Denoising Convolutional Neural Network for Wideband Frequency Modulation Signals Based on Microwave Photonic Down-Conversion

The denoising convolutional neural network (DnCNN) algorithm is introduced into the microwave photonics down-conversion system to alleviate the system noise and reduce the number of radar false alarms, which is proposed and demonstrated. The noise characteristics of the broadband signal after fast Fourier transform are learned by training the network, and the noise reduction and enhancement of the signal are realized. The simulation results show that the peak-to-floor ratio (PFR) of the triangular frequency modulation down-conversion signal is increased by about 23.2 dB, which greatly improves the detection ability of the radar, especially the weak target detection ability.