LFM Signal Detection and Estimation Based on Deep Convolutional Neural Network

Abstract

Linear frequency modulation (LFM) signal detection and estimation are important for radar, communication, or spectrum analysis etc.. As the generalized form of Fourier transform (FT), Fractional FT (FRFT) has good energy aggregation ability for LFM signal and can reflect the Doppler variation, which is suitable for LFM signal detection and estimation. However, it needs two-dimensional parameters searching and for multiple signals it requires searching one by one and easily affected by strong signals with poor resolution. In this paper, the convolutional neural network (CNN) is applied for replacing the FT and FRFT and used for signal frequency signal and LFM signal detection and estimation. The pre-trained CNN model can establish the relations among various single frequency signal or LFM signal and the two dimensional parameters domain. By simulation, it is found that the CNN based method can also achieve the function of FRFT and has the advantages of high precision and resolution. And it is proved that the CNN based method can achieve good recognition performance even at lower signal-to-noise ratio (SNR) combined with the denoising method. The proposed method would provide a novel solution for radar moving target detection, as well as speech intelligent signal processing, sonar signal processing, etc..