Non-Stationary Aerial Channel Spectrum Sensing Dataset for CNN Generalization Studies
Bildiriler > Tam Metin Bildiri
Yazarlar: Almamoori, Hayder, Münif ZEYBEK
Zenodo (CERN European Organization for Nuclear Research), , , Uluslararası, Tam Metin Bildiri
Özet
This dataset contains simulation-generated data for evaluating convolutional neural network (CNN)-based spectrum sensing under domain shift in non-stationary aerial communication channels. The dataset is constructed using a controlled signal generation framework that models realistic UAV channel conditions, including Doppler shifts, time-varying multipath fading, and varying signal-to-noise ratio (SNR). Three distinct domains are provided to represent different channel regimes: Domain A: Mild channel conditions (training domain) Domain B: Harsh out-of-domain conditions (evaluation domain) Domain C: Mixed channel conditions (robustness evaluation) Each domain contains balanced samples corresponding to signal-present and noise-only hypotheses. The input representation consists of in-phase and quadrature (IQ) components arranged as two-channel tensors suitable for deep learning models. The dataset is intended to support research on generalization, robustness, and domain adaptation in machine learning-based spectrum sensing. It is provided for peer review purposes and will be publicly documented and expanded upon acceptance. All data are stored in NumPy compressed (.npz) format. A preview subset and example usage may be included to facilitate inspection.