IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO.

3, MARCH 2011

901

Interference Cancellation and Detection for More than Two Users

Feng Li, Student Member, IEEE, and Hamid Jafarkhani, Fellow, IEEE
AbstractIn multiple access channels, when users know each others channels, precoders can be designed utilizing channel information to cancel the interference at the receiver without sacricing the diversity or the complexity of the system. Recently, it was shown that when there are only two users, a receiver can completely cancel the interference of the two users and provide full diversity for each user. Unfortunately, the scheme only works for two users. In this paper, we extend the scheme to more than 2 users. In other words, we propose a system to achieve interference cancellation and full diversity with low complexity for any number of users. Then, we extend the results to any number of users with any number of transmit and receive antennas. Our main idea is to design precoders, using the channel information, to make it possible for different users to transmit over orthogonal directions. Then, using the orthogonality of the transmitted signals, the receiver can separate them and decode the signals independently. We also analytically prove that our system provides full diversity to each user. Simulation results conrm our analytical proof and show that our proposed scheme outperforms other existing interference cancellation schemes. Index TermsMulti-user detection, multiple antennas, interference cancellation, precoder, orthogonal designs.

be increased signicantly using limited feedback[8][14]. In addition, the performance of each user in multi-antenna multiuser systems can also be enhanced using limited feedback. For example, in [15], post-processing SNR is maximized for a given linear receiver by selecting the QOSTBC with the minimum quaternionic angle as well as realizing interference cancellation. In [16], limited feedback is utilized to adapt the phase of a transmitted signal and improve the performance of the system. The common goal and the main characteristics of the above multi-user systems are the small number of required receive antennas and the low complexity of the array-processing decoding. However, as mentioned before, full diversity for each user is only achieved using maximum-likelihood detection. On the other hand, maximum-likelihood detection is usually not practical, because its complexity increases exponentially as a function of the number of antennas, the number of users, and the bandwidth efciency (measured in bits per channel use). To overcome this drawback, in [17], channel information is utilized at the transmitters to increase the diversity of the system while keeping the low complexity of the decoding. In other words, unlike the above-mentioned methods, we do not use receive antennas to cancel the interference. Instead, we use the channel information at the transmitter to design precoders that align different groups of signals along orthogonal directions. As a result, interference suppression is achieved without utilizing the receive antenna resources and therefore full diversity is achieved naturally. However, the scheme only works for two users. In this paper, we extend the work in [17] and propose a scheme that works for any number of users. The outline of the paper follows next. In Section II, we introduce our scheme for 4 users each with 4 transmit antennas. Our scheme can simultaneously achieve interference cancellation and full diversity for each user. We provide the details of decoding in Section III and the proof of full diversity in Section IV. We show that our scheme can be extended to any number of users each with any number of transmit antennas and any number of receive antennas in Section V. Section VI provides simulation results to validate our theoretical analysis and Section VII concludes the paper. Notation: We use boldface letters to denote matrices and vectors, super-scripts () , () , () to denote transpose, conjugate and transpose conjugate, respectively. We denote the element in the th row and the th column of matrix X by (, ). Also, we denote the th column of a matrix X by x(). The real and imaginary parts of a scalar are denoted by and , respectively.

ECENTLY, a lot of attention has been given to multi-user detection schemes with simple receiver structures. Multiple transmit and receive antennas have been used to increase rate and improve the reliability of wireless systems. In this paper, we consider a multiple-antenna multi-access scenario where interference cancellation is achieved by utilizing channel information. When there is no channel information at the transmitter, simple array processing methods using orthogonal space-time block codes (OSTBCs) [1] and quasi-orthogonal space-time block codes (QOSTBCs) [2] have been proposed in [3][7]. For users each with transmit antennas, it has been shown that in order to achieve interference cancellation, a receiver does not need more than receive antennas. In addition, the diversity of each user is equal to using maximum-likelihood detection and ( + 1) using lowcomplexity array-processing schemes[7]. Even a few bits of feedback at the transmitter can be used to improve the system performance. For example, the capacity and performance of the point-to-point MIMO systems can
Paper approved by N. Jindal, the Editor for MIMO Techniques of the IEEE Communications Society. Manuscript received June 2, 2009; revised July 6, 2010. This work was supported in part by ARO under the Multi-University Research Initiative (MURI) grant number W911NF-04-1-0224. The authors are with the Center for Pervasive Communications and Computing, University of California, Irvine (e-mail: {fengl, hamidj}@uci.edu). Part of this paper has been presented at the 2010 IEEE International Conference on Communications (ICC). Digital Object Identier 10.1109/TCOMM.2010.011811.090307

I. I NTRODUCTION

c 0090-6778/11$25.00 2011 IEEE

902

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

1 1

where H = H A and y = 2 denotes the received

1 3

C
2 2

signals of the four receive antennas at time slot . denotes

the transmit energy of each user. n = 2 denotes the
1 3

4 4

noise at the receiver at time slot . We assume that 1 , 1 , , are i.i.d complex Gaussian noises with mean 0 and 2 3 4 variance 1. Applying some simple algebra to Equation (4), we have
y =

( ( 1 ) ( 1 ) ( 1 ) ( 1 )) 2 2 2 + n H1 3 + H2 3 + H3 3 + H4 2 3
4 4 4 4

(5)

Fig. 1.

Block diagram of the system.

where

II. I NTERFERENCE C ANCELLATION FOR F OUR U SERS EACH WITH F OUR T RANSMIT A NTENNAS In this paper, we assume a quasi-static at Rayleigh fading channel model. The path gains are independent complex Gaussian random variables and are xed during the transmission of one block. In addition, a short-term power constraint is assumed. For the sake of simplicity, we only present the scheme for four users each with four transmit antennas and one receiver with four receive antennas. By adjusting the dimensions of channel matrices, our proposed scheme can be easily applied to users with transmit antennas and one receiver with receive antennas. The block diagram of the system is shown in Figure 1. We assume the channel matrices for Users 1, 2, 3, 4 are H1 = [1 (, )]44 , H3 = [3 (, )]44 , H2 = [2 (, )]44 H4 = [4 (, )]44 (1) H =

y =

y1 (y2 ) 3 y (y4 )

1 (1,1) 1 (2,1) 1 (3,1) (4,1) 1 2 ( (1,2)) (2 (2,2)) (2 (3,2)) (2 (4,2)) 3 (1,3) 3 (2,3) 3 (3,3) (4,3) 3 4 ( (1,4)) (4 (2,4)) (4 (3,4)) (4 (4,4))

1 (1,2) 1 (2,2) 1 (3,2) 1 (4,2) (2 (1,1)) (2 (2,1)) (2 (3,1)) (2 (4,1)) 3 (1,4) 3 (2,4) 3 (3,4) 3 (4,4) (4 (1,3)) (4 (2,3)) (4 (3,3)) (4 (4,3))

1 (1,3) 1 (2,3) 1 (3,3) 1 (4,3) (2 (1,4)) (2 (2,4)) (2 (3,4)) (2 (4,4)) 3 (1,1) 3 (2,1) 3 (3,1) 3 (4,1) (4 (1,2)) 4 ( (2,2)) (4 (3,2)) (4 (4,2))

n =

n1 (n2 ) n3 (n4 )

1 (1,4) 1 (2,4) 1 (3,4) 1 (4,4) (2 (1,3)) (2 (2,3)) (2 (3,3)) (2 (4,3)) 3 (1,2) 3 (2,2) 3 (3,2) 3 (4,2) (4 (1,1)) (4 (2,1)) (4 (3,1)) (4 (4,1))

(6)

respectively. At the th time slot, = 1, 2, 3, 4, the precoders for Users 1, 2, 3, 4 are A = [ (, )]44 , 3 3 A = [ (, )]44 , 1 1 A = [ (, )]44 2 2 A = [ (, )]44 4 4 (2)

respectively. In every four time slots, Users 1, 2, 3, 4 send Quasi Orthogonal Space-Time Block Codes (QOSTBCs) [2]
C= 2

T= respectively.

1 2 1 3 4 4 3 2 1 4 3

3 4 1 2

4 3 2 1

S= 2

1 2 3 4

3 4 1 2

4 3 2 1

, Z=

1 2 1 3 4 4 3
2 1 4 3

3 4 1 2

4 3 2 1

(3)

1 2 3 4

3 4 1 2

4 3 2 1

At time slot , = 1, 2, 3, 4, we have the following inputoutput equation y = (H1 A c()+H2 A s()+H3 A t()+H4 A z())+n 1 2 3 4 = (H1 c() + H2 s() + H3 t() + H4 z()) + n (4)

Now we choose precoders that can realize full diversity and interference cancellation for each user. First, we illustrate our main idea. To realize interference cancellation, a straightforward idea is to transmit the symbols of the four users along four orthogonal directions. By doing so, it is easy to achieve interference cancellation at the receiver using zero-forcing. However, the difculty lies in how to achieve full diversity as well. In [17], a scheme based on Alamouti structure has been proposed to achieve interference cancellation and full diversity for two users. When we have four users, the method does not work because four-dimensional rate-one complex orthogonal designs do not exist. An alternative is to use the quasiorthogonal structure, but it cannot achieve full interference cancellation for each user due to its non-orthogonality. To tackle all the above problems together, we propose a new precoder design scheme as follows. At each of the rst 2 time slots, we design precoders such that symbols of User 1 and symbols of User 2 are transmitted along two orthogonal directions, respectively, as illustrated in Figure 2. In addition, because of the characteristic of our designed precoders, each element of the equivalent channel matrices for Users 1 and 2 is still Gaussian. This property is critical to achieve full diversity for Users 1 and 2 as we will show later. Then we design precoders for Users 3 and 4, such that the transmit directions of their signals are orthogonal to each other. Note that it is impossible to obtain this orthogonal

LI et al.: INTERFERENCE CANCELLATION AND DETECTION FOR MORE THAN TWO USERS

903

structure and make each element of the equivalent channel matrices for Users 3 and 4 still Gaussian. This is the main difference between the precoders for Users 1, 2 and the precoders for Users 3, 4, at the rst 2 time slots. At the second 2 time slots, we also design precoders to make the transmit directions of signals orthogonal to each other. However, we design the precoders for Users 3 and 4 rst, such that each element of the equivalent channel matrices for Users 3 and 4 is Gaussian. Then we design the precoders for Users 1 and 2 to obtain the orthogonal structure. As a result, elements of the equivalent channel matrices for Users 1 and 2 will not be Gaussian at the second 2 time slots. Later we will prove that by using such precoders, we can achieve interference cancellation and full diversity for each user. In what follows, we will describe the details of our precoder designs. At time slot 1, in order to have orthogonality between User 1 and User 2, we design the precoders such that 1 1
2 (1,1) 1 2 (2,1) 1 (3,1) 2 1 (4,1) 2

User 1

User 3

User 2
User 4

Fig. 2.

Orthogonal structure of signal vectors in 4-dimensional space.

1 (1,1)
1 (2,1) 2
2 2

1 = u() , 2 (3,1) 2 1 + 3 1 =1 (4,1)

1 (1, )
1 (2, ) 2 1 (3, ) 2 1 (4, ) 2
2

1 (1,1)
2 1

= 1 2 (2,1) 1
2 (3,1) 1 (4,1) 2

1 (2,1) 1 (1,1) 1 1 (4,1) 1 1 (3,1) 1

(7)

where 1 (, ) and 1 (, ) are elements of the equivalent 1 2 channel matrices in Equation (6). Equation (7) can be rewritten as 1 1 H2 2 (3,1) = H1 1 (3,1) 1 1
2 1 (4,1) 2 1 1 (4,1) 1

2 (1,1) 1 (2,1)

1 (1,1) 1 (2,1)

(8)

where = 2, 3, 4 and is the same as that in Equation (12). As we will discuss later, we choose parameters 1 , 2 , 3 to maximize the coding gain. The choice of 1 , 2 , 3 will complete the precoder design for Users 1 and 2 at time slot 1. Note that the designed precoders A1 , A1 satisfy A1 2 = 1 2 1 A1 2 = 1 and the signals of User 1 and User 2 will be 2 transmitted along two orthogonal directions as shown in Figure 2. In order to derive the orthogonality among Users 1, 2, 3 at time slot 1, we design precoder A1 to satisfy the following 3 properties: 1. Complex vectors
1 (1,1) 1 1 (2,1) H1 1 (3,1) , 1 1 1 (4,1) 1 1 (1,1) 2 1 (2,1) H2 2 (3,1) , 1 2 1 (4,1) 2

(14)

where H1 = Now let

( 1 (2,1) 1 (2,2) 1 (2,3) 1 (2,4) )

1 (1,1) 1 (1,2) 1 (1,3) 1 (1,4) 1 (4,1) 1 (4,2) 1 (4,3) 1 (4,4) 1 (3,1) 1 (3,2) 1 (3,3) 1 (3,4)

(9)

Q=

H1 H1 2

= UV

(10)

1 (1,1) 3 1 (2,1) H3 3 (3,1) 1 3 1 (4,1) 3

are orthogonal to each other. =

2.

where we have made the singular value decomposition. It has been proved in [17] that 1 1 1 (1,1) 2 (1,1) 1 1 1 (2,1) = v() , 2 (2,1) = u(), = 1 (11) 1 (3,1) 1 (3,1) 1 2 (, ) 1 1
1 (4,1) 2 (4,1)

1 (1,) 3 1 3 (2,) 1 (3,) 3 1 (4,) 3

1 (1,1) 3 1 (2,1) 1 3 (3,1) , 1 3 1 (4,1) 3

= 2, 3, 4

(15)

where = 1, 2, 3, 4, will satisfy Equation (8). There are four 1 1

(2,1) (2,1) different choices for 1 (3,1) and 2 (3,1) depending 1 1 1 (1,1)
1 1

2 (1,1)
1 2

on which we pick. Different choices of result in different performances. For given channel matrices H1 and H2 , at time slot 1, we let v = v() , {1, 2, 3, 4}, such that the norm of H1 v is the largest, i.e., v = arg
1 (1,1)
1 1 1

1 (4,1) 1

1 (4,1) 2

3. The Frobenius norm of complex matrix A1 is equal to 1. 3 In order to maximize the coding gain, A1 can be further 3 chosen numerically such that the norm of H3 A1 is maximized. 3 Similarly, for User 4, at time slot 1, in order to derive the orthogonality as shown in Figure 2, we choose precoder A1 4 to satisfy the following properties: 1. Complex vectors
1 (1,1) 1 1 (2,1) H1 1 (3,1) , 1 1 1 (4,1) 1 1 (1,1) 2 1 (2,1) H2 2 (3,1) , 1 2 1 (4,1) 2

v() ,=1,2,3,4

max

H1 v() 2

(12)
1 (1,1)
1 1

1 (1,1) 3 1 (2,1) H3 3 (3,1) , 1 3 1 (4,1) 3

Then for User 1, at time slot 1, we let

v 3 ,
1 1

2.

1 (1,1) 4 1 (2,1) H4 4 (3,1) 1 4 1 (4,1) 4

are orthogonal to each other.

1 (1, )

1 (2,1) = 1 (3,1)
1 (4,1) 1

1+

=1

1 (2, ) = 1 1 (2,1) 1 1
1 (3, ) 1 (4, ) 1 1 (3,1) 1 (4,1) 1

1 (1,) 4 1 4 (2,) 1 (3,) 4 1 (4,) 4

= 1

1 (1,1) 4 1 4 (2,1) , 1 (3,1) 4 1 (4,1) 4

= 2, 3, 4

(16)

where = 2, 3, 4. For User 2, at time slot 1, we let

(13)

3. The Frobenius norm of complex matrix A1 is equal to 1. 4 Similarly, in order to improve the coding gain, A1 can be 4

904

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

further chosen numerically such that the norm of H4 A1 is 4 maximized. By choosing A1 , A1 , A1 , A1 , the precoder design 1 2 3 4 at time slot 1 is complete. At time slot 2, the precoder design is similar to that at time slot 1. The difference is that we choose u = u(), {1, 2, 3, 4}, such that H2 u is the largest, i.e., u = arg Then we let
2 (1,1)
2 2 (2,1) 2 2 (3,1) 2 2 (4,1) 2

In order to decode symbols from User 1, we multiply both sides of Equation (20) by matrix H1 to achieve ( 1 ) H1 y = H1 H1 2 + H1 n (22) 3
4

u(),=1,2,3,4

max

H2 u()2

(17)
2 (1,1)
2 2

1+

u 3

2 (1, )
2

=1

2 2 (2, ) 2 2 (3, ) 2 2 (4, ) 2

= 1 2 (2,1) 2
2 (3,1) 2 (4,1) 2

Note that the noise elements of H1 n are correlated with covariance matrix H1 H1 . We can whiten this noise vector by multiplying both sides of Equation (22) by the matrix 1 (H1 H1 ) 2 as follows ( 1 ) 1 1 (H1 H1 ) 2 H1 y = (H1 H1 ) 2 2 + n (23) 3
4

where = 2, 3, 4. For User 1, at time slot 2, we choose

2 (1,1)
2 (2,1) 1
1

(18)

where = 2, 3, 4 and is the same with that in Equation (17). Design of A2 , A2 is similar to that of A1 , A1 . By switching 3 4 3 4 the terms related to Users 1 and 2 with those of Users 3 and 4, respectively, we can design the precoders at time slots 3 and 4. Till now, the precoder design for each user at the rst 4 time slots is complete. When there are users, at time slots 2 1 and 2, we rst design precoders for Users 2 1 and 2 similar to what we do for Users 1 and 2. Then we design precoders for other users such that all of them transmit along orthogonal directions. Therefore, the above idea for 4 users can be easily extended to any number of users. In the next two sections, we will illustrate how to decode and why our scheme can realize interference cancellation and full diversity for each user. III. D ECODING Using our precoders, Equation (5) becomes
y=
4 4 4

2 (1, ) 2 (1,1) 1 1 v() 2 (2, ) 2 (2,1) 2 = , 1 (3, ) = 1 1 (3,1) 2 2 1 (3,1) 1 1 2 1 + 3 =1 2 (4,1) 2 (4, ) 2 (4,1) 1 1 1 (19)

where = (H1 H1 ) 2 (H1 n) has uncorrelated elements n (0, 1). Equation (23) can be further rewritten as 1 ( 1 2 3 4 ) 2 ( 1 ) 2 5 6 7 2 1 + (24) n (H1 H1 ) 2 H1 y = 3 6 8 9 3
4 7 9 10 4

where
2 2 2 1 = + 1 + 2 + 3 ,

2 = 1 1 + 2 3 2 3 6 = 1 2 3 + 3 1 2
2 2 2 8 = 2 +3 ++1 2 2 2 10 = 3 +2 +1 + (25)

3 = 2 +1 3 +2 +1 3 , 4 = 3 1 2 +1 2 3
2 2 2 5 = 1 + + 3 + 2 ,

7 = 1 3 +2 +1 3 +2 , 9 = 2 3 2 3 +1 1 ,
4 =1

= = Now let

1 (, 1)2 , 1 3 (, 1)2 , 1 ( 1

= =

4 =1 4 =1

2 (, 1)2 1 4 (, 1)2 1 (26)

4 =1

H=

2 3 4 2 5 6 7 3 6 8 9 4 7 9 10

)1 2

(27)

( ( 1 ) ( 1 ) ( 1 ) ( 1 )) H1 2 + H2 2 + H3 2 + H4 2 +n 3 3 3 3
4

(20)

where

H =

(21) Here y and n are the same with y and n in Equation (5). Note that using our precoders, each column of matrix H1 is orthogonal to each column of matrices H2 , H3 , H4 .

1 (1,1) 1 (2,1) 1 (3,1) (4,1) 1 1 (2 (1,1)) ( (2,1)) 1 2 1 (2 (3,1)) 1 (2 (4,1)) 2 3 (1,1) 2 3 (2,1) 2 3 (3,1) 2 (4,1) 3 3 (4 (1,1)) ( (2,1)) 4 3 3 (4 (3,1)) 3 (4 (4,1))

1 1 (1,1) 1 1 (2,1) 1 1 (3,1) 1 1 (4,1) 2 ( (1,1)) (2 (2,1)) (2 (3,1)) (2 (4,1)) 3 3 (1,1) 3 3 (2,1) 3 3 (3,1) 3 3 (4,1) 2 (4 (1,1)) 2 (4 (2,1)) 2 (4 (3,1)) 2 (4 (4,1))

2 1 (1,1) 2 1 (2,1) 2 1 (3,1) 2 1 (4,1) 2 3 ( (1,1)) 3 (2 (2,1)) 3 (2 (3,1)) 3 (2 (4,1)) 3 (1,1) 3 (2,1) 3 (3,1) 3 (4,1) 1 (4 (1,1)) 1 (4 (2,1)) 1 (4 (3,1)) 1 (4 (4,1))

3 1 (1,1) 3 1 (2,1) 3 1 (3,1) 3 1 (4,1) 2 (2 (1,1)) 2 (2 (2,1)) 2 (2 (3,1)) 2 (2 (4,1)) 1 3 (1,1) 3 1 (2,1) 1 3 (3,1) 1 3 (4,1) (4 (1,1)) 4 ( (2,1)) (4 (3,1)) (4 (4,1))

From Equation (24), we can see that User 1 transmits 4 different codewords along 4 different equivalent channel vectors in the 4 time slots. So the rate is 1. If 1 , 2 , 3 are all real, from (27), it is easy to see that the equivalent channel matrix H is real. So if QAM is used, Equation (24) is equivalent to the following two equations ( 1 ) 1 Real{H y} = H 2 + Real{} H n (28) 1 3
1 H Imag{H1 y} =

( 1 )
2 3 4

+ Imag{} n

(29)

Then we can use the Maximum-Likelihood method to detect the real and imaginary parts of these 4 codewords separately. For example, by Equation (28), we can detect 1 , . . . , 4 by
(
2 3 4
1

= arg

1 ,...,4

min

1 H Real{H1 y}

( 1 )
2 3 4

( 1 ) (30) Similarly, using Equation (29), we can detect 2 . Note 3 that the decoding complexity is pair-wise decoding. In order
4

LI et al.: INTERFERENCE CANCELLATION AND DETECTION FOR MORE THAN TWO USERS

905

to detect codewords of Users 2, 3, 4, we can multiply both sides of Equation (20) with matrix H2 , H3 , H4 , respectively, to remove the signals of other users and use a similar method to complete the decoding. IV. P ROOF OF F ULL D IVERSITY In this section, we prove that we can achieve diversity 16, i.e., full diversity, using our proposed precoding scheme. We only present the proof for User 1, since the proof for Users 2, 3, 4 is the same. Diversity is dened as = lim

So we have (d d) = [ (d dH)] [ ( )] 4 4 =1 =1 1 (, )2 exp 16 = 16 where = 1 + ( /16)]

=1 [1

(37)

log log
1

(31)

1 + 1 2 + 2 3 + 3 4 2 3 2 1 + =1

(38)

User 1. Here we add a unitary rotation R to 2 . Thus, the 3 4 ( 1 ) 2 data vector is d = R 3 and we dene the error vector 4 ( 1 ) ( 1 ) = 2 2 . By (24), the pairwise error probability 3 3 (PEP) can be derived using the Gaussian tail function as[18] 2 HR (d dH) = 4 R H2 R = 4 ) ( R H2 R (32) exp 4 ( 1 ) Now we assume R = 2 . Substituting R and H in 3 Equation (27) into (32), we have
4 4 4

where denotes the SNR and represents the probability of error. We rst consider (24) to analyze ( ) the diversity for

At high SNRs, one can neglect the one in the denominator and get ( )16 (39) (d d) 16 Then, it is easy to show that the diversity is 16 if we simply choose KR such that = KR = 0 where K =
1 2 1+ 3 =1

(40)

( 1 1

2 3

) is a normalized vector.

( ) exp (d dH) 4 = 1 + 1 2 + 2 3 + 3 4 2 +3 1 2 + 1 3 2 4
2

(33)

where

Therefore, by using our scheme, User 1 can achieve full diversity. In addition, in order to maximize the coding gain, we need to choose KR such that the minimum possible norm of KR is maximized. For QAM, it is not hard to do so. For example, when QPSK is adopted, we can simply 1 choose KR = 85 ( 1 2 4 8 ). It is easy to check that the minimum possible norm of KR is maximized. Similarly, we can also prove that the diversity for Users 2, 3, 4, is 16 as well. Therefore, our scheme can achieve full diversity for each user. When we use Equations (28), (29) to simplify the decoding complexity, similar techniques can be used to complete the proof of full diversity. Note that our precoding design procedure itself does not rely on the channel statistics. So using our scheme, the pairwise error probability can always be upper bounded by [ ( )] H1 2 (41) (d d) exp 16 like Equation (37), where H1 is the channel matrix for User 1. This means that the proposed procedure is universal in that it can achieve the maximum possible diversity over any fading distribution. V. E XTENSION TO U SERS WITH T RANSMIT A NTENNAS AND O NE R ECEIVER WITH R ECEIVE A NTENNAS In this section, we show that the presented scheme can be extended to a general case of users each with transmit antennas and one receiver with receive antennas. For the simplication of presentation, we discuss 3 cases where among parameters , and , two are the same and the third one is larger than the other two. It is easy to extend the results to a general case. In addition, we just show our schemes when , and are all even. By some simple antenna or user removals, our proposed scheme can also be used when not all of , and are even.

+2 1 + 3 2 + 3 + 1 4 2 +1 1 2 2 + 3 3 4 2 Further, we have

(34)

( ) 2 ) exp 1 + 1 2 + 2 3 + 3 4 (d dH 4 ) ( 4 =1 1 (, 1)2 1 + 1 2 + 2 3 + 3 4 2 1 = exp 4 (35) Note that

4 =1 4 4

1 (, 1)2 = H1 v 2 1

1 H1 V2 4 (36)

1 1 H1 2 = 1 (, )2 4 4 =1 =1

906

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

A. More Transmit Antennas, i.e., > = First, we consider the case > = . For simplication, we assume = 6, = = 4. Also we take User 1 for example. Each user transmits QOSTBCs and the precoder will be a 6 4 matrix. Similar to Equation (7), to make User 1 and User 2 orthogonal to each other, we design precoders such that 1 (1,1) 1 (1,1) 2 (2,1) 1 (2,1) 1 (3,1) 1 (3,1) 2 H2 1 (4,1) = H1 1 (4,1) 1 2 1 1 1
2 (5,1) 1 (6,1) 2 1 (5,1) 1 (6,1) 1
2 1 1 1

Then by Equation (45), we have

4 =1

1 (, 1)2 > 1
=1

1 H1 2 > H1 2 4 6 (48)

6 Substituting (48) in (46), we have (d dH)

( exp 4
=1

=1

1 (, )2

(42)

=1

1 (, )2 1 + 1 2 + 2 3 + 3 4 2 24

(49)

where H1 and H2 are all 4 6 matrices. Since H1 and H2 are not square matrices, we cannot take the inverse as in Equation (10). Instead, we multiply both sides of Equation (42) by H 2 and (H H2 )1 resulting in 2 Q = (H H2 )1 H H1 2 2 (43) Then we can calculate the singular value decomposition of Q and use the same method used in Section II to design the precoders. For the sake of brevity, we do not include the decoding and the proof of full diversity. They are similar in nature to what we presented earlier for users with 4 transmit antennas.

Using the same techniques presented in Section II, we have (d d) < 24 1

=1 [1 + ( /24)]

(50)

At high SNRs, one can neglect the one in the denominator and get ( )24 (51) (d d) 24 Then, it is easy to show that the diversity of User 1 is 24, i.e., full diversity. Similarly, we can prove that the diversity of any other user is also full diversity. Note that although we can achieve full diversity for each user, we only use = receive antennas at each time slot. In other words, we do not use all receive antennas. In what follows, we show that besides achieving full diversity, we can further increase the array gain by a simple iterative decoding method. ( ) ( ) We let
1 2 3 4 , 2 , 4
3

B. More Receive Antennas, i.e., > = For the case of > = , we can pick the = receive antennas with the best channel conditions among all receive antennas for User at time slot . In what follows, we illustrate our selection criterion and prove that it provides full diversity. For simplication, we assume = 6, = = 4. We assume the channel matrix for User , = 1, . . . , 4, is H = [ (, )]64 (44) Then we pick the 4 rows in H with the largest norms and put them in a matrix H . So we have 2 H 2 (45) 3 We just need to pick the 4 antennas corresponding to these 4 rows to nish the antenna selection. Then we can use the scheme proposed in Section II to design the precoder. Next we prove that each user achieves full diversity. We take User 1 for example and start with Equation (35), i.e., H 2 > (d dH)
( exp 4
1 2 =1 1 (, 1) 1 2

1 2 3 4

, denote the detected signals (

1 2 3 4

) ,

of Users 2, 3, 4, respectively. We subtract the term H2 ( ) H3 2 , H4

4
3

of Users 2, 3, 4 to have ( ) ( ) 1 1 1 y H2 2 H3 2 H4 2 3 3 ( 1 ) = H1 2 + n + 3
4 4 4
3

1 2 3 4

, from Equation (20) to remove the effect

(52)

where = 1 + 2 + 3 and (( ) ( )) 1 1 2 1 = H2 2 3 3
4

+ 1 2 + 2 3 + 3 4 4

) (46)

2 = H3 3 = H4

( 1 )
2 3 4

Note that
4 =1

(( H1 2 H1 V2 = 4 4 (47)

1 (, 1)2 = H1 v 2 > 1

1 2 3 4

1 2 3 4

4 )) (53)

1 2 3 4

where we have used Equation (12) and the fact that multiplying by a unitary matrix does not change the norm of a matrix.

denote the residual error. Then we can multiply both sides of Equation (52) by H1 and use the same method in Section III to detect the signals of User 1. In what follows, we rst show

LI et al.: INTERFERENCE CANCELLATION AND DETECTION FOR MORE THAN TWO USERS

907

that the method still provides full diversity to User 1. There are two factors that result in an error for User 1. The rst one is error in decoding symbols of User 1 after removing the effect of other users and the second one is the error in detecting the symbols of other users at the rst time, i.e., error propagation. Let Pr(d1 d1 ) denote the pairwise error probability for User 1, we separate these two events to have Pr(d1 d1 ) = Pr{d1 d1 = 0} Pr{ = 0} + Pr{d1 d1 = 0} Pr{ = 0} = Pr{d1 d1 = 0}(1 Pr{ = 0}) (54) + Pr{d1 d1 = 0} Pr{ = 0} Since Pr{d1 d1 = 0} 1 and 1 Pr{ = 0} 1, we have Pr(d1 d1 ) Pr{d1 d1 = 0}(1 Pr{ = 0}) + Pr{ = 0} Pr{d1 d1 = 0} + Pr{ = 0} (55) Note that when = 0, we can follow the steps in Section III to detect the signals of User 1 and by the same technique used in Section IV, we can easily derive ( )24 Pr{d1 d1 = 0} = 1 24 (56) 24 where 1 is a constant. Further, from (51), we know that
Pr{ = 0} Pr{1 = 0} + Pr{2 = 0} + Pr{3 = 0} 2 24 + 3 24 + 4 24 (57)

the following equivalent channel equation

c2 H c3 c4 c5 c6

c1 +n=

H1
1 2

y=

4 H1
0 0

H1 H2 H3 H4 0 0 H3 H3 0 0 H3 H3 5 6 1 2
H2 0 0 0
4 5 H3 6 H3 5 H4 6 H4

H2 H3 H4 0
2 2 2

0 H5
5 6

H5 H6 H5 H6
6

c1 c2 c3 + n c4 4 H6 c5 5 c
6

(61)

where the equivalent channel matrix H is a 24 24 matrix, noise vector n is a 241 vector and H denotes the equivalent channel matrix of User in time slot . Both H and 0 are 44 matrices. Note that with our proposed precoder design in Section II, each column in matrix H is orthogonal to all other columns. So we can use the method in Section III to decode the symbols of each user. We can also prove that we can achieve full diversity for each user by the same method used in Section IV. More specically, when we prove full diversity for User 1, similar to Equations (33), (34), we derive ) ( ( ) R H1 2 R = exp (d dH) exp 4 4 (62) where ) ( 1 2 3 4 H1 = H1 H1 H1 H1 0 0 (63) = + +
+
4 =1 4 =1 4 =1
4 =1

1 (, 1)2 1 + 1 2 + 2 3 + 3 4 + 4 5 + 5 6 2 1

where 2 , 3 , 4 are all constants. Substituting (56) and (57) in (55), we get Pr(d1 d1 ) (1 + 2 + 3 + 4 )24 (58)

2 (, 1)2 5 1 + 2 + 1 2 + 3 4 + 3 5 + 4 6 2 1 3 (, 1)2 4 1 + 5 2 + 3 + 1 4 + 2 5 + 3 6 2 1
4 (, 1)2 3 1 + 4 2 + 5 3 + 4 + 1 5 + 2 6 2 (64) 1

Using (58), it is easy to show that the diversity 24. Also, it is easy to show that the diversity 24. So the diversity for User 1 is still full diversity. Similarly, we can show that all the other users can also achieve full diversity. In addition, since all the receive antennas are used in the decoding efciently, it is obvious that the coding gain will be increased. C. More Users, i.e., > = In this section, we consider the case that > = . For simplication, we assume = 6, = = 4. First, we assume User transmits codewords 1 2 3 4 5 6 c = 3 4
2 3 4 5 5 6 6 1 4 5 6 1 2 5 6 1 2 3 6 1 2 3 4 1 2 3 4 5

Using the techniques in Section IV, we show ( )24 (d d) 24 where = 1 + 1 2 + 2 3 + 3 4 + 4 5 + 5 6 2 2 1 + 5 =1

(65)

(66)

(59)

It is easy to see that full diversity for User 1 is achieved. Similarly, we can prove that we can achieve full diversity for other users. VI. S IMULATION R ESULTS In this section, we provide simulation results that conrm our analysis in the previous sections. The performance of our proposed scheme is shown in Figures 3, 4, 5, 6 and 7. In Figure 3, we consider 4 users each equipped with 4 transmit antennas and a receiver with 4 receive antennas. We compare our scheme using QPSK and Equations (28), (29) with the scheme proposed in [6] for the same conguration without channel information at the transmitter. With 4 receive antennas, the

Channel matrix and precoder for User are given by H = [ (, )]44 and A = [ (, )]46 , respectively, where denotes the time slot. Note that we can only have four orthogonal directions at most since there are four receive antennas. In order to get the orthogonal structure, we let = A4 = A4 = A5 = A5 = A6 = A6 = 044 (60) 3 4 1 2 1 2 Then we can use the method proposed in Section II to complete the remaining precoder design. Finally, we can have A1 = A1 = A2 = A2 = A3 = A3 5 6 5 6 3 4

908

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

multi-user detection (MUD) method offered in [6] cancels the interference and provides a diversity of 4. Our scheme also cancels the interference completely but provides a diversity of 16 by utilizing the channel information at the transmitter. We also compare our scheme with two other schemes that can realize interference cancellation and full diversity as well. In both of these two schemes, we assume there are 4 users each with 4 transmit antennas and one receiver with 4 receive antennas. The rst scheme is to let only one user transmit using single beamforming at each time slot[19]. In order to have the same bit-rate, 256-QAM is used. The second scheme is to let the rst 2 users transmit using the scheme for 2 users in [17] at the rst 4 time slots and the second 2 users transmit at the second 4 time slots. Similarly, in order to have the same bit-rate, 16-QAM is used. The results, in Figure 3, show that our scheme outperforms both of these two schemes, by 13 dB and 8 dB, respectively. Finally, we compare our scheme with the following TDMA multiple beamforming scheme. At each time slot, only one user transmits using multiple beamforming, i.e., sends 4 symbols along the 4 eigenvectors of the channel matrix[19]. QPSK is used to match the rate. From the simulation, our scheme outperforms this scheme whose diversity is only 1. In addition, in Figure 4, we compare our scheme with all mentioned schemes when the rate is changed to adapt with the received SNR. In the simulation, the single beamforming scheme switches between 16-QAM and 256-QAM. The multiple beamforming scheme, the MUD scheme in [6] and our scheme all switch between BPSK and QPSK. The interference cancellation scheme in [17] switches between QPSK and 16-QAM. The threshold to switch between the two rates is properly chosen such that the two constellations are used with the same probability in each scheme. So the rate of all these schemes is 1.5 bits/sec/Hz. We have also provided a xed rate" set of simulation results. In all cases, for 1.5 bits/sec/Hz, what we mean by xed rate" is the average between the performance of two xed-rate systems using BPSK and QPSK. From Figure 4, we can see that adapting the rate can improve the performance compared with using a xed rate. Also we can see that even with variable rate, our scheme provides the best performance. Next, we present results for 4 users with 4, 6 and 8 transmit antennas and one receiver with 4 receive antennas in Figure 5. When the number of users and the number of receive antennas are xed, by increasing the number of transmit antennas from 4 to 8, we will have a higher diversity. As we have proved, the diversity is always full diversity using our proposed scheme in Section V. Further, we show the results for 4 users each with 4 transmit antennas and one receiver with 4, 6 and 8 receive antennas in Figure 6. By increasing the number of receive antennas from 4 to 8, the diversity increases from 16 to 32. Therefore, extra receive antennas will provide extra diversity and the resulting diversity of the system is always , i.e., full diversity, which conrms our theoretical analysis. Finally, we provide results for different number of users each with 4 transmit antennas and one receiver with 4 receive antennas in Figure 7. We can see that although we can still achieve full diversity for each user, the coding gain for each

10

MultiUser, 2 bits/sec/Hz
Single beamforming with 256QAM/TDMA Multiple beamforming with QPSK/TDMA Scheme in [17] with 16QAM/TDMA Scheme in [6] with QPSK Our scheme with QPSK

10

Signal to Noise Ratio(dB)

10

10

10

10

10

10

15

20

Bit Error Rate

25

30

35

40

Fig. 3. Simulation results for four users each with four transmit antennas and one receiver with four receive antennas.

10

MultiUser, 1.5 bits/sec/Hz

Single beamforming: fixed rate Single beamforming: variable rate Scheme in [17]: fixed rate Scheme in [17]: variable rate Scheme in [6]: fixed rate Scheme in [6]: variable rate Our Scheme: fixed rate Our scheme: variable rate Multibeamforming: fixed rate Multibeamforming: variable rate

10

Bit Error Rate

10

10

10

10

10

10

15

Signal to Noise Ratio(dB)

20

25

30

35

40

Fig. 4. Simulation results for four users each with four transmit antennas and one receiver with four receive antennas when the rate can be adapted.

user will be reduced signicantly. In order to keep the coding gain on a satisfactory level, the number of transmit antennas and the number of receive antennas should be larger than the number of users as shown in the previous gures. VII. C ONCLUSIONS We have considered interference cancellation for a system with more than two users when users know each others channels. We have proposed a system to achieve the maximum possible diversity of 16 with low complexity for 4 users each with 4 transmit antennas and one receiver with 4 receive antennas. Besides diversity, our proposed scheme also provides the best performance among all existing schemes with simple array processing decoding. Our main idea is to design precoders, using the channel information, to make it possible for different users to transmit over orthogonal directions. Then, using the orthogonality of the transmitted signals, the receiver can separate them and decode the signals independently. We have analytically proved that the system provides full diversity

LI et al.: INTERFERENCE CANCELLATION AND DETECTION FOR MORE THAN TWO USERS

909

10

MultiUser, 2 bits/sec/Hz 4 transmit antennas 6 transmit antennas 8 transmit antennas

10

MultiUser, 2 bits/sec/Hz 4 users 6 users 8 users

10 Bit Error Rate

10

Bit Error Rate

10

10

10

10

10

10

10

15 20 Signal to Noise Ratio (dB)

25

10

10

15

20 25 30 Signal to Noise Ratio (dB)

35

40

Fig. 5. Simulation results for four users each with different number of transmit antennas and one receiver with four receive antennas.

Fig. 7. Simulation results for different number of users each with four transmit antennas and one receiver with four receive antennas.

10

MultiUser, 2 bits/sec/Hz 4 receive antennas 6 receive antennas 8 receive antennas

10 Bit Error Rate

10

10

10

10

15 20 Signal to Noise Ratio (dB)

25

Fig. 6. Simulation results for four users each with four transmit antennas and one receiver with different number of receive antennas.

to each user and extended the results to any number of users each with any number of transmit antennas and one receiver with any number of receive antennas. Further work includes the extension of our scheme to the situation with only limited feedback. R EFERENCES
[1] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, Space-time block codes from orthogonal designs," IEEE Trans. Inf. Theory, vol. 45, pp. 1456-1467, July 1999. [2] H. Jafarkhani, A quasi-orthogonal space-time block code," IEEE Trans. Commun., vol 49, no. 1, pp. 1-4, Jan. 2001. [3] V. Tarokh, A. Naguib, N. Seshadri, and A. R. Calderbank, Combined array processing and space-time coding," IEEE Trans. Inf. Theory, vol. 45. pp. 1121-1128, May 1999. [4] A. F. Naguib, N. Seshadri, and A. R. Calderbank, Applications of space-time block codes and interference suppression for high capacity and high data rate wireless systems," in Proc. 32nd Asilomar Conf. Signals, Syst. Comput., pp. 1803-1810, 1998.

[5] A. Stamoulis, N. Al-Dhahir and A. R. Calderbank, Further results on interference cancellation and space-time block codes," in Proc. 35th Asilomar Conf. Signals, Syst. Comput., pp. 257-262, Oct. 2001. [6] J. Kazemitabar and H. Jafarkhani, Multiuser interference cancellation and detection for users with more than two transmit antennas," IEEE Trans. Commun., vol. 56, no. 4, pp. 574-583, Apr. 2008. [7] J. Kazemitabar and H. Jafarkhani, Performance analysis of multipleantenna multi-user detection," in Proc. 2009 Workshop Inf. Theory Applications, Feb. 2009. [8] A. Scaglione, P. Stoica, S. Barbarossa, G. Giannakis, and H. Sampath, Optimal designs for space-time linear precoders and decoders," IEEE Trans. Signal Process., vol. 50, no. 5, pp. 1051-1064, 2002. [9] D. Love and R. J. Heath, Limited feedback unitary precoding for orthogonal space-time block codes," IEEE Trans. Signal Process., vol. 53, no. 1, pp. 64-73, 2005. [10] A. Ghaderipoor and C. Tellambura, Optimal precoder for rate less than one space-time block codes," in Proc. IEEE International Conf. Commun., Glasgow, Scotland, June 2007. [11] H. Sampath and A. Paulraj, Linear precoding for space-time coded systems with known fading correlations," IEEE Commun. Lett., vol. 6, no. 6, pp. 239-241, June 2002. [12] G. Jongren, M. Skoglund, and B. Ottersten, Combining beamforming and orthogonal space-time block coding," IEEE Trans. Inf. Theory, vol. 48, pp. 611-627, Mar. 2002. [13] L. Liu and H. Jafarkhani, Application of quasi-orthogonal space-time block codes in beamforming," IEEE Trans. Signal Process., vol. 53, no. 1, pp. 54-63, Jan. 2005. [14] S. Ekbatani and H. Jafarkhani, Combining beamforming and space-time coding using quantized feedback," IEEE Trans. Wireless Commun., vol. 7, no. 3, pp. 898-908, Mar. 2008. [15] C. W. Tan and A. R. Calderbank, Multiuser detection of Alamouti signals," IEEE Trans. Commun., vol. 57, no. 7, pp. 2080-2089, 2009. [16] Y. Wu and A. R. Calderbank, Code diversity in multiple antenna wireless communication," in Proc. IEEE International Symp. Inf. Theory, Toronto, Canada, July 2008. [17] F. Li and H. Jafarkhani, Multiple-Antenna Interference Cancellation and Detection for Two Users Using Precoders," IEEE J. Sel. Topics Signal Process., vol. 3, no. 6, pp. 1066-1078, Dec. 2009. [18] M. K. Simon and M.-S. Alouini, Digital Communication over Fading Channels, 1st ed. NY: Wiley, 2000. [19] E. Sengul, E. Akay and E. Ayanoglu, Diversity Analysis of Single and Multiple Beamforming", IEEE Trans. Commun., vol. 54, no. 6, pp. 990-993, June 2006.

910

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

Feng Li (S 08) received the B.Eng. degree from Harbin Institute of Technology, Harbin, China, in 2005 and the M.Phil. degree from City University of Hong Kong, Hong Kong, in 2007. He is currently working toward the Ph.D. degree at the University of California, Irvine. His research interests include wireless communications and signal processing, focusing on precoder design for multi-user multiantenna wireless communication systems. Hamid Jafarkhani received the B.S. degree in electronics from Tehran University in 1989 and the M.S. and Ph.D. degrees both in electrical engineering from the University of Maryland at College Park in 1994 and 1997, respectively. In 1997, he was a Senior Technical Staff Member at AT&T Labs-Research and was later promoted to a Principal Technical Staff Member. He is currently a Chancellors Professor at the Department of Electrical Engineering and Computer Science, University of California, Irvine, where he is also the Director of Center for Pervasive Communications and Computing and the ConexantBroadcom Endowed Chair.

Dr. Jafarkhani ranked rst in the nationwide entrance examination of Iranian universities in 1984. He was a co-recipient of the American Division Award of the 1995 Texas Instruments DSP Solutions Challenge. He received the best paper award of ISWC in 2002 and an NSF Career Award in 2003. He received the UCI Distinguished Mid-Career Faculty Award for Research in 2006 and the School of Engineering Fariborz Maseeh Best Faculty Research Award in 2007. Also, he was a co-recipient of the 2006 IEEE Marconi Best Paper Award in Wireless Communications and the 2009 best paper award of the Journal of Communications and Networks. He was an Associate Editor for IEEE C OMMUNICATIONS L ETTERS from 2001-2005, an editor for the IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS from 2002-2007, an editor for the IEEE T RANSACTIONS ON C OMMUNICATIONS from 2005-2007, and a guest editor of the special issue on MIMO-Optimized Transmission Systems for Delivering Data and Rich Content" for the IEEE J OURNAL OF S ELECTED T OPICS IN S IGNAL P ROCESSING in 2008. Currently, he is an area editor for the IEEE T RANS ACTIONS ON W IRELESS C OMMUNICATIONS . He is listed as a highly cited researcher in http://www.isihighlycited.com. According to the Thomson Scientic, he is one of the top 10 most-cited researchers in the eld of computer science" during 1997-2007. He is an IEEE Fellow and the author of the book Space-Time Coding: Theory and Practice.