Enhancing RF Fingerprinting for Indoor Positioning Systems Using Data Augmentation

Abstract

Indoor Positioning Systems (IPS) have recently emerged as a crucial technology in the Internet of Things (IoT), with widespread applications in smart cities and homes. Radio frequency-based fingerprinting, enabling location estimation through signal observations, requires manual surveys for constructing location maps. This process involves annotating radio signatures with corresponding locations, rendering it time-consuming and labor-intensive. To address this challenge, our paper proposes a data augmentation method that leverages a conditional generative adversarial network with LSTM and CNN. This approach effectively captures patterns in the training data, generating synthetic data that aligns with the distribution. Experiments in a real scenario demonstrate an average localization error of 1.966 and 1.218 m for Wi-Fi and Bluetooth low energy (BLE), surpassing traditional fingerprinting and comparable to the baseline data augmentation methods.