Common Filter Performance Analysis In Bimodal Noise Background

In recent years, signal processing gradually be a new subject as the rapiddevelopment of communication, radar, sonar and auto-control, etc. Especially digitalsignal processing, which has touched various areas in theory and application, in itsdomain it provide a tool to basic analysing, modeling and colligating. Modern signalprocessing, which object is obtaining the best performance in information’s optimalutilization, it analyse and dispose signal in non-Linear, non-Gaussian andnon-stationary background, and remarkable advances have been made in complexityof arithmetic.On the basis of bimodal noise，this dissertation complement and perfect the signalprocessing theory in communication, and it solving some problems about bimodalnoise, by using modern signal processing method. The main work is as follows:(1) It reviews present conditions of development of communication systems, andintroduces the history and current situation of the processing technology of noise,and provides the perspective of the theory of bimodal noise.(2) Research and analysis of the bimodal noise model and the basic statisticalcharacteristics on it.(3) Elaborated systematically signal estimation theory, mainly introduced maximum aposteriori probability estimation criterion, criterion of maximum likelihoodestimation, minimum mean-square error(MMSE) estimation. Minimum meanabsolute error estimation criterion, Bayesian estimation criterion, Least squaresestimation criterion. In the practical application, quantitative inference one orseveral signals from the obtained observed signals or observation of the samplesis a estimation problem.(4) Study of the extended Kalman filter and particle filter algorithm in bimodal noisebackground, and an improved algorithm is proposed. For nonlinear, non Gauss problem, based on the Bayesian estimation theory to signal state estimation,filtering and estimation on the system with bimodal noise measurements.(5) Study of the adaptive algorithm and affine projection algorithm, in response to theinput data of the system has certain correlation, the convergence speed of LMSalgorithm and normalization of LMS algorithm will be low, this paper presents animproved algorithm. For the identification of unknown system shows goodfiltering performance.