Machine Learning Techniques for 5G and Beyond

Table of Contents [show]

Introduction

In today's world, wireless communication technologies are essential for applications in business, commerce, health, and security. These systems continuously advance from generation to generation, and the fifth generation (5G) wireless networks are currently being deployed globally. The physical, network, and application layers, as well as every other component and the building block of a wireless system that we currently understand from our understanding of wireless technologies up to 5G, will each contain one AI/ML technique.

What is 5G?

All US mobile carriers have now launched some form of 5G.It means a fifth-generation mobile wireless network. The initial standards were created in late 2017. 5G services can be divided into three categories: low-band, medium-band, and high-band. They are all incompatible now, and they all work differently. While all US carriers "have" 5G right now, it will be a few more years before we see significant changes from it. In comparison, 4G first appeared in 2010, and it wasn't until 2012/2013 that mainstream applications that required 4G to function became popular. But according to Ericsson, a top supplier of information and communication technology (ICT) for service providers, 40% of the world will have access to 5G by 2024.

Simply put, the "G" in 5G stands for "generation." 1G was an analog mobile network. The initial generation of digital mobile technologies was known as 2G. 3G technologies improved the speed from 200 kbps to several megabits per second. 4G technologies currently offer hundreds of Mbps and even gigabit speeds. 5G offers several new aspects: larger channels with higher speeds, lower latency for excellent responsiveness, and the ability to connect multiple devices simultaneously.

The Connection Between 5G and ML

5G networks integrate a more comprehensive range of connected nodes, such as smart devices and sensors, by catering to many different applications. To enable low-latency applications and maintain data privacy, there has been a paradigm shift in mobile computing from centralized mobile cloud computing to mobile edge computing (MEC). The need to draw meaningful insight from large amounts of decentralized data while preserving privacy gave birth to distributed ML at the mobile edge. Large-scale efforts have been made recently in academia and business to create these technologies.

To adopt machine learning in 5G networks and MEC applications, we need to (i) decompose the model itself, train (or infer) its components individually, or (ii) scale or parallelize the training process to perform model updating at distributed locations associated with the data containers. This makes distributed ML approaches critically essential.

The main issues MEC faces when applying AI in context are workload decisions, resource allocation, server deployment, and overhead.

The authors argue that MEC's effectiveness depends on workloads of high complexity and dimensionality. Approaches such as heuristics greatly simplify the scenario. The work focuses on the combination of edge computing and Deep Learning (DL), precisely segmenting the relationship into four categories: DL application at the edge, DL inference at the edge, DL training at the border, and edge optimization with DL. DL applications at the edge explored technical frameworks for providing smart services, such as real-time video analytics, smart manufacturing, and smart homes/cities. DL on edge derivation focuses on architecture requirements and optimization methods to reduce latency without compromising model accuracy. The techniques discussed included early termination, model selection, optimization, and computation sharing. Regarding DL training at the edge, the authors explored the role of FL and its enhancements when working with a distributed learning environment, with a particular focus on FedAvg. 

To address the gaps observed in the literature, we thoroughly explore distributed ML concerning computation, communication, privacy, and resource allocation. This article offers a complete overview of all interconnects for ML use for 5G. It helps ensure the success of adopting such a concept in mitigating the challenges of 5G and addressing the demands of the diverse array of distributed applications that 5G enables.

Our Learners Also Read: What's the Role of Artificial Intelligence in the Future of 5G and Beyond?

ML Needs in 5G Network

A study that examined the present state of the art and aspirations for AI adoption by mobile network operators and international service providers was led by Ericsson, one of the top producers of mobile systems in the world. According to the survey, operators mostly embraced AI to help them make the switch to 5G and ensure efficient expenditures. Additionally, the management of a huge number of devices and enormous volumes of data has already become more challenging with the advent of 4G/4.5G. Operators anticipate that AI and ML will help simplify this situation. The study also came to the following other important conclusions:-

• Ml is already being incorporated into networks with a primary focus on reducing capital expenditure, optimizing network performance, and building new revenue streams. The advantages of incorporating AI into their networks are already being reaped by operators all around the world. By the end of 2020, 53% of service providers anticipate having fully incorporated some part of ML into their networks.
• ML will be critical to improving customer service, and customer experience generally referred to as “Quality of Experience (QoE)”. ML is expected to help providers enhance the customer experience in several ways, including improving network quality and delivering personalized services.
• ML will help recoup the investment Communications Service Providers (CSPs) are making in their networks for the transition to 5G. Reducing operational costs and ensuring return on network investment are vital priorities that service providers are trying to achieve with AI. Network intelligence and automation are critical to developing 5G, IoT, and industrial digitization. As technologies supporting 5G evolve, operators must increase their network capacity. However, the added ability introduces additional complexity.
• Adopting ML creates new data problems even as it addresses network complexities. Network providers agree that they must develop efficient mechanisms for collecting, structuring, and analyzing the vast volumes of data that ML can manage.

Conclusion

Machine Learning (ML) technology is currently used in industry and academia due to its data-driven function to achieve high network performance. ML algorithms can be used across the network to predict potential problems without needing external resources. This makes ML a promising technique for detecting attacks in and outside of 5G networks, such as those that are similar to known attacks but difficult to see with traditional algorithms. More importantly, ML in 5G security and beyond brings new capabilities in network threat analysis, modeling, and detection. It also makes learning about unprecedented attacks in and beyond the emerging 5G networks, so they can be detected and addressed without needing external resources.

About The Author: