[Paper Review] Real-time Localization Using Radio Maps

Çağkan Yapar, Ron Levie+2 more|arXiv (Cornell University)|Jun 9, 2020

Indoor and Outdoor Localization Technologies20 references4 citations

TL;DR

This paper proposes a real-time, radio map-based localization method for urban cellular networks using received signal strength (RSS) measurements from base stations. By leveraging RadioUNet—a deep learning-based pathloss simulator—it constructs accurate radio maps to estimate user equipment (UE) location via signal strength contour summation, achieving sub-2-meter accuracy even in complex urban environments with dynamic obstacles like cars.

ABSTRACT

This paper deals with the problem of localization in a cellular network in a dense urban scenario. Global Navigation Satellite System typically performs poorly in urban environments when there is no line-of-sight between the devices and the satellites, and thus alternative localization methods are often required. We present a simple yet effective method for localization based on pathloss. In our approach, the user to be localized reports the received signal strength from a set of base stations with known locations. For each base station we have a good approximation of the pathloss at each location in the map, provided by RadioUNet, an efficient deep learning-based simulator of pathloss functions in urban environment, akin to ray-tracing. Using the approximations of the pathloss functions of all base stations and the reported signal strengths, we are able to extract a very accurate approximation of the location of the user.

Motivation & Objective

• Address the poor performance of GNSS in urban environments due to non-line-of-sight conditions.
• Overcome limitations of traditional RSS-based localization that oversimplify urban radio propagation.
• Enable real-time, accurate localization using only standard RSS measurements from base stations.
• Develop a scalable, computationally efficient method that avoids expensive physical ray-tracing simulations.
• Integrate learned pathloss models into a robust localization framework adaptable to real-world urban dynamics.

Proposed method

• Use RadioUNet C to generate pathloss maps from city geometry and base station locations without ground truth measurements.
• Use RadioUNet S to refine pathloss maps by incorporating sparse ground truth measurements, improving accuracy in complex environments.
• Construct radio maps for each base station using the output of RadioUNet, representing pathloss as a function of spatial position.
• Estimate UE location by summing level sets of received signal strength across multiple base stations, enhancing localization stability.
• Apply a voting-based algorithm that combines multiple estimates with varying noise tolerance (epsilon) to improve robustness.
• Optimize hyperparameters (epsilon and standard deviation) using training data to minimize localization error on unseen test maps.

Experimental results

Research questions

• RQ1 Can a deep learning-based radio map simulator (RadioUNet) replace computationally expensive ray-tracing for urban pathloss modeling?
• RQ2 How accurately can UE localization be achieved using only RSS measurements and precomputed radio maps in dense urban scenarios?
• RQ3 To what extent does incorporating real-world measurements into RadioUNet S improve localization accuracy compared to purely simulated models?
• RQ4 How does the proposed method perform under dynamic urban conditions, such as the presence of moving vehicles?
• RQ5 What is the optimal trade-off between localization accuracy and computational efficiency in real-time deployment?

Key findings

• The method achieved a localization error of 1.3 meters on the DPM dataset using RadioUNet C and a sum of level sets.
• With RadioUNet S trained on simulations including cars, the method achieved a localization error of 1.6 meters, demonstrating robustness to dynamic obstacles.
• On high-accuracy IRT4 simulations, the method achieved a localization error of just 2.24 meters, with zero standard deviation (Q=0 m), indicating high consistency.
• The method outperformed coarse-simulation-based models when tested on IRT2 simulations, showing the benefit of adaptation to higher-fidelity data.
• The second algorithm, which selects epsilon based on optimal standard deviation, achieved better accuracy than the fixed-epsilon approach, especially with increasing numbers of base stations.
• The method remains effective even with limited ground truth data, as RadioUNet S adapts well to real-world conditions with minimal calibration.