Radio Modulation Classifier

CNN-based automatic modulation classification (AMC) for radio signals. Classifies common digital modulations (e.g. BPSK, QPSK, 8PSK, 16QAM) from I/Q samples.

Motivation

Automatic modulation recognition is used in spectrum monitoring, cognitive radio, and signal intelligence. Deep learning can learn discriminative features from raw I/Q data without hand-crafted feature design.

Dataset

You can use the RadioML 2018.01A dataset (available from Deepsig) or similar public AMC datasets.
Alternatively, run generate_synthetic_data.py to create a small synthetic dataset for quick experiments.
Data format: (N, 2, L) — N samples, 2 channels (I, Q), L time steps per sample.

Project structure

model.py — ResNet-style 1D CNN.
train.py — training loop; uses synthetic data if nothing is in ./data.
generate_synthetic_data.py — writes synthetic I/Q to ./data/synthetic_iq.pt.

Model

Small ResNet-style CNN that takes (2, L) I/Q input and outputs class logits.
Implemented in model.py; training script in train.py.

Usage

pip install -r requirements.txt
python train.py

Training expects data in ./data (or set DATA_DIR). For RadioML, place the .hdf5 or processed files in that directory and adjust paths in train.py if needed.

Limitations / future work

Trained and evaluated on a single dataset; performance may drop under different SNR or channel conditions.
Could be extended with attention or 1D variants for longer sequences.

Author

Alireza Aminzadeh

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support