Classification and Root-Cause Analyzing Tool
CARAT is focusing on mobile communication networks starting from 2G up until 5G. The analysis is performed for speech and data transactions, so that a thorough result of recorded issues is reported. The analysis includes incomplete transactions (failed or dropped) as well as successful ones. It covers correctness, completeness, and quality for the events.
Based on our expertise in wireless communication networks we identified patterns in traces that identify problems or failures of measurement equipment or measured network. This offers an in-depth analysis of the recorded behavior in the traces and allows the user to work on efficiently on solving identified issues.
With CARAT’s analysis, many different patterns are detected and assigned to the following classes.
AI / ML in CARAT
The integration of more advanced machine learning in CARAT is done to improve the ability to classify unknown patterns and improve the classification accuracy. This is achieved using a combination of supervised and unsupervised ML Algorithms, enabling us to get the best of both worlds. While with the unsupervised ML Algorithms our aim is to unveil the hidden patterns among the complex datasets, the supervised part ensures the validity of our results.
Data sources considered in CARAT
- Drive/Walk-Test-Data by Swissqual, Infovista, Focus Infocom …
- Crowd-Data (UE-based measurements)
- Protocol-Traces (PCAP) from Core and RAN
- Applications-Test Data (Voice, Interactivity, Performance, Capacity, …)