MultiChannel-16-ModSet: A Multi-Channel Modulation Recognition Dataset for Channel Generalization

This dataset contains recordings of 12 digital modulation schemes across **16 distinct channel conditions** — one real **over-the-air (OTA)** channel and fifteen **3GPP-compliant Tapped-Delay-Line (TDL)** channels — designed for studying the **channel-generalization** capability of **automatic modulation classification (AMC)** models. The data generation and collection were performed by the **Software Defined System Studio (SDS Studio)**, a platform independently developed by the Broad-band Communication and Network Group (BCNG) of the National University of Defense Technology. The OTA channel was captured between Universal Software Radio Peripheral (USRP) devices in a controlled indoor environment, while the 15 TDL channels were produced through a hardware-in-the-loop (HIL) setup in which SDS Studio baseband waveforms pass through a channel emulator configured with 3GPP TR 38.901 TDL profiles. All channels retain real radio-frequency front-end impairments. The dataset is stored in a single HDF5 (`.h5`) file with predefined training, validation, and test splits. Each sample consists of **I/Q complex samples** along with its modulation label, SNR value, and **channel ID**.