High-Capacity Complex Convolutional Neural Networks For I/Q Modulation Classification

TL;DR

This paper introduces high-capacity complex convolutional neural networks with residual and dense connections for I/Q modulation classification, achieving state-of-the-art accuracy and outperforming comparable models in speed and complexity.

Contribution

It demonstrates that high-capacity architectures with complex convolutions significantly improve I/Q modulation classification performance over prior shallow CNNs.

Findings

Peak accuracy of 92.4% on RadioML 2016.10a dataset

Statistically significant improvements with complex convolutions

Models outperform comparable architectures by over 10% in speed and complexity

Abstract

I/Q modulation classification is a unique pattern recognition problem as the data for each class varies in quality, quantified by signal to noise ratio (SNR), and has structure in the complex-plane. Previous work shows treating these samples as complex-valued signals and computing complex-valued convolutions within deep learning frameworks significantly increases the performance over comparable shallow CNN architectures. In this work, we claim state of the art performance by enabling high-capacity architectures containing residual and/or dense connections to compute complex-valued convolutions, with peak classification accuracy of 92.4% on a benchmark classification problem, the RadioML 2016.10a dataset. We show statistically significant improvements in all networks with complex convolutions for I/Q modulation classification. Complexity and inference speed analyses show models with complex convolutions substantially outperform architectures with a comparable number of parameters and comparable speed by over 10% in each case.