Pulse train deinterleaving: algorithms and cost criteria

Consider the problem where pulse trains transmitted from a known number of sources are received on a single communications channel. These pulses are corrupted with noise. The deinterleaving problem is to determine which source contributed which pulse and the periods and phases of each source. This paper explores the performance of a number of deinterleaving algorithms. We propose an alternative to the existing forward dynamic programming (FDP) technique: simulated annealing (SA). It can use either the same cost function as for FDP, or an L₁ or L₂ norm output error cost function. We also investigate modelling the noise by heavy-tailed distributions, in addition to white Gaussian noise (WGN)