基于多域时序解耦注意力网络的雷达信号分选算法OA

To address the problem that radar emitter signals adopt increasingly complex modulation modes in modern electronic warfare and their essential attributes cannot be effectively characterized by a single feature domain,the Multi-domain Temporal Disentangled Attention Network(MD-TDAN)is proposed as a radar signal sorting algorithm.The method first constructs a six-dimensional holographic feature matrix in the data preprocessing stage.This matrix includes time-domain orthogonal components,frequency-domain spectral components,and instantaneous amplitude-phase features.Subsequently,a multi-view decoupling attention module is designed,utilizing a stride-wise self-attention mechanism to capture the interval patterns of pulse repetition interval(PRI)parameters,effectively addressing the high computational complexity of long sequence signals and the generalization issue of arbitrary PRI adjustments.Finally,the method fuses local texture features and global semantic information through a deep residual convolutional network to achieve high-precision sorting of 8 types of complex system radar signals,such as conventional,staggered,jittered,and sliding signals.Experimental results show that MD-TDAN achieves an average sorting accuracy of 96.25%for 8 types of radar signals,and exhibits excellent robustness and generalization ability in low signal-to-noise ratio and complex modulation environments.