RFID techniques have been extensively used in sensing systems due to their low cost. However, limited by the structural simplicity, collision is one key issue which is inevitable in RFID systems, thus limiting the accuracy and scalability of such sensing systems. Existing anti-collision techniques try to enable parallel decoding without sensing based applications in mind, which can not operate on COTS RFID systems. To address the issue, we propose COFFEE, which enables parallel channel estimation of COTS passive tags by harnessing the collision. We revisit the physical layer design of current standard. By exploiting the characteristics of low sampling rate and channel diversity of RFID tags, we separate the collided data and extract the channels of the collided tags. We also propose a tag identification algorithm which explores history channel information and identify the tags without decoding. COFFEE is compatible with current COTS RFID standards which can be applied to all RFID-based sensing systems without any modification on tag side. To evaluate the real world performance of our system, we build a prototype and conduct extensive experiments. The experimental results show that we can achieve up to 7.33x median time resolution gain for the best case and 3.42x median gain on average.
REINFORCEMENT LEARNING BASED ANTI-COLLISION ALGORITHM FOR RFID SYSTEMS