On the Accuracy of Frequency Determination by an Autoregressive Spectral Estimator

Abstract

AbstractThe accuracy of frequency determination by a least squares technique for an autoregressive spectral estimator is studied and compared with the Fourier method. Using numerical tests the probability distribution function of the peak location is calculated. The autoregressive filter order is optimized in the sense of minimum variance of the peak location. Simple sinusoidal signals with additive Gaussian noise are considered and the effect of other components is only indicated. Generally, a filter order between 1/3 and 1/2 of the total data number and a not very dense data sampling, gives the most stable spectrum. The results are numerical.