RateCSI: A Public CSI Dataset for Wi-Fi Rate Adaptation

RateCSI is a large-scale Wi-Fi channel-state-information (CSI) datasetcollected on commodity ESP32C5 radios for research on wireless sensing and linkadaptation. It contains 133 measurement runs — 44 line-of-sight (LOS), 49non-line-of-sight (NLOS), and 40 dynamic "rapid-sweep" sessions with random humanmobility — spanning more than ten distinct indoor and outdoor environments(classrooms, hallways, offices, an auditorium, a rooftop, and an open court) attransmitter–receiver separations from 1 to 8 m. In aggregate it provides nearly2.9 million parsed CSI records paired with over 11 million per-packet transmissionand packet-delivery-ratio (PDR) observations, totalling roughly 40 hours ofcapture. Each run bundles time-synchronized CSI, transmission, and PDR tables withstandardized run-level metadata, so per-subcarrier channel measurements can beassociated with the corresponding MCS decisions and delivery outcomes — supportinglink-quality prediction, rate adaptation, and LOS/NLOS classification. All runs follow a single documented schema and ship with a searchable manifest,per-file SHA-256 checksums, and a validation script. Device addresses and siteidentifiers are anonymized. Two archives are provided: the full corpus(esp-csi-scenarios-v1.tar.gz) and a small three-run subset(esp-csi-scenarios-samples.tar.gz) for quick inspection.