Machine Learning Approach to Chirp Rate Estimation of Linear Frequency Modulated Radars

The detection and parametric estimation of low-SNR radar signals, particularly linear frequency modulated (LFM) radar signals, is a problem of considerable interest. In prior work, this problem has been investigated using various signal processing techniques, such as maximum likelihood estimation, fractional Fourier transform and Wigner-Ville-based methods, to analyze the signal parameters of a complex linear frequency modulated signal. Other work has focused on applying deep learning to automatically recognize various radar waveform types and their features, such as linear frequency modulation (LFM), Barker code and rectangular waveforms. In this paper, we investigate this problem from a machine learning perspective for multiple LFM radar signals given a priori information. We explore the use of naive Bayes, support vector machine and neural network classifiers to identify the LFM chirp rate, out of a set of known chirp rates, from a specific radar emitter under varying SNR conditions. Simulation results demonstrate the viability of this technique to identify the radar LFM mode in very low signal-to-noise ratio conditions down to -20 dB where using existing approaches (e.g., Wigner-Ville) fail.