WiFi RSS and RTT Indoor Positioning with Graph Temporal Convolution Network.

Area of Science:

• Computer Science
• Electrical Engineering
• Robotics

Background:

• Indoor positioning systems (IPS) are crucial for various applications.
• Achieving sub-meter accuracy with commodity WiFi is hindered by multipath fading and Non-Line-Of-Sight (NLOS) propagation.
• Existing methods struggle with diverse indoor layouts and dynamic environments.

Purpose of the Study:

• To develop a novel hybrid Graph-Temporal Convolutional Network (GTCN) for high-accuracy indoor positioning.
• To enhance robustness by jointly utilizing WiFi Received Signal Strength (RSS) and Round-Trip Time (RTT) features.
• To create a computationally efficient model suitable for real-time deployment on edge devices.

Main Methods:

• Proposed a hybrid GTCN model integrating graph convolutions for Access Point (AP) geometry and dilated temporal convolutional networks for signal dynamics.
• Implemented a lightweight gating mechanism for adaptive per-AP importance learning.
• Evaluated the model using both WiFi RSS and RTT measurements across diverse indoor environments.

Main Results:

• The GTCN model demonstrated high positioning accuracy across lecture theatres, offices, corridors, and building floors.
• Positioning accuracy improved with increased AP density, particularly in large-scale mixed environments under Line-Of-Sight (LOS) and NLOS conditions.
• The model requires fewer than 105 trainable parameters and tens of MFLOPs per inference, enabling real-time performance.

Conclusions:

• The proposed GTCN model offers a robust and accurate solution for indoor positioning using commodity WiFi.
• The hybrid approach effectively addresses challenges posed by multipath fading and NLOS effects.
• The model's computational efficiency makes it suitable for real-time applications on embedded and edge computing platforms.