Radio frequency impairments in transmitters : characterising power amplifiers for radio frequency fingerprinting identification

Abstract

The explosive growth of wireless connectivity brings with it significant cybersecurity challenges. In response, there have been some research efforts on radio frequency fingerprinting identification (RFFI) technology that has become one of the promising physical (PHY) layer wireless security solutions. The radio frequency fingerprinting (RFF) refers to the unique fingerprint/characteristics inherently presented in analogue transmit RF chains. These distinctive RFF features are extracted and used as device identities (IDs) to facilitate secure network access in the authentication processes. They introduce minor but unique distortions to the transmitted signal waveforms, detectable by receivers. Compared to traditional authentication methods, RFF does not rely on shared secrets like passwords or cryptographic keys, as they can be compromised through various attacks. The RFF features in an RF transmitter chain can arise from various components, e.g., RF oscillator, digital-to-analogue converter (DAC), RF power amplifier (PA), and antenna. Given that PAs are major contributors to signal non-linearity, they are extensively studied for RFF applications. However, achieving high classification performance in low SNR regions remains a challenge. The main objective of this research is the development of the RFFI schemes based on non-linear features of the PA. In our proposed RFFI schemes, we first exploit the unique non-linear memory effect of the transmitter RF chains, which consist of matched pulse shaping filters and non-linear PAs. We also introduce a hybrid classification method that can significantly enhance the classification performance while at the low SNR region. However, it remains a challenge to classify the device under test (DUT) with the same model/ or with similar RFF features. This is expected as the differences among the same models are more minute, compared to those of different models. To overcome this limitation, we devise a strategy that involves deliberately and random adjustments to the operating conditions of the PA, specifically using the active load-pulling technique to modify the output impedance of the PAs. Upon experimental observation, the non-linear characteristics of the PA have been significantly altered. This strategy results in improved classification performances, particularly in low SNR scenarios, when classifying the PA of the same model. Furthermore, this thesis explores the RFF features in a RF multi-antenna array transmitter chain. A key aspect of this exploration is the mutual coupling (MC) effect, which is caused by electromagnetic interactions between adajacent antenna elements in the antenna array, and this effect may impact the performance of the PA. In the multi-antenna RF chain, a reconfigurable power divider (RPD) is employed as a feeding network that can distribute the signal to each antenna element, which, in this work, is also utilised to control the MC effect in the RF chain. This approach results in varied behaviours of the non-linear RFF features in the RF transmitter chain. Through these methodologies, the distinctions in RFF features across different models and even the same model of wireless devices are more pronounced.