LED-PUF: Physical Unclonable LED Signatures For Unique Identification of IoT Nodes
Shab Naz, Ayesha Daniya Mulla, H Sumayya, Rifah Sameen Sarang, Kurian Polachan · 2024
We present LED-PUF, a novel Physical Unclonable Function (PUF) which derives signatures from the light patterns emitted by LEDs for identifying both the LEDs themselves and the nodes to which they are connected. LED-PUF operates on the hypothesis that LEDs exhibit unique light patterns at low brightness levels. Smartphone cameras can capture these patterns, and Convolutional Neural Networks (CNN) can effectively differentiate them, enabling the identification of LEDs. LED-PUF will find applications in IoT networks where identifying nodes is critical. Since LEDs are common in many IoT nodes, a PUF based on LEDs will require minimal hardware changes. Also, LED-PUF can be conveniently integrated into IoT nodes that use LEDs for Visible-Light Communication (VLC), repurposing LEDs used as transmitters for device identification. We validated LED-PUF by differentiating nine different LEDs from their images with a custom-designed CNN. The experiments yielded high identification accuracies of 96.4% during training and 94.2% during validation. Additionally, we generated PUF keys from the LED images using CNN. Subsequent analysis of their inter-hamming and intra-hamming distances, including the computation of false acceptance and rejection rates, validated the quality of the PUF.