Doppler Invariant CNN for Signal Classification

Current deep learning models for automatic signal classification rely on brute-force Doppler augmentation to achieve real-world generalization, undermining training efficiency and interpretability. We propose a convolutional neural network (CNN) with complex-valued layers and adaptive polyphase sampling (APS) for pooling to achieve provable frequency-bin shift invariance. Using a synthetic dataset of common interference signals, experimental results demonstrate that unlike a vanilla CNN, our model maintains consistent classification accuracy with and without random Doppler shifts despite being trained on no Doppler-shifted examples.