The holy grail of zero-touch automation

Telcos are awash with promises of efficiencies to be made through the adoption of automation. The concept of ‘self-organizing networks (SON) has been around for some time. Back in 2012, for example, papers such as that published in Mobile Networks and Management prophesied that “Future wireless access networks, e.g. LTE and LTE-Advanced, will be empowered by Self-Organizing Network (SON) mechanisms with the objective to increase performance, reduce the cost of operations, and simplify the network management.” 20 years ago, the concept of “A self-organizing network that grows when required, 2002” was published by Science Direct.

A further 10 years back, in 1992, a research paper from Science Direct described how “Organizations acquire decision support systems to deliver the information needed for decision making.” We must go as far back as 1988, to identify when SON was first introduced as a means of increasing productivity for telecommunications networks. In this paper, the author explains “Until now, we have generally designed protocols for complex telecommunication networks, founded on the assumption that these networks are static. This assumption facilitates the design of routing algorithms, algorithms for network management, and network topology control.” The fundamental principle was the requirement to move to open networks, standardized APIs, and dynamic allocation of network resources.

34 years on, these principles are still as relevant today. Furthermore, many of the SON use cases defined to deliver on productivity and efficiency gains, present the same business value promise as they did 20 years ago. A Light Reading article in 2017 described ‘autonomous networks’ as the ‘endgame’ for telcos. So why has it taken so long and why has it been so difficult to realize the benefits of autonomous networks for Telecommunications?

According to Steve Saunders (founder of Light Reading), automation of processes requires the coming together of multiple ingredients (virtualization, AI, machine learning, telemetry, robotics, analytics, et al), and quite simply, only now, with the maturity of all these elements can the dream of autonomous networks or ‘zero touch automation’ be realized. The other big barrier to reaching the holy grail of network automation is the fact that it is complicated. Autonomous networks consider not only the infrastructure management but also the services, applications, underlying processes, and applied intelligence to carefully balance the decisions involved to reach desired business intents and outcomes.

Let’s not forget however that the fuel, feeding the automation machine is ultimately the data acquired from the various element sources. The richer the data the greater the intelligence and the insight, garnered from it. And with 5G, operators can expect lots of it. Data volume increases in the order of 7 to 10 times current volumes are expected, together with an increase of 100 times the number of connected entities (in the form of VPN, enterprises, and edge networks), and 1000 times the number of actual devices. This is all great news for the data-hungry AI beasts, however, one of the key challenges operators face continues to be access to relevant data promptly – thus impeding the ability to act on that data on time.

Analysys Mason, 2022 observed that “CSPs’ AI implementations are typically executed in silos; they address specific departmental issues and use data that is only accessible within these domains. As such, these implementations are affected by limited data access”. The analysts maintain that the power of AI is maximized when tools are used to analyze disparate, cross-departmental data sets to yield meaningful insights. Siloed data, therefore, makes it challenging to make the most of AI.

According to Analysys Mason, storing data in silos also increases the time taken to generate insights. For applications that have fed off mediation platforms and data storage systems built up over 20 years or more, removing the silos is not a trivial task. The data preparation, crucial to AI modeling, generally currently resides in multiple formats and must be processed and categorized consistently across the organization.

The best way forward is to consolidate the data into a unified data platform, so that it can be accessed and processed by a unified data model. The AI-powered Network Data Analytics Function (NWDAF) and the RAN Intelligent Controller (RIC) must be run as platforms supported by a unified data environment with core, access (fixed and mobile), and transport data for 5G, as well as data from other legacy environments. 5G networks will continue to co-exist with legacy networks for a long time, so management and monetization should be driven by a consolidated data environment upon which robust AI models can be developed.

Arriving at the holy grail of zero-touch automation is a lofty goal. Saunders highlights the ambition of the task, when he explains the challenge from the viewpoint of the operators, stating, “most of whom are still operating in super manual mode, using thousands of human admins and operators, this stuff is pretty much pie in the sky right now.” With the motivation to consolidate the data ingested, perform aggregation, unify the data formats, and enable fundamental data processing and AI preparation can operators begin this quest in earnest. There are no shortcuts to achieving this. Staying true to this course, operators can begin to eat the elephant- piece by piece.