Power Control of 5G-Connected Vehicular Network Using PPO-Based Deep Reinforcement Learning Algorithm

In this paper, we propose a novel power control in vehicular 5G-connected network using Deep Reinforcement Learning (DRL) algorithm. We investigate power allocation for Connected Autonomous Vehicles (CAVs) on uplink connections in mm-wave bands between the CAVs and Roadside Units (RSUs). Our objective is to achieve the desired uplink transmission capacity using the minimum required power and minimize co-channel interference for neighboring cells. To achieve this goal, we use the Proximal Policy Optimization (PPO) algorithm implemented by modified actor-critic architecture to solve the problem. In the proposed architecture, a Deep Neural Network (DNN) model is used to gain the desired outputs of the problem. The suggested approach is fully compatible with the existing 3GPP-based 5G architecture and uses the available quantized information in cellular users’ measurement reports which provides seamless integration within existing RAN architectures. The performance of the proposed algorithm is compared with multiple power control algorithms in various road conditions. Simulation results show that the proposed algorithm outperforms the 3GPP-based power control algorithm in the dynamic road environment.