Anomaly Detection

How do we detect anomalies in the telecom base stations

We have already presented what we do for O2SK in our previous blog, ( Tracking functions of Base stations ). We at LibertyAces figured out how to monitor functions and track irregularities of the transmitters without having to set thresholds for every one of them. 

So how do we actually do it?

We were tasked to track the utilization of base stations, which means establishing ERL metrics that state how many hours of calls have gone through the base station in a fifteen-minute time window. 

The model compares the ERL to the normal behaviour of the site, which is stored in a model calculated from the data gathered in last 30 days. 

These two graphs show outputs from two base stations. The blue, jagged line depicts the ERL throughout time. The yellow, comparatively flatter line expresses the normality of the signal. 

The higher the yellow line is, the more it indicates that the base station is performing as expected.

Whenever the yellow line approaches zero, it indicates increasing irregularities. 

As an illustration of how this monitoring functions, on the 25th of August, one of these base stations flatlined and failed to transmit calls until midday. At the same time, the second base station had to pick up the work and double the usual amount of transmitted calls. 

This is a textbook anomaly and the yellow line shows just that. The First base station has reached anomaly score values close to 0. Consequently, the second base station is detected doing the work of two at once, transmitting twice the amount of calls than the norm would be, and therefore reaching the mid-way mark between the usual Anomaly score and zero. 

Fortunately, we at LibertyAces track these anomalies and get alerted whenever a base station exceeds the thresholds. This particular case was resolved in a matter of a couple of hours. Had it not been for our expertise and diligent monitoring, O2SK might not have detected this severe problem until after the second base station overheated.