Publication

Abstract : Radio Frequency Fingerprinting (RFF) has emerged as a promising physical layer authentication technique that is capable of exploiting the inherent hardware impairments for uniquely identifying the wireless transmitters. In this study, a complete end-to-end RFF framework is proposed that synthesizes highly realistic device signals, constructs hybrid time and frequency feature representations, and also evaluates multiple Deep Learning architectures for the multi-device classification. Using QPSK and BPSK base-band modulation schemes, a dataset is generated for 20 RF devices, each modeled with distinct impairments that include Sampling Clock Offset, Carrier- Frequency Offset, Phase Noise, Static Phase Rotation, Non-linear Power Amplifier distortion, DC Offset, Stochastic Phase Noise, and I/Q imbalance. These signals are then segmented into a fixed length and normalized I/Q samples, and FFT-based spectral magnitudes are also extracted. The experimental result achieved an accuracy of 94.32% for the Deep Convolutional Neural Network (CNN) model, with BPSK, significantly outperforming other CNN variants. The ROC analysis further validates the superior discriminative power of Deep CNN. The proposed framework highlights the potential of a Deep CNN for reliable RF device identification, and it also establishes a robust foundation for future studies involving real-time RF captures and dynamic channel conditions for secure wireless authentication.