深度学习框架下的移动感知预缓存策略

Mobility-aware Precaching in Small Cell Network with Deep Learning Framework

为了在移动流量需求不断增长的条件下提高用户体验,本文针对小基站网络提出了一种基于深度学习的移动感知预缓存策略.该策略采用条件变分自动编码器根据大量历史数据建立用户移动模型,然后预测用户将来可能经过的基站,并且在这些基站上预缓存用户正在下载的文件的一部分.本文定义了缓存效用用以评估缓存策略的性能.通过在真实GPS轨迹数据上的仿真实验,验证了所提出的缓存策略与典型对比策略相比能够为用户提供更高的平均下载速度,具有更大的缓存命中率,产生更大的缓存效用.

In order to improve users’QoE under the condition of increasing mobile traffic demand,this paper proposes a mobility-aware precaching strategy based on deep learning for small cell networks.The strategy applies conditional variational autoencoder to train a user mobility model from a large number of historical trajectories,then predicts the base stations which user probably associates with in the future,and precache a portion of the file that the user is downloading on these base stations.This paper defines caching utility to evaluate the performance of caching strategies.The simulation on real GPS trajectories show that the proposed strategy can provide higher average download speed,has a higher cache hit rate and yields better caching utility compared with some typical strategies.