Multi-Station Cooperative Radar Signal Deinterleaving Based on Deep Information Fusion

Multi-station radar signal deinterleaving, which aims to distinguish signals from different radar targets with multiple electronic reconnaissance stations, can get more discriminative features compared with single-station deinterleaving. However, it is still a challenging task to find the matched Pulse Description Word (PDW) pairs to obtain the time differences, especially for close-parameter and imbalance signals. To address this problem, it is the first time to propose a multi-station cooperative pulse signal deinterleaving method based on deep information fusion, which can better explore the relationships between multi-station data in a high-dimensional feature space. Firstly, the method initially incorporates position encoding for the pulse sequences from multiple stations, using the arrival times of the pulses. Secondly, it utilizes the self-attention mechanism to fuse pulses from different stations. Finally, U-Net is used to perform deep feature extraction on the associated information and map features to the corresponding radar labels, thus completing the radar signal deinterleaving. Simulation experiments demonstrate that this method effectively deinterleaves closeparameter and imbalance signals, achieving higher deinterleaving accuracy compared to single-station and traditional multi-station cooperative deinterleaving methods.