Deep-learning based noise-resistant broadband signal recovery for fiber-connected radar networks

A deep learning based broadband radar signal recovery method for fiber-connected radar networks is proposed, which is implemented by using a residual convolutional neural network (RCNN) to deal with the defects of analog fiber transmission systems. In order to enhance the robustness of the proposed scheme, a method for training the network with data superimposed with extra noise is also demonstrated, based on which the RCNN is able to compensate for the nonlinear spurs, noises, and signal distortions in low-noise conditions. In the experiment, signal recovery of 3-GHz bandwidth radar echoes after 8-km fiber transmission is realized with the average mean square error (MSE) of the waveforms reduced by ~19 dB, and the dynamic range for maintaining a good pulse compression performance is increased by 35 dB.