MUSIC-group delay based methods for robust DOA estimation using shrinkage estimators

摘要

TheMUSIC-group delay spectrum has been hitherto used for high resolution direction of arrival estimation using a limited number of sensors. In this work we discuss the significance of the MUSIC-group delay spectrum for robust direction of arrival estimation using shrinkage estimators. The MUSIC-group delay spectrum computed using shrinkage estimation of the covariance matrix, is found to be robust under reverberant conditions. A minimum phase analysis of the proposed method is also presented. Results of CRLB analysis on MUSIC based methods are also illustrated. A filter and sum beamformer is then trained using the DOAs estimated from the MUSIC-group delay spectrum using shrinkage estimation. Experiments on distant speech recognition are conducted in reverberant conditions using this filter and sum beamformer on the TIMIT and the MONC databases. The experimental results indicate a reasonable improvement over several conventional beamforming methods.