USRP-based Audio Transmission and Activity Recognition
Kaile Cui, Quanquan Liang, Yan Zhu · 2024
In the past few years, contactless activity recognition has gained widespread attention, demonstrating potential applications in both the medical and health monitoring fields as well as in smart homes. It has become common practice to utilize Channel State Information (CSI) for human activity recognition. In this paper, we propose activity recognition through audio transmission. Unlike the traditional way of sending random sequences to extract CSI, we choose to transmit a fixed sequence of audio and utilize the transmitted audio for activity recognition while ensuring accurate audio reception. Two Universal Software Radio Peripherals (USRP) are used as transceivers, and the audio is transmitted and received through a shared Simulink platform. Finally, the assessment of audio quality is not only subjective based on hearing but also objective by analyzing the time-frequency characteristics extracted from the received audio. In the context of activity recognition, the CSI of the daily movements of an individual (such as tiptoeing, standing up, walking, and squatting) is assessed for classification. Machine learning and deep learning algorithms are utilized for classification, and the results show that the classification can achieve a peak recognition accuracy of 98%.