Advanced Algorithms for 1-D Adaptive Filtering

摘要: In adaptive filtering practice, the Least Mean Squares (LMS) algorithm is widely used due to its computational simplicity and ease of implementation. However, since convergence rate depends on eigenvalue ratio autocorrelation matrix input noise signal, an LMS filter converges rather slowly when trained with colored as signal. continuing increase power that currently available in modern integrated signal processors (simply called “DSP chips” throughout following discussion), designers should be free future use more computationally intensive algorithms can perform better than simple real time applications. The objective this chapter explore several these intensive, but potentially performing, algorithms. particular, we will consider four classes have received attention by research community over last few years: 1) data-reusing algorithms, 2) orthogonalization pseudo-random (PR) modulation, 3) Gauss-Newton optimization for FIR filters, 4) block IIR filters using preconditioned conjugate gradient techniques.