Prewhitening of colored noise fields for detection of threshold sources

Le Cadre's (1989) algorithm for the estimation of parametrically defined noise correlations in a mixed spectra model is studied. A spatial autoregressive (AR) process with complex coefficients models the continuous noise component. The complex AR formulation solves the general problem for continuous noise sources that are arbitrarily oriented (non-broadside) to a line array of sensors. The parametric definition allows a gradient algorithm which maximizes a properly constructed likelihood functional to be formulated analytically, thus avoiding computationally expensive grid searching. MUSIC, when applied to the prewhitened observed correlation matrix, illustrates the performance enhancement in detecting low SNR sources. Computational examples utilize the standardized test case (STC), a realistic model with a challenging range of SNRs and difficult source locations.<>