Machine Learning Algorithms for 6G Wireless Networks: A Survey

Machine Learning Algorithms for 6G Wireless Networks: A Survey

Anita Patil, Sridhar Iyer, Rahul J. Pandya
DOI: 10.4018/978-1-6684-3921-0.ch003
(Individual Chapters)
No Current Special Offers


Over the past decade, in view of minimizing network expenditures, optimizing network performance, and building new revenue streams, wireless technology has been integrated with artificial intelligence/machine learning (AI/ML). Further, there occurs dramatic minimization of power consumption and improvement in system performance when traditional algorithms are replaced with deep learning-based AI techniques. Implementation of ML algorithms enables wireless networks to advance in terms of offering high automation levels from distributed AI/ML architectures applicable at network edge and implementing application-based traffic steering across access networks. This has enabled dynamic network slicing for addressing different scenarios with varying quality of service requirements and has provided ubiquitous connectivity across various 6G communication platforms. Keeping a view of the aforementioned, in this chapter, the authors present a survey of various ML techniques that are applicable to 6G wireless networks. They also list open problems of research that require timely solutions.
Chapter Preview


Overview of machine learning in wireless communication networks

With the exponential increase in the bandwidth demand and data traffic, there is an immediate requirement to serve this traffic through high-speed wireless communication networks. In turn, this requisites efficient software enabled intelligent algorithms, advanced physical layer solutions, and spectral bands at a higher frequency to fulfil the requirements of the next-generation users. The wireless communication research community has recently shown that the Tera-Hertz (THz) band is one of the promising bands to enable ultra-broadband wireless communication and minimize the spectrum scarcity issues (Zhao et. al., 2021).

The current wireless systems rely heavily on mathematical models; however, such models do not define the system structure accurately. Hence, the use of Machine learning (ML) techniques for wireless communication has gained momentum as these methods enable the attainment of the quality of service functionalities with advanced solutions (Ali et. al., 2020). Moreover, ML techniques provide the replacement of heuristic or Brute Force Algorithms for optimizing localized tasks and can also present adequate solutions that the existing mathematical model are unable to obtain. Currently, the ML algorithms are being deployed and trained statically at different management layers, core, radio base stations, and mobile devices. The dynamic deployment is envisioned to yield enhanced performance and utilization.

In general, the ML algorithms help in tasks such as, classification, regression, the interaction of an intelligent agent with the wireless environment (Syed et. al. 2019). In such operations, ML algorithms work in three different versions viz., supervised learning, unsupervised learning, and reinforcement learning. Few ML models such as, non-parametric Bayesian methods (Gaussian approach), are promising, especially in handling small, incrementally growing data sets; however, they have increased complexity compared to the parametric methods. Further, the Kernel Hilbert Space-based solutions have shown encouraging results in generating improved data rate, which is 10-100 times higher in comparison to the ones shown in the 5G wireless networks, simultaneously being computationally simple and scalable with lower approximation error. Federated Learning (FL) is an alternate distributed ML algorithm which enables mobile devices to collaboratively learn a shared ML model without data exchange among mobile devices (Marmol et. al., 2021). It is being analysed further to be considered as a next-generation solution for orientation, intrusion detection, mobility, and extreme event prediction. Reinforced Learning algorithms help in coding scheme selection, modulation, beam forming, and power control. In addition, physical layer optimization also exploits ML for multi-input and multi-output downlink beam forming. Implementing all the aforementioned ML algorithms at the end-user devices, needs the consideration of key parameters such as, cost, size, and power. Additional considerations in the simulation and the prototyping of ML at the end-user devices are to optimize the physical realization of the design and finding the inputs to the model (Dalal & Kushal, 2019).

The focus of this chapter is to bring out the importance of AI and ML in 6G wireless communication. ML is a component of AI although it endeavors to solve the problems based on historical or previous examples. Unlike AI applications, ML involves learning of hidden patterns within the data (data mining) and subsequently using the patterns to classify or predict an event related to the problem. Simply put, intelligent machines depend on the knowledge to sustain their functionalities and ML offers the same. In essence, ML algorithms are embedded into machines and data streams provided so that knowledge and information are extracted and fed into the system for faster and efficient management of processes (Ali et. al., 2020).

Demand for radio spectrum is increasing as the data traffic is increasing, and hence, massive connections with high quality of service have to be provided. Recent advances in ML have shown that ML will play a major role in solving multiple issues in wireless communication networks. To mention few, ML will provide ease in all sort of applications which were not enabled in the earlier generations such as, Augmented Reality (AR), Virtual Reality (VR), holographic telepresence, eHealth, wellness applications, Massive Robotics, Pervasive connectivity in smart Environment, etc. It is envisioned that ML will enable real time analysis and zero-touch operation, and will provide control in in the 6G networks (Zhao et. al., 2020). Mobile devices can assist and report to the network regarding the ML actions and predictions to aid efficient resource management. In order to manage the connection density, dynamic spectrum management has been proposed in the literature. The key enabling techniques for dynamic spectrum are i) Cognitive Radio ii) Symbiotic Radio, and iii) Blockchain Technology (Hong et. al., 2014; Hewa et. al. 2020).

The scarcity of available spectrum and underutilization of the allocated spectrum necessitates efficient techniques to manage the spectrum dynamically. In dynamic spectrum management, the concept of primary and secondary users exists wherein; secondary users do not have the authority to access the spectrum; however, they can access it whenever the primary spectrum is idle, and it can even be shared with the protection of primary users’ service. This process enables the secondary users to transmit their data without the licence of spectrum.

In order to achieve dynamic spectrum allocation many algorithms are proposed in the literature. These algorithms not only address the issues of spectrum allocation but also issues such as, data security, optimization, power-efficiency, cost-efficiency, etc. Following are the ML algorithms which help in addressing all the aforementioned issues:

  • 1.

    Supervised Learning

  • 2.

    Unsupervised learning

  • 3.

    Reinforced Learning

  • 4.

    Federated Learning

  • 5.

    Kernel Hilbert Space

  • 6.

    Block Chain Technology

  • 7.

    Cognitive Radio

  • 8.

    Symbiotic Radio.

  • 9.

    THz Technology

  • 10.

    Free Duplex

  • 11.

    Index Modulation

Key Terms in this Chapter

SVM: Super Vector Machine

NOMA: Non-Orthogonal Multiple Access

RKHS: Reproducing Kernel Hilbert Space

FL: Federated Learning

AMBC: Ambient Backscattering Communications

SSK: Space shift Keying

THz: Tera Hertz

APM: Amplitude Phase Modulation

QoS: Quality of Service

SR: Symbiotic Radio

MDP: Markov Decision Process

GaN: Generative Adversarial Networks

DL: Deep Learning

IM: Index Modulation

SWIPT: Simultaneous wireless information and Power Transfer

DRL: Deep Reinforcement Learning

ML: Machine Learning

SM: Space Modulation

OFDMA: Orthogonal Frequency Division Multiple Access

Complete Chapter List

Search this Book: