Indoor Positioning Deep Learning Network based on CNN-GAN for High-precision Bluetooth

In recent years, Bluetooth Low Energy (BLE) positioning technology has garnered significant attention from researchers in the field of indoor localization. This study leverages IQ signal data collected from BLE beacons for positioning that construct spatial covariance matrix images as the dataset. However, the availability of Bluetooth datasets with precise location labels remains limited. To address this challenge, we propose a new architecture that combines Convolutional Neural Networks (CNN) and Generative Adversarial Networks (GAN), termed CNN-GAN, in order to augment fingerprint datasets, enhance data diversity, and ultimately improve the localization accuracy. The proposed model undergoes extensive testing in indoor environments with varying surface materials, room layouts, and density of collection points, followed by a comprehensive performance analysis. The experimental results show that the CNN-GAN architecture reduces the average localization error by approximately $19 \%$. Furthermore, the model trained on the dataset augmented with CNN-GAN achieves significantly higher localization accuracy than the model trained solely with the CNN, demonstrating the effectiveness and reliability of the proposed approach.