When DSRC Meets With C-V2X: How to Jointly Predict and Select Idle Channels?

The Dedicated Short Range Communications (DSRC) terminator requires extra idle channels to ensure reliable broadcasting of safety information, especially when traffic congestion occurs under high vehicle density. In the DSRC and Cellular Vehicle-to-Everything (C-V2X) mixed vehicular communication networks, it is straightforward to adopt the dynamic spectrum access of the DSRC over idle C-V2X channels. In this paper, we propose a novel joint spectrum prediction and selection scheme for DSRC users (DUEs), aiming to improve the data rate of DUEs and mitigate the interference to C-V2X terminators (CUEs). Firstly, we define the process of DUEs' spectrum selection as an Indian buffet process (IBP), which fits the practical scenarios and also implies that the channel's availability is predictable. Secondly, we integrate channels' available probability predicted by IBP with the corresponding historical reward obtained by deep reinforcement learning (DRL) to acquire a more accurate available channel list. Finally, we exploit a deep Q-learning network (DQN) to select high-profit channels for DSRC communications in the candidate channels. Simulation results show that the proposed algorithm can improve the average data rate of DUEs by 15% and maintain low interference to CUEs under different transmission power levels and densities of DUEs, compared with the other four typical methods.