HFDigiSet: A Broad-Coverage HF Digital-Mode I/Q Dataset for Signal Recognition

This data descriptor presents HFDigiSet, a dataset of complex baseband I/Q vectors generated from Fldigi WAV recordings. Real-valued audio modem signals are converted into analytic baseband signals using the Hilbert transform, shifted from the configured audio carrier to baseband, resampled to 6 kHz, and passed through a CCIR 520 Watterson fading channel. Random frequency offset, phase offset, and complex additive white Gaussian noise are then added. The public release contains 139 fine-grained HF digital-mode configurations, including distinct Fldigi modes and parameter variants. A 17-class parent-level benchmark is also provided by grouping closely related variants from the full release. Each sample is a fixed-length vector of 2048 complex samples with mode, SNR, channel-profile labels, and generation metadata. Signal-quality checks and baseline classification experiments on the 17-class benchmark are included to assess the use of HFDigiSet for HF digital-mode recognition experiments.