The Implementation Of Blind Signal Separation Based On Digital Signal Processor

The blind source separation (BSS) origins from the famous Cocktail-Partyproblems, is refers the process of restoring each independent source signal from theobserved signals, in the situation of not knowing the system transfer function,transmits channel parameters and the probability distribution of the source signals. TheIndependent Component Analysis (ICA), as a kind of effective blind source separationmethods, is gradually developed in the recent 20 years, it is a statistical method, and itsgoal is to separate the independent source signals from the mixed signals, making theoutput signals to be mutually independent as far as possible, from a collects ofobversed signals from the sensors. The method of ICA has widespread applications inthe signals processing domain of voice signals processing, the mobile communication,the picture processing and the biomedicine signal processing and so on at present. Ithas become a research hot spot of signal processing principle in the blind signalsprocessing, artificial nerve network, and a lot of mature algorithms have alreadyappeared each day by day. The digital signal processor (DSP), took as one kind ofmicroprocessor, has some unique structures and characteristics, that extremely suits tothe digital signal processing, specially real-time processing, It has the extremely highrunning speed and the nimble programmable characteristics. The appearance and therapidly developments of the DSP, provides an ideal hardware platform, making it to bepossible that moves those blind signal separation algorithms into the practicalapplication from the laboratory simulation. Therefore, the question of the hardwarerealization of blind signal separation algorithms has received the more and morewidespread attention, is a valuable topic which is worth studying.The main work and innovation of this paper are as follows:(1) On the basis of introducing the basic theory of the blind source separationand independent component analysis, this paper focuses on the study of the FastICAalgorithm and verifies the separation effect of the algorithm using two-way voicesignals in Matlab. The simulation results show that the algorithm is feasible.(2) After studying the structural characteristics of the DSP hardware, this paperdesigns a real-time voice blind signal processing hardware platform. This real-timevoice processing platform is based on TI's TMS320C5402 DSP hardware. The designof essential part such as Clock and power supply module, memory expansion module,voice data acquisition and data transmission module are in-depth studied. And also the boot mode for the system are analyzed.(3)After familiaring with the TI Corporation's DSP Integrated DevelopmentEnvironment CCS and the eXpressDSP software structure initiated by TI, this papergives thorough research on the Trs DSP/BIOS, an embedded operating systemkernel, pays much attentions to the key technologies, such as thread scheduling,datacommunications and so on. Then we constructed a software platform of an embeddedreal-time voice blind signals separation system, the specific voice blind separationalgorithm can be inserted only as a thread into the platform. The procedures of theFastICA algorithms, based on the DSPLIB (digital signal processing library),programmed in the C programming language, are simulatd on the Code ComposerStudio IDE, and the desired results are achieved.