Adaptive data reduction for signals observed in spatially colored noise. (Q1575765)

This paper treats adaptive data reduction in array processing for the spatially colored noise case. The purpose is to reduce the computational complexity of the applied signal processing algorithms by mapping the data into a space of lower dimension by means of a linear transformation. The derived transformation preserves the Cramér-Rao bounds for the parameters of interest. We discuss ways to implement the transformation and show that it suffices to estimate the array covariance matrix instead of the noise covariance matrix in the design process of the optimal transformation. Computer simulations are given that illustrate the problem of interference from out-of-band-sources that result when a beamspace transformation is designed to focus on a particular sector. When compared to other methods, such as the method of spheroidal sequences, these simulations show that significant improvements are gained in terms of mean-square error performance.