Channel Estimation in C-V2X using Deep Learning

Channel estimation forms one of the central component in current Orthogonal Frequency Division Multiplexing (OFDM) systems that aims to eliminate the inter-symbol interference by calculating the Channel State Information (CSI) using the pilot symbols and interpolating them across the entire time-frequency grid. It is also one of the most researched field in the Physical Layer (PHY) with Least-Squares (LS) and Minimum Mean Squared Error (MMSE) being the two most used methods. In this work, we investigate the performance of deep neural network architecture based on Convolutional Neural Networks (CNNs) for channel estimation in vehicular environments used in 3GPP Rel.14 Cellular-Vehicle-to-Everything (C-V2X) technology. To this end, we compare the performance of the proposed Deep Learning (DL) architectures to the legacy LS channel estimation currently employed in C-V2X. Initial investigations prove that the proposed DL architecture outperform the legacy C-V2X channel estimation methods especially at high mobile speeds.