R2-D2D: A Novel Deep Learning based Content-Caching Framework for D2D Networks

The explosive growth of wireless data and trafficaccompanied by the rapid advancements in intelligence andprocessing power of user equipments (UEs) has paved the way fordevice-to-device (D2D) communication to surface as a promisingsolution. One major benefit is that users can collaborativelycache and share content to reduce costs associated with backhaullinks. In this paper, we explore different approaches to cachecontent for users in a D2D enabled environment and propose anovel two-stacked approach to achieve a higher cache-hit ratiowhile leveraging advancements in deep learning. We propose the‘R2-D2D’framework, wherein we use long short-term memory(LSTM) networks stacked with a recently developed omni-Scaleconvolutional neural network (CNN) for making the cachingdecision. Unlike most previous works, the proposed system modelworks without any apriori knowledge like file popularity distribu-tion, or any assumptions like stationarity of the environment. Ourexperiments show that the proposed framework learns well fromhistorical information, obtaining a D2D cache hit ratio of0.418when5000timesteps of historical information was provided,outperforming a recently proposed neural network collabortivefiltering (NCF) framework by approximately10%.

Keywords: deep reinforcement learning, online learning, Markov decision process, modeling financial markets, algorithmic trading

View pre-print here