International Journal of Computer Trends and Technology Volume 68 Issue 11, 43-48, November 2020

ISSN: 2231 – 2803 / https://doi.org/10.14445/22312803/IJCTT-V68I11P105 © 2020 Seventh Sense Research Group®

Original Article

Machine Learning (ML) in a 5G Standalone

(SA) Self Organizing Network (SON)
Srinivasan Sridharan1
1
Principal Engineer T-Mobile USA Inc – Senior Engineer, IEEE.

Received Date: 03 October 2020

Revised Date: 14 November 2020
Accepted Date: 17 November 2020

Abstract - Machine learning (ML) is included in Self- cases through new devices. 5G Standalone is considered a
organizing Networks (SONs) that are key drivers for key enabler by the service providers as it improves the
enhancing the Operations, Administration, and Maintenance efficacy of the throughput that edges the network. It also
(OAM) activities. It is included in the 5G Standalone (SA) assists in advancing new cellular use cases like ultra-reliable
system is one of the 5G communication tracks that low latency communications (URLLC) that supports
transforms 4G networking to next-generation technology combinations of frequencies. However, the benefits of 5G get
that is based on mobile applications. The research's main restricted as its implementation along the devices cannot be
aim is to an overview of machine learning (ML) in 5G estimated (Andrews, Buzzi, Choi, Hanly, Lozano, Soong &
standalone core networks. It was found that 5G intentions Zhang, 2014).
revere heterogeneous demands of phrases of data-rate,
reliability, latency, or efficiency. Mobile operators shall be
Moreover, to provide 5G coverage includes developing a
in a position in imitation of revere whole of these
new infrastructure that is time-consuming and expensive.
requirements using shared network infrastructure’s
resources. 5G Standalone is considered a key enabler by the The implementation of Machine Learning (ML) in the 5G
service providers as it improves the efficacy of the Standalone System increases network services' efficacy by
throughput that edges the network. It also assists in enabling dynamic allocation through network slicing. The
advancing new cellular use cases like ultra-reliable low ML inclusion in the 5G Standalone System helps in real-time
latency communications (URLLC) that supports source administration of the virtual network functions and
combinations of frequencies. Radio Access Network. It also facilitates advanced analytics
of the network data that helps in improving the performance
Keywords - 5G, Machine learning (ML), Self-organizing of the network. ML in 5G Standalone System optimizes
Networks (SONs), 5G Standalone, Artificial Intelligence (AI) services along with the mobile networks and Verizon's fixed
networks. It also contributes to developing cost-efficient and
I. INTRODUCTION scalable architecture that helps move the workloads to use
The 5G Standalone (SA) system is one of the 5G case requirements. However, crucial computational issues
communication tracks that transforms 4G networking to such as supervised, unsupervised, or reinforcement learning
next-generation technology based on mobile applications. problems of ML techniques for 5G networks impede ML
The 5G applications have been in practice since its release in techniques' application to 5G Standalone networks (Mumtaz,
2017 after the standard specification was provided by the Huq & Rodriguez, 2014). The current research examines the
global mobile communications standards organization 3GPP. facts related to machine learning (ML) in 5G standalone core
The 5G networks released in the form of Non-standalone networks. It provides valuable information about the
(NSA) majorly focused on bringing mobile broadband challenges faced in applying ML techniques to 5G
improvements. It enhanced data-bandwidth capacity and technologies.
promoted reliable connections (Mars, Abadleh & Adi, 2019).
II. LITERATURE REVIEW
While focusing on the 5G Standalone, it is based on 5G According to Moysen & Giupponi (2018), machine
architecture and uses simplified RAN to establish learning (ML) is included in Self-organizing Networks
connections. It also includes other technologies such as (SONs) that are key drivers for enhancing the Operations,
cloud-native 5G Core, ultra-low latency, and device Administration, and Maintenance (OAM) activities. SON
architecture to provide 5G coverage for low band. It also simplifies operational tasks that reduce the cost of installing
network-slicing functionaries and facilitates wider use of the 4G and 5G networks. However, to meet the 5G network

This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

 Srinivasan Sridharan et al. / IJCTT, 68(11), 43-48, 2020

management requirements, it is essential to extend the the low latency expected for 5G and creates dynamic
autonomous management vision to end network support. models. Lateef, Imran & Abu-Dayya (2013) analyzed that
Under such conditions, when ML is included in SON, it acts SON has several encapsulating functionalities because of
as an efficient tool to facilitate autonomous adaptation and which it is widely accepted in 3G and 4G SON products.
helps in the efficient decision-making process. For example, However, due to this approach's application, there is a
when ML-enabled SON is included in the Mobility Load creation of network operators' transparency and control.
Balancing (MLB), there is an improvement in the Quality of Another major issue with 5G is that it requires a faster SON
Service (QoS) and an increase in capacity. compared to the SON functionaries in 3G and 4G. It requires
On the other hand, ML-enabled SON enhances the converting the SON paradigm from reactive to proactive
Coverage and Capacity Optimization (CCO) functionary by mode so that there is support to 5G requirements.
optimizing the coverage and edge throughput. ML also
improves Inter-Cell Interference Coordination (ICIC) with Bashir, Arul, Basheer, Raja, Jayaraman & Qureshi
the help of UE and OI measurements. It helps in transmitting (2019) examined that 5G network includes several
power and Coordinating ABBS. It also minimizes intercell technologies such as Mobile Edge Computing (MEC),
interference and provides improved antenna parameters. Software Defined Networking (SDN), and Multiple Input
Amirijoo, Frenger, Gunnarsson, Kallin, Moe & Zetterberg and Multiple Output (MIMO) to work properly. It cannot
(2008) analyzed that ML-enabled SON such as E-UTRA function under the existing system such as Base Band Units
Control plane protocols and interfaces help controlling (BBUs), Radio Units (RUs), and Radio Access Networks
information and to regularize the short-term network (RANs) as they do not support the colossal requirements of
operations. It also facilitates QoS, session set-up and the 5G system. Therefore, machine learning algorithms are
establishes connected mobility mode and radio resource included in mobile networks to improve connectivity to
control. support 5G. It led to the development of C-RAN
architectures that perform as intelligent transmitters by
Imran, Zoha & Abu-Dayya, (2014) analyzed that certain including radio propagation medium. C-RAN with ML
challenges such as underutilized intelligence, self- forms a new design that facilitates the development of the
coordination issues, and lack of transparency are faced while 5G network architecture. It includes adopting Multi-tier H-
enabling SON for 5G. Aliu, Imran, Imran & Evans (2012) CRAN ML-based techniques to address the issues related to
examined that a generic method is followed in implementing the BBU traffic pool. The proposed architecture helps in
2G, 3G, 4G SON. improving end-to-end optimization, augmented granularity,
and QoS. Chabbouh, Rejeb, Choukair & Agoulmine (2016)
examined that interlinking of cloud technology in RAN
leads to C-RAN formation. When C-RAN is included in the
HetNets, there is the achievement of flexibility and
reconfiguration in small cells. As a result, there is an
improvement in the terminals' information exchange
process, creating fairness in QoS parameters such as
throughput.

ML is applied to different 5G network applications such

as massive MIMO channels, location prediction, and
channel estimation. For instance, ML is applied in
information-theoretic regression and decision tree models to
recognize radio access networks' issues. O’Shea & Hoydis
(2017) examined that the deep learning approach is
associated with categorization of modulation to achieve
Fig.1 KPI Analysis in Self Organizing Networks (SON) in 5G competitive precision. It was also analyzed that deep neural
networks help optimize the algorithm related to wireless
SON's solutions are currently based on Spatio-temporal network resource administration. Deep learning algorithms
knowledge in which partial information is available about such as deep Q networks play a major role in improving
the problem. For example, information related to coverage gaming options, especially Atari games. The integration of a
holes and congestion spot locations is presumed to be known deep Q network helps enhance algorithm and high-
by the SON system. On the other hand, in state-of-the-art dimensional sensory input workings so that high levels of
networks, learning is gained through logs of client human performance are achieved in Atari games. (30)
grievances or operation, and maintenance center (OMC) analyzed that Q-learning is applied to maximize revenue and
reports. However, the approach is inefficient in providing increase flexibility to change environments. It helps in

44
 Srinivasan Sridharan et al. / IJCTT, 68(11), 43-48, 2020

achieving optimal performance by addressing issues related

to admission control for 5G architecture. It includes
developing 5G network models by incorporating eMBB,
mMTC, and URLLC to develop adequate 5G infrastructure.
The adoption of eMBB helps meet flexibility requirements
for resource usage, and URLLC helps meet capacity
requirements.

When deep Q-learning is integrated with Network

Function Virtualization (NFV), there is a dynamic reliability
service provision application. Zhang, Yao, & Guan (2017)
examined the use of cloud technology in the resource
management framework and found that it led to deep Fig.2 U-RAN Drones in an Internet of Things (IoT) based 5G SA
learning reinforcement. The integration of Virtual Network Network.
Function (VNF) in the GPU server also helps analyze and The automatic clustering process and MLP-LSTM
classify packets and applications that led to the incorporation approach also impact the application of ML tools in the
of deep neural networks in forming a resource management positioning of UAV. Considering different environments
framework. Additionally, Q-learning is also included in the such as urban, suburban, and rural areas also impacts
SFC deployment process to resolve issues related to service UAV's outdoor and indoor positioning.
mapping. Using different applications such as NFV, VNF,
and Q-learning helps eliminate the issues related to VNF
resource allocation and implementation of machine learning Bithas, Michailidis, Nomikos, Vouyioukas & Kanatas
in the resource management framework. The study also (2019) analyzed that when ML techniques are applied to the
emphasized that ML in the network function placement will U-RAN, it leads to improvements in the functional and
help identify the best placement essential for VNF that design aspects of the channel survey. ML adoption in U-
facilitates SFC formation. The Ml techniques also help RAN also enhances the security management process,
acquire previous learning about the placements, because of channel modeling, positioning, and resource management
which it becomes easy to make comparative decisions aspects. Moreover, ML approaches are based on wireless
regarding future server selection for placements. sensor networks that help in improving application and
Implementing machine learning also helps acquire networking parameters. It also helps in resolving sensor-
information about the operational constraints required to based applications and enhances the use of UAVs in BSs.
meet the best server and QoS requirements. Nishiyama, Kawamoto & Takaishi (2017) analyzed that
Network function virtualization (NFV) is responsible for
virtualizing the network functions that serve as switches and
Machine learning tools efficiently extract hidden storage devices. It also includes using software-defined
information from the available data and provide facts networking (SDN) in NFV to control the control data planes.
related to several parameters that could not be acquired SDN helps in simplifying the reconfiguring and network
using other model-based approaches. ML-based models functioning as it includes using flow-based networking. It is
can simplify the data and be easily incorporated into also used in the cellular network functioning to facilitate
different metrics like energy consumption of UAVs. It radio resource and mobility administration system
helps in acquiring valuable information about transmitted performance. It also plays an important role in enhancing the
and received facts in a single feature. The positioning of capability of the U-RANs by establishing a load balance in
UAV is essential to optimize the usage of DL tools. the multitier U-RANs. It was also ascertained that when U-
RANs are implemented with cloud computing technologies,
It also includes implementing long short-term memory there is the conduction of remote computing and processing.
(LSTM), and Multi-layer perceptron (MLP) based It reduces the major cloud center load and allocates the
approaches to determine the optimized position of UAV damaged parts to the terrestrial networks in natural disasters.
and user throughput.
U-RANs, when implemented with the integration of cloud
computing and hierarchical SDN controllers, develop
network-based architectures. The adoption of hierarchical
SDN controllers helps in the granular administration of the
UAV cells related to routing procedures.

45
 Srinivasan Sridharan et al. / IJCTT, 68(11), 43-48, 2020

Foerster, Assael, De Freitas & Whiteson (2016) consumers' and producers' concept. Alawe, Ksentini,
examined that ML algorithms execute keep categorized as Hadjadj-Aoul, Bertin & Kerbellec (2017) examined that 5G
much supervised, unsupervised, yet reinforcement learning. core functions include core AMF, Session Management
In a supervised study (SL), the goal conforms with Function (SMF), Policy Control Function (PCF), User Plane
mechanically accumulate a mapping besides the center to the Function (UPF), and Unified Data Management (UDM).
output, whose right values (referred in conformity with as
like labels) are provided through a director for the duration
AMF is based on the UE authentication and mobility
of a coaching phase. In absolutely complicated scenarios
management that help assess several technologies by
with a couple of UAVs, even are plenty about parameters so
connecting to a lone AMF. On the other hand, SMF is
much perform lie considered as entering records training
related to the allocation of IP addresses and session
because of SL. In the litigation over SL, an ML container
management. It is responsible for controlling and selecting
can stand ancient base as much enter the features and
UPF so that there is a secure transfer of data. When UE has
variables related to the defined trade-offs over the
several sessions, there is an allocation of SMEs in each
community. A ruin feature can be described based on the
session to manage individuality and ensure the provision of
given scenario. In unsupervised learning, in that place is no
different sessions as per the session's need.
certain supervisor, then the algorithm must autonomously
discover similarities, patterns, yet differences in the
reachable statistics except somebody previously training.
However, an adequate aggregation about coaching statistics
may additionally now not stay available in deep cases. In
incomplete applications, the output concerning a dictation is
a sequence of actions.

Xu, Yin, Xu, Lin & Cui (2019) analyzed that 5G

perform to keep a solution enabler in imitating pressure the
ML then AI integration into the network edge. The
determine below suggests what 5G allows simultaneous
connections following a couple of IoT units generating
tremendous quantities concerning data.5G, deployed using
mm-wave, has beam-based mobile phone coverage
compared to 4G as has sector-based coverage. A computer
discovered algorithm can help the 5G cell website online in
conformity with count a engage of postulant beams, made
both out of the attention or its close telephone site. A perfect
accept is the set up to expectation includes fewer beams and Fig. 3 5G SA Core Network Functions (NFs)
has a high probability of containing the superior beam. The
superior filament is the filament, together with the very best
sign energy, a.k.a. RSRP. The extra activated beams present, Hochreiter, S. Schmidhuber (1997) analyzed the use of
the greater the chance of discovering the excellent beam, two Neural-network based techniques such as Deep learning
although the higher variety regarding activated beams Neural Network (DNN) and Long Short Term Memory cell
increases the system's useful resource consumption. The (LSTM) in the traffic forecasting process that included 5G
consumer tools (UE) measures yet reviews the candidate network system. In the case of DNN, it was found that it
beams in imitation of the attention telephone site, which helped to inactivate a specific neuron that contributed to
desire afterward decide agreement the UE wishes in multiple inputs. As a result, it did not help in the
imitation of lie handed on after a near phone website yet in identification of any real traffic pattern. On the other hand,
conformity with as candidature beam. LSTM was based on a recurrent neural network that
successfully received traffic inputs. It included the use of the
III. LEVERAGING ML TECHNIQUES FOR 5G memory and omission mechanism that helped in managing
Machine learning-based scalable mechanism for 5G and recording traffic patterns.
core network found that it is associated with scalability Additionally, Neural networks-based can be used to
issues of the Access and Mobility Management Function make predictions about the load class in the successive
(AMF). 5G architecture includes two main aspects, which duration. It includes two techniques, which are DNN and
are 3GPP communication interferences and service-based RNN. While focusing on DNN, it includes three layers: the
architecture. In the 5G network, communication is input layer, the middle layer, and the output layer. It is also
established with a common bus's help by developing known as Feed Forward Neural Network (FFNN), in which

46
 Srinivasan Sridharan et al. / IJCTT, 68(11), 43-48, 2020

the data is preceded from the input layer and passed through unsupervised learning includes the observations were
the middle layer before it finally reaches the outer layer. In presented only the input values. It finds similarities among
the second technique which RNN is used as an efficient these values in conformity with similar group data between
neural network. RNN is composed of loops that inject the clusters. The algorithm does not target any moving in
previous data and forecast future data. accordance with estimate capability and label associated
5G cell networks arrive with many recent services with information factors. This algorithm is used because
compared after the legacy cellular networks, certain as much organizing the crew's records about clusters to draw its
network facts analytics characteristic (NWDAF), as enables shape that is bunch the records to reveal meaningful
the community operators in imitation of either implement partitions then hierarchies. It makes statistics look simple yet
their very own computing device learning (ML) primarily equipped because of analysis. Examples concerning
based information analytics methodologies or combine third- unsupervised discipline include K-means, Fuzzy clustering,
party solutions after their networks. In that paper, the shape Hierarchical clustering.
and the protocols of NWDAF to that amount are defined in On the other hand, semi-supervised learning is based on
the third manufacture Partnership Project (3GPP) honor the observations in the form concerning input-output pairs.
documents are preceding described. Then, cell-based However, the outputs values are no longer regarded between
synthetic statistics employ for 5G networks primarily based a full-size total of observations. The motive is following
over the fields defined with the aid of the 3GPP specs is usage these recent following improve the concept modeling
generated. Further, some anomalies are added to its records among supervised discipline.
engage (e.g., suddenly increasing traffic in a precise cell).
Then, these anomalies inside each cell, subscriber category, VI. CONCLUSION
and user equipment are classified. Afterward, three ML As per the above-discussed facts, it can be said that
models, namely, linear regression, long-short term memory, presently, the academic and pragmatic communities are
or recursive neural networks, are implemented to lesson rite collaborating in conformity because of future 5G networks.
data addition (e.g., anomalies among the community traffic) The vast aggregation concerning associated literature
yet network lay count applications about NWDAF. For the foresees advancements that may be leading according to
account regarding community load, ternary exclusive models better community capability and decreased delay, new ways
are old to limit the low utmost error, which is deliberated of the spectrum get admission to and sharing, the
employing subtracting the genuine generated data besides applicability of virtualization then software program
the mannequin account value. For the classification of described networks (SDN) concepts, and the development
anomalies, twain ML fashions are back according to the about latter verbal exchange requirements (e.g., mmWave)
increased area under the recipient working characteristics and Massive MIMO methods. The study found that 5G
curve. Namely, logistic regression and severe gradient intentions revere heterogeneous demands of phrases of data-
boosting. According to the simulation results, neural rate, reliability, latency, or efficiency. Mobile operators shall
network algorithms fail linear regression into community be in a position in imitation of revere whole of these
burden prediction since the tree-based gradient boosting requirements using shared network infrastructure’s resources.
algorithm outperforms logistic regression into exclusion It includes resource orchestration for the 5G network
detection. These estimations are expected after enlarge the enforcing IV Quality as service pillars. It helps in traffic
overall performance concerning the 5G community through classification by committed logical digital networks referred
NWDAF. to as Network Slices (NSs). To optimally adore multiple
Machine learning can be classified into three main NSs, an equal physical network is implemented with
aspects, such as supervised learning, unsupervised learning, Machine Learning (ML). It helps address demands change
and semi-supervised learning. While focusing on supervised dynamically that serves as a mere recursive optimization
learning, it includes the observations that are partial in shape leading in the foremost slice direction. It includes using
concerning input-output pairs. It analyzes a function among regression bushes, namely an ML-based strategy for both
the inputs and outputs. Machine lesson algorithm makes classification and prediction, outdo lousy alternative
predictions about employ pattern in the form of supervised solutions within phrases concerning prediction truth then
learning algorithms searches. The supervised learning throughput (Perez, Jayaweera & Lane, 2017).
algorithm includes an outcome alternative that anticipates a
given set of predictors that are impartial to variables. Using REFERENCES
this set of variables, there is the creation of a function based [1] Alawe, I., Ksentini, A., Hadjadj-Aoul, Y., Bertin, P., & Kerbellec, A.
(2017, September). “On evaluating different trends for the virtualized
on favored outputs. The supervised learning networks are and sdn-ready mobile network”. In 2017 IEEE 6th International
Neural Networks, Regression, Decision tree, k-Nearest Conference on Cloud Networking (CloudNet) (pp. 1-6). IEEE.
Neighbor, Logistic Regression, Support Vector Machine, [2] Aliu, O. G., Imran, A., Imran, M. A., & Evans, B. (2012). “A survey
Naive Bayes, etc. of self-organization in future cellular networks”. IEEE
Communications Surveys & Tutorials, 15(1), 336-361.
Khorsandroo, Noor & Khorsandroo (2013) analyzed that

47
 Srinivasan Sridharan et al. / IJCTT, 68(11), 43-48, 2020

[3] Alnoman A, Anpalagan A (2017) “Towards the fulfillment of 5G [20] Moysen, J., & Giupponi, L. (2018). “From 4G to 5G: Self-organized
network requirements: technologies and challenges”. Telecommun network management meets machine learning”. Computer
Syst 65(1):101–116 Communications, 129, 248-268.https://arxiv.org/pdf/1707.09300.pdf
[4] Amirijoo, M., Frenger, P., Gunnarsson, F., Kallin, H., Moe, J., & [21] Mumtaz, S., Huq, K. M. S., & Rodriguez, J. (2014). “Direct mobile-to-
Zetterberg, K. (2008, May). “Neighbor cell relation list and physical mobile communication: Paradigm for 5G”. IEEE Wireless
cell identity self-organization in LTE”. In ICC Workshops-2008 IEEE Communications, 21(5), 14-23.
International Conference on Communications Workshops (pp. 37-41). [22] Dr.V.V.Narendra Kumar, T.Satish Kumar, "Smarter Artificial
IEEE. Intelligence with Deep Learning" SSRG International Journal of
[5] Andrews JG, Buzzi S, Choi W, Hanly SV, Lozano A, Soong AC, Computer Science and Engineering 5.6 (2018): 10-16.
Zhang JC (2014) “What will 5G be?” IEEE J Sel Areas Commun [23] Nadeem Q-U-A, Kammoun A, Alouini M-S (2018) “Elevation
32(6):1065–1082 beamforming with full dimension mimo architectures in 5G systems: a
[6] Andrews, J. G., Buzzi, S., Choi, W., Hanly, S. V., Lozano, A., Soong, tutorial”. arXiv preprint arXiv:1805.00225
A. C., & Zhang, J. C. (2014). “What will 5G be?”. IEEE Journal on [24] Nishiyama, H., Kawamoto, Y., & Takaishi, D. (2017, September).
selected areas in communications, 32(6), 1065-1082. “On OFDM-based resource allocation in LTE radio management
[7] Bashir, A. K., Arul, R., Basheer, S., Raja, G., Jayaraman, R., & system for uncrewed aerial vehicles (UAVs)”. In 2017 IEEE 86th
Qureshi, N. M. F. (2019). “An optimal multitier resource allocation of Vehicular Technology Conference (VTC-Fall) (pp. 1-5). IEEE.
cloud RAN in 5G using machine learning”. Transactions on Emerging [25] Zhang, Y., Yao, J., & Guan, H. (2017). “Intelligent cloud resource
Telecommunications Technologies, 30(8), e3627. management with deep reinforcement learning”. IEEE Cloud
[8] Bithas, P. S., Michailidis, E. T., Nomikos, N., Vouyioukas, D., & Computing, 4(6), 60-69.
Kanatas, A. G. (2019). “A survey on machine-learning techniques for [26] Srinivasan Sridharan. (, 2020). “A Literature Review of Network
UAV-based communications. Sensors”, 19(23), 5170. Function Virtualization (NFV) in 5G Networks”. International Journal
[9] Chabbouh, O., Rejeb, S. B., Choukair, Z., & Agoulmine, N. (2016, of Computer Trends and Technology, 68.10, 49-55.
May). A novel cloud RAN architecture for 5G HetNets and QoS [27] https://patents.google.com/patent/US20160088677A1/en
evaluation. In 2016 International Symposium on Networks, [28] https://patents.google.com/patent/US9972037B1/en
Computers, and Communications (ISNCC) (pp. 1-6). IEEE. [29] https://patents.google.com/patent/US20160088678A1/en
[10] Foerster, J., Assael, I. A., De Freitas, N., & Whiteson, S. (2016). [30] Srinivasan Sridharan. (, 2020). “5G Cloud Network Resource Slicing
“Learning to communicate with deep multi-agent reinforcement – A Literature Review”. International Research Journal of Engineering
learning”. In Advances in neural information processing systems (pp. and Technology (IRJET) Volume 7, Issue 11, November 2020
2137-2145). https://irjet.net/volume7-issue11 S.No: 01
[11] Hochreiter, S. JA1 4 rgen Schmidhuber (1997).“Long Short-Term [31] https://www.researchgate.net/publication/345225295_5G_Cloud_Net
Memory”. Neural Computation, 9(8). work_Resource_Slicing_-A_Literature_Review
[12] Hossain E, Hasan M (2015) “5G cellular: key enabling technologies [32] https://www.researchgate.net/publication/344979015_A_Literature_R
and research challenges”. IEEE Instrum Meas Mag 18(3):11–21 eview_of_Network_Function_Virtualization_NFV_in_5G_Networks
[13] Imran, A., Zoha, A., & Abu-Dayya, A. (2014). “Challenges in 5G: [33] https://www.academia.edu/44397850/A_Literature_Review_of_Netw
how to empower SON with big data for enabling 5G”. IEEE Network, ork_Function_Virtualization_NFV_in_5G_Networks
28(6), 27-33. [34] https://www.academia.edu/44423417/5G_Cloud_Network_Resource_
[14] Jiang C, Zhang H, Ren Y, Han Z, Chen K-C, Hanzo L (2017) Slicing_A_Literature_Review
“Machine learning paradigms for next-generation wireless networks”. [35] O’Shea, T., & Hoydis, J. (2017). “An introduction to deep learning
IEEE Wirel Commun 24(2):98–105 for the physical layer”. IEEE Transactions on Cognitive
[15] Khorsandroo, S., Noor, R. M., & Khorsandroo, S. (2013). “A generic Communications and Networking, 3(4), 563-575.
quantitative relationship to assess the interdependency of QoE and [36] Perez, J. S., Jayaweera, S. K., & Lane, S. (2017, June). “Machine
QoS”. KSII Transactions on Internet and Information Systems (TIIS), learning aided cognitive RAT selection for 5G heterogeneous
7(2), 327-346. networks”. In 2017 IEEE International Black Sea Conference on
[16] Lateef, H. Y., Imran, A., & Abu-Dayya, A. (2013, September). “A Communications and Networking (BlackSeaCom) (pp. 1-5). IEEE.
framework for the classification of self-organizing network conflicts [37] Sultan K, Ali H, Zhang Z (2018) “Big data perspective and challenges
and coordination algorithms”. In 2013 IEEE 24th Annual International in next-generation networks”. Future Internet 10(7):56
Symposium on Personal, Indoor, and Mobile Radio Communications [38] Valente KP, Imran MA, Onireti O, Souza RD (2017) “A survey of
(PIMRC) (pp. 2898-2903). IEEE. machine learning techniques applied to self-organizing cellular
[17] Ma Z, Zhang Z, Ding Z, Fan P, Li H (2015) “Key techniques for 5G networks”. IEEE Commun Surv Tutor 19:2392–2431
wireless communications: network architecture, physical layer, and [39] Valente KP, Imran MA, Onireti O, Souza RD (2017) “A survey of
mac layer perspectives”. Sci China Inf Sci 58(4):1–20 machine learning techniques applied to self-organizing cellular
[18] MacCartney GR, Zhang J, Nie S, Rappaport TS (2013) “Path loss networks”. IEEE Commun Surv Tutor 19:2392–2431
models for 5G millimeter-wave propagation channels in urban [40] Wei L, Hu RQ, Qian Y, Wu G (2014) “Enable device-to-device
microcells.” In: 2013 IEEE global communications conference communications underlaying cellular networks: challenges and
(GLOBECOM), pp 3948–3953 research aspects”. IEEE Commun Mag 52(6):90–96
[19] Mars, A., Abadleh, A., & Adi, W. (2019). “Operator and Manufacturer [41] Xu, Y., Yin, F., Xu, W., Lin, J., & Cui, S. (2019). “Wireless traffic
Independent D2D Private Link for Future 5G Networks.” arXiv prediction with scalable Gaussian process: Framework, algorithms,
preprintarXiv:1911.00303.https://arxiv.org/ftp/arxiv/papers/1911/1911 and verification”. IEEE Journal on Selected Areas in Communications,
.00303.pdf 37(6), 1291-1306.

48