Radio Signal Augmentation for Improved Training of a Convolutional Neural Network
Abstract
This technical report presents the findings of an experiment to evaluate the effectiveness of our technique for improving the accuracy of identifying which type of digital modulation is present in a sample of radio signal data. We use a convolutional neural network (CNN) to identify the modulation type from raw digitized radio signal input. The CNN is trained using our technique of dataset augmentation, which applies a transformation specific to the sensory domain of radio (and potentially, closely related signal types). This augmentation simulates a receivers clock offset or error. Digital radio signal receivers will have a clock frequency slightly different than the transmitter, even if each is tuned to the same frequency. This is usually accounted for in the receiver design, referred to as carrier clock recovery, since it is designed for a known signal type. Our method is to apply varying amounts of clock frequency offset to a training dataset, and use it to train the machine learning algorithm (in this case, a CNN). The trained CNN model is compared to a baseline model in which no clock offset was used during training. Classification performance increases to nearly 100 when trained with frequency offset, compared to the baseline of 58 . Two real-world signals were captured from car remote keyless entry fobs. These signals contain an unknown receiver clock offset. The network trained with our method classified nearly 100 of the samples correctly, while the baseline network did not correctly identify the on-off keying (OOK) modulation. A recommended action is to further investigate dataset augmentation, especially in the domain of radio signals. This domain could benefit from very specific, but very useful, transforms to further improve performance of machine learning techniques.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2016
- Accession Number
- AD1026932
Entities
People
- Benjamin Migliori
- Daniel Gebhardt
- Logan Straatemeier
- Michael Walton
Organizations
- Naval Information Warfare Systems Command