering fromThe University of Texas (UT), Austin, TX, USA in 2008,
where he was a member of the Wireless Networking and Communications
Group (WNCG). Prior to joining UT, he was a Research Engineer at the
Telecommunications R&D Center, Samsung Electronics, Suwon, Korea, from
2001 to 2005. He was a Visiting Scholar at the WING Lab, Aalborg
University, Denmark in 2004 and at University of Minnesota, MN, USA in
August 2007. While working at Samsung, he participated in the IEEE 802.16e
standardization, where he made several contributions and led a number of
related patents from 2004 to 2005. His current research interests include
capacity analysis and interference management in energy-efcient wireless
mobile networks and nano (molecular) communications. He serves as an
editor for the IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS and
the IEEE TRANSACTIONS ON SMART GRID.
Dr. Chae was the recipient of the IEEE Dan. E. Noble Fellowship Award
in 2008, the Gold Prize (1st) in the 14th Humantech Thesis Contest, and the
KSEA-KUSCO scholarship in 2007. He also received the Korea Government
Fellowship (KOSEF) during his Ph.D. studies
Kai Yang received the M.S. and Ph.D. degree in
Electrical Engineering from Columbia University in
2007 and 2009, respectively.
Dr. Yang joined Alcatel-Lucent Bell Labs at
Murray Hill, NJ in 2009. He is also an adjunct
faculty member at Columbia University. In 2007
and 2008, he worked at NEC Labs America on
LTE/WIMAX wireless networks and 100G long-
haul ber optical systems. Dr. Yangs research in-
terests include wireless communications, smart grid,
coding and optimization theory. He was awarded the
Eliahu Jury Award for the best Ph.D. thesis work from the E.E. department
of Columbia University. He is also the recipient of the 2011 Bell Labs
Teamwork Award. Dr. Yang serves as an editor for European Transaction
on Telecommunications, and the demo/poster co-chair of IEEE INFOCOM
2012.
Doru Calin (S95-M99-SM05) is the Director
of the High Performance Wireless Technologies
and Networks Group in Bell Laboratories, Alcatel-
Lucent, Murray Hill, New Jersey. He also serves as
an Adjunct Professor at the Columbia University,
NYC. He received the M.S. degree in electrical engi-
neering from the University of Bucharest, Romania,
and the M.S. and Ph.D. (with high honors) degrees
in electrical and computer engineering from the Uni-
versity of Versailles & TELECOM SudParis, France.
He has over 16 years of professional experience in
the telecommunication industry, cutting edge and applied research, standards,
products and services, built in both Europe and the USA. His current research
interests are in the areas of broadband wireless systems design, LightRadio
wireless networks design, creation of high value wireless services, cross-layer
optimization, converged networks, and performance analysis. Prior to joining
Bell Laboratories in early 2001, he was a researcher at the Institut National
des Telecommunications (now TELECOM & Management SudParis), Paris,
France, a technical consultant for Bouygues Telecom, and a Senior Research
Engineer at Motorola Labs in Paris. He has published over 50 papers in
scientic journals, book chapters, and refereed conferences, and holds 12
issued patents and over 20 pending patents. Dr. Calin has received a number
of major awards, including two Bell Laboratories Presidents Gold Awards,
four Bell Labs Teamwork Awards and Motorola 3GPP standard award. He is
a senior member of the IEEE and a member of the Alcatel-Lucent Technical
Academy.