Improvement on the estimation of covariance matrices by incorporating cross-correlations

Addresses the problem of improving the estimate of a covariance matrix from one set of multivariate random processes when there exist non-zero cross-correlations with another set of random processes. The improvement is obtained by linearly combining the first set's sample covariance matrix with covariance matrices predicted via the cross-correlations. The superiority of the proposed method is demonstrated by an application to spatial smoothing for the DOA estimation of coherent narrowband signals using a uniform linear array.<>