Machine learning (ML) has evolved to the point that this technique enhances communications and enables fifth-generation (5G) wireless networks. ML is great to get insights about complex networks that use large amounts of data, and for predictive and proactive adaptation to dynamic wireless environments. ML has become a crucial technology for mobile broadband communication. Special case goes to deep learning (DL) in immersive media. Through this chapter, the goal is to present open research challenges and applications of ML. An exploration of the potential of ML-based solution approaches in the context of 5G primary eMBB, mMTC, and uHSLLC services is presented, evaluating at the same time open issues for future research, including standardization activities of algorithms and data formats.
TopIntroduction
5G is a wireless communication system based on a collection of advanced technologies and architectures that can be configured for different purposes. As 5G networks are software-based and programmable, they can be configured to meet virtually any set of requirements. As consequence of various requirements, and heterogeneity in devices and applications, the 5G network is complex system. In service-driven network, a single infrastructure efficiently and flexibly supports (enhanced) mobile broadband, (ultra-reliable) low-latency communications and (massive) machine type communications (Bojkovic, 2017; Shafi, 2017). Moreover, mobile operators straggle to extend service coverages and increase network capacity. Traditional approach for complex networks planning, control, operation and optimization is centrally-managed and reactive. A novel proactive, adaptive and predictive networking is necessary (Jiang, 2017). The 5G system is self-adaptive and operate autonomously. We explore in this chapter, challenges and open issues for research in ML communications.
Overall, 5G networks are more challenging in evolving service requirements and complicated configuration issues. As 5G standard specifications have been developing quickly and commercially deployed worldwide, mobile operators straggle with network complexity and quality of user experience. Network complexity refers to deployment of densely distributed 5G base stations, configuration of large-scale antenna arrays, and global scheduling of virtualized cloud networks. Networks are maintained and operated in a smarter and more agile manner. 5G is at an advanced stage of technical development and standardization. The first set of technical specifications were completed in 2018 and further standardization is in progress until 2020. The 3G Partnership Project (3GPP) Release 15 includes New Radio (NR) technical specification for non-standalone architecture (NSA) and eMBB services. The standalone (SA) version of the NR standard is formally completed in June 2018. Current Release 16 study on further NR enhancements and transition to the 5G core network until March 2020. Technical specifications enhances massive MIMO technology, introduces the non-orthogonal multiple access (NOMA), and network virtualization and network slicing based on cloud computing (Wang, 2015; Cui, 2018; Awan, 2018; Gui, 2018).
5G communication system jointly optimizes key performance indicators (KPIs), including connection density, latency, reliability, and user experience. Recently, applied machine learning (ML) provides new concepts and possibilities (Fig. 1) for research in academia, industry and standardization organizations (Simeone, 2018; You, 2019; Li, 2017). The partnership project 3GPP eNA (Enablers for Network Automation) and focus group ITU ML5G (Machine Learning for Future Networks) have proposed research projects on ML communications. 5G supports ultra-reliable, high-speed, low-latency communication, and massive Internet of Things (IoT) services in real-time. Operation in dynamic wireless environment requires quality of service (QoS) guarantees, while ML technology is integrated on network infrastructure and end-user devices (Cayamcela, 2018).