Sinusoidal frequency estimation by approximate MUSIC method

Abstract

Two efficient methods avoiding eigenvector computation are proposed for approximating the signal subspace in terms of the Fourier transform. The resulting approximations are used to substitute for the signal eigenvectors in MUSIC. The proposed methods perform almost the same as MUSIC at high SNRs and provide often clearly better results at low SNRs. They seem to be more robust than MUSIC against overestimation of the number of sinusoids.<>