Detection of coherent signals using weighted subspace smoothing

A new approach to detecting the number of coherent signals incident upon a uniformly-spaced linear array is presented. The approach combines a modified spatial smoothing scheme and a modified MDL criterion. The modified spatial smoothing scheme, referred to as "weighted subspace smoothing," is actually a generalization of the "post-smoothing" approach proposed by Krim and Proakis in 1994. It is shown that the noise eigenvalues obtained with weighted subspace smoothing are more accurate than those obtained with the original spatial smoothing. We thus attempt to improve the detection performance using these more accurate eigenvalues. A novel modification of the minimum description length (MDL) criterion is proposed to accomplish this. Computer simulations are presented to study the performance improvement of this new approach.