Real-Time Adaptive STFT-CNN Fusion for EW Passive Guidance
Somnath Mondal, Harminder Singh Johar, Sabbineni Srinivasa Rao · 2024
Passive guidance systems in Electronic Warfare (EW) encounter challenges stemming from modern radar techniques like Pulse-on-Pulse (POP) signals and environmental interference. Detecting intra-pulse modulated signals necessitates real-time Short-Time Fourier Transform (STFT), particularly crucial due to maneuvering challenges when the EW system is in close proximity to the radar. To address the computational latency inherent in radar parameter extraction, our approach integrates Machine Learning (ML) based signal recovery with a Convolutional Neural Networks (CNN) empowered STFT solution. This framework encompasses signal recovery, STFT processing, and simultaneous input into a CNN model for predictive analysis. Initially, test data undergoes STFT and prediction logic, with discrepancies prompting data inclusion into the training dataset. This adaptive strategy leverages processing time for analysis, augmenting fault tolerance without additional hardware. Our proposed design surpasses other software-based CNN implementations by 2X and 19.27X, respectively. Furthermore, our STFT hardware implementation achieves a 2X higher clock rate and 5.33X lower DSP utilization compared to other FPGA implementations.