Service Based Proactive Storage At Caching-Enable Base Stations

In recent years, with the development of wireless broadband services and the growth of traffic, there is an increasing demand for higher data rate transmission and larger system capacity. How to reduce the network flow of cellular system based on limited resources becomes the focus of research. Traffic localization is one of the effective ways to reduce the network flow. According to the characteristics of wireless link in cellular system, traditional P2P technology will significantly increase the uplink load in the air interface. But for proactive storage at caching-enabled base stations (BS), BSs can not only forward the data but also store them in their cache. Thus traffic load of backhaul links is reduced and the performance of cellular system is improved. With random linear network coding, BS never chooses packets from its neighbours but from other BSs whose link conditions are the best. As a result, it no longer matters to BSs whether content needed is available or not, the risk of network congestion is effectively avoided. In this paper, we investigate the technology of proactive storage and cache replacement at caching-enabled base stations. We also analyse the storage allocation for proactive storage combined with D2D communication.Firstly, the problem of storage allocation for caching-enabled base stations is studied. In cel-lular networks, the proactive storage at caching-enabled BSs is an efficient way to reduce traffic load of backhaul links. We investigate the storage allocation problem with network coding. The storage allocation problem is proved to be an NP-hard problem. By decomposing the NP-hard problem, we propose a low-complexity storage allocation algorithm to solve two subproblems in an iterative way. Combining a heuristic initial allocation scheme with iterative process, we can obtain the storage allocation result very close to optimal result. The convergence of the proposed algorithm is proved and its computation complexity is analyzed. For the case that BSs may not provide the data as we expected, e.g., because of its overload or link congestion, the proposed stor-age allocation algorithm is extended by interval programming to address the network uncertainty. We adopt interval programming according to different degrees of satisfaction which depends on the probability of successful transmission and decoding. Secondly, the problem of distributed cache replacement based on service information is stud-ied. Data of services could be stored in the caching-enabled base stations to reduce the network traffic load. Taking the limited caching space into account, it is necessary for the base stations to adjust the service data stored in their cache based on service popularity in order to achieve bet-ter caching efficiency. We model the cache replacement problem as a Markov Decision Process (MDP). As BSs adjust services stored in their cache in a distributed way and the exact information of system cannot be easily obtained, we propose a Q-learning algorithm to minimize the average transmission cost of system. As it is a multi-agent situation, convergence properties of Q-learning are not applicable. We model the problem as a sequential stage game and Q-learning is applied in each single stage. When state changes, the game moves on to the next stage. Thus convergence property of Q-learning is proved. By utilizing Q-learning, the average transmission cost of service in cellular system is minimized.At last, the problem of storage allocation combined with Device-to-Device (D2D) communi-cation is studied. In traditional mode of cellular communication, a user transmits data to another user through BSs. While in D2D mode communication, users are allowed to transmit data directly between each other with the advantage of spectrum efficiency and offloading. We need to consider the influence caused by D2D communication and adjust storage allocation of BSs according to the storage information of D2D users. However, enabling D2D communication in a cellular network presents a challenge in resource allocation due to the potentially severe interference it may cause to the cellular network by reusing the spectrum with the cellular users. We consider two scenar-ios according to the mobility of users. For the condition of low mobility, the topology of D2D users is determined. For the condition of high mobility, we model the motion patterns of users as space poisson point process. Thus we separately calculate the probability that users choose D2D mode communication for each scenario. According to the probability and algorithms above, we can obtain the strategy of the storage allocation and cache replacement combined with D2D mode communication at caching-enabled base stations.