CrowdFi: A Communication Efficient Multi-device Wi-Fi Sensing System

In this paper, we propose a novel multi-device wireless sensing system, called CrowdFi, to balance the sensing performance and the transmission cost. In the CrowdFi, from the perspectives of devices, data, and bits, we propose the adaptive priority based transmission scheme for the heterogeneous data importance and time-varying channel of each device. Moreover, we design a two-stage training procedure and the loss functions to achieve a good tradeoff between the sensing accuracy and the transmission delay. We develop a prototype of the CrowdFi, and validate its performance by employing gait recognition as the application case. Experimental results demonstrate that the proposed CrowdFi system can reduce the transmission delay by 86.9%, while achieving the comparable or even improved recognition accuracy.