Prewhitening of Colored Noise Fields for Detection of Threshold Sources

摘要

This report presents an algorithm for the estimation of noise correlations for anarray of sensors. The algorithm assumes a mixed spectra model composed of discrete sinusoidal sources and continuous noise components. A spatial AR process models the continuous component. Previous methods use spatial AR models with real coefficients, restricting the utility of the model to noise sources that impinge on an array at broadside. The complex formulation presented here solves the general problem of arbitrarily oriented noise sources. The technique uses a gradient algorithm for maximization of a likelihood functional to solve for the complex AR coefficients. Once the algorithm determines the noise covariance matrix, prewhitening techniques allow detection of threshold sources. The multiple signal classification (MUSIC) direction finder when applied to the prewhitened observed correlation matrix illustrates the usefulness in detecting low signal-to-noise ration (SNR) sources. Computational examples used the standardized test case (STC), a realistic model with a challenging range of SNR and difficult source locations. AR Model, Colored noise field, Mixed spectra model, MUSIC, Noise field, Prewhitening, SNR, Standardized test case, Threshold sources.