High-Accuracy Radar Parameter Estimation Under Low SNR Environments

This paper proposes a high-accuracy radar parameter estimation method for extremely low signal-to-noise ratio (SNR) environments, integrating deep learning and connected component analysis (CCA). Accurate estimation of parameters such as time of arrival, pulse width, pulse repetition interval, bandwidth, and carrier frequency is critical for various applications, but remains challenging under high noise conditions. The proposed method consists of three key steps. First, frequency-domain pulse detection isolates radar signals from noise-contaminated samples. Second, time-frequency analysis using short-time Fourier transform is followed by UNet-based denoising and CCA to suppress noise and enhance signal features. Finally, edge-based parameter computation is applied to extract signal boundaries and estimate parameters precisely. Simulation results show that the method achieves approximately 3 dB improvement in time-domain estimation accuracy and lower root mean square error in frequency-domain parameters compared to conventional methods, across SNR levels from –20 dB to 10 dB. Test-bed experiments using Universal Software Radio Peripheral and GNURadio further validate its robustness, especially at –5 dB SNR. These findings confirm the method’s practical viability for reliable radar signal processing in challenging low-SNR environments.