Waveform Recognition in Multipath Fading using Autoencoder and CNN with Fourier Synchrosqueezing Transform
Gyuyeol Kong, Minchae Jung, Visa Koivunen · 2020
In this paper the problem of recognizing radar waveforms is addressed for multipath fading channels. Waveform classification is needed in spectrum sharing, radar-communications coexistence, cognitive radars, spectrum monitoring and signal intelligence. Different radar waveforms exhibit different properties in time-frequency domain. We propose a deep learning method for waveform classification. The received signal is first equalized to mitigate the effect of multipath fading channels by using a denoising auto-encoder (DAE). Then, the equalized signal is processed with Fourier synchrosqueezing transform that has excellent properties in revealing time-varying behavior, rate of, strength and number of oscillatory components in signals. The resulting time-frequency description is represented as a bivariate image that is fed into a convolutional neural network. The proposed method has superior performance over the widely used the Choi-Williams distribution (CWD) method in distinguishing among different radar waveforms even at low signal-to-noise ratio regime.