Performance Improvement of DPLL's in Non-Gaussian Noise Using Robust Estimators

Abstract

This paper deals with the analysis of DPLL's operating in the presence of non-Gaussian noise. The loop model developed here allows for multiple input samples each loop sampling period, complex processing of the input sample set, and various noise distributions. In particular, six different robust estimators are used as processors of the input: these are the trimmed mean estimator, a one-step Huber M -type estimator, a new M -type estimator, and three versions of a zero-memory nonlinear (ZNL) estimator: hole punch, limiter, and sign detector. The tracking performance of a second-order DPLL is simulated and analyzed for five different types of noise: Gaussian, contaminated Gaussian, Laplace, Middleton Class A , and generalized exponential. Comparisons are made for signal-to-noise ratios ranging from 0 to - 10 dB. It is found that the ZNL estimators provide performance superior to that of the other robust estimators, especially for noise densities that are heavy tailed.