‘We are trying to drive the notion of QoE,’ says AT&T’s VP of network analytics and automation
“In 5G, our focus is the customer experience. We are trying to drive the notion of QoE,” says Raj Savoor, vice president of network analytics and automation for AT&T. “The quality of experience is in the heart of everything we do.”
That manifests itself in two ways, he continues, which are linked but have to be approached separately: Planning and design, and the network life cycle.
“We are making massive investments in spectrum, transport and backhaul capacity, and then deploying the Standalone core, the user plane functions and bringing them in closer proximity to the user; the edge cloud deployments. You want to be able to very efficiently plan that. It is not just a step-function of, hey, let’s increase the capacity,” Savoor says. “It’s, how does that design come together? So, a deliberate thing that we began a few years ago as part of our transformation is a digital twin strategy. The idea being, let’s build a construct of our network and think about it as the ‘N’ state that we in, and ‘N-plus-one’ where we want to be.”
He described that network digital twin as having three attributes: Parametric modeling and network simulation capabilities that enable the carrier to understand its parameters and deterministic and non-deterministic behaviors that might occur if it were to make changes; it is fueled by data from LTE and 5G Non-Standalone to reflect the network as it is, including regional differences, content provider interconnect points, where user plane functions reside as well as local traffic hot spots in various markets; and thirdly, it incorporates AT&T’s engineering rules for optimization, considering QoE add physical and logical resiliency (particularly important, given AT&T’s relationship as the network provider for FirstNet).
“We went down this path, and it’s one of those ‘the destination is the journey,” he says. “We’re continuing to learn and evolve from that, and it is the cornerstone and our first area of planning and design and build.”
When that network is planned, designed and built, then the next challenge is its life cycle: Meaning, Savoor says, “How do you continually optimize the network behavior … based on the dyamics?” The answer there, he says, is “fine-grain data collection.”
AT&T, he adds, is applying artificial intelligence and machine learning in both its planning and design as well as in its life cycle management of the network. In addition to its digital twin environment, he says the carrier is using AI for ray tracing to improve its propagation models for coverage and do so in three dimensions. “It’s literally in seconds, compared to what we were doing in days—it’s that fast, for some of the metros, and in much better precision.” He sees opportunity to expand the use of intelligent 3D planning outdoors as well as indoors and in large venue coverage.
Leading from that, then, because the carrier understands more characteristics of the network, “we are able to have feedback loops with fine-grain data,” Savoor says. “In life-cycle management, we’ve matured from those closed loops where we were making decisions with fairly static information, to now having much more finer grain data and applying a lot of the learnings around, ‘Hey, this configuration has a higher risk profile. This parameter change has a probability to impact X, Y, Z” much more effectively.”
In particular, he says he is bullish about the use of predictive analytics with ML in the network lifecycle aspect: Predictive KPIs that can give insights into where performance will be in the next five minutes, 15 minutes and so on—which is significant in a dynamic mobile network environment. AT&T is already using security/threat analytics and wants to expand that as well, he adds. So given the multiple networks that AT&T operates and the many layers of those networks, is there a magic bullet tool for optimizing? Not really. “We’ve had to be very realistic and practical that when we think of the Layer 2, Layer 3 network, managing our evolving metro Ethernet network and our backbone 400 gig interconnection network, [there are] very different modeling and simulation tools there,” Savoor says. “Then you take the access network, our fiber build and RAN – very different simulation tools. The real challenge and opportunity is … mapping the data set and the problem, to the right tool. And that’s true with even selecting the right AI and ML model for the right problem.” He brings up the cliche that when you have a hammer, everything looks like a nail – and trying mindfully to avoid that. “We are trying to be very careful about understanding the problem, seeing if we can bootstrap something we already have, but at the same time, looking for a better mousetrap,” Savoor says. “There’s partners, we have our ecosystem, but also, a lot of homegrown innovation as well.”
