Abstract: Due to the weak energy and low signal-to-noise ratio of radar echo signals, it is difficult to obtain the frequency domain characteristics of high quality medium frequency signals, which often leads to large measurement error. To address this issure, a signal time-frequency analysis method combining synchrosqueezing wavelet transform (SST) with median filtering is proposed to achieve strong random noise suppression. Firstly, a simulation model of 24 G frequency modulated continuous wave (FMCW) radar echo signal is constructed. By adding Rayleigh clutter and Gaussian random noise, the frequency analysis diagrams from short-time Fourier transform (STFT), continue wavelet transform (CWT), and SST are compared, and the instantaneous frequency identification accuracy is proposed as an evaluation index I. Secondly, STFT, CWT and SST algorithms are used to analyze the time spectrum of the echo signals with an actual distance of 6 meters. The research results indicate that SST can better highlight the instantaneous frequency characteristics and compress the random noise into low-energy short lines and strips. Based on the distribution characteristics, SST combined with median filtering can effectively suppress the noise, with I value decreasing by 0.21, 0.55 and 0.71 compared with SST, STFT and CWT, respectively. Compared with the traditional SST, STFT and CWT, the instantaneous frequency recognition accuracy of the proposed method is highest, while the peak and valley values of the reconstructed signals through inverse transform can be well maintained, at the same time, and the detection accuracy error at different distances is also the smallest.

Key words: millimeter wave radar, synchrosqueezing wavelet transform, median filtering, time frequency analysis, identification accuracy, random noise