1 Release 1 Tag

RSS Feed

• v0.1.0 9979d84e29

  Compare

  ModrecWorkflow v0.1.0 – Initial Release Stable

  Liyux released this 2025-07-08 11:10:15 -04:00 | 18 commits to main since this release

  An end-to-end CI/CD pipeline for modulation recognition model training and deployment. This workflow enables reproducible data generation, model development, and deployment into inference systems, all in a streamlined DevOps-style fashion.

  🌟 Workflow Contents

  Automated Data Generation: Generates labeled synthetic IQ samples across digital modulation types like BPSK, QPSK, and 16QAM using domain-configurable parameters.

  Dataset Construction: Aggregates generated IQ data into structured datasets, ready for training and evaluation.

  Model Training Pipeline: Trains deep learning models on synthetic datasets using a reproducible and customizable training loop with built-in profiling.

  Model Profiling & Export: Profiles the trained model's performance (latency, accuracy) and exports the model to ONNX format.

  ORT Export: Converts ONNX model to optimized ONNX Runtime (ORT) format for fast, production-ready inference.

  CI/CD Integration: End-to-end GitHub Actions workflow automates the pipeline—data generation, training, export, and packaging.

  Modulation Configuration: Modulation types, channel conditions, and signal parameters are easily defined via YAML or environment variables.

  🛠️ Performance Improvements
  Parallelized data generation for faster synthetic dataset creation.

  Model training optimized for GPU when available.

  🚧 Known Issues
  No built-in evaluation set for real-world signals—only synthetic support for now.

  Model performance may vary based on SNR range and modulation overlap.

  🙌 New Contributors
  This initial release is a collaborative effort from the Qoherent team, incorporating utilities and insights gained
  from numerous projects.

  Of special mention are the following team members who have authored contributions to the project:

  • Michael Luciuk

  • Liyu Xiao

  💡 Future Plans

  • Add TensorRT Runtime Support

  • Package RIA Scripts as Mini CLI Tools

  ℹ️ Additional Information

  We’re excited to share this resource under the AGPLv3 License, helping others to accelerate their own intelligent radio development and research.

  Downloads

  • Source Code (ZIP)
  • Source Code (TAR.GZ)