Performance Evaluation of FPGA-Based LSTM Neural Networks for Pulse Signal Detection on Real-Time Radar Warning Receivers.

This article focuses on evaluating the implementation of Long Short-Term Memory (LSTM) Artificial Neural Network (ANN) for radar signal detection on Field Programmable Gate Array (FPGA) hardware, comparing it with the conventional Constant False Alarm Rate (CFAR) method used in real-time Radar Warning Receiver (RWR) systems. While LSTM-based detection demonstrates superior accuracy at low signal-to-noise ratio (SNR) levels (e.g., 94% success at -5 dB SNR), the FPGA implementation exhibits high latency, significant resource consumption, and requires prior training with knowledge of the noise environment, limiting its practicality for real-time applications. In contrast, CFAR offers low latency, adaptability without prior training, and efficient FPGA resource usage, making it more suitable for real-time radar signal detection despite lower performance at very low SNR. The study concludes that although LSTM shows promise for offline analysis, it is currently not a viable alternative to CFAR for real-time radar warning receivers.