Deep Reinforcement Learning-Based Method of Mobile Data Offloading

• Published in: 2018 Eleventh International Conference on Mobile Computing and Ubiquitous Network (ICMU) pp. 1 - 6
• Main Authors: Mochizuki, Daisuke, Abiko, Yu, Mineno, Hiroshi, Saito, Takato, Ikeda, Daizo, Katagiri, Masaji
• Format: Conference Proceeding
• Language: English
• Published: IPSJ 01-10-2018
• Subjects: Bandwidth; Data communication; Delay tolerant; Delays; Load management; Mobile data; Offloading; Quality of service; Reinforcement learning; Servers
• DOI: 10.23919/ICMU.2018.8653588

The demand for mobile data communication is increasing due to diversification and the increase in the number of mobile devices accessing mobile networks. This demand is likely to increase further. In a mobile network, communication quality deteriorates due to the congestion of the cellular infrastructure because of the concentration of demand for mobile data communication. Therefore, improving the cellular infrastructure bandwidth utilization efficiency is crucial. To improve the cellular infrastructure bandwidth utilization efficiency, we previously proposed the mobile data offloading protocol. Although this method balances the load by focusing on the delay tolerance of contents in the uplink, accurately balancing the load is challenging. In this paper, we propose a mobile data offloading method using deep reinforcement learning for increasing offloading performance of the uplink. The proposed method can balance the load appropriately by learning what the bandwidth and transmission timing provide to the user equipment when the previous method does not work properly.