Challenges in 5G: how to empower SON with big data for enabling 5G
he advent of the first generation (1G) wireless telephony changed the world by connecting people to people, as its predecessor technology could only connect places to places. Now 5G aims to change the world by connecting anything to anything. Moreover, unlike its predecessors, 5G needs to be conceived as a set of technologies that are efficient and economical in terms of an array of key performance indicators (KPIs) that are of interest to all stakeholders in an omnium-gatherum of applications.
... These KPIs, from an operator's perspective, include capacity, quality of service (QoS), capital expenditure (CAPEX), and operational expenditure (OPEX). From a user's perspective, the KPIs include seamless connectivity , spatio-temporal uniformity of service, perception of almost infinite capacity or zero latency, and, last but not least, the cost of service. Obviously, no technology can offer infinite capacity or zero latency, but by maintaining a latency shorter than the human sensory and physiological delay in the type of application under use, a false perception of infinite capacity or zero latency can be provided. For, example if the network can provide a latency below 100 ms, 10 ms, and 1 ms for audio, video, and tactile applications, respectively, limited by the intrinsic latency of the pertinent human sensory organs and associated neural circuitry, the user will have a perception of infinite capacity and zero latency . However, designing the complete 5G network, only for extremely low latency requirements might it be inefficient in addition to difficult, if not impossible. The more logical approach is to design 5G to be fully self-organizing with end-to-end network behavior intelligence, from the perspective of a self-organizing network (SON) engine, so that it can exploit the cognition of the context of application as well as that of the state of the network to divert and focus the right amount of network resources when and where needed such that users will perceive seamless and limitless connectivity. 5G also has to take into account the recent marriage between Moore's law backed computing power and the wireless technology that has triggered a new era in human history. In this new era the use of wireless communications for novel applications is only bound by imagination. There is hardly an aspect of human life that will not benefit from high-speed T Abstract While an al dente character of 5G is yet to emerge, network densification, miscellany of node types, split of control and data plane, network virtualization, heavy and localized cache, infrastructure sharing, concurrent operation at multiple frequency bands, simultaneous use of different medium access control and physical layers, and flexible spectrum allocations can be envisioned as some of the potential ingredients of 5G. It is not difficult to prognosticate that with such a conglomeration of technologies, the complexity of operation and OPEX can become the biggest challenge in 5G. To cope with similar challenges in the context of 3G and 4G networks, recently, self-organizing networks, or SONs, have been researched extensively. However, the ambitious quality of experience requirements and emerging multifarious vision of 5G, and the associated scale of complexity and cost, demand a significantly different, if not totally new, approach toward SONs in order to make 5G technically as well as financially feasible. In this article we first identify what challenges hinder the current self-optimizing networking paradigm from meeting the requirements of 5G. We then propose a comprehensive framework for empowering SONs with big data to address the requirements of 5G. Under this framework we first characterize big data in the context of future mobile networks, identifying its sources and future utilities. We then explicate the specific machine learning and data analytics tools that can be exploited to transform big data into the right data that provides a readily useable knowledge base to create end-to-end intelligence of the network. We then explain how a SON engine can build on the dynamic models extractable from the right data. The resultant dynamicity of a big data empowered SON (BSON) makes it more agile and can essentially transform the SON from being a reactive to proactive paradigm and hence act as a key enabler for 5G's extremely low latency requirements. Finally, we demonstrate the key concepts of our proposed BSON framework through a case study of a problem that the classic 3G/4G SON fails to solve.