Sean Kennedy, Data and AI Lab Leader at Nokia Bell Labs, tells us how telcos are being transformed with Artificial Intelligence.
AI is helping telcos effectively deal with complexity from data and solve problems very efficiently – at the millisecond speed you need. It is changing the way software is written; building up the necessary data pipelines is changing software developers’ fundamental approach. Rather than building fixed algorithms and code bases, we’re building models that will learn from the data.
Here are some key use cases for the use of AI for telcos:
AI for predictive hardware maintenance
A key benefit of AI for telcos is predictive hardware maintenance. By following the paradigm that webscalers are already actively using – collecting lots of data and looking for patterns, AI is able to predict the failure of hardware in the future. We often find warning signals as far as 14 days ahead of failure with high confidence. Currently, implementation exists at small scale and we have field trials with major customers in progress.
AI for Self-Organizing Networks
5G networks are becoming increasingly complex. There are more parameters to tune than ever before, there’s increased frequency range, an increased number of users to schedule, as well as users that can be scheduled simultaneously. AI machinery is excellent at driving these automations and making them more efficient.
At Nokia, we are currently working on a project which uses AI for Self-Organizing Networks (SON). Modern wireless networks have many parameters including for things like power control and energy savings etc. It is extremely difficult for humans to tune these parameters, let alone continuously tune them as your network conditions change, as there is dependency between them all.
AI helps you look at the data and performance values as you’re training information and optimize these processes over time automatically using principal techniques, including from mathematics and statistics.
In other words, AI can autonomously learn SON parameters by watching network traffic and radio conversations, using complex mathematical methods such as Bayesian optimization and Markov decision processes. AI outperforms humans at these tasks. Nokia is firmly committed to evolving towards a sustainable world and tools like these will be critical for this evolution.
AI for 5G packet scheduling
Packet scheduling becomes increasingly complex as we move to 5G due to an increase in frequency range, the number of users to schedule, the number of users that can be scheduled simultaneously and the number of ‘wireless beams’ used for transmission.
Moreover, decisions on how to do this packet scheduling has to happen really fast – in the sub millisecond range – to be useful. From a mathematical point of view, it is too hard to find the optimal set of transmissions in that time. By going through the data, AI can help us learn the best set of transmissions for a given set of wireless conditions.
AI for Mechanized Inventory
At Nokia we’re using AI imaging to determine what is on a job site before technicians arrive to roll out a new network. By using images of equipment and training deep nets to identify different objects, AI removes the strenuous (and often impossible) task of keeping inventory up-to-date.
Our research shows a high level of accuracy is possible with deep supervised learning and additional algorithms to automatically adjust images based on differing distances, heights and angles at which images are taken from the same set, be these the back of a rack, a cell tower or other enclosures containing telecom equipment.
AI is already fundamentally changing many industries. For the telco sector, AI is one of the best tools for things like anomaly detection, network optimization and predictive maintenance. We’re seeing this across all areas of the stack – from the physical layer all the way up to applications that run on top of the network, even in the deployment of our networks.
There is no question that AI is driving improvements across the telco industry, and it will only become more important going forward.
