SVD based information theoretic criteria for detection of the number of sinusoids

Following the information theoretic approach to model selection, criteria for detecting the number of sinusoids are developed. The criteria are well matched to singular-value-decomposition (SVD-) based methods of spectral estimation in that the additional computations required over and above those needed for the SVD calculation are marginal. The detection performance of these criteria is compared with that of the J.J Fuchs (1988) method. The results show that the minimum description length (MDL)-based criterion, developed in the paper, and the Fuchs method perform similarly. However, the latter method involves substantially more computations than the former.<
>