AN OPEN-SOURCED TIME-FREQUENCY DOMAIN RF CLASSIFICATION FRAMEWORK

In this paper we present a machine learning-based approach to solving the radio-frequency (RF) signal classification problem in a data-driven way. To this end, we propose an efficient and easy-to-use graphical user interface (GUI) for researchers to collect their own data to build a customized RF classification system. The GUI operates in the time-frequency (TF) domain, which is achieved by applying short-time Fourier transform to the in-phase and quadrature (IQ) time domain signals. Using the proposed GUI, a radio frequency (RF) dataset is collected from the ultra high frequency industrial, scientific, and medical (ISM) bands using commercial-off-the-shelf (COTS) transceivers, and COTS transceiver modules. We train three different variants of convolutional neural network models, such as VGG and ResNet, using the collected dataset and show that they can perform acceptable test-time classification (up to 95% accuracy) on unseen real-world RF signal recordings. Our experimental results also show that a carefully prepared TF domain without a loss of information can achieve better performance than a magnitude-only representation that loses phase information during the TF transformation. We open-source our project to provide the public with access to the labeled datasets, programming code, and the GUI software that can expedite the labeling process.