Dataset for "IriSig-Spoof: A Real-World Benchmark for Time-Robust Satellite RF Fingerprinting and Spoofing Detection"

IriSig-Spoof is a labelled dataset of Iridium ring alert (IRA) downlink messages and SDR-generated spoofing signals designed for satellite radio-frequency fingerprinting (RFF), open-set identification, and spoofing detection. The legitimate dataset contains approximately 5.17 million IRA messages from 66 Iridium satellites, collected continuously over 32 days from June 27 to July 28, 2025. Legitimate satellite signals were received using an RST740 rooftop antenna and a USRP B210 at the Xidian University, in Xi'an, Shaanxi Province, China. The signals were sampled at 250 kS/s, corresponding to 10× oversampling of the IRA symbol rate, to preserve transmitter-dependent physical-layer characteristics for RFF analysis. Each model-ready signal sample is represented as a 2 × 2000 real-valued I/Q sequence, where the two channels correspond to the in-phase (I) and quadrature (Q) components. Each sequence is extracted around the detected IRA unique-word position (uw_start), spanning 500 samples before and 1500 samples after the synchronization point. The released data are stored in NumPy format and can therefore be directly loaded for signal processing and machine-learning experiments. The legitimate dataset is organized hierarchically by acquisition date and satellite identity. Each date is distributed as a separate ZIP archive following the MMDD.zip naming convention (e.g., 0627.zip for June 27 and 0728.zip for July 28). Within each date, NumPy files are indexed by satellite identity, and each file contains all valid IRA samples from the corresponding satellite on that date in an N × 2 × 2000 array, where N denotes the number of available messages. The chronological order of the legitimate data should be determined from the archive names rather than from their display order on the Zenodo webpage. The spoofing dataset contains IRA-compatible signals generated with a USRP B210 transmitter and collected under multiple controlled indoor and outdoor propagation scenarios, including different line-of-sight and non-line-of-sight conditions and transmitter–receiver configurations. Spoofing samples use the same signal representation as the legitimate data and are organized by spoofing scenario, enabling controlled evaluation across different propagation environments. The dataset was originally collected for the paper “IriSig-Spoof: A Real-World Dataset and Benchmark for Time-Robust Satellite RF Fingerprinting and Spoofing Detection.” Its date-, identity-, and scenario-aware organization is designed to support reproducible evaluation of cross-day RFF robustness, open-set satellite identification, unknown-signal rejection, and cross-scenario spoofing detection, while reducing the risk of overly optimistic results caused by random train/test splits within the same acquisition environment. Researchers are welcome to contact [e-mail] for questions, discussion, or collaboration related to the dataset and benchmark.