Scaling RFID positioning systems using distributed and split computing

Abstract

Fine-grained tracking of objects in the physical world at scale has a broad potential impact in health care, retail, manufacturing, supply chain, and consumer product industry. In this thesis, I focus on using RFID-based technology for such applications due to its low-cost and growing prevalence of RFID tags. In contrast to current RFID systems that focus on a monolithic reader, I propose a distributed sensor node architecture that can scale by combining distributed and split computing techniques. On the distributed computing front, I introduce an architecture that enables extending the operation range and coverage from an end user's perspective while improving the manageability aspect via high-level semantic API. On the split computing front, I develop a framework to offload expensive tasks to the cloud or an edge server; the framework enables the use of small, cheap commodity compute devices as hosts at the edge while maintaining the high accuracy of fine-grained positioning. The thesis describes the design and implementation of these techniques. Moreover, through a hybrid evaluation of simulation and practical systems, the thesis demonstrates how these techniques enable us to design a scalable, manageable, and accurate RFID positioning system.