Research On Intelligent Detection Techniques Of Multi-input Multi-output Communications System

Multi-input multi-output (MIMO) communications system is an overwhelming topicin the realm of research and engineering. Signal detection is important in the whole systemwhich is recognized as the key technology. Signal detection technology recognizes differentusers’ signal through classification of the received data stream, which is a mixture ofdifferent users’(multiuser) information. This technology mainly comprises two categories,namely the optimal detectors and suboptimal detectors, of which suboptimal techniquespossesses a prevalent role. Besides, intelligent signal detection techniques are novel topicand research frontier. Thus it is essential and meaningful to undertake research within thecontext of intelligent signal detection.Main works of this thesis are as follows.(1) Study on intelligent signal detection techniques that deploys support vectormachine(SVM). A study is undertaken on intelligent signal detection utilizing SVM andparticle swarm optimization algorithm (PSO). This study tries to find the balance betweenlow error rate and fast detection speed under the circumstance of a few training samples inaddition to the insufficient information for channel estimation. This learning firstly trainsthe machines through regression to obtain optimized models in combination of PSOalgorithm. Then the incoming unknown data stream is detected by the already trainedmodels. Experiment shows that the proposed PSO-SVM detector enjoys a relatively highdetection accuracy which meets the requirements of MIMO communication system.(2) Study on intelligent signal detection techniques based on relevance vector machine(RVM). In order to overcome some of the drawbacks of SVM, RVM signal detectortechnique is proposed. Experiment result shows that RVM is capable of ensuring highdetection accuracy via a small number of relevant vectors. The RVM detector has bettergeneralization ability and requires less computation owing to the advantage of smallcollection of relevant vectors, which makes it stand out among MIMO detectors.(3) Study on kernel function selection. Discussion is made on the influence of RVMdetector performance inserted by different kernel functions, such as Cauchy kernel,multinomial kernel and Gaussian RBF kernel. Numerical experiment investigates themulticlass classification method. It analyses the classification performance using differentkernels and derives meaningful conclusion. Therefore it provides theoretic reference to thedesign of intelligent signal detectors for MIMO communication system.