Adaptive Constant False Alarm Rate Detector Based on Long Short-term Memory Network

Abstract

To solve the problem of degradation of detection performance of adaptive constant false alarm rate (CFAR) detectors due to low accuracy of environment recognition, an automatic clipping adaptive CFAR detector based on long short-term memory (LSTM) network is proposed. LSTM network is used to recognize the environmental type information contained in radar echo signals, and the appropriate detector is determined based on the recognition results. When there are interferences in both the leading and lagging reference windows, the interferences are clipped, and an ordered statistics CFAR detector is used to detect the target. Simulation results show that the designed adaptive CFAR detector, compared to the variability index CFAR detector, achieves an average improvement of 0.31% in detection probability in homogeneous environment. In the environment with interferences in a single-sided reference window, the average improvement in detection probability is 5.43%. In the environment with interferences in both the leading and lagging reference windows, the average improvement in detection probability is 41.57%. The automatic clipping adaptive CFAR detector based on LSTM network can more accurately recognize background environments and clipping interferences when interferences exist in both the leading and lagging reference windows, so its detection performance can be enhanced.