See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/359277495

Machine Learning Algorithms for 6G Wireless Networks: A Survey

Preprint · June 2022

DOI: 10.4018/978-1-6684-3921-0.ch003

CITATION READS

1 219

3 authors:

Anita Patil Sridhar Iyer

Visvesvaraya Technological University KLE Technological University Dr. MSSCET
8 PUBLICATIONS   14 CITATIONS    113 PUBLICATIONS   346 CITATIONS   

SEE PROFILE SEE PROFILE

Rahul Pandya
Indian Institute of Technology Dharwad
43 PUBLICATIONS   247 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Machine Learning Oriented Optimization Strategies and Methods for Next Generation Optical Transport Networks View project

ESSON: Elastic Spectral and Spatial Optical Networks View project

All content following this page was uploaded by Sridhar Iyer on 29 May 2022.

The user has requested enhancement of the downloaded file.

 A Survey of Machine Learning Algorithms for
6G Wireless Networks
Anita Patil
Department of ECE, S G Balekundri Institute of Technology, Belagavi, KA, India- 590010.
[email protected]

Sridhar Iyer
Department of ECE, KLE Dr. M.S. Sheshgiri College of Engineering and Technology, Belagavi, KA,
India- 590008.
[email protected]

Rahul Jashvantbhai Pandya

Indian Institute of Technology, Dharwad, WALMI Campus, PB Road, KA, India- 580011.
[email protected]

Abstract:
The primary focus of Artificial Intelligence/Machine Learning (AI/ML) integration within the wireless
technology is to reduce capital expenditures, optimize network performance, and build new revenue streams.
Replacing traditional algorithms with deep learning AI techniques have dramatically reduced the power
consumption and improved the system performance. Further, implementation of ML algorithms also enables the
wireless network service providers to (i) offer high automation levels from distributed AI/ML architectures
applicable at the network edge, (ii) implement application-based traffic steering across the access networks, (iii)
enable dynamic network slicing for addressing different scenarios with varying quality of service requirements,
and (iv) enable ubiquitous connectivity across the various 6G communication platforms.
In this chapter, we review/survey the ML techniques which are applicable to the 6G wireless networks. and
also list the open problems of research which require timely solutions.

Keywords: 6G, spectrum, THz, AI/ML

List of Abbreviations used in the chapter.

AMBC Ambient Backscattering Communications
AI Artificial Intelligence
APM Amplitude Phase Modulation
CR Cognitive Radio
DSA Dynamic Spectrum Adaptation
DL Deep Learning
DRL Deep Reinforcement Learning
FL Federated Learning
GAN Generative Adversarial Networks
IM Index Modulation
MDP Markov Decision Process
ML Machine Learning
NOMA Non-Orthogonal Multiple Access
OFDMA Orthogonal Frequency Division Multiple Access
QoS Quality of Service
RKHS Reproducing Kernel Hilbert Space
 SM Space Modulation
SSK Space shift Keying
SWIPT Simultaneous wireless information and Power Transfer
SR Symbiotic Radio
SVM Super Vector Machine
THz Tera Hertz

1. 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.

i. Supervised Learning

ii. Unsupervised learning

iii. Reinforced Learning

iv. Federated Learning

v. Kernel Hilbert Space

vi. Block Chain Technology

vii. Cognitive Radio

viii. Symbiotic Radio.

ix. THz Technology

x. Free Duplex

xi. Index Modulation

1. Supervised Learning and Unsupervised Learning Algorithms

The ML algorithms can be mainly classified as supervised or unsupervised. These two

classes mainly differ in the labels of training data sets. In supervised ML output and input attributes
are predetermined (Amanpreet et. al. 2016). The algorithms perform prediction and classification of
the predetermined attributes, and their accuracy. For instance, if the input variable is X and output
variable is Y then, the mapping function using by supervised learning will be Y  F ( X ) .The learning
process stops when the algorithm achieves an acceptable level of performance. As detailed by
(Chowdhury et. al., 2020), first analytical tasks are performed by supervised algorithms using the
training data, and subsequently contingent functions are constructed for mapping variations of the
attribute. These algorithms need pre-specifications of maximum settings to obtain the desired
outcome and performance. With this approach in ML, it has been observed that the training subset
of approximately 66% is rationale, and without demanding more computational time, outcome will
be achieved. Further, the supervised learning algorithms can be further classified into the
classification and regression algorithms.

 Classification: If output variable is a category such as, “red” or “blue” or “disease” and “no
disease” the classification problem can be used (Amanpreet et. al. 2016).
 Regression: If the output value is a real value such as, “rupees” then, the regression problem can
be used. Few popular examples of supervised ML algorithms are:

 Linear regression for regression problems.

 Random forest for classification and regression problems.
 Support vector machines (SVM) for classification problems.
 For resource allocation, and coding scheme selection in wireless communication, supervised
learning algorithms are utilized.

2. Semi-Supervised Machine Learning

In these algorithms, large amount of input data X and small amount of output data Y are
labelled. These problems lie in between the supervised and the unsupervised learning (Sheena &
Sachin, 2019; Yogesh et. al., 2020).

3. Unsupervised Learning
In contrast to the supervised learning, unsupervised data learning comprises of pattern
recognition without having a target attribute. All the variables used in the analysis are used as inputs,
and the techniques are suitable for clustering and association mining techniques. According to
(Amruthnath at. al., 2018), unsupervised learning algorithms are used to create labels in the data that
are subsequently used to implement the supervised learning tasks. Clustering algorithms identify
inherent groupings within the unlabeled data, and subsequently assign label to each data value.
Further, unsupervised association mining algorithms identify rules that accurately represent the
relationships between the attributes. Overall, the aim of unsupervised learning is to model the
underlying structure or distribution in the data so as to learn more about the data. These algorithms
do provide correct answers and there is no teacher; hence, they are called as ‘Unsupervised
Algorithms’. These algorithms can also be grouped into Clustering and Association Algorithms
(Zhao et. al., 2021).

 Clustering: To discover the inherent groupings in the data. An example is based on purchasing
behaviour grouping of customers.
 Association: To discover rules that describe large portions of the given data such as, people who
buy X will also buy Y.

Few additional examples of the unsupervised learning Algorithms are:

 K-means for clustering problems.
 A priori algorithm for association rule learning problems.
  The commonly used techniques are: clustering, auto encoders (Kien et. al., 2020), deep belief
nets, generative adversarial networks, and the expectation–maximization algorithm. It is also
used in the physical layer for optimal modulation, channel-aware feature extraction, anomaly
detection, localization etc.

4. Reinforcement Algorithm

Reinforcement learning (RL) is one of the basic ML paradigms along with supervised and
unsupervised learning. It is concerned with the manner in which intelligent devices are required to
make decisions or take actions in an environment in order to maximize the notion of cumulative
reward. It differs from supervised learning as it does not need labelled input/output pairs to be
presented, and does not need sub-optimal actions to be explicitly corrected. Instead, the focus is on
finding a balance between exploration and exploitation. Markov Decision Process (MDP) is
typically used to state the environment, as reinforcement learning algorithms make use of dynamic
programming techniques. Reinforcement learning algorithms do not assume knowledge of an exact
mathematical model of the MDP; however, they target large MDPs where exact mathematical
models become infeasible (Vasileios. Et. al., 2021).

RL finds an application in many disciplines such as, game theory, control theory, simulation
based optimization, multi-agent systems, swarm intelligence and wireless communication. For
example, in the operational research and control literature, reinforcement learning is referred as
approximate dynamic programming, or neuro-dynamic programming. The theory of optimal control
includes reinforcement learning problems which is concerned with the characterization and
existence of optimal solutions and algorithms for their exact computation. The problems are less
concerned with approximation or learning, specifically in the absence of a mathematical model of
the environment. Function approximation and use of samples are the powerful elements of
reinforcement learning use to deal with large environments and optimization of performance.
Reinforcement learning problem consists of an agent to interact with an environment and learns how
to take actions. At each step of the learning process, state of the environment is observed by the
agent which then takes action from the available set of actions, receives a numeric reward, and moves
to the next state. Hence, the aim of the agent is to maximize long term cumulative reward. Problem
such as resource allocation in wireless domain can be formulated as a reinforcement learning
problem wherein, neural networks can be used as function approximators to learn values of each
state or the rewards that are generated by the environment. Many problems in wireless networks
such as, power control, beamforming, and modulation and coding scheme selection are solved by
various deep reinforcement learning architectures. However, as a drawback, reliance on training is
a major limitation of reinforcement learning. In this regard, recently, there has been advances
towards reducing this reliance, specifically when dealing with extreme network situations. In
particular, the concept of experienced deep reinforcement learning was proposed in which
reinforcement learning is trained using Generative Adversarial Networks (GAN) that generate
synthetic data to complement a limited, existing real dataset (Iqbal, 2021).

Two elements make RL powerful viz., the use of samples to optimize the performance and
the use of function approximation to deal with large environments. RL can be used in following
situations:

i. Only a simulation model of the environment is known i.e., subject of simulation optimization.

ii. A model of the environment is known and analytic solution is not available

iii. In 6G, RL can be used for resource allocation in cognitive network, beam forming, coding scheme
selection, power control, channel modulation, etc.

RL Algorithms can also be categorised as described in the following sections.

5. Associative Reinforcement Learning

The learning system interacts in a closed loop with its environment and combines the facets
of stochastic learning automata tasks.

6. Deep Reinforcement Learning

RL is extended in this approach by using a deep neural network without explicitly designing
the state space. Learning ATARI games by Google Deep Mind has created increased attention
to Deep Reinforcement Learning (DRL). It influences Markov decision models for selecting next
‘action’ based on the state transition models. In DRL, instead of mapping every solution, states are
approximated or estimated by a neural network. In 6G wireless communications, efficient solutions
for radio resource allocation are required and this is challenging as 6G wireless network will aim to
serve wider variety of users in future radio resources which will be in extreme scarcity. As a solution,
DRL can be implemented to obtain efficient solutions for the radio resource problems (Zhao et. al.,
2020).

7. Inverse Reinforcement Learning

Purpose of inverse reinforcement learning is to imitate observed behaviour which is often

optimal solution or close to optimal. Inverse reinforcement learning approach will have no reward
function but a reward function will be inferred as an observed behaviour from expert.
 8. Safe Reinforcement Learning

Safe RL is the process of learning policies that maximize the expectation of the return in
problems wherein, it is important to ensure reasonable system performance, safety constraints during
learning and deployment processes.

9. Federated Learning:

Federated learning (FL) is helpful is cases when it is difficult to assign the ML models to
each mobile device and data center because in traditional centralized ML algorithms, it is required
for each mobile device to transmit its collected data to the data center for training purpose. It is
impractical for wireless mobile devices to transmit their local data for training ML models due to
privacy issues. To overcome this traditional ML problem, FL, which is a distributed ML algorithm
that enables devices to learn a shared ML model without data exchange among mobile devices, is
implemented (Li et. al., 2021). In this approach, each mobile device and the data center will have
their own ML models, and these ML models are referred to as ‘local Model’ and ‘Global Model’ for
mobile device and data center, respectively.

Further, the training model of ML in FL approach is as follows:

a. Each mobile device uses the collected data to train the local FL model and sends the trained local
FL model to the data center.
b. The data center integrates the local FL models to generate the global FL model and broadcasts it
back to all mobile devices.

As it can be inferred from the FL process, the mobile devices are required to transmit the
training parameters over the wireless links. The limited wireless bandwidth, imperfect transmission
and dynamic wireless channels will affect the FL performance in a significant manner. However,
many studies have shown that implementation of FL process by optimizing the wireless network.
The main objective of FL is to protect the data owner’s privacy. It aims to train a ML model with
training data kept distributed at the clients in order to protect the data owners’ privacy. The working
idea illustration includes devices of the users are used to train local models and then the trained local
models are sent to the base station for aggregation. In this process, the privacy of the user data is
well preserved as the data are still maintained in the devices. In 6G networks, the architecture will
be distributed wherein, the FL technology of AI from centralized cloud based model to decentralized
devices will be necessitated. The AI computing tasks can be distributed to multiple decentralized
edge nodes from a central node, and hence, FL is one of the vital ML methods to enable the
deployment of accurately generalized models across multiple devices (Du et. al., 2020).
10. Kernel Hilbert space

High interference, which is a result of massive connectivity, will be a major performance

bottleneck in 6G. Massive connectivity will involve serving extensive range of devices of various
manufacturing qualities and this will lead to introduction of impairments which develop due to
diverse objects introduced by non-ideal hardware. The major objects may be I/Q imbalance, non-
linear characteristics, high-mobility especially in the industries where fixed solution may not be
applicable, etc. To accomplish the promise of improvement in the data rate of 10-100 times in 6G
as compared to the scenarios of 5G, the Reproducing kernel Hilbert space (RKHS) based solutions
are predominantly useful owing to their computational simplicity (Salh et. al., 2021). RKHS methods
have significantly lower approximation error and are scalable. 6G will potentially encounter the high
interference non-Gaussian environments; however, RKHS based solutions will provide efficiency
with lower approximation error. RKHS based approaches have recently appeared as a solution for
many impairments in the framework of numerous applications in the next-generation wireless
communication systems. Hence, the major issues in 6G such as, detection, tracking and localization
will be addressed by various RKHS based solutions. In recent years, technological advances have
ensured that DL method offers tremendous solutions, and it is largely used in wireless
communications problems. Simultaneously, further improvement in the performance of RKHS
based approaches could be achieved by the feature extraction methods using Monte-Carlo sampling.
The extracted features can be utilized as input to the DL based slants to boost the act of models used
in 6G. As compared to the classical DL algorithms, RKHS based DL algorithms ensure enhanced
performance as there is an intrinsic regularization and support with strong analytical framework
(Guo, 2021).

11. Cognitive Radio

The most widely accepted key technology to empower dynamic spectrum allocation is
cognitive radio. Based on the interaction with surrounding environment, and also based on the
awareness of its internal states such as, hardware and software architectures, user needs, spectrum
use policy, cognitive radio can autonomously and dynamically adapt the system transmission
strategies, bandwidth, transmit power, antenna beam, carrier frequency and modulation scheme. The
transmission strategies are adjusted by computer software in the Software Defined Radio. Cognitive
radio is capable to observe, analyze the observed environment through sensing, and process
information through learning. It decides the transmission strategy through reasoning. Recent
advances in research have shown that cognitive radio technology can be explored more for its
inherent potential towards AI, and also its capacity to facilitate Dynamic Spectrum Allocation (DSA)
very efficiently (Hong et. al., 2014).

Although most of the existing cognitive radio researchers till date have been focusing on the
exploration and realization of cognitive capability to facilitate the DSA, recent research has been
conducted to explore additional potential which is inherent in the cognitive radio technology through
AI. A typical cognitive cycle for cognitive radio comprises of a Secondary User having cognitive
capability which is essential to consistently and periodically observe the environment, and it obtains
the information regarding interference temperature, spectrum holes, etc. It also determines the best
operational features to optimize its own performance subjected to protecting the primary users, and
according to these operational parameters, system configurations are conducted. The traffic statistics
and channel fading statistics of the radio environment can be analyzed by the information time. Due
to this, the cognitive radio equipment is able to learn and perform better in future Dynamic Spectrum
Adaptation (DSA). However, implementing DSA with cognitive radio involves efforts from various
research communities such as, communications, signal processing, computer networking,
information theory and ML. Lastly, combining DSA with cognitive radio and its realization also
fundamentally hinge on the inclination of regulators to open the spectrum for unlicensed access.
Providentially, over the past decades, it has been observed that universal determinations from
regulatory bodies are eradicating regulatory barriers to expedite DSA (Mollah et. al., 2019).

12. Block Chain Technology

In addition to the cryptocurrency, with its prominent features, blockchain has many uses
including, smart contracts, financial services and IoT. Besides, blockchain can fetch new openings
to improve the competence, and to lessen charge in the dynamic spectrum management. Also,
blockchain is utilized to manage and share the spectrum resources, due to which central authority
may get eliminated. Blockchain is fundamentally a distributed and open archive, in which
transactions are firmly recognized in blocks. In the existing block, an exclusive indicator determined
by transactions in the preceding block is recorded. Tampering with any transaction deposited in a
previous block can be detected efficiently. The transactions commenced by one node are broadcast
to other nodes, and a consensus algorithm is opted to conclude which node is approved to authorize
the new block by affixing it to the blockchain. With the distributed authentication and record
mechanism, blockchain will be robust against single point of failures, verifiable and transparent.
Depending on the level decentralization, blockchain can be classified into consortium blockchain,
public blockchain and private blockchain. A public blockchain can be tested and retrieved by all
nodes in the network; whereas, a private blockchain or a consortium blockchain maintained by the
permissioned nodes only. A smart contract, reinforced by the blockchain technology, is a self-
executable connection with its sections being converted to programming writings and deposited in
a transaction. Blockchain has been explored to provide support to various applications of IoT. As a
decentralization entity, blockchain can be useful in integrating the various IoT devices and strongly
stock the massive data formed by them. Moreover, blockchain is also used to accomplish the mobile
edge computing resources to preserve the IoT devices with limited computation capacity.
Blockchain provides security to dynamic spectrum access enabled by the spectrum auctions. In
opportunistic spectrum access provided by sensors, block chain is used to intermediate the spectrum
sensing service (Abdualgalil et. al., 2020). Specifically, blockchain can be used to:

i. Establish a self-organized Spectrum Market: In the management of spectrum resources, a self-

organized spectrum market is essential with enhanced efficiency, with the reduced cost when
compared to depending on a centralized authority. Dynamic spectrum management, transactions and
their verification and smart contract combination will be used to block chain based self-organized
spectrum market. The blockchain security assures that users can make the transactions without trust
in each other. Thus, high bar to achieve spectrum resource accessibility can be reduced.

ii. Create a secured database: The information such as, outcomes of spectrum auctions and access
records, historical sensing results can also be stored in the blockchain.

Overall, block chain technologies have been assumed to convey new opportunities to the
dynamic spectrum management, to ensure progress in the avenues of security, autonomy,
decentralization and administration cost reduction. Moreover, investigation on challenges such as,
deployment, energy consumption, and scheme of blockchain network over the conventional
cognitive radio network should also be carried out.

13. Symbiotic Radio

Symbiotic radio (SR) is an innovative technique that proposes the rewards and overcomes
shortcomings of ambient backscattering communications (AmBC) and cognitive radio (CR). In
building 6G as a spectrum and energy efficient communication system, SR is considered as one of
the promising approach. In addition, Intelligent Reflecting Surfaces (IRS tool) can further play a role
in improving the concert of the transmission by enhancing the backscattering link signal (Mingzhe
et. al., 2021). Additional technologies, related to spectrum, in 6G are as follows:

Terahertz: Terahertz (THz) is measured as one of the vital technologies for the 6G wireless
communications. 5G is described under 100 GHz as the millimeter-wave bands, whereas, 100 GHz-
3 THz is considered as the THz band in 6G. The beyond 90 GHz band is purely used for scientific
service which has not been fully discovered. Therefore, it is proposed to support the increased
wireless network capacity. THz also empowers the ultra-low latency communication and ultra-high
bandwidth paradigms, which caters the desires of several evolving applications such as, IoT and
autonomous driving. THz is especially adequate for high bit-rate short-range communications since,
path loss increases with increase in frequency which makes it challenging for long-range
communications.

Free Duplex: 6G will eradicate the difference between Frequency Division Duplex and Time
Division Duplex, and the frequency sharing will be grounded on the necessities, which is known as
free duplex. Hence, the spectrum resource allocation in 6G will be more effective and efficient. With
the Free duplex technology, transmission rate and throughput in 6G can be increased and
transmission latency will be reduced (Zhao et. al., 2021).

Index Modulation: Index modulation (IM) can progress the transmission rate as it carries the source
information bits through the classical amplitude-phase modulation (APM) signals. Therefore, it can
potentially be used in 6G. Studies suggest that information bits can be communicated through the
index of the antennas in MIMO systems. This technique is known as space shift keying (SSK), and
if SSK is combined with classical linear modulation, it is called as amplitude-phase modulation
(APM). For example, space modulation (SM) is proposed based on the same idea as in SSK. In SM
technology, the base information bits are distributed into two parts: the index of the transmit antennas
and the other parts for the APM. Consequently, SM can meaningfully increase the transmission rate
by sending the extra information bits through the traditional APM transceiver’s antenna index. On
one side of the antennas, further resource units can also be indexed to transmit the additional
information bits. These resource units include, channel state, sub-carriers, and time slots. Index
Modulation, along with Orthogonal frequency division multiplexing access (OFDMA), will be the
key technology to suggestively upsurge the quantity output for supporting more users to access the
6G network (Zhao et. al., 2020).

Overall, 6G is expected to be a composite network where a huge selection of smart devices

is connected to the system and are mandatory to interconnect with others anytime, and the life period
of the battery-charging components is also vital to fulfill the restraints of ultra-low power
consumption. To extend the life duration of numerous devices in the network, simultaneous wireless
information and power transfer (SWIPT) technology is anticipated. SWIPT empowers sensors to be
charged using wireless power transfer; thereafter, battery-free devices can be maintained in 6G,
dropping the network’s power consumption considerably. Consequently, performance on the sum
rate, throughput, and outage probability for non-orthogonal multiple access (NOMA) networks with
SWIPT are derived.

2. Applications of 6G

In the case of 6G networks, a varied series of AI applications will develop into ‘connected
intelligence (Shrestha et. al., 2019), later smoothing every facet of our day-to-day life. For instance:
i. Innovative AI methods can be hired in autonomy to save manpower or for network
management. Overall, 6G invigorates smart healthcare by providing high-precision medical
treatment, real-time health monitoring, and reliable privacy protection.
ii. With the arrival of 6G, Industry 4.0 will be completely comprehended as smart
manufacturing and it will achieve high precision manufacturing.
iii. Intelligent robots coupled by pervasive 6G network empower manufacturing structures to
carry out multifaceted and hazardous tasks devoid of risking people’s life.
iv. The smart home that furnishes with smart IoT devices will deliver a contented living
atmosphere to people, and 6G permits the smart home to confirm the occupants’ safety.
v. In terms of traffic and transportation, the sophisticated sensing and planning algorithms can
be deployed for traffic optimization.
vi. Other applications such as, smart grid and unmanned aerial vehicle will also be enhanced
with the aid of 6G.

Explainable Artificial Intelligence: An enormous number of applications such as, remote surgery
and autonomous driving exist in 6G eon. As these applications are attentively connected to humans’
life, a small error may invite miserable accidents. Hence, it is very essential to mark an AI
explainable for building faith between individuals and machines. Currently, most AI tactics in MAC
and PYH layers of 5G wireless networks are incomprehensible. AI applications such as, remote
surgery and autonomous driving will be an extensive part of 6G, which needs explaining ability to
qualify trust. AI decisions should be understood by human experts, and must be explainable so as to
be considered as reliable. Prevailing approaches, including didactic statements, visualization with
case studies, hypothesis testing can improve the explaining ability of DL (Guo, 2019; Abhishek &
Neha, 2021).

3. Open Research Avenues

Finally, we list the various open problems of research in regard to ML in 6G wireless networks
 a. Implementation of automation requires higher data rates, flexibility, link preservation,
scheduling, on demand beam forming-deep neural network with reinforcement learning
algorithms need to be analysed for usage and deployment constraints have to be studied. In such
case, solution to a challenge ‘how to use deep reinforcement learning for the automation of 6G
wireless network’ will be an observable one.

b. Ultra-reliable low latency and interference management are the major requirements in open data
access. To support the emerging machine critical applications, the transmission time delay of E-
to-E (Equipment to Equipment) is expected to be less than 1ms. Satisfying low latency and high
reliability needs with open data access in business oriented mobile network is an open problem
in 6G.

c. In Application and Platform dependent models a design of device is necessary which can sense
the environment and its history so that resources can be allocated dynamically and spectrum
selection should also happen dynamically to avoid the congestion. Hence the transfer of models
to highly resource constrained platforms and designing a method to dynamically select
application and platform based models is a research challenge.

d. Implementation of ML algorithms for enhancing, automating and managing the 6G network

performance is another research challenge.

References

Abdualgalil, B., & Abraham, S. (2020). Applications of Machine Learning Algorithms and Performance Comparison: A
Review. IEEE 2020 International Conference on Emerging Trends in Information Technology and Engineering (ic-
ETITE), 1–6. DOI:10.1109/ic-ETITE47903.2020.490
Abhishek, D., & Neha, G. (2021). A Systematic Review of Techniques, Tools and Applications of Machine Learning.
Proceedings of the Third International Conference on Intelligent Communication Technologies and Virtual Mobile
Networks (ICICV 2021). IEEE Xplore Part Number: CFP21ONG-ART; 978-0-7381-1183-4
Ali, S., Saad, W., & Rajatheya. (2020). 6G White Paper on Machine Learning in Wireless Communication Networks.
arXiv:2004.13875v1.
Amanpreet S., Narina,T., & Aksha , S. (2016). A review of supervised machine learning algorithms. 3rd International
Conference on Computing for Sustainable Global Development (INDIACom), Corpus ID:27775741
Amruthnath, N., & Gupta, T. (2018). A research study on unsupervised machine learning algorithms for early fault detection
in predictive maintenance. 5th International Conference on Industrial Engineering and Applications (ICIEA).
DOI:10.1109/iea.2018.8387124.
Chowdhury, S., & Schoen, M. P. (2020), Classification using Supervised Machine Learning Techniques, Intermountain
Engineering Technology and Computing (IETC). DOI:10.1109/ietc47856.2020.924921.
Dalal R., and Kushal, R. (2018). Review on Application of Machine Learning Algorithm for Data Science. IEEE 3rd
International Conference on Inventive Computation Technologies (ICICT), 270-273.
DOI:10.1109/icict43934.2018.9034256.
Du, J., Jiang, C., WANG, J., Ren, Y., & Debbah, M. (2020). Machine Learning for 6G Wireless Networks: Carry-Forward-
Enhanced Bandwidth, Massive Access, and Ultrareliable/Low Latency. IEEE Vehicular Technology Magazine.
DOI:10.1109/mvt.2020.3019650.
Guo, W. (2019). Explainable artificial intelligence (XAI) for 6G: Improving trust between human and machine. arXiv preprint
arXiv:1911.04542.
Hewa, T., Gurkan, G., Kalla, A., & Ylianttila. (2020). The Role of Blockchain in 6G: Challenges, Opportunities and Research
Directions, 2nd 6G wireless Summit (6G SUMMIT), DOI: 10.1109/6GSUMMIT49458.2020.9083784.
Hong, X., Jing, W., & Jianghong, S. (2014). Cognitive radio in 5G: a perspective on energy-spectral efficiency trade–off.
IEEE communications Magazine. DOI: 10.1109/MCOM.2014.6852082.
Iqbal H. S. (2021). Machine Learning: Algorithms, Real World Applications and Research Directions. SN Computer Science,
2:160 Available on https://doi.org/10.1007/s42979-021-00592-x
Kien, T., Long Ton, T., & Nguyen, G.M.T. (2020). Plant Leaf Disease Identification by Deep Convolutional Autoencoder as
a Feature Extraction Approach. Proceedings of The IEEE 17th International Conference on Electrical Engineering /
Electronics, Computer, Telecommunications and Information Technology (ECTI-CON), paper ID: #1570639150,
Thailand, Available on https://ieeexplore.ieee.org/document/9158218/
Li, T., Sahu, A. K., Talwalkar, A., & Smith, V. (2020). Federated Learning: Challenges, Methods, and Future Directions.
IEEE Signal Processing Magazine, 37(3), 50–60. DOI:10.1109/msp.2020.2975749.
Marmol, C., Pablo, F.S., Aurora, G.V., Jose, L., Hernandez, R., Jorge, B., Gianmarco B., & Antonio, S. (2021). Evaluating
Federated Learning for Intrusion Detection in Internet of Things: Review and Challenges Enrique. arXiv:2108.00974v1.
Mingzhe, C., Deniz, G., Kaibin, H., Walid, S., & Mehdi, B.(2021). Distributed Learning in Wireless Networks: Recent
Progress and Future Challenges. IEEE Journal on Selected Areas in Communications, 39(12): 3579 – 3605. DOI:
10.1109/JSAC.2021.3118346.
Mollah, M.B., Zeadally, S., & Azad, M.A.K. (2019). Emerging wireless technologies for Internet of Things applications:
Opportunities and challenges, Encyclopaedia of Wireless Networks. Springer International Publishing Cham, pp. 1–11.
Salh, A., Audah, L., & Shah, NSM. (2020). A Survey on Deep Learning for Ultra-Reliable and Low-Latency Communications
Challenges on 6G Wireless Systems. Proceedings of Future of Information and Communication Conference (FICC)
2021, arXiv: 2004.08549v3.
Sheena, A. & Sachin, A. (2017). Machine learning and its applications: A review. International Conference on Big Data
Analytics and Computational Intelligence (ICBDAC). DOI: 10.1109/ICBDACI.2017.8070809
Shrestha, A., & Mahmood, A., (2019). Review of Deep Learning Algorithms and Architectures. IEEE Access, 7, 53040–
53065. DOI:10.1109/ACCESS.2019.2912200.
Syed, J.N., Shree, K.S., Shurjeel, W., Mohammad, N., & Md. Asaduzz, A. (2019). Quantum Machine Learning for 6G
Communication Networks: State-of-the-Art and Vision for the Future. IEEE Access, 7: 46317-46350.
Vasileios, P.R., Sotirios, S., Panagiotis, S., Shaohua, W., George, K. K., & Sotirios K.G.(2021). Machine Learning in Beyond
5G/6G Networks—State-of-the-Art and Future Trends. Electronics, 10, 2786.Available on https://doi.org/10.3390/
electronics10222786.
 Yogesh, K., Komalpreet K., & Gurpreet, S. (2020). Machine Learning Aspects and its Applications Towards Different
Research Areas. IEEE International Conference on Computation, Automation and Knowledge Management (ICCAKM).
DOI: 10.1109/ICCAKM46823.2020.9051502
Zhao, Y., Zhai, W., Zhao, J., Zhang, T., Sun, S., Niyato, D., & Yan Lam, K. (2020). A Survey of 6G Wireless
Communications: Emerging Technologies. arXiv:2004.08549v3.
Zhao, Y., Zhai, W., Zhao, J., Zhang, T., Sun, S., Niyato, D., & Yan Lam, K. (2021). A Comprehensive Survey of 6G Wireless
Communications. arXiv:2101.03889v2.

View publication stats