Stability of adaptive FIR filters

Abstract

Svoboda's matrix inversion algorithm is used to analyze the adaptive FIR (finite-impulse response) filter. To get stable response, conditions of convergence and solvability have to be met. They determine the maximum correction step of the adaptive loop, the lower bound of the norm of output signal vector and the statistical properties of the reference signal. Methods of signal decorrelation are discussed and the upper bounds of resolution of a digital filter are derived. The long-term instability caused by parasitic DC voltages in the correction loop is analyzed and stabilization methods are suggested.<>