Recursive Neural Networks as an Hypothesis Tester for Deinterleaving Repetitive Sequences

for Deinterleaving Repetitive Sequences G. Noone, S. D. Howard Electronic Warfare Division, Defence Science and Technology Organisation PO Box 1500 Salisbury, South Australia 5108 email: [email protected] & [email protected] Abstract| Conventional algorithms for deinterleaving repetitive sequences, such as radar pulse trains, experience great di culties when two or more of the sequences are signi cantly jittered. Such algorithms are usually variations of simple sequence search type techniques which are unable to intelligently test whether the captured sequence is actually periodic or not. In addition, such scenarios tend to lead to overwhelming ambiguity and fragmentation di culties when using traditional techniques. In this paper we propose a modi ed sequence search technique combined with a novel recursive neural network approach. The sequence search forms the hypothesis that a given chain of events is periodic and the neural network is used to intelligently test this hypothesis. This approach has been used to allow the satisfactory deinterleaving of multiple jittered radar pulse trains.