The Design of Optimal Processors for Arrays with Non-Gaussian Noise Inputs

Abstract

Although the derivation of an expression for the maximum likelihood estimator for detecting signals embedded in noise is well known, the complexity of the result has prevented its general use except for Gaussian noise fields. This paper assumes that the input signals and noise are stationary and that the signal-to-noise ratio is small. The theory of Symmetry Groups is then used to simplify the expression for the optimal processor. In particular, it is shown that using the covariance matrix to optimize the weights of a classic sum and square processor is not optimum unless the input noise is Gaussian. For such cases, practical alternative processors are given, which in some situations are easier to implement than the non-optimal processors commonly used.