AT&T turns up AI for drones, load balancing, 5G build out

As part of its close embrace network artificial intelligence (AI), AT&T is testing drones in New Jersey to inspect cell sites, and researching the use AI to define network policies.

AT&T's Mazin Gilbert outlined his company's blueprint for artificial intelligence and machine learning on Tuesday during a keynote address at the TM Forum Action Week conference.

While AT&T has successfully embraced virtualization of its network using software-defined network (SDN) and network function virtualization (NFV)—it's on track to reach its goal of 75% virtualization of the core network by next year—Gilbert said that's not enough in a world of 5G and IoT and increasing consumption of bandwidth.  

"The network can't be just software," said Gilbert, vice president, advanced technology systems, AT&T Labs. "The network needs to be autonomous and pretty much zero touch. It needs intelligence to know when it repairs itself, when it secures itself. The network needs to be contextual, personalized."

RELATED: AT&T travels the open-source road to help build an industry ecosystem

Working in tandem with open source groups such as ONAP, Acumos, Akraino and ORAN, AT&T is building its network AI platform to scale from the core to the edge. It's also deploying white boxes with its DANOS operating system as it works towards disaggregation.

AI is the key ingredient for implementing numerous projects and platforms for AT&T. Gilbert said AT&T is putting more intelligence into its mobile edge compute (MEC) at the customer edge and into its radio access network (RAN.) It's using AI to manage its third-party cloud arrangements, such as Microsoft, and in its internal cloud and hybrid clouds.

"We're rethinking the RAN completely," he said. "We're pushing some of that intelligence to the RAN, where it's needed, all the way to the data centers, national, local, regional data centers."

AT&T is using AI and machine learning (ML) to build its 5G infrastructure to map its cell towers, fiber lines, and other transmitters that exist today, and to pinpoint the best location for 5G build outs in the future.

Gilbert said that AT&T has more than 75,000 macro cells in its network and plans to build hundreds of thousands of small cells as well adding picocells, all of which will be guided by AI, AT&T's internal network data and third-party data.

"We have fiber now that passes over 12 million, 13 million businesses," Gilbert said. "And, you can imagine that figuring out where to put the fiber, where to put the cell, a small cell, or picocell as well, that's not an easy exercise. So, we use data and machine learning to take into account our current infrastructure, our competitor's infrastructure.

"We take into account the local policies for the township that allow us to build. We identify utility poles in the country all automatically using machine learning. We identify the type of pole, the material, and the height of the pole all automatically using the data with satellite images."

If AI detects a cell site isn't functioning properly, it will signal another tower to pick up the slack. If one area is experiencing a high volume of usage, AI will trigger lower-use cell sites to ensure that speed isn't compromised.

A "perfect marriage" of AI, ML and SDN will help enable the speeds and low latency of 5G, according to Gilbert.

"I'm really excited about this because really it's the data behind this infrastructure, behind this transformation, is what's really going to open up a lot of opportunities," Gilbert said.

Gilbert said AT&T is testing the use of flying drones in New Jersey. The drones fly around cell sites and AT&T's premises testing and validating the RF signals, and testing whether the cell itself is functional.

"Is there rust? Are the wires connected? Is there dirt, which is one of the primary reasons why some of these cells malfunction, etc.," Gilbert said "We're doing that with purely drones where the intelligence is sitting at the edge. The drones simply have video capabilities to take video in real time. We're able to diagnose whether humans really are required to go and climb the radio cell site."

AT&T is also researching using AI to define policies that are currently set by systems and employees. AI and the data analytics tell AT&T if if any of the policies have conflicts prior to defining them.

AT&T is using AI to load balance traffic, such as video, on its network and use machine learning to detect congestion on small cells on 5G networks before it happens. By putting intelligence closer to the edge, AT&T is starting to load balance traffic across these small cells and move traffic around when needed.

AI, data and open source, mainly ONAP, are used to for operation automation in AT&T's networks.

"Clearly, our biggest cost is in planning, design, installing the network, and also maintaining the network and securing the network," Gilbert said. "So, we have built these templates of intelligent agents. These are nothing more than closed-loop systems. Closed loop systems that capture data that can be configured for different problems. We push those in our network to collect data."

AI will also help enable SLAs from networking slicing offerings to AT&T's customers. AI is being pushed into customer premises, such as IoT sensors in a factory. AI can remove malfunctioning IoT sensor in milliseconds, according to Gilbert.

While AI is playing a key role in the core network and the RAN, Gilbert said "There's a lot of intelligence that needs to happen closer and closer to the edge."