Channel-State-Aware Deep Reinforcement Learning for Adaptive Routing in Wireless Mesh Networks

In the given paper, a channel-state-aware deep reinforcement learning (CSA-DRL) method of adaptive and intelligent routing in dynamic wireless mesh networks (WMNs) modification is proposed.The proposed system is centered around a deep Q-network (DQN) that can utilize a multidimensional state vector based on real-time physical-layer measurements such as signal-to-noise ratio (SNR), signal-to-interference-plus-noise ratio (SINR), received signal strength indicator (RSSI), link stability, interference index, and local buffer occupancy.These metrics enable the agent to develop a rich context account of wireless environment, which enables it to take proactive decisions, working with nondeterministic and highly dynamic network conditions.The major contribution in this work is that one explicitly considers effects of antenna beamwidth and gain, which is incorporated in DRL-based mesh routing framework.Observing the properties of directional antennas in the channel metrics, for example, radiation pattern, gain distribution, and beam orientation, the CSA-DRL agent is able to learn the signal propagation patterns and shape the routing policies accordingly.The antenna-aware improvement allows better modeling RSSI and SINR which are orientation and interference area-sensitive.The success of the method can be proved by using simulations with directional antenna modules with the NS-3 simulator, showing that the CSA-DRL model outperforms traditional routing protocols and baseline DRL models in most indicators of performance, such as packet delivery ratio, average throughput, route stability, and latency.The system is demonstrated to be both scalable and resilient to dense and sparse node topology as well as differing conditions of small-scale fading including Rayleigh and Nakagami-m, thus making it a robust system.In addition, because it aligns with physical-layer realities, the CSA-DRL frame work forms the basis of a future RF-aware and beam-adaptive routing that can be deployed in next-generation wireless networks, such as 6G, vehicular ad hoc networks, and edge-based smart infrastructure.On the whole, the proposed work shows an innovative and feasible contribution to the development of the field of smart wireless communication based on reinforcement learning in the form of an antenna.