Machine Learning based Distributed Dynamic Spectrum Access

With innovations in the field of wireless communication many advanced technologies are coming up keeping in point data rates and mobile traffic so that users can efficiently communicate. Different coexistence scenarios between Wi-Fi and long-term evolution technique as 4G-LTE, spectrum-sensing techniques and spectrum-access method have been studied, proved, and simulated so far. This paper is mainly focused on performing spectrum-access in dynamic and distributed manner. This can be achieved using deep-reinforcement-learning model. It gives us information about overall channel-utilization and response of model applying different policy methods. In the proposed methodology policy functions as Boltzmann distribution, epsilon greedy, upper confidence bound, and Thompson sampling are used to obtain a reward in learning algorithm. Results shows that Thompson sampling policy performs superior than others.