The cumulant based MUSIC and its use in frequency estimation

A technique for estimating the frequency components of a signal with additive colored noise is presented. This technique is a cumulant-based MUSIC-type method that effectively operates based on good estimates of the cumulants and the separation of signal and noise subspaces of the fourth-order cumulant matrix. The cumulant-based MUSIC offers the same performance as the noiseless case and provides infinite resolution. Results for simulated colored noisy data with decreasing signal-to-noise ratio values are presented, and a comparison with the correlation-based MUSIC method is made to demonstrate the validity of proposed technique. Due to its insensitivity to Gaussian and non-Gaussian processes, the cumulant-based MUSIC provides better performance and higher frequency resolution than the correlation-based MUSIC.<
>