Article Preview
TopIntroduction
As communications technology advances, there will be an increase in the number of users and user equipment (UE) that will be connected to cellular networks. In addition, more demanding applications are becoming a larger percentage of total network traffic, placing a higher burden on cellular networks than conventional traffic. 5G cellular networks need to be developed and maintained to deal with these challenges. The major change with 5G networks is the development of SON. SON allows the network to configure, SO and SH itself are based on the data that it receives. The SON would be able to analyze the data it obtains, identify anomalies, and implement corrective measures. These corrective measures mostly take the form of SH functions which would attempt to reduce the impact of anomalies on network performance.
The challenges that 5G faces are explained in (Zhu, et al., 2019) and (Asghar, Farooq, & Imran, 2018). 5G networks need to cope with increasing data rates, increasing complexity of network architecture, and an increase of network density. User-centric approaches to implement a SON with the use of ML could help design a future 5G network which solves many of these problems. A user-centric approach is defined as the network design and strategies that are developed and implemented where the end-user is the point of focus (Murudkar & Gitlin, 2019). SON solutions are modified based on the end-users’ needs and feedback. Currently, there are few optimization networks that use user-centric approaches, and this can be a problem.
SON has three different properties: self-configuration, self-optimization, and self-healing (Klaine, Imran, Onireti, & Souza, 2017), (Scheit, 2014). The main objective of SON is to prevent any communication network from failing by determining any anomalies within the system then concurrently applying SH measures before the station or system reaches a failure state by providing intelligence and resilience. If a cell were to fail for any reason, SH would be implemented to offload users onto neighboring cells without putting the other cell(s) at risk of failure.
The increasing volume and complexity of data makes more traditional methods of partitioning and network optimization ineffective. To solve this, the implementation of big data technology is essential. Various platforms and architectures are available to efficiently process large quantities of data. The paper in (Londe & Rao, 2017) analyzes these platforms and describes the concept of scaling, which is a system’s ability to handle increasing amounts of data. Horizontal scaling is the type that is generally used for big data processing. Software framework implementations for horizontal scaling were limited in the past, but programs have been developed that help solve this challenge. Horizontal scaling, as well as more platforms and methods for achieving big data analytics are detailed in (Mahmud, Huang, Salloum, Emara, & Sadatdiynov, 2020).
Large datasets which include geographical location, type of movement (i.e., static, walking, driving), UE demand, etc., can be used to implement a sophisticated data analytics function. One of the recently proposed data analytic functions in 5G networks is known as network data analytics function (NWDAF), as it provides other network functions (NF). Although NWDAF has several capabilities, the paper in (Gökarslan, Sevgican, Tugcu, Turan, & Yilmaz, 2020) investigates three of these capabilities: abnormal behavior information for a group of UE, expected behavior information for a group of UE, and network load performance in an area of interest. These three capabilities are chosen to strictly focus on creating a network that can be sustainable and have a quality of service (QoS).
Other important aspects for SH and SO would be handovers (HO) and the usage of user mobility prediction. By calculating the projected position of a user, a cell can inform neighboring cells of the users that potentially are moving into their cells. 5G load balancing becomes an optimal solution to handle these scenarios. The details and important KPIs regarding HO are discussed in papers (Ma, Yang, Zhu, & Zhang, 2020), (Qin, et al., 2018), and (Wang, Liu, & Fang, 2020).