Real-Time Indoor Localization for Smartphones Using Tensor-Generative Adversarial Nets.

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• Accurate indoor localization is crucial for mobile computing and social networks.
• Existing radio frequency (RF) fingerprinting methods face challenges with real-time needs, limited data, and storage.
• Current state-of-the-art methods (kNN, DeepFi, AutoEncoder) have limitations in accuracy and efficiency.

Purpose of the Study:

• To propose a novel Tensor-GAN scheme for high-accuracy, real-time indoor localization.
• To address limitations of existing RF fingerprint-based localization techniques.
• To improve both localization accuracy and storage efficiency.

Main Methods:

• Modeling RF fingerprints as 3-D low-tubal-rank tensors to capture latent structures.
• Developing a three-player Tensor-GAN (regressor, generator, discriminator) for localization.
• Utilizing a tensor completion algorithm as the generator to create synthetic RF fingerprints for training.

Main Results:

• Achieved a localization accuracy of 0.19 m for 80% of test points, outperforming state-of-the-art methods (0.42 m).
• Demonstrated significant improvements in localization accuracy on a real-world dataset.
• Developed a prototype Tensor-GAN Android app with a small memory footprint (57 KB).

Conclusions:

• The proposed Tensor-GAN scheme offers a significant advancement in real-time indoor localization accuracy.
• The method effectively handles challenges of limited data and storage requirements.
• Tensor-GAN provides a practical and efficient solution for mobile indoor positioning systems.