White Paper

Ignite 5G Network Analytics

with Machine Learning &
Artificial Intelligence

Network data is a rich source of

insights that can be used to make
highly informed and impactful business
decisions that not only improve the
customer experience, but also assist
with new product development.
However, as more and more 5G
services come online, the demands
of consuming, processing, and taking
action with that data have outpaced
traditional databases and now require
sophisticated data visualization
tools and applications. Time-series
databases and Artificial Intelligence/
Machine Learning (AI/ML) are needed
to build, train, and interpret network
models capable of understanding
and emulating human experiences at
machine scale.
Whitepaper Ignite 5G Network Analytics with Machine Learning & Artificial Intelligence

Insert headline
Why do network operators need
analytics powered by AI/ML?
Over time, while network models Artificial Intelligence
and automated recovery reduce
the number of catastrophic events,
no human being can perform the Responsive
kind of atomized and granular Actions
network monitoring that complex
Train & Analyze
5G ecosystems create. Providers
Models
need AI/ML so they can focus their
time and energy on ef ficiencies, Neural Network
directional items, and customer Models
experience while the AI system Deep Learning Machine Learning
manages typical operational tasks
with a neural network brain. Figure 1: Network operators need AI/ML

Monetizing network data with multivendor,

multilayer analysis
To monetize network data and create business value, Beyond building customer loyalty
providers need AI/ML solutions that facilitate management through improved service quality,
of the entire data pipeline of the multilayer, multivendor
network. That management encompasses ingestion, data new use cases make it possible to
processing, clustering, model training, and automation for 5G monetize network data and create
RAN, edge and mobile core, as well as optical transport and
the MPLS network. This comprehensive form of network data
business value by aligning network
analytics yields much more sophisticated insights into pricing value directly to revenue. This can be
and consumer segmentation as it relates to network usage, a crucial dif ferentiator for new B2B
enabling dynamic, on-demand capabilities such as:
and B2C service of ferings.
• Partitioning and prioritizing traf fic
• Optimizing throughput and transmission speeds
• Managing high-traf fic users and their impact on other
customers’ service quality
• Handof f capabilities between networks
• Managing low- and high-density geographies dif ferently
• Pricing structures based on dif ferent types of traf fic

© 2023 Fujitsu. Proprietary and Confidential. All Rights Reserved. Page 2 of 6

Whitepaper Ignite 5G Network Analytics with Machine Learning & Artificial Intelligence

Building the business case for AI/ML

To get started with AI/ML in a provider network, keep the business case simple. Learn to look for
problems and then use that knowledge to solve future problems faster and more economically. And it’s
easier than you might think. Here are five real-world examples of how a network operator can harness
AI/ML to manage their network.

Advanced anomaly detection

Use a digital twin, also known as a neural network model, and AI/ML techniques to identify events,
behaviors, and observations that vary significantly against steady-state network data. Artificial neural
networks excel at finding and learning data patterns by modeling complex relationships between network
inputs and outputs. Imagine network inputs that include network Operations and Maintenance (OAM) data
and network outputs like analysis for root cause and reinforcement learning. Those outputs can enrich
other programs, like ticketing systems, personnel notifications, and updated orchestration workflows.

Predictive network planning

With AI/ML tools, the right network expansion in the right place with the right equipment can deliver
higher ROI. Predictive network planning allows providers to reduce downtime and increase revenue by
pinpointing exactly where network congestion and bottlenecks occur, and then automatically allocating
new resources, either with idle capacity or with new equipment recommendations that match the need.
Providers and operators can drive revenue growth by optimizing their CAPEX budgets for customer
experience and new service delivery, rather than relying on business-blind metrics like capacity and reach.

Consistent problem solving

Solutions to complex network problems should not depend on the experience of the person who
catches the ticket. That makes improving the consistency of problem-solving a significant goal of every
Network Operations Center (NOC.) For example, on the back end of network events, like an outage or
planned maintenance, ef fective Machine Learning and Artificial Intelligence tools can provide NOCs
with consistent measurements and automated responses that returns the network to steady-state
convergence.

Adaptive network solutions

5G networks are constantly evolving, and NOCs need adaptive problem-solving approaches that include
fast and accurate root cause identification and remediation. But when a problem occurs, it’s impossible to
sift through network events, alarms, and obscure behaviors without advanced tools. Machine Learning can
automatically identify and sort relevant performance data, classify it appropriately, and pinpoint systemic
issues and causes quickly. Using “fuzzy logic,” AI returns possible solutions and pre-implementation
predictive modeling and testing for all potential matches. Operations staf f can use those testing outputs
to choose the best solution for the specific problem at hand, and then AI tools can re-provision the
network automatically.

Fewer executive escalations

Solving network problems consistently and faster, with well-documented and automated solutions, will
inevitably reduce the number of escalations the NOC must negotiate between customers and executives.
With adaptive network solutions powered by AI/ML, NOC staf f and administrators will spend less time
firefighting and explaining themselves, and more time creating greater ef ficiencies, planning, and
optimizing the network.

© 2023 Fujitsu. Proprietary and Confidential. All Rights Reserved. Page 3 of 6

Whitepaper Ignite 5G Network Analytics with Machine Learning & Artificial Intelligence

AI/ML technologies every operator should know

Insights and remediation related to slowdowns, degradations,

and failures in network elements inevitably cascade into Here are a few tools and techniques to know:
major usability problems at customer endpoints, setting
of f alarm storms. As a result, modern applications running • ML uses anomaly detection, a type of outlier analysis,
on those endpoints are highly dependent on extreme real- to identify data points, events, or observations that
time visibility into usage, the network and its connections, vary significantly from a global data set. Operations
supporting compute elements, and storage. staf f uses anomaly detection to recognize a critical
incident or to surface opportunities for new service
development.
• Auto-encoder is a deep-learning neural network
Predict
system that compresses data and then reconstructs
Fuzzy that same data with minimal errors.
logic Layered
Models
• Correlation is the basis of statistical modeling that
shows the degrees to which two variables change in
coordination with one another.

Auto- Artificial • Classification references predefined categories the

system uses to sort data into specific categories.
action
Intelligence • Clustering is a collection of data points gathered
Deep LSTM
together because of certain similarities.
Learning • Long Short-Term Memory (LSTM) is a recurrent neural
network model with memory blocks that provide
Machine context for the system’s information. ML algorithms
Supervised use that context to understand data and help inform
learning Learning the next steps.
• Principal Component Analysis (PCA) is an ML technique
Auto-
Encoder
that reduces data dimensionality and increases
interpretability while minimizing information loss.
Unsupervised
learning • Time-series prediction uses any data source, including
Reinforcement streaming data, to forecast future behavior.
learning
• Sequence mining is a technique that examines
sequences of data groups and identifies statistically
relevant patterns.
Figure 2: AI/ML techniques

These AI/ML techniques completely upend loosely

AI/ML techniques track these network measurements and
documented “if-then-else-do-this” fixes. Using
events and analyze changes, trends, seasonality, cycles, and
multidimensional observations from the network itself,
fluctuations. While ML isolates issues, AI clusters, classifies,
transformative analytics and network automation deliver
and correlates the data, draws conclusions, and triggers
carefully orchestrated sequences of remediation that provide
network automation to deliver a closed-loop system that
adaptive network management solutions.
fixes root causes before severely impacting the end-user
experience. As networks evolve, behaviors fluctuate, and
interdependencies between data and behaviors fluctuate
correspondingly. Only AI/ML tools that leverage these
techniques can close the gaps that would otherwise take a
person hours or days to sort through.

© 2023 Fujitsu. Proprietary and Confidential. All Rights Reserved. Page 4 of 6

Whitepaper Ignite 5G Network Analytics with Machine Learning & Artificial Intelligence

Putting AI/ML techniques to work in the

communications network
Today, providers and their NOCs are most likely monitoring Time-series prediction and closed-loop actions
network behaviors with manual thresholds. When an event
crosses a threshold, the system sends a notification, and the Closed-loop actions based on time-series predictions can
person who receives that notification may or may not take provide better management solutions than traditional,
corrective action – again, manually. But it doesn’t have to reactive fixes. An example is span-loss prediction. Using time-
work that way. series prediction, the system can project when span loss will
hit a pre-configured threshold in the future. Then, the system
can be pre-programmed to reprovision the circuit before
performance is impacted.

Customer Post-event health checks

Experience Latency Post-maintenance event health checks also benefit from
Management Fuzzy
logic
Predict

Monitoring AI/ML. Before maintenance events, the logical layers of a

Layered
Models

Auto- Artificial healthy network are converged optimally in a steady state.

Using multidimensional modeling, AI/ML tools develop a
action
Intelligence

Automated Deep
Learning
LSTM Perform RCA
Convergence Supervised
Machine
Learning on alarms snapshot of the network. That includes log data, PM data, and
alarm patterns. After the maintenance event is performed,
learning

Auto-
Encoder

staf f can compare the new network state to the steady-state

Unsupervised
learning

Traffic Port/Circuit
Reinforcement
learning

model. If the network does not re-converge, the snapshot

Prediction Utilization uses anomaly detection and correlation techniques to help
identify discrepancies and missed configurations that can be
Provisioning & Steady-state addressed immediately. Multidimensional modeling might
Re-provisioning Snapshots also be used before and after snapshots, surfacing missed
configurations before a maintenance event is performed.
Figure 3: AI/ML in the provider network Monitoring rising alarm storms
Significant alarms for catastrophic events can go unnoticed
in a sea of meaningless data. Or worse, network conditions
create an alarm storm with no obvious indications of where
Streaming data analysis and model training to start solving the problem. Sorting through all the relevant
Using techniques like LSTM and the mathematical models data to find the root cause might take days and maybe a
that make up time-series prediction, AI/ML analyzes all the bit of luck. AI/ML surfaces those events identifies their root
information a network, its equipment, and its endpoints cause more quickly, and provides consistent and accurate
create. This can include of fline data and/or streaming data resolutions within hours.
from a network that is in service.
Recognizing disruptive network behaviors before
The system uses the information to forecast behaviors and they cause major problems
enforce policies for handling service disruption.
AI/ML tools also use the information to build and maintain An added benefit to solving alarm storms using AI/ML tools
multidimensional neural network models. Maintaining is that, post-event, providers know what caused the event,
network models includes model training, which is critical to understand what behaviors contributed to that event, and
minimizing loss of function within a prediction range that can tag those behaviors for future reference and pre-emptive
could impact the accuracy of the forecast. remediation. A disk overrun, equipment failure, or signal loss
can be tagged with potential resolution suggestions for
For example, circuit and port utilization predictions and future events.
forecasts can be made using streaming data analysis and
online/of fline model training. Latency-sensitive applications Managing the budget for growth rather than
that traverse network paths can combine predictions and remediating problems
forecasts with latency monitoring. This combination can
AI/ML tools help create network planning recommendations
proactively identify path fluctuations in longer routes that
that generate more ef ficiencies and budgeting insights.
might impact the application’s service. Once identified,
Accurate network inventory, progression, and growth
operations staf f might choose to provision a new circuit, re-
forecasts can be used maximize shelf utilization before new
provision an existing circuit with fewer fluctuations, and even
investments are made, and help ensure assets are added in
assess applications that appear unnecessarily sensitive.
the right places.

© 2023 Fujitsu. Proprietary and Confidential. All Rights Reserved. Page 5 of 6

Whitepaper Ignite 5G Network Analytics with Machine Learning & Artificial Intelligence

AI/ML with Virtuora® AX

Virtuora AX delivers explainable AI for sophisticated data sets,

with tools that operations, planning, engineering, and services
teams can leverage to improve service and network quality
of experience. Combine any or all of the Virtuora AX Network
Intelligence toolset and Virtuora AX MicroApplications to
deliver specific automated network control and management
across domains, layers, and systems.
Virtuora AX is a collection of cloud-native microapplications
that uses actionable and automated network intelligence to
extend traditional network monitoring. It includes:
• Application-building infrastructure capabilities for DevOps
• Network Intelligence tools
• Neural network models
• Learning engines
• A library of Virtuora AX microapplications for
network operations

The Virtuora AX leverages cloud-native architecture that

scales to fit data consumption and operation needs, and
runs in public or private clouds. Virtuora AX is a member of
Fujitsu’s Virtuora cloud family of products. The Virtuora cloud
family unifies orchestration, control, and management of the
multilayer, multivendor network stack, unwinding complexity
and enabling powerful functionality.
Virtuora cloud products are part of the Fujitsu hybrid CT/IT
infrastructure, expanding the reach, productivity, and value
of the communications network within the enterprise. With
Virtuora cloud, anyone can build a network around of ferings
and services powered by software.

us.fujitsu.com/networkautomation

Contact
Fujitsu Network Communications, Inc. ©Copyright 2023 Fujitsu Network Communications, Inc. FUJITSU (and design)®, 1FINITY™, and Virtuora® are
2801 Telecom Parkway, trademarks of Fujitsu Limited in the United States and other countries. FLASHWAVE® and NETSMART® are
Richardson, TX 75082 trademarks of Fujitsu Network Communications, Inc. (USA). All Rights Reserved. All other trademarks are the
Tel: 888.362.7763 property of their respective owners.
Configuration requirements for certain uses are described in the product documentation. Features and
us.fujitsu.com/telecom specifications subject to change without notice.

© 2023 Fujitsu. Proprietary and Confidential. All Rights Reserved. 1.0/08.23 Page 6 of 6