Adaptive linear prediction of autoregressive models in the presence of noise

Abstract

An alternative algorithm is developed for adaptive linear prediction of autoregressive (AR) models in the presence of noise. Central to this algorithm is that the variance of the corrupting noise, which determine the bias in the standard least-squares (LS) parameter estimator, is estimated by using the expected LS errors under the assumption of the known ratio between the driving noise variance and the corrupting noise variance. Then the adaptive linear prediction (ALP) algorithm is established via the bias correction principle. While achieving estimation unbiasedness, the proposed ALP algorithm exhibits computational and algorithmic advantages over the previously developed LS based algorithms.