Research On D2D Cache Technology Based On User Preference

The explosive growth of mobile data has brought unprecedented challenges to traditional cellular networks.In order to alleviate the pressure caused by massive wireless traffic,edge caching technology has become a hot research topi c.Edge caching places part of the network content in edge nodes with caching capabilities,which can achieve rapid content distribution during peak traffic and relieve the traffic pressure on traditional cellular networks.Generally,these edge nodes can be base stations or mobile devices.With the rapid increase in the number of mobile devices and the gradual decrease in the cost of storage media in mobile devices,the combination of D2D(Device-to-Device)communication technology and edge caching has become a hot research direction.D2 D communication has advantages such as improving system throughput,reducing the delay of content acquisition,and improving spectrum utilization.In order to make full use of the advantages of D2 D communication in edge caching,it is necessary to design a reasonable and efficient caching strategy.In a specific application scenario,designing a caching strategy usually needs to consider what content to cache and where to place the content.In real applications,the needs of users are usually different,that is,there are differences between user preferences.Considering the differences in user preferences is helpful to the design of caching strategies,so this thesis will study how to use user preferences to design a reasonable and efficient D2 D caching strategy.In the centralized D2 D cache system,the central base station assumes the roles of information processing center and decision center.In this thesis,we consider dynamically changing file popularity,model user prefer ences as user preferences for file types and the global popularity of files in the corresponding types,and use EM(Expectation Maximization)algorithm to learn user preferences.Simulations prove that this method has better high prediction accuracy.In addition,the sliding window method is used to predict the real-time user activity,and the user preference and real-time user activity are jointly considered to predict the real-time file popularity.In the content placement stage,due to the high time comp lexity of using the greedy algorithm to place cached content,this thesis proposes a low time complexity content placement algorithm.The simulation results show that the centralized cache strategy proposed in this thesis can achieve a higher cache hit rat e.The centralized cache strategy will bring a lot of computational pressure to the central base station.Therefore,this article considers distributed D2 D caching.In a distributed cache system,each mobile device is an information processing center and a decision center.Considering that each device cannot obtain global information,this thesis uses a logistic regression model and a gradient descent algorithm to learn user preferences,and uses a sliding window method to predict user activity.Each user can selectively perform cache update operations based on the results obtained by his own prediction.The simulation results prove that the distributed caching strategy proposed in this thesis can not only guarantee a higher D2 D cache hit rate,but also greatly reduce the load of the cache update on the central base station.