AI in next-gen mobile networks

Eighteen years ago, on 07 July 2005, I was interning at an investment bank over the summer. I needed to earn some money to help fund my third year at University, and I wanted to work out whether banking is what I wanted to do when I graduated. My partner was doing the same, but at a different bank.?On that morning, we both got onto the London Underground (a.k.a. "the Tube") like we did every other day and travelled a few stops to Canary Wharf where we worked.?Everything seemed fairly normal, perhaps a tiny bit quieter.?When I got to my desk, I noticed one of my fellow interns, who had quickly become my “work bestie”, had not yet arrived.?When she still hadn’t arrived after an hour, I assumed she was unwell or running late.?But then our manager came over and asked if I had heard from her and seemed a bit worried.?I assumed he was just worried she may be unwell too.

Separately, I received an email from my partner’s father asking if we were both okay as he couldn't reach either of us via phone.?I had no idea what was going on, why people seemed so worried, or why he was trying to contact us.?

I used my flip-phone to send SMS messages to my friend and my partner, but didn’t receive any replies.?I checked that my phone’s memory wasn’t full so that I could actually receive texts.?(Some of you may have no idea what I'm talking about here!) I tried to call them, but couldn’t get through.?I honestly didn’t think much of it – she was probably on the Tube where there was no mobile reception, my partner was probably busy working, and his father was probably just checking-in.

Then my manager made an announcement to the whole team about what was going on…

It was a terrifying and tragic day.?We were told to go home and not come in the next day because Canary Wharf was a possible target too.

My friend was safe – she had actually walked quite a long way from her home to Canary Wharf because all the Tube stations near her apartment had been mysteriously closed that morning. My partner was safe. But for the longest time, I didn’t know that.?Nobody knew anything.?The news reports were patchy, and you had to have access to a television or radio to hear any news, or be at a computer to access a news website.

We assumed that the mobile phone issue was just like at New Year’s, when the network used to become overloaded by everyone trying to call and text at the same time.?Later, it was revealed that the mobile networks in a certain part of London had actually been shut down to the general public so that the emergency services were prioritised. ?One mobile network operator estimates that several hundred thousand, possibly even more than a million attempted calls by members of the public were lost that day.?Today, it’s difficult to imagine being so disconnected, from each other and from what's going on around us.?

So, in this month’s brAIn blog, I’m taking a look at a few of the many ways that AI is being used in mobile communication networks. These technologies enable us to be much more connected to each other, and enable us to find out breaking news so much faster, whether via official sources, social media, or via communication apps.

Example 1 – EP3788814A1 (Samsung)

This patent application relates to a method an apparatus for machine learning based wide beam optimisation in cellular networks.?5G networks are implemented using higher frequency bands (millimetre wavelengths or mmWaves), which enable higher data rates.?There are multiple demands on 5G networks.?For example, it is necessary to decrease propagation loss of the radio waves and increase the transmission distance.?Furthermore, the networks need to cope with the high bandwidth demands of users, which arise because we often now make video calls rather than audio calls, and because we use multiple apps on our mobile phones at the same time (such as to stream music or video, while also checking emails or online shopping).?To ensure the networks are able to keep us connected, a variety of techniques such as beamforming are required.

Beamforming involves using multiple radiating elements that transmit the same signal at an identical wavelength and pulse.?These are combined to create a single antenna that has a longer and more targeted stream (using constructive interference).?The more radiating elements that are used to form the antenna, the narrower and more focussed the beam. Once beams are formed, it is necessary to decide which beam a base station is going to use to communicate with a specific user equipment (i.e. a smartphone or any computing device that can communicate using the mobile/cellular network).

The claims of this patent application relate to, among other things, a central controller in a wireless communication system which selects, from a set of beams in a candidate beam pool, a first beam for each base station. The central controller then instruct the base stations to transmit signals to user equipment using the selected first beam.?The central controller may use a deep neural network to aid the selection of the best beam for each base station.? The best beam may vary depending on the required quality of service, number of connected user equipments, and so on. Using AI enables the best beam to be selected dynamically for potentially rapidly changing scenarios.

Example 2 – EP3963745A1 (Qualcomm)

This patent application relates to beam management using adaptive learning.?Typically, each base station is able to simultaneously support communication for multiple user equipment.?Beam management is the process to form beams, control beams and detect beams. In 4G systems, user equipment periodically monitors the radio link to determine channel quality. This allows the user equipment to report a fault if the link is below an acceptable link quality. However, in 5G systems, hybrid beamforming is used at both the base station and the user equipment for the reasons mentioned above in example 1. It is necessary to enable precise alignment of beams and to identify the optimal beams at any given time. Beam management procedures are used both when a user equipment is idle and not actively transmitting data, and when a user equipment is active and data exchange is taking place. The user equipment may also be moving within a cell of the network when it is active, which means that the beam alignment/direction will need to adapt.

The claims of this patent application relate to, among other things, a method for wireless communication by a node which involves determining one or more beams to utilise for a beam management procedure and then using those determined beams in the procedure.?An adaptive learning algorithm is used to enable the determining.?The algorithm may be trained using information obtained from user equipment deployed in simulated communication networks, or historical information from user equipment deployed in real communication networks, or from elsewhere in the network.

Example 3 – EP3956992A1 (Nokia)

This patent application relates to beam prediction for wireless networks.?A base station may configure one or more parameters of the beams used to communicate with user equipment based on feedback or reports received from the user equipment.?It is desirable to improve the performance of beam switching decisions by a base station so that the base station can more consistently select an appropriate or best beam prior to any connection or performance drop.?Relying on the reports from the user equipment may be difficult in rapidly changing environments such as cities, where user equipment moves around a lot.

The claims of this patent application relate to, among other things, determining a past beam sequence for a user equipment from a base station, and predicting, using this past beam sequence and a beam sequence model, a future beam sequence.?The predicted future beam sequence can then be used by the base station to perform a variety of beam-related actions.?For example, the base station may pre-emptively switch to a next beam or may handover to another target base station.

Example 4 - EP4156742A1 (Intel)

This patent application relates to a system for quarantining and recovery of a network after an outage or in advance of a potential outage.?Outages could be innocent and caused by equipment failure or extreme weather conditions, or they could be malicious and caused by a denial of service (DoS) attack or via a virus.?It is necessary to determine what corrective action to take in real time so that the negative effects can be mitigated.?

The claims of this patent application relate to, among other things, a network apparatus which is able to track a predetermined number of events related to access of a predetermined network at a network element.?An AI model is used to process sensor data and determine whether a Fault-Attack-Failure-Outage (FAFO) event has occurred at the network element which has resulted in or may result in failure of the network event, and determine?the type of attack.?The network apparatus is then able to switch, based on the determined type of attack, to using a quarantine physical network function or quarantine virtual network function instead.?These quarantine network functions are entirely isolated from other network functions, and as they contain the functionality affected by the FAFO event, enable rapid restoration of the affected parts of the network and provide services to any affected user equipment.???

Key Take-Aways

It is really cool to see how AI is being integrated into next-gen mobile communication networks, and how it can be used to ensure we remain connected and have a good connection even when we're on the move. Faults or attacks can be detected and mitigated more quickly, which again ensures we can stay connected.

As always, if you want to know more or want to know whether your software or AI invention - in any area, not just telecoms - could be protected using patents, please contact me via email ([email protected]) and I would be happy to have a confidential discussion with you!