Deep Multi-Modal Neural Receiver for 6G Vehicular Communication

Deep Learning (DL)-based neural receiver models are used to jointly optimize the physical (PHY) layer of receiver for cellular vehicle to everything (C-V2X) system in next generation (6 G) communication. However, there has been no exploration on how varying training parameters affect the model's efficiency. Moreover, a comprehensive evaluation of its performance on multi-modal data remains largely unexplored. To overcome these issues, we propose a multi-modal neural receiver to optimize Bit Error Rate (BER) of vehicle to network (V2N) scenarios in 6 G network. The main contributions of our proposed neural receiver is that it is not scenario-limited and is multi-modal. The proposed neural receiver finds the best-fit channel model by training at vast and varying channel parameters. Therefore, it can be applied to large scale scenarios. The proposed model gets signal in frequency domain at the base station (BS) as input and generates optimal log likelihood ratio (LLR) at the output. It estimates the channel parameters based on the received signal, equalizes and demodulates the higher order modulated signal. The multimodality of the proposed receiver has been demonstrated by testing it across diverse V2X data flows e.g., image, video, GPS, lidar cloud points and radar detection signal. The simulation results clearly indicates that our proposed multi-modal neural receiver achieves approximately 25 % improvement to achieve minimal BER compared to the state-of-the-art of DeepRx and 60 % improvement to achieve the same compared to Baseline.