In complex electromagnetic environment, pulse signals from different emitters are highly overlapped in time, spatial and frequency domains. Traditional methods perform poor in deinterleaving mixed signals having similar pulse parameters, arrival directions and frequency sets. In this paper, a deinterleaving method employing the clustering and the machine leaning is proposed to solve this problem. The proposed method first clusters pulse signals based on multiple parameters, and trains a supervised learning model using features representing pulse sequences' variation trend. The model is used to predict whether different clustered signal groups belong to the same emitter. As a result, mixed pulse signals are deinterleaved into different emitter clusters. A verification test is presented to measure the performance of the proposed method.
展开▼
