Synthetic Data for Indoor Positioning Systems: Reducing Offline Collection Costs

Indoor Positioning Systems (IPS) using fingerprinting require extensive data collection, which is costly and time-consuming. This study proposes a method to generate synthetic fingerprint data, reducing reliance on real-world acquisition by simulating both noise-free and noisy signal conditions with Bluetooth Low Energy (BLE) Received Signal Strength Indicator (RSSI) values. To simulate real-world constraints, we apply data reductions to the Position-Annotated BLE RSSI Dataset and evaluate the impact of synthetic data under four training configurations: real-only (baseline), synthetic-only, synthetic with real-data finetuning, and a combined approach. The impact is assessed under varying levels of real-data availability to analyze trade-offs between data collection effort and positioning accuracy. Results show that synthetic data effectively reduces dependency on real data while maintaining or improving accuracy in low-data scenarios. Baseline performance remains stable at 2–3 m error when using the full dataset but degrades to 5–7 m when data is reduced below 27,187 samples. Introducing synthetic data mitigates this loss, reducing error by 0.77–1.14 m on average and up to 2.65 m in extreme cases. These findings position our method as a cost-effective and scalable alternative, minimizing offline data collection efforts while preserving positioning accuracy.