Crowdsensed Performance Benchmarking of Mobile Networks

In recent years, the boom of mobile-network-measurement apps has stimulated the growth of publicly available datasets, which contain up to billions of measurements conducted by anonymous end-users. Although the crowdsensing has proven its value in different areas, in the context of performance monitoring in cellular mobile networks, convincing applications are scarce. A portion of already published research focuses on modeling and predicting achievable throughput as a function of observed signal indicators. Due to the system complexity and a large number of possible predictors, the performance of these models is low since the throughput varies greatly even at a constant signal power level. In this paper, we introduce a new method for evaluating an empirical network-centric upper bound on the throughput-performance of different cellular mobile networks. Based on simulations and reference measurements, we propose a model function that characterizes throughput as a function of signal power. The critical point that increases the quality of fit when processing the crowdsourced measurements is the removal of the system-inherent high noise level by finding a method to exclude biased samples systematically. We apply our method to crowdsourced measurements provided by several national regulatory bodies, and benchmark the performance of LTE networks of different mobile operators in different countries.