Deep-Learning-Based Wi-Fi Indoor Positioning System Using Continuous CSI of Trajectories.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Wi-Fi indoor positioning systems (IPS) performance relies on channel state information (CSI).
• Multipath fading and non-line-of-sight (NLOS) propagation in indoor environments limit CSI quality.
• Existing IPS methods often collect CSI at stationary locations, which is inefficient.

Purpose of the Study:

• To propose a novel IPS that utilizes trajectory CSI for enhanced performance.
• To reduce the cost and complexity of CSI data collection.
• To leverage deep learning for improved spatial and temporal exploitation of trajectory CSI.

Main Methods:

• Proposed an IPS using trajectory CSI collected along predetermined paths.
• Employed a generative adversarial network (GAN) to augment the CSI training dataset.
• Utilized a 1D convolutional neural network-long short-term memory (1DCNN-LSTM) deep learning network.
• Hardware implementation using digital signal processors and universal software radio peripherals.

Main Results:

• The proposed trajectory CSI-based IPS significantly outperforms traditional stationary CSI-based IPS.
• The use of GAN effectively reduced the cost of trajectory CSI collection.
• Extensive experiments and simulations validated the superior performance.

Conclusions:

• Trajectory CSI-based IPS offers a more robust and accurate solution for indoor positioning.
• Deep learning models like 1DCNN-LSTM are effective in exploiting spatio-temporal CSI data.
• The proposed method presents a significant advancement over current state-of-the-art IPS techniques.