EVALUATION OF MAXIMUM ENTROPY METHOD OF SPECTRUM ESTIMATION

Abstract

The parametric models autoregressive (AR)/AR-moving average (MA)/MA are sometimes not capable of finding out the power spectral densities of random sequences. Under such circumstances, the non-parametric methods outperform the parametric ones because of the sensitivity of the latter to model specifications. The maximum entropy method (MEM) is regarded as the non-parametric method of spectrum estimation; it suggests one possible way of extrapolating the autocorrelation sequence so that a more accurate estimate of the spectrum can be obtained with better resolution. This paper investigates the work of realizing MEM method and evaluating its performance with minimum variance method.