مديريت حافظه نهان لبه براي سيستم‌هاي توزيع محتوا در شبكه‌هاي بي‌سيم سلولي هوشمند آينده

One of the key solutions employed to mitigate the adverse effects of traffic growth in mobile networks is content caching. By caching data at the network edge, backhaul traffic is reduced, and the Quality of Service (QoS) for users is improved. Developing an effective caching algorithm necessitates precise prediction of future content popularity, which is a challenging task requiring the utilization of available network data and employing creative algorithms for accurate forecasting. In recent years, deep learning models have achieved high prediction accuracy due to advancements in data accessibility and increased computational power. In this research, a new content caching strategy based on deep learning, called User Preference-Aware Caching Strategy (UPCS), is introduced. UPCS consists of three key algorithms. The first algorithm is a content popularity prediction method that forecasts future content preferences based on usersʹ past requests, relying on two Variational Autoencoder (VAE) based models named the Gated Residual Variational Autoencoder Collaborative Filtering model (GRVCF) and the Parallel Variational Autoencoder Collaborative Filtering (PVCF). The second algorithm is a dynamic online cache content replacement algorithm that considering the predicted content popularity, the number of content requests, and the time of the last request, performs the content replacement process. The third algorithm is a collaborative caching approach, wherein if the requested content is not available at the Base Station (BS), it searches for it in the neighboring BS before sending a request to the remote server. To prove the applicability of the proposed method, UPCS is considered in the context of two video content distribution scenarios in mobile networks, where users at the edge of the network use multimedia services provided by remote video content providers. To eva‎luate the proposed caching mechanism, various experiments have been designed for both scenarios. The results show that the proposed popularity prediction algorithm used in UPCS provides better performance in terms of precision in the first scenario than the PopSel and Random methods by 1.375% and 2.056%, respectively, and in the second scenario by 1.323% and 1.515%, respectively. Additionally, UPCS outperforms the ECC, NMF, ITBCF, LRU, and FIFO algorithms by 1.189%, 1.272%, 1.384%, 1.668%, and 1.727%, respectively, in the first scenario and by 1.127%, 1.171%, 1.283%, 1.735%, and 1.844%, respectively, in the second scenario. Furthermore, in terms of content retrieva‎l delay (CRD) compared to ECC, NMF, ITBCF, LRU, and FIFO algorithms, UPCS is 1.228%, 1.266%, 1.322%, 1.458%, and 1.479% better in the first scenario, and 1.424%, 1.474%, 1.549%, 1.824%, and 1.899% better, respectively, in the second scenario.

6

75902