Adaptive subspace selection using subband decompositions for sensor array processing

This paper considers the steady state behavior of adaptive processors. The use of a time-varying, minimum mean-square error searching, tree-structured filterbank is introduced to select the best subspace, in a least squares sense, for the adaptive processor with a predetermined number of adaptive coefficients. With the realistic assumption that the quantity of adaptive coefficients is the major computational restriction on the adaptive processor, it is shown that the time-varying filterbank allows the processor to achieve a lower minimum mean-square error than that which could be realized otherwise. The simple transform-domain least mean-square algorithms may be utilized to improve the dynamic behavior of the adaptive processor. These ideas are extended to the multidimensional filter case, and an example is presented via simulation for sensor array processing utilizing the generalized sidelobe canceller.<>