Increasing the Trust Factor in Cognitive Radio Networks Driven by Software Defined Radio

A current network research trend in is the employment of high algorithmic ML tools such as Reinforcement Learning (RL) for analysis and adaptation to various network situations. The selection of a suitable radio channel in a Cognitive Radio wireless multi-channel spectrum, example 802.11 networks, with outside interference is one such situation. The objective of the agent learning the environment, or the pattern is to accurately predict an interference-free channel for the next slot i.e., the channel should not be scheduled for transmission by a primary user. For the scope of this paper, a case of external interference following and predicting a periodic pattern is considered for simulation. NS3 is the simulation platform on which the channel spectrum pattern is created. OpenAI Gym is an Artificial Intelligence toolkit for Reinforcement Learning (RL) research that enables agent-like simulation on NS3 modules. TensorFlow is used to provide libraries for the RL optimizer algorithm employed for emulating random RL input characteristics during channel selection per iteration. A basic reward arrangement (+1 for a correct prediction, -1 for interference) allows for the RL algorithm to learn from mistakes and improve the iteration performance. The results showcase the prediction success rate of the RL algorithm for predicting a free slot, as well as open new windows to RL analysis for network situations.