On the Physical Plausibility and Distribution Alignment for Sim-to-Real RF Positioning

2026-07-05

AI Summary

This work addresses the poor generalization of RF-based localization models to unseen urban streets by pretraining on large-scale synthetic data generated via ray-tracing simulation (using the Sionna platform) over the city of Rome. The study systematically investigates the impact of base station calibration, physical plausibility, dataset scale, and RSSI distribution on simulation-to-reality transfer. It finds that aligning RSSI distributions is more critical than physical fidelity or data volume, leading to an effective distribution normalization strategy. Experiments show that synthetic pretraining consistently improves localization accuracy on known streets, with city-scale unconstrained data yielding the best performance. Crucially, on previously unseen streets, only simulations with aligned RSSI distributions significantly enhance real-world localization accuracy.

Abstract

Reliable radio frequency (RF) positioning from cellular measurements is limited by the high cost and limited coverage of real drive-test data, especially when models must work on streets not seen during training. Previous work showed that ray tracing simulations can provide useful synthetic data for pretraining deep positioning models. In this paper, we focus on the simulation side and study how base-station calibration, physical realism, synthetic-data scale, and RSSI distribution alignment affect transfer to real data. Using a Sionna reconstruction of a Rome deployment, we calibrate each base station by adjusting its location, height, azimuth, and transmit power. We compare physically plausible calibrations with unconstrained ones that allow unrealistic base-station placements. We also compare deployment-specific synthetic data with much larger city-scale datasets. Although unconstrained calibration matches measured RSSI better, it does not always improve positioning accuracy. All synthetic pretraining approaches improve performance on known streets, with the best result obtained using city-scale unconstrained data. However, larger synthetic datasets alone do not improve performance on unseen streets. The best results on held-out streets are achieved only after normalizing simulated RSSI values to better match the real distribution. Overall, the results suggest that distribution alignment is more important than physical realism or dataset size for sim-to-real RF positioning.

Problem

• sim-to-real
• RF positioning
• distribution alignment
• physical plausibility
• synthetic data

Innovation

• distribution alignment
• sim-to-real transfer
• RF positioning
• ray tracing simulation
• base station calibration