A new estimation algorithm for AR signals measured in noise

Abstract

Albeit several least-squares (LS) based methods have been developed for noisy autoregressive (AR) signal identification, none is "in closed form", in that an iterative procedure is needed for estimating the AR parameters and the measurement noise variance alternately. A new formulation with respect to the measurement noise variance is presented, leading to the development of a new estimation algorithm for noisy AR signals. In addition to the eminent algorithmic difference from its predecessors, the developed algorithm achieves a better estimation accuracy while requiring an almost identical amount of computation.