Research On Maximum-likelihood Detection Algorithms For MIMO Wireless Communications Systems

Multiple-input Multiple-output (MIMO) wireless systems with multipleantennas at both the transmitter and receiver sides have proven to have thepotential of greatly increasing the capacity and the reliability of the wirelesslinks. With the transmitted data streams from different transmit antennasinterfering with each other, it is a great challenge to achieve low-complexitymaximum-likelihood signal detection at the MIMO receiver side. In this thesis,we present several low-complexity exact maximum-likelihood (ML) detectionalgorithms which are very helpful in bringing the high-performancemaximum-likelihood detection for MIMO system into reality. We alsoinvestigate the problem of ML joint channel estimation and data detections fornon-coherent orthogonal Space-time Block codes (OSTBC) systems.Through transforming the maximum-likelihood detection for MIMOsystem into an equivalent problem of searching for the closet lattice pointover a tree graph, we propose two kinds of optimal or near optimal algorithms:ML stack detection algorithms and metric-guided algorithm. ML stackdetection algorithms can achieve exact maximum-likelihood performance withmuch lower complexity than exhaustive search. Metric-guided algorithm canalso achieve the near-optimal detection performance efficiently and, moreover,by adjusting two parameters inherent in metric-guided algorithm, differentlevels of tradeoff between performance and complexity can be naturally made.Simulation results have shown that ML stack algorithms and metric-guidedalgorithm have great performance and complexity advantage overconventional MIMO detection algorithms.Sphere decoder is another low-complexity approach proposed for MLdetection in MIMO systems. However, a sphere radius has to be specifiedbeforehand heuristically in conventional sphere decoders and how tooptimally specify such a sphere radius remained an open problem. In thisthesis, we propose a novel sphere decoder called IR-SD with a new radiusupdate mechanism and have solved the problem of optimally specifyingsphere radius. It has been theoretically proven that the IR-SD is optimalamong sphere decoders in reducing the computational complexity of findingthe exact ML solution. Theoretical results of the complexity of IR-SD havebeen derived and simulation results show that IR-SD greatly reduces thecomputational complexity of MIMO detection when compared with the fastestsphere decoder ever known before this. In addition, we design a parallelsphere decoder (P-SD) with competitive extensions and P-SD can acceleratethe speed of finding the ML solution compared with conventional spheredecoders while requiring storage only linear with the search dimension.Finally, we study the problem of optimal receiver signal detection fornon-coherent orthogonal space-time block codes (OSTBC) wireless systems.We show that the joint optimal channel estimation and data detection forOSTBC system can be reduced to an integer least square problem and proposeutilizing the newly developed IR-SD in finding the maximum-likelihood datasequence. Simulation results show that considerable performance gain can beachieved with this new method of ML joint channel estimation and datadetection and that IR-SD requires much lower computational complexity thannaive exhaustive search method.