On Deep Learning Classification of Digitally Modulated Signals Using Raw I/Q Data

TL;DR

This paper investigates how well deep neural networks, specifically residual and convolutional architectures, can classify digitally modulated signals from raw I/Q data across different datasets, highlighting generalization challenges.

Contribution

It compares the generalization ability of residual and CNN models on independently generated datasets for digitally modulated signal classification.

Findings

Residual networks show better generalization than CNNs.

Models trained on one dataset perform poorly on a different dataset.

Study highlights importance of dataset diversity for training robust classifiers.

Abstract

The paper considers the problem of deep-learning-based classification of digitally modulated signals using I/Q data and studies the generalization ability of a trained neural network (NN) to correctly classify digitally modulated signals it has been trained to recognize when the training and testing datasets are distinct. Specifically, we consider both a residual network (RN) and a convolutional neural network (CNN) and use them in conjunction with two different datasets that contain similar classes of digitally modulated signals but that have been generated independently using different means, with one dataset used for training and the other one for testing.