AI come “cervello” della 5G Digital Platform di TIM

AI come “cervello” della 5G Digital Platform di TIM
 

AI come “cervello” della 5G Digital Platform di TIM

 

While AI (artificial intelligence) is among today’s most popular topics, it was born in 1950 and went through a first hype cycle between 1956 and 1982. What seemed to truly spark the field of AI was a question proposed by Alan Turing in 1950: can a machine imitate human intelligence?

 

Figura 1 - The long history of AI ( AI past, present, future) - (Source: Ovum)

This triggered debates and research that culminated in the conference organized by professor John McCarthy in Dartmouth in 1956 that gathered twenty pioneering researchers to "explore ways to make a machine that could reason like a human, was capable of abstract thought, problem-solving and self-improvement". At this conference the term artificial intelligence was coined and where AI gained its vision, mission, and hype.
Notwithstanding exemplary research that followed, it wasn’t' until 2005 that the pace of research and development picked up substantially thanks to important technology developments that led to an AI rebirth, especially with respect to machine learning and, more specifically, deep learning. Chips were faster and cheaper and could support the processing speeds needed by AI and cloud-based storage and compute capacity were increasingly available on an on-demand basis.
We are currently in yet another Artificial Intelligence hype triggered by the wonderful opportunities that all the data generated by the IoE (Internet of Everything) will produce and by the increased complexities in telco infrastructure deployments: AI is consistently listed among the major technological trends alongside 5G.
In this article we will explore how AI plays a fundamental role on telco’s digital transformation and why it is at the heart of TIM’s 5G Digital Business Platform approach.

 

Figura 2 - The Second Machine Age (Source: Brynjolfsson.McAfee-Cummings)

Telco challenges

As Kodak’s story teaches us, when your core business is going well, the incentives for change are weak. The situation Telcos are in today however, is very different from the one Kodak was in at the dawn of the digital photography age, and digital transformation seems to be a necessary evolution.
Over the past ten years competition and the proliferation of over-the-top players has significantly impacted telco revenues. According to GSMA Intelligence data, in Europe the average ARPU has decreased 37% in the past ten years.

 

Figura 3

In the same period Telcos have experienced an increase in Opex and Capex investments due to the increase in traffic in the networks and to the introduction of newer generations of mobile technology.
Since the introduction of data in the mobile network technology that triggered the merger between telecommunications and internet, telecom providers have seen an increased decoupling of mobile revenue and network cost.

 

Figura 4 - Decoupling of mobile revenue and network cost (Source: Analysys Mason)

5G will open opportunities to capture value from new 5G use cases and widespread adoption of the IoT (Internet of Things), but it will increase infrastructure investments in 5G while telcos are still investing in upgrading 4G infrastructure to cope with growing traffic demand. In an analysis of one European country, McKinsey & Company [note 1]  predicted that network-related capital expenditures would have to increase 60 percent from 2020 through 2025, roughly doubling TCO (Total Cost of Ownership) during that period.
Unlike what we’ve learned from the Kodak experience, the conditions are mature for Telcos to take the opportunity of 5G to introduce a wider deep technological transformation in their technical approach to infrastructure deployment and operations.

AI for telcos

AI is a broad term that includes different areas spanning from natural language, to deep learning passing by robotics process automation. In general, we can think of it as a human-like intuition that links analytics - which is a way to make correlations - and automation - which is a final action in a sequence.
The most common form of AI today is ML (Machine Learning). Rooted in statistics and mathematical optimization, Machine Learning is the ability of computer systems to improve their performance by exposure to data without the need to follow explicitly programmed instructions. Machine Learning is the field of study that gives computers the ability to learn without being explicitly programmed, sort of like teaching a dog how to catch a ball as opposed to teaching someone to follow a cake recipe. The first one is by trying giving feedback and letting the dog figure it out; the second is a step-by-step instructions to be followed.
Going forward ML will be key to making sense of the explosion of data generated by IoE as it is able to automatically spot patterns in large amounts of data that can then be used to make predictions.
5G, in fact,  is the first mobile standard built natively to enable both human and non human use cases. 5G will enable the IoE through connected hardware with embedded sensors that will digitalize all sorts of information: air humidity, energy consumption, number of people in transit, heartbeat, driving pattern, etc…  GSMA estimates that by 2025 there will be 25 billion IoT connections. Making sense of all this data will require advanced techniques such as Machine learning to drive out patterns and deviations.
Besides effectively enabling a whole new stream of IoE, 5G also presents an important opportunity for telcos to drive digital transformation within their own infrastructure.  5G in fact, introduces innovative technological features that enable a more flexible network. The Radio Access will be composed of a combination of small cells alongside macro cells, antennas will become active with the possibility of beam forming, virtualisation of the RAN will enable the separation of the antenna and the baseband unit with the possibility of centralising the latter. As a consequence, topologies will grow more complex with small cells and new antennas, usage patterns will become less predictable from humans alone, the radio propagation models will become harder to compute with new radio spectrum bands and denser topologies.
On the Core Network side, the SDN and NFV trends will develop further to mold into a new cloud native infrastructure based on micro-services and containers. Today’s networks are hugely complex already, and, going forward, they will acquire a whole new level of sophistication.
Network Operators will require more dynamic methods and Ai-based technology offer a useful support to get the best from the infrastructure. We are at the brink of AI/ML development whereby AI technologies will increasingly permeate processes and procedures supporting human activities and providing a variety of services on the networks and a higher quality, for the benefit of all.How the telco infrastructure is changing (and the 5G Digital Business Platform).
TIM’s approach to 5G, builds on the paradigm of the 5G Digital Business Platform that was discussed in the previous number of the Notiziario Tecnico. What we will address here is how and why AI is a fundamental element of the 5G Digital Business Platform.

 

Figura 5 - The 5G Digital Business Platform: Functional view

Radio

Mobile networks are designed and managed by telecom experts who rely on their extensive knowledge of the network topology, the subscriber’s mobility/usage patterns and the radio propagation models to design and configure the orchestration of the network. With 5G however, this traditional approach will be challenged as the variables to take into consideration will grow significantly.
First and foremost, 5G introduces the use of mmWaves therefore enabling the flexibility of using a combination of macro cells or small cells for coverage). This will introduce diversification and possible densification, allowing a wider choice of options for radio planning.
Radio components will be more flexible in their deployment as openness and virtualisation are introduced. Today, radio architectures are distributed with Radio and Baseband Units part of an integrated solution that is deployed as a whole allowing for quick rollouts, ease of deployments and standard IP-based connectivity. Several operators however are testing and deploying centralised architectures  whereby the Baseband Unit is closer to the core network and serves potentially several Radio units. The connectivity between the Radio Unit and the Baseband Unit needs to ensure very low latency which can imply the deployment of fiber directly to the Radio Unit.

 

Figura 6 - Optical Communications (Source: Corning)

Having to bring fiber directly to the Radio Unit opens up a new element to take into consideration in the radio planning and clearly the advantages of a centralised architecture need to be balanced out with the additional complexity of new elements that need to be factored into radio planning.
5G will also see an increase use of beamforming whereby the antenna lobe can be dynamically oriented horizontally of vertically. Active sharing will also be possible allowing operators to contain capex investments by sharing the radio units. Without getting into radio technical details, this implies that the time synchronization and coordination with adjacent bands/operators has to be carefully considered in the TDD radio deployment.
Ericsson estimates that the number of parameters per base station that need to be taken into consideration in the radio planning has jumped from 500 in the 2G to 1500 with 3G and close to 3500 with 4G. A dramatic increase in the number of parameters in 5G can easily be imagined and AI/ML techniques will be necessary to assist good radio planning and dynamic (re)configuration.
Usually complexity has a cost: the study below by McKinsey shows how the total cost of ownership of the access network will increase with the deployment of 5G.

 

Figura 7 - Total cost of ownership for mobile access networks will increase (Source: McKinsey&Company)

Core

On the core network side the trend that started several years ago around NFV and SDN will continue evolving towards a cloud native infrastructure.
The networks will be studded with sources of data integrated with different network functions. These varied "perspectives" can provide rich insights upon correlated analysis when an architecture component to enable the ML functionalities to collect and correlate data from these varied sources in the network is introduced. The fact that the sources may report in different formats leading to heterogeneous data will need to be managed.
5G introduces the concept of network slicing that allows a network operator to dynamically provide a dedicated virtual network with functionality specific to the service or customer over a common network infrastructure.
This technical feature enables interesting opportunities from a commercial point of view, but also introduces some challenges operators have yet to address. How dynamic will the slices be? Who will be responsible for creating them? Will the slices simplify current approaches to traditional businesses such as roaming and MVNO?

 

Figura 8 - Source: C.K. Chung, Georgia Tech)

The role of AI

With 5G, topologies will grow more complex with small cells and new antennas, usage patterns will become less predictable for humans alone, the radio propagation models will become harder to compute with new radio spectrum bands and denser topologies. Future networks will have multiple technologies coexisting side by side, e.g., licensed and unlicensed wireless technologies, fixed mobile convergence (FMC) technologies, legacy and future technologies..
Telcos will have to base their investment decisions on an increased number of variables and on very granular and complex return of investment assessments.
Machine intelligence will play a key role in assisting operators in engineering and operating networks. More and more policies will be machine-learned, leveraging on constant measurements from the field and best-in-class simulators, together with a constant supervision and training by the best human experts.
AI will need to be at the heart of networks as it is in TIM’s 5G Digital Business Platform approach. And its many facets make it applicable in several areas:

  • Help manage the increased complexity in infrastructure engineering and optimizing investments
  • Facilitate and improve infrastructure operations
  • Enable new approaches and opportunities by bringing and linking together different components
  • Increasing customer knowledge and the quality of the service provided
  • Enhance security

The figure below produced by GSMA shows different areas of AI applications for telcos.

 

Figura 9 - Operators’ Strategic positioning on AI – Three levels of activity (Source: Courtesy to Javier, GSMA)

Machine intelligence capabilities can also have a local or “global” flavor. ML will be added to several layers in the 5G architecture to enable data processing for various purposes, both locally (close to where data is created) and centrally (where data can be consolidated).

  • Closer to the edge, in distributed sites, local learning and decision making can be done; Each local site is a rich source of data about the state of the different components, the time series of events and associated contextual information. This can be used to build models for local behavior;
  • Across sites, data and knowledge can be blended for a comprehensive global understanding of networks, services and functions; reasoning is required for the knowledge gathered across sites in order to infer system-wide insights.

As the increased complexities of engineering and operating telco infrastructure lead to a natural introduction of AI technologies, it is important to address what AI feeds upon: data.

 

Figura 10 - Local and global learning and decision making in large distributed networls (Source: Ericsson “AI and ML in next generation systems”)

The AI fuel: Data

We addressed the aspect of technological availability of cheaper processing power and cloud solutions as being an enabler to the recent hype in AI. However, the increased availability of data is another key element that is enabling a spreading of ML solutions.
Data is the foundation for artificial intelligence and machine learning as these are based on training algorithms by feeding them data and providing feedback. Just like many children are taught that “practice makes perfect”, ML needs training and the more data you train the ML algorithms with, the better the algorithm.
In their 1998 book “Information Rules”, economists Carl Shapiro and Hal Varian define digitalization as encoding information as a stream of bits. Digitization, in other words, is the work of turning all kinds of information and media—text, sounds, photos, video, data from instruments and sensors, and so on—into the ones and zeroes that are the native language of computers.

 

Figura 11 - The Data Science Hierarchy of Needs Pyramid (Source: "The AI hierarchy of needs" Monica Rogati)

In a world of connectivity and internet, zeros and ones are nearly instantly transferrable to anywhere globally and have close to zero marginal cost of reproduction. Data is the new gold and ML needs plenty of it.
According to a July 2012 story in the New York Times, “The combined level of robotic chatter on the world’s wireless networks . . . is likely soon to exceed that generated by the sum of all human voice conversations taking place on wireless grids.”. This seems to be confirmed by Cisco’s Visual Networking Index Forecast and trends 2017-2022 according to which annual global IP traffic will reach 4.8 ZB per year by 2022, or 396 exabytes (EB) per month. In 2017, the annual run rate for global IP traffic was 1.5 ZB per year, or 122 EB per month. Also, the same study, suggests that by 2022 there will be 12.3 billion mobile-connected devices enabled to produce data, which approximately amounts to 1.5 device per capita.
As Ovum puts it [note 2], architecting AI begins with data. Enterprises on the path to adoption must focus first on understanding all data points, integrating them, complying with regulations, and understanding the customer journeys.

 

Embracing AI is more than introducing the technology

Introducing AI/ML within consolidated business goes beyond simply introducing a new technology as it requires a new approach based on Design Thinking paradigm.
AI/ML can be applied to many different areas such as supporting self-organising networks, providing robust predictions to enable pro-active data-driven strategies, making sense of the increased complexity, simplifying existing processes, offering better customer experiences, enhancing security, predictive maintenance, and the list goes on.
The context and the goal of every AI solutions needs to be discussed and designed. Thanks to knowledge of the field of application it is possible to identify a goal and set the appropriate KPIs; data then needs to be selected appropriately to ensure correct training of the algorithm. At last the training of the algorithm entails tweaking parameters in order to get as close as possible to the goals and kpis. The first phase is fundamental and requires subject matter experts to provide the context and the goals and AI experts will support in identifying the best AI algorithms and the necessary data to train the algorithm.
The process of shaping the “problem” to be addressed with AI is a very important phase that is better addressed if approached through a Design Thinking paradigm. Design Thinking addresses the issue of shaping the “problem” and includes three subsequent sub-phases: empathise, define and design. What comes after having framed the problem in the Design Thinking phase, is the solution development phase which can take on the more common techniques of Lean Startup and Agile paradigms.
Therefore embracing AI/ML also means embracing new ways of working and new culture set around Design Thinking paradigm, possibly bringing the methodology onward, taking into account a deeper insight into the problem underlying “data”, using proper exploratory tools,  instead, for example, of the collection of end user feeback.
Another important aspect of ML algorithms is that they are living algorithms that can keep evolving as they process more and more data. As a consequence, it is important to accompany them in their evolution. Unlike more traditional software algorithms based on linear models that, once implemented, will remain unchanged unless updated, ML algorithms will evolve as they process data.
Last but not least, AI is not the solution to all problems. There may be some instances where ML algorithms will not offer improvements with respect to traditional algorithms and it is not always obvious to predict when this will happen. It is therefore important to approach AI with an open mind accepting the possibility that sometimes the outcome will not be the desired one. On the other hand, the lack of control and explainability of the results obtained may be deemed not acceptable, either from an organizational, legal or Customer Experience point of view.

 

Conclusion

The convergence of mobile and internet that has taken place since 2006, has created a new wave of innovation based on the so called over-the-top. The paradigm of the always connected mobile technology with the power of smartphones has triggered innovation in digital applications and social networks. These have boosted the amount of available digitalized personal data and contributed to the initial introductions of AI.
Going forward it is estimated that a new wave of data will be generated by the IoE which will enable new opportunities for AI applications.
But fundamentally AI will be key to future business and digital transformation. For the telco industry, it will be a necessity in order to drive increasingly autonomous and intelligent networks and improve customer experience through greater learning of customer behavior.
As AI is based on data, networks will become studded with sources of data that may be very heterogeneous. Also, new ways of working will slowly spread based on Design Thinking paradigm or its evolutions.
Networks will be interlaced with AI algorithms to help make sense of the increasing complexity and provide a better, more flexible infrastructure.