A Domain-Independent Generative Adversarial Network for Activity Recognition Using WiFi CSI Data.

Area of Science:

• Computer Science
• Artificial Intelligence
• Signal Processing

Background:

• Device-free sensing using WiFi Channel State Information (CSI) is gaining traction for unobtrusive context recognition.
• Existing deep learning techniques for CSI-based activity recognition suffer performance degradation due to domain changes (environment, configuration, subjects, tasks).
• Addressing domain change typically requires extensive, time-consuming data collection, which is impractical for WiFi CSI signals.

Purpose of the Study:

• To propose a novel domain-independent generative adversarial network (GAN) for WiFi CSI-based activity recognition.
• To develop a simplified data pre-processing module to complement the GAN.
• To enhance robustness against domain shifts and improve activity recognition accuracy while reducing model complexity.

Main Methods:

• A domain-independent generative adversarial network (GAN) architecture was developed.
• A simplified data pre-processing module was integrated with the GAN.
• The proposed approach was evaluated against state-of-the-art methods for WiFi CSI activity recognition.

Main Results:

• The proposed domain-independent GAN demonstrated superior robustness against domain change compared to existing methods.
• The approach achieved higher accuracy in WiFi CSI-based activity recognition.
• The model complexity was significantly reduced, making the system more efficient.

Conclusions:

• The developed domain-independent GAN offers a more robust and accurate solution for WiFi CSI activity recognition.
• The simplified pre-processing module contributes to the overall efficiency and effectiveness of the system.
• This work addresses the critical domain change problem in WiFi CSI sensing, paving the way for more practical applications.