CNN-Based DMRS Pattern Optimization for 5G Vehicle-to-Everything Communications

In this paper, we optimize the demodulation reference signal (DMRS) pattern of the physical sidelink shared channel (PSSCH) based on a deep learning approach to maximize the channel estimation performance for 5G vehicle-to-everything (V2X) communication systems. For the DMRS optimization, we design a convolutional neural network (CNN) which classifies the optimal DMRS pattern index for the given received signals. It is shown that the proposed CNN structure achieves 91% accuracy for the optimal DMRS pattern index. The simulation results verify that the proposed scheme achieves 4 dB performance gain in terms of mean square error (MSE) compared to the legacy DMRS pattern.