Synthetic Radio Signal Datasets for Machine Learning Benchmarking

A survey paper introduces several public open source synthetic datasets for radio machine learning tasks. The datasets are intended to provide a robust common baseline for model learning, testing, and evaluation in the field. The paper is authored by T. J. O’Shea from Virginia Tech.

Use Cases

• Benchmarking radio signal classification models based on synthetic signal data.
• Evaluating model generalization for radio frequency tasks using common baselines.
• Training machine learning models for signal modulation recognition using provided datasets.
• Comparing the performance of different radio ML techniques against a standard measure.

Strengths

• Datasets are described as public and open source, facilitating access.
• Datasets are intended to provide a robust common baseline for the field.
• The paper discusses the critical importance of good datasets for model evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Virginia Tech (author T. J. O’Shea), via paperswithcode.

Collection Method

Synthetic dataset generation, likely using GNU Radio as indicated in the title.