A Semi-Supervised 3D Indoor Localization Using Multi-Kernel Learning for WiFi Networks.

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• Indoor localization is crucial for location-based services.
• Radio frequency (RF) based localization offers low-energy, simple solutions.
• Kernel learning has been applied to 2D indoor localization but not yet to 3D environments.

Purpose of the Study:

• To propose and evaluate a multi-kernel learning scheme for 3D indoor localization.
• To enhance localization accuracy in 3D environments by leveraging multiple kernels.
• To address limitations of single-kernel approaches and improve upon existing methods.

Main Methods:

• A multi-kernel learning scheme is proposed for 3D indoor localization.
• WiFi signals are filtered for quality, proximity, and to mitigate multipath effects and interference.
• User location is determined by collecting wireless received signal strengths (RSS) and signal-to-noise ratios (SNR) using multiple kernels.

Main Results:

• The multi-kernel learning scheme demonstrated improved localization accuracy in a 3D indoor environment.
• The proposed method outperformed both multi-deep neural network (DNN) and existing kernel-based localization schemes.
• Experimental validation used real RSS and SNR data from multiple wireless access points (AP).

Conclusions:

• The proposed multi-kernel learning scheme is effective for 3D indoor localization.
• This approach offers superior accuracy and reduced error compared to existing methods.
• The technique successfully maps signal data to a location database through multi-kernel training.