Deep Learning-Based Wi-Fi Signal Fingerprinting Indoor Positioning Technology

With the rapid growth of demand for location-based services in indoor environments, Wi-Fi signal fingerprint positioning, as a mainstream indoor positioning technology, is gaining increasing attention due to its simple system equipment and high positioning accuracy. Indoor positioning technology based on Received Signal Strength Indicator (RSSI) utilizes wireless signal propagation models and plays a crucial role in building RSSI fingerprint databases. The quality and quantity of the database significantly impact the entire fingerprint positioning system. Therefore, it is necessary to construct a dense offline fingerprint database that meets accuracy requirements to fulfill people’s demand for location services. However, this approach can lead to increased costs, substantial workload, and high system complexity. We have found that in deep learning (DL), deep generative models are beneficial for data augmentation, addressing the significant impact of dataset scarcity on learning performance. Data augmentation or synthesis promises to solve the problem of sparse data in databases. Thus, this paper proposes using deep generative models to generate synthetic data to address these issues. We choose the Conditional Tabular Generative Adversarial Network (CTGAN) to generate a synthetic fingerprint database for indoor positioning. Using CTGAN not only maximally retains the original information during the training process but also makes the generated data closer to the real data distribution, accelerating the training speed of the dataset. By collecting a small amount of fingerprint data, higher positioning accuracy can be achieved. Therefore, using a small dataset, data synthesis through CTGAN can mitigate the challenges of offline fingerprint databases. Index Terms: Deep Learning, Fingerprint Localization, Conditional Tabular Generative Adversarial Network (CTGAN), Generative Adversarial Networks (GAN), Synthetic Data