Detection Performance of a Modified Generalized Likelihood Ratio Processor forRandom Signals of Unknown Location

摘要

A search region of N bins in time-frequency space contains either noise only ornoise plus a signal which occupies a subset of M bins of unknown location. The unknown signal strengths are estimated by means of a maximum likelihood procedure modified by a breakpoint, in order to suppress the weaker estimates. The stronger signal estimates are then used to implement a modified form of generalized likelihood ratio processor. The highly nonlinear character of this generalized likelihood ratio processor debilitates against an exact analysis of its performance. To circumvent this difficulty, all the nonlinear components are replaced by piecewise-linear devices which best match the given characteristics according to minimum mean square error. The false alarm and detection probabilities of this piecewise-linear processor are then derived exactly. Receiver operating characteristics for a general set of parameter values are presented and verified by means of simulations.