MVDR Spectral Estimation by Spectral Peak Dichotomous Search

Canonical correlation analysis (CCA) and minimum variance distortionless response (MVDR) are typical nonparametric spectral estimation methods based on matched filterbank theory. To avoid the common signal mismatch problem, an algorithm combining CCA and MVDR is proposed for peak frequency modification. It consists of two stages: a coarse-peak search in CCA spectrum, and followed with a fine-peak search using dichotomous search strategy in MVDR spectrum. Furthermore the new algorithm is extended to the magnitude squared coherence (MSC) spectral estimation. Simulations show that the peak frequency estimation accuracy is improved simply and efficiently, with only slight computation increased.