Interleaved Hidden Markov Processes Inference for Deinterleaving Radar Pulse Sequences

Abstract

The Hidden Markov Process (HMP) has been widely used to model radar pulse sequences. For the radar signal deinterleaving task in an electronic reconnaissance system, the intercepted radar pulse sequences are assumed to be interleaved hidden Markov processes (IHMP). In this context, this paper proposes a generative model to represent the IHMP and reformulates the deinterleaving problem as a posterior inference task. To compute the posterior probability, we first design an exact inference algorithm. However, due to the combinatorial nature of the hidden state representation, exact inference becomes computationally intractable. To address this limitation, we further develop a sampling-based method and two variational-based methods, yielding tractable solutions for the posterior computation. Finally, a theoretical lower bound on the error probability is derived based on the likelihood ratio test, with the proposed methods shown to get reasonably close to the bound. Simulations on diverse radar pulse signal datasets verify that variational inference with a structured approximation delivers a superior balance between deinterleaving accuracy and computational efficiency, making it a promising alternative to exact inference methods and search-based methods.