NONLINEAR EFFECTS OF THE LMS PREDICTOR FOR CHIRPED INPUT SIGNALS

This paper shows that it is possible for an adaptive transversal prediction filter to outperform the fixed Wiener predictor of the same length for narrowband input signal embedded in Added White Gaussian Noise (AWGN). The error transfer function approach, which takes into account of the correlation of predictor error feedback and input signal, is derived for stationary and chirped input signals. It shows that with a narrowband input signal, the nonlinear effect is small for a 1-step predictor, but increases in magnitude as the prediction distance is increased. We also show that the LMS predictor uses information from the past errors more effectively than the Recursive Least Square (RLS) predictor, as a consequence, the magnitude of nonlinear effects of the LMS predictor are more significant than for the RLS predictor.