上海科技大学知识管理系统

Caching is a promising technique to reduce communication load during peak hours. In practice, users may not communicate directly with the server, but through intermediate relays. This paper considers a two-hop cache-aided network where a server wishes to communicate files with users with the help of relays. We first propose centralized coded caching schemes that improve the state-of-the-art schemes. Unlike the existing schemes that separately design the transmission of the two hops, we jointly design the delivery coded symbols in the two hops via a superposition coding manner, to achieve a larger multicast gain at the server. Since our centralized scheme requires the users attached to different relays to cache the same contents, not feasible for the decentralized setting where all the relays and users independently store contents, we propose a genius decentralized scheme that intricately designs the coded symbols depending on the cached status of subfiles among the users and relays. Theoretical analysis and numerical results demonstrate that our schemes can greatly reduce the communication load compared to the existing schemes. Finally, we analyze the impact of caching size on the communication latency for full-duplex hierarchical networks, and show that if each relay’s caching size is sufficiently large but strictly smaller than the size of the file library, increasing the relay’s caching size could be wasteful.