IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO.

3, MARCH 2011

767

Linear Precoding for Orthogonal Space-Time Block Coded MIMO-OFDM Cognitive Radio
A. Punchihewa, Student Member, IEEE, Vijay K. Bhargava, Fellow, IEEE, and Charles Despins, Senior Member, IEEE

AbstractThe paper presents design of a linear precoder for orthogonal space-time block coded orthogonal frequency division multiplexing (OFDM)-based multiple-input multipleoutput (MIMO) antenna cognitive radio (CR) when operating in correlated Rayleigh fading channels. Unlike previous studies on precoder design for CR, this proposed linear precoder is capable of handling both transmit and receive correlation in a multi-carrier based CR system. The linear precoder is designed to minimize an upper bound on the average pairwise error probability, constrained to a set of per subcarrier transmit power constraints at the CR transmitter and a set of interference power thresholds at primary user receivers. The CR transmitter exploits the knowledge of transmit and receive correlation matrices while designing the precoder. It is shown that the linear precoder design problem is convex with these constraints, and convex optimization techniques are exploited to derive an efcient algorithm to obtain the optimal precoder matrices. Computer simulations are performed to investigate the performance of the proposed linear precoder in a CR system. Index TermsCognitive radio, convex optimization, orthogonal space-time block coding, multiple-input multiple-output antennas, orthogonal frequency division multiplexing.

I. I NTRODUCTION

HE electromagnetic radio spectrum is a precious resource available for wireless communications, which demands efcient usage. However, it has become increasingly scarce due to a wide deployment of wireless services. According to the Federal Communications Commissions spectrum policy task report [1], the usage of allocated spectrum varies from fteen to eighty-ve percent at specic time and geographical location. This low spectrum utilization coupled with spectrum scarcity motivates the development of novel spectrum-sharing technologies with the aim of improving spectrum utilization. Cognitive radio (CR) has emerged as a promising technology to improve spectrum utilization, while accommodating the growing amount of services and applications in wireless communications [2]. CR is capable of dynamically sensing and identifying unoccupied spectrum bands that are initially allocated to licensed (primary) users (PUs), and allowing unlicensed (secondary) users (SUs) to communicate through
Paper approved by I. Lee, the Editor for Wireless Communication Theory of the IEEE Communications Society. Manuscript received September 8, 2009; revised June 2, 2010 and August 16, 2010. A. Punchihewa and V. K. Bhargava are with the Department of Electrical and Computer Engineering, University of British Columbia, Vancouver, BC, Canada (e-mail: anjana@ece.ubc.ca, vijayb@ece.ubc.ca). C. Despins is with the Universit du Qubec, Montral, QC, Canada (email: cdespins@promptinc.org). Digital Object Identier 10.1109/TCOMM.2011.011811.090548

these available spectrum segments without causing harmful interference to PUs, thus having the potential to efciently improve spectrum utilization. Since CR operates with opportunistic spectrum sharing in dynamically changing environments, managing the quality of services (QoS) offered by a CR system while maintaining the QoS of the PUs, is challenging. Hence, proper design of a transmission scheme for CR to facilitate high data rate access and better performance along with high spectral efciency is very important. To achieve this objective, it is crucial to integrate recent physical layer technical advances into the CR systems. Multiple-input multiple-output (MIMO) antenna systems and space-time block coding (STBC) in wireless communications have attached considerable attention due to their ability to increase capacity and improve system performance over hostile wireless channels [3]-[5]. Orthogonal frequency division multiplexing (OFDM) is a promising transmission technique in CR systems due to its several advantages such as scalability, robustness against multipath fading, multiple access mechanisms, simplicity in channel equalization and coding [6]. Therefore, with these valuable features, incorporating MIMO, STBC and OFDM into CR would promise enhanced performance in terms of spectral efciency, capacity and bit error rate over hostile wireless channels. It is shown in previous studies that the performance of conventional MIMO systems is degraded in spatially correlated channels based on the available channel state information (CSI) at the transmitter [7], [8]. However, efcient precoding techniques in combination with STBC can be used to further improve the system performance in such channel conditions, when the knowledge of CSI is available at the transmitter [9]-[12]. Therefore, linear precoding is a vital technique to combat the correlation effect of MIMO channels. In practice a perfect CSI is seldom available and is difcult to obtain at the transmitter. Thus, a common practice is to assume partial channel knowledge at the transmitter, for example, in terms of transmit or both transmit and receive correlation matrices [9]-[12]. In conventional MIMO systems [9]-[12], the linear precoder is designed with the knowledge of transmit or both transmit and receive correlation matrices at the transmitter by minimizing a metric related to average error probability and constrained only to total transmit power. Although this topic has been extensively studied for conventional MIMO systems, less attention is given in previous studies for design of a linear precoder for MIMO-based CR, where additional constraints need to be incorporated in pre-

c 2011 IEEE 0090-6778/11$25.00

768

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 3, MARCH 2011

coder design. The linear precoder designed in [13] for CR with the intention of improving the error rate performance assumes only the SU transmit antenna correlation, single antenna at the PU receiver and considers the single carrier transmission scheme. However, this paper extend the linear precoder design for CR in several ways. The theoretical analysis of the linear precoder design is based on a comprehensive signal model that takes into account of multiple antennas at both SU and PUs, multi-carrier transmission scheme and the correlation effects of SUs both transmit and receive antennas. A linear precoder is designed for orthogonal space-time block coded (OSTBC) MIMO-OFDM based CR, when operating in frequency-at correlated Rayleigh fading channels. The linear precoder is designed to minimize an upper bound on the average pairwise error probability (PEP) when the SU transmitter has the knowledge of transmit and receive correlation matrices, while imposing a set of interference power constraints at the PUs and a set of per subcarrier transmit power constraints at the SU transmitter. It is shown that the precoder design problem for CR is convex with these constraints. Furthermore, an efcient algorithm based on the Lagrange dual-decomposition is proposed to obtain the linear precoder. The individual effects of the SU transmit and receive antenna correlation on the linear precoder design for CR is also addressed in this paper. A closed-form solutions for power loading in each OFDM subcarrier for simplied correlation scenarios are also presented. The rest of the paper is organized as follows. The system model and OSTBC MIMO-OFDM transmission scheme are introduced in Section II. The optimal linear precoder design problem is formulated in Section III. The linear precoder design with SUs different correlation scenarios are investigated in Section IV, and the Lagrangian dual-decompositionbased efcient algorithm is proposed to obtain the linear precoder. Simulation results are provided in Section V. Finally, conclusions are drawn in Section VI. Proofs of the theorems are given in the Appendix. The following notations are used throughout the paper. Vectors are denoted by boldfaced lowercase letters, e.g., a, b, and matrices are denoted by boldface uppercase letters, e.g., A, B. The superscripts ()1 , () , () , and ()1/2 stand for inverse, transpose, conjugate transpose, and square root, respectively. {} is the trace of a square matrix, det[] is the determinant of a matrix, E{} is the expectation operation, vec() is the vectorization operation, is the Frobenius norm of a matrix and is the Kronecker product. I is the identity matrix and A 0 indicates that the square matrix A is positive semi-denite. II. S YSTEM M ODEL A. System Description A CR network as illustrated in Fig. 1 is considered, where a single pair of SU transmitter and receiver coexist with the PUs receivers. We assume that PUs and SU share the same bandwidth. The SU pair are equipped with and transmit and receive antennas, respectively, while the PUs have receive antennas each. The SU transmitter employs subcarriers to modulate the signal using OFDM.

H sp2

H sp1

PU Receiver 2
H spL
H ss

PU Receiver 1

PU Receiver L SU Transmitter
Fig. 1. A cognitive radio network.

SU Receiver

Furthermore, we assume that only the transmit and receive correlation matrices for the MIMO channels, between the SU transmitter and PU receivers, and between the SU transmitter and SU receiver, are available at the SU transmitter. These matrices are obtained at the SU transmitter by periodically sensing the transmitted signals from the PU and SU receivers. In addition, perfect CSI is assumed at the SU receiver. B. Correlated Channel Model Quasi-static frequency-at correlated Rayleigh MIMO channels between the SU transmitter and PU receivers, and between the SU transmitter and SU receiver are considered. Under the assumption of SU transmit and SU, PUs receive scattering radii are large compared to the distance between the SU transmitter and SU, PUs receivers, the MIMO channels between the SU transmitter and SU receiver and between the SU transmitter and PU receivers for the th subcarrier can be respectively written as [14]
/2 H ( ) = R1 , ( )H, ( )R, ( ), = 1, . . . , , (1) 1/2

and
/2 H ( ) = R1 , ( )H, ( )R, ( ), 1/2

= 1, . . . , , = 1, . . . , ,

(2)

where R, ( ), R, ( ) are the transmit and receive correlation matrices of sizes and , respectively, for the MIMO channels between SU transmitter and SU receiver; and R, ( ), R, ( ), are the transmit and receive correlation matrices of sizes and , respectively, for the MIMO channels between SU transmitter and PU receivers. H, ( ) and H, ( ) are matrices of sizes and , respectively, with independent and identically distributed (i.i.d.) zero-mean circular symmetric complex Gaussian (CSCG) entries with unit variance. The transmit and receive correlation matrices for the MIMO channels between the SU transmitter and SU receiver, and between the SU transmitter and PU receivers can be respectively written as [14] { } 1/2 1/2 R, ( ) = R, ( ) R, ( ) = H ( ) H ( ) , (3) = {, }, = 1, . . . , , = 1, . . . , ,

PUNCHIHEWA et al.: LINEAR PRECODING FOR ORTHOGONAL SPACE-TIME BLOCK CODED MIMO-OFDM COGNITIVE RADIO

769

and R, ( ) =
1/2 1/2 R, ( )R, ( )

{ } = H ( )H ( ) ,

= {, }, = 1, . . . , , = 1, . . . , .

(4)

Input Bits

S/P

STBC Encoder

S/P, IFFT P/S, CP Addition Precoder F(k) S/P, IFFT P/S, CP Addition

The full autocorrelation matrices, R ( ) and R ( ), can be obtained in terms of the Kronecker product of the transmit and receive correlation matrices: R ( ) = R, ( ) R, ( ), = 1, . . . , , R ( ) = R, ( ) R, ( ), = 1, . . . , , = 1, . . . , . (5) (6)
Received Bits CP Removal, S/P FFT, P/S ML Decoder CP Removal, S/P FFT, P/S

H ss (k )

By applying the vectorization operation to (1) and (2), we have /2 vec (H ( )) = R1 ( )vec (H, ( )), and vec(H ( )) = 1/2 R ( )vec(H, ( )), respectively. This Kronecker model has been widely exploited in the previous studies for correlated MIMO systems [10]-[12], [15], [16]. C. Transmission Scheme A block diagram of the proposed OSTBC MIMO-OFDM based CR transmission scheme with linear precoding is shown in Fig. 2. The SU transmitter includes an OSTBC encoder and a linear precoder followed by an OFDM modulator. At the SU transmitter, the input symbols are rst serial to parallel (S/P) converted and fed into the OSTBC encoder. Secondly, the output symbols of the OSTBC encoder are multiplied by precoder matrices F( ), = 1, . . . , . Third, the precoded data to be transmitted by each transmitting antenna are subjected to typical OFDM transmit processing, such as S/P conversion, inverse fast Fourier transform operation, parallel to serial (P/S) conversion and the addition of a cyclic prex (CP). Fourth, the precoded data are transmitted over the wireless MIMO channel. Then at the SU receiver, the received signal in each antenna is subjected to typical OFDM processing such as removal of CP, S/P conversion, fast Fourier transform operation and P/S conversion. Finally, the maximum likelihood (ML) detector recovers the received data symbols. For this system, the received signal at the SU receiver for the th subcarrier can be written as ( ) + N( ), = 1, . . . , , Y ( ) = H ( )F( )C (7) ( ) is the transmitted OSTBC matrix of size where C ofdm , with ofdm as the total OFDM symbols transmitted in a ( ) are drawn from block of data. Individual data symbols of C a nite complex signal constellation with unit energy. N( )1 is the complex additive white Gaussian noise matrix of size 2 I . In addition, ofdm with zero-mean and variance the same noise statistics are assumed for all subcarriers. Under the assumption of perfect CSI at the SU receiver, the ML decoding of codeword C from the received signal matrices Y ( ), = 1, . . . , , yields C = arg min
C =1

Fig. 2. System block diagram of the precoded OSTBC MIMO-OFDM based CR transmission scheme.

III. O PTIMAL L INEAR P RECODER D ESIGN P ROBLEM F ORMULATION The main objective is to nd a set of linear precoder matrices F( ), = 1, . . . , , at the SU transmitter, so as to minimize an upper bound on the average PEP under a set of per subcarrier power constraints at the SU transmitter and a set of PUs interference power thresholds. Upper bounds on the average PEP have been extensively exploited as design criteria of the linear precoder in conventional MIMO systems [9]-[12]. In our framework, an upper bound on the average PEP is also adopted as a design criteria, but for OSTBC MIMO-OFDM based CR transmission scheme. The PEP, (C C ) is the probability that ML decoding decides in favor of the codeword C instead of the actually transmitted codeword C . Theorem 1: An upper bound of the average PEP of the OSTBC MIMO-OFDM, when SUs transmit and receive antennas are correlated, can be written as: (C C ) { [ ]}1 ( ), ( ) det I + , ( ) F ,
=1 =1

(9)

( ) Y ( ) H ( )F( )C

(8)

1 Note that the noise at the SU receiver also contains the interference from the PU transmitters in the network and therefore non-white in general. By applying the noise-whitening lter at the SU receiver and incorporating the lter effects into MIMO channel matrix between the SU transmitter and SU receiver, the effective noise is assumed to be approximately white Gaussian.

where is a factor that depends on the codeword pair C and C , , ( ) is the th eigenvalue of the receive eigenvalue ( ) = F( )F( ) , and , ( ) is the matrix , ( ), F transmit eigenvalue matrix of the transmit correlation matrix R, ( ). Proof: See the Appendix. In CR networks, CR users may coexist with PUs either on a non-interfering basis or on an interference tolerance basis [17]. Therefore, one fundamental challenge of the CR is to maintain the QoS of the PUs while maximizing the SUs performance. Since PUs have a higher priority than the SUs while opportunistically sharing the spectrum in an interference tolerance basis, SUs have to maintain interference introduced to the PUs by SUs below a certain threshold, known as the interference temperature constraint and dened by regulatory bodies. Therefore, the QoS of the PUs in the CR network is maintained by introducing the additional interference power constraints, measured at the PUs receivers [18], [19] into the precoder design problem for CR. The average interference

770

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 3, MARCH 2011

power measure is appropriate for delay-insensitive communications and has been extensively exploited in the previous studies to limit the interference from SU transmission to the PUs [18], [19]. Following considers the average interference power introduced to the PU receivers by SU transmission. The interference power introduced by SU transmission at the th PU receiver, conditioned on the input signal constellation and the channel realization, can be written as ( ) { ( ), H( ) = H ( )F( )C ( )C ( ) C } (10) F( ) H ( ) , = 1, . . . , , = 1, . . . , . Therefore, the total average interference power introduced by the SU transmission to each PU can be obtained under the assumption of independent channel realization for each subcarrier and by taking the expectation of (10) with respect to input signal and the channel realization as , =
=1

{ } { ( )C ( ) F( ) F( )E C }} { E H ( ) H ( ) , = 1, . . . , , (11) { } ( )R, ( ) , = 1, . . . , , F

=1

where the unit variant input signal constellation is assumed ( )C ( ) } = I ). (i.e., E {C In OFDM transmission schemes, the transmitted signal power should be limited to avoid generating strong interference to other active users and the systems, and to avoid requirement of linear ampliers with large dynamic range. Furthermore, decreasing the transmit power will prolong the battery lifespan. Therefore, in this CR network per subcarrier power constraint is imposed at the SU transmitter. Under the assumption of unit variant constellation, the transmit power from the th subcarrier can be written as P( ) = ( )}, = 1, . . . , . {F An upper bound on the average PEP of OSTBC is exploited to obtain a set of linear precoder matrices F( ), = 1, . . . , at the SU transmitter. By exploiting the property that the logarithmic function is monotonic increasing for nonnegative values, we can obtain the optimum linear precoder matrices F( ), = 1, . . . , , that minimize an upper bound on the average PEP by solving the following optimization problem (P1) as presented in (12)-(15). In the previous equations (12)-(15), I, is the interference power threshold specied by the th PU and P ( ) is the transmit power available
=1 =1 =1

for the th subcarrier at the SU transmitter. Equations (13) and (14) represent the average interference power constraint over all receive antennas for th PU receiver and the per subcarrier transmit power constraint at the SU transmitter, respectively. The third constraint implies that the matrices ( ), = 1, . . . , , should be positive semi-denite. Note F that this linear precoder design problem is different from the precoder design for conventional systems due to the additional interference power constraints in (13). Therefore, the precoder obtained by standard multi-level water-lling is not optimal for this problem. Based on the convexity of the optimization problem P1 and the structure of the optimal precoder, the following theorem and the lemma can be stated. Theorem 2: The linear precoder design problem P1 is convex with constraints (13)-(15). Proof: See the Appendix. Lemma: If F( ), = 1, . . . , is an optimal solution to the problem P1, then the linear precoder F( )U( ), = 1, . . . , , where U( ), = 1, . . . , is a unitary matrix is also optimal. Proof: Since U( ), = 1, . . . , is a unitary matrix and by insertion of F( )U( ), = 1, . . . , , into the objective function (12) and the constraints (13)-(15) remain unchanged. Therefore, F( )U( ), = 1, . . . , , is also an optimal solution to the problem P1. Since the optimization problem P1 is convex, standard numerical optimization techniques, e.g., the interior-point method [20] can be employed to obtain the optimal linear precoder. The details of this method are omitted for brevity. However, to get more insight into the system performance, Lagrange dual-decomposition-based algorithm is proposed in the next section to obtain the set of linear precoder matrices. IV. P RECODER D ESIGNS FOR CR IN C ORRELATED MIMO C HANNEL Standard convex optimization techniques are exploited to derive efcient algorithms to obtain the optimal precoder matrices for SUs different transmit and receive correlation scenarios. This section will present the solution to the precoder design problem P1 using eigen-beamforming. In this scenario, the linear precoder F( ) functions as a multi-mode beamformer based on the knowledge of the transmit and receive correlation matrices of the MIMO channels between the SU transmitter and SU receiver and between the SU transmitter and PU receivers. The optimal precoder F( ) has its orthogonal beams directions with the left eigenvectors

minimize
(), =1,..., F

[ ] ( ), ( ) , log det I + , ( ) F

(12) (13) (14) (15)

subject to

{ } ( )R, ( ) I, , = 1, . . . , , F

{ } ( ) P ( ), = 1, . . . , , F

( ) 0, = 1, . . . , F

PUNCHIHEWA et al.: LINEAR PRECODING FOR ORTHOGONAL SPACE-TIME BLOCK CODED MIMO-OFDM COGNITIVE RADIO

771

of the SUs transmit correlation matrix R, ( ), and the power loading across the beams as the square values of the ( ). Thus, by taking eigenvalue deeigenvalues of the matrix F ( ) = U ( ) ( )U ( ) , choosing optimal composition of F F F F eigen beam directions to be UF ( ) = U, ( ), and using the properties of eigenvalue, the optimization problem (P2) in eigen-beamforming can be rewritten as in (16)-(19). In equations (16)-(19), , ( ) , F ( ) are the th eigenvalues ( ) , respectively; and u, ( ) is the of , ( ) and F th eigenvector of the U, ( ). Next, the Lagrange dualdecomposition method is applied to obtain the optimal power allocation across each antenna and for each subcarrier in different SUs transmit and receive antenna correlation scenarios. In previous studies, the Lagrangian dual-decomposition method has been extensively exploited for resource allocation in communication systems [21], [22]. A. Precoder Design With the SUs Both Transmit and Receive Correlation The general scenario, considering both SU transmit and receive antenna correlation is presented. This is the common situation encountered in the uplink of a communication link, where a multi-antenna access point is located high above the multi-antenna subscriber transmit units. Since the access point is installed at a high location, the receive antennas are having a small spread of angle of arrival due to the less scatters around it. Therefore, causing a high receive antenna correlation at the receive side. In the case of multi-antenna subscriber transmit unit, the possible causes of transmit correlation are lack of spacing between antennas, antenna arrangement and the antenna congurations. It can be noticed from the optimization problem P2 that the SUs transmit and receive antenna correlations have different effects on the linear precoder design for CR. Particularly, the objective function in (16) depends on the SUs both transmit and receive correlation matrices R, ( ) and R, ( ) through the eigenvalues , ( ) , and , ( ) , respectively. Furthermore, the interference power constraint depends on the transmit correlation matrices R, ( ). The Lagrange dual-decomposition method is applied here to obtain the optimal values of the F ( ) . The optimal power allocation policy for the th subcarrier and the optimal precoder matrix, F ( ), can be obtained according to the following theorem.
=1 =1 =1

Theorem 4: The optimal power allocation for the th + transmit antenna and for the th subcarrier, ( ) , = F 1, . . . , , with + = max{0, } can be obtained by solving the following set of equations:
=1

, ( ) ( ) = 1 , ( ) + , ( ) ( ) F ( ) u, ( ) R, ( )u, ( ) , = 1, . . . , , (20)

=1

where and are the non-negative Lagrange multipliers associated with per subcarrier power constraint and the interference power constraint, respectively. Therefore, the optimum linear precoder matrix F ( ) for the th subcarrier can be obtained as F ( ) = U, ( ) ( ). F Proof: See the Appendix. It can be noticed from equation (20) that for this scenario, the optimal power loading across the th transmit antenna and for th subcarrier depends on the eigenvalues of the SUs transmit and receive correlation matrices, eigenvectors of the SUs transmit correlation matrix and the transmit correlation matrices of the MIMO channels between the SU transmitter and the PU receivers. Furthermore, the interference power introduced at the PU receivers are controlled by and , which are calculated based on the constraints (17) and (18). Normally, the cross-correlation between pairs of antennas is much smaller than one and, as a result for a well behaved receive correlation matrix, the values of , ( ) are close , ( ) = to each other and can be approximated as 1 {, ( )}. In this scenario, the optimal solution + ( ) , = 1, . . . , , for the th subcarrier can be obtained F from the following closed-form solution
F ( ) = ( +
=1

u, () R, ( )u, ( ) )+
1

) 1

( {, ()})

, = 1, . . . , . (21)

To summarize, the complete algorithm for precoder design with both transmit and receive correlation is given below. Note that the solution derived for the linear precoder problem in this paper is for a general scenario. The solution for conventional MIMO systems can be straightforwardly

minimize
{ F ()}

( ) log 1 + , ( ) , ( ) F ( ) ,

(16) (17) (18) (19)

subject to

=1 =1 =1

( ) F ( ) u, ( ) R, ( )u, ( ) I, , = 1, . . . , ,

F ( ) P ( ), = 1, . . . , ,

F ( ) 0, = 1, . . . , , = 1, . . . ,

772

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 3, MARCH 2011

Algorithm 1 Computation of optimum precoder matrix

1:

in eigen-beamforming for this scenario can be written as minimize

{ F ()} =1 =1

2: 3: 4:

5:

6: 7: 8:

Given , , initial ellipsoids centered at (0) , (0) , which contain the optimal dual solutions and , respectively. Set = 0. repeat For each subcarrier = 1, . . . , obtain the optimal power allocation ( ) , = 1, . . . , by solving F (20). (+1) (+1) () () Update , using , and the sub() () gradients of , = 1, . . . , , and , = 1, . . . , , respectively [23]. Set (+1) and (+1) as the centers of new ellipsoids (+1) (+1) and , respectively. Set + 1. until the stopping criteria of the ellipsoid method is satised [23].

(0)

(0)

) ( log 1 + , ( ) F ( ) , (22)

subject to

(17), (18), and (19).

obtained by simply setting the Lagrange multipliers corresponding to average interference power constraints to zero (i.e., = 0). Furthermore, the linear precoder design for the single carrier systems in a CR system can be obtained with = 1. In addition, in the case of a single PU, there is only one Lagrange multiplier for the interference power constraint. In such a scenario, this Algorithm 1 can be further simplied and the can be updated by the bisection method [20].

It can be seen that the problem P4 has a similar structure as problem P2. Therefore, the Lagrange dual-decomposition method can be exploited to solve problem P4 and derive an efcient algorithm to obtain the optimal precoder matrices F( ) for = 1, . . . , . A similar analysis can be performed as described in Section IV.A. However, the analysis is presented to obtain the optimal precoder matrix for a OFDM subcarrier as described in subproblem P3, in order to obtain the dual function ( , ), for given and . Therefore, subproblem P5 for the th OFDM subcarrier can be written as in (23). The optimal power allocation, ( ) , = 1, . . . , , for this F scenario can be obtained as in the subproblem P3, by applying the KKT condition for the convex subproblem P5. Therefore, the optimal power allocation, ( ) , = 1, . . . , , for F the th subcarrier with no receive correlation is given by the following closed-form water-lling like solution for given and :
( F ( ) = +
=1

( ) u, () R, ( )u, ( ) )+
1

) 1

(, () )

, = 1, . . . , . (24)

B. Precoder Design With Only the SU Transmit Correlation The linear precoder design considering only the SU transmit antenna correlation is presented in this subsection. In practice, this situation is encountered in a downlink of a communication system when the multi-antenna subscriber units have sufcient antenna spacing between them. In this scenario, the multiantenna base station is situated high above the ground in a low scattering environment and thus, results a high transmit antenna correlation. The multi-antenna subscriber unit is situated in rich scattering environment and with sufcient spacing between antennas, thus no receive antenna correlation. For this scenario, the SU receive correlation matrix is equivalent to the identity matrix, i.e., R, ( ) = I , or equivalently , ( ) = I . Therefore, using the properties of the determinant, the Kronecker product, and applying some mathematical manipulations to P1, the optimization problem (P4)
=1

Algorithm 1 proposed in Section IV.A to obtain the precoder matrices F( ) for = 1, . . . , now needs to be slightly modied for this scenario. In Step 4, the optimal solution ( ) , F = 1, . . . , , for each OFDM subcarrier = 1, . . . , , can now be obtained by (24). Therefore, the main advantage of this method in multicarier transmission scheme is that the same computational routine can be simultaneously applied to all subcarriers in order to obtain the optimal F( )s. Thus, the overall computational time will be maintained regardless of the number of subcarriers . This is a signicant advantage in CR networks since the number of subcarriers are dynamically assigned and the overall computational time of the linear precoder design algorithm will remain approximately the same as the single carrier systems. Furthermore, for this scenario the convergence time of the overall algorithm is further improved due to the closed-form solution on ( ) , = 1, . . . , in F (24). ( (
=1

minimize
F () , =1,...,

( ) log 1 + , ( ) F ( ) + =1 =1

F ( ) u, ( ) R, ( )u, ( )

) )

+ subject to

F ( ) ,

(23)

F ( ) 0, = 1, . . . ,

PUNCHIHEWA et al.: LINEAR PRECODING FOR ORTHOGONAL SPACE-TIME BLOCK CODED MIMO-OFDM COGNITIVE RADIO

773

C. Precoder Design With Only the SU Receive Correlation In this subsection, the linear precoder design considering the SU receive correlation and no SU transmit correlation is explored. This situation is encountered in an uplink of a communication system when the multi-antenna subscriber units are located in a rich scattering environment with sufcient antenna spacing. There are two possible cases involving the receiveside correlation, i.e., either R, ( ) = I or R, ( ) = R, ( ) = I , = 1, . . . , . The latter is considered to be the case when only the SUs receive-side correlation is considered, with no transmit side correlation. The former case is considered rst. For this scenario, R, ( ) = I , or equivalently , ( ) = I . Therefore, using the properties of the determinant, the Kronecker product, and some trivial calculations to problem P1, the optimization problem (P6) in eigen-beamforming for this scenario can be written as minimize
{ F ()} =1 =1 =1

system of equations:
=1

( )1 , ( ) 1 + , ( ) = ( ) F (28) , = 1, . . . , .

=1

It can be noticed from the equations (27) and (28) that the optimal matrix F ( ) for the th subcarrier has equal diagonal ( ) , elements = 1, . . . , for given and . Since this F is the uncorrelated scenario and as CSI is not available at the SU transmitter, an equal diagonal precoder is expected. The algorithm proposed in Section IV.A, to obtain the precoder matrices F( ), = 1, . . . , now needs to be modied for these scenarios. Step 4 of Algorithm 1 can now be obtained by (27) and (28), for the cases R, ( ) = I and R, ( ) = R, ( ) = I , = 1, . . . , , respectively. V. N UMERICAL R ESULTS In this section, numerical results are presented to illustrate the performance of the proposed linear precoder in an OSTBC MIMO-OFDM based CR system. Throughout the simulations, the bit error rate (BER) is used as the performance measure. In all the simulations, quasi-static frequency-at correlated Rayleigh fading MIMO channels and zero-mean uncorrelated 2 are assumed. The elements CSCG noise with variance of the MIMO channel matrices H, ( ) and H,1 ( ), for = 1, . . . , , are generated as i.i.d. samples of CSCG distribution with zero-mean and unit variance. Furthermore, it is assumed that the transmit and receive antennas form linear arrays for both SUs and PUs. The correlation coefcient between the th and the th transmit antennas for the th subcarrier with a small angle spread can be approximately obtained as [7]: [R, ( )], 1 2 2 2 sin , (), sin , 0 exp 2 1 = , 0 (2 sin , ( ), ), = {, }, = 1, . . . , , = 1, . . . , . Here, 0 () is the zeroth-order Bessel function of the rst kind, , ( ) is the transmit angle spread for the th subcarrier, , is the spacing between the transmit antennas, is the wavelength of the 2 is the transmit antenna array power gain. The carrier and , , ( ), = {, }, for each subcarrier is generated from , + /2] , /2, a uniform distribution in the range [ , , = {, } as the mean transmit angle spread with and = 60 . The receive correlation matrices R, ( ) and R, ( ), = 1, . . . , , = 1, . . . , are also obtained similarly as the transmit correlation matrices. Furthermore, ( (
=1

) ( log 1 + , ( ) F ( ) , (25)

subject to

(17), (18), and (19).

It is apparent from problem P6 that it has a similar structure to problems P2 and P4. Therefore, as previously discussed, the Lagrange dual-decomposition method can also be exploited for this scenario, to solve and derive an efcient algorithm to obtain the precoder matrices F( ), = 1, . . . , . However, only the procedure to obtain the precoder matrix F( ) for a OFDM subcarrier, similar to the subproblem P3, is explored here in order to obtain the dual function ( , ), for given and . Therefore, the subproblem (P7) for the th subcarrier in this scenario can be written as in (26). Applying the Lagrangian multiplier method to subproblem P7, one can nd the optimal ( ) , = 1, . . . , for the th subcarrier by F solving the following system of equations:
=1

( )1 , ( ) 1 + , ( ) = ( ) F ( ) , = 1, . . . , . u, ( ) R ( ) u ( ) , , (27)

=1

Similarly, the optimal ( ) , = 1, . . . , for the th F subcarrier for the latter case where, R, ( ) = R, ( ) = I , = 1, . . . , , can be obtained by solving the following
=1 =1

minimize
F () , =1,...,

) ( log 1 + , ( ) F ( ) + =1

F ( ) u, ( ) R, ( )u, ( )

+ subject to

=1

F ( ) ,

(26)

F ( ) 0, = 1, . . . ,

774
0

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 3, MARCH 2011

10

2
tx,ss = 300 Iint = 100 mW tx,sp = 200 tx,ss = l

4 10
1

Received Interference Power at a PU (dB)

BER

10

10

2 1 , No Precoding 10
3

2 1 , Precoding 2 2 , No Precoding 2 2 , Precoding 4 2 , No Precoding 4 2 , Precoding 2 4 6 8 SNR (dB) 10 12 14 15

12 Iint = 100 mW 14 Iint = 200 mW Iint = 300 mW Iint = 400 mW 16 0 2 4 6 8 SNR (dB) 10 12 14 15

10

Fig. 3. The average BER versus SNR for the CR transmission scheme with and without precoding.
10
0

Fig. 5. The average received interference power versus SNR for different interference power thresholds.

tx,sp = 200 tx,ss = l

Iint = 100 mW Iint = 200 mW Iint = 300 mW Iint = 400 mW

10

10

10

10

8 SNR (dB)

10

12

14

15

Fig. 4. The BER performance of the precoded OSTBC MIMO-OFDM based CR transmission scheme with different interference power thresholds.

antennas at transmitter and receiver are assumed to be uniformly spaced with half wavelength distance between them for both SUs and PUs. Two PUs are assumed to be present in the system. The number of subcarriers is set to 16. The number of receive antennas for both PUs and SU are either set to 1 and 2. The total available transmit power for a subcarrier at the SU transmitter is 1/ W. The PU receiver interference temperature power threshold is set to 100 mW [13]. Binary phase shift keying symbols are generated with unit variant signal constellation for signal transmission. The Alamouti code was exploited as the OSTBC for the SU transmission [5]. Simulation results are obtained by averaging 1000 trials, with each block consisting of 10000 OSTBC OFDM symbols. Finally, the signal to noise ratio (SNR) is dened as the signal power to noise power at the SU receiver. The average BER of the CR system is plotted versus SNR in Fig. 3 for proposed precoded system and with non-precoded system. In order to have a fair compression, the transmit power is uniformly loaded in non-precoding such that the interference power constraint satised. In this scenario, the transmit power from the th transmit antenna is obtained

( ) 1 u, ( ) , by F R ( ) u ( ) ,u ( ) = I, , , = 1, . . . , , = 1, . . . , . This will ensure in uniform power loading that the total interference introduced is below the prescribed threshold value. The BER results are presented for 2 1, 2 2 and 4 2 SU antenna congurations with a 100 mW interference power threshold. From the plots, it can be seen that the proposed precoder in CR outperforms the non-precoding scheme. The precoding gain is around 1.2 dB for the presented SNR range. Similar performance results are achieved for all three antenna congurations. Furthermore, it is noticed that signicant improvement in performance by adding more antennas to the SU receiver with the same interference power constraint. Thus, employing multiple antennas at the CR transmitter and receiver, system error performance can be improved signicantly for a xed interference power threshold. In Fig. 4, the average BER is plotted against SNR for the 2 2 CR system with different interference power thresholds. As expected, BER performance improvement can be seen with increased interference power thresholds. The increase of the interference power limit allows the CR transmitter to allocate higher power to the precoder matrices. However, interference power cannot be increased signicantly, since it could increase an unacceptable interference receive at the PU receivers. In Fig. 5, the average received interference power at a PU receiver is plotted against the SNR for different interference power thresholds. It is apparent from Fig. 5 that as the interference power threshold increases the amount of received interference also increases. In addition, it can be seen from Fig. 5 that for all the cases, the received interference power at the PU is below or equal to the respective interference power thresholds, thus allowing CR and PU to coexist in the same frequency band. The average received interference power at a PU receiver is plotted against SNR in Fig. 6 for different number of subcarriers. From Fig. 6 it is apparent that for all the subcarriers the received interference power at the PU remain below or equal to the interference power threshold. Furthermore, as the number of subcarriers increases, the received interference power at the PU receiver reach the interference power threshold at a higher

BER

PUNCHIHEWA et al.: LINEAR PRECODING FOR ORTHOGONAL SPACE-TIME BLOCK CODED MIMO-OFDM COGNITIVE RADIO
0

775

tx,sp = 200 tx,ss = l

3

K=8 K = 12 K = 16

10

Iint = 100 mW

tx,ss = tx,sp = 15 0 l tx,ss = tx,sp = 20 0 l tx,ss = tx,sp = 25 0 l

Received Interference Power at a PU (dB)

10

BER 10
2

SNR (dB)

10

12

14

15

10

8 SNR (dB)

10

12

14

15

Fig. 6. The average received interference power versus SNR for different number of subcarriers.
10
0

Fig. 8. The average BER versus SNR of the precoded OSTBC MIMO , . OFDM based CR transmission scheme with different

tx,ss = tx,sp = 20 0 l tx,ss = tx,sp = 20 0 and rx,ss = 30 0 l

10

10

10

8 SNR (dB)

10

12

14

15

correlation matrices at the CR transmitter and by minimizing an upper bound on the average pairwise error probability, while imposing a set of per subcarrier transmit power constraints at the CR transmitter and a set of interference power constraints specied by the PUs. We have shown that the precoder design problem with these constraints is convex and have proposed a Lagrange dual-decomposition-based efcient algorithm to obtain the optimal precoders. Furthermore, the individual effects of the transmit and/or receive correlation on the linear precoder design for CR were investigated. The current study further reveals that for uncorrelated CR transmit antennas, the precoder consists of equally weighted diagonal elements. Simulation results illustrate that the proposed linear precoder outperforms uniform power loaded systems in a correlated MIMO channel even with the presence of additional interference power constraints. A PPENDIX A. Proof of Theorem 1 An upper bound of the average PEP of OSTBC MIMOOFDM, when the SUs transmit and receive antennas are correlated is derived here. Applying the Chernoff bound to (8), a tight upper bound on the PEP conditioned on channel matrices H ( ), = 1, . . . , , can be obtained as [24] (C C H (1), . . . , H ( )) where d2 (C C ) =
1 ( ) C ( )) H ( )F( )(C 2 2 =1 2
d2 (C C ) 1 2 exp , (29) 2

Fig. 7. The average BER performance of a CR system for SUs transmit correlation and both transmit and receive correlation scenarios.

SNR value. In Fig. 7, the average BER is plotted against SNR for a 2 2 CR system for SU transmit and receive correlations, and for only SU transmit correlation, at a 100 mW interference power threshold. Comparing the plots in Fig. 7 clearly shows that the precoding gain decreases due to SU receive antenna correlation. Fig. 8 depicts the average BER versus SNR for , for a 100 mW a 2 2 CR system with different interference power threshold. As expected, BER performance , at higher SNR improves signicantly with decreased regimes. VI. C ONCLUSION This paper presents a design of a linear precoder for orthogonal space-time block coded orthogonal frequency division multiplexing-based multiple-input multiple-output (MIMO) antenna cognitive radio (CR). The CR coexists with the primary user (PU) network by opportunistically sharing the originally allocated PU spectrum in correlated Rayleigh fading channels. The optimum linear precoder with respect to error probability performance is obtained by exploiting the partial channel information in the form of transmit and receive

BER

is the Euclidian distance between two codeword matrices. Due to the orthogonality of the codeword error product matrix, ( ), OSTBC has the appealing property ( ) C E( ) = C that E( )E( ) = ( )I , = 1, . . . , , where ( ) is the codeword distance that depends on the two code ( ). Therefore, exploiting these properties ( ) and C words C and the standard relations, {AB} = vec(A ) vec(B) and

776

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 3, MARCH 2011

vec(ABC) = (C A)vec(B) for any arbitrary matrices A, B, and C [25], the conditional PEP of OSTBC can be written as (C C H (1), . . . , H ( )) 1 exp 2

=1 () 2 4

vec(H () ) (I F()F() )vec(H () )

(30)

An upper bound on the average PEP of OSTBC can then be obtained by taking the expectation of (30) with respect to H ( ), = 1, . . . , , as (C C ) (1) h(1) (I F(1)F(1) ) h(1) 2 (1))h (1) . . . exp 4 (h ( ) h( ) (I F( )F( ) ) h( ) 2 ( ))h ( ), (h exp 4 (31) ( ) = vec(H( ) ), = 1, . . . , . Note that the where h probability distribution function (pdf) of the MIMO channel for the th subcarrier is a complex Gaussian and can be 1 () 1 h ()R h() ( )) = , written as (h ()] exp det[R where R ( ) = R, ( ) R, ( ), = 1, . . . , , is the ( ). Therefore, an upper bound on the covariance matrix of h average PEP can be rewritten as in (32). The integrals in (32) can be easily by making use solved 1 h() () h() of the fact that det[ exp h( ), 1 ()] 1 () where ( ) = 42 ((I F( )F( ) ) + R ( )), = 1, . . . , , is the integral of a complex Gaussian pdf and thus equals one. Therefore, an upper bound on the average PEP of the OSTBC MIMO-OFDM can be written as in (33). It can be noticed from (33) that the dependence on the codeword pair is now only through the codeword distance ( )s. Furthermore, we can notice that the (33) is a decreasing function of ( )s and PEP is dominated by the codeword pairs corresponding to minimum ( ). Consequently, only one such pair is considered in the linear precoder optimization procedure, i.e., = min (),= { ( )I = ( ) C ( )(C ( ) C ( )) }, = 1, . . . , . The (C depends on several factors, such as the modulation format of the input signal constellation, the variance of the input signal constellation and the OSTBC generator matrix [4]. Therefore, using this property of OSTBC, the equality (A B)(C D) = (AC) (BD), and the eigendecomposition of R, ( ) = U, ( ), ( )U, ( ) ,

R, ( ) = U, ( ), ( )U, ( ) , = 1, . . . , , an upper bound on the average PEP in (33) can be rewritten as (34), where = 2 ; F( ) = F( )F( ) ; and U, ( ), 4 U, ( ) are the matrices of the eigenvectors; and , ( ), , ( ) are the diagonal eigenvalue matrices of the SUs transmit and receive correlation matrices, respectively. Using the properties of Kroneker product, one can easily write [ ] ( ), ( ) det [( )] = det I + , ( ) F ( ), ( ). as in (35). In equation (35), ( ) = F [ ] A 0 Furthermore, by using the property that det = C D det [A] det [D], an upper bound on the PEP can be further written as in (9). B. Proof of Theorem 2 The convexity of the linear precoder design problem P1 given in (12)-(15), for OSTB MIMO-OFDM based CR is proven here. First, we dene the function (( )), for subcarrier as (( )) = log det [( )], = 1, . . . , , ( ), ( ). One can where ( ) = I + , ( ) F prove the convexity of (12) by showing that (( )) is convex over the set of positive denite matrices. Thus, the convexity of (( )) is proven using the following theorem [26]. Theorem 3 [26]: If 1 is Hermitian and 2 is positive denite, then there exists a nonsingular matrix A such that A 2 A = I, and A 1 A = D, where D is a diagonal matrix with all diagonal elements, > 0. The function (( )) is convex, if (1 ( ) + (1 )2 ( )) (1 ( )) + (1 ) (2 ( )) for any two positive denite Hermitian matrices in and for any 0 1. Equality holds when 1 ( ) = 2 ( ). Using the Theorem 3 and the properties of the logarithm, one can show that (( )) is convex, if (I + (1 )D( )) (1 ) (D( )) for all 0 1 and for any diagonal matrix D( ) with positive diagonal entities. This can be easily proven by using the properties of the determinant and the strict concavity of the logarithmic function itself as log ( + (1 )( ) ) (I + (1 )D( )) =
=1

(1 ) log

=1

) ( ) (36)

(1 ) (D( )).

1 (C C ) ( )] det[R 1 ( )] det[R
{ =1

exp exp

h(1)

) (( ) (1) 1 F(1)F(1) +R I 4 (1) h(1) 2

(1) . . . h (32) ( ) h

h( )

) (( ) ( ) 1 F( )F( ) +R I ( ) h( ) 42

(C C )

]}1 [ ) ( ) ( F ( ) F ( ) ( ) R ( )) det I + I ( R , , 2 4

(33)

PUNCHIHEWA et al.: LINEAR PRECODING FOR ORTHOGONAL SPACE-TIME BLOCK CODED MIMO-OFDM COGNITIVE RADIO { ]}1 [ 1/2 ( )1/2 ( ) det I + , ( ) , ( ) F , =1

777

(C C )

(34)

I + , ( )1 ( ) . .. . det [( )] = det . . 0

0 . . . I + , ( ) ( )

(35)

Thus, (( )) is convex over the set . Furthermore, using the property that the sum of convex functions is also a convex function [20], one can easily show that (12) is convex. In addition, it can be easily veried that the constraints (13)-(15) are convex [20]. Therefore the entire optimization problem P1 is convex. C. Proof of Theorem 4 The proof of the Theorem 4 is presented here. First, the non-negative Lagrange multipliers = [1 , . . . , ] and = [1 , . . . , ] associated with the average interference power constraints and per subcarrier power constraints in (17) and (18) are introduced and the Lagrangian of the primal problem is written as [20], [22] ) ( {F ( )}, , =
=1 =1 =1

obtained by rst minimizing the Lagrangian ({F ( )}, , ) in order to obtain the dual function ( , ) for some given dual variables , , and then maximizing ( , ) over all non-negative values of and . Therefore, consider the minimization of ({F ( )}, , ) with respect to the variables {F ( )} and for some given xed values of , to obtain the dual function ( , ). One can notice from (37) and (38) that the dual function ( , ) has the following form: ( , ) =
=1

( , )

=1

P ( )

=1

I, , (39)

( ) log 1 + , ( ) , ( ) F ( ) ) P ( ) )

+ +

F ( ) =1 =1 ( ( (
=1 =1 =1

F ( ) u, ( ) R, ( )u, ( )

)) I, , (37)

( , ) is given as in (40). where Therefore, it is interesting to note that the dual function ( , ) can be obtained by solving independent ( , ), each for OFDM subcarrier subproblems (P3)s, = 1, . . . , . This implies that the same computational routine can be repeatedly applied for solving each subproblem P3. Thus, the convergence time of the overall algorithm can be dramatically improved. Now it is required to nd the optimal solution for a subproblem P3. Since the subproblem P3 is convex, the globally optimal solution can be found by solving the system of Karush-Kuhn-Tucker (KKT) conditions [20]. Therefore, the Lagrangian of the subproblem P3 can be formulated as ( ) , ( ) , ( )
F

= +

=1 =1

log(1 + , ( ) , ( ) F ( ) )
F ( ) u, ( ) R, ( )u, ( ) =1

where {F ( )} = {F ( ) , = 1, . . . , , = 1, . . . , } is the set of eigenvalues of the eigenvalue matrices F ( ), = 1 . . . , . The Lagrange dual function is then dened as [20], [22] ( , ) = minimize
{ F ()}

( (
=1

) )

( ) {F ( )}, , , = 1, . . . , .

(38)

=1 =1

F ( )

( ) F ( ) , (41)

subject to

F ( ) 0, = 1, . . . , ,

Therefore, the Lagrange dual problem of the primal problem is dened as maximize 0, 0 ( , ) [20], [22]. The optimal value of the dual problem is achieved by the optimal dual variables and . Furthermore, the dual function ( , ) provides a lower bound of the optimal value, , of the primal problem. Since the original problem P1 is indeed convex and also satises Slaters condition, the duality gap ( ) is zero [20], [22]. Therefore, these interesting results suggest that the optimal solution, , for the primal problem can be

where ( ) are the Lagrange multipliers associated with the inequality constraints in (40). Then, the KKT conditions for the subproblem P3 can be written as in (42)-(44). In equations (42)-(44), ( ) represents the optimal solution. Therefore, F + + = the optimal solution ( ) , = 1, . . . , , with F max{0, } for the th subcarrier can be obtained by solving the set of equations as given in (20). Once the dual function ( , ), is obtained for given and , the next step of the Lagrange dual-decomposition is to maximize the dual function ( , ), over all possible values of and . Searching for the optimal value

778

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 3, MARCH 2011

( , ) =

minimize
F () , =1,...,

=1 =1

( ) log 1 + , ( ) , ( ) F F ( ) + ( )

+ subject to

=1

( (
=1

F ( ) u, ( ) R, ( )u, ( )

=1

, (40)

F ( ) 0, = 1, . . . ,

( ) , ( ) , ( ) F F ( )

= +

=1

, ( ) , ( ) ( ) + ( ) 1 + , ( ) , ( ) ( ) F (42) (43) (44)

=1

( ) u, ( ) R ( ) u ( ) , = 0, = 1, . . . , , ,

( ) ( ) F ( )

= 0, = 1, . . . , , 0, = 1, . . . ,

of ( ) and ( ) can be done by using, for example, the ellipsoid method [23], which exploits the fact ( ( ) u, ( ) that R, ( )u, ( ) ) I, and =1 F ( ) P ( ) are sub-gradients of , = 1, . . . , =1 F and , = 1, . . . , , respectively. ACKNOWLEDGMENT This research is supported by the Natural Science and Engineering Research Council of Canada (NSERC) under a strategic project grant. The authors like to acknowledge the valuable feedback given by Dr. Zouheir Rezki. R EFERENCES
[1] Federal Communication Commission, Spectrum Policy Task Force, ET Docket No. 02-155, Nov. 2002. [2] S. Haykin, Cognitive radio: brain-empowered wireless communications, IEEE J. Sel. Areas Commun., vol. 23, no. 2, pp. 201220, Feb. 2005. [3] E. Telatar, Capacity of multi-antenna Gaussian channels," AT&T Bell Labs Tech. Memo, Mar. 1995. [4] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, Space-time block coding for wireless communications: performance results, IEEE J. Sel. Areas Commun., vol. 17, no. 3, pp. 451460, Mar. 1999. [5] S. Alamouti, A simple transmit diversity technique for wireless communications, IEEE J. Sel. Areas Commun., vol. 16, no. 8, pp. 14511458, Oct. 1998. [6] T. Weiss and F. K. Jondral, Spectrum pooling: an innovative strategy for the enhancement of spectrum efciency, IEEE Commun. Mag., vol. 43, no. 3, pp. S8S14, Mar. 2004. [7] D. Shiu, J. G. Foschini, M. Gans, and J. M. Kahn, Fading correlation and its effect on the capacity of multi-element antenna systems, IEEE Trans. Commun., vol. 48, no. 3, pp. 201220, Mar. 2000. [8] E. Bjornson, B. Ottersten, and E. Jorswieck, On the impact of spatial correlation and precoder design on the performance of MIMO systems with space-time coding, in Proc. ICASSP, Apr. 2009. [9] H. Sampath and A. Paultry, Linear precoding for space-time coded systems with known fading correlations, IEEE Commun. Lett., vol. 6, no. 6, pp. 239241, June 2002. [10] H. R. Bahrami and L. N. Tho, Precoder design based on correlation matrices for MIMO system, IEEE Trans. Wireless Commun., vol. 5, no. 12, pp. 35793587, Dec. 2006. [11] A. Hjorungnes and D. Gesbert, Precoding of orthogonal space-time block codes in arbitrarily correlated MIMO channels: iterative and closed-form solutions, IEEE Trans. Wireless Commun., vol. 6, no. 3, pp. 10721082, Mar. 2007.

[12] A. Hjorungnes, D. Gesbert, and J. Akthar, Precoding of space-time block coded signals for joint transmit-receive correlated MIMO channels, IEEE Trans. Wireless Commun., vol. 5, no. 3, pp. 492497, Mar. 2006. [13] M. H. Islam and Y. C. Liang, Space-time coding in MIMO cognitive networks with known channel correlations, in Proc. IEEE EuWiT, Oct. 2008, pp. 97102. [14] A. Paultry, R. Nabar, and D. Gore, Introduction to Space-Time Wireless Communications, 2nd edition. Cambridge University Press, 2003. [15] K. Yu, M. Bengtsson, B. Ottersten, D. McNamara, P. Karlsson, and M. Beach, Modeling of wide-band MIMO radio channels based on NLoS indoor measurements, IEEE Trans. Veh. Technol., vol. 53, no. 3, pp. 655665, May 2004. [16] R. Stridh, K. Yu, B. Ottersten, and P. Karlsson, MIMO channel capacity and modeling issues on a measured indoor radio channel at 5.8 GHz, IEEE Trans. Wireless Commun., vol. 4, no. 3, pp. 895903, May 2005. [17] X. Hong, C. X. Wang, H. H. Chen, and J. Thompson, Performance analysis of cognitive radio networks with average interference power constraints, in Proc. IEEE ICC, May 2008, pp. 35783582. [18] G. Bansal, M. J. Hossain, and V. K. Bhargava, Optimal and suboptimal power allocation schemes for OFDM-based cognitive radio systems, IEEE Trans. Wireless Commun., vol. 7, no. 11, pp. 47104718, Nov. 2008. [19] W. Wei, P. Tao, and W. Wenbo, Optimal power control under interference temperature constraint in cognitive radio network, in Proc. IEEE WCNC, 2007, pp. 116120. [20] S. Boyd and L. Vandenberg, Convex Optimization. Cambridge University Press, 2004. [21] M. Codreanu, M. Juntti, and M. Latva-Aho, On the dualdecomposition-based sum capacity maximization for vector broadcast channels, IEEE Trans. Veh. Technol., vol. 56, no. 6, pp. 35773581, Nov. 2007. [22] D. P. Palomar and M. Chiang, A tutorial on decomposition methods for network utility maximization, IEEE J. Sel. Areas Commun., vol. 24, no. 8, pp. 14391451, Aug. 2006. [23] R. G. Bland, D. Goldfarb, and M. J. Todd, The ellipsoid method: a survey, Operations Research, vol. 29, no. 6, pp. 10391091, 1981. [24] J. G. Proakis, Digital Communications, 4th edition. McGraw-Hill, 2000. [25] A. Graham, Kronecker Products and Matrix Calculus With Applications. Ellis Horwood Ltd., 1981. [26] R. A. Horn and C. A. Johnson, Matrix Analysis, 1st edition. Cambridge University Press, 1985.

PUNCHIHEWA et al.: LINEAR PRECODING FOR ORTHOGONAL SPACE-TIME BLOCK CODED MIMO-OFDM COGNITIVE RADIO

779

Anjana Punchihewa received the B.Sc. degree in electrical and electronics engineering from the University of Peradeniya, Peradeniya, Sri Lanka, in 2004 and the M.Eng. degree in electrical engineering from Memorial University of Newfoundland, Canada, in 2007. He is currently pursuing the Ph.D. degree with the Department of Electrical and Computer Engineering at the University of British Columbia, Canada. His research interests lie in the areas of wireless communications and signal processing, with particular emphasis on space-time coding, cognitive radio, cooperative network, blind parameter estimation, and modulation classication. Vijay Bhargava an IEEE volunteer for three decades, is a professor in the Department of Electrical and Computer Engineering at the University of British Columbia in Vancouver, where he served as department head from 2003-2008. Previously, he was with the University of Victoria (1984-2003) and Concordia University (1976-84). He received his Ph.D. from Queens University. As a distinguished speaker for the IEEE Communications Society and the IEEE Information Theory Society, and as a senior level IEEE volunteer, he has lectured in 66 countries and assisted IEEE Presidents in negotiating sister society agreements in India, Japan, and Russia. He has rudimentary knowledge of several languages and is an eager student of different cultures and societies. Vijay served as the founder and president of Binary Communications Inc. (1983-2000). He has provided consulting services to several companies and government agencies. He is a co-author (with D. Haccoun, R. Matyas, and P. Nuspl) of Digital Communications by Satellite (New York: Wiley: 1981), which was translated into Chinese and Japanese. He is a co-editor (with S. Wicker) of Reed Solomon Codes and their Applications (IEEE Press: 1994); a co-editor (with V. Poor, V. Tarokh, and S. Yoon) of Communications, Information and Network Security (Kluwer: 2003); a co-editor (with E. Hossain) of Cognitive Wireless Communication Networks (Springer: 2007); and a co-editor (with E. Hossain and D. I. Kim) of Cooperative Wireless Communications Networks, a forthcoming Cambridge University Press Publication. Vijay has served on the Board of Governors of the IEEE Information Theory Society and the IEEE Communications Society. He has held important positions in these societies and has organized conferences such as ISIT95, ICC99, and VTC 2002 Fall. He has served as an editor of the IEEE T RANSACTIONS ON C OMMUNICATIONS. He played a major role in the creation of the IEEE Wireless Communications and Networking Conference (WCNC) and IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS, for which he served as editor-in-chief during 2007, 2008, and 2009. In 2010, he was appointed for a two year term as the IEEE Communications Society Director of Journals. He is a past President of the IEEE Information Theory Society. Vijay Bhargava has been elected to serve as IEEE Communications Society President-Elect during 2011 and as President during 2012 and 2013.

Charles Despins received a bachelors degree in electrical engineering from McGill University, Montral, QC, Canada, in 1984, and masters and Ph.D. degrees from Carleton University, Ottawa, ON, Canada, in 1987 and 1991, respectively. He was, from 1992 to 1996, a Faculty Member of the Institut National de la Recherche Scientique (INRS), Universit du Qubec, Montral, Canada, following employment in 1984-1985 with CAE Electronics as a Member of the Technical Staff, and in 1991-1992 with the Department of Electrical and Computer Engineering, cole Polytechnique de Montral, Canada, as a Lecturer and a Research Engineer. From 1996 to 1998, he was with Microcell Telecommunications Inc., a Canadian GSM operator, and was responsible for industry standard and operator working groups, as well as for technology trials and technical support for joint venture deployments in China and India. From 1998 to 2003, he was Vice-President & Chief Technology Ofcer of Bell Nordiq Group Inc., a wireless and wireline network operator in northern and rural areas of Canada. Since 2003, he has been President and CEO of Prompt Inc., a university-industry research consortium in the eld of information and communications technologies. He remains a faculty member (on leave) at cole de Technologie Suprieure (Universit du Qubec), Montral, QC, Canada, with research interests in wireless communications. He is also a guest lecturer in the Master of Business Administration (MBA) program at McGill University, Montreal, QC, Canada. Dr. Despins was awarded the IEEE Vehicular Technology Society Best Paper of the Year prize in 1993, as well as the Outstanding Engineer Award in 2006 from IEEE Canada. He is a Member of the Order of Engineers of Qubec and is also a Fellow (2005) of the Engineering Institute of Canada.