Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large-Scale Classification of Shortwave Communication Signals with Machine Learning

Stefan Scholl

TL;DR

The paper tackles large-scale shortwave signal classification using a blind CNN trained on synthetic and real IQ data with extensive augmentation to mimic ionospheric propagation and noise. It demonstrates a 28-layer CNN with 1.7 million parameters trained end-to-end on 4 kHz IQ signals and evaluated on real-world data from worldwide SDRs, achieving about 90% accuracy (top-3 ~95%) for 143 modes at a 1-second observation window. The results show strong real-world performance across analog and digital HF modes and highlight remaining confusions among similar signals and the importance of explainability. The work provides a scalable baseline for large-class RF signal classification in the HF band with practical implications for spectrum monitoring and cognitive radio.

Abstract

This paper presents a deep learning approach to the classification of 160 shortwave radio signals. It addresses the typical challenges of the shortwave spectrum, which are the large number of different signal types, the presence of various analog modulations and ionospheric propagation. As a classifier a deep convolutional neural network is used, that is trained to recognize 160 typical shortwave signal classes. The approach is blind and therefore does not require preknowledge or special preprocessing of the signal and no manual design of discriminative features for each signal class. The network is trained on a large number of synthetically generated signals and high quality recordings. Finally, the network is evaluated on real-world radio signals obtained from globally deployed receiver hardware and achieves up to 90% accuracy for an observation time of only 1 second.

Large-Scale Classification of Shortwave Communication Signals with Machine Learning

TL;DR

The paper tackles large-scale shortwave signal classification using a blind CNN trained on synthetic and real IQ data with extensive augmentation to mimic ionospheric propagation and noise. It demonstrates a 28-layer CNN with 1.7 million parameters trained end-to-end on 4 kHz IQ signals and evaluated on real-world data from worldwide SDRs, achieving about 90% accuracy (top-3 ~95%) for 143 modes at a 1-second observation window. The results show strong real-world performance across analog and digital HF modes and highlight remaining confusions among similar signals and the importance of explainability. The work provides a scalable baseline for large-class RF signal classification in the HF band with practical implications for spectrum monitoring and cognitive radio.

Abstract

This paper presents a deep learning approach to the classification of 160 shortwave radio signals. It addresses the typical challenges of the shortwave spectrum, which are the large number of different signal types, the presence of various analog modulations and ionospheric propagation. As a classifier a deep convolutional neural network is used, that is trained to recognize 160 typical shortwave signal classes. The approach is blind and therefore does not require preknowledge or special preprocessing of the signal and no manual design of discriminative features for each signal class. The network is trained on a large number of synthetically generated signals and high quality recordings. Finally, the network is evaluated on real-world radio signals obtained from globally deployed receiver hardware and achieves up to 90% accuracy for an observation time of only 1 second.

Paper Structure

This paper contains 11 sections, 9 figures, 3 tables.

Table of Contents

  1. Introduction
  2. RF Signal Classification
  3. Challenges for Shortwave Signal Classification
  4. Related Work
  5. Large-Scale Signal Classification with Deep Learning
  6. This Work
  7. Training Data
  8. Neural Network and Training
  9. Real-World Test Data
  10. Classification Results
  11. Summary

Figures (9)

  • Figure 1: Signal classification automatically determines the type or mode of a received signal.
  • Figure 2: Exemplary part (here some 100 kHz) of the shortwave spectrum with a variety of different signal types densely packed and only loosely regulated.
  • Figure 3: Training data augmentation
  • Figure 4: Real-world accuracy and top-3 accuracy as average over all signal classes
  • Figure 5: Real-world accuracy for some exemplary digital modes
  • ...and 4 more figures