Reinforcement Learning for Quality of Experience Optimization in Tactical Networks

摘要

The future tactical battlespace will require a variety of services that are deployed at the company or platoon level. Tactical radios provide the communications between these elements, which must operate in a disconnected intermittent, limited bandwidth environment. The projected computing and communications requirements are significantly greater than what tactical radios and their devices currently offer. Capacity is the result of a bottom-up approach, where routes and links are determined by the nodes locations. This leads to the situation where the capacity does not take the applications requirements into account. We propose a top-down approach that considers application requirement in the network and attempts to reconfigure the network to improve overall network performance through reinforcement learning techniques. The learning considers understanding of the quality of experience demands or requirements of the communications services in a network domain and learns optimizations of the Quality of Service in the network. We present validation of this approach to use of tactical node deployments in Switzerland.