CH A RTI N G TH E FUTU R E O F I N N OVATI O N | VO LU M E 97 I 2018–02

ERICSSON
TECHNOLOGY

TECHNOLOGY TRENDS
AUGMENTING THE
CONNECTED SOCIETY
DIGITAL
CONNECTIVITY
MARKETPLACES

FIXED WIRELESS
ACCESS IN LTE AND 5G
 CONTENTS ✱

08 OPEN, INTELLIGENT AND MODEL-DRIVEN: EVOLVING OSS

08 GBR QOS
flows

Ericsson is leveraging the open-source approach to build open and intelligent

operations support systems (OSS) that are designed to support autonomic
networks and agile services. Our concept benefits from our ability to combine the
open-source approach with our unique network and domain knowledge. QFI and

20
DSCP tagged
IP packets
Non-GBR
20 LOW LATENCY, HIGH FLEXIBILITY: VIRTUAL AQM QOS flows
Bufferbloat – the excessive buffering of packets – can cause significant delays in
mobile networks for the application consuming the data, which is why it is so
important for the network to be able to detect and properly manage it. Virtual Active
Queue Management is an innovative approach to buffer management that is
deployed as a centralized network function.

Dequeuing Flow Flow

30
scheduler queueing hashing
30 FEATURE ARTICLE
Five technology trends augmenting
the Connected Society
Consumer Consumer Consu
Ericsson’s CTO Erik Ekudden presents the five technology trends that are
most relevant to the future development of network platforms capable of
supporting the continuous evolution of industries and societies around the world.
Digital connectivity marke
The use of machines to augment human intelligence will play a key role. Application & Information Bu
enablement market se
market

42 DIGITAL CONNECTIVITY MARKETPLACES TO ENRICH

5G AND IoT VALUE PROPOSITIONS Inferred
The ongoing digitalization of industry represents a key growth opportunity for the knowledge
telecom sector. The main challenge is to fully understand the needs of a new market.
42 Induced
models
Cloud services market
Our solution is to establish a platform model where capabilities from many providers Ope
can be effectively packaged and exposed in attractive ways to buyers from different se
industries. Producer Connectivity services market
Knowledge
transfer
52 COGNITIVE TECHNOLOGIES IN NETWORK
Machine Machine
52
AND BUSINESS AUTOMATION Reasoning Digital connectivity platfP
Planning
While the terms artificial intelligence and machine learning are often used reasoning learning F
Actionscontrol of
synonymously, achieving automation that goes beyond low-level A
network parameters also requires machine reasoning. The best way to create (Symbolic) (Numeric)
the intelligent agents that are necessary to automate operational and
business processes is to capitalize on the strengths of both technologies. Knowledge
extraction
64 LEVERAGING LTE AND 5G NR NETWORKS FOR
FIXED WIRELESS ACCESS
The high-speed mobile broadband coverage that LTE and 5G New Radio (NR) Expert Training
enables has created a commercially attractive opportunity for operators to use knowledge examples
64 fixed wireless access (FWA) to deliver broadband services to previously unserved
homes and small and medium-sized enterprises around the world. 64
 ✱ EDITORIAL EDITORIAL ✱

Ericsson Technology Review brings you

insights into some of the key emerging
innovations that are shaping the future of ict.
Our aim is to encourage an open discussion
about the potential, practicalities, and benefits
of a wide range of technical developments,
and provide insight into what the future
TRANSFORMING THE AUGMENTATION OF HUMAN
INTELLIGENCE IS ONE OF THE KEY THEMES
TO FIT THE
has to offer.

address
IN THIS YEAR’S TRENDS ARTICLE
Ericsson

FUTURE REALITY
se-164 83 Stockholm, Sweden
Phone: +46 8 719 00 00

publishing
All material and articles are published on the
Ericsson Technology Review website: one billion households around the world that are still adapting to link rate fluctuation and considering
www.ericsson.com/ericsson-technology-review unconnected, as well as to many small and medium- the current bottleneck link bandwidth to the user.
■ technology development keeps getting
publisher faster and more interconnected, with new innovations sized businesses. The FWA article in this issue Our testing has shown that when vAQM is
Erik Ekudden appearing every day. As a result, we’re swiftly moving highlights the key principles for combined mobile centralized upstream, rather than being deployed
toward the realization of the “Augmented Connected broadband and FWA deployments, and presents in the bottleneck nodes as it is in classic AQM,
editor
Society” – a world characterized by ubiquitous a use case that illustrates the recommended there is a substantial reduction in bufferbloat.
Tanis Bestland (Nordic Morning)
internet access for all, self-learning robots and truly deployment approach.
[email protected]
intuitive interaction between humans and machines. I hope you get a lot of value out of this issue of
editorial board But how can our industry best prepare for this future? In my discussions with people from different Ericsson Technology Review, and that the articles
Håkan Andersson, Anders Rosengren, In my role as CTO, I have the challenging and industries this past year, I’ve noticed a growing can serve as the basis for stimulating future-focused
Mats Norin, Erik Westerberg, awareness that telecom infrastructure can bring discussions with your colleagues and business
Magnus Buhrgard, Gunnar Thrysin, exhilarating annual task of identifying the five
Håkan Olofsson, Dan Fahrman, Robert Skog, technology trends of the future that are (or will be) significant value to the digital transformation of partners. If you would like to share a link to the whole
Patrik Roseen, Jonas Högberg, most relevant to our industry. You can find my insights industries. However, many providers need help magazine or to a specific article, you can find both
John Fornehed and Sara Kullman and reflections on this year’s trends on page 30. to understand the requirements of this new PDF and HTML versions at https://www.ericsson.
market and to create solutions for it. The Digital com/en/ericsson-technology-review
f e at u r e a r t i c l e
The augmentation of human intelligence is one Connectivity Marketplaces article in this issue
Five technology trends augmenting the
Connected Society by Erik Ekudden of the key themes in this year’s trends article. explains our concept for a platform model in
Creating the highly automated environment that which capabilities from many providers can be
art director
network operators and digital service providers effectively packaged and exposed in attractive
Liselotte Eriksson (Nordic Morning)
will need in the future requires the support of ways to buyers from different industries.
production leader intelligent agents that are able to work
Susanna O’Grady (Nordic Morning) collaboratively. The two proofs of concept There’s no doubt that operations support systems
presented in the Cognitive Technologies article (OSS) have a critical role to play in the future of our
l ay o u t
in this issue demonstrate how the combination industry. Check out the OSS article in this issue to
Liselotte Eriksson (Nordic Morning)
of machine reasoning and machine learning learn about our future OSS concept, which is built on
i l lu s t r at i o n s techniques makes it possible to create intelligent a solid implementation architecture that enables the
Nordic Morning Ukraine agents that are able to learn from diverse inputs, use of industrydefined interfaces and open-source
and share or transfer experience between contexts. modules, as well as integration with full component
chief subeditor
compatibility.
Ian Nicholson (Nordic Morning)
I believe that fixed wireless access (FWA) is likely to play
subeditors an important role on our journey toward a world Finally, I want to encourage those of you who are
Paul Eade and Penny Schröder-Smith characterized by ubiquitous internet access for all. concerned about bufferbloat to check out the
(Nordic Morning) Virtual Active Queue Management (vAQM) article
Already today, LTE and 5G New Radio are opening up ERIK EKUDDEN
issn: 0014-0171 significant commercial opportunities for operators to in this issue. In our innovative concept, vAQM SENIOR VICE PRESIDENT
Volume: 97, 2018 use FWA to bring the internet to many of the more than is centralized and applied per user and per flow, AND GROUP CTO

ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 7
✱ EVOLVING OPERATIONS SUPPORT SYSTEMS EVOLVING OPERATIONS SUPPORT SYSTEMS ✱

OPEN, INTELLIGENT AND

model-driven: tool chains that support automation. Microservice

architecture supports the DevOps paradigm and
enables highly scalable, extendable and flexible
Our OSS architecture principles
The key principles guiding Ericsson’s OSS
architecture concept are that it is service oriented,

evolving OSS
systems. that the automation is analytics and policy driven,
Ericsson’s approach to evolving OSS is based on and that it uses virtualization and abstraction of
combining these technologies and solutions to create network functions. By following these principles,
a modern OSS architecture that is optimized to meet we can evolve OSS to become programmable and
the requirements of service providers in the mid to able to leverage the interfaces offered by the
long term. production domains.

Definitions of key concepts

〉〉 Analytics is the discovery, interpretation and communication of meaningful patterns in data. It is the
The simplicity required to deploy and manage services in future network and umbrella for AI, ML and MI.
infrastructure operations is driving the need to rethink traditional operations 〉〉 Cloud native is a term that describes the patterns of organizations, architectures and technologies that
consistently, reliably and at scale take full advantage of the possibilities of the cloud to support cloud-
support systems (OSS). To unleash the potential of 5G, Network Functions oriented business models.
〉〉 Microservice is a small service supporting communication over network interfaces. Several such services
Virtualization (NFV) and software-defined networking (SDN), future OSS constitute an application. Each microservice covers a limited and coherent functional scope and is
independently manageable by fully automated life-cycle machinery. A microservice is small enough to be
need to be open, intelligent and able to support model-driven automation. the responsibility of a single, small development team.
〉〉 OSS business layer is a conceptual layer of the OSS that guides the service provider through the
operational support subset of its business process.
they enable greater efficiency in development and 〉〉 OSS control layer is a layer of the OSS containing functionality that controls and manages production
MALGORZATA
The long-standing paradigm of bundling domains.
vendor and domain-specific management operations. 〉〉 Policy is a function that governs the choices in the behavior of a system. Policy can use a declarative or
SVENSSON,
with network functions has been challenged For example, model-driven automation eliminates imperative approach.
MUNISH AGARWAL ,
in recent years by a new approach that is built the need for manual interaction by externalizing 〉〉 Programmability is the ability to externally influence the behavior of a system in a defined manner.
STEPHEN TERRILL ,
behavioral aspects of specific domains from 〉〉 Resource is a means to realize a service, and can be either physical or non-physical.
JOHAN WALLIN on the concept of an open and model-driven 〉〉 Service is a means to deliver value to a customer or user; it should be understandable to a customer.
platform. This new approach leverages open application logic. Real-time and near-real-time
Service is also a representation of the implementation aspects to deliver the value.
source and uses standardized interfaces to analytics redefine the implementation and scope of
enable a horizontal management platform. support system functions like performance, fault
management, assurance and optimization, where
analytics models act on the collected and correlated Terms and abbreviations
■ Today’s service providers expect rapid network data, producing insights that are far richer in scope AI – artificial intelligence | API – application programming interface | BSS – business support systems |
deployments, agile introduction of new services than the outcome of the traditional functions. ETSI – European Telecommunications Standards Institute | MANO – Management and Orchestration |
and cost-efficient operation and management – Machine intelligence (MI) algorithms enable the MI – machine intelligence | ML – machine learning | NFV – Network Functions Virtualization | ONAP – Open
requirements that are even more important when transition from reactive to proactive decision making. Network Automation Platform | OSS – operations support systems | PNF – Physical Network Function |
planning for future 5G capabilities. Recent The DevOps paradigm simplifies development SDN – software-defined networking | SLA – Service Level Agreement | TOSCA – Topology and Orchestration
technology trends make it possible to redefine and operational processes. It requires development Specification for Cloud Applications | UDM – Unified Data Management | VNF – Virtual Network Function
traditional support systems at the same time as and deployment-friendly software architecture and

8 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 9
✱ EVOLVING OPERATIONS SUPPORT SYSTEMS EVOLVING OPERATIONS SUPPORT SYSTEMS ✱

Regardless of the generalization of the behavior MI and autonomic networks As the level of MI increases, the role of policy shifts
AUTOMATION CAN BE between various production domains, the domain- Automation can be further enhanced by applying from governance of the decisions to be made, to
FURTHER ENHANCED BY specific aspects remain. Some examples of MI, which is the combination of machine learning ensuring that the insights produced from MI are
domain-specific extensions are: managing and (ML) and artificial intelligence (AI). MI adapts to appropriate. In this sense, MI is an enabler for
APPLYING MI, WHICH IS THE optimizing radio frequency, designing and handling the situation in a network and learns to provide the autonomic networks.
COMBINATION OF MACHINE data center capacity, or designing L2/L3 service best insights for a given network situation.
overlay. Specific production domain competencies
LEARNING AND ARTIFICIAL and processes are required to provide the necessary
INTELLIGENCE capabilities in these examples, even though the
capabilities are realized by the same OSS.
OSS business layer
Abstraction, programmability From automatic management to
and domain-specific extensions management of automation
Networks are complex. As the physical resources The goal of automation is to provide a zero-touch Design Onboard Test Deploy Operate Retire
within them transform into virtual resources such network, which means that automation moves
as Virtual Network Functions (VNFs), virtual from automating via scripting what can be done
switches, virtual networks and virtual cloud manually, to an autonomic network that takes care OSS control layer
resources, they will expose only what is necessary of itself and handles previously unforeseen
and be managed in a uniform way. This leads to a situations. This changes the approach to Control
logical representation of resources and networks, management, from managing networks manually Analytics Insights Policy Decisions orchestration
which is referred to as abstraction. The to managing automation. management
capabilities of these virtual resources are exposed A key part of managing automation is the control
through their interfaces, while their behavior is loop paradigm, as shown in Figure 1. This is achieved

Actions
influenced by these interfaces. This is referred to by designing the insights and policies required for a

Data
as programmability. The methodology of use case, and the decisions that need to be made.
abstraction and programmability is continually Insights are the outcome when data is processed by
and recursively applied at all levels of the OSS and analytics. Policies represent the rules governing the
networks, from the resources’ interfaces to the decisions to be made by the control loop. The decisions Production domains
exposed services. are actuated by COM (Control, Orchestration,
Abstraction and programmability are the main Management), which interacts with production
prerequisites for uniform management of domains by requesting actions such as update,
Cloud system
automation, easing the shift from managing configure or heal. The control loops are implemented PNF PNF VNF VNF VNF VNF VNF infrastructure Transport

networks to managing automation. They enable by multiple OSS functions and can act in a
simplification by exposing only the required hierarchical way.
capabilities of a domain, while still supporting an By leveraging analytics, MI and policy, control
efficient interaction between management system loops become adaptive and are central to assuring
scopes, and management systems and networks. and optimizing the deployed services and resources. Figure 1 Analytics and policy-driven automation: closed control loop

10 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 11
✱ EVOLVING OPERATIONS SUPPORT SYSTEMS EVOLVING OPERATIONS SUPPORT SYSTEMS ✱

OSS business layer

BSS
Knowledge
Interaction management management
Design Onboard Test Deploy Operate Retire

Capacity management

Services and API exposure

OSS control layer Security
management
Service and User service and
Assurance and

Studio suite
Network slice RAN resource life-cycle resource life-cycle
Studio suite life-cycle VNF scaling SLA assurance ... optimization
optimization management management

Development

Marketplace
management
Procurement

OSS information layer OSS catalogs and inventories

Network slice VNF scaling policy RAN optimization

deployment model model policy model Forecasting
Mediation and analytics

Production
Catalog Analytics Policy Orche- Inventory Mediation Packet
Communi- domains
stration Cloud system cation
RAN core infrastructure Transport services UDM

Figure 2 Model-driven OSS architecture Figure 3 Ericsson’s OSS architecture

Model-driven architecture (specifically, orchestration in OpenStack [1]) – and Modular architecture and managed production domains. The studio suite
To secure efficient and consistent management, TOSCA [2]. The “operate” process can cover implementation capabilities framework includes business, engineering and
the behavioral aspects of specific domains are activities such as scale, upgrade, heal, relocate, stop The optimal OSS of the future will be based both on operational studios.
externalized from application logic to models. and start. the principles outlined above and on modular The business studio is the environment where
There are models for configuration, orchestration, The configuration and instrumentation of the architecture and implementation capabilities. business and operational processes are maintained.
policy rules and analytics. executing resource and service instances require The business studio automatically orchestrates the
Resource and service life-cycle management is a resource and service views expressed also as Studio suite process models by invoking operations on the
good example of a model-driven approach. models. IETF YANG [3] has emerged as the Figure 3 illustrates the studio suite, which provides appropriate functions in the OSS stack.
As shown in Figure 2, the models describe the mainstream modeling language for this purpose. a single, coherent environment to design, test, The engineering studio provides a model-driven
services and the resources. These models are Policy rule sets define the behavioral aspects of operate, control and coordinate all the business, platform where the OSS functions and the relevant
designed and onboarded in a studio suite. They are service assurance, resource performance and engineering and operational processes of a service information and information models associated
then used to test, deploy, operate and retire the scaling. Since the models are the way to capture provider. This is done by expressing the processes, with those functions are constructed and extended.
services and the resources. The models are based business logic and intellectual property, they will interface extensions, policies, deployments and It lets a user (human or machine) realize various
on standard modeling approaches. For example, become sellable objects themselves and be subject configurations as models and using them for management tasks by extending the OSS functions
deployment models are described using HEAT to commercial agreements. automation of all the OSS functions and the with use-case specific models for configuration,

12 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 13
✱ EVOLVING OPERATIONS SUPPORT SYSTEMS EVOLVING OPERATIONS SUPPORT SYSTEMS ✱

orchestration, data and policies. It provides software by surrounding functions like interaction operations on the networks and infrastructures of the environments, capable of scaling to tens of thousands
development kits for defining the extension models management, assurance and optimization. They production domains. of self-healing, multi-tenant nodes. Cloud native
and then instantiates the needed components, provide both closed loop optimization capabilities The assurance function makes it possible to check systems are container packaged, dynamically
provisions the components with the models and (where decision making is done based on policies and that resources and services behave as they have been managed and microservice oriented. The Ericsson
wires them together to realize the desired use case. MI) and an open loop optimization feature (where a designed, published and offered. Assurance makes OSS of the future will be based on cloud-native
An example could be a set of ML functions that human user decides based on a set of automatic use of analytics to discover insights about the architecture to enable better reuse, simplified
provide scalable machinery that can be assigned to recommendations). This dynamic, model-driven and performance of services and resources. Policy operations and improved efficiency, agility and
multiple uses by applying different configuration and real-time interaction is what will make this future engines are then triggered to act on the insights to maintainability.
data models. OSS solution different from the traditional one. determine actions that ensure the desired The microservice architecture enables fine-grained
The operational studio offers capabilities to We can already see the need for seamless performance level offered on services and resources. reuse by having several small components instead of a
manage resources and services throughout their life interaction between OSS software development and Our OSS architecture is designed to use single large one. By being loosely coupled and
cycle. Resources represent the infrastructure and the operations, which suggests that the core DevOps interaction management, services and application backward compatible, the services can be developed
networks of production domains. Services use the capabilities of continuous integration, delivery, programming interface (API) exposure to interact independently, enabling efficient DevOps. It also
infrastructure and networks to deliver value to release and deployment on-demand will be an with external functions. A few examples of those allows each service to choose the most efficient
customers. One of the first activities in the resource integral part of the future OSS. external functions are development, marketplaces, development and runtime technology for its purpose.
and service life cycle is to construct a specification of OSS inventories will be used to store instantiated business support systems, advanced security and The microservice architecture makes it possible to
a resource or a service, then assign the required resources and services. They track the presence, knowledge management systems, procurement and compose various business solutions, then deploy them
policy rules and configurations defining the behavior capacity and configuration of resources and services, forecasting. automatically, significantly reducing deployment cost
of the network, or the infrastructure or the service. as well as their relationships to other resources and The production domain example shown in Figure and time to market. The architecture will use the
Once they are defined, the specifications, policies services. They provide the data for capacity 3 includes a RAN, packet core and communication cloud-native ecosystem as a portability layer to
and configurations are stored in the OSS catalogs management, which enables capacity planning, services, Unified Data Management, cloud system support multiple deployment options.
and distributed in real time to be used by the OSS design and monitoring over time. By using analytics infrastructure, fixed access and transport networks.
functions. in capacity management, it is possible to make it work Our OSS manages multi-vendor network and Security
in real time and be predictive and self-learning. infrastructure. Supporting multiple deployment models requires
Managing services and resources The purpose of the user service and resource life- The OSS functions are based on an API-first multiple security capabilities like credential
Service and resource life-cycle management cycle management function is to manage resources and approach that mandates that an API be defined management, secure communication, encryption,
functions provide a programmable way of managing services that have been assigned to a user, tenant or before the function-exposing services can be authentication and authorization, which can be
the services and resources of the production subscriber. This capability also supports user, tenant implemented. The approach enables parallel selected based on the deployment. Our OSS
domains based on specifications, policies and and subscriber provisioning. The closed and open development and provides the ability to adapt to architecture will include the capabilities needed to
configurations stored in the catalogs. The core control loop architecture patterns apply here as well. external implementations. The API management support the multiple deployment models and to
capabilities are orchestration and policy engines The user service and resource life-cycle management framework facilitates the effective use of APIs during protect the models.
that parse the defined models to realize all the life- function has the same properties as the service and development, discovery of internal API endpoints at
cycle steps of the services and resources. The policy resource life-cycle management function. runtime and efficiently controlled exposure of APIs Embracing industry initiatives
and orchestration engines automate the network Mediation provides an open and adaptable to external applications. Due to NFV and programmable networks such as
management by taking over the burden of low-level interface to collect data, process events generated by SDN, there is huge industry interest in automation to
decision making and execution of life-cycle the production domains, and do the required Cloud native decrease complexity and achieve rapid introduction
operations from human users. translation, transformation and filtering. It also Cloud native is the new computing paradigm that is of services and resources. This has resulted in several
These functions are triggered by events generated provides a set of flexible interfaces to invoke life-cycle optimized for modern, distributed systems significant industry initiatives, most notably

14 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 15
✱ EVOLVING OPERATIONS SUPPORT SYSTEMS EVOLVING OPERATIONS SUPPORT SYSTEMS ✱

ETSI-MANO [4] and the Open Network appropriate models on a compatible OSS platform.
Automation Platform (ONAP) [5]. Since a model-driven architecture is essential to References
Initiatives that call for standardization provide an drive automation, we are working toward ensuring 1. OpenStack, Heat Orchestration Template (HOT) Guide, available at: https://docs.openstack.org/heat/ocata/
environment to develop strong concepts and identify that our models are usable within ONAP. template_guide/hot_guide.html
important interfaces, while open source provides
Conclusion 2. OASIS, OASIS Topology and Orchestration Specification for Cloud Applications (TOSCA) TC, available at:
reference implementations, as well as
https://www.oasis-open.org/committees/tc_home.php?wg_abbrev=tosca
implementation interfaces. As both an established infrastructure and OSS
supplier, Ericsson has a vision for OSS in which 3. Internet Engineering Task Force (IETF), October 2010, YANG – A Data Modeling Language for the Network
ONAP autonomic networks enable intelligent and Configuration Protocol (NETCONF), available at: https://tools.ietf.org/html/rfc6020
By bringing together the resources of its members, automated service and resource life-cycle 4. ETSI, Network Functions Virtualization, available at: http://www.etsi.org/technologies-clusters/technologies/nfv
ONAP has sped up the development of a globally management. We are using analytics, MI, policy 5. ONAP, available at: https://www.onap.org/
shared architecture and implementation for and orchestration technologies to create the OSS of
network automation. ONAP provides both design the future, and to offer greater efficiency in
time and runtime reference implementations, development and operations. We apply a DevOps
covering service design and creation, catalog, paradigm to our software development to keep us Further reading
inventory, orchestration and control, policies and close to customers and speed up our software
〉〉 Ericsson Technology Review, Technology trends driving innovation – five to watch, 2017, Ekudden, E:
analytics. It also includes service management and delivery.
https://www.ericsson.com/en/publications/ericsson-technology-review/archive/2017/technology-trends-2017
automation capabilities for Physical/Virtual As our approach is model-driven, OSS behavior is
〉〉 Ericsson Technology Review, Generating actionable insights from customer experience awareness,
Network Functions (PNFs/VNFs), transport and externalized from the management platform and
September 2016, Niemöller, J; Sarmonikas, G; Washington, N: https://www.ericsson.com/en/publications/
cloud infrastructure. materialized through a combination of models and
ericsson-technology-review/archive/2016/generating-actionable-insights-from-customer-experience-awareness
Ericsson’s modular and model-driven OSS configurations. As a result, the developed models are
〉〉 Ericsson Technology Review, DevOps: fueling the evolution toward 5G networks, April 2017, Degirmenci,
approach is well aligned with ONAP and extends the independent of the execution platform.
F; Dinsing, T; John, W; Mecklin, T; Meirosu, C; Opsenica, M : https://www.ericsson.com/en/publications/ericsson-
ONAP vision to automation beyond network Our OSS concept is modular by design and follows
technology-review/archive/2017/devops-fueling-the-evolution-toward-5g-networks
management to include interaction management, microservice architecture principles that make it
〉〉 Ericsson, Gearing up support systems for software-defined and virtualized networks, June 2015: https://www.
workforce management, user service and resource possible to replace software components with open
ericsson.com/en/news/2015/6/gearing-up-support-systems-for-software-defined-and-virtualized-networks-
life-cycle management, advanced service source implementations. The concept is built on a
〉〉 Ericsson, Zero touch networks with cloud-optimized network applications, Darula, M; Más, I: https://www.
optimization and assurance as a few examples. solid implementation architecture that enables the
ericsson.com/assets/local/narratives/networks/documents/zero-touch-networks-with-the-5g-cloud-optimized-
To interact with ONAP, we are in the process of use of industry-defined interfaces and open source
network-applications.pdf
adopting the necessary interfaces and model modules, as well as integration with full component
〉〉 Ericsson Technology Review, Architecture evolution for automation and network programmability, November
definitions in our product portfolio (both in OSS and compatibility.
2014, Angelin, L; Basilier, H; Cagenius, T; Más, I; Rune, G; Varga, B; Westerberg, E: https://www.ericsson.com/
network functions). Ericsson is committed to ONAP Ericsson has embraced open source
en/publications/ericsson-technology-review/archive/2014/architecture-evolution-for-automation-and-network-
and heavily engaged with it across a broad range of implementations through active contributions and
programmability
topics. We aim to use the ONAP-provided open usage, as well as driving important
source implementations in accordance with standardizations. We are committed to driving
Ericsson’s architecture principles. industry alignment that will help create a healthy
Managing multiple production domains requires a ecosystem.
model-driven architecture because achieving
domain-specific behavior requires the application of

16 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 17
✱ EVOLVING OPERATIONS SUPPORT SYSTEMS

Malgorzata for OSS/BSS. He has 19 engaged in open source on

the authors
Svensson years of experience in ONAP’s Technical Steering
◆ is an expert in OSS. She telecommunications working Committee. Terrill holds an
joined Ericsson in 1996 with product development, M.Sc., a B.E. (Hons.) and a
and has worked in various common platforms and B.Sc. from the University of
areas within research and architecture evolution. He Melbourne, Australia
has hands-on experience
with mediation, rating, Johan Wallin
order management and ◆ is an expert in network
interactive voice response. management. He joined
Agarwal holds a Bachelor Ericsson in 1989 and has
of Technology degree worked in several areas in
from the Indian Institute of R&D in various positions
Technology Kharagpur. covering design, product
management and system
Stephen Terrill
development. For the past 10 ◆ has more than 20 years
years, her work has focused of experience working
on architecture evolution. with telecommunications
She has broad experience in architecture, implementation
business process, function and industry engagement.
and information modelling; His work has included both
information and cloud architecture definition and
technologies; analytics;
DevOps processes; and tool
chains. She holds an M.Sc. in management. He has broad
technology from the Silesian experience of systems
University of Technology in architecture in various
Gliwice, Poland. mobile telephony systems
including GSM, TDMA and
Munish Agarwal CDMA. For the past 10
The authors
◆ is a senior specialist in years, he has been working
would like to
implementation architecture with network management
acknowledge
posts within standardization architectures from an overall
the contribution
organizations such as Ericsson perspective. He
their colleagues
ETSI, 3GPP, ITU-T (ITU holds an M.Sc. in computer
John Quilty, Bo
Telecommunication engineering and computer
Åström, Ignacio
Más, Jan Friman, Standardization Sector) and science from the Institute
and Jaco Fourie IETF (Internet Engineering of Technology at Linköping
made to the Task Force). In recent years, University, Sweden.
writing of this his work has focused on the
article. automation and evolution
of OSS, and he has been

18 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018

✱ VIRTUAL ACTIVE QUEUE MANAGEMENT VIRTUAL ACTIVE QUEUE MANAGEMENT ✱

LOW LATENCY, HIGH FLEXIBILITY

Virtual
AQM find the ideal send rate for a given end-to-end link.
Mobile networks are provisioned with large buffers
to handle bursty traffic, user fairness and radio
channel variability. Even though these oversized
buffers prevent packet loss, the overall performance
degrades, as bufferbloat has major impacts on
Video streaming with adaptive bitrate through a
continuous sequence of small HTTP downloads
suffers when packets are lost. Under good
conditions, video is transferred over the network
at the same rate as it is played back. If the network
can’t keep up with video playback, the player
Virtual Active Queue Management is an innovative approach to buffer congestion control at endpoints. It misleads loss- usually switches to a lower video rate representation
management that does not require deployment in the network nodes where based congestion control algorithms, resulting and reasonable video quality can be achieved
in packet overshooting on the sender side. Over- with less bandwidth.
the buffering takes place. Instead, it is deployed as a centralized network buffering also results in large jitter of end-to-end The main challenges in video streaming are
latency, which potentially brings in timeout events packet loss and excessive buffering. When a packet
function, which allows for simpler configuration and increased flexibility and eventually leads to throughput degradation. is lost, the video player stops receiving data until the
compared to traditional Active Queue Management deployments. lost packet has been retransmitted, even though data
Application impact packets continue to arrive. Video players normally
High latency and jitter cause significant problems buffer video segments, allowing them to continue
To minimize the effects of bandwidth tends to work well in fixed networks, where bottle- for interactive applications such as VoIP, gaming playing from their buffer while downloading video
MARCUS IHLAR , neck link bandwidth varies little over short time and videoconferencing. Web browsing is also segments that might involve retransmissions.
ALA NAZARI, fluctuations in mobile networks, radio base
stations are typically provisioned with large durations. However, the deployment of traditional highly dependent on latency, and page load times Buffering is normally used for video on demand and
ROBERT SKOG
buffers. However, since excessive buffering of AQM is not widespread in mobile networks. grow linearly as round-trip time (RTT) increases. timeshifted video but is limited for live video.
packets causes a large perceived delay for the
The problem of bufferbloat
application consuming the data, it is important
for the network to detect and properly manage Bufferbloat is an undesirable phenomenon
cases in which buffering becomes excessive. prevalent in packet networks in which excessive Terms and abbreviations
buffering results in unnecessarily large end-to- ACK – acknowledgement | AMBR – aggregate maximum bitrate | AQM – Active Queue Management | CoDel
end latency, jitter and throughput degradation. It – Controlled Delay | CPE – Customer-premises equipment | DN – Data Network | DNS – Domain Name System
■ Active Queue Management (AQM) is a well- occurs due to large queues that absorb a huge | DSCP – Differentiated Services Code Point | DRR – Deficit Round Robin | EPC – Evolved Packet Core | EPS
established technique for managing the frequent amount of traffic in a congested path and usually – Evolved Packet Switched System | FQ – Flow Queue | FQ-CoDel – Flow Queue Controlled Delay | GBR –
and substantial fluctuations in bandwidth and causes a long queuing delay. To prevent packet guaranteed bitrate | IETF – Internet Engineering Task Force | PDN GW– public data network gateway | PDU –
delay that are common in networks. It has been losses, networks nodes rely on large buffers. protocol data unit | QFI – QoS Flow Identifier | QUIC – Quick UDP Internet Connections | RTT– round-trip time |
applied to network buffers for many years with the Bufferbloat occurs mostly at edge network nodes SDF – Service Data Flow | SMF – Session Management Function | SYN – synchronization | TCP – Transmission
purpose of detecting and informing data senders and defeats the built-in TCP congestion avoidance Control Protocol | UE – user equipment | UPF – user plane function | vAQM – virtual Active Queue Management
about incipient traffic congestion. Traditional AQM mechanism, which relies on dropped packets to

20 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 21
✱ VIRTUAL ACTIVE QUEUE MANAGEMENT VIRTUAL ACTIVE QUEUE MANAGEMENT ✱

However, if multiple TCP packets are lost, TCP variable of the minimum delay to see where With FQ-CoDel, it is possible to reduce OUR INNOVATIVE
may slow down its sending rate to well below the it is relative to the standing queue delay. bottleneck delays by several orders of magnitude.
data rate of the video, forcing the player into the Third, instead of measuring queue size in bytes It also provides accurate RTT estimates to CONCEPT FOR VIRTUAL
rebuffering state. or packets, it measures the packet sojourn time elephant flows, while allowing priority access to AQM IS APPLIED PER USER
In short, bufferbloat degrades application in the queue and uses it as a metric to detect shorter flow packets [3].
performance and negatively impacts the user incipient congestion. In combination with improvements in Linux AND PER FLOW
experience. There is therefore a pressing need CoDel works as follows: upon arrival, a packet network stacks, CoDel and FQ-CoDel can
to control latency and jitter in order to provide is enqueued if there is room in the queue. While eliminate the problem of bufferbloat on Ethernet, Virtual AQM consists of two separate but
desirable QoS to users. enqueuing, a timestamp is added to the packet. cable, digital subscriber lines and fiber, resulting in interrelated components: a bottleneck modeler
When dequeuing, the packet sojourn time is network latency reductions of two to three orders of and a queue manager. A separate vAQM instance
Mitigation calculated. CoDel is either in a dropping state or magnitude, as well as improvements in goodput [4]. is required for each bottleneck. In LTE, this
Traditional AQM mitigates bufferbloat by a non-dropping state. If the packet sojourn time In mobile networks, however, AQM needs to corresponds roughly to a bearer (dedicated or
controlling queue length through dropping or remains above the target value (default 5ms) for a adjust to link bandwidth fluctuation that varies for default), while in 5G it corresponds to a Protocol
marking packets from the bottleneck buffer when specified interval of time (default 100ms), CoDel different users. This is due to the fact that bottleneck Data Unit (PDU) session or a QoS flow. To effectively
it becomes full or when the queuing delay exceeds enters the dropping state and starts dropping link bandwidth can vary significantly in mobile model the bottleneck, the vAQM measures the RTT
a threshold value. AQM reacts to congestion by packets at the head of queue. While CoDel is in networks even over short time durations, which of packets between itself and the user equipment
dropping packets and thereby signaling to the dropping state, if the packet sojourn time falls below causes queue build-up that can add significant (UE), as well as the amount of inflight data.
sender that it should restrict the sending data rate the target or if the queue does not have sufficient delays when link capacity decreases. Virtual AQM maintains four variables: Path RTT,
due to the imminent overload. The short wait time packets to fill the outgoing link, CoDel leaves the Our innovative concept for virtual AQM (vAQM) Average RTT, Target Queue Delay and Target
also improves the response time for the transport dropping state. is applied per user and per flow, adapting to link rate RTT. Path RTT represents the propagation delay
protocol error handling. The CoDel parameters target and interval are fluctuation and considering the current bottleneck between the vAQM and the UE at the receiving end,
fixed parameters and their values have been chosen link bandwidth to the user. Bottleneck link without queuing delay. It is calculated by feeding
based on the observations from several experiments. bandwidth is determined by correlating the number RTT measurement samples to a min filter. Due to the
WITH FQ-CODEL, Interval is used to ensure that the measured of bytes in flight with RTT measurements. vAQM is varying nature of radio networks, the Path RTT must
IT IS POSSIBLE TO REDUCE minimum delay does not become too stale. It should similar to FQ-CoDel but it uses the measured time- be occasionally revalidated.
be set at about the worst-case RTT through the varying link bandwidth to the user as the dequeuing Average RTT is an exponentially weighted
BOTTLENECK DELAYS BY bottleneck to give endpoints sufficient time to react. rate instead of non-varying deficit/quantum as in moving average of the measured RTT, which
SEVERAL ORDERS OF The default value is 100ms. The target parameter is FQ-CoDel. represents the current observed delay. The use of
the acceptable queue delay for the packets, including a weighted moving average reduces the impact of
MAGNITUDE capacity to cope with traffic spikes. A target of 5ms Virtual AQM inherent noise in the RTT signal, which typically
works well in most situations; lower values can Traditional AQM is typically designed to operate arises from path layer characteristics such as Radio
Controlled Delay reduce the throughput. If the target is more than on queues directly. However, due to deployment Link Control layer retransmissions.
Controlled Delay (CoDel) is an emerging AQM 5ms, there is minor or no improvement in utilization. constraints and other factors, it is not always A certain bottleneck queue delay must be
scheme designed by the IETF (Internet straightforward to deploy new AQMs at the tolerated to ensure smooth delivery of data over the
Engineering Task Force) [1]. Its goal is to contain Flow Queue Controlled Delay network edges where congestion is likely to be radio. Target Queue Delay represents the delay that
queuing latency while maximizing the throughput. Flow Queue Controlled Delay (FQ-CoDel) [2] is a experienced. One way of getting around this is tolerated on top of Path RTT before vAQM takes
CoDel attempts to limit bufferbloat and minimize standard AQM algorithm that ensures fairness in problem is to move the bottleneck to a more some sort of action. Target Queue Delay is defined as
latency in saturated network links by distinguishing CoDel. It is useful because AQM algorithms central point in the network and apply the AQM a fraction of Path RTT. The sum of the Path RTT and
good queues (the ones that empty quickly) from working on single queues exacerbate the tendency there. In a fixed network, it is easy to achieve this the Target Queue Delay is the Target RTT.
bad queues that stay saturated and slow. CoDel of unfairness between the flows. FQ-CoDel is a by shaping traffic to a rate slightly below the These four variables are used to determine
features three major innovations that distinguish hybrid scheme that classifies flows into one of bottleneck peak capacity. Since mobile networks whether queuing is taking place in network
it from prior AQMs. First, it uses the local minimum multiple queues, applying CoDel on each queue exhibit vastly different properties of fluctuating bottlenecks. The information can be used in two
queue as a measure of the standing/persistent and using modified Deficit Round Robin (DRR) bandwidth, a more refined approach is required distinct ways: either as direct input to an AQM
queue. Second, it uses a single state-tracking scheduling to share link capacity between queues. to apply centralized AQM. algorithm or as input to a bitrate shaper.

22 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 23
✱ VIRTUAL ACTIVE QUEUE MANAGEMENT VIRTUAL ACTIVE QUEUE MANAGEMENT ✱

The AQM makes its drop/mark decisions based However, at the time of writing, neither the
on the relation between Average RTT, Path RTT QUIC protocol nor the described mechanism
DRR scheduling and the Target Queue Delay. If the Average RTT is a finished standard.
continuously exceeds the Target RTT for an
interval amount of time, the AQM enters a The additional benefits of vAQM
vAQM dropping state, like that in CoDel. The interval is The benefits of traditional AQM are well known,
AQM algorithm per queue determined as a function of Path RTT. If multiple namely: fast delivery of short-lived flows and
Origin flows are traversing the same bottleneck, a drop increased performance of latency-critical
server will occur on the flow that is determined to be the applications. There is an increased responsiveness
most significant contributor of queue delay. This is to short-lived flows due to scheduling at flow level
effectively determined by comparing the flight and minimal queue delay. For example, Domain
sizes of the respective flows. Name System (DNS) responses and TCP SYN/
ACK packets do not have to share a queue with
... Origin Handling of non-responsive flows a TCP download of bulk data.
server The basic principle of AQM is that it provides In recent years, it has become increasingly
Different flows
feedback to the congestion control mechanism at common for different types of applications to
the sending endpoint. However, there is a share a single bottleneck buffer. Latency-critical
possibility that senders will not react as expected applications that run in parallel with bulk download
to the signals that the AQM provides. The vAQM applications such as video streaming or file
Origin can detect such flows and apply flow-specific downloads typically experience severe performance
server shaping to a level where the flow does not generate degradation. By ensuring that the standing queues at
Dequeue rate determined congestion in the bottleneck. This feature makes it the bottleneck are small or non-existent, large flows
by measurement of RTT Flows hashed into separate queues
and flight size
safe to deploy and experiment with new will not degrade latency.
congestion control algorithms, as it reduces the In addition to the benefits of traditional AQM,
potential harm to the network and other flows vAQM is adaptive to link rate fluctuation and
sharing the same bottleneck. considers the current bottleneck link bandwidth to
Figure 1: vAQM with dynamic bitrate adjustment and flow separation the user. It also provides the benefits of centralized
Rate and RTT measurement AQM and modularity for increased flexibility.
Virtual AQM relies on passive measurements vAQM is able to simplify the deployment of AQM
vAQM with dynamic bitrate adjustment In such cases, a standard AQM algorithm such as of throughput and RTT to detect congestion. in an operator network because embedding it in a
In vAQM with dynamic bitrate adjustment, the FQ-CoDel can be applied to those queues. Path TCP is unencrypted at the transport level and single (aggregation) node is sufficient to enable the
traffic shaper dynamically adjusts its dequeuing RTT can be used as input to the AQM algorithm to can be measured trivially by storing sequence management of the entire network downstream
rate based on the measured sending rate and the determine the CoDel interval and other variables. numbers and observing acknowledgements from the point of deployment. This means that
relation between Average RTT and Path RTT. The Figure 1 demonstrates how vAQM assigns (ACKs) on the reverse path. legacy routers and radio schedulers do not need
rate of incoming data is measured and applied to a different flows to different queues managed by When the transport layer network protocol to be upgraded to include AQM. While vAQM
max filter, while the average RTT is less than or CoDel and dequeues with DRR using the current QUIC (Quick UDP Internet Connections) is used, currently uses a scheme resembling FQ-CoDel, it
equal to Target RTT. When Average RTT exceeds rate to the UE. a cryptographic envelope hides most transport is built to support pluggable AQM modules, so that
the Target RTT, the shaper will set the dequeue mechanisms. Most notably, the ACK frame is advances in the field of AQM and congestion control
rate to the largest bitrate observed while Average vAQM without bitrate adjustment completely hidden from on-path observers. can have rapid impact in live networks.
RTT was below target. Furthermore, the The vAQM without bitrate adjustment approach A specific mechanism that allows passive on-path
difference between Average RTT and Target RTT sacrifices the flow isolation that is achieved by FQ measurement of RTT is being proposed in QUIC vAQM testing
is used as an additional bounding factor on the for reduced implementation and runtime standardization [5]. This mechanism makes use of We have tested our vAQM concept on the
dequeue rate. complexity. This approach uses the RTT a single bit in the QUIC short header, the value of Ericsson lab LTE network, embedding it in a
Internal queues will start to form if packets arrive measurements as direct input to an AQM the bit cycles with a frequency corresponding to multi-service proxy that inspects traffic to
at the vAQM at a higher rate than the dequeue rate. algorithm without applying the shaping step. the RTT between two communications endpoints. determine optimal bottleneck link capacity to UEs.

24 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 25
✱ VIRTUAL ACTIVE QUEUE MANAGEMENT VIRTUAL ACTIVE QUEUE MANAGEMENT ✱

RTT (ms)
1000
4G UE 4G RAN
800
EPC
600 PDN GW
5G UE 5G RAN Serving GW vAQM
PDN
400

200
vAQM per PDN connection
and per EPS bearer
0
TCP QUIC TCP vAQM 4G UE 4G RAN
Median 95th percentile

5GC
RTT (ms) UPF UPF
1000 5G UE 5G RAN vAQM vAQM
Data network
800

600 Wi-Fi vAQM at centralized UPF

Wi-Fi UE access per PDU session and per
Local Data QoS-flow
400
Network
200
Fixed vAQM at edge UPF per
CPE
0 access session and per QoS flow for
TCP QUIC TCP vAQM local breakout PDU session
Median 95th percentile

Figure 2: vAQM in full signal strength testing scenario (top) and in weak signal Figure 3: vAQM at PDN GW of EPC and at UPF of 5GC
strength scenario (bottom)

We conducted the testing according to two We ran the tests for each scenario in three vAQM in packet core QoS flows during the PDU session establishment/
multiuser cell scenarios: at full-signal strength different ways: Figure 3 illustrates vAQM in two scenarios: Packet modification phase. Configuration will occur when
and at weak-signal strength. 〉〉 TCP with vAQM disabled Data Network Gateway (PDN GW) of Evolved SMF provides the UPF with a QoS Enforcement
The test application consisted of loading a 〉〉 QUIC with vAQM disabled Packet Core (EPC) in 4G and in the User Plane Rule that contains information related to QoS
Google Drive page and downloading a 16MB image. 〉〉 TCP with vAQM enabled Function (UPF) of 5G core. enforcement of traffic including an SDF template
We used three mobile devices to generate the In 5G, vAQM will be deployed as a network function (in other words, packet filter sets), the QoS-related
background traffic for the multiuser cell, with The results, shown in Figure 2, demonstrate a by the UPF of the 5G core. It will be configured per information including GBR and maximum bitrate,
each of them downloading 1GB files, so that reduction of the median RTT by roughly three PDU session, per SDF (Service Data Flow)/QoS and the corresponding packet marking information
we could test latency under load for multiple streams. times with vAQM enabled in the strong-signal flow – that is, the guaranteed bitrate (GBR) or non- – in other words, the QoS Flow Identifier (QFI),
The page was fetched using both TCP and QUIC, scenario. In the weak-signal scenario, the median GBR QoS flow. The non-GBR QoS flows of one the transport-level-packet-marking value. vAQM
and the eNodeB was configured with a drop-tail RTT with vAQM is not much different from the PDU session will be aggregated and handled by is self-learning, making it fully automated without
queue. The server was configured to use the strong-signal scenario, whereas the traffic without one vAQM, whereas each GBR QoS flow will be any parametrization. When sending the packets
CUBIC congestion control algorithm for both AQM displays increased latency with a 95th allocated a vAQM with a dedicated virtual buffer. on the PDU session to the UE, vAQM will read
TCP and QUIC. percentile RTT of approximately one second The Session Management Function (SMF) out the packets from the queues, recalculate the
for both QUIC and TCP traffic. will configure vAQM and its virtual buffers for the rate to the UE and adjust the bottleneck. In cases

26 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 27
✱ VIRTUAL ACTIVE QUEUE MANAGEMENT VIRTUAL ACTIVE QUEUE MANAGEMENT ✱

IETF standardization in the working with GPRS, 3G trainee program for system

the authors
Transport Area Working and IP transport. More engineers. Since then, he
GBR QOS Group. He holds a B.Sc. recently, he has worked as has mainly been working in
flows PF1
in computer science from a senior solution architect
SDF1 SDF1 Stockholm University in and engagement manager.
PF2
Sweden. Prior to joining Ericsson,
Nazari spent several years
Ala Nazari at Televerket Radio working
PF3 with mobile and fixed
QFI and Marcus Ihlar broadband and transport. He
DSCP tagged SDF2 SDF2 DN
Non-GBR ◆ is a system developer holds an M.Sc. in computer
IP packets PF4
QOS flows in the field of traffic science from Uppsala
optimization and media University in Sweden
delivery. He joined Ericsson the service layer and media
PF5
in 2013 and initially did Robert Skog delivery areas, on everything
SDF3 SDF research on information- ◆ is a senior expert in the from the first WAP solutions
centric networking. field of media delivery. After to today’s advanced user
PF6
Since 2014, he has been earning an M.Sc. in electrical plane solutions. In 2005,
working on transport ◆ is a media delivery engineering from KTH Royal Skog won Ericsson’s
Dequeuing Flow Flow UE-AMBR MBR policing Packet filter PDU session layer optimization and architecture expert. He Institute of Technology prestigious Inventor of the
scheduler queueing hashing policing on (MBR for GBR SDFs sets IP flows media delivery. Ihlar also joined Ericsson in 1998 as in Stockholm in 1989, he Year Award.
non-GBR SDFs and session AMBR participates actively in a specialist in datacom, joined Ericsson’s two-year
for non-GBR SDFs)

Figure 4: vAQM for the PDU session QoS flows at the UPF References
1. Controlled Delay Active Queue Management, RFC8289, January 2018, available at:
https://tools.ietf.org/html/rfc8289
where the PDU session has a local breakout, vAQM Conclusion 2. The Flow Queue CoDel Packet Scheduler and Active Queue Management Algorithm, RFC8290, January
is configured at the edge UPF serving the local The ongoing evolution of networks and applications 2018, available at: https://www.ietf.org/mail-archive/web/ietf-announce/current/msg17315.html
breakout; in other words, at the uplink classifier or is increasing the requirements on end-to-end 3. CoDel Overview, available at: https://www. bufferbloat.net/projects/codel/wiki
branching point. performance. Specifically, it is important to be
4. Making WiFi Fast (blog), Dave Täht, available at: http://blog.cerowrt.org/post/make-wifi-fast/
In downlink, as shown in Figure 4, the UPF binds able to serve data at high rates without causing
the IP flows of the PDU session to SDFs/QoS flows unnecessary delay due to bloated buffers. Our 5. The Addition of a Spin Bit to the QUIC Transport Protocol, IETF Datatracker (2017), available at:
based on the SDF templates using the IP packet filter testing has shown it is possible to mitigate https://datatracker.ietf.org/doc/draft-trammell-quic-spin
set (for example, 5 tuples). The UPF then enforces bufferbloat substantially through the use of a virtual
the PDU session aggregate maximum bitrate form of AQM (vAQM) that is centralized upstream
(AMBR) across all non-GBR QoS flows of the PDU rather than being deployed in the bottleneck
session. It also enforces the GBR for the GBR QoS nodes, as it is in classic AQM. This approach greatly Further reading
flows. The UPF finally tags the packets with QoS simplifies the deployment and configuration of 〉〉 E nding the Anomaly: Achieving Low Latency and Airtime Fairness in WiFi (conference paper), Toke
Flow Identities (QFIs) and hands over the packets to AQM in mobile networks, and it ensures consistent Høiland-Jørgensen et al., USENIX ATC (2017), available at: https://www.usenix.org/system/files/conference/
vAQM, which is responsible for the QoS flow. behavior across bottleneck nodes, which in many atc17/atc17-hoiland-jorgensen.pdf
The different vAQM instances associated cases are supplied by a multitude of vendors. In light 〉〉 Ending the Anomaly: Achieving Low Latency and Airtime Fairness in WiFi (slide set), Toke Høiland-
with QoS flows can contain implementations of these benefits, vAQM will be an important user Jørgensen et al., USENIX ATC (2017), available at: https://datatracker.ietf.org/meeting/99/materials/slides-
of completely different AQM algorithms or be plane function in 5G core. 99-iccrg-iccrg-presentation-1/
configured with different default parameters.

28 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 29
✱ FEATURE ARTICLE FEATURE ARTICLE ✱

five technology
trends augmenting
the connected
society
Rapid advancements in the use of machines to augment human intelligence are creating a
new reality in which we increasingly interact with robots and intelligent agents in our daily
lives, both privately and professionally. The list of examples is long, but a few of the most
common applications today are found in education, health care, maintenance and gaming.
My vision of the future network is an intelligent platform that enables this new reality
by supporting the digitalization of industries and society. This network platform consists
of three main areas: 5G access, automation through agility, and a distributed cloud.
A set of intelligent network applications and features is key to hiding complexity from
the network’s users, regardless of whether they are humans or machines.
The ability to transfer human skills in real time to other humans and machines located
all around the world has the potential to enable massive efficiency gains. Autonomous
operation by machines with self-learning capabilities offers the additional advantage of
continuous performance and quality enhancements. High levels of cooperation and trust
between humans and machines are essential. Building and maintaining trust will require
decision transparency, high availability, data integrity and clear communication of intentions.
The network platform I envision will deliver truly intuitive interaction between humans
and machines. In my view, there are five key technology trends that will play critical roles
in achieving the vision:

#1 The realization of zero touch

#2 The emergence of the Internet of Skills
#3 Highly adaptable, cyber-physical systems
#4 Trust technologies for security assurance
#5 Ubiquitous, high-capacity radio

by erik ekudden, cto

30 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 31
✱ FEATURE ARTICLE FEATURE ARTICLE ✱

#1
the realization
of zero-touch
T H E Z E R O -T O U C H networks the complexity is expected to increase. As created from unstructured data with the process. The interaction between humans with other digitalized components or continually receives feedback from
of the future will be characterized by machine learning and artificial intelligence support of semantic web technologies, and machines occurs using natural digital twins to receive direct feedback. the environment) and supervised/
the fact that they require no human continue to develop, efficiently integrating such as ontologies. The models are language processing as well as syntactical However, further advancements in robot unsupervised learning (such as
intervention other than high-level learning and reasoning, the competence created and evolved with new knowledge and semantic analysis. design and manufacturing will be needed classification, regression and clustering)
declarative and implementation- level of machine intelligence will grow. to make informed predictions and enhance to improve their dexterity. from multiple data streams. An agent can
independent intents. On the road to automated decision making. ROBOTS AND AGENTS COLLABORATE A software agent in a zero-touch be pre-trained in a safe environment,
zero touch both humans and machines AUGMENTATION OF HUMAN To maximize human trust and improve WITH HUMANS network acts in the same way as a human as within a digital twin, and transferred
will learn from their interactions. INTELLIGENCE decision quality, there is a need for In a collaborative scenario, a robot will operator. The agent should be able to learn to a live system. Domain knowledge is a
This will build trust and enable the The realization of zero touch is an iterative transparency in the machine-driven be able to anticipate human intentions the role in real time, as well as the pattern key success factor when applying agents
machines to adjust to human intention. process in which machines and humans decision-making process. It is possible and respond proactively. For example, an and the proper actions for a given task. to complex tasks.
collaborate reciprocally. Machines build to gain insights into a machine’s decision assembly-line robot would automatically In particular, it should be able to handle Techniques such as neural networks
Compute and intelligence will exist in the intelligence through continuous learning process by analyzing its internal model and adapt its pace to the skills of its human a wide range of random variations in the offer significant advantages in learning
device, in the cloud and in various places in and humans are assisted by machines determining how that model supported coworkers. Such interactions require task, including contaminated data from patterns, but the current approach is too
the network. The network will automatically in their decision-making processes. particular decisions. This serves as a basis the introduction of explainable artificial the real world that originates from rigid. Differential plasticity is another
compute the imperative actions to fulfill In this collaboration, the machines for generating explanations that humans intelligence to cultivate human trust in incidents and mistakes. These agents technique that looks promising.
given intents through a closed loop gather knowledge from humans and the can understand. Humans can also evaluate robots. Robots will work alongside humans will learn through a combination of
operation. Today’s complex networks environment in order to build models decisions and provide feedback to the to aid and to learn. Robots can also interact reinforcement learning (where the agent
are designed for operation by humans and of the reality. Structured knowledge is machine to further improve the learning

32 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 33
✱ FEATURE ARTICLE FEATURE ARTICLE ✱

THE INTERNET OF SKILLS and rendering. These capabilities make it INSTANT INTERACTION
allows humans to interact in real time possible to create a 3D representation of AND COMMUNICATION
over great distances – both with each the captured world and provide the Communication between humans and
other and with machines – and have experience of being immersed in a remote machines will become more natural, to the
similar sensory experiences to those or virtual environment. While today’s user point that it is comparable to interpersonal
that they experience locally. Current devices don’t yet provide the necessary interaction. Natural user interfaces such as
application examples include remote resolution, field of view, depth perception, voice and gesture will be commonplace.
interactive teaching and remote repair wearability and positioning capabilities, The use of vision-based sensors will allow
services. A fully immersive Internet of the quality and performance of these for an intuitive type of interaction. To better
Skills will become reality through a technology components is steadily understand human-machine interaction
combination of machine interaction improving. there is a need to evolve the understanding
methods and extended Spatial microphones will be used to of kinesiology, ergonomics, cognitive
communication capabilities. Internet separate individual sound sources in the science and sociology, and to incorporate
of Skills-based systems are space domain. This implies that there will them into algorithms and industrial design.
characterized by the interplay of be an increased amount of data needed to This would make it easier to convey a
various devices with sensing, capture the audio spatial aspects. Spatial machine’s intent before it initiates actions,

#2
processing and actuation capabilities audio rendering performance is very much for example.
near the user. tied to efficient head-related filter models. Large volumes of 3D visual information

the emergence
New formats for exchanging spatial audio impose high network capacity demands,
Current systems lack the audio, visual, streams have been specified and making ultra-low latency and high
haptic and telecommunication capabilities compression techniques are being bandwidth communication technologies
necessary to provide a fully realistic developed. essential. Enabling the best user
experience. To enable the Internet of Skills, Haptic components allow users to feel experience requires the use of network

of the internet
the interplay between humans and robots, shapes, textures, motion and forces in 3D. edge computers to process the large
and between humans and virtual content, Devices will also track the motions and volumes of 3D visual, audio and haptic
is of particular importance. Both industry forces applied by the user during information. This setup saves device
and consumers are showing great interest interaction. With current technologies the battery lifetime and reduces heat
and openness in using these new user needs to wear or hold a physical dissipation, as well as reducing network

of skills
capabilities. device, but future ultrasound based haptic load.
devices will offer a contact-free solution.
HUMAN SKILLS DELIVERED Standardization efforts for haptic
WITHOUT BOUNDARIES communication will allow for a quicker
An authentic visual experience requires adoption of haptic capabilities.
real-time 3D video capturing, processing

34 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 35
✱ FEATURE ARTICLE FEATURE ARTICLE ✱

A C Y B E R - P H Y S I C A L system on a wider operating scale, supervising system. Physical processes are

is a composite of several systems of the operation of the mostly automated compositions of many things occurring in
varying natures that will soon be and self-organizing process. parallel. A model of time that is consistent
present in all industry sectors. It is a A cyber-physical system contains with the realities of time measurement and
self-organizing expert system created different heterogeneous elements such as time synchronization needs to be
by the combination of model of models, mechanical, electrical, electromechanical, standardized across all models.
dynamic interaction between models control software, communication network
and deterministic communication. and human-machine interfaces. It is a EXAMPLE: INDUSTRY 4.0
A cyber-physical system presents a challenge to understand the interaction of The factory of the future implements the
concise and comprehensible system the physical, cyber and human worlds. concept of Industry 4.0, which includes the
overview that humans can understand System models will define the evolution of transformation from mass production to
and act upon. each system state in time. An overarching mass customization. This vision will be
model will be needed to integrate all the realized through large-scale industrial
The main purpose of a cyber-physical respective system models while contem- automation together with the digitalization
system is to control a physical process and plating all possible dynamic interactions. of manufacturing processes.

#3
use feedback to adapt to new conditions This implies a control program that delivers Humans assume the role of supervising
in real time. It builds upon the integration of a deterministic behavior to each the operation of the automated and self-
computing, networking and physical subsystem. Current design tools need to organizing production process. In this

highly
processes. An example of a cyber-physical be upgraded to consider the interactions context it will be possible to recognize all
system is a smart factory where between the various systems, their the system models that need to interact:
mechanical systems, robots, raw materials, interfaces and abstractions. • Physical and robotic systems such as
and products communicate and interact. conveyors, robotic arms and automated

adaptable
This interaction enables machine MODEL OF MODELS CREATES guided vehicles
intelligence to perform monitoring and THE CYBER-PHYSICAL SYSTEM • Control systems such as robot
control of operations at all plant levels. Within the cyber-physical system all controllers and programmable logic
system dynamics need to be considered controllers for production
SYNERGISTIC INTEGRATION OF through a model that interacts with all the • Software systems to manage all the

cyber-physical
COMPUTATION, NETWORKING AND sub-models. Many factors impact the operations
PHYSICAL PROCESSES dynamics of the interactions between the • Big data and analytics-based
The main challenge is the orchestration systems, including latency, bandwidth and software systems
of the networked computational resources reliability. For a wireless network, factors • Electrical systems to power
for many interworking physical systems such as the device location, the propagation machines and robots

systems
with different levels of complexity. Cyber- conditions and the traffic load change over • Communication networks
physical systems are transforming the way time. This means that networks need to be • Sensors and devices.
people interact with engineered systems, modeled in order to be integrated in the
just as the internet has transformed the model of models. The master model consists of and interacts
way people interact with information. The time it takes to perform a task may with all the listed processes above, resulting
Humans will assume responsibility be critical to enable a correctly functioning in the realization of the final product.

36 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 37
✱ FEATURE ARTICLE FEATURE ARTICLE ✱

T R U S T T E C H N O L O G I E S will behavior analytics. Artificial intelligence their combined data inputs while keeping
provide mechanisms to protect assists security analysts by collecting and those inputs private. In addition to
networks and offer security assurance sifting through threat information to find protecting the confidentiality of the input
to both humans and machines. relevant information and computing data, multi-party protocols must
Artificial intelligence and machine responses. However, there is a need to guarantee that malicious parties are not
learning are needed to manage the address the current lack of open able to affect the output of honest parties.
complexity and variety of security benchmarks to determine the maturity of Although multi-party computation is
threats and the dynamics of networks. the technology and permit comparison of already used in special cases, its limited
Rapidly emerging confidential products. functionality and high computation
computing – together with possible While the current trend is to centralize complexity currently stand in the way of
future multi-party computation – will data and computation, security wide adoption. Time will tell if it becomes
facilitate secure cloud processing of applications for the Internet of Things and as promising as confidential computing.
private and confidential data. future networks will require more
Performance and security demands distributed and hierarchical approaches to PRIVACY REQUIRES SECURE
are driving the development of support both fast local decisions and IDENTITIES
algorithms and protocols for identities. slower global decisions that influence local Digital identities are crucial to maintaining
policies. ownership of data and for authenticating

#4
The use of cloud technologies continues to and authorizing users. Solutions that
grow. Billions of new devices with different CONFIDENTIAL COMPUTING address identities and credentials for
capabilities and characteristics will all be TO BUILD TRUST machines are equally important. The

trust
connected to the cloud. Many of them are Confidential computing uses the features widespread use of web and cloud
physically accessible and thus exposed of enclaves – trusted execution technologies has made the need for
and vulnerable to attack or to being environments and root of trust efficient identity solutions even more
misused as instruments of attack. Digital technologies. Code and data is kept urgent. In addition, better algorithms and

technologies
identities are needed to prove ownership confidential and integrity protection is new protocols for the transport layer
of data and to ensure that services only enforced by hardware mechanisms, which security provide improved security, lower
connect to other trustworthy services. enable strong guarantees that data and latency and reduced overhead. Efficiency
Flexible and dynamic auditing and processing are kept confidential in the is particularly important when
compliance verification are required to cloud environment and prevent orchestrating and using identities for many

for security
handle new threats. Furthermore, there is a unauthorized exposure of data when doing dynamic cloud systems, such as those
need for automated protection that adapts analytics. Confidential computing is realized via microservices, for example.
to operating modes and performs analytics becoming commercial in cloud systems. When quantum computers with enough
on the system in operation. Research is underway to overcome the computational power are available, all
remaining challenges, including improving existing identity systems that use public-

assurance
PROTECTION DRIVEN BY the efficiency of the trusted computing key cryptography will lose their security.
ARTIFICIAL INTELLIGENCE base, reducing context switch overheads Developing new secure algorithms for this
Artificial intelligence, machine learning and when porting applications and preventing post-quantum cryptography era is an
automation are becoming important tools side channel information leakage. active research area.
for security. Machine learning addresses Multi-party computation enables
areas such as threat detection and user parties to jointly compute functions over

38 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 39
✱ FEATURE ARTICLE FEATURE ARTICLE ✱

#5
ubiquitous,
high-capacity
radio
T H E W I R E L E S S access network technologies and access nodes is required. deployments of network nodes. High-
is becoming a general connectivity Wireless technology will be used for the accuracy positioning is important for
platform that enables the sharing of connectivity between the network nodes, enhanced network performance and is an
data anywhere and anytime for anyone as a complement to fiber-based networks. enabler for new types of end-user services.
and anything. There is good reason to Device cooperation will be used to create The positioning of mobile devices, both
believe that rapidly increasing data virtual large antenna arrays on the device indoor and outdoor, will be an integrated
volumes will continue in the side by combining the antennas of multiple part of the wireless access networks.
foreseeable future. Ultra-reliable devices. The borderline between devices Accuracy will be well below one meter.
connectivity requires ultra-low and network nodes will be more diffuse.
latency, which will be needed to Massive heterogenous networks will A NEW TRADE-OFF BETWEEN
support demanding use cases. have a much more mesh-like connectivity. ANALOG AND DIGITAL RADIO
The focus will be on enabling high Advanced machine learning and artificial FREQUENCY HARDWARE
data rates for everyone, rather than intelligence will be key to the management For the past 20 years there has been
support for extremely high data rates of this network, enabling it to evolve and a continuous trend toward moving
that are only achievable under specific adapt to new requirements and changes in functionality from the analog to the digital
conditions or for specific users. the operating environment. radio frequency domain. However, the
trend is reversed for very wide band
A few technologies will need to be NO SURPRISE – EXPONENTIAL transmission at very high frequencies,
enhanced in order to create a ubiquitous, INCREASED DATA RATES in combination with a very large number
high-capacity radio network. The common Meeting future bit rate demands will of antennas. This means that a new
denominator for these technologies is their require the use of frequency bands above implementation balance and interplay
capability to enable and utilize high 100 GHz. Operation in such spectrum will between the analog and digital radio
frequencies and wide bandwidth enable terabit data rates, although only for frequency domains will emerge.
operations. Coverage is addressed short-range connectivity. It will be an Increasingly sophisticated processing is
through beamforming and flexibility in implementation challenge to generate already moving over to the analog domain.
device interworking. The challenge is to substantial output power and handle heat This will soon also include utilizing
support data volumes and demanding- dissipation, considering the small correlations between different analog
traffic use cases, without a corresponding dimensions of THz components and signals received on different antennas, for
increase in cost and energy consumption. antennas. Spectrum sharing will be further example. The compression requirements
enabled by beamforming, which is made on the analog-to-digital conversion is
DEVICES ACT AS NETWORK NODES possible by the high frequency. reduced. The split between analog and
To enhance device coverage, performance Integrated positioning will be enabled by digital radio frequency hardware
and reliability, simultaneous multi-site high-frequency and wide-bandwidth implementation will change over time as
connectivity across different access operation in combination with very dense technology and requirements evolve.

40 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 41
✱ DIGITAL CONNECTIVITY MARKETPLACES DIGITAL CONNECTIVITY MARKETPLACES ✱

Digital
connectivity
marketplaces
consume networks just as other ICT capabilities services to consumers under long-term contracts
are consumed. The business model of the future and competed within national boundaries.
will look more like current aaS models for other parts The collaboration between communication
of the ICT sector. The current telecom business model service providers has generally been limited to
must evolve significantly to work in this new market interoperability and roaming for the purpose
reality to place telecom companies in a better position of global service reach and to drive technology
TO ENRICH 5G AND IoT to capitalize on new business opportunities. and market scale. It is reasonable to expect the
traditional modus of collaboration to continue
VALUE PROPOSITIONS The case for business model transformation to serve the industry in the years ahead, but it is
For many years, the telecom industry has successfully also clear that this framework is not fit to explore the
been operating in what is best described as a vertical full value of network infrastructure. Rich configuration
One of the key growth opportunities for the telecom industry is to provide business model. In this model, telecom operators variances and more complex services will in many
have delivered end-to-end (E2E) standardized cases push demand beyond the capability of a single
network capabilities that support the digital transformation underway
in most businesses and industries. Already today, we have a powerful
technology foundation in place, and this will become even stronger with 5G.
Definitions
Now is the ideal time to evolve the business side of the equation toward 〉〉 Consumer: a user associated with a company that purchases services/products exposed by the platform/
platform business models, which will enable the telecom industry to prosper marketplace. A consumer can be an individual and/or an organization.

in multisided business ecosystems as well. 〉〉 Producer: a supplier of services/products to a platform/marketplace. A producer can be an individual and/
or an organization.
〉〉 Platform provider: an organization that manages the platform/marketplace.
〉〉 Platform/marketplace: the entity that exposes the services of the platform. The exposed services are a
Digitalization is undoubtedly one of the most requires network capabilities to support a broad composition of platform services and services supplied by producers.
MALGORZATA
significant forces for change of our time, spectrum of different user scenarios. Coverage,
SVENSSON, 〉〉 Digital connectivity marketplace: the marketplace that offers various connectivity services and connectivity
LARS ANGELIN, characterized by a wide variety of smart speed, latency and security are some of the key
parameters, along with new network functions service enablers to enterprises. The enterprises may require IoT capabilities.
CHRISTIAN OLROG, devices and applications that touch all
PATRIK REGÅRDH, possible areas of life. Now, new technologies for effective processing of data and distribution of 〉〉 Cloud services: a set of services including storage and computation.
BO RIBBING such as the Internet of Things (IoT), artificial applications and functions far out in the network. 〉〉 Digitalization: the use of digital technologies to change a business model and provide new revenue and value-
intelligence, wearables and robotics are 5G brings the ability to realize a wide variety producing opportunities. It is the process of moving to a digital business.
driving an accelerating digital transformation of connectivity and network services to meet
in which core processes across virtually all the performance requirements of tomorrow’s
businesses and industries are becoming more digital industries.
distributed, connected and real-time optimized. With networks capable of supporting an
Terms and abbreviations
unlimited set of services and use cases, it has become
increasingly clear that traditional mass-market aaS – as a Service | B2B – business-to-business | B2C – business-to-consumer | BSS – business support systems
■ Businesses need networks that can provide services will soon be a thing of the past. Looking | DCP – Device Connection Platform | E2E – end-to-end | IoT – Internet of Things | OSS – operations support
them with the foundation to operate the new forward, networks must be viewed as a horizontal systems | SLA – Service Level Agreement
transforming business models. Digital transformation foundation or platform on which businesses can

42 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 43
✱ DIGITAL CONNECTIVITY MARKETPLACES DIGITAL CONNECTIVITY MARKETPLACES ✱

provider, as well as beyond the scope of what is

possible to realize with traditional interoperability MULTISIDED PLATFORMS
and federation models. BRING TOGETHER TWO OR MORE
In light of this, it is necessary to evolve toward
a new collaboration framework – one that offers DISTINCT BUT INTERDEPENDENT
Consumer Consumer Consumer Consumer
service providers an environment where they can GROUPS OF CONSUMERS AND
dynamically aggregate capabilities from multiple Digital connectivity marketplace
sources into the tailored solution they require to SUPPLIERS
Application & Information Business support
enable their unique digitalization journey. enablement market services market
Such a model must excel in serving the unique and business that is designed like a platform where a market
varying needs of consumers as effectively as possible multitude of players, producers and consumers can
and include value contribution from innovation coexist and gain mutual business outcomes. In this
partners. It must also meet a new set of requirements scenario, the platform is both a marketplace that
to effectively create and capture value. Some of the supports the business of its stakeholders in the most
most critical aspects are: effective way and also an infrastructure that Cloud services market
Operation support
facilitates and simplifies these multisided services market Platform
〉〉 simplicity for service providers to define and configure relationships with a range of key functionalities. Producer Connectivity services market provider
their own connectivity and network solution regardless
of underlying infrastructure and player boundaries. Multisided connectivity platforms
〉〉 openness for service providers to easily integrate and Multisided platforms bring together two or more Digital connectivity platform
dynamically adjust their solution, including the needed distinct but interdependent groups of consumers
management, into their digital business processes. and suppliers. Such platforms are of value to one
〉〉 well-structured exposure of enhanced connectivity group (consumers, for example) only if the other Figure 1: Digital connectivity marketplace: multisided platform
and network capabilities such as distributed cloud, group (suppliers, in this case) is also present. A
edge analytics, and so on, for rapid use-case innovation platform creates value by facilitating interactions diverse characteristics will be necessary for (marketplaces), as shown in Figure 1. In this way, the
and value creation. and matchmaking between different groups. A enterprises within the industries to digitalize their services – and consequently the industries – will
〉〉 easy exchange of digital assets such as safe and secure multisided platform grows in value to the extent operational and business processes [2]. benefit from the presence of a vast range of other
trade of data across different players. that it attracts more users and it matchmakes one Digitalization of enterprises’ business and services such as information and application
〉〉 attractive complement for operators to explore currently consumer’s needs with services and products operation requires not only connectivity services but enablement services, as well as business and
unaddressable business/markets beyond the vertical offered by multiple suppliers; a phenomenon known also various cloud-based services like computation operation support capabilities. The connectivity
business they traditionally run today. as the network effect. and storage, where companies’ operations and marketplace will complement the federation model
〉〉 creation of an environment where innovation thrives The telecom industry already offers a vast variety business applications can be hosted, deployed and for connectivity and bridge the gap where the
to the benefit of all business model participants. of connectivity services, and 5G technologies operated, or where enterprises can benefit from the existing federation model is not enough.
will accelerate the development of new ones [1] cloud services’ capabilities offered in the software as The digital connectivity marketplace will enable
To effectively meet these and other requirements, it characterized by the different capabilities that the service business model. new business models by providing capabilities
is obvious that the establishment of such a business various industries require. The connectivity services The full potential of 5G connectivity and cloud for the exchange of information and connectivity
model is beyond the ability of a single player or of 5G will be realized by means of network slicing. services can be reached when they are exposed enabling services to ecosystems of consumers
company. Instead, it makes more sense to look at a We believe that tailored connectivity services with on fully digital multisided connectivity platforms and suppliers.

44 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 45
✱ DIGITAL CONNECTIVITY MARKETPLACES DIGITAL CONNECTIVITY MARKETPLACES ✱

communication (also known as massive IoT), explosion in usage and implementation in social
Application
Capability Subscription mission-critical machine type communication networks in recent years. The model in Figure 2
Edge (also known as mission-critical IoT) and extreme shows an example of how consumer requirements,
Consumer
compute
Latency Technology Subscription mobile broadband services. There will be a need modeled as graph edges, are linked with services,
x
to complement the connectivity services provided then technologies and consequently subscriptions.
Logistics 5G by others to be able to offer the desired value By defining each service as a vertex in a graph
Capability Subscription Producer proposition and satisfy industry demands. where edges constitute a consumer/producer
Security will remain one of the top priorities relationship, it is possible to model complex value
Vehicle Throughput Subscription
y
Producer
y
as societies continue to connect everything from chains or relationships between services.
Technology
autonomous trucks to elevators and ventilation
Assets systems. Isolation techniques such as automatic
4G
Producer
x
Capability Subscription network slicing are likely to play a key role in SECURITY WILL REMAIN
Service reducing security risks, together with machine-
Producer
Data
volume Subscription
z
learning-based ingress firewalls, possibly embedded ONE OF THE TOP PRIORITIES
Connectivity as virtual network functions in operators’ networks. AS SOCIETIES CONTINUE
Making the platform attractive TO CONNECT
Digital connectivity platform and efficient to operate
Matchmaking At Ericsson, we foresee the digital marketplace as
the foundation of an ecosystem of many consumers Given a set of requirements modeled as vertexes
Figure 2: Graph model: how industry requirements can be linked with connectivity services
and providers from a multitude of industries. with consumption needs as “dangling edges,”
This implies that industries will require capabilities we can find an efficient subset of the graph which
to express their requirements in languages and fulfills these needs, if at all possible. By also modeling
Our vision of the future digital connectivity services. The consumers representing various contexts that are specific to their business and non-functional aspects such as location, latency
marketplace is that it will offer interaction between industries will require connectivity and cloud operational environments, as opposed to needing and throughput, we get a generically applicable
producers and consumers by exposing the services to digitalize their business and telco and cloud-domain knowledge to be able to mechanism. This gives rise to tremendous
connectivity and cloud services on one side, and operational processes [2]. choose or define the requirements on the possibilities where a multitude of loosely connected
collecting the requirements from consumers and connectivity and cloud services. players together enable an efficient E2E service,
matchmaking them against the capabilities of the Consumer services offered by producers The multisided character of a platform – and its as a dynamically constructed value chain, capable
services on the other. The platform will facilitate the As discussed above, the future multisided matchmaking abilities – will make it well suited of reaching far beyond what is currently possible.
interactions by providing business and operation connectivity marketplace will offer capabilities to to provide these capabilities. The platform will There are many implementation options, including
support services to both producers and consumers. expose connectivity and cloud services that may match industry-specific requirements with capabilities for life-cycle management to ensure that
The services exposed in the marketplace could be be provided by multiple producers. In the context service capabilities. consumers and producers, joined by market forces
supplied by multiple producers owning the networks of 5G the following types of connectivity services The matchmaking function uses the relatively and the platform, provide a suitable mix of stability
and infrastructures required to produce these may be exposed: massive machine type old research area of graph theory, boosted by its and innovation.

46 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 47
✱ DIGITAL CONNECTIVITY MARKETPLACES DIGITAL CONNECTIVITY MARKETPLACES ✱

relationship with the platform; the entire service life healing, based on orchestration and policies, is one
cycle of the onboarded producer and consumer way to to mitigate outage situations.
Digital connectivity platform services sets; and self-services for the aforementioned. The service configuration and activation will make
One of the most critical capabilities is contracting, it possible for enterprises to define and configure
Business support systems due to the fact that the mix of services, SLAs and their connectivity services settings simply and

Cloud service integration Connectivity service integration

Security management
Interaction management Consumer/producer management
pricing schemes is complex and contracts are independently of underlying infrastructure and
frequently negotiated and renegotiated. The business player boundaries.
Consumer/producer service management support platform simplifies business by using only

Asset management integration

bilateral contracts between parties – that is, a Device Connection Platform

Service exposure
Operations support systems business-to-business (B2B) contract between One of the first realizations of a multisided platform
any two of the three: the producer, the consumer is Ericsson’s Device Connection Platform (DCP).
Service capability matchmaking
and the marketplace provider. B2C contracts are This is a platform where mobile network connectivity
Service activation Service configuration Service assurance applicable if the consumer is a private individual services are exposed to be consumed by applications.

Identity management
and not an enterprise. While business value chains Its fundamental purpose is to make connectivity
Matching engine are often referred to as B2B2x, the contracting available for enterprises to include in their offerings
arrangement supporting them is a set of linked and provide the means to manage this connectivity.
B2B and B2C contracts. Two examples of typical enterprises in this context
Data lake and analytics In addition, the marketplace allows for a party to would be an original equipment manufacturer
act as a consumer to purchase services, orchestrate of some sort (such as an automotive company or
them into a new service, and offer the orchestrated a camera manufacturer) or a service provider of a
service on the platform as a producer. A party in service where a device is an integral part of the
a producer role can also act as a consumer for its offering (such as a utility company using automatic
Figure 3: Digital connectivity platform
resource needs. This results in a series of B2C and meter reading or a point-of-sales terminal provider).
B2B contracts, which is the essence of a vibrant As mobile networks are always of a geographically-
Data lakes and analytics are used to execute Security solutions will enable the safe and secure ecosystem built on the platform. The contracting and limited nature, normally by country or license area,
matchmaking, as shown in Figure 3. Analytics trade of data between different players and the contract management processes are fully automated. there is a need to harmonize the way the enterprise’s
services will make the future platform data driven, exchange of digital assets [3] . Trust facilities will connectivity is used across several networks. Ideally,
as they will be exposed for use by either consumers enable secure transactions between consumers Operations support platform the enterprise needs to have the same connectivity
or producers to help them evolve their services. and producers. Analytics insight in the shape of The consumer and producer services that enable the experience wherever it launches and runs its service.
At the same time, the consumers or producers fraud management, combined with a novel form network capabilities such as distributed cloud The only practical way to achieve this is to standardize
provide the platform with extended insight into how of analytics insight predicting likelihood of successful and edge analytics will be exposed using service on a centralized platform, which brings together
producers’ services are used, potentially helping service activation, will be the key for the platform exposure, thereby enabling innovation and the connectivity of all providers and exposes it in
improve the recommendations supporting service provider to be able to manage risk when bridging value creation. a uniform way. This makes the experience truly
discovery and matchmaking. the trust divide. The dependencies between services are harmonized, in terms of functionality as well as service
A common solution for providing identification Service exposure will secure openness for automatically inferred through service-level levels and operational procedures. Furthermore, it
is yet another important component of the platform consumers to easily integrate and dynamically adjust assurance, and the combined analytics of the also minimizes the cost of integration between the
to truly support scale and innovative combinations. their solutions and services, including the needed different services provides insight into performance enterprise systems and the access networks, saving
Federation of identities is a key service when management, into their digital business processes. at several levels. This may allow fine-tuning of SLAs money both initially and continuously, as well as
considering new service combinations. Authorization based on heuristics and help fuel innovation by shortening time to market for the enterprise’s service.
for a connection between two things will likely take Business support platform managing and providing transparency with regard
both software fingerprinting and behavioral analysis The business support platform comprises many things to risk. A network or service outage must be dealt Conclusion
into account if this functionality is available in a for both producers and consumers: their entire life with swiftly, as they may be used in consuming The ongoing digitalization of businesses and
“thing management” system in a trusted relationship. cycle, interactions and management while in a business enterprises’ mission-critical production. Automatic industries is one of the key growth opportunities

48 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 49
✱ DIGITAL CONNECTIVITY MARKETPLACES DIGITAL CONNECTIVITY MARKETPLACES ✱

for the telecom industry. Telecom infrastructure, for the telecom industry as the provider of the
encompassing both current networks and soon-to-be- marketplace and brings opportunities for multisided
launched 5G, is rapidly evolving into a very powerful business relations and transactions to prosper. Malgorzata the company in 1994 and development and wireless architectures and the

the authors
resource that can bring significant value to the digital On one side of the platform are the businesses Svensson his previous positions have enterprise connectivity with software architectures to
transformation of most industries. Looking ahead, and industries that are consuming services from ◆ is an expert in operations ranged from strategy and a security focus to cloud and implement the systems.
some of the telecom industry’s key value levers are: the underlying infrastructure, and on the other side support systems (OSS). operations support systems/ He holds an M.Sc. in
quality differentiated connectivity; distributed are providers of both network assets and enabling She joined Ericsson in 1996 business support systems engineering physics and
computing and storage; analytics of data flows; functions. The platform organizes the relationships and has worked in various (OSS/BSS). Olrog currently a Tech. Licentiate in tele-
and security solutions. With the technology already for optimal fit and usefulness for the players involved. areas within research and sits in the Technology & traffic theory from the
in place, the main challenge is to fully understand In so doing, the platform provides a number of useful development. For the past Industry group at Business Faculty of Engineering at
the needs of a new breed of consumers, and use that functions with one of the key values being the scale 10 years, her work has Area Digital Services at Lund University in Sweden,
information to organize the business, develop new of business and the networking effect that it has to focused on architecture Ericsson. He holds an M.Sc. and an honorary Ph.D. from
relationships and establish efficient operations. offer. Instead of competing for the same consumers evolution. Svensson in general physics from KTH Blekinge Institute
At Ericsson, we believe that the best way to with the same offer, operators that restructure has broad experience in Royal Institute of Technology. of Technology, Sweden.
overcome this challenge is to establish a platform according to a logic that enables full participation
model where capabilities from many providers can in an ecosystems platform will find themselves business development to Bo Ribbing
be effectively packaged and exposed in attractive well positioned to fully capitalize on new market account management. ◆ is head of Product
ways to buyers from different industries. This opportunities. Currently based at the global Management for
approach creates a new and much-needed role headquarters in Stockholm, Connectivity Management.
he has also worked He joined Ericsson in 1991
extensively in Brazil, Thailand and has worked in the
References and Germany. Regårdh holds networks business since
an M.Sc. from KTH Royal 1995. During this time, he has
1. Ericsson white paper, January 2017, 5G – enabling the transformation of industry and society, available at:
Institute of Technology in gained broad international
https://www.ericsson.com/en/white-papers/5g-systems--enabling-the-transformation-of-industry-and-society
business process, function Stockholm, Sweden. Lars Angelin experience, particularly
2. Ericsson white paper, October 2017, Telecom IT for the digital economy, available at: and information modeling, ◆ is an expert in BSS at
https://www.ericsson.com/en/publications/white-papers/telecom-it-for-the-digital-economy information and cloud Christian Olrog Business Area Digital
3. Ericsson Technology Review, November 2017, End-to-end security management for the IoT, available at: technologies, analytics, ◆ is an expert in cloud Services. He has more
https://www.ericsson.com/en/ericsson-technology-review/archive/2017/end-to-end-security-management-for-the-iot DevOps processes and tool service delivery architecture. than 30 years of experience
chains. She holds an M.Sc. in He joined Ericsson in 1999 in the areas of concept
technology from the Silesian and has been a technical development, architecture
University of Technology leader and architect in many and strategies within the
in Gliwice, Poland. different areas, ranging telco and education
Further reading
from embedded client industries. He joined
〉〉 E  ricsson white paper, November 2015, Operator service exposure – enabling differentiation and Patrik Regårdh Ericsson in 1996 as a
innovation, available at: https://www.ericsson.com/en/white-papers/operator-service-exposure--enabling- ◆ is head of strategy research engineer, and in from Latin America and Asia
differentiation-and-innovation for Solution Area OSS, 2003 he moved to a position Pacific, and has held several
The authors
〉〉 Ericsson Technology Review, April 2017, Tackling IoT complexity with machine intelligence, available at: where his work focuses as concept developer in the management positions.
would like to
https://www.ericsson.com/en/publications/ericsson-technology-review/archive/2017/tackling-iot-complexity- on market development, machine-to-machine and Ribbing holds an M.Sc.
thank Frans
with-machine-intelligence industry dynamics and OSS/BSS areas. Since 2006, in applied physics from
de Rooij for his
〉〉 Ericsson press release, December 2017, 2018 hot consumer trends: technology turns human, available at: driving strategies and Angelin has been focusing on Linköping University
contribution to
https://www.ericsson.com/en/press-releases/2017/12/2018-hot-consumer-trends-technology-turns-human initiatives for Ericsson’s BSS – specifically business in Sweden.
this article.
digital business. He joined support, enterprise

50 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 51
✱ COGNITIVE TECHNOLOGIES COGNITIVE TECHNOLOGIES ✱

Cognitive
technologies Known
Knowing

Previous
IN NETWORK AND BUSINESS AUTOMATION facts experience

Forward-looking network operators and digital service providers require an

automated network and business environment that can support them in the
transition to a new market reality characterized by 5G, the Internet of Things,
virtual network functions and software-defined networks. The combination
of machine learning and machine reasoning techniques makes it possible to Sensing Thinking Acting
build cognitive applications with the ability to utilize insights across domain
borders and dynamically adapt to changing goals and contexts.
Figure 1: The model of mind

implementation of artificial intelligence and decision from input and knowledge.

JÖRG NIEMÖLLER , The need to support emerging technologies ■ Automation is an essential part of the solution. machine learning). It perceives its environment and An intelligent agent needs a model of the
LEONID MOKRUSHIN will soon lead to radical changes in the At Ericsson, we envision a new infrastructure for takes actions to maximize its success in achieving environment in which it operates. Technologies
operations of both network operators and network operators and digital service providers in its goals. The term cognitive technologies refers to used to capture information about the environment
digital service providers, as their businesses which intelligent agents operate autonomously with a diverse set of techniques, tools and platforms that are diverse and use-case dependent. For example,
tend to be based on a complex system minimal human involvement, collaborating to reach enable the implementation of intelligent agents. natural language processing enables interaction
of interdependent, manually-executed their overall goals. These agents base their decisions The model of mind shown in Figure 1 illustrates with human users; network probes and sensors
processes. These processes span across on evidence in data and the knowledge of domain the main tasks of an intelligent agent, and thus deliver measured technical facts; and an analytics
technical functions such as network experts, and they are able to utilize knowledge from the main concerns of cognitive technologies. The system processes data to provide relevant insights.
operation and product development, support various domains and dynamically adapt to changed model describes the process of deriving an action or The purpose of intelligent agents is to perform
functions such as customer care, and contexts.
business-level functions such as marketing,
product strategy planning and billing. Cognitive technologies Terms and abbreviations
Manually-executed processes represent a Software that is able to operate autonomously and CPI – Customer Product Information | eTOM – Enhanced Telecom Operations Map |
major challenge because they do not scale make smart decisions in a complex environment SID – Shared Information/Data | SLI – Service Level Index | TOVE – Toronto Virtual Enterprise
sufficiently at a competitive cost. is referred to as an intelligent agent (a practical

52 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 53
✱ COGNITIVE TECHNOLOGIES COGNITIVE TECHNOLOGIES ✱

machine reasoning (illustrated in Figure 2). Both Hybrid approaches to symbolic neural networks
involve making predictions and planning actions also exist. These are deep neural networks with a
Inferred Induced toward a goal. Each has its own strengths and numeric and statistics-based core and an implicit
knowledge models weaknesses. mapping of the model’s numeric variables to a
Machine learning relies on statistical methods symbolic representation.
to numerically calculate an optimized model based
on the training data provided. This is driven by Designing intelligent agents
Knowledge wanted characteristics of the model, such as low Autonomous intelligent agents support human
transfer average error or the rate of false positive or negative domain experts by fully taking over the execution of
predictions. Applying the learned numerical operational tasks. Doing this convincingly requires
Reasoning Machine Machine Predictions
model to new data leads to predictions or action them to react and execute faster than humans and
Planning reasoning learning Features
Actions Actions recommendations that are statistically closest be able to overcome unexpected situations, while
(Symbolic) (Numeric) to the training examples. making fewer errors and scaling to a high number of
An example of a learned model is the Service managed assets and tasks.
Knowledge Level Index (SLI) [2] [3] implemented in Ericsson Intelligent agents are developed and deployed
extraction Expert Analytics, which predicts a user’s level of in a software life cycle. As such, they profit from
satisfaction. The training input is measurements the encapsulation provided by a microservice
from network probes that show the QoS delivered to architecture, comprehensive and performant
Expert Training the user combined with surveys in which users state data routing and management, and a dynamically
knowledge examples their level of satisfaction. The learned model predicts scalable execution environment. The ability to create
this satisfaction level from new QoS measurements. an optimal thinking core for an intelligent agent
Machine reasoning generates conclusions from requires a good understanding of the fundamental
Figure 2: Machine reasoning and machine learning [1] symbolic knowledge representation. Widely used characteristics of machine learning and machine
techniques are logical induction and deduction. reasoning.
actions and communicate solutions. Acting The thinking phase derives its decisions It relies on a formal description of concepts in a
complements sensing in interaction with the from facts and previous experiences stored in a model, often organized as an ontology. Knowledge AUTONOMOUS INTELLIGENT
environment. The choice of techniques and tools knowledge base. The key is a machine-readable about the environment is asserted within the model
is equally diverse and use-case dependent. knowledge representation in the form of a model. by connecting abstract concepts and terminology
AGENTS SUPPORT HUMAN
For example, speech synthesis enables convenient Graph databases and triple stores are frequently to objects representing the entities to be used and DOMAIN EXPERTS BY FULLY
communication with users, robotics involves used for efficient storage. Formal knowledge managed. For example, “customer satisfaction,”
mechanical actuation, and an intelligent network definition can be achieved using concepts of “user” and “quantifies” are abstract concepts. Based
TAKING OVER THE EXECUTION
manager can act by executing commands on the RDF (the Resource Description Framework) or on these, we can assert that “Adam” is a user and “4” OF OPERATIONAL TASKS
equipment or changing configuration parameters. description languages, such as UML (the Universal is the SLI value representing his satisfaction. We
The thinking phase in the model of mind is the Markup Language) or OWL (the Web Ontology can further assert inference rules: “SLI quantifies The role of abstraction
source of the intelligence in an intelligent agent. Language). satisfaction,” “SLI below 5 is low,” “low satisfaction A person uses abstraction to distill essential
Thinking can be implemented, for example, as a causes churn”. Based on this knowledge, a machine- information from the input presented. Abstraction
logic program in Prolog, in an artificial neural Machine learning and machine reasoning reasoning process would logically conclude that provides focus and easier-to-grasp concepts as a
network, or in any other type of inference engine, There are two technological pillars on which an Adam is about to churn. It would trace the reason to base for reasoning and decisions. It also facilitates
including machine-learned models. intelligent agent can be based: machine learning and the low SLI value. communication.

54 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 55
✱ COGNITIVE TECHNOLOGIES COGNITIVE TECHNOLOGIES ✱

Interacting with a person or with another make domain experts unnecessary. Instead, the patterns in data and classifying them symbolically. domain expertise by asserting further concepts and
intelligent agent requires an intelligent agent to task of the expert shifts from direct involvement in The intelligent digital assistant’s use of image inference rules. They also design the applications
have the ability to operate on the same level of operational processes to maintenance of the models recognition and its ability to read natural language that assess data source and automatically assert
abstraction with a shared understanding of concepts that dictate the operation of autonomous agents. documents show that not all knowledge for machine knowledge. This requires staff to be well trained
and terminology. This includes, for example, how The abstraction of the models contributes to reasoning needs to originate from a human domain in knowledge management, with efficient processes
goals are formulated and how the intelligent agents the efficiency of the domain expert. A practical expert. Machine-learning-based processes can add and tools for knowledge life-cycle management.
present insights and decisions. example is the design of decision processes of expert knowledge and keep it up-to-date based on what is A well-designed meta model establishes a
systems proposing actions. These systems reach an learned from data. standard for consistent knowledge representation.
answer by checking a tree of branching conditions. In this respect, it is important to differentiate Any knowledge management competence gap
BUSINESS STRATEGY Even with a small number of variables, manually between data and knowledge. Data is values as can usually be filled by knowledge engineers, who
PLANNING IS A GOOD designing these conditions is a time-consuming and provided by the environment. Knowledge is the can listen to the domain experts and transfer their
unintuitive task. An intelligent agent can compile interpretation of these values with respect to the knowledge into a model.
EXAMPLE OF A HIGHLY the tree from knowledge about the reasons for semantics that are applied to give the data its A major task in modeling is assembling
ABSTRACT DOMAIN proposing an action. Managing the abstract rules meaning. Data and information models categorize a knowledge base according to use-case
is a considerably more intuitive because the data objects. Analytics creates further knowledge requirements. Ontologies can integrate and
Machine-learned models are numerical. They abstraction rises to the level the expert from multiple data elements and the domain context. interconnect any formally defined model allowing
manage abstraction by mapping meaning to is used to thinking at. A knowledge base preserves this knowledge for extensive reuse. For example, data and information
numerical values. This constitutes an implicit reasoning. When facing continuously changing data, models used in application programming interface
translation layer between the numerical Obtaining and managing knowledge a swarm of specialized intelligent agents can keep design constitute a foundation for asserting data
representation and the abstract semantics. The intelligent digital assistant example (see proof the knowledge up-to-date. objects. eTOM [4] and SID [5] are industry-standard
Ontology-based models are symbolic. Within of concept #1 on page 8) demonstrates an automated In machine learning, the learned model is the models contributing common telecommunication
an ontology, objects are established and linked to process that contributes knowledge. The assistant is knowledge, and training examples are the main terminology. TOVE [6] [7] or Enterprise Ontology [8]
each other using predicates. Machine reasoning generated from product manuals written in natural source. Domain experts are involved in selecting can cover business concepts. They were used in the
draws inference from this representation by logical language by a document-crawler application. Based variables and data sources, and in configuring the business analytics orchestration example [9]
induction and deduction. on existing knowledge, it identifies and classifies learning processes according to use-case goals (see page 9) for interpreting business-level questions.
The symbolic representation assigned to objects, the information provided in the documents. It and constraints. The success of learning – and An important part of the knowledge of
predicates and numeric values is convention. It is asserts this information as additional knowledge. consequently, the performance of a learning-based autonomous intelligent agents is their goals.
chosen to use the same abstraction and the same Furthermore, site data stored in catalogs and intelligent agent – mainly depends on the availability The domain expert uses goals to tell the intelligent
terminology as the domain it reflects. This facilitates inventories is automatically and continuously and quality of training data. agent what it is supposed to accomplish. Ideally,
an intuitive experience when users create and asserted in the knowledge base. This keeps the Reinforcement learning is a variant of machine they are formulated as abstract business-level goals
maintain the knowledge base. knowledge up-to-date, and the reasoning results learning that learns from a set of rules and a derived directly from the business strategy of the
Business strategy planning is a good example of a adapt dynamically to changed facts. simulation of the environment. Therefore, it does organization. This requires broad knowledge and
highly abstract domain. It deals with concepts such The intelligent digital assistant also uses image not necessarily depend on example data. However, adaptability to be built into the intelligent agents,
as growth, churn, customers, satisfaction and policy. recognition. It identifies physical elements and the learned model is also not based on experience. but it promises a high level of autonomy.
Numerical data needs to be interpreted to deliver a the current situation from images and asserts its The manual design of knowledge by domain
meaningful contribution at this level. An intelligent findings in the knowledge base. This demonstrates experts remains a major source of knowledge for IT IS IMPORTANT TO
agent performing this interpretation of data is a
valuable assistant in business-level processes.
a transformation of numeric data into symbolic
knowledge. Deep-learning based neural networks
machine reasoning. The domain experts create
a stable core framework of asserted terminology
DIFFERENTIATE BETWEEN
The introduction of intelligent agents will not are particularly successful at this task of identifying and concepts. Based on this, they express their DATA AND KNOWLEDGE

56 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 57
✱ COGNITIVE TECHNOLOGIES COGNITIVE TECHNOLOGIES ✱

Proof of concept Proof of concept

#1: INTELLIGENT DIGITAL ASSISTANT #2: BUSINESS ANALYTICS ORCHESTRATION
The intelligent digital assistant (see Figure 3) is The intelligent digital assistant is currently The business analytics orchestration use case analytics and orchestrates the execution of suitable
designed to assist field technicians who service a proof of concept implemented by Ericsson (see Figure 4) was implemented at Ericsson as a analytics applications. When the results are available,
base stations [10]. The technician interacts with the Research. We have implemented and deployed the proof of concept within a master thesis project [9]. the intelligent agent reasons about their meaning in
assistant through a mobile device. The assistant uses machine-reasoning system on backend servers. It demonstrates how the abstract level of business the context of the question and explains the answer
augmented reality to derive the base station type, The system collects sensed input, analyses concepts can be linked with the technical level of to the user.
configuration and state through object detection symptoms and presents corresponding data-driven analytics, so that intelligent agents can The inference is based on a knowledge base
and visible light communication. For example, it can maintenance procedures as a proposed series of operate across the levels. The use case starts with that contains a combination of a business concept
read the status LED of the device. The assistant actions. Domain experts have manually designed a business question that can be solved through ontology and abstract service descriptions of
provides instructions and visual guidance to the the procedural knowledge for problem resolution. analytics. An intelligent agent acts as a business analytics applications. It was built using existing and
technician during maintenance operations. Additionally, a document crawler automatically consultant, providing analytics-based assistance freely available business ontologies combined with
It downloads contextual data about the site and reads operational documentation, which allows the to a user. It analyzes the question, plans the needed manually-designed knowledge.
requests any additional information that could assistant to present documents that are relevant for
not be retrieved automatically through a query the current tasks to the technician for reference.
and answer dialogue.

Knowing / Modeling Sensing Thinking Acting

Expert rules Query Query Analytics

Object interface analysis orchestration
detection
Document CPI
crawler store Knowing
Context Knowledge Business domain concepts
data base Processing
Analytics service descriptions
Linked data Site Knowledge
adaptation data base Data interpretation rules

Query & Inference & Explanation Results Assertion

answer dialog planning Presentation of results analysis of insights

Sensing Thinking Acting Acting Thinking Sensing

Figure 3: Intelligent digital assistant Figure 4: Business consulting through analytics

58 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 59
✱ COGNITIVE TECHNOLOGIES COGNITIVE TECHNOLOGIES ✱

Machine learning and machine reasoning maintaining model consistency and the performance of the model is available, the mapping application Tiered implementation
hybrid solutions of the inference generation due to the number of interprets it in the domain context, determines The tiered implementation approach uses
Good decisions and plans are often based on knowledge elements to process. its meaning and generates a respective symbolic machine learning on the layer of specialist models
understanding multiple domains. For example, representation. This constitutes new knowledge and machine reasoning for consolidation across
experts in network operation know about network The role of machine learning that is asserted into the knowledge base. domains. This assignment of roles reflects strengths
incidents and the appropriate procedures to solve In machine learning, each additional domain Alternatively, an application incorporating a of the technology families, although a different
them. They can analyze technical root causes and contributes yet another set of variables adding machine-learned model can be linked directly selection is possible depending on the use case and
apply corrective and preventive actions. The same further numerical dimensions to the model. into the knowledge base acting as a proxy for a environment. For example, machine learning may be
experts usually also know some facts about the This introduces challenges such as the need for knowledge object. A reasoning process would successfully applied for cross-domain consolidation
broader business environment. Knowing about training examples that contain consolidated data call the linked application when the respective if training data is available. And machine reasoning
financial goals and Service Level Agreements helps samples from all domains. There is also an increase knowledge is needed. The application generates a can implement a specialist intelligent agent, for
them to prioritize tasks. By understanding the in the number of data points required to reach reply based on all currently available data. example, if it incorporates the manually-designed
application domain of a device or the concerns of a acceptable statistical characteristics. The combination Both methods create a hybrid of machine learning rules of a human domain expert.
user, they can customize the solution. They might of more dimensions and higher data volume increases and machine reasoning that enables dynamic
also know about marketing efforts or products in the processing cost. Furthermore, each change in adaptation of the reasoning results based on
development and proactively provide consulting. scope requires a full life-cycle loop including data learning and the latest data. Asynchronous assertion COGNITIVE TECHNOLOGIES
All this knowledge allows an expert to make the right selection, implementation, deployment and learning acts like a domain expert continuously updating WILL MAKE THIS FLEXIBLE,
decisions. For intelligent agents, it is a challenge to until a new model is available for productive use. knowledge. A knowledge proxy application
operate with the same amount of diverse knowledge Considering these challenges, machine-learned synchronously generates knowledge on demand. AUTONOMOUS ENVIRONMENT
and to provide an equally diverse range of actions. models are best suited to be specialists in confined However, this comes at the cost of delaying the A REALITY
tasks. A secondary layer of models can then build reasoning process.
The role of machine reasoning on the specialist insights and evaluate them in
The knowledge used in machine reasoning is a broader context. The second tier operates on Symbolic neural networks Conclusion
pure data decoupled from the implementation of higher abstraction with concepts from multiple Symbolic neural networks specialize in learning Intelligent agents with the ability to work
the inference engine. Changes in behavior and domains. However, since training examples at this about the relationships between entities. They collaboratively present the best opportunity for
extensions of scope must therefore be reached level are broad in scope, they tend to be hard to implicitly abstract from an underlying statistical network operators and digital service providers
by changing the model data rather than the obtain. Domain experts are still available, though, model, which allows them to answer abstract to create the extensively automated environment
implementation of the intelligent agent. Therefore, so using machine reasoning is always feasible. In questions directly. One example is image processing that their businesses will require in the near
machine-reasoning models are well suited to general, machine learning excels at inference that combining multiple machine-learned models. future. Cognitive technologies – and in particular a
integrating ontologies and inference rules from results from processing large amounts of data, while One model identifies the objects seen. Another combined use of machine reasoning and machine
multiple domains, if formal and semantic machine reasoning works very well in drawing learns about the relationship between the objects. learning – provide the technological foundation for
consistency is preserved. conclusions from broad, abstract knowledge. A third has learned to interpret questions asked. developing the kind of intelligent agents that will
Ideally, a layer of core concepts and terminology Due to the implicit abstraction and use of symbolic make this flexible, autonomous environment a
common to all domains should be used to anchor Hybrid solutions representation, the insights generated by these reality. These agents will have a detailed semantic
domain-specific models. This allows inference The result is an environment comprised of models would integrate seamlessly into a knowledge understanding of the world and their own individual
engines to traverse across domain borders and draw orchestrated or choreographed intelligent agents. base and further reasoning. However, getting contexts, as well as being able to learn from diverse
conclusions from all constituent domain models. Coordination and collaboration is done through reference data for learning is a challenge in this inputs, and share or transfer experience between
If the models from different domains already use the knowledge. A machine-learned model can scenario and would usually be dependent on human contexts. In short, they are capable of dynamically
similar concepts, but define them differently, a “glue” contribute its findings through asynchronous experts creating samples. As this setup has machine adapting their actions to a broad range of domains
model can relate them by introducing knowledge assertion. A mapping application is designed to learning at its core, it also does not scale well to a high and goals.
about the differences. monitor the numeric output of a machine-learned number of concerns and variables. Nevertheless,
The drawbacks of the multi-domain knowledge model or analyze the learned numeric model itself. it can find and contribute knowledge about new
base described here are the complexity of When new output is generated, or a new version relationships that was hitherto unknown to experts.

60 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 61
✱ COGNITIVE TECHNOLOGIES COGNITIVE TECHNOLOGIES ✱

Jörg Niemöller the introduction of these

the authors
◆ is an analytics and technologies into Ericsson’s
References customer experience expert portfolio of Operations
in solution area OSS. He Support Systems / Business
1. Acadia University, On Common Ground: Neural-Symbolic Integration and Lifelong Machine Learning
(research paper), Daniel L. Silver, available at: http://daselab.cs.wright.edu/nesy/NeSy13/silver.pdf joined Ericsson in 1998 Support Systems solutions.
and spent several years at Niemöller holds a degree in
2. Ericsson Technology Review, Generating actionable insights from customer experience awareness,
Ericsson Research, where electrical engineering from
September 30, 2016, Niemöller, J; Sarmonikas, G; Washington N, available at: https://www.ericsson.com/
he gained experience TU Dortmund University
en/ericsson-technology-review/archive/2016/generating-actionable-insights-from-customer-experience-awareness
of machine-reasoning in Germany and a Ph.D. in
3. Annals of Telecommunications, Volume 72, Issue 7-8, pp. 431-441, Subjective perception scoring: technologies and developed computer science from of industrial and telco use
psychological interpretation of network usage metrics in order to predict user satisfaction, 2017, Niemöller, J; an understanding of their Tilburg University in the cases. He joined Ericsson
Washington, N, abstract available at: https://link.springer.com/article/10.1007%2Fs12243-017-0575-6 business relevance. Netherlands. Research in 2007 after
4. TM Forum, GB921 Business Process Framework (eTOM), R17.0.1, available at: He is currently driving postgraduate studies at
https://www.tmforum.org/resources/suite/gb921-business-process-framework-etom-r17-0-1/ Leonid Mokrushin Uppsala University, Sweden,
5. TM Forum, GB922 Information Framework (SID), Release 17.05.1, available at: ◆ is a senior specialist in with a background in real-
https://www.tmforum.org/resources/suite/gb922-information-framework-sid-r17-0-1/ cognitive technologies time systems. He received
at Ericsson Research. an M.Sc. in software
6. Berlin: Springer-Verlag, pp. 25-34, The TOVE project towards a common-sense model of the enterprise,
Industrial and Engineering Applications of Artificial Intelligence and Expert Systems, 1992, Fox, M.S., His current focus is engineering from Peter
available at: https://link.springer.com/chapter/10.1007/BFb0024952 on investigating new the Great St. Petersburg
opportunities within artificial Polytechnic University,
7. University of Toronto, TOVE Ontologies, available at: http://www.eil.utoronto.ca/theory/enterprise-modelling/
intelligence in the context Russia, in 2001.
tove/
8. Cambridge University Press, The Knowledge Engineering Review, Vol. 13, Issue 1, pp. 31-89, The Enterprise
Ontology, March 1998, King, M; Moralee, S; Uschold, M; Zorgios, Y, abstract available at: https://www.
cambridge.org/core/journals/knowledge-engineering-review/article/enterprise-ontology/17080176D5F06DEAEA8
DBB2BAA9F8398 Further reading
9. Tilburg University, Mediating Insights for Business Needs, A Semantic Approach to Analytics Orchestration 〉〉 C  IO, Artificial intelligence is about machine reasoning – or when machine learning is just a fancy plugin,
(master’s thesis), June 2016, Alhinnawi, B. November 3, 2017, Rene Buest, available at: https://www.cio.com/article/3236030/machine-learning/
10. Ericsson Mobility Report 2018, Applying machine intelligence to network management, Stephen Carlsson, artificial-intelligence-is-about-machine-reasoning-or-when-machine-learning-is-just-a-fancy-plugin.html
available at: https://www.ericsson.com/en/mobility-report/reports/june-2018 〉〉 Microsoft Research, From machine learning to machine reasoning – An essay, February 13, 2013, Léon
Bottou, available at: https://www.microsoft.com/en-us/research/wp-content/uploads/2014/01/mlj-2013.pdf

62 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 63
✱ FIXED WIRELESS ACCESS IN LTE AND 5G FIXED WIRELESS ACCESS IN LTE AND 5G ✱

LEVERAGING LTE AND 5G NR NETWORKS FOR

Fixed
wireless
access such as xDSL. Here, the need is for high data rate
and capacity, with a corresponding level of ARPU.
Typical sold data rates are 50 to 200Mbps and
monthly ARPU levels are around USD 20-60. The
Subscriptions, data rates and consumption
The paradigms for fixed broadband and MBB are
different, both in terms of subscription offerings and
dimensioning. Fixed broadband subscriptions tend
Globally, there is a huge underserved market for broadband connections, with FWA sweet spot for this segment is in suburban or to focus on maximum data rates that are achieved
rural villages or towns that are currently under normal circumstances – that is, at low to
more than one billion households still unconnected. The growth in high-speed underserved. Some more sparsely populated areas medium load. The user traffic is often shaped so that
mobile broadband coverage enabled by LTE and 5G New Radio is opening are also addressable. it does not exceed the sold data rate. Hence, for fixed
The Connect the Unconnected segment is made broadband, the sold data rate is the normal value
up much more commercially attractive opportunities for operators to use up of cases in which fixed broadband competition is that household subscribers relate to.
fixed wireless access to deliver broadband services to homes and small and virtually non-existent, and smartphones that use By contrast, for MBB, peak rates are sometimes
MBB are the dominant way of accessing the internet. used for marketing, and normally the network
medium-sized enterprises. User expectations of access speed are relatively low. transmits the maximum rate that the mobile device
Typical sold data rates are 10 to 100Mbps and can handle. Monthly data buckets dominate the
monthly ARPU levels are around USD 10-20. Even subscription paradigm, and additional monetization
HÅK AN OLOFSSON, Unlike the country-wide decisions typically offering, the availability of fixed access and the though ARPU levels are limited in this segment, it is achieved through upgrades to larger data buckets,
ANDERS ERICSSON, made for mobile broadband (MBB), decisions corresponding average revenue per user (ARPU) has a FWA sweet spot that stretches from urban all the way to unlimited data. Hence, for MBB,
FREDRIC KRONESTEDT, about fixed broadband and targeted fixed that can be expected from customers [1]. The environments to rural villages, due to limited monthly data buckets are the normal subscription
SVEN HELLSTEN wireless access (FWA) deployments tend Wireless Fiber segment consists of those cases in investment needs. value that mobile subscribers relate to.
to be made at the local market level, and which there is a need for very high-rate offerings and
operators have a critical role to play. A number capacity as a direct alternative to high-end fixed
of different drivers govern local market broadband. The ambition is to provide fiber-like
attractiveness, as outlined in Ericsson’s speeds and handle households’ TV needs, matched Terms and abbreviations
recently published FWA handbook [1]. with a correspondingly high ability to pay. Typical ADSL – Asymmetric Digital Subscriber Line | ARPU – Average Revenue per User | CAT – Category (in LTE) |
sold date rates are 100 to 1,000+ Mbps and monthly CPE – Customer Premises Equipment | dav – Average Busy-hour Data Consumption | DL – Downlink | DSL
■ We have organized the FWA market ARPU levels of USD 50-100. The FWA sweet spot – Digital Subscriber Line | FDD – Frequency Division Duplex | FWA – Fixed Wireless Access | MBB – Mobile
opportunities into three distinct segments that we for this segment is typically suburban environments. Broadband | MIMO – Multiple-input, Multiple-output | mmWave – Millimeter Wave | NR – New Radio |
call ‘Wireless Fiber’, ‘Build with Precision’, and The Build with Precision segment is comprised of Rmin – Minimum Data Rate | TDD – Time Division Duplex | Tx/Rx – Radio Transmitter/Radio Receiver |
‘Connect the Unconnected’. Each of these has those cases where there is competition from WCDMA – Wideband Code Division Multiple Access | xDSL – DSL family (e.g. ADSL)
different characteristics mainly based on the performance-limited fixed broadband alternatives,

64 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 65
✱ FIXED WIRELESS ACCESS IN LTE AND 5G FIXED WIRELESS ACCESS IN LTE AND 5G ✱

FWA toolbox Add radio network capabilities

An existing mobile radio network, normally In an MBB RAN, radio capabilities are continuously
Desired Desired addressed Desired sold data rate Minimum data rate designed for voice and MBB, is an excellent base for added to handle more traffic, more customers and
ARPU households/site (or km2) X Mbps Rmin, dav (design criteria) offering an FWA service. Depending on the radio better app coverage. To handle FWA as an extra
network starting point and the operator’s ambitions service, some of these additions may have to be made
for FWA, there is a toolbox available to make the sooner to achieve a combined network with
network capable of handling a combination of voice, sufficient capabilities.
Network design MBB and FWA. An existing mobile operator has the significant
1. Low/medium load analysis => These tools fall into three main categories: utilize, advantage of being able to add the following radio
Make use of RAN add and densify. The particular needs of each local network capabilities and co-finance them for MBB
Sold data rate ‘utilize, add, densify’
2. Max. load analysis =>
situation can be met by deploying a well-planned and FWA:
deployment toolbox
mix of these tools. • Spectrum – upcoming wide spectrum bands in
Max. system load =>
3-6GHz and millimeter wave (mmWave) open up
Max. number of households Utilize existing radio network assets potential for providing high data rates and
The ability to utilize existing radio network assets is capacity, benefiting both MBB and FWA
Network cost a fundamental advantage that sets mobile operators • Higher-order modulation, multiple-input,
apart from start-ups or greenfield competitors in the multiple-output (MIMO) and beamforming –
Calculate business case FWA market. However, the advantage is only fully offering the potential to squeeze out the most from
realized if all relevant RAN assets are efficiently each spectrum band
Refine if needed combined for voice, MBB and FWA. If the operator • FWA-tailored software features – to enhance
chooses not to utilize existing assets built for voice performance for FWA users and to provide
Figure 1: FWA deployment analysis flow and MBB, the number of economically viable local adequate quality to MBB and FWA in shared
areas for FWA will be smaller, and the operator risks deployments
facing unnecessary competition with standalone • Additional sectors on existing sites
It is important that both consumers and operators (when available), we recommend that FWA is FWA providers. • 5G New Radio (NR) access – designed for low
(fixed and mobile) understand this crucial difference. brought into general RAN dimensioning. The radio network assets that should be utilized latency and for wide spectrum bands, creating an
Our view is that FWA will inherit the subscription Note that for fixed broadband, FWA and MBB include existing radio sites, spare capacity in excellent overall network together with LTE.
paradigms of fixed broadband rather than those of alike, there is transport aggregation above the last deployed spectrum (including associated
MBB. That is, households should pay for FWA on hop, which is dimensioned according to standard equipment), and acquired but undeployed spectrum.
the basis of data rate and not be concerned about principles and can also contribute to a varying user Existing radio sites are critical assets, whether they
THE ABILITY TO UTILIZE
data consumption. experience. are operator-owned or rented. The ‘tool’ of utilizing EXISTING RADIO NETWORK
In short, while FWA will inherit the subscription existing sites is not used by itself, but in combination
Last-hop dimensioning paradigm of fixed broadband, due to the radio with other actions to make those more cost-efficient.
ASSETS IS A FUNDAMENTAL
In FWA the last hop is wireless, so all the properties of the last hop to households, it must use Spare capacity in deployed spectrum and associated ADVANTAGE
characteristics of a wireless network apply to the modified dimensioning methods and terms from the deployed radio, baseband and transport network
dimensioning. Unlike fiber, but similar to digital MBB paradigm. equipment is quite common in FWA target areas, Densify the radio network grid
subscriber line loop length, there will be varying Figure 1 illustrates a typical FWA analysis flow. It and making use of it requires no new capital When the ‘utilize’ and ‘add’ tools have been used to
connection quality to different households. And, starts with input on the subscription and offering, expenditure. Acquired but undeployed spectrum is their full potential, densification can offer further
unlike fixed broadband overall, the last hop is radio including dimensioning criteria, which triggers a also common in FWA target areas, which makes gains. In these cases, MBB enhancements tend to be
and therefore shared, which means that speeds will selected, maximally efficient network design that radio deployment in new bands possible without the necessary as well, so the upgrade needs of MBB and
degrade with increasing network load. All these depends on the offering ambitions and network cost of acquiring new spectrum. The geographical fit FWA should be considered together and the
characteristics must be taken into account when starting point. A business case can be calculated by for FWA is excellent, since FWA targeted areas are densification of the network should be co-financed.
dimensioning an FWA network. Further, since balancing the resulting cost items of the deployment often suburban and rural, where unused spectrum is The two options for densifying the radio network
Ericsson promotes the sharing of assets with MBB with the extra revenues foreseen. most prevalent. grid are macro site densification and small cell site

66 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 67
✱ FIXED WIRELESS ACCESS IN LTE AND 5G FIXED WIRELESS ACCESS IN LTE AND 5G ✱

check status remotely. Converged operators have the needed to serve indoor-only FWA households. A • Based on the operator’s experience from similar
QUALITY ACROSS BOTH choice of reusing the fixed access CPE management final advantage of outdoor CPE is that the relative FWA areas, the average household’s consumption
SERVICES IS ENSURED system or deploying a separate one for FWA. Both performance difference between the best, median during busy hours is 0.9GB/h, corresponding to an
CPE and CPE management systems are separate and worst five-percentile users is significantly lower. average data flow of 2Mbps during busy hours.
THROUGH EXISTING network entities that generally have limited In terms of performance, indoor CPE units With the assumption that 10 percent of data is
SOFTWARE FEATURES integration with cellular networks, meaning that the normally start with CAT 6 capabilities of up to being consumed during busy hours, this would
operator can acquire best-of-breed products and 300Mbps. More advanced devices could support correspond to 270GB per month.
densification on poles. Macro site densification is an expect them to work using standard protocols. The CAT 16 up to 1Gbps and offer rank-4 MIMO. More
opportunistic approach: where new macro sites can biggest difference between the CPE alternatives is advanced CPE architectures are also being Network starting point
be found, such opportunities can be taken. Small cell the ability to achieve promised service levels, discussed, such as a split design, where an outdoor Coverage is provided by a macro network with
site densification on poles may be necessary if the especially during busy hours. window antenna is connected to an indoor unit via three-sector sites and an inter-site distance of about
macro grid is sparse and performance requirements An outdoor CPE provides the best performance, induction through the window glass. 1km. The operator has access to six FDD bands:
are high. as it has a built-in directional antenna (3.5GHz, three bands below 1GHz (typically 700, 800 and
10-14dBi) and is installed with a predictable radio Case study: the country town 900MHz), and three bands in the 1-3GHz range
Spectrum sharing across MBB and FWA link quality to the selected base station. The typical The country town example represents a market (typically 1,800, 2,100 and 2,600MHz). The MBB
Sharing spectrum across FWA and MBB enables antenna configuration has two Rx antennas, but within our Build with Precision segment, traffic in this area is handled using a subset of the
significant gains in overall spectral efficiency devices with four Rx antennas are also available. The characterized by relatively mature LTE MBB and available bands. The majority of smartphones are
because higher utilization is possible with one normal transmission mode is rank-2 MIMO, as the decent fixed broadband offerings, complemented by LTE-capable, and there is also GSM and WCDMA
‘bigger pipe’. This is explained by the trunking gain modem is expected to be installed with good line-of- terrestrial or satellite broadcast services to meet coverage to handle simpler phones. A typical macro
effect, which has been known and used in mobile sight. Most outdoor LTE devices support CAT 6 and households’ linear TV needs. The typical monthly site has two LTE carriers (800 and 1,800MHz) as
systems since their infancy, all the way from voice 20+20MHz carrier aggregation but more advanced ARPU for MBB is around USD 20, and the well as a WCDMA carrier in the 2,100MHz band,
channel capacity to LTE carrier aggregation for devices up to CAT 16 support are also available. predicted willingness to pay is USD 40 for a and a few GSM carriers in the 900MHz band.
MBB. It is also applicable to FWA. Inter-band carrier aggregation between FDD and dedicated household FWA internet service with a High-level analysis [2] has shown that the
The logical consequence of this is that spectrum TDD is especially useful, as services can be started sold rate of 50-200Mbps and unlimited data. deployed LTE capacity in western and central
assets should be shared as one pool, employing on existing FDD bands and later expanded as FWA The operator uses the following as the basis for Europe is less than 40 percent utilized, given the
carrier aggregation for LTE and dual connectivity subscribers and traffic increase. dimensioning the system: LTE smartphone subscriber density in the area. This
for LTE/NR to ensure that all resources are utilized A correctly installed outdoor CPE is directed to • The network should be designed to be able to means that there is spare radio capacity that can be
to the maximum, while securing good user the best-serving cell, leading to a lower path loss and connect at least 30 percent of households. In utilized by FWA.
experience for both MBB and FWA. Quality across increasing the value of mid-band and mmWave TDD contrast to the extensive upfront investments
both services is ensured through existing software spectrum. The large gain in signal quality is a result required in a fiber deployment, the ability to
features such as RAN slicing. of the 10dB difference in antenna gain and the design and invest for a limited market share from THE ABILITY TO DESIGN
By contrast, any artificial split of spectrum avoidance of 10-15dB in wall/window attenuation the beginning and expand later as the subscriber
resources for different services would result in losses suffered by indoor devices. Another contributor base grows is a useful property of FWA.
AND INVEST FOR A LIMITED
under-utilization of the spectrum assets. to signal attenuation for indoor devices is the deep • There is no ambition to offer IPTV over FWA, as MARKET SHARE... IS A USEFUL
indoor loss, as the device is likely to be placed in a household TV needs are assumed to be served by
Performance differences of FWA CPE types hidden location or to provide optimum Wi-Fi coverage. satellite or terrestrial access.
PROPERTY OF FWA
Using FWA to deliver broadband services requires This could contribute another 5dB in path loss. • The dominant use case is meeting all the
new FWA customer premises equipment (CPE), Whereas an indoor CPE is comparable to a households’ internet needs. Overall solution
from simple indoor nomadic devices to fixed smartphone in terms of spectrum efficiency, an • For video streaming support, households should, We recommend utilizing the existing sites, radios
outdoor-installed units, provisioned through outdoor CPE is two to three times more efficient. To when needed, experience at least a minimum data and baseband deployed to provide MBB, and
standard device retail or new methods. A CPE put it another way, for the same data consumption, rate (Rmin) of 10Mbps even during busy hours. sharing these resources across FWA and MBB
management system is likely to be needed to manage around two to three times as many households can This corresponds to one high definition TV video users. Current deployments have spare capacity
CPE in the fixed broadband sense – enabling the be served using outdoor rather than indoor units – or stream, with some margin, or a combination of both in LTE carriers and in baseband units.
operator to log in to the devices, configure them and two to three times as much spectrum would be multiple standard definition TV streams. In addition, we recommend utilizing the acquired

68 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 69
✱ FIXED WIRELESS ACCESS IN LTE AND 5G FIXED WIRELESS ACCESS IN LTE AND 5G ✱

Proportion of households served

Experienced user
LTE/NR 20MHz TDD (430 households per site)
3.0GHz data rate (Mbps)
300
LTE 20MHz FDD Add (step 4) 100%
LTE/NR 20MHz TDD
Add (step 1) 250 5% best households 90%
Sold data rate 225Mbps Add (step 3)
WCDMA
to average households LTE/NR 20MHz TDD 80%
200
Add (step 2) 70%
3.0GHz
LTE 20MHz FDD LTE 20MHz FDD
150 Average 60% Step 4
households Add (step 1)
WCDMA 50% Step 3
1.0GHz 100
40%
GSM
Step 2
LTE 10MHz FDD
Minimum data rate
50
5% worst households Served traffic load
LTE 20MHz FDD
30% Dimensioning target
(0.9GB/h at 30%)
•
(Mbps/site) 1.0GHz 20%
Rmin = 10Mbps 0 Step 1
0 50 100 150 200 250 300 350 400 GSM 10%
LTE 10MHz
0%
Capacity, C = 300Mbps per site
For example, 30% of 430 households per site with
0.25 0.5 0.75 1.0 1.25 1.5 1.75 2.0 2.25 2.5
average busy hour consumption of 1GB/h Average busy hour consumption (GB/h)

Figure 2: Performance and spectrum use of FWA deployment step 1 Figure 3: FWA deployment solution evolution to steps 2-4: spectrum use and performance

but undeployed band below 3GHz (such as Performance analysis meaning that households with worse radio some 130 households per site. At 135GB/h capacity,
2,600MHz), with a new 2 Tx/Rx radio, together with Although MBB and FWA services share spectrum environments will likely always have worse than this market can be served with an average busy hour
the existing LTE bands by means of carrier in the country town case, to simplify the presentation average data rates. In this scenario, the five percent consumption of slightly above 1GB/h – that is, above
aggregation for both FWA and MBB. Carrier of the performance analysis, our evaluation only worst-performing households experience close to the dimensioning target of 0.9GB/h (2Mbps). In
aggregation improves peak speeds as well as shows FWA. Further, we have chosen to focus on the 175Mbps at best. Therefore, it may be worth addition, MBB will benefit from the additional 20MHz
coverage for both services. The left side of Figure 2 downlink (DL) because the FWA traffic (and considering having different subscription categories; spectrum, for example in terms of increased peak rates.
shows the spectrum use of the FWA deployment at broadband traffic in general) is DL-heavy and so it may not be possible for all households to subscribe
this first step deployment. A RAN slicing feature can capacity is DL-limited. to the higher service level. Solution evolution
be applied to ensure that there is no negative impact The performance is illustrated in Figure 2. The To dimension the system, the Rmin is set to It is important that the solution is future-proof and
on MBB services (and vice versa) during peak experienced DL data rate for a specific household 10Mbps. This means that the five percent worst- can evolve to handle more connected households
loading as a result of FWA and LTE users sharing depends on its location, as with xDSL services, and performing households should experience at least and higher demand per household over time. To
the same carriers. may be up to 270Mbps in this scenario. An average 10Mbps DL data rate during busy hours. This results provide higher capacity and cope with greater
There is no need to densify the network in this household would experience around 225Mbps at in a capacity of 300Mbps, or 135GB/h, per site. As demands, operators can acquire and add a new TDD
case. With regard to CPE choices, we suggest using low system load. This could be used as the sold data long as the total traffic in all three sectors does not band above 3GHz (such as 3.5GHz) using 8 Tx/Rx
high-end 4 Rx outdoor (roof-top mounted) CPE, as rate to a typical customer. exceed 300Mbps, the Rmin requirement will be advanced antenna system radios. The multi-user
FWA speeds need to be high in this case to compete Note however that, unlike MBB, where users fulfilled. MIMO feature can be activated to provide
with xDSL services in the area. Indoor CPE may be move around and experience both good and bad Assuming there are 500 households per square additional capacity. Figure 3 illustrates how
deployed as a complement for households where radio environments, in this scenario the CPE is fixed kilometer, and an inter-site distance of 1,000m, an additional capacity can be provided in several
their performance is acceptable. and variation in the radio environment is smaller, FWA market share of 30 percent corresponds to evolution steps.

70 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 71
✱ FIXED WIRELESS ACCESS IN LTE AND 5G FIXED WIRELESS ACCESS IN LTE AND 5G ✱

Initially, the system is dimensioned to serve 30 Step 1. Assuming a 30 percent market share, an
percent of households with an average busy hour average busy-hour consumption of up to 4.1GB/h
consumption of 1GB/h. The area of the graph in can be met (outside graph range).
Figure 3 marked as Step 1 indicates the possible In summary, by using the FWA toolbox and limited
combinations of percentages of households and initial investments, and then adding TDD spectrum
average busy hour consumption for this solution. as needed, the chosen deployment is able to support
The area of the graph that is marked as Step 2 high data rates and consumption immediately at
indicates the capacity provided by an additional launch. Then, through a series of smooth solution and is currently working working at Development

the authors
20MHz. This shows that the system can serve a evolution steps, capacity can grow to more than four as a system designer Unit Networks, where he is
customer base of 30 percent with an average busy- times the initial offering. in Development Unit focusing on radio network
hour consumption of 1.9GB/h. Alternatively, the Networks. During his time deployment and evolution
higher capacity can be used to serve an increased Conclusion at Ericsson, he has worked aspects for 4G and 5G.
market share (up to 58 percent) with an unchanged The large number of underserved households around at Ericsson Research and Kronestedt holds an M.Sc.
average busy hour consumption. the world represents a profitable FWA growth in system management, in electrical engineering
Increasing the bandwidth with another 20MHz of opportunity for current 3GPP operators. Mobile-only as well as heading up the from KTH Royal Institute
TDD spectrum provides a system capacity operators can explore a new business opportunity Algorithm and Simulations of Technology, Stockholm,
represented by the area marked Step 3 in the graph. with FWA, and converged operators can add FWA Håkan Olofsson department at Ericsson Sweden.
This will serve 30 percent of households in the area as a complement to their fixed broadband strategy ◆ has worked in the mobile Mobile Platforms/ST-
with an average busy-hour consumption of 3GB/h. for certain locations as a more cost-efficient solution industry for 25 years, with a Ericsson. He previously
Again, the higher capacity could instead be used to with faster time to market. Segmented solutions are particular focus on its RAN worked at the Swedish
serve an increased market share with an unchanged needed, with subscriptions and dimensioning based aspects. He joined Ericsson National Defense Research
average busy-hour consumption, or a combination of on fixed and mobile paradigms. We believe that the in 1994 and has served Establishment (FOI).
increased market share and increased average best way to deliver future-proof broadband solutions in a variety of capacities, Ericsson holds a Licentiate
consumption. is based on the evolution of LTE and 5G NR, and mostly dealing with strategic Eng. in automatic control and
Finally, Step 4, the darkest grey area of the graph that the most promising approach is shared The authors would technology development an M.Sc. in applied physics
in Figure 3, indicates what can be achieved when a investment using the same ecosystem, assets and like to thank the and the evolution from 2G all and electrical engineering
total of 60MHz of TDD spectrum is added beyond spectrum bands for both MBB and FWA. following people for the way to 5G. He is currently from Linköping University,
their contribution head of the System Concept Sweden. Sven Hellsten
to this article:
program at Development ◆ joined Ericsson in 1993
Tomas Dahlberg,
Unit Networks. Olofsson and over the years he
References Hani Elmalky, Bo
holds an M.Sc. in physics has worked with radio
1. Ericsson, Fixed Wireless Access Handbook (extracted version), available at: https://www.ericsson.com/assets/ Göransson, Henrik
engineering from Uppsala technologies ranging from
local/narratives/networks/documents/fwa-handbook.pdf Johansson, George
University, Sweden. analogue AMPS, GSM,
Jöngren, Michael
2. Ericsson Mobility Report, November 2017, available at: https://www.ericsson.com/assets/local/mobility-report/ WCDMA, to LTE and 5G/
Kühner, Per Lindberg,
documents/2017/ericsson-mobility-report-november-2017.pdf NR. His main focus has been
Staffan Lindholm,
on product management
Reiner Ludwig,
of base stations, but he has
Claes Martinsson,
Björn Möller, Richard also worked with signal
Further reading Fredric Kronestedt processing design and
Möller, Per Arne
〉〉 E  ricsson Technology Review, Fixed wireless access on a massive scale with 5G, 2017, Furuskär A; Nilsson, Christoph ◆ joined Ericsson in 1993 to systems management.
Laraqui, K; Nazari, A; Skubic, B; Tombaz, S; Trojer, E, available at: https://www.ericsson.com/assets/local/ Schrimpl-Rother, work on RAN research. Since Hellsten holds an M.Sc. in
publications/ericsson-technology-review/docs/2017/2017-01-volume-94-etr-magazine.pdf Sibel Tombaz, Henrik then, he has taken on many physics engineering from
〉〉 Ericsson ConsumerLab, Connected homes, June 2015, available at: https://www.ericsson.com/assets/ Voigt, David Waite different roles, including Uppsala University.
local/news/2015/6/ericsson-consumerlab-connected-homes.pdf and John Yazlle. Anders Ericsson system design and system
◆ joined Ericsson in 1999 management. He is currently

72 ERICSSON TECH NOLOGY REVIEW ✱ #02 2018 #02 2018 ✱ ERICSSON TECHNOLOGY REVIEW 73
ISSN 0014-0171
284 23-3322 | Uen

© Ericsson AB 2018
Ericsson
SE-164 83 Stockholm, Sweden
Phone: +46 10 719 0000