Research On Spectrum Allocation Based On Deep Reinforcement Learning

The rapid development of wireless communication technology has made the number and types of cognitive users continue to increase rapidly,and spectrum resources have become increasingly tight.In order to improve the utilization of the spectrum,the introduction of cognitive radio technology enables wireless personal mobile devices and highly intelligent wireless network environments to calculate wireless resources based on user needs,providing users with wireless resources and services that best meet their communication needs.However,on the one hand,in the process of realizing the rational allocation of the spectrum through the cognitive radio network,in order to reduce the interference of the secondary user on the cognitive environment and the energy consumption of the receiver,its transmission power needs to be strictly controlled or accurately judge the time for users to access and leave the channel,so as to reduce the influence between users and improve the utilization of idle spectrum.On the other hand,for a large and complex cognitive radio network,it is difficult to accurately establish a mathematical model of the network environment.In this article,a spectrum allocation method that combines cognitive radio technology with deep reinforcement learning is adopted.The characteristics of reinforcement learning,model-free control,is utilized to solve the spectrum allocation problem in complex cognitive environments.At the same time,deep learning will make up for the shortcomings of reinforcement learning when dealing complex problems and highdimensional data.Therefore,applying deep reinforcement learning to the spectrum allocation problem will realize intelligent control of transmit power,some problems such as the complexity of the network environment or difficulty in modeling in cognitive radio networks will be solved.The main work of this article includes the following points:We first analyze the problem of power control when using deep reinforcement learning to accomplish transmission,and a power control method that combines social relationship networks and physical transmission networks is proposed,the social credibility determine the access order of secondary user.Secondly,the spectrum allocation problem is further analyzing in the VANETs,where the location of users changes rapidly and the surrounding environment is more complicated.According to the characteristics of vehicle driving and the rapidly changing topology,a “multi-hop broadcast protocol based on vehicle mobility” is proposed.“Global optimization reward accumulation algorithm" is proposed that combine the capable of processing sequence inputs of RNN with deep Q network algorithms,and change the reward mode which in traditional reinforcement learning.Finally,the simulation results on the Tensor Flow show that in ordinary cognitive radio environments,deep reinforcement learning control algorithms combined with social relationships between users can effectively improve the transmission success rate and the reliability of information.In VANETs,the forwarding protocol and channel allocation algorithm proposed in this paper improve the channel utilization while reducing network congestion.