Adaptive Kalman smoothing of AR signals disturbed by impulses and colored noise

Abstract

In this paper, we describe a new and computationally efficient adaptive system for the enhancement of autoregressive (AR) signals which are disturbed by additive white or colored noise and impulsive noise. The system is comprised of an adaptive Kalman filter operating as a fixed lag smoother and a subsystem for AR parameter estimation. A superior performance is achieved by implementing a feedback loop between the Kalman filter output and the parameter estimation. Accordingly, the AR parameters are obtained from the enhanced signal and the influence of the disturbing noise on the parameter estimation is damped down. Impulsive noise is suppressed by an outlier detection scheme and by freezing the Kalman filter update during presence of impulses. The subsystem for AR parameter estimation can operate in a block processing mode or on a sample per sample basis. Another advantage of the adaptive Kalman filter is its tracking capability for short-time stationary signals.