Uniquely Identifying Device Fingerprint via Deep Learning over Ethernet Communication Signals

The widespread adoption of electronic devices has made imitation attacks a significant security concern, especially in the context of Ethernet communication. In this paper, a deep learning model is trained by using data from both received and transmitted Ethernet communication signals of devices to extract unique fingerprints and employs this model to provide system security. To accomplish this, we initially conducted a theoretical analysis of the LAN9514 Ethernet integrated circuit, followed by the acquisition of Ethernet communication signals from two Raspberry Pi 3 devices with identical models and software through an Analog-Digital Converter (ADC). These signals were then processed and labeled on CUDA cores, resulting in the creation of a dataset from analog signals. When this dataset was trained using the LSTM algorithm, the system exhibited a %98 accuracy rate.