Research on Neighbor Discovery Algorithms Based on Reinforcement Learning with Directional Antennas for Ad Hoc Networks
Changhao Sui, Huaiyu Tang, Jianyin Gao, Liang Liu, Rui Wang, Hao Xu · 2021
The use of directional antennas in Ad Hoc networks can improve the performance of networks, but it will make neighbor discovery challenging. The neighbor discovery algorithms based on reinforcement learning improve the efficiency of neighbor discovery by learning the experience of the process according to the reward obtained by the nodes without the prior location information of neighbors. Four reinforcement learning algorithms, including Q-Learning, SARSA, Q(λ), and SARSA(λ), are used to simulate the time to discover all neighbors and the neighbor discovery ratio in different conditions. According to the simulation results, compared with the completely random algorithm, the neighbor discovery algorithm based on Q-Learning has the largest increase in efficiency with the 1-way handshake, especially in the network with high node density and low speed movement of nodes.