Anti-main lobe suppression jamming using signal separation network

The main lobe suppression jamming seriously damages radar detection by covering the target echo in multiple domains. When the target and the jammer are in the same direction, the traditional blind source separation-based methods are ineffective. To effectively suppress jamming and achieve target detection, this paper proposes an end-to-end framework implemented by a complex-valued dual-path convolutional shrinkage time-domain signal separation network (CVDPCS-TssNet) to automatically separate mixed signals and recover target signals for jamming suppression. The jamming suppression framework is designed with an encoder-separation-decoder structure. Firstly, the encoder converts the received mixed signal into a representation in a separable feature domain. Then, the separation module learns the optimal separation weights in the feature domain to extract the jamming and target signal representations. Finally, the weighted signal representations are recovered into independent jamming signals and target signals by the decoder. Utilizing the advantage of the integrated multiple network components in signal sequence modeling and robust weak information representation, the CVDPCS-TssNet uses only the single-channel observed signal to recover the time-domain target signal. It is applicable to the scenario where the target and jammer are in the same direction. Experimental results on noise modulation jamming verify that the proposed method is superior in signal separation, jamming suppression, target detection performance and robust to varying signal-to-noise ratios and jamming-to-signal ratios.