JAM Journal of Applied Mathematics 1687-0042 1110-757X Hindawi Publishing Corporation 10.1155/2015/201369 201369 Research Article MIMO Detection for High Order QAM by Canonical Dual Approach http://orcid.org/0000-0002-4156-3200 Tian Ye 1 http://orcid.org/0000-0003-1034-677X Dang Jr-Fong 2 Mathiopouloss Panayotis Takis 1 School of Business Administration Southwestern University of Finance and Economics Chengdu 611130 China swufe.edu.cn 2 Department of Industrial Engineering & Management National Chiao Tung University Hsinchu 300 Taiwan nctu.edu.tw 2015 542015 2015 25 10 2014 23 03 2015 542015 2015 Copyright © 2015 Ye Tian and Jr-Fong Dang. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We develop a canonical dual approach for solving the MIMO problem. First, a special linear transformation is introduced to reformulate the original problem into a { 1 , 1 } constrained quadratic programming problem. Then, we derive a canonical dual problem which is piecewise continuous problem with no duality gap. Under certain conditions, the canonical problem becomes a concave maximization dual problem over a convex feasible domain. By getting the stationary point of the canonical dual problem, we can find either an optimal or approximate solution of the original problem. A gradient decent algorithm is proposed to solve the MIMO problem and simulation results are provided to demonstrate the effectiveness of the method.

1. Introduction

In the recent decade, multiple antennae communication systems have developed very fast since it could provide substantial performance gain over their single antenna counterparts . Therefore, how to efficiently detect the signal vector of transmitted symbols becomes an important issue. As a consequent result, the problem of multiple-input-multiple-output MIMO detection has been raised up and received considerable attention over several years.

Note that, in the communication scenarios, the signal model is always written in the following form: (1) y = H x + v , where y R m is a received signal vector, x R n is a transmitted symbol vector whose elements are drawn from a symbol constellation set, H is m × n real matrix that characterizes the input-output relation, and v R m is an additive white Gaussian noise with unit variance. It is worth pointing out that the signal model is sometimes represented in a complex-valued form. But it is easy to reformulate the complex-valued model to a real-valued model (please see  for details). The MIMO problem aims to detect the transmitted vector x based on the observations y and H . And the optimal symbol vector which minimizes the error probability can be found by solving the maximum-likelihood (ML) detection problem . It is a very fundamental problem in the communication area. The corresponding problem is as follows: MIMO min W y - H x 2 i s . t . W i I x i ± 1 , ± 3 , , ± u , i = 1 , , n , where · denotes the 2-norm and each symbol x i is drawn from a ( u + 1 ) -QAM constellation set S = { ± 1 , ± 3 , , ± u } (please see  for details). In this paper, we suppose u 3 . Unfortunately, the MIMO problem is NP-hard . Therefore, researchers have developed a lot of approximation algorithms.

Lattice decoding is an important research direction for the MIMO detection. It has received a lot of attention for its good tradeoff between detection accuracy and complexity . Naive lattice decoding (NLD) method relaxes the symbol bound constraints and finds the closest lattice point to the received signal over the whole lattice generated by the channel . In order to further improve the efficiency, some suboptimal lattice decoding methods, such as sampled decoding , embedded decoding , and lattice reduction-aided (LRA) methods , can be combined with lattice decoding method to accelerate the lattice point search. However, since the NLD method completely ignores the symbol bounds, it fails to achieve the optimal diversity-multiplexing tradeoff (DMT) under general MIMO system models [6, 8]. Then, in order to prevent the lattice points going too far away from the origin point, some researchers developed the regularized lattice decoding (RLD) method which adds a quadratic penalization term to the lattice decoding metric . Though the RLD method has been empirically found to be computationally fast for small to moderate problems sizes, its complexity would be prohibitive for large n and higher order QAM .

Another big family of MIMO detection algorithms is based on semidefinite relaxation (SDR). The SDR method relaxes the ML detection problem into a convex semidefinite programming (SDP) problem which leads to a polynomial-time complexity in the problem dimension. The SDR detector was first developed for the binary phase-shift keying (BPSK) constellation  and then extended to QPSK (4-QAM) constellation . Researchers have verified that the SDR detector can provide a constant factor approximation to the optimal log-likelihood value in the low signal-to-noise ration (SNR) region almost surely . Based on that, Wiesel et al.  proposed a polynomial-inspired SDR (PI-SDR) method for 16-QAM and proved that PI-SDR achieves an optimal Lagrangian dual lower bound of the ML. Sidiropoulos and Luo  designed a bound-constrained SDR (BC-SDR) method which has a special structure. Thus, compared to PI-SDR, BC-SDR makes fast implementations more favorable. Moreover, Mao et al.  developed a virtually antipodal SDR (VA-SDR) method for any 4 q -QAM (where q > 1 ). For the relationship and comparisons between these SDR detectors, please see . Though SDP problem has a theoretical low polynomial computational complexity, due to large problem size and slow SDP solvers, the actual computation time is very high in practice.

Besides, there are some other algorithms for the MIMO problem. Sphere decoder method is a classical one [4, 20]. However, it exhibits exponential complexity with respect to the problem size. Moreover, Goldberger and Leshem  proposed a new detection algorithm based on an optimal tree approximation in an unconstrained linear system. They showed that this algorithm outperforms other methods for the loop-free factor graph situation. Recently, Pan et al.  proposed a Lagrangian dual relaxation (LDR) for the MIMO problem. This method finds the best diagonally regularized lattice decoder to approximate the ML detector. They proved that the LDR problem yields a duality gap no worse than that of the SDR method.

In this paper, we present a canonical duality approach to the MIMO problem. The canonical duality theory is originally proposed for handling general nonconvex and/or nonsmooth systems . Canonical dual transformation and associated triality theory play a key role in the implementation. It is worth pointing out that the canonical dual transformation may convert a nonconvex and/or nonsmooth primal problem into a piecewise smooth canonical dual problem. And this dual problem has no duality gap in each subregion. Therefore, this powerful tool has a big potential in some global optimization problems and nonconvex nonsmooth analysis . Particularly, the canonical duality theory can be applied to the quadratic programming problem with integer constraints. Fang et al.  proposed a more general global optimization condition using the canonical duality approach. Wang et al.  developed a canonical duality approach for solving multi-integer quadratic programming problems. Thus, this paper adopts the canonical duality approach to study the MIMO problem.

The paper is arranged as follows. In Section 2, we introduce a special linear transformation to reformulate the original problem into a { - 1,1 } constrained quadratic programming problem. In Section 3, we develop the canonical dual problem for the new reformulation. Then we show that it has no duality gap under certain conditions. Moreover, some global optimality conditions are presented. In Section 4, we propose a gradient decent method to find the stationary points. Comparisons by simulations are provided in Section 5. The last section summarizes the paper and points out some future research directions.

2. Linear Transformation

The MIMO problem can be written as follows: MIMO 1 min W x T H T H x - 2 y T H x + y 2 i s . t . W I x i ± 1 , ± 3 , , ± u , i = 1 , , n . Let Q = H T H and f = 2 H T y . Note that Q is n × n real symmetric matrix and f is n -dimensional real vector. Since y 2 is a fixed scalar, problem MIMO 1 has the same optimal solutions with the following problem: MIMO 2 min W x T Q x - f T x i s . t . W I x i ± 1 , ± 3 , , ± u , i = 1 , , n . We first reformulate problem MIMO 2 into a { - 1,1 } constrained quadratic programming problem. Note that the set { ± 1 , ± 3 , , ± u } forms an arithmetic series; that is, the adjacent two elements in the set have a constant gap 2. Therefore, we can take advantage of this special structure.

Let y i = x i + u / 2 for i = 1 , , n ; then y i { 0,1 , , u } . Let e denote the n -dimensional vector with all elements being 1. Then, it is easy to verify that problem MIMO 2 is equivalent to the following problem: MIMO 3 min W 4 y T Q y - 4 u e T Q + 2 f T y + u 2 e T Q e + u f T e i s . t . W I y i 0,1 , , u , i = 1 , , n . Problem MIMO 2 and problem MIMO 3 have the same optimal value. Moreover, once we get an optimal solution of problem MIMO 3 , we can use x = 2 y - u e to get the corresponding optimal solution of problem MIMO 2 . Now, let Q ¯ = 4 Q and f ¯ = 4 u Q T e + 2 f . Since u 2 e T Q e + u f T e is also a fixed scalar, problem MIMO 3 has the same optimal solutions with the following problem: MIMO 4 min W y T Q ¯ y - f ¯ T y i s . t . W I y i { 0,1 , , u } , i = 1 , , n . Let Y = y R n y i    is    an    integer , 0 y i u denote the feasible domain of problem MIMO 4 . Set t = log ( u + 1 ) ; then t 2 and u + 1 2 t . Now, we can define a new set Z as follows: (2) Z = z R t + 1 n z - 1,1 t + 1 n . Note that each element in set Z is a ( t + 1 ) n -dimensional vector with all elements being −1 or 1. Then, we have the following theorem to show the relationship between Y and Z .

Theorem 1.

The transformation y i ( z ) = j = 1 t 2 j - 2 z ( j - 1 ) n + i + 1 / 2 ( u + 1 - 2 t ) z t n + i + u / 2 is a full mapping from Z to Y .

Proof.

We first show that, for any z Z , the corresponding y belongs to Y . Note that y i ( z ) can also be written as y i ( z ) = j = 2 t 2 j - 2 z ( j - 1 ) n + i - 2 t - 1 z t n + i + 1 / 2 z i + u + 1 / 2 z t n + i + u / 2 . The first part j = 2 t 2 j - 2 z ( j - 1 ) n + i - 2 t - 1 z t n + i is always an integer; the possible fraction can only occur in the second part 1 / 2 z i + u + 1 / 2 z t n + i + u / 2 . However, it is easy to verify that this part is always an integer as z i , z t n + i { - 1,1 } for u 3 , t 2 . Moreover, for i = 1 , , n , max { y i } = j = 1 t 2 j - 2 + u + 1 - 2 t / 2 + u / 2 = u and min { y i } = - j = 1 t 2 j - 2 - u + 1 - 2 t / 2 + u / 2 = 0 . Therefore, y i is an integer between 0 and u and y belongs to Y .

Then, we show that, for an arbitrary integer 0 y i u , there exist z ( j - 1 ) n + i - 1,1 for j = 1 , , t + 1 such that the transformation holds. If 0 y i 2 t - 1 , then y i can be written as y i = 2 t - 1 a t - 1 + 2 t - 2 a t - 2 + + 2 a 1 + 1 a 0 for particular a i 0,1 and j = 0 , , t - 1 . Then, let z j n + i = 2 a j - 1 for j = 0 , , t - 1 and y t n + i = - 1 . Thus, z ( j - 1 ) n + i { - 1,1 } for j = 1 , , t + 1 . It is easy to verify that j = 1 t 2 j - 2 z ( j - 1 ) n + i + u + 1 - 2 t / 2 z t n + i + u / 2 = j = 0 t - 1 2 j a j = y i . If 2 t - 1 < y i u , since 2 t u + 1 2 t + 1 , we have - 2 t 2 t - ( u + 1 ) 0 . Therefore, 0 y i + 2 t - ( u + 1 ) 2 t - 1 . Then following the similar way, we can write y i + 2 t - ( u + 1 ) as y i + 2 t - ( u + 1 ) = 2 t - 1 a t - 1 + 2 t - 2 a t - 2 + + 2 a 1 + 1 a 0 for particular a i 0,1 , j = 0 , , t - 1 . This time, let z j n + i = 2 a j - 1 for j = 0 , , t - 1 and z t n + i = 1 . Thus, z ( j - 1 ) n + i { - 1,1 } for j = 1 , , t + 1 . Consequently, we have j = 1 t 2 j - 2 z ( j - 1 ) n + i + u + 1 - 2 t / 2 z t n + i + u / 2 = j = 0 t - 1 2 j a j + ( u + 1 ) - 2 t = x i .

Therefore, the transformation is a full mapping from Z to Y .

It is worth pointing out that this transformation is a linear mapping. Moreover, since we use the advantage of the special structure of the original problem, the size of the reformulated problem is smaller than the problems derived by some traditional transformation methods .

Now, replacing y by z in problem MIMO 4 , we can get the following problem: MIMO 5 min W z T M z - b T z + c i s . t . W I z i - 1,1 , i = 1 , , t + 1 n , where (3) M = 1 4 Q ¯ 1 2 Q ¯ 2 t - 3 Q ¯ 1 4 u + 1 - 2 t Q ¯ 1 2 Q ¯ Q ¯ 2 t - 2 Q ¯ 1 2 u + 1 - 2 t Q ¯ 2 t - 3 Q ¯ 2 t - 2 Q ¯ 2 2 t - 4 Q ¯ 2 t - 3 u + 1 - 2 t Q ¯ 1 4 u + 1 - 2 t Q ¯ 1 2 u + 1 - 2 t Q ¯ 2 t - 3 u + 1 - 2 t Q ¯ 1 4 u + 1 - 2 t 2 Q ¯ , b = 1 4 [ 2 f ¯ - u Q ¯ e ] 1 2 [ 2 f ¯ - u Q ¯ e ] 2 t - 3 [ 2 f ¯ - u Q ¯ e ] 1 4 ( u + 1 - 2 t ) [ 2 f ¯ - u Q ¯ e ] , c = 1 8 u 2 e T Q ¯ e - u 2 f ¯ T e . Therefore, from Theorem 1, solving problem MIMO 5 is equivalent to solving problem MIMO 4 .

Let r = ( t + 1 ) n and ignore the fixed scalar c ; we focus on the following problem: MIMO 6 min W F ( z ) = z T M z - b T z i s . t . W I z i - 1,1 , i = 1 , , r .

Note that problem MIMO 6 has no linear constraints and its feasible domain is merely defined by Z = z R r z - 1,1 r .

3. Canonical Dual Problem and Global Optimality

Let ρ R r be a vector and let Diag ( ρ ) denote an r × r diagonal matrix with ρ i being the i th diagonal element. Let (4) M ρ = M + Diag ( ρ ) , P ( ρ ) = - 1 4 b T M ρ - 1 b - e T ρ .

Moreover, we define a set as follows: (5) D M = ρ R r ρ 0 , M ρ 0 , where M ρ is the determinant of the matrix M ρ .

Then following the work of [30, 31], we can get the canonical dual problem of problem MIMO 6 as follows: CDP sta P ρ ρ D M , where e is an r -dimensional vector with all elements being 1. sta P ρ ρ D M represents finding all stationary points (critical points) of P ( ρ ) over D M .

Moreover, for two vectors a , b R n , let a b denote the standard Hadamard product a b ( a 1 b 1 , a 2 b 2 , , a r b r ) T . Then, for the relationship between the primal problem and the dual problem, we have the next two important theorems.

Theorem 2.

If ρ D M is a stationary point of the dual objective function P ( ρ ) , then z ( ρ ) = 1 / 2 ( M ρ ) - 1 b is an integer vector in { - 1,1 } r . Moreover, if M ρ is positive definite, then the Hessian matrix of P ( ρ ) is negative definite.

Proof.

Note that since z = 1 / 2 ( M ρ ) - 1 b , we have M ρ z = 1 / 2 b . Thus, d M ρ z / d ρ ρ = ρ = d M z + Diag ρ z / d ρ ρ = ρ = M d z / d ρ ρ = ρ + Diag ( z ) + Diag ρ d z / d ρ ρ = ρ = 0 . Hence, we have d z / d ρ ρ = ρ = - ( M ρ ) - 1 Diag ( z ) . Therefore, d P ρ / d ρ ρ = ρ = - 1 / 2 d [ ( b T ( M ρ ) - 1 M ρ z ) + 2 e T ρ ] / d ρ ρ = ρ = - 1 / 2 d b T z / d ρ ρ = ρ - e = - 1 / 2 ( d z / d ρ ) T ρ = ρ b - e = 1 / 2 Diag ( z ) ( M ρ ) - 1 b - e = Diag ( z ) z - e = z z - e . Therefore, if ρ is a stationary point of P ( ρ ) , then z ( ρ ) z ( ρ ) - e = 0 . This implies that the vector z ( ρ ) = 1 / 2 ( M ρ ) - 1 b { - 1,1 } r .

Moreover, the Hessian matrix d 2 P ρ / d ( ρ ) 2 ρ = ρ = d z z - e / d ρ ρ = ρ = 2 Diag z d z / d ρ ρ = ρ = - 2 Diag ( z ) ( M ρ ) - 1 Diag ( z ) . Thus, if M ρ is positive definite, then the Hessian matrix of P ( ρ ) is negative definite.

Theorem 2 indicates that a stationary point ρ of P ( ρ ) over D M is corresponding to a feasible solution of the primal problem MIMO 6 .

Theorem 3.

The canonical dual problem CDP is perfectly dual to the primal problem MIMO 6 in the sense that if ρ ¯ D M is a stationary point of P ( ρ ) , then z ¯ = 1 / 2 ( M ρ ¯ ) - 1 b is a KKT point of problem MIMO 6 and F ( z ¯ ) = P ( ρ ¯ ) .

Proof.

Note that problem MIMO 6 can be written as follows: (6) min W F ( z ) = z T M z - b T z i s . t . W I z z - e = 0 , W W W W W W W W i z R r . Its corresponding Lagrangian function is (7) L ( z , ρ ) = z T M z - b T z + ρ T ( z z - e ) . Therefore, the KKT conditions are as follows: (8) L z = 2 M z - b + 2 Diag ( ρ ) z = 2 M ρ z - b = 0 , L ρ = z z - e = 0 . Thus, from Theorem 2 and the definition of z ¯ ( ρ ¯ ) , it is easy to verify that z ¯ ( ρ ¯ ) is a KKT point if ρ ¯ is a stationary point of P ( ρ ) .

Moreover, we have (9) z ¯ T M z ¯ - b T z ¯ W W = z ¯ T ( M + Diag ( ρ ¯ ) ) z ¯ - b T z ¯ - z ¯ T Diag ( ρ ¯ ) z ¯ W W = 1 2 z ¯ T b - b T z ¯ - z ¯ T Diag ( ρ ¯ ) z ¯ W W = - 1 2 b T z ¯ - z ¯ T Diag ( ρ ¯ ) z ¯ W W = - 1 2 b T z ¯ - e T ρ ¯ W W = - 1 4 b T M ρ ¯ - 1 b - e T ρ ¯ W W = P ( ρ ¯ ) . The result follows.

KKT conditions provide necessary conditions for local minimizers in a nonconvex programming problem. Next, we show that the canonical dual problem is a concave maximization dual problem over a convex feasible domain under certain conditions. First, we define a subset of set D M as follows: (10) D M + = ρ R r ρ 0 , M ρ 0 , where M ρ 0 indicates that M ρ is positive definite. Then, we present the sufficient global optimal conditions.

Theorem 4.

Assume that ρ ¯ D M is a stationary point of P ( ρ ) and z ¯ ( ρ ¯ ) = 1 / 2 ( M ρ ¯ ) - 1 b . If ρ ¯ D M + , then z ¯ is a unique global minimizer of F ( z ) over Z and ρ ¯ is a global maximizer of P ( ρ ¯ ) over D M + with (11) z ¯ T M z ¯ - b T z ¯ = min z Z z T M z - b T z = max ρ D M + C ( ρ ) = C ( ρ ¯ ) .

Proof.

Note that if ρ D M + , then M ρ 0 . Thus, Theorem 2 indicates that the Hessian matrix of P ( ρ ) is negative definite over D M + . Then we know that the canonical dual function P ( ρ ) is strictly concave over D M + . Therefore, a stationary point of P ( ρ ) must be a global maximizer of P ( ρ ) over D M + . Since the stationary point ρ ¯ D M + , Theorem 3 indicates that F ( z ¯ ) = P ( ρ ¯ ) .

On the other hand, since z ¯ is a KKT point assured by Theorem 3, we have z ¯ z ¯ - e = 0 . Therefore, F ( z ¯ ) = z ¯ T M z ¯ - b T z ¯ + ( ρ ¯ ) T ( z ¯ z ¯ - e ) = z ¯ T ( Q + Diag ( ρ ¯ ) ) z ¯ - b T z ¯ - e T ρ ¯ . Note that Q + Diag ( ρ ¯ ) is positive definite; therefore z ¯ = 1 / 2 ( M ρ ¯ ) - 1 b is a global minimizer. Therefore, z ¯ T ( Q + Diag ( ρ ¯ ) ) z ¯ - b T z ¯ - e T ρ ¯ = min z R r { z T ( Q + Diag ( ρ ¯ ) ) z - b T z - e T ρ ¯ } min { z R r , z z = e } z T Q z - b T z . Therefore, z ¯ is a global minimizer of F ( z ) over Z .

Above all, we transform the original problem which is a multi-integer quadratic programming problem into a piecewise continuous canonical dual problem by using the canonical duality theory. It is worth pointing out that the original problem and the canonical dual problem have no duality gap over D M . Moreover, if ρ D M + , the canonical dual problem becomes a concave maximization problem over a convex feasible domain which can be solved very efficiently. To the best of our knowledge, among all the existing methods, strong duality only holds for the Lagrangian dual relaxation (LDR) in the special 2-PAM case . However, the canonical dual problem holds for any order PAM problem over D M . Therefore, in some sense, the canonical dual problem provides the best dual problem of all.

4. Algorithm

Note that if ρ is a stationary point of P ( ρ ) over D M , then z ( ρ ) = 1 / 2 ( M ρ ) - 1 b is a feasible solution of problem MIMO 6 . Moreover, if ρ D M + , the corresponding solution z ( ρ ) is the global optimal solution of problem MIMO 6 . Therefore, the solution of the dual problem CDP is corresponding to an optimal or approximation solution of problem MIMO 6 .

The key issue for solving the problem CDP is to avoid the matrix M ρ becoming a singular matrix. Since z ( ρ ) = 1 / 2 ( M ρ ) - 1 b , we have 2 ( M + Diag ( ρ ) ) z ( ρ ) = b . Thus, 2 Diag ( z ( ρ ) ) ρ = b - 2 M z ( ρ ) . Note that ( Diag ( z ) ) - 1 = Diag ( z ) for z Z . Therefore, ρ = 1 / 2 ( Diag ( z ( ρ ) ) b - 2 Diag ( z ( ρ ) ) M z ( ρ ) ) . Hence, ρ = 1 / 2 ( z ( ρ ) b - 2 ( M z ( ρ ) ) z ( ρ ) ) . Therefore, if the matrix M ρ is close to a singular matrix, we can perturb the vector b to change the value of ρ and avoid the situation.

Let z denote the infinity norm of vector z . Then, we propose an algorithm based on the gradient methods to solve problem CDP .

Canonical Dual Algorithm

Step 1 (initialization step).

Let T = max i = 1 n j = 1 n M i j , ϵ > 0 be a sufficiently small number. Let k = 0 , ρ 0 = 4 ( T + ϵ ) e be the initial feasible solution of problem CDP .

Step 2.

Let z ( ρ k ) = 1 / 2 ( M ρ ) - 1 b . If P ρ k ϵ , go to Step 4. Otherwise, ρ k + 1 = ρ k - α k g k , where α k = P ρ k is a nonnegative scalar and g k = P ( ρ k ) ; k = k + 1 .

Step 3.

If M ρ k < ϵ , then let b = ( 1 + 1 / b ) b and ρ k = 1 / 2 ( z ( ρ k ) b - 2 ( M z ( ρ k ) ) z ( ρ k ) ) . Go back to Step 2.

Step 4.

Return z ( ρ k ) and use Theorem 1 to get the corresponding solution y .

5. Comparisons by Simulations

In this section, we use simulations to compare the canonical dual method (CDM) with some other benchmarked approximating methods, such as inexact ML sphere decoding (ML-SD) , semidefinite relaxation (SDR) , and inexact MMSE lattice decoding (MMSE-LD) .

The channel matrix H comprises i.i.d. elements drawn from a zero-mean normal distribution of unit variance. The symbol vector x is element-wise i.i.d. uniformly distributed with each element s i drawn from the standard u + 1 2 -QAM constellation set. Moreover, v is the additive white Gaussian noise with zero mean and variance σ v 2 . Note that the signal-to-noise ratio (SNR) is defined as E ( H x 2 ) / E ( v 2 ) = n σ x 2 / σ v 2 , where σ x 2 is the variance of the elements of x .

We use two different problem sizes as ( m 1 , n 1 ) = ( 8,8 ) and ( m 2 , n 2 ) = ( 16,16 ) . And for each case, we test every numerical example under two situations: 16-QAM and 64-QAM. All the simulations are implemented using MATLAB 7.9.0 on a computer with Intel Core 2 CPU 2.50 Ghz and 2 G memory. Moreover, the solvers of cvx  are incorporated in solving the SDP problems.

Figures 1 and 2 plot the SERs of the various methods versus SNR under different problem sizes for 16-QAM constellations and 64-QAM constellations, respectively.

Symbol error rate comparison for different methods, ( m , n ) = ( 8,8 ) .

16-QAM

64-QAM

Symbol error rate comparison for different methods, ( m , n ) = ( 16,16 ) .

16-QAM

64-QAM

Table 1 provides the average computational time in seconds for each method under different cases. Note that the notation “8 16-QAM” denotes the 16-QAM problem whose size is ( m , n ) = ( 8,8 ) . The other notations follow the same way.

Average computational time for different methods under different cases.

Methods Cases
8 16-QAM 16 16-QAM 8 64-QAM 16 64-QAM
ML-SD 8.04 s 31.2 s 15.39 s 38.44 s
SDR 6.21 s 28.17 s 10.32 s 36.45 s
MMSE-LD 10.42 s 35.76 s 18.47 s 54.53 s
CDM 7.27 s 36.89 s 14.23 s 45.17 s

From the simulation results, we can see that the CDM method outperforms all other methods in all situations. And the SER gaps between the CDM method and other methods are significantly wide. Moreover, the computation time indicates that the CDM method is quite efficient compared with other methods. It is worth pointing out that the CDM method is much more effective for the high order QAM problem.

6. Conclusion

In this paper, we have developed a canonical dual method to solve the MIMO problem. By introducing a tricky linear transformation, the original problem can be reformulated as a { - 1,1 } constrained quadratic programming problem with the smallest size. Then, a canonical dual problem has been developed. The stationary points of the canonical dual problem correspond to the KKT points of the original problem. And there is no duality gap between the primal problem and its canonical dual problem under some conditions. Solving the canonical dual problem can lead to either an exact or approximate solution of the original problem. The simulation results indicate that the canonical dual method performs quite well compared with other benchmark methods.

The canonical dual approach offers a different angle to study the MIMO problem. It sheds some light on designing a new but more efficient algorithm. For the future study, some searching methods can be combined in the algorithm to improve the efficiency for finding the stationary points. And we aim to figure out which subclass of the MIMO problem can be exactly solved by the canonical duality theory.

Conflict of Interests

The authors declare that there is no conflict of interests related to this paper.

Acknowledgment

Ye Tian’s research has been supported by the Chinese National Science Foundation no. 11401485.

Tse D. Viswanath P. Fundamentals of Wireless Communication 2005 Cambridge, UK Cambridge University Press Ma W.-K. Su C.-C. Jalden J. Chang T.-H. Chi C.-Y. The equivalence of semidefinite relaxation MIMO detectors for higher-order QAM IEEE Journal on Selected Topics in Signal Processing 2009 3 6 1038 1052 10.1109/JSTSP.2009.2035798 2-s2.0-76149118448 Verdú S. Multiuser Detection 1998 Cambridge, UK Cambridge University Press Damen M. O. El Gamal H. Caire G. On maximum-likelihood detection and the search for the closest lattice point IEEE Transactions on Information Theory 2003 49 10 2389 2402 10.1109/tit.2003.817444 MR2027993 2-s2.0-0242365586 Verdú S. Computational complexity of optimum multiuser detection Algorithmica 1989 4 3 303 312 10.1007/bf01553893 MR1000216 2-s2.0-0010185175 El Gamal H. Caire G. Damen M. Lattice coding and decoding achieve the optimal diversity-multiplexing tradeoff of MIMO channels IEEE Transactions on Information Theory 2004 50 6 968 985 10.1109/tit.2004.828067 MR2094862 2-s2.0-2942653060 Singh A. K. Elia P. Jaldèn J. Achieving a vanishing SNR gap to exact lattice decoding at a subexponential complexity IEEE Transactions on Information Theory 2012 58 6 3692 3707 10.1109/tit.2012.2190709 MR2924395 2-s2.0-84861359992 Taherzadeh M. Khandani A. K. On the limitations of the naive lattice decoding IEEE Transactions on Information Theory 2010 56 10 4820 4826 10.1109/tit.2010.2059630 MR2808651 2-s2.0-77956687566 Liu S. Ling C. Stehlè D. Decoding by sampling: a randomized lattice algorithm for bounded distance decoding IEEE Transactions on Information Theory 2011 57 9 5933 5945 10.1109/tit.2011.2162180 MR2857944 2-s2.0-80052349133 Luzzi L. Stehlè D. Ling C. Decoding by embedding: correct decoding radius and DMT optimality IEEE Transactions on Information Theory 2013 59 5 2960 2973 10.1109/tit.2012.2236144 MR3053388 2-s2.0-84876778639 Wübben D. Seethaler D. Jaldèn J. Matz G. Lattice reduction IEEE Signal Processing Magazine 2011 28 3 70 91 10.1109/msp.2010.938758 2-s2.0-79955367003 Jaldèn J. Elia P. DMT optimality of LR-aided linear decoders for a general class of channels, lattice designs, and system models IEEE Transactions on Information Theory 2010 56 10 4765 4780 10.1109/tit.2010.2059493 MR2808647 2-s2.0-77956706410 Jaldèn J. Ottersten B. On the complexity of sphere decoding in digital communications IEEE Transactions on Signal Processing 2005 53 4 1474 1484 2-s2.0-17444400753 10.1109/tsp.2005.843746 MR2128255 Tan P. H. Rasmussen L. K. The application of semidefinite programming for detection in CDMA IEEE Journal on Selected Areas in Communications 2001 19 8 1442 1449 10.1109/49.942507 2-s2.0-0035416437 Ma W.-K. Davidson T. N. Wong K. M. Ching P.-C. A block alternating likelihood maximization approach to multiuser detection IEEE Transactions on Signal Processing 2004 52 9 2600 2611 10.1109/TSP.2004.832062 MR2091806 2-s2.0-4344696766 Kisialiou M. Luo Z.-Q. Performance analysis of quasi-maximum-likelihood detector based on semi-definite programming 3 Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP '05) March 2005 III433 III436 2-s2.0-33646764918 10.1109/icassp.2005.1415739 Wiesel A. Eldar Y. C. Shamai S. Semidefinite relaxation for detection of 16-QAM signaling in MIMO channels IEEE Signal Processing Letters 2005 12 9 653 656 2-s2.0-27644579613 10.1109/LSP.2005.853044 Sidiropoulos N. D. Luo Z.-Q. A semidefinite relaxation approach to MIMO detection for high-order QAM constellations IEEE Signal Processing Letters 2006 13 9 525 528 10.1109/lsp.2006.874443 2-s2.0-33747804290 Mao Z. Wang X. Wang X. Semidefinite programming relaxation approach for multiuser detection of QAM signals IEEE Transactions on Wireless Communications 2007 6 12 4275 4279 10.1109/TWC.2007.060418 2-s2.0-37049013911 Viterbo E. Boutros J. A universal lattice code decoder for fading channels IEEE Transactions on Information Theory 1999 45 5 1639 1642 10.1109/18.771234 MR1699088 2-s2.0-0032636505 Goldberger J. Leshem A. MIMO detection for high-order QAM based on a Gaussian tree approximation IEEE Transactions on Information Theory 2011 57 8 4973 4982 10.1109/tit.2011.2159037 MR2849099 2-s2.0-79961015397 Pan J. Ma W.-K. Jaldèn J. MIMO detection by Lagrangian dual maximum-likelihood relaxation: reinterpreting regularized lattice decoding IEEE Transactions on Signal Processing 2014 62 2 511 524 10.1109/tsp.2013.2292040 MR3158383 2-s2.0-84892391939 Gao D. Y. Duality Principles in Nonconvex Systems: Theory, Methods and Applications 2000 39 Berlin, Germany Springer Nonconvex Optimization and its Applications 10.1007/978-1-4757-3176-7 MR1773838 Gao D. Y. Canonical duality theory and solutions to constrained nonconvex quadratic programming Journal of Global Optimization 2004 29 4 377 399 10.1023/b:jogo.0000048034.94449.e3 MR2104405 2-s2.0-9244233832 Gao D. Y. Solutions and optimality criteria to box constrained nonconvex minimization problems Journal of Industrial and Management Optimization 2007 3 2 293 304 10.3934/jimo.2007.3.293 MR2329373 ZBL1171.90504 2-s2.0-42649126195 Gao D. Y. Watson L. T. Easterling D. R. Thacker W. I. Billups S. C. Solving the canonical dual of box- and integer-constrained nonconvex quadratic programs via a deterministic direct search algorithm Optimization Methods & Software 2013 28 2 313 326 10.1080/10556788.2011.641125 MR3003146 2-s2.0-84870888624 Latorre V. Gao D. Y. Canonical dual solutions to nonconvex radial basis neural network optimization problem Neurocomputing 2014 134 189 197 10.1016/j.neucom.2013.06.050 2-s2.0-84896542852 Zhou X. Gao D. Y. Yang C. Canonical primal-dual algorithm for solving fourth-order polynomial minimization problems Applied Mathematics and Computation 2014 227 246 255 10.1016/j.amc.2013.11.013 MR3146313 2-s2.0-84889677127 Fang S.-C. Gao D. Y. Sheu R.-L. Wu S.-Y. Canonical dual approach to solving 0-1 quadratic programming problems Journal of Industrial and Management Optimization 2008 4 1 125 142 MR2379923 ZBL1180.90195 2-s2.0-56649105067 10.3934/jimo.2008.4.155 Wang Z. Fang S.-C. Gao D. Y. Xing W. Global extremal conditions for multi-integer quadratic programming Journal of Industrial and Management Optimization 2008 4 2 213 225 2-s2.0-43949105773 ZBL1161.90457 MR2386071 10.3934/jimo.2008.4.213 Wang Z. Fang S.-C. Gao D. Y. Xing W. Canonical dual approach to solving the maximum cut problem Journal of Global Optimization 2012 54 1 341 351 10.1007/s10898-012-9881-8 MR2979633 2-s2.0-84867270705 Grant M. Boyd S. CVX: Matlab Software for Disciplined Programming, Version 1.2 2010 http://cvxr.com/cvx/