HeteroSignal-7: A Multi-Task Electromagnetic Signal Routing Benchmark
Year: 2026
Type: dataset
Abstract: HeteroSignal-7 is a multi-task electromagnetic signal routing benchmark comprising 140,000 samples from seven datasets across six task types. The benchmark includes automatic modulation classification using RML2016.10a and RML2018.01a, signal denoising using SD-BENCH3, wireless interference identification using EM-Infer-Comm, radio-frequency fingerprint identification using FS-SEI, radar waveform classification using RadChar, and blind source separation using EM-Radar-Mix. Each dataset contains 20,000 samples with a unified 8:1:1 training, validation, and test split. Stratification by class and SNR is applied where applicable, while denoising samples are grouped by clean reference to prevent reference leakage across splits. Cross-task source-overlap checks are conducted for the validation and test splits, resulting in 13,971 validation samples and 13,975 cleaned test samples. The associated evaluation framework covers both classification and waveform-recovery tasks and uses a heterogeneous pool of seven expert models. Source code and reproduction instructions are available at https://github.com/Dawn-try/GLCR.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225756869