Research On D2D Network Caching Design Based On User Preference Prediction

The explosive growth of mobile data traffic in recent years has brought unprecedented pressure and challenges to traditional cellular communication networks.In order to alleviate the traffic congestion of communication links,edge caching technology has gradually received significant attention in the industry.Edge cache is to build relay communication nodes with large-capacity storage space.In the smooth period of traffic,the nodes cache some network contents in advance,and then these contents will be quickly distributed during the peak traffic period.In general,edge cache nodes can be either base stations or various mobile devices.When utilizing mobile devices to assist the cellular network for content distribution,direct data transmission between devices must be achieved by without base stations,and that is D2 D communication technology.D2 D communication multiplexes the bandwidth resources of cellular networks,so that devices closer to each other can directly establish D2 D links.To make full use of edge caching,a reasonable design of the cache strategy is necessary.The caching strategy design is to decide which network content the edge nodes need to store in advance.The design basis includes the current network topology,the probability of different network content being requested,and the communication mechanism.Therefore,this paper first studies how to predict the probability of users requesting network files(materialized network content),then proposes a distributed algorithm for D2 D network caching strategy design,and finally tests and verifies the combined effect of the two parts of work.The key to predicting the probability of a user's request for a file is the measure of the user preference.Most recommendation algorithms use the user's rating of the file as a measure of the user's preference.However,the user's rating on the file occurs after the user requests the file,and the probability of the file request mapped by the score is not accurate.Therefore,starting from the judgment of whether the features of a file meet the user's taste,this thesis proposes a user preference prediction scheme which includes three steps: feature selection,preference measurement and compression-mapping.Feature selection can select valuable file features to calculate similarity.Preference measurement uses 'overall similarity' to measure users' preferences for new files.Compression-mapping normalizes users' preferences more accurately into the probabilities of users requesting new files.The evaluation based on real data shows that the proposed scheme has higher file hit rate and more accurate file request probability for caching design than the traditional scheme using user rating or file popularity as a measure of user preference.Most of the existing works use centralized algorithms to design D2 D network caching strategy,but centralized algorithm not only causes huge computational pressure to the central computing unit,but also does not make use of the computing resources of mobile devices.Thus when designing the cache strategy,this thesis first constructs an optimization problem aiming at minimizing the average download time of files,and then proposes a distributed algorithm to solve it.Based on the traditional BP(Belief Propagation)algorithm,Randomized Gaussian Message Passing scheme and message weighting scheme are introduced to improve the convergence and optimization performance of the distributed algorithm.The simulation results show that the improved BP algorithm has the optimization performance close to that of the centralized greedy algorithm with lower complexity.To verify the overall performance of the proposed user preference prediction scheme and the D2 D caching design algorithm,we built a simulation environment based on partial real data.The simulation results show that the average download time of the files can be minimized by using the proposed prediction scheme to obtain the file request probability for designing the cache,and then using the proposed distributed algorithm to complete the cache design.