The role of AI in 5G, SDN and NFV
There is talk that AI will be used to run network scaling capabilities and to operate zero touch service operations centers. But is this realistic?
31 Oct 2019
The role of AI in 5G, SDN and NFV
For many people, artificial intelligence (AI) still seems like part of the future, not the present, but its use is already becoming a reality in telco networks. There is talk that AI will be used to run network scaling capabilities (in&out or up&down) and to operate zero touch service operations centers. But is this realistic?
Well, according to research from 2018 by the TM Forum Digital Transformation Tracker 3, one-third of communications service providers (CSPs) have started to deploy virtualized network functions (NFV) at various parts of their networks. We will only see this trend grow. A number of operators have announced plans to become fully software-defined in the future including T-HT, a part of the Deutsche Telekom group in Croatia, recently announced on LinkedIn that it would be in that position by the end of 2022.
If AI is part of the future—or even the present—where will CSPs look for a partner to provide network analytics? TM Forum polled a group of CSPs and found that:
I think there are three possible outcomes of this. The first, that network equipment providers, traditional telco-oriented software and appliance vendors will incorporate AI and machine learning (ML) technologies (mainly open source) into their products. This will obviously have to be at the level of today’s leading AI and ML specialists, otherwise they will fall behind.
The second option is that the AI and ML specialists will figure out how to communicate with, and appeal to telcos— understanding the complexity of telecoms and learning to speak the language. They can then help CSPs take advantage of their currently-superior technology to support networking.
There is, however, a third, riskier option. CSPs bring in a combination of open source technology and new data scientists fresh out of university into network operations and planning teams. Drawing on big data, this innovation-led combination could actually start to rule the world. After all, great risk comes great reward — in this case, a competitive advantage.
However, the price of failure could be falling behind competitors that went in the ‘commercial off the shelf’ (COTS) direction. Every COTS company playing in the AI and ML domain is currently throwing billions of dollars into research and development. Why pay for your own development and worry about changing versions of different integration points, when specialists could do it for you?
This issue is probably why Gartner is predicting that by 2022, 75% of new end-user solutions leveraging AI and ML techniques will be built with commercial solutions, rather than open source platforms. However, open source functionalities will be part of the AI and ML learning solutions to run on those platforms. Business get value from data by being able to feed it to virtualized network function managers, who can then to scale the throughput up or down as necessary. The only systems that can do this are those that can forecast in real-time, detect anomalies using advanced neural networks, and use deep learning and other advanced machine learning algorithms via standardized APIs.
Right now, some Tier 1 and a few Tier 2 CSPs are starting to think about these options, experimenting with partnerships with research departments at universities, using their big data resources. Leading AI and ML software vendors have the necessary capabilities, but may not have the specific telco and network data knowledge. Both seem to have the wheels and engine on one hand, and the steering wheel on the other. The best option would likely be to bring the two sides together, rather than each trying to build the missing parts from scratch.
According to Facebook speech on TIP summit from 2018, capex savings of up to 10% can be achieved just by using AI to drive more efficient network planning. Much more can be achieved if – instead of provisioning networks to function for peak traffic, providers use scaling capabilities running on commercial software in the cloud. Knowing when to switch on and off in advance, and even detecting anomalies before customers notice the impact, will make all the difference between good and bad customer experiences.
I am often asked which statistical method or ML algorithm is best for predicting a particular type of network problem but it’s almost impossible to answer this. The many challenges and anomalies in any network that are below the alarm threshold make it almost impossible to determine which algorithm to use without actually testing some of them out. By finding the one that delivers the best results, we can run automatic optimizations of predictive models and results. Many analysts agree that the best option is likely to be some sort of enterprise AI and ML platform, coupled with the flexibility of open source technologies running on top of this platform. This has both the power and the flexibility to handle complex problems, like detecting and predicting a problem with a VoLTE call and establishing its root cause. This may sound simple enough, but even incompatibility between different voice codecs in different VoLTE networks can lead to a problem, and that’s before you have even considered a VoLTE call to a non-VoLTE device.
How, therefore, should telcos make decisions about where and how to invest in AI?
First, it depends on your company’s business goals, because AI and ML can provide results that are measurable in dollars. But, it also depends on the AI and ML competences available to you, either in-house or via strategic partners. The best option may be to trust the technology and your platform of choice, and create an agile team involving both your people and those from your technology provider to move fast and solve each challenge as they arise. Throwing all your data at the vendor might get better results in the short-term, but it may also mean giving up control of the business’s best asset to somebody else. To avoid this will mean investing in new skills and capabilities internally.
Well, according to research from 2018 by the TM Forum Digital Transformation Tracker 3, one-third of communications service providers (CSPs) have started to deploy virtualized network functions (NFV) at various parts of their networks. We will only see this trend grow. A number of operators have announced plans to become fully software-defined in the future including T-HT, a part of the Deutsche Telekom group in Croatia, recently announced on LinkedIn that it would be in that position by the end of 2022.
Predicting the future
If AI is part of the future—or even the present—where will CSPs look for a partner to provide network analytics? TM Forum polled a group of CSPs and found that:
- 54% said that they would manage with existing resources;
- 40% would find a new technology partner with a traditional telco-oriented portfolio; and
- 38% were planning to find a new technology partner specializing in AI.
I think there are three possible outcomes of this. The first, that network equipment providers, traditional telco-oriented software and appliance vendors will incorporate AI and machine learning (ML) technologies (mainly open source) into their products. This will obviously have to be at the level of today’s leading AI and ML specialists, otherwise they will fall behind.
The second option is that the AI and ML specialists will figure out how to communicate with, and appeal to telcos— understanding the complexity of telecoms and learning to speak the language. They can then help CSPs take advantage of their currently-superior technology to support networking.
There is, however, a third, riskier option. CSPs bring in a combination of open source technology and new data scientists fresh out of university into network operations and planning teams. Drawing on big data, this innovation-led combination could actually start to rule the world. After all, great risk comes great reward — in this case, a competitive advantage.
The price of failure
However, the price of failure could be falling behind competitors that went in the ‘commercial off the shelf’ (COTS) direction. Every COTS company playing in the AI and ML domain is currently throwing billions of dollars into research and development. Why pay for your own development and worry about changing versions of different integration points, when specialists could do it for you?
This issue is probably why Gartner is predicting that by 2022, 75% of new end-user solutions leveraging AI and ML techniques will be built with commercial solutions, rather than open source platforms. However, open source functionalities will be part of the AI and ML learning solutions to run on those platforms. Business get value from data by being able to feed it to virtualized network function managers, who can then to scale the throughput up or down as necessary. The only systems that can do this are those that can forecast in real-time, detect anomalies using advanced neural networks, and use deep learning and other advanced machine learning algorithms via standardized APIs.
Right now, some Tier 1 and a few Tier 2 CSPs are starting to think about these options, experimenting with partnerships with research departments at universities, using their big data resources. Leading AI and ML software vendors have the necessary capabilities, but may not have the specific telco and network data knowledge. Both seem to have the wheels and engine on one hand, and the steering wheel on the other. The best option would likely be to bring the two sides together, rather than each trying to build the missing parts from scratch.
Getting the best value from AI
According to Facebook speech on TIP summit from 2018, capex savings of up to 10% can be achieved just by using AI to drive more efficient network planning. Much more can be achieved if – instead of provisioning networks to function for peak traffic, providers use scaling capabilities running on commercial software in the cloud. Knowing when to switch on and off in advance, and even detecting anomalies before customers notice the impact, will make all the difference between good and bad customer experiences.
I am often asked which statistical method or ML algorithm is best for predicting a particular type of network problem but it’s almost impossible to answer this. The many challenges and anomalies in any network that are below the alarm threshold make it almost impossible to determine which algorithm to use without actually testing some of them out. By finding the one that delivers the best results, we can run automatic optimizations of predictive models and results. Many analysts agree that the best option is likely to be some sort of enterprise AI and ML platform, coupled with the flexibility of open source technologies running on top of this platform. This has both the power and the flexibility to handle complex problems, like detecting and predicting a problem with a VoLTE call and establishing its root cause. This may sound simple enough, but even incompatibility between different voice codecs in different VoLTE networks can lead to a problem, and that’s before you have even considered a VoLTE call to a non-VoLTE device.
How, therefore, should telcos make decisions about where and how to invest in AI?
First, it depends on your company’s business goals, because AI and ML can provide results that are measurable in dollars. But, it also depends on the AI and ML competences available to you, either in-house or via strategic partners. The best option may be to trust the technology and your platform of choice, and create an agile team involving both your people and those from your technology provider to move fast and solve each challenge as they arise. Throwing all your data at the vendor might get better results in the short-term, but it may also mean giving up control of the business’s best asset to somebody else. To avoid this will mean investing in new skills and capabilities internally.
