Researches On Blind Signal Separation Methods And Their Applications

Recently, Blind Source Separation (BSS) methods for Multi-Input Multi-Output(MIMO) systems are hot research topics in many fields such as wirelesscommunication, hand-free speech communication and EEG signal processing, etc.However, everything is just at the beginning. Through understanding and utilizationof the previous achievements, this dissertation focuses on developing innovative BSSmethods for MIMO systems. The dissertation begins with defining MIMO systems. Then the basic BSS modelfor MIMO systems is described.It is shown that BSS can be achieved in time-main,space-domain and space-time domain. Thereafter, the dissertation presents itscontributions to all three directions as follows: 1). The time domain BSS for MIMO systems is investigated from the perspectiveof wireless communication systems. Researches are focusing on Innovation andSecond–Order Statistics (SOS) based blind channel identification/equalizationalgorithms and their relationships to the basic BSS model. The main contributionsinclude: a). Five classic algorithms, LPA, OPDA, MSLP, LSS and CMOE areanalyzed theoretically. And their natures are discussed in details. Through computersimulation, their characteristics and complexities are compared. b). Problems inherent in SOS based methods are discussed and the reasons areanalyzed. In order to solve the problems, a new semi-blind channel identification andequalization algorithm is developed by using of the prior knowledge in wirelesscommunication systems. Compared with traditional training-sequence basedequalization methods, the new algorithm can efficiently improve the system capacity.Compared with blind methods, it is easier to be implemented. Thus, it has greatpotential for applications in wireless communication systems. 2). The space domain BSS for MIMO systems is investigated from the 第 III 页Abstractperspective of hand free speech communication systems. Researches are focusing onmicrophone array based blind beam-forming algorithms. The primary contributionsinclude: a). A smart microphone array system for speaker tracking and speech signalseparation is developed. Computer simulation verifies the validity and effectivenessof the scheme. b). Properties of speech signals are incorporated to improve the blindbeam-forming algorithm and the thernel technique of the microphone system.Computer simulation results show that the improved algorithm has lower complexityand overall system performance is greatly improved. c). To compensate the distortion effect caused by non-ideal microphones, twomicrophone calibration algorithms were developed. d). The corresponding software and hardware design is implemented and testedon TMS320C6711 DSP platform. 3). The space-time BSS for MIMO systems is investigated in this part. A newalgorithm for separating spatial speech mixture utilizing Innovation commonly usedin blind channel identification and equalization is developed. It uses innovation modelto whiten the mixture and then makes use of space domain BSS method to estimatethe mixing matrix. The new algorithm does not depend on either the geometry of themicrophone array or estimating the Direction of Arrival (DOA). Furthermore, it doesnot require the sources to be white signals. It can be applied in both hand-free speechcommunication systems and wireless communications.