Related Experiment Video
Updated: May 25, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
2.5K
Enhanced Human Activity Recognition Using Wi-Fi Sensing: Leveraging Phase and Amplitude with Attention Mechanisms
Thai Duy Quy1, Chih-Yang Lin2, Timothy K Shih1
1Department of Computer Science and Information Engineering, National Central University, Taoyuan City 320317, Taiwan.
Sensors (Basel, Switzerland)
|February 26, 2025
Summary
This study introduces the Phase-Amplitude Channel State Information Network (PA-CSI) for robust Wi-Fi-based human activity recognition (HAR). The novel model effectively integrates amplitude and phase features, achieving state-of-the-art accuracy in diverse environments.
Area of Science:
- Computer Science
- Electrical Engineering
- Signal Processing
Background:
- Wi-Fi-based human activity recognition (HAR) offers a privacy-preserving method using Channel State Information (CSI).
- Existing HAR methods face challenges in robust feature extraction, particularly in dynamic or multi-environment settings, and struggle to integrate CSI's amplitude and phase data.
- Effective integration of both amplitude and phase features is crucial for improving HAR accuracy and robustness.
Purpose of the Study:
- To propose a novel model, the Phase-Amplitude Channel State Information Network (PA-CSI), for enhanced Wi-Fi-based HAR.
- To address limitations in existing approaches regarding feature extraction and the integration of dual CSI features (amplitude and phase).
- To improve the robustness and accuracy of HAR systems in complex and varied environments.
Main Methods:
- Developed the Phase-Amplitude Channel State Information Network (PA-CSI) incorporating a dual-feature approach.
- Implemented an attention-enhanced feature fusion mechanism combining multi-scale convolutional layers and Gated Residual Networks (GRN).
- Evaluated the model's performance on three distinct datasets: StanWiFi, MultiEnv, and a MINE lab dataset.
Main Results:
- The PA-CSI model achieved state-of-the-art performance across all tested datasets.
- Achieved high accuracy rates: 99.9% on StanWiFi, 98.0% on MultiEnv, and 99.9% on the MINE lab dataset.
- Demonstrated the effectiveness of the dual-feature approach and attention-enhanced fusion for robust feature extraction.
Conclusions:
- The PA-CSI model significantly advances Wi-Fi-based HAR by effectively utilizing both amplitude and phase CSI features.
- The proposed model offers a robust solution for human activity recognition in dynamic and multi-environment scenarios.
- Findings highlight the potential of PA-CSI for practical, real-world HAR applications.
