OPTIMALLY WEIGHTED MUSIC FOR FREQUENCY ESTIMATION

This paper introduces a weighted MUSIC (multiple signal classification ) algorithm for estimating the frequencies of sinusoidal signals from noise-corrupted measurements. The large-sample variance of the weighte d MUSIC is determined, and the optimal weighting matrix which minimize s that variance is derived. The optimally weighted MUSIC is shown to p rovide more accurate frequency estimates than the unweighted MUSIC and ESPRIT (estimation of signal parameters via rotation invariance techn iques).