sigkit 0.0.3

The Python DSP Toolkit

SigKit

SigKit is a modular digital signal‐processing toolkit built on top of NumPy. It has additive impairments to simulate OTA impairments. We include a full Machine Learning toolkit with methods for PyTorch designed for training Modulation Classification Models that are capable of generalizing over the air (OTA). It provides:

• Core abstractions (Signal, Impairment, Modem, …) for working in complex baseband
• NumPy ipairments, PyTorch Transforms & tools (AWGN, fading, filtering, SNR & BER calculators)
• PyTorch Transforms so you can drop signal operations straight into Compose
• PyTorch Lightning Pipeline includes our pretrained model and methods for retraining
• (WIP) GNURadio Blocks wrapping our tools
• Synthetic data generators & torch.utils.data.Dataset classes

---

🚀 Getting Started

Try the example notebook

A quick way to explore SigKit is to run the Jupyter notebook in:

examples/notebooks/basic.ipynb

The notebook is a guide that covers:

• Generating a signal with a Modem
• Adding an Impairment like AWGN
• Calculating Signal Metrics
• Visualizing the waveform

Be sure to restart your notebook's kernel after installing the package.

Classification

There is also an example notebook for running the pretrained model.

examples/notebooks/classification.ipynb

Pull down LFS objects first with git lfs pull

Installing SigKit

Package Installation

Installing the pypi package

pip install sigkit

Installing from source (recommended for ML tasks)

git clone https://github.com/IsaiahHarvi/SigKit.git
cd SigKit
pip install -e .

Sanity Check for source installs

You can be gauranteeed your installation is sound by running pytest without failure from the root of the repository.

(Optional) DevContainer for VS Code

If you use VS Code and would prefer to isolate your SigKit installation, we’ve provided a DevContainer in the repository. It is designed to be OS agnostic but it is confirmed to support: Ubunutu >22.04, ARM MacOS, Windows

To setup:

1. Install the Remote – Containers extension in VS Code.
2. Clone, open the project, and run chmod +x .devcontainer/setup.sh.
3. Run Reopen in Container from the VSCode console.

Inside the container you’ll have all dependencies installed and SigKit ready to run.

---

State of Implementations

The below table tracks major planned features. Smaller features like Signal Impairments and Torch Transforms are not tracked here.

Feature	Implemented	Torch Transform
Architecture	-	-
Rust bindings	No	-
GNURadio	-	-
ClassificationBlock	No	-
Modulator	No	-
Modems	-	-
PSK	Yes	-
FSK	Yes	-
QAM	WIP	-
OFDM	WIP	-
MSK	No	-
ASK	No	-
Machine Learning	-	-
Training Pipeline	Yes	-
Pretrained Model	Yes (WIP)	-
Procedural Dataset	Yes	-
File Dataset	No	-
Signal Impairments	-	-
AWGN	Yes	✅
Frequency Shift	Yes	✅
Phase Shift	Yes	✅
Polyphase Resampling	No	❌
Rayleigh Fading	No	❌
Random Resampling	No	❌
IQ Imbalance	No	❌
Time Shift	No	❌

---

🛠️ Development

• Tests:

  pytest
  

• Lint & Format:

  ruff check . --fix
  ruff format .
  

• Update docs:

  # Install the optional dependencies [docs] or:
  pip install sphinx piccolo_theme tomli
  
  ./docs/gen.sh
  

---

License

This project is released under the MIT License. See LICENSE for details.