Simultaneous Data Rate and Transmission Power Adaptation in V2V Communications: A Deep Reinforcement Learning Approach

Abstract

In Vehicle-to-Vehicle (V2V) communications, channel load is key to ensuring the appropriate operation of safety applications as well as driver-assistance systems. As the number of vehicles increases, so do their communication messages. Therefore, channel congestion may arise, negatively impacting channel performance. Through suitable adjustment of the data rate, this problem would be mitigated. However, this usually involves using different modulation schemes, which can jeopardize the robustness of the solution due to unfavorable channel conditions. To date, little effort has been made to adjust the data rate, alone or together with other parameters, and its effects on the aforementioned sensitive safety applications remain to be investigated. In this paper, we employ an analytical model which balances the data rate and transmission power in a non-cooperative scheme. In particular, we train a Deep Neural Network (DNN) to precisely optimize both parameters for each vehicle without using additional information from neighbors, and without requiring any additional infrastructure to be deployed on the road. The results obtained reveal that our approach, called NNDP, not only alleviates congestion, leaving a certain fraction of the channel available for emergency-related messages, but also provides enough transmission power to fulfill the application layer requirements at a given coverage distance. Finally, NNDP is thoroughly tested and evaluated in three realistic scenarios and under different channel conditions, demonstrating its robustness and excellent performance in comparison with other solutions found in the scientific literature.

Keywords

• Vehicular ad-hoc networks
• connected vehicles
• Vehicle-to-Vehicle (V2V) communications
• congestion control
• power control
• data rate control