Applying AI and Machine Learning
to these 5G Challenges

Machine intelligence is needed for 5G networks to be self-optimizing and self-healing. Dense networks of small cells with advanced antenna technologies need to be continuously tuned based on the devices connected, application data flow, bandwidth utilization and RF signal analysis. Fronthaul and backhaul capacity needs to be managed dynamically based on varying bandwidth demand and network utilization. Network failures can be remediated automatically, applying machine intelligence to detect and isolate the root cause and then determine the appropriate fix or workaround, which may involve reallocating network resources or shifting network demand onto other resources in the underlying infrastructure.

Failure prediction for individual hardware or software components is a good example of an application of machine intelligence within a silo. Recently, I was involved in a trial which focused on prediction of failure in RAN networks. The trial showed that ML could identify the 1.2 percent of network RAN elements responsible for 75 percent of outages and within this group correctly predict 334 out of 337 actual incidents. Further, 50 percent of these incidents could be identified three hours or more in advance, which would be sufficient time to prevent them, most especially as approximately 74 percent of failures could be addressed remotely without a truck roll. 

AI Across Multiple Domains

AI will play a critical role in enabling operations teams to rapidly detect, isolate and remediate problems that are manifested by events triggered across multiple 5G domains. AI algorithms enable you to take insights gleaned from individual operational silos and correlate those data points across multiple domains. 

For example, a group of subscribers report poor application performance while an alert indicates the backhaul network is experiencing a bottleneck, but the root cause is actually a malfunction in a small cell that is flooding the network with a stream of bad packets. Without machine intelligence, different operations teams might investigate the same incident by accessing multiple dashboards and examining various log files to determine what is happening within their silo. Only after sharing information between teams, investigating further and ruling out possible root causes will they arrive at a definitive explanation. The intent of applying machine intelligence is to remove human operators from this process as much as possible, leveraging operations tools that use ML and AI to perform the required data analysis and correlation of insights across multiple domains. The goal is not only to have machines figure out what is happening but to automatically take corrective action. While some operators may deem this to be too risky, for many scenarios in 5G networks this will be a necessity. Services for ultra-reliable machine-to-machine communication will depend on the ability of machines to fix machines.

Today, diagnosis of the root cause of a complex incident is the province of small teams of expensive experts. Machine intelligence multiplies the effectiveness of these teams of skilled people and allows a larger proportion of complex incidents to be handled by less-skilled teams augmented by machines. For example, augmenting traditional methods with ML enabled the same team mentioned above to eliminate 98 percent of previous alarm noise, reduce the average time to resolve an incident by almost an hour, and increase the number of incidents found per hour by 40 percent.

Key Usage Scenarios

5G Roaming

The advent of 5G places additional constraints on mobile operators when selecting a visited network. A smartphone user who frequently streams 4K video will have high expectations for the performance of an enhanced mobile broadband service, whether delivered on the subscriber’s home network or when roaming onto the network of another operator. Ultra-reliable, low-latency M2M communications will also have stringent QoS requirements that must be satisfied when a smart device roams onto a visited network.