Today, a single smartphone generates large data traffic daily – and an amazing amount of data is transported on a whole mobile network – more than 10 Tbit/s. The mobile network operators should not leave the huge and continuously moving data assets unused, as they can not only get valuable information about the operation of the network from those, but they also remedy or prevent any problems, using those data.
The Ericsson Expert Analytics (EEA) software family, based on Machine Learning, has been developed for that purpose. It analyses the real-time data of more than 100 million subscribers with operators located at different sites all over the world at the same; allowing the acquisition of key conclusions or insights regarding the operation, maintenance, and development of the network. This is done without the analysis of the content of the user data; the system has no such sight of the content accessed on the clients.
Of course, this real-time operation could not be realized without the use of state-of-the-art data mining and cloud-based technologies, 90% of which are designed and developed in Hungary, in the Ericsson building at the Danube bank. Thus, the EEA is a real “made in Hungary” product, which grew from a research project launched at Ericsson in Hungary in 2008, to a globally available service, used all over the world today. The majority of the developments is carried out in Hungary also today; the company working on the fourth generation of EEA at the moment, which will relocate the whole service to a cloud-based platform, terminating bare metal solutions – and all that is of course performed focusing on machine learning.
Repair, even without human intervention
The service is essentially the immune system of the mobile network; accordingly, it can not only make recommendations for the repairing of detected errors as needed, but can independently intervene and remedy problems. Although this functionality is already used by some service providers, it is still in an early stage: most locations have points now where the system passes over the reviewing and repair tasks of problems to human operators.
After an appropriate amount of testing, human intervention can be excluded, and the “immune system” can work individually. The algorithms of EEA can learn how the professionals of a given operator manage certain anomalies, problems – different companies use different practices for similar cases – and will later individually remedy problems in accordance with the samples learned from the company’s procedures. Of course, Ericsson regularly re-measures the automatized repairs, to ensure that the operation will be smooth.
In the long run, the industry will probably go this way, and network troubleshooting will become fully automatized.
No errors will slip through the net
The service is modular, so the mobile service providers can select the insights they need. The analytics solution uses a number of technologies from very simple checking mechanisms to complex machine learning algorithms to analyze the network, predict network failures, search for anomalies, and identify samples of the behavior or use of devices. The system tracks in real time the activities of all users and components of the given mobile network, and can tell what problems have been faced by them, and even what problems may occur later, quasi predicting future events on the basis of the past samples.
The solution continuously tracks network errors, whether services like VoLTE, VoWiFi; but 5G and OTT data traffic is also monitored at all times. Expert Analytics promptly notifies the service provider if: a call is disconnected, any cases of call fails, poor sound quality, and even the decrease of quality of streamed videos; and it can detect the cases where a user cannot access a web site. It summarizes the collected data and makes remedy recommendations to the service provider, describing in detail where and how many subscribers are affected, how serious the problem is, and whether a similar problem has occurred before or can be expected in the future.
The analysis of the above-mentioned OTT data traffic, such as the quality of the streamed videos or the video calls, the number of which increases due to the epidemic, poses significant challenge to the developers. While quality fluctuations of IMS-based calls can be relatively easily and accurately detected, in most cases it is not that easy in case of IP-based data traffic. That is because the companies behind the various services usually encrypt their data traffic, for which the network provider only provides the channel and cannot look into the content. The EEA tries to solve this problem through various heuristic algorithms, which allow the detection of samples indicating the deterioration of service quality within the encrypted data traffic.
In this regard, actually there is a race between the external content providers and the mobile service providers: the former ones try to hide the data which are valuable also to themselves, and the latter ones try to extract from them the signals referring to service quality – even though both parties are interested in the improvement of user experiences.
From a social point of view, this technology can be used for very useful purposes: in some countries, the service providers used the EEA-based mobility data of the population also for the creation of models for the analysis of the proliferation of the Coronavirus.
One step ahead of problems
Expert Analytics also offers predictive algorithms to customers, which can predict problems expected within a few days – and these are useful not only for the prevention of failures, but also deliver valuable information in case an expected problem does not occur. These machine learning algorithms continuously analyze the operation of the network, thus making a profile of its everyday operation, including errors occurring. If they later detect behavior samples that caused problems before, they will notify the service provider’s professionals; however, in the future they may also remedy those problems before the deterioration of the service quality would affect a wider range of subscribers. The algorithms are created, tested, and integrated into the software environment by a team of technological professionals, mathematicians, data scientists, and experienced software developers.
Meanwhile, the extensive analytics does not compromise sensitive user data; analysis is fully anonym; the product of Ericsson is wholly managed by the service providers and is run on their own systems; and it does not store any personal data or any data that can identify any users, it operates with encrypted identifiers not referring to any individual user.
Ericsson recommends EEA not only for the operation of existing networks but also for the building of new technologies and systems; and the service is very relevant now, at the time of the expansion of 5G networks. Proper analysis of the data traffic and speed of new generation networks cannot be solved with the traditional technologies, but EEA is prepared for their extensive, real-time monitoring, both regarding commissioning and user satisfaction.