Cellular-Assisted, Deep Learning Based COVID-19 Contact Tracing

The Coronavirus disease (COVID-19) pandemic has caused social and economic crisis to the globe. Contact tracing is a proven effective way of containing spread COVID-19. In this paper, we propose CAPER, Cellular-Assisted deeP lEaRning based COVID-19 contact system on cellular network channel state information (CSI) measurements. CAPER leverages deep neural feature extractor map CSI space, within which Euclidean distance between points strongly correlates with proximity devices. By doing so, maintain user privacy by ensuring that never propagates one client's data its server or other clients. We implement prototype using software defined radio platform, evaluate performance in variety real-world situations including indoor outdoor scenarios, crowded sparse environments, differing traffic patterns configurations common use. Microbenchmarks show our model runs 12.1 microseconds OnePlus 8 smartphone. End-to-end results demonstrate achieves an overall accuracy 93.39%, outperforming BLE approach 14.96%, determining whether two devices are six feet not, only misses 1.21% close contacts. also robust environment dynamics, maintaining 92.35% after running for ten days.