A prior knowledge-driven reference signal extraction network towards measured data in DRM-based passive radar

Abstract

We propose a novel intelligent reference signal extraction system for DRM-based passive radar. By fully using prior knowledge of the received signal, a pilot conditional reference signal extraction network (PCRSENet) is designed to replace both conventional channel estimation and signal equalization. The PCRSENet estimates the channel state information (CSI) implicitly and outputs reference signals directly. Meanwhile, what makes it challenging to utilize deep learning in reference signal extraction of passive radar is not only the limited measured data but also the truth transmission signals and CSI being unknown. So, the least square (LS) estimator-aided training sets generation method is employed, which obtains the CSI from measured data by LS estimator as prior knowledge and then simulates DRM signals as accurate labels of training sets. We evaluate the proposed method on measured data, and the experimental results show that the proposed method works effectively for target detection in DRM-based passive radar with different working parameters when the training and test sets are in the same transmission mode.