Deinterleaving Pulse Trains via Interleaved Markov Process Estimation

Abstract

A pulse train deinterleaving framework based on interleaved Markov process (IMP) estimation is presented. The approach models interleaved pulse trains as a collection of Markov processes interleaved by a Markov switch. Deinterleaving is performed through learning an IMP representation of the interleaved pulse trains. We apply a recent unsupervised information-theoretic technique that infers the underlying Markov processes by minimizing a penalized maximum likelihood (PML) entropy cost function. The technique is able to associate groups of disparate clusters in the pulse parameter space that are produced by the same parameter-agile emitter. This enables the correct determination of the number of emitters in scenarios where traditional multi-parameter clustering techniques produce many more clusters than there are actual emitters. Pulse timing information is incorporated to the extent that it is reflected in pulse ordering; in contrast to traditional time-of-arrival analysis techniques, the approach does not require precise timing resolution and can accommodate highly complex pulse repetition interval (PRI) patterns. An experimental demonstration and performance characterization on synthetic pulse train datasets is provided.