Autonomous Traffic Offloading in Heterogeneous Ultra-Dense Networks Using Machine Learning

The scarcity of network resources and the contention between resources and traffic volume have been the most critical network performance bottlenecks due to the booming growth of various applications in mobile Internet and Internet of Things. Consequently, effectively matching traffic with resources is of great importance and poses significant challenges. The 5G mobile communications networks will be heterogeneous, dense, and smart with various resources autonomously matching with the traffic demands. Although various traffic offloading schemes have been extensively investigated, applications in 5G present new characteristics such as interference-awareness on licensed or unlicensed bands, autonomous spectrum utilization, and delay-tolerant or delay sensitive traffic. In this article, we focus on HUDNs, which comprise dense small cells on the licensed band, WiFi AP on the unlicensed band, D2D communications, and V2V communications coexisting together to address the ever-increasing performance demands on both the user and network sides. We first summarize the recent research findings in this area and the technical challenges. We further present emerging traffic offloading frameworks and discuss the implementation issues including traffic offloading from virtualization, user-centric caching, and network selection in V2V communications. Furthermore, we propose an autonomous traffic offloading scheme based on big data and machine learning and also highlight future research directions.