Cross Layer Routing in Cognitive Radio Networks using Deep Reinforcement Learning

Cognitive radio networks (CRNs) implementing spectrum sharing between primary and secondary users are able to provide a needed solution for the increasing spectrum scarcity problem. This paper presents a cross layer resource allocation scheme for underlay dynamic spectrum access (DSA) in cognitive radio networks with a goal to improve end-to-end Quality of Experience (QoE) for video traffic, as measured through the Mean Opinion Score (MOS). The presented solution implements a deep Q-network (DQN) that allows a CR to perform resource allocation by observing the environmental variables over physical and network layers and takes actions to update its own parameters across these layers in order to maximize the MOS for its interactive video stream, all while maintaining interference to the primary users below a threshold limit. Simulation results show that perceived quality for the transmitted video using the proposed cross-layer scheme outperforms a benchmark comparable DQN scheme that adapts only the physical layer, often yielding an improvement of one grade scale in the MOS. Moreover, the proposed cross-layer DQN scheme achieves a balanced load on a network with routers with unequal service rates.