Joint Data-model Driven Content Caching In Wireless Edge Networks

Both the spread of the smart mobile devices and the development of the diversified mobile services have greatly promoted the social progress and transformation.At the same time,the transmission and computing tasks in the core network are increasing exponentially,which bring heavy loads to the wireless communication systems and drives the reformation of the wireless network architecture.The content caching technology of wireless edge networks has emerged.Through exploring the potential of the cache resources in wireless edge networks,the data and computing tasks can be offloaded from the cloud servers to the edge devices.Then,the heavy loads of the core networks can be relieved.Meanwhile,the users can be provided with the high-quality services in short distances.However,the communication resource and cache resource are rather scarce in wireless edge networks,which limits the popularizing of the content caching technology.This thesis enhances the content caching performance from the following four aspects:1.Enhancing the popularity of the cached contents through the content caching-oriented popularity prediction.We consider the content caching oriented popularity prediction through a weighted clustering approach in order to improve the caching performance.Some critical factors are taken into consideration,e.g.,the cache capability,the transmission capacity,the user dynamics and the time-varying user interests.For depicting the explicit relationship between the caching performance and the popularity prediction accuracy,we derive the popularity prediction error distribution of each content,and design the caching threshold.By extracting the insights in the relationship between the popularity prediction accuracy and the user clustering strategy,we develop a weighted clustering-based popularity prediction algorithm,which takes the caching regret probabilities of files as the weights.The results of the analysis show that by embracing the content caching-oriented popularity prediction,the cache hit ratio can be significantly enhanced under both the constraints of the communication resource and the cache resource.2.Improving the cache resource utilization through the clustering-based cooperative content caching.We design a clustering algorithm for the sectionalized caching with the consideration of the time-varying nature of content popularity.Some critical factors are taken into consideration,e.g.,the cache capability and the user dynamics.By measuring the cooperative caching gain of each user pair,we design the piecewise interest similarity as the clustering criterion.Then,we propose an affinity propagation based clustering algorithm to partition the users into different clusters.Moreover,we establish the clustered cooperative caching as a survival process model,and derive the re-clustering decision through the tradeoff between the decay of the cooperative caching gain and the extra cost of re-clustering.The results of the analysis show that the clustering-based cooperative content caching can utilize the discrete cache resources in the edge networks efficiently,and further reduce the average transmission delay.3.Improving the communication resource utilization through the cache-aided content pushing.We propose a multicast content pushing strategy to maximize the offloaded traffic with the cost on content caching based on structured deep learning.Some critical factors are taken into consideration,e.g.,the cache cost and the transmission capacity.Through analyzing the spatiotemporal correlations of the pushing decisions,we design a structured convolution neural network for multicast content pushing.Through deriving the optimal pushing strategy of the transmission constraint-relaxed problem,we design a performance upper bound of content pushing for guiding the training direction.The results of the analysis show that the cache-aided content pushing can utilize the transmission resource flexibly,and achieve the balance between the offloaded data amount and the caching cost.4.Promoting the edge intelligence through the cache-aided edge learning.We propose a novel distributed knowledge-aware edge learning framework by taking both the benefits of the edge learning and the knowledge incorporation.Some critical factors are taken into consideration,e.g.,the transmission capacity,the learning state and the knowledge base caching modes.Considering both the characteristics of the edge learning and the knowledge-aware learning,we derive the influence of the transmission scheduling to the global learning performance.Through analyzing the caching modes of the local knowledge bases,based on both the learning state and the channel state of the edge devices,we design the optimal and asymptotically-optimal transmission scheduling policies for different caching modes respectively.The results of the analysis show that the learning performance can be enhanced through the cache-aided edge learning.