Abstract
<jats:p>Abstract. This paper develops conceptual foundations for a wireless device identification method based on RF fingerprinting of their signals using Software-Defined Radio (SDR) receivers. It is shown that unique hardware imperfections of transmitters — carrier frequency offset (CFO), IQ imbalance, power amplifier nonlinearity, and phase noise — form a stable radio frequency fingerprint suitable for physical-layer device authentication without protocol modification. A comparative analysis of affordable SDR receivers (RTL-SDR V3, HackRF One, ADALM-PLUTO, LimeSDR Mini 2.0, BladeRF 2.0 Micro, USRP B210) is conducted based on key parameters: ADC resolution, maximum sample rate, frequency range, and cost. The system architecture is proposed that includes an SDR receiver, IQ sample preprocessing module, Siamese CNN-based neural network feature extractor, and a database of RF fingerprints of authorized devices. An algorithm is developed that supports device registration, real-time identification, and self-learning through incremental addition of reference embedding vectors. The choice of Siamese network is justified as a model that enables adding new device classes without full retraining. Analysis of literature results demonstrates that modern CNN architectures trained on IQ samples captured by SDR receivers achieve identification accuracy of 95–99% for sets of 10 to 100+ devices under sufficient signal-to-noise ratio conditions. Keywords: radio frequency fingerprint, RF fingerprinting, Software-Defined Radio, SDR, wireless device identification, IQ samples, convolutional neural network, Siamese network, physical layer authentication, Internet of Things, HackRF One, machine learning.</jats:p>