An RFID Tag Localization Method Based on Hologram Mask and Discrete Cosine Transform

Indoor radio frequency identification (RFID) tag positioning in strong multipath environments is critical in numerous industrial application scenarios, especially in production-related item monitoring. Among others, synthetic aperture radar (SAR)-based schemes have been developed, reconstructing tag positions from different angles. However, they usually compute entire holograms, containing lots of noncandidate grids and nonrelevant information to the tag position. This causes not only unnecessary overhead in position estimation, but also low generalization ability for a learning-based model. In this article, we propose a simple yet effective SAR-based tag localization model for fast and robust performance in strongly affected scenarios. In this model, we designed a two-step position estimation module based on a discrete cosine transform (DCT) for dimensionality reduction and an extra-trees model for position regression. Our model first preprocesses a mask to filter out noncandidate grids from holograms. This mask information is subsequently exploited in the regressor to further increase the positioning accuracy. Our experimental results showed that the proposed model surpasses existing holographic methods in both performance accuracy and inference time and can achieve 2 cm mean lateral accuracy in a strongly affected scenarios, with about a minute over training and testing phases.