Enabling Coexistence of Indoor Millimeter- Wave Networking and Human Activity Sensing

We propose mNetS, a system that enables the coexistence of indoor millimeter-wave (mmWave) networking and sensing for at-home physical rehabilitation and continuous health diagnostics. Although mmWave signal has been shown to effectively capture human activities/posture, running networking and sensing applications simultaneously remains a challenge. Typically, sensing can only be performed opportunistically to minimize the impact on networking, resulting in low-rate data capture with irregular spacing that affects sensing accuracy. To overcome this challenge, mNetS leverages idle times within the data transfer process to perform sensing opportunistically. Next, using a customized Dynamic Graph Convolutional Neural Network (DGCNN), mNetS extracts relevant features from low-rate sensing samples that are adaptively combined and regressed to estimate high-rate sensing samples. To demonstrate the effectiveness of mN etS, we collect signal reflections from various human activities that involve diverse movements of different body parts. We train and test mNetS on these data samples, and our results show a significant improvement in estimating high-rate signal samples, qualitatively and quantitatively. Such high-rate signal samples estimated by mN etS improves the performance of a typical human activity sensing application such as the classification of activity markedly over reduced rate samples or high-rate samples estimated through a naive linear interpolation.