Machine Learning Approach for Spectrum Sharing in the Next Generation Cognitive Mesh Network

Nowadays, there is an unexpected explosion in the demand for wireless network resources. This is due to the dramatic increase in the number of the emerging web-based services. For wireless computer network, limited bandwidth along with the transmission quality requirements for users, make quality of service (QoS) provisioning a very challenging problem. To overcome spectrum scarcity problem, Federal Communications Commission (FCC) has already started working on the concept of spectrum sharing where unlicensed users (secondary users, SUs) can share the spectrum with licensed users (primary users, PUs), provided they respect PUs rights to use spectrum exclusively. Cognitive technology presents a revolutionary wireless communication where users can exploit the spectrum dynamically. The integration of cognitive technology capability into the conventional wireless networks is perhaps the significant promising architectural upgrade in the next generation of wireless network that helps to solve spectrum scarcity problem. In this work, we propose integrating cognitive technology with wireless mesh network to serve the maximum number of SUs by utilizing the limited bandwidth efficiently. The iv architecture for this network is selected first. In particular, we introduce the cluster-based architecture, signaling protocols, spectrum management scheme and detailed algorithms for the cognitive cycle. This new architecture is shown to be promising for the cognitive network. In order to manage the transmission power for the SUs in the cognitive wireless mesh network, a dynamic power management framework is developed based on machine learning to improve spectrum utilization while satisfying users requirements. Reinforcement learning (RL) is used to extract the optimal control policy that allocates spectrum and transmission powers for the SUs dynamically. RL is used to help users to adapt their resources to the changing network conditions. RL model considers the spectrum request arrival rate of the SUs, the interference constraint for the PUs, the physical properties of the channel that is selected for the SUs, PUs activities, and the QoS for SUs. In our work, PUs trade the unused spectrum to the SUs. For this sharing paradigm, maximizing the revenue is the key objective of the PUs, while that of the SUs is to meet their requirements and obtain service from the rented spectrum. However, PUs have to maintain their QoS when trading their spectrum. These complex conflicting objectives are embedded in our machine learning model. The objective function is defined as the net revenue gained by PUs from renting some of their spectrum. We use a machine …