Meta extreme learning machine based performance analysis for IoT device identification using RF fingerprint

The IoT revolution has led to exponential growth in wireless devices, creating intricate communication networks. Ensuring the security of these networks is crucial to prevent unauthorized access and attacks. However, IoT devices have limited resources, making it challenging to implement customized security measures. To address this, RF fingerprints, which capture the unique hardware characteristics of devices, are used for identification and authentication. In this study, a system was developed using cost-effective software-defined radio (SDR), a single-board computer, and open-source software to capture and record RF signals. Features of the signals were extracted by identifying the transient region and applying the Hilbert Transform (HT) to obtain Instantaneous Amplitude (IA), Instantaneous Phase (IP), and Instantaneous Frequency (IF) values. Rather than traditional statistical methods, the feature dimension was reduced using the Minimum Redundancy Maximum Relevance (MRMR) technique. For IoT device identification, classifiers such as ELM-based Meta-ELM, Multilayer Meta-ELM (ML-Meta-ELM), and Constrained Mixed Meta-ELM (CM-Meta-ELM) network structures were used, along with ELM-based learning algorithms. The ML-Meta-ELM network achieved an 88% accuracy in distinguishing devices, while the Meta-ELM network achieved around 90%. The CM-Meta-ELM network demonstrated the highest performance with a 92% accuracy rate.