Wireless backhaul, consisting of point-to-point microwave links, has been the backbone of modern mobile telecom networks from 2G to 5G. Furthermore, it plays a key role in the ongoing transformation to a digital society.
Global communications service providers (CSPs) control thousands of links in complex configurations with potentially high operational costs. Network operations centers (NOCs) are often centralized into one or two locations in or outside a country and have the ability to setup alarms to indicate when the network is not performing as expected or to monitor individual links. However, due to the large size of networks, and the large number of possible alarms, NOCs are often overloaded with information. It is not uncommon that networks trigger thousands of alarms per minute and this makes it an almost impossible task for a NOC to manually analyze and prioritize which should be acted upon first.
Due to the vast amount of data produced, a root cause analysis could take even a skilled expert hours or days to perform. This leads to increased OPEX, reduced quality of experience for the end-user, and a tie up of resources.
Therefore, the NOC often needs to send out expensive service teams to sites to replace the hardware – without knowing the exact cause of the problem. This often results in costly unnecessary site visits and the replacement of radio hardware with perhaps no fault found.
A single trouble report is estimated to cost ~10,000€ and a day’s cost for a microwave expert is 1,000-2,000€. Thus, reducing the number of trouble reports and limiting the needed manual work for a skilled expert is a highly efficient way to reduce OPEX for CSPs.
On the other hand, if presented with a limited and selected dataset covering a short time period, it is often a quick and easy task for a skilled expert to come up with a high-probability root cause for a troubled microwave link. Going through all data to find the correct small dataset and to do this continuously in networks with tens of thousands of links is an undesirable task.
This is an area where AI can step in and do the tedious work for us. We can train algorithms to do both the quick and easy tasks that experts carry out, as well as continuously browse vast amounts of data looking for root causes. This would be an excellent use case for AI to help CSPs manage their network infrastructure in a more time-efficient way with lower OPEX.
Research into training algorithms for NOCs started back in 2015 and transitioned three years later into a collaboration with two large European CSPs. For the first time ever, Ericsson introduced real-time AI based on convolutional neural network algorithms trained to recognize events in harvested data from microwave links.
The algorithms were introduced into the two CSPs’ networks to monitor how the received signal for each individual microwave link in their network varies over time and provide an estimate of the root cause of that signal variation. A signal fade could be the result of rain, line of sight blockage due to an obstacle, reflections in the environment, hardware failure, and so on. The system provides a detailed real time analysis of the potential root causes of all signal fades in the network with a 10-second time resolution. Trained and verified against a database with tens of thousands of manually verified fading events, the algorithms is more than 98% accurate for the supported events. This is on par with (or better than) the accuracy reached by human microwave experts in a fraction of time.
Presenting this information to NOCs in combination with the information from alarms gives network experts the possibility to filter out events that need to be prioritized, as well as events that require dedicated site visits. For example, different actions need to be taken if an alarm is caused by hardware failure (such as a loose connector, unstable mast, or degraded radio electronics) than if the signals is blocked by a tree during windy time periods or intense rain.
The system has been running for more than three years with more than 10 million operational hours analyzed by the AI algorithms. The successful results confirm that it is possible to automatically identify links with degraded availability and propose the root cause. This is something that has previously not been possible due to the enormous amount of manual work required.
Ericsson released the AI analytics tool Advanced Microwave Insight in spring 2021, making the AI algorithms available to CSPs everywhere.
Related content
Explore our 5G Transport offerings
