Code and data for "Learned Spectrum Sensing Under Noise-Power Uncertainty: Invariance and Coherence"

Python code, trained models, result tables, figures and reproduction logs accompanying the manuscript “Learned Spectrum Sensing Under Noise-Power Uncertainty: Invariance and Coherence”. The package compares eighteen classical and learned spectrum-sensing detectors under a common false-alarm-calibrated protocol with controlled noise-power offsets. It also includes a sensing–throughput analysis for a cognitive radio network. All data are synthetic and generated by the seeded code. The main numerical results were obtained on the first author’s Windows computer. Running python run_all.py from the code directory executes the main experiments. Two runs of the main experiment pipeline on that computer produced bit-identical results apart from measured computing times. Results can differ in other environments, especially for trained networks. The reproducibility/ folder includes the development run and its comparison with the reported results. See README.md for setup instructions, file descriptions and software versions. Code is licensed under the MIT licence. Result tables, figures, trained models and logs are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0).