A new decision-directed algorithm for nonstationary priors

A new decision-directed (DD) procedure is introduced to address the binary detection problem for nonstationary priors. Using a martingale approach, an explicit, recursive finite-dimensional filter is applied to the problem of computing the conditional expectation of the stochastic rate of the discrete-time point process (DTPP) representing the output of the detector. Conventional DD techniques strictly apply only when the rate is constant. The approach described herein allows the rate to be modeled as a finite-state Markov chain, and allows a further extension to address the situation where the Markov transition matrix possesses an observation-dependent component.