Rail Twin: Scalable Estimation of Train Cabin RF Attenuation Using Crowdsourced Measurements

Abstract

Train cabins introduce significant radio frequency (RF) attenuation, degrading wireless connectivity for passengers and onboard systems. While drive test campaigns can quantify this effect, they are costly and unscalable for comprehensive, fleet-wide modeling. Crowdsourced datasets, such as those collected via 3GPP Minimization of Drive Test (MDT) mechanisms, offer broader coverage but lack ground truth labels, device context, or train configuration metadata, making direct attenuation estimation difficult. In this paper, we present Rail Twin, a scalable Digital Twin framework for estimating cabin RF attenuation from heterogeneous, passively collected mobile network measurements. Our method combines two MDT datasets: (i) modem measurements from train-mounted rooftop antennas, serving as consistent outdoor references, and (ii) user equipment (UE) measurements, which we classify as indoor or outdoor based on mobility and spatial continuity features. We estimate attenuation as the perfrequency difference between in-train and rooftop signal levels, aggregated by 100-meter track segments. We validate our approach against controlled reference measurements, achieving sub-7 dB mean absolute error (MAE) in mid-band frequencies for most segments. Rail Twin produces frequency-resolved, train-type-agnostic attenuation maps, enabling practical applications such as deployment planning, performance benchmarking, and anomaly detection, without requiring dedicated measurement campaigns or prior knowledge of the train fleet.