I attended the 2024 Ultra-Broadband Forum 2024on 30 October through 1 November in Istanbul, Türkiye. My blog on the overall program can be found at https://www.acgcc.com/blogs/b/recap-ultra-broadband-forum-2024/. 

I was the moderator of the ADN Summit Advancing Toward AN l4: Empowering Superior UBB Operations, sponsored by Huawei. I also gave a talk on AN evolution and the challenges I see in future network automation work. Here I present what I found to be the highlights of the presentations along with selected charts from the speakers. I offer MyTake on the important issues, these are purely my opinion.

Presentations

Dr. Philip (Xiaodi) Song, Huawei, Chief Marketing Officer, Carrier Business Unit
Welcome

Huawei welcomed approximately 100 participants to the ADN Summit and invited them to see the demonstrations available in the adjoining exhibition area. The summit focused on Huawei’s Autonomous Driving Network (ADN), the Huawei implementation of the TM Forum Autonomous Network vision.

Aaron Boasman-Patel, TM Forum, VP Innovation
How Do We Pave the Way to AN L4?

There now exists a corpus of TM Forum detailed information and proven plans for implementing Autonomous Networks. There is growing consensus on what AN L4 is and how to implement it, with many of the TM Forum member CPSs publicly committing to implementing L4 by 2026 in one or more of what are now 15 use cases.

Over the last few years, however, AN L4 has been re-defined, pushing out some of the features that were anticipated to be in AN L2 and L3 operations. In particular, AI helpers (called CoPilots) have been found to greatly increase the efficiency of technicians by providing them easy access to real-time network information and suggestions for actions based on that information (“machines helping humans” was the hallmark of AN L2). AN L3 has also been focused more on service assurance, with proactive service assurance found to be very valuable.

AN L4 has also been redefined and broken into two phases. AN L4 Phase1 focuses on providing closed-loop operations within a domain (previously in L3) with CoPilot help, when needed. AN L4 Phase 2 provides the cross-domain automation that has always been the hallmark of AN L4.

The TM Forum continues the work on AN, with multiple working groups further defining AN L4 as well as expanding the use cases of AN L3.

MyTake: This has come about as the quick rise of AI LLM technology has been found to be of great use in providing help in the service assurance area, both proactive and monitoring and in providing CoPilots to help technicians more easily and quickly understand the root cause of issues. It is a tribute to the TM Forum AN project that it pivoted to take advantage of this new technology in the program, bringing quick, real value to the CSPs. I am also not surprised that several other features originally defined in L3 were moved to L4. Cross-domain orchestration is hard, especially in a multi-vendor environment, and much of the overall value of AN is more easily gained through good domain automation.

Dang Wenshuan, Huawei, Chief Strategy Architect
Envisioning Telcos in the Age of AI

AI will permeate most areas of CSPs’ AN operations and service offerings in two aspects:

• AI used in CSPs in their own operations and as a part of new service capabilities (“AI for Networks”) and
• Networks support for AI increased bandwidth and lower latency needs, especially as AI is distributed across the architecture, from end-user devices, through private clouds, edge computing resources, and regional and centralized data centers (“Networks for AI”).

The six areas constitute the framework for the intelligent operations and offering of the future.

Key to realizing autonomous operations are four components:

• Intelligent network elements that provide a rich set of telemetry data and can be remotely configured,
• A digital twin of the network that provides a detailed, real-time view of the network equipment,
• AI CoPilots that can provide intelligent assistance to technicians and field personnel,
• AI Agents that can autonomously perform many tasks.

Several specific examples were cited:

• An AI Agent that reduces the KPIs of low-rate cells by 20% and shorten the network optimization period from one day to one hour.
• An AI Agent that can provide closed-loop operations for 70% of the change scenarios and reduce the change processing duration from 60 hours to 4 hours.
• An AI CoPilot that can reduce the average home broadband troubleshooting time from 60~90 minutes to 30 minutes and reduce the second visit rate from 10%–15% to 5%.

More details can be found in the newly released Huawei document, Strriding Toward the Intelligent World 2024 —Autonomous Driving Networkwhite paper. 

MyTake: AI will, indeed, permeate all areas of CSPs and most enterprises, bringing much more flexibility and efficiency. The use cases cited are quite compelling. However, the complexity of creating, training, retraining, and evolving the multiple AIs and multiple digital twins (optical, IP, wireless, core network, etc.) will be a challenge and will lead to focusing on specific use cases, as is being done. The existence of legacy equipment in the network that does not provide the required rich set of telemetry and remote configuration capabilities, and the complexity of a multi-vendor environment will also hamper AN implementation. Cross-domain orchestration for service provisioning and assurance will be particularly hard in these circumstances.

Adnan Al Alawi, AWASR, CEO
HBB Digital Experience Operation Drives AWASR Continuous Leadership

AWASR, a leading home broadband CSP in Oman, has prided itself on providing an exceptional customer experience. However, a POC of the Huawei iMaster-FAN automation product demonstrated that much more could be done to increase customer satisfaction while reducing operations costs and increasing ARPU.

MyTake: This was an impressive talk. AWSR’s POC was completed quickly, and it is now going into implementation. With CSPs usually rated poor in Net Promoter Scores, the push for more proactive operations is compelling and will gain market share.

William Yue, Huawei, President of NCE Optical Network Domain
Intelligence Empowers F5.5G Premium All Optical Network

The use of iMaster NCE with the Huawei all-optical network has many benefits.

Key to this approach is to create a synthetic “Customer Experience Index” (CEI) score for areas and even individual customers. This is an algorithm that uses internal network KPIs to rate the experience from the users’ perspective.

MyTake: Yes. The proactive approach of improving CEI is a more effective way of focusing on customers than the traditional focus on (usually somewhat arbitrary) internal metrics.

John Ji, Huawei, Vice President of NCE Data Communication Domain
AI Powers to Build an IP Bearer Network with High Reliability and Deterministic Experience

The IP layer is a separate operation domain in CSPs and has its own versions of the AIs, digital twins, and algorithms. Huawei has released the HUAWEI Xinghe Intelligent ADN solution that marries Huawei NetMaster with the iMaster NCE and the IP network equipment.

Three use cases demonstrate the capabilities of the solution:

• Automated optimization of the IP network provides greater efficiency and customer satisfaction, leading to a 25% uplift in traffic carried, without additional network resources.
• Automation of more than 3,000 network changes with no errors, which used to take 6 hours of five-year expert’s time.
• Auto-troubleshooting IP network faults, reducing trouble durations by 90% and overall, by 30% mean time to repair.

MyTake: IP configuration errors are still the major source of long-term network outages or slowdowns. Automated optimization with simulations of proposed network changes on digital twins represents the best hope to reduce these outages. The era of “real men use command lines for IP equipment” is finally over.

Dr. Mark H Mortensen, ACG Research, Principal Analyst
How Do Carriers Start Their AN L4 Journey?

The implementation of many use cases to achieve AN L3 autonomous operations is known art. The TM Forum has a rich set of information and training to support that. If you have not started AN L3 activities, you should. Deployment of L3 autonomy can garner about two-thirds of the benefits. The rest will come from L4. Implementation of AN L4 capabilities in selected use cases is going on now.

There are challenges ahead, including:

• The costs of creating, training, and retraining AI CoPilots and Agents: The best approach today is a layered one: create an LLM that knows how to talk to humans, imbue it with telecom knowledge, add the equipment manual information and APIs, train it in local network processes and procedures.
• Determining how effective a given AI agent is versus other options is equivalent to “what school did you go to and what certifications do you have” in a human.
• Managing the overall life cycle of AIs as they “drift” over time.  When to retrain? Retraining can be very costly if the AI is not architected correctly or is done too often. But if not done when needed, the quality of its work degrades.
• Implementing the AIs for new services: The AIs are not smart enough to figure out new procedures and processes. This will have to come from human experts in pilot programs between vendors and CSPs.
• Technology training: When the easy problems have been automated, how do we train the next generation of network engineers to handle the tough problems? This is a fundamental challenge of automation in all industries.

MyTake: There are challenges, but the CSP and vendor leaders in AN have shown the way and generously shared their knowledge of how to achieve this.

Conclusion

AN is becoming known art. It is difficult, but it can increase customer satisfaction and ARPU, decrease operations costs, and speed operations, making CSPs more competitive.