AI Automation of Telco Networks Requires Vision


At Huawei’s Analyst Summit last week in Shenzhen, the company shared in great detail the capabilities of its SoftCOM AI platform. This platform enables network automation across domains, from the core through the RAN, in the data center or within fixed line networks. Clearly, Huawei has put a great deal of thought, effort, and investment into SoftCOM AI. The solution required a great deal of vision, first in understanding the future trends facing communications service providers (CSPs), second in understanding the role of AI in expanding the potential of CSP networks, and third in harnessing the power of AI and productizing it.

But it was also apparent at the summit that, understandably, it is early days for AI automation for telecom networks. The level of vision, risk, and investment that Huawei has made in creating an AI-driven automation platform is a great deal higher than the level of vision, risk, and investment most CSPs have for network automation today.

Part of that hesitancy is likely due to prioritization. Tractica believes most CSPs won’t significantly embrace AI-driven network automation until they have deployed software-defined networks (SDN), but that’s only part of the story. More on what that might be in a bit.

The SoftCOM AI Pitch and Vision

CSP hesitancy wouldn’t spring from CSPs that have heard Huawei’s SoftCOM AI pitch. Huawei announced the live availability of SoftCOM AI in early 2018 and has been marketing it since. In breakout sessions through various speakers, the SoftCOM AI story unfolded:

  • A data lake for collecting, cleaning, and organizing training data with options for public cloud, private cloud or on premises.
  • An AI training platform with built-in algorithms for “exception detection, root cause analysis, optimization control and service prediction.” No coding required.
  • An AI inference framework for deploying, monitoring, scaling, and fault self-healing AI-fueled applications. The framework also has incremental learning and self-optimization capabilities, in theory to cut down on having to manually retrain with new data.

Telecom operators can use the platform for any network optimization use case they see fit, but Huawei also designed applications for the platform:

  • Traffic prediction for wireless cells: The model can predict traffic, load, and number of users in a wireless cell.
  • Alarm correlation: Analyzes network alarms, topologies, and work orders to extract alarm rules and generate an alarm correlation model. Reduces/eliminates duplicate work orders for network faults.
  • Data center energy management.
  • Fault prediction and management based on key performance indicator (KPI) anomalies.

Huawei also has outlined what it sees as the key early use cases for using the platform and tools and provided some insight into real life results:

  • Fault location and prediction for passive optical networks (PONs): Faults of passive PON components can be hard to locate, and there is no efficient remote locating method. Operations and maintenance costs are high. According to Huawei, the SoftCOM AI solution “uses NMSs… to obtain historical data from ONT/OLT optical modules … , extract features from the data, and label the features for AI training. The model obtained after the AI training can be used for inference based on real-time data to obtain the fault type and fault demarcation information, thereby implementing accurate fault location and prediction.” Huawei claims the solution improves fault location accuracy from 30% to 80%.
  • Intelligent energy saving for base stations: By some estimates, base station power consumption accounts for more than 16% of network operation costs. The SoftCOM AI solution predicts base station cell radio resource usage to develop a customized sleeping time for each cell. Accurate timing for this requires the historical data, such as time, neighboring cell relationship, event, and radio resource usage features, from tens of thousands of cells and the monitoring and dynamic adjustment of changes in KPIs and key quality indicators (KQIs). According to Huawei, in a first office application, the solution was able to reduce base station power by 15% to 25% while coverage areas remained unchanged and KPIs and KQIs were not affected.
  • Antenna parameter optimization: Using traffic prediction, the solution automates antenna parameters to create an optimal network topology. In this way, network paths are determined by traffic instead of just by physical connections. Antenna optimization is difficult and time consuming in the 4G era, where there are hundreds of antenna parameter combinations. With 5G, there will be thousands of potential combinations, making it nearly impossible to do without automation. Huawei said an operator that has deployed the solution reached optimal initial value at a site in a few days.

Long term, Huawei is striving for self-driving network automation in the 5G era and shared a vision for how that will look over time. Nearer term, the company said “we will implement 5G base station deployment intelligence, 5G slicing intelligence and 5G DC intelligence. During the deployment of a 5G site, the location and deployment parameters will be automatically recommended based on multi-attribute site planning. Low granularity features are automatically enabled, and parameters automatically optimized. For network slicing, the system predicts resource consumption based on service scenarios to recommend the optimal slicing model and orchestration policy, implementing real time resource awareness, on-demand adjustment and simplified configuration.” For 5G data centers, the vision is “intelligent scale-out configuration … energy saving, intelligent inspection, hard disk fault prediction and maintenance and intelligent O&M through mobile edge computing.”

A Lack of Urgency on AI-Driven Network Automation

SoftCOM AI appears to be a comprehensive and flexible solution for AI-fueled network automation, ready to go. Nokia, Ericsson, Juniper, and others are diligently building variations on the theme, though it is unclear how long their investments in AI have been in place. It is unlikely that any of them are as mature in their thinking as Huawei.

So why are CSPs moving slowly on AI-driven network automation? In addition to waiting on SDN, Tractica has observed that currently, there appears to be a lack of urgency around the primary pain point – the significant reduction of operating expenses. Driving down costs is the key selling feature of the near-term use of AI-driven network automation. (Longer term, automation is necessary in 5G networks because the networks are too complex and legion for humans to manage.)

Backing up this point both in Tractica’s recent conversations with Chinese and European CSPs and in a case study shared by Huawei at the summit from Hong Kong Telecom, many CSPs kick the tires on network automation solutions looking for top-line growth opportunities – and not cost savings. This is interesting since CSPs have bought into the cost savings benefit of AI when it has been applied to automating customer service and call centers. Service assurance use cases leveragable within 4G networks appear to have some momentum.

Another likely barrier is scope blindness. In Tractica’s research, we have found in many cases that CSP decisions to explore AI-driven automation solutions are driven by specific departments and not centrally. This approach lends itself naturally to a narrowly focused ROI.

However, none of this is outside our current guidance on the market. Over the next 2 years, CSPs will centralize their vision, strategy, and investment around AI capabilities. Once SDN is in place, AI-driven network automation makes more sense. And as 5G networks materialize, AI-driven network automation will become a necessity.

Comments are closed.