Machine Learning Based Cognitive Radio Sensor Networks Spectrum Sensing And Access Technology

As a novel technology of information acquisition and processing,wireless sensor networks(WSNs)has the characteristics of low power,low cost,distributed and self-organization and have become a hot research area.The issue of constrained spectrum resource in WSNs can be efficiently addressed by introducing cognitive radio(CR)technologies.Spectrum utilization can be improved significantly by making it possible for a CR user to access a spectrum hole unoccupied by the licensed user.The key technologies in CR are spectrum sensing,spectrum decision and spectrum handoff,where spectrum sensing detects spectrum holes in real time,spectrum decision decides which channel to move and spectrum handoff provides resilient service for the secondary users.In this paper,we mainly focus on applying machine learning in CR-WSN,designed to improve the spectrum sensing accuracy by using the method of learning,and reduce resource consumption.First,a typical concern in CR-WSN is energy consumption due to resource-constrained nature of sensor nodes.Moreover,additional energy is consumed in a CR-WSN to support CR-exclusive functionality such as spectrum sensing and switching,which could shorten sensor node lifetime.However,some sensor nodes could receive similar signal due to similar channel condition such that they probably have same spectrum sensing results.Consequently,we propose a clustering based scheme for spectrum sensing in CR-WSN,which reduces energy consumption by involving less nodes in spectrum sensing.With our improved clustering algorithm,sensor nodes are grouped into different sets based on their similarity in sensing result.In order to identify the optimal cluster number,a new objective function,based on new intra-cluster and inter-cluster proximity measures has been proposed in our study.The simulation results show that the proposed scheme can effectively reduce the energy consumption of sensor node and improve global detection probability.Then we propose a reinforcement learning(RL)based scheme for spectrum access in CR-WSN.The key point of spectrum access is to avoid interference to licensed users,while reducing the probability of conflict between cognitive users and seek to maximize system throughput.By using RL,cognitive users interact with the environment by gradually.The experiments show that our proposed spectrum access scheme learning the optimal spectrum access policies intelligently,reduce the impact of licensed users and avoid the random access degrade system performancebased on RL can improve overall system throughput in cognitive radio networks.Finally,this paper complies with the current trend of CR-WSN technologies and contributes to expanding the applications of CR-WSNs in various areas.