Social-Aware D2D Video Delivery Method Based on Mobility Similarity Measurement in 5G Ultra-Dense Network

The huge amounts of network traffic generated by the ultra-high-definition video playback of super-large-scale video users results in tremendous pressure for front-haul and back-haul in the 5G network, which brings severely negative influence for the large-scale deployment and scalability of video systems and video delivery performance (e.g. transmission delay and packet loss) related to user quality of experience. The direct D2D communications between mobile devices with adjacent position in geographical area can offloading huge video traffic into the underlying networks, which reduces load of cellular base station in the edge networks and relieves traffic pressure in the core networks. In this paper, we propose a novel Social-Aware D2D Video Delivery Method based on Rapid Sample-Efficient Measurement of Mobility Similarity in 5G Ultra-Dense Network (DMSEM). By investigation for one-hop D2D pair relationship, DMSEM builds a social state transition model of user movement, which makes use of encounter duration and shared video length between encountered users to define the state transition condition. A cluster algorithm of encounter events is proposed, which achieves initial clusters of encounter events by calculating similarity between encounter events from the two aspects of both variation rate of geographical distance between mobile users and encounter duration time. DMSEM makes use of the Fuzzy C-Means to refine the initial clusters and extracts encounter patterns of mobile users. DMSEM designs a sample-efficiency rapid recognition algorithm of encounter pattern, which can use small number of encounter distance samples to achieve fast heuristic recognition of encounter pattern. Extensive tests show how DMSEM achieves better results in comparison with other state-of-the-art solutions in terms of packet loss rate, average freeze time, cache utilization, average bitrate, buffer level and control overhead.