3901029 1

“MACHINE LEARNING FOR 5G MOBILE AND WIRELESS COMMUNICATION”

BABATUNDE ADERIBIGBE
DEPT. OF COMPUTER AND INFOMATICS
LONDON SOUTH BANK UNIVERSITY
[email protected]

Abstract — The main reason of 5G was to significantly systematically improve the routine [12]. An artificial neural
provide assets for ever growing users of communication network [10] (ANN) is an important ML procedure used to
resources. In other to realize this goal, ML is seen as a design objective functions that involve optimization[10] or
major facilitator of this concept. This survey paper will approximation [16]. Similarly, as with every other case,
direct it focus on various option of 5G in ML viewpoint. ML has its drawback. Some of the downsides restricting its
This include looking into the various categories of ML and usage with wireless communication [2] are the
their impact on 5G wireless Network. interpretability of the tests, Difficulty in obtaining the
Firstly, the paper sets out the basic introduction of relevant data, the necessary computing resources, the
supervised, unsupervised and reinforcement learning and difficulty added, the amount of time needed to train certain
how they can be used as an enhancer of 5G wireless algorithms, etc. Many types of research have highlighted
network. The survey then proffers promising incite to how the potential of ML as a cellular network facilitator[12],
ML will contribute to meeting each target of the 5G networking functions [12], or radio communications].
network criteria, outlining its distinct usage, and However, there is little evidence in the literature on how
appreciating the effectiveness and constrains of ML on ML can help meet the basic and realistic requirements of
network activity. Finally, this report analyses the possible 5G. Critical concerns have been stated for potential
features of B5G, which offers Potential insight into future research to examine the drawbacks, disputes and trade-offs
directions to undertake study on how ML can lead to the [1] between accuracy and interoperability [1].
realization of the next inline technology,6G. The paper sets out the basic introduction of supervised,
unsupervised and reinforcement learning and how they can
Keywords—Wireless communication, 5G , ML, B5G be used as an enhancer of 5G wireless network. The survey
then proffers promising incite to how ML will contribute to
meeting each target of the 5G network criteria, outlining its
I. INTRODUCTION distinct usage, and appreciating the effectiveness and
constrains of ML on network activity. Finally, this report
Machine Learning(ML) is dominant in the fields of analyses the possible features of B5G [1], which offers
medicine [1] ,image recognition [1] , cars [1], etc. Potential insight into future directions to undertake study
Preliminary curiosity and debates on the viability of 5G on how ML can lead to the realization of the next inline
through the incorporation of ML algorithms have proved technology,6G.
great importance in improving social infrastructure [2], The next section of this survey will investigate the various
Strengthening our practices as well as expanding ML classifications (Figure 1) utilized in mobile wireless
digitalization [4].Even though ML [1] and 5G wireless is networks. Section III and IV discusses the various ML
[1] are often viewed as distinct fields of study, they have a needs of 5G mobile wireless networking and factors that
powerful effect when combined. The beneficial effect of influences ML deployment on 5G respectively. Section V
ML and 5G mobile network have already been explains future trends of technology while VI contains the
demonstrated by some paradigms like Mobile edge caching conclusion.
[1], big data analysis [1], mobile edge computing [6],
context aware networking [8] and network traffic control
[6] . ML has also proven to be an important area in II. MACHINE LEARNING APPLICATION IN WIRELESS
providing solutions to complicated problems that require a MOBILE COMMUNICATIONS
lot of manual adjustment, or issues that are not addressed at
all by a primitive solution. Also, ML algorithms has the Machine learning [1] algorithms and models are widely
capability to automatically learn from previous data and used in several technical issues, particularly recent mobile
technologies and wireless networks (Figure 2) [2]. Against B. UNSUPERVISED LEARNING
this backdrop, this review paper starts with the analyses of In unsupervised machine learning, [1] the sets of data
the categories learning techniques. Explicitly, the following employed in the training ML algorithms unmarked set of
sections considers the supervised unsupervised, [1] and the form x1, x2,... xn, and the program is trying to identify
reinforcement learning [1] together with their contributions subsets with the same attributes within the variables,
to mobile wireless networks[6]. without human interference. Procedures like this can be
especially helpful in pattern identification dataset
associativity.
Groups of similar attributes are automatically detected at
each point. Nevertheless, in some situations, the number of
clusters you want may be chosen for the algorithm to be
generated. Clustering[1] is a popular ML method that has
shown fantastic results in edge devices clustering . Table 2
summarizes the applications of unsupervised learning.

A. SUPERVISED LEARNING
Supervised learning[1] involves utilization of classified
dataset that have been assigned tags to influence the output
of a learning algorithm. The process of supervised learning
involves achieving an optimal solution of a problem by
using samples of the problem whose outputs are known to
train a learning algorithm of the problem Supervised
learning [1], [2] activities are also sub-divided into
classification and regression[8]. The concept of
classification defines the function of projecting the discrete
outcome of a tag class for input, whereas regression is the
method of determining the outcome of a continuous
quantity for a given example. For every possible case, we
can opt for a Model of low complexity. Amongst other ML
strategies with capabilities for wireless technology are k-
nearest neighbors [8], decision tree [15] and logistic
regression [8] . Table 1 offers a description of various C. REINFORCEMENT LEARNING
supervised learning models and the related implementation. Reinforcement learning describe a procedure where the
improvement of the performance is attained through several
trials with the output observed. The process is repeated
severally until optimization is achieved. Wireless networks
operate under instability in stochastic environments. In an
unpredictable environment, system dynamics is formulated
through Markov Decision Process (MDP).
MDP mathematical process modelling is employed to
maximize target objective. The learning object, often
referred to as the agent, is dealing with the environment.
When it is necessary to make a decision, the agent selects
the "a" which is the available action in the existing state
“s”. The system reacts to the action "a" by producing an
equivalent reward or penalty ,R(s , a) and then move to a
new state 0. For any chosen state s, and action a, based on
the Markov property, the probability of translating the state
is in the form P(s 0, a) and it does not depend on all the
former states and acts. RL also employ feedback
mechanism, although it is possess less insight when
3901029 3

compared to supervised learning where the agent is A. MOBILE BROADBAND ENHANCEMENT

provided with the correct action to adopt. Nevertheless, Strengthening the existing MBB service would allow new
when compared with unsupervised learning , it is more implementations with greater data rate requirements over a
insightful because the agent would not be assisted with any standardized coverage area. The basic specifications to
reasonable performance enhancing inputs in taking correct facilitate eMBB are set out below:
action. The approach employed by RL has been deployed
in several areas ( Table 3). Load balancing, management of 1) Peak Data Rate
resources, mobility, dynamic channel selection and This was conceived with the intention of increasing and
admission control are some of the application scheme of supporting users with high demand for data usage. For
RL in 5G wireless networks. instance, IMT-2020 is required to avail users with the data
rate capacity 20-time more than specification of all
previous technologies. For example , the current 4G
delivers 1 GB/s while 5g is expected to deliver 20GB/s. to
be able to deploy 5G networks, service providers require
access substantial amount of spectrum resources which will
bring about greater functional benefits of radio technology
of 5G. Multiple-Input Multiple-Output (MIMO) a integral
component for the realization of effectiveness in 5G. the
impact of ML MIMO technology can be seen in Channel
Estimation and Arrival Direction (DOA).Classification is
another ML element in the MIMO on the channel state
(CSI) for selecting indices for antennal optimization
III. ML AND 5G NEEDS employing supervised learning algorithm.
This section discusses the application of ML algorithms to 2) Users’ Experience
the specifications of 5G [1]. 5G is not an upgrade over 4G, Improving Quality of Experience(QofE) through Wireless
instead it is perceived as a major advancement of mobile Network Virtualization(WNV) [6] is another invention
communications technology with increased efficiency. brought about by 5G. the realization of (QofE) through
Such specifications are intended to ensure that IMT-2020 WNV heavily depend Network function Virtualization
provides greater versatility, security and reliability than in (NFV) and Software Define Network (SDN) in order to
the past. Technology, offering a broad range of resources meet the various related network standards. SDNs and NFV
and implementation instances for several application provides the platform for network programmability with
area[1]. In this segment, the key 5G requirements have unending possibilities that facilitate the automation of
been grouped into 3 basic services (figure 3) and network activity and management tasks through the
researched how ML can be used to achieve these services application of cognitive intelligence and ML
requirements. Every service emphasizes a specific subset algorithms[41]. Network resource allocation systems based
of specifications and applications relevant to such extent. on SDN and NFV have been developed to allow
The generic services are highlighted below. autonomous network and network resources management.
In this approach, systems utilize “Classifiers” to forecast
users’ need and dynamically distribute resources, topology
configuration, and most acceptable bitrate based on
communication efficiency (e.g. bandwidth) and also
improve users’ observed data rate.

3) Area traffic capacity

Unsupervised learning algorithm has been regarded as the
primary ML solution always employed to generating
Clusters in a Structured Multi-Point (CSMP) that focuses
on various features like load balancing, intercellular
interference mitigation and capacity enhancement. One
interesting developments for improving the efficiency
traffic is through Multi-Input Multi-Output (MIMO) and
Figure 2: Machine Learning for 5G in Mobile and Wireless comm. Synchronized Multi-Point (SMP) antenna systems. Spectra
effectiveness and greater coverage can be achieved through
eMMB Multi-antenna systems.
 4) Spectrum Efficiency 2) Energy Efficiency
The least acceptable specification for optimal efficiency in
spectrum utilization in IMT-2020 are 30bit/s/Hz(uplink) Significance of network energy efficiency for eMBB
and 15bit/s/Hz downlink. Currently, the spectrum capacity cannot be over emphasized and is projected to increase
in 4G is 3GHz but it is expected to extend to 30GHz in 5G. from 1x in 4G to 100x for IMT-2020. An area that requires
in order to achieve this target, there is need to utilize further research is how to reduce the amount of energy
flexible techniques that would maximize spectrum consumed by devices that are connected to mMTC
efficiency and effectiveness. Inadequacies in wireless applications. With the availability of wireless devices in 5G
network spectrum is being tackled by Adversarial Auto applications, developing energy-efficient wireless networks
Encoder(AAE) which is a routine of ML that detects is now a crucial and highly tasking problem for academia
inadequacies in wireless spectrum through the use of power and will remain a popular field of research in the
spectral density data present in unsupervised learning. foreseeable future .

C. ULTRA-RELIABLE LOW-LATENCY COMMUNICATIONS

Reliability, low latency, and mobility are the main priority
of future network services rather than data rates. The main
criteria for URLL communication are:
1) Latency
Latency is potentially one of the most important 5G
efficiency indicators. A stable 5G network requires
incredibly low latency, and only a few milliseconds (ms)
will make a big difference, making it an extremely
interesting area for 5G researchers and engineers alike.
There is a direct relationship between Latency , distance
and connectivity congestion. There is an increase in latency
with distance and congestion in the connections which is
why it is imperative to bring some storage closer to the end
user.
Figure 3 Fixed Services Categorization in 5G

2) Mobility
The maximum speed (in Km/h) a mobile station requires to
B. MASSIVE MACHINE TYPE COMMUNICATION achieve a defined QoS is referred to as Mobility. In order to
be able to assist highly mobile usage, introduction of
Another main aspect of 5G networking service is the need optimal identification of beamforming vector is of great
for flexible access for a growing number of wireless [6] importance. ML methods uses the uplink starting signal
devices, concentrating on efficient transmission rather than gotten from the BSs terminal and learn the definite
greater coverage. Smart homes, Internet of Things (IoT) is mapping function for the setup to foretell and coordinate
likely to cause irregular traffic between huge number of the beamforming vectors at the BSs. Networks that are
geographically distributed devices, requiring mMTC to be sensitive to mobility architecture also plays an important
able to accommodate new, but unexpected applications. role in the mutual contact of high-mobility network. Other
The two main requirements to allow mMTC are: promising implementations of ML to optimize wireless
Connection Density network energy efficiency network mobility include attribute extraction identify,
anticipate, and reduce interference-inflexible network
1) Connection Density implementations . Mobile networking is one of the most
An enormous obstacle for 5G networks involves linking a important elements for 5G mobile communications, but
vast number of devices to the internet, taking IoT and smart luckily, anticipatory mobility management by ML is
cities or homes or buildings to the standard prevalent in 5G. throwing up a new pathway to URLL networking.
The advent of IoT has resulted in a large volume of data
generated from sensors, devices and BSs .Traditional
dilemma present in association of cell resulting from IV. DECIDING ML DEPLOYMENT ON 5G / B5G
increase in user density was not effectively solved before
the advent of ML. Q-learning algorithms which is ML The implementation of 5G comes with a particular set of
technique is being touted to facilitate users’ BS service disputes that requires attention and comprehension before
selection by leveraging local information and learning initiating any project. The predominant concerns of ML
outcomes of adjacent users. will be summarized below.
3901029 5

A. IMPACT OF DATA
One of the fundamental parts in the application of ML is
the achievement of high-quality data. This is a major
consideration when selecting the king of learning to V. FUTURE DEVELOPMENT
employ specifically when it involves the deployment of 5G As the implementation of 5G technologies begins in
for applications. Data generation has been a major various parts of the world and its arrival widely celebrated,
challenge in wireless telecommunication and data Beyond 5G (B5G) [1] systems have received
simulation is not often encourage when applying ML massive attention and many research groups have recently
algorithm because ML learn from such data and most time begun investigating 6G taking into account the promising
doesn’t reflect real world scenario. features of 5G. As fixed service categorization is always
complicated and complex, the B5G is expected to resolve
B. NO FREE LUNCH THEOREM the problem of reliability and latency requirements of the
current eMBB, mMTC and uRLLC. In respect of energy
This concept in ML emphasizes that by averaging the efficiency, researchers anticipate 6G would assist ultra-long
outcome of data obtained from a collection in real world, battery life in order to reduce the number of times a device
every ML algorithm will give similar achievement from would require charging. As regards the continuous increase
deducing inference from unobserved data. specifically, it in the complication of network and the advent of multi-
has been observed that none of the ML algorithm is better service apps, it will be important to improve network
than the others, rather understanding the type of knowledge base in order to realize self- functionality. In the
distribution in a definite5G implementation and the ML future, all new spectrum will be exchanged, and so work
algorithm that most suit the data is of importance. into integrated usage of various frequencies will come into
play. ML is expected to exhibit a leading role in
C. CHOOSING HYPERPARAMETERS addressing a wide range of questions together with the
The collection of default values that are inputted into an optimization of the use of unlicensed spectrum,
algorithm before it commences training is referred to as opportunistic use of white spaces, adaptive smart contracts,
“HYPERPARAMETERS”. The final parameters of an adaptive leasing between carriers, etc.
algorithm is mostly influenced by the chosen
hyperparameters. For the fact that an ML algorithm VI. CONCLUSION
produces a magnificent output in a problem doesn’t imply
it will be efficient in another. Therefore, it is imperative to First and foremost, the report established how ML has
undertake a research into developing a customized ML shown great promise in importance as an intrinsic feature
algorithm that solves 5G challenges. of wireless communication. Nevertheless, it is worth noting
that the implementation of ML all around us may not be
possible at this time due to the cost, time, latency and delay
D. INTEROPRETABILITY AND ACCURACY in some ML algorithm. Therefore, the field of ML and 5G
Most times it is important to understand the rationale needs to improve greatly. Secondly, it was also established
behind the procedure BS employs in distributing network how difficult it could be s to actually ascertain the hurdle
resources amongst various users such that one gets more ML may add to a network so, extreme caution is
than the other. The most important thing is detecting the encouraged in its applications. Future research should take
optimal balance in a model that produces high level of up an important assessment of the trade-off between rising
interoperability and better accuracy. accuracy of the wireless connection using an approach
based on ML and interoperability of the method.
E. PERFORMANCE METRICS Conclusively, it will be of great interest as to the effect ML
technology will have in defining the future of wireless
After the deployment of an ML procedure or method in
network standards. Consequently, the fundamental question
solving particular real-world problem, it is important to
now is when will the full usage of ML to wireless
measure the level of achievement on untagged data.
communication system eventually take place.
Normally, the achievement metrics interpreted about the
task performed. The percentage of samples for which an
algorithm gives a correct result is referred to accuracy.
Also, performance information can be deduced from the
measurement of the error rate, which is the ratio of samples
from which the model produces wrong result. However, it
is more challenging deciding what to be measured than
chosen the metric to apply in measurement.
 IEEE INSTRUM. MEAS. MAG., VOL. 18, NO. 3, PP. 11–21,
JUN. 2015.
[12] X. CAO, L. LIU, Y. CHENG, AND X. S. SHEN,
References ‘‘TOWARDS ENERGY-EFFICIENT WIRELESS
NETWORKING IN THE BIG DATA ERA: A SURVEY,’’ IEEE
[1] M. Morocho-Cayamcela, H. Lee and W. Lim, "Machine
Learning for 5G/B5G Mobile and Wireless COMMUN. SURVEYS TUTS., VOL. 20, NO. 1, PP. 303–332,
Communications: Potential,Limitations,andFuture 1ST QUART., 2018.
Directions", IEEE Access, vol. 7, pp. 137184-
137206,2019.Available: 10.1109/access.2019.2942390. [13] Y. MAO, C. YOU, J. ZHANG, K. HUANG, AND K. B.
LETAIEF, ‘‘A SURVEY ON MOBILE EDGE COMPUTING: THE
[2] C. Jiang, H. Zhang, Y. Ren, Z. Han, K. Chen and L. COMMUNICATION PERSPECTIVE,’’ IEEE COMMUN.
Hanzo, "Machine Learning Paradigms for Next- SURVEYS TUTS., VOL. 19, NO. 4, PP. 2322–2358, 4TH
Generation Wireless Networks", IEEE Wireless QUART., 2017.
Communications, vol. 24, no. 2, pp. 98-
105,2017.Available:10.1109/mwc.2016.1500356wc.
[14] K.-L. A. YAU, P. KOMISARCZUK, AND P. D. TEAL,
‘‘REINFORCEMENT LEARNING FOR CONTEXT AWARENESS
[3] M. Eugenio, M. Cayemcela and W. Lim, "Artificial AND INTELLIGENCE IN WIRELESS NETWORKS: REVIEW,
Intelligence in 5G Technology: A Survey - IEEE NEW FEATURES AND OPEN ISSUES,’’ J. NETW. COMPUT.
Conference Publication", Ieeexplore.ieee.org, 2018. APPL., VOL. 35, NO. 1, PP. 253–267, 2011.
[Online].Available:https://ieeexplore.ieee.org/document/
8539642. [Accessed: 12- Mar- 2020]
[15] G. C. SOBABE, Y. SONG, X. BAI, AND B. GUO, ‘‘A
[4] E. BALEVI AND R. D. GITLIN, ‘‘UNSUPERVISED MACHINE COOPERATIVE SPECTRUM SENSING ALGORITHM BASED ON
LEARNING IN 5G NETWORKS FOR LOW LATENCY UNSUPERVISED LEARNING,’’ IN PROC. 10TH INT. CONGR.
COMMUNICATIONS,’’ IN PROC. IEEE 36TH INT. PERFORM. IMAGE SIGNAL PROCESS., BIOMED. ENG. INFORM. (CISP-
COMPUT. COMMUN. CONF. (IPCCC), DEC. 2017, PP. 1–2. BMEI), VOL. 1, OCT. 2017, PP. 198–201.

[5] K.-L. A. YAU, P. KOMISARCZUK, AND P. D. TEAL,

‘‘REINFORCEMENT LEARNING FOR CONTEXT AWARENESS
AND INTELLIGENCE IN WIRELESS NETWORKS: REVIEW,
NEW FEATURES AND OPEN ISSUES,’’ J. NETW. COMPUT.
APPL., VOL. 35, NO. 1, PP. 253–267, 2011.

[6] S. WANG, X. ZHANG, Y. ZHANG, L. WANG, J. YANG, AND

W. WANG, ‘‘A SURVEY ON MOBILE EDGE NETWORKS:
CONVERGENCE OF COMPUTING, CACHING AND
COMMUNICATIONS,’’ IEEE ACCESS,

[7] G. BERARDINELLI, N. H. MAHMOOD, I. RODRIGUEZ,

AND P. MOGENSEN, ‘‘BEYOND 5G WIRELESS IRT
FOR INDUSTRY 4.0: DESIGN PRINCIPLES AND
SPECTRUM ASPECTS,’’ IN PROC. IEEE GLOBECOM
WORKSHOPS (GC WKSHPS), DEC. 2018, PP. 1–6

[8] J. QUINLAN, "LEARNING DECISION TREE

CLASSIFIERS", ACM COMPUTING SURVEYS (CSUR),
VOL. 28, NO. 1, PP. 71-72, 1996. AVAILABLE:
10.1145/234313.234346. JCDL05, PP. 100-109.

[9] VAPNIK, V.N. (1998). THE NATURE OF STATISTICAL

LEARNING THEORY (2ND ED., PP. 23–57).
[10] A. JIVANI, K. SHAH, S. KOUL AND V. NAIK, "THE ADEPT
NEAREST NEIGHBOUR ALGORITHM, AN OPTIMIZATION
TO THE CONVENTIONAL K-NEAREST NEIGHBOUR
ALGORITHM", TRANSACTIONS ON MACHINE LEARNING AND
ARTIFICIAL INTELLIGENCE, VOL. 4, NO. 1, 2016.
AVAILABLE: 10.14738/TMLAI.41.1876
[11] E. HOSSAIN AND M. HASAN, ‘‘5G CELLULAR: KEY
ENABLING TECHNOLOGIES AND RESEARCH CHALLENGES,’’