Sparsity-based Time-Frequency Analysis for Automatic Radar Waveform Recognition

Abstract

In this paper, we develop a novel pre-processing algorithm to achieve effective signal denoising for improved recognition of noisy radar signals. The algorithm is considered in the instantaneous autocorrelation function domain in which time or lag slices are converted to a Hankel matrix, and an atomic norm-based method is applied to mitigate the impacts of noise. Cross-terms are suppressed by using a time-frequency kernel, such as the Choi-Williams distribution, and a sparsity-based reconstruction technique is utilized to obtain a high-resolution time-frequency distribution of the radar waveforms. Simulation results verify the effectiveness of the proposed method. The proposed denoising algorithm for radar waveform recognition enables a substantial increase of the overall successful recognition rate from 90.24% to 97.76%.