The generalized ACM-music without estimation of the number of sinusoidal components

The family of MUSIC estimators is often used in the case of the spectral analysis of sinusoidal signals. One of those estimators, called ACM-MUSIC, is based on the direct estimation of the vectors which constitute the signal and noise subspaces thanks to the autocorrelation matrix. In this paper, we suggest using the iterated power method in order to improve the estimation of those vectors. Besides, instead of estimating the number of signal components with criteria such as AIC and MDL, we resort to a judicious weighting which enables us to separate the signal and noise subspaces. The results show frequency spectra close to those obtained by the MUSIC estimator to which has been associated the decomposition of the autocorrelation matrix into eigenvalues. However, the numerical calculation of the proposed method can be lower, and the choice of the number of sinusoidal components can be avoided. That approach can also be applied to the ROOT-ACM-MUSIC estimator, which is a direct extension of the ACM-MUSIC.