Dataset for Machine Learning on RFID tags: 4326 VNA measurements with 3 tags and 4 different environments

This dataset contains 4,326 S-parameter measurements of retransmission-based chipless RFID tags, acquired to support machine learning-based tag identification or localization in realistic, interference-affected conditions. Three fabricated tag samples encoding the 3-bit identifiers 001, 011 and 111 were measured with a two-port VNA and a broadband dual-polarized horn antenna over the 3.2–5.2 GHz band, using 20,004 uniformly spaced frequency samples per sweep (99.985 kHz step). Each sweep is released as the raw VNA CSV export and provides magnitude and phase for all four scattering parameters, giving nine values per frequency sample. Measurements were repeated across four environments: an RF-shielded controlled environment with absorbers (2,940 sweeps, tag–antenna distances of 1–49 cm in 1 cm steps, 10 repetitions), and three uncontrolled real-world environments introducing a concrete wall, a large metal obstacle, and complex indoor clutter (462 sweeps each, 11 distances between 3 and 49 cm, 7 repetitions). Every configuration was recorded at two tag heights above the ground plane, 4 cm and 11.5 cm. The three tag identifiers are exactly balanced with 1,442 sweeps each. All acquisition parameters are encoded in the file names following the convention M_Sp_XY_ID_S_Dcm_H_A.csv (see readme), and the environment is given by the folder structure. A separate angle study is also included, comprising 315 additional sweeps acquired in the controlled environment with the tag rotated to −30°, 0° and +30° relative to the reader antenna, at five distances between 2 and 18 cm and at a single height, using a coarser 201-point sweep; these sweeps are provided for completeness and are not part of the 4,326 identification measurements. The archive totals approximately 8.2 GB, with an average file size of 1.8 MB. A README documents the naming convention, the file format, and a reference loading script.