Lightweight Detection of Tariff Limits in Cellular Mobile Networks

Abstract

Tariff limits in place are one of the major challenges for crowdsourced benchmarks, blocking us from determining the network performance. We propose a new, lightweight method for traffic shaping detection and compare it to an already existing algorithm that relies on the detection of a level shift in the achieved throughput.

The level shift detection is suitable for online processing during a throughput measurement and allows for an estimation of additional parameters characterizing the traffic shaper. Our method has a lower computational complexity that offers us the opportunity to analyze large datasets. Another advantage of our method is that it requires only two parameters - threshold and bin size, whereas the level shift detection needs four tunable parameters.

The level shift detection may fail in mobile networks where the throughput exhibits higher fluctuations. Based on reference LTE measurements, which we have made publicly available, we show that our method achieves better performance in terms of detection rates and false alarm rates. We propose a modification of the level shift detection algorithm to improve its performance.

Finally, our method is validated in the context of mobile operator benchmarking. We apply the method to a large crowdsourced dataset containing approximately 150 000 crowdsourced throughput measurements in the LTE networks of three major Austrian operators. After removing the tariff limited tests, the operator ranking changes dramatically and better reflects the quality of the network rather than the share of tariff limited users.

Tariff limits in place are one of the major challenges for crowdsourced benchmarks, blocking us from determining the network performance. We propose a new, lightweight method for traffic shaping detection and compare it to an already existing algorithm that relies on the detection of a level shift in the achieved throughput.

The level shift detection is suitable for online processing during a throughput measurement and allows for an estimation of additional parameters characterizing the traffic shaper. Our method has a lower computational complexity that offers us the opportunity to analyze large datasets. Another advantage of our method is that it requires only two parameters - threshold and bin size, whereas the level shift detection needs four tunable parameters.

The level shift detection may fail in mobile networks where the throughput exhibits higher fluctuations. Based on reference LTE measurements, which we have made publicly available, we show that our method achieves better performance in terms of detection rates and false alarm rates. We propose a modification of the level shift detection algorithm to improve its performance.

Finally, our method is validated in the context of mobile operator benchmarking. We apply the method to a large crowdsourced dataset containing approximately 150 000 crowdsourced throughput measurements in the LTE networks of three major Austrian operators. After removing the tariff limited tests, the operator ranking changes dramatically and better reflects the quality of the network rather than the share of tariff limited users.