Evaluation of a two-step clustering method for radar pulse deinterleaving

In electronic warfare, deinterleaving is the task which sorts radar pulses in order to separate radar waveforms, without prior knowledge. This paper intends to evaluate two versions of a two-step deinterleaving method. It starts with a RF-PW clustering, then tries to merge some clusters by defining a distance related to their set of pulse times of arrival. After a parameter optimization, the method is evaluated with five metrics on a dataset consisting of simulated frequency-agile and PRI-modulated signals, and on a real data observation.