Novel Transformer-Based Wi-Fi Sensing Method Handles Variable Traffic Patterns
Researchers have introduced a new approach to Wi-Fi-based motion recognition that addresses the challenge of variable transmission traffic, which often disrupts existing sensing systems. Traditional Wi-Fi sensing models, which rely on Channel State Information (CSI), typically struggle with poor generalization when faced with fluctuating sampling rates and intervals caused by network traffic variations. To solve this, the team proposed a Sampling Rate Versatile Neural Network (SRV-NN) built on transformer architecture, designed to efficiently process variable input-sized sensing signals. Additionally, they employed dynamic sampling rate augmentation to enhance model robustness. The method was validated through extensive experiments using two self-collected datasets (SRV activity and SRV gesture) and two public datasets. Results demonstrated exceptional performance and stability, showing substantial improvements in average accuracy and significantly reduced variance across different sampling rates compared to baseline models. This advancement promises more reliable Wi-Fi sensing for applications like gesture and activity recognition in real-world environments with unpredictable network conditions.
Wire timeline
Novel Transformer-Based Wi-Fi Sensing Method Handles Variable Traffic Patterns
Researchers have introduced a new approach to Wi-Fi-based motion recognition that addresses the challenge of variable transmission traffic, which often disrupts existing sensing systems. Traditional Wi-Fi sensing models, which rely on Channel State Information (CSI), typically struggle with poor generalization when faced with fluctuating sampling rates and intervals caused by network traffic variations. To solve this, the team proposed a Sampling Rate Versatile Neural Network (SRV-NN) built on transformer architecture, designed to efficiently process variable input-sized sensing signals. Additionally, they employed dynamic sampling rate augmentation to enhance model robustness. The method was validated through extensive experiments using two self-collected datasets (SRV activity and SRV gesture) and two public datasets. Results demonstrated exceptional performance and stability, showing substantial improvements in average accuracy and significantly reduced variance across different sampling rates compared to baseline models. This advancement promises more reliable Wi-Fi sensing for applications like gesture and activity recognition in real-world environments with unpredictable network conditions.
cs.AI updates on arXiv.org