A Wide Dynamic Range Multichannel Spectral Estimator

A multichannel extension of the AR matrix prewhitened power spectral density estimator, which was originally developed for a single channel, is proposed. In order to make the extension, the Cholesky decomposition of the inverse autocorrelation matrix for a multichannel autoregressive process is discussed and the autoregressive model order selection for a multichannel process based on the exponentially embedded families criterion is introduced. The asymptotic mean and variance of the proposed estimator are derived. Compared to a filter-based autoregressive prewhitened multichannel power spectral estimator, the new estimator has less bias, i.e. higher resolution, and less overall mean square error for short data records due to the amelioration of end effects by the matrix prewhitener. It can serve as an excellent multichannel spectral estimator for processes exhibiting a wide dynamic range. Simulation results are given which show the advantage of the new estimator over a variety of common multichannel power spectral density estimators.