Cross layer design for RFID systems
Date of Issue2017-01-12
School of Computer Science and Engineering
Parallel and Distributed Computing Centre
Recently, Radio Frequency IDentification (RFID) systems are widely adopted in real world applications, such as inventory management, item tracking, and access control. Some fundamental RFID operations underlying almost every application, like identification, population estimation, missing tag detection should be executed with high efficiency and accuracy. Although a rich amount of works are devoted to improve the performances of these fundamental operations, most of these works suffer greatly from unconscious and frequent tag collisions -- especially in a large tag set -- and hence produce limited performance gains. In this thesis, I adopt a cross-layer design principle to facilitate these traditional MAC-layer operations with physical-layer information extracted from each colliding tag response signal, which is largely ignored by previous works. Specifically, two types of RFID operations are considered: population (or cardinality) estimation and identification. Since pinpointing the exact number of tags in a given set is expensive and unnecessary in most application scenarios, typical methods often take probabilistic models to estimate the cardinality within an accuracy requirement specified by the application. While most of existing probabilistic estimate methods detect and collect only binary states (i.e., empty or colliding) from the series of tag response slots, in this thesis, I propose to detect extra integer states (i.e., the exact number of colliding tags) in each colliding slot, by counting the number of clusters formed in the corresponding constellation map. As each integer state can infer an independent cardinality, I combine estimations based on all states into an optimal one (i.e., error is minimized) as the ultimate result. I name the proposed scheme after PLACE (Physical LAyer Cardinality Estimation). Experiments based on USRP2/WISP testbed and large-scale simulations show that PLACE achieves around 3x to 4x gains over state-of-the-art cardinality estimation schemes. RFID identification identify all tags in an unknown set with their binary IDs. Although current tree-based schemes improve the identification speed over classic ALOHA-based ones, their common underlying assumption on uniform ID distribution is not justified in most practical scenarios. In this thesis, I propose to exploit two types of physical layer information in each colliding slot and collect from them the hints on tag ID distribution in the binary tree. First, the RFID reader detects if all colliding tags in the same slot transmit the same bit at each ID index; a positive result at an index indicates no tags reply to a certain prefix pattern and prefixes matching the pattern can thus be skipped for subsequent queries. Second, the reader estimates the number of colliding tags in one slot and compute accordingly the optimal number of bits appended to the current prefix for subsequent queries. I name the proposed scheme after PHY-Tree. Experiments based on USRP2/WISP testbed and trace-driven simulations demonstrate that PHY-Tree reduce the number of reader queries of state-of-the-art schemes by 1.79x, on average.