Duelling Q-Learning Based Path Selection for 5G D2D Networks

Reinforcement learning is a promising approach for path selection in 5G mobile D2D networks. However, the conventional Q-learning algorithm faces limitations in handling large state spaces and high-dimensional action spaces. The dueling Q-learning algorithm has been proposed to overcome these limitations by decoupling the state and action value functions. In this paper, we recommend a path selection algorithm using dueling Q-learning for mobile 5GD2D networks. The goal of the algorithm is to locate a path from source to the destination node that has lowest possible delay and the highest possible throughput. The proposed algorithm utilizes a reward function that considers both delay and throughput. The algorithm is evaluated on a simulation of a 5G multi-hop (MH)D2D communication (D2DC) network, and the results demonstrate its superior performance in finding an optimal path compared to the conventional Q-learning algorithm.