现代数字信号处理AdvancedDigitalSignalProcessingch3adaptivefilter.ppt

Advanced Digital Signal Processing(Modern Digital Signal Processing)Chapter 3 Adaptive Linear Filter  3.1 Introduction The Classes of Adaptive Linear Filter By the length of linear filter FIR: always stable; good convergence properties; possibly linear-phased IIR: probably less estimation error (residual) than FIR By the structure of linear filter Transversal Lattice: fast convergence; insensitive to finite word-length effects; modular structure By the adaptive algorithm Least mean square (LMS) Recursive least square (RLS) Other variances of LMS or RLS Performances of Adaptive Filter Convergence rate of adaptive algorithm Misadjustment Computational complexity of adaptive algorithm Expected properties of adaptive filter structure: high modularity, parallelism, concurrency (suitable for implementation with VLSI) Numerical stability and numerical accuracy Robustness Adaptive algorithm is insensitive to the initial values 3.2 Transversal Adaptive FIR Filter Multiple Input Adaptive Linear Combiner Single Input Adaptive FIR Filter Optimum Solution (MMSE) of Adaptive FIR Filter 3.3 MSE Performance Surface MSE Performance Function Weight Deviation Vector Principle Axes Coordinate System Searching the Performance Surface 3.4 LMS Adaptive Algorithm The Gradient of Performance Surface Natural coordinate system Shifted coordinate system Steepest Descent Method (B. Widrow, 1959) Basic principles Searching the optimum point along the negative gradient direction, such a direction is the one with the steepest descent of the performance function. Natural coordinate system Sufficient condition for convergence Transition process Limitations of steepest descent algorithm Least Mean Square (LMS) Algorithm Estimation of the gradient Learning curves of a weight The average of 50 learning curves Misadjustment Excess mean square error Misadjustment Comments on LMS algorithm Simplicity and low computation loads Relatively slow convergence