Dual-Path GRU with Encoder-Decoder Optimization: A Resource-Efficient Framework for Airborne Signal Demixing

An unmanned aerial vehicle (UAV) is a partially or fully autonomous aircraft that has been extensively utilized in target detection, disaster rescue, and other complicated systems. However, the associated unknown airborne avionics signal interference under complicated electromagnetic environments would greatly impact the UAV's normal operation. Given the limited computational capabilities of UAVs and the high demands for real-time processing, it is essential to develop lightweight algorithms. In this work, we have proposed a novel Dual-Path Gated Recurrent Unit (DP-GRU) based lightweight algorithm for the airborne avionic signals separation problem. We constructed a novel block named DP-GRU, which combines the Gate Recurrent Unit model with dual-path processing and improved the model's efficiency through an optimized end-to-end encoding-decoding learning methodology, providing a new approach to solving the airborne avionic signals separation problem. Results show the proposed algorithm has the best performance on the constructed dataset, which Scale-Invariant Signal-to-Noise Ratio (SI-SNR) and SI-SNR Improvement (SI-SNRI) could reach 18.52dB and 26.53 dB, respectively. Compared with classical deep learning-based signals separation algorithms, the proposed DP-GRU module reduces the model size by approximately 50%.