A multichannel cross-spectrum density estimator based on canonical correlation analysis

In this work, the problem of multichannel cross-spectrum density (MCSD) estimation is studied. Based on a multichannel data model, the classic periodogram based MCSD estimator and the minimum variance (MV) based MCSD estimator are tested for cross-spectrum density estimation. Our major contribution in this work is a canonical correlation analysis (CCA) based MCSD estimator, relying on the inherent maximization arguments of CCA. It is demonstrated through a model based multichannel simulation example that the newly proposed CCA-MCSD estimator is of high frequency resolution as well as good sidelobe suppression properties, and can serve as an excellent candidate for MCSD estimation.