A reduced-complexity quadratic structure for the detection of stochastic signals

Quadratic detection of discrete-time stochastic signals in additive stationary Gaussian noise is considered. A banded-quadratic detector structure is introduced to reduce the multiplicative complexity and data storage requirements of the optimum full-quadratic detector, and the optimization of this reduced-complexity structure is studied. The issue of performance versus complexity is explored for the specific problems of detecting wide-sense Markov and triangularly correlated signals in white noise, with the conclusion that performance of the reduced complexity detector can be very close to optimum if an adequate quadratic-form bandwidth is chosen.