A high-dimensional, multi-transceiver channel state information dataset for enhanced human activity recognition.

Area of Science:

• Computer Science
• Signal Processing
• Machine Learning

Background:

• Human Activity Recognition (HAR) is crucial for real-world applications.
• Existing Channel State Information (CSI)-based HAR datasets lack diversity and complexity.
• Limited datasets hinder the development of robust HAR models due to insufficient training data.

Purpose of the Study:

• To introduce a novel, diverse, and complex CSI-based dataset for Human Activity Recognition (HAR).
• To capture spatial diversity using multiple transceiver orientations and high-dimensional subcarriers.
• To address limitations of existing datasets by incorporating real-world factors like varied activities, movements, body composition, and environmental conditions.

Main Methods:

• A controlled laboratory experiment utilizing four ESP32-S3-DevKitC-1 devices as transceivers under Wi-Fi IEEE 802.11n standard.
• CSI data collection from multiple transceiver orientations capturing spatial diversity across 166 subcarriers.
• Inclusion of diverse human activities, micro/macro movements, body composition variations, and environmental noise/interference.

Main Results:

• A novel dataset with enhanced spatial diversity and high dimensionality (166 subcarriers) was created.
• The dataset captures a wide range of activities, movements, and environmental factors.
• The collected data supports the development of more robust and generalizable CSI-based HAR models.

Conclusions:

• The novel dataset overcomes limitations of existing CSI-based HAR datasets by providing greater diversity and complexity.
• Leveraging this dataset enables the development of more accurate and robust HAR models for real-world scenarios.
• CSI-based HAR models utilizing multi-perspective spatial variations and high-dimensional data show promise for enhanced performance.