RF Fingerprinting of IoT Multi-Modulation Wireless Devices

Deep learning-based RF fingerprinting is a recent research area in wireless transmitter authentication. This paper presents results demonstrating the use of convolutional neural networks (CNNs) to classify RF transmissions of an IoT multimodulation wireless device. The model is trained on I/Q baseband samples captured by a software-defined radio. Results show that each unique combination of wireless device and physical layer corresponds to a distinct class, allowing differentiation based on device and physical layer parameters. Experimental results demonstrate classification accuracy of $88 \%$ at an SNR of 6 dB and $\mathbf{9 8 \%}$ at $\mathbf{1 8 \mathrm {dB} \text {, indicating the methodprimes ability to differentiate}}$ the pair (wireless device, physical layer).