Research And System Implementation Of NFC Relay Attack Detection Scheme Based On Machine Learning

As a radio frequency identification(RFID)technology,RF/NFC technology is widely used in smart cards,mobile payments,access control or passive keyless entry systems(PKES)due to its convenience and high security.However,in an open wireless channel,the transmitted wireless signals can be easily eavesdropped or relayed.NFC is vulnerable to relay attacks which poses threat to the security of NFC.Relay attacks simply forward the transmitted signals and do not involve analyzing or modifying the transmitted data,Relay attacks do not involve analyzing or modifying the transmitted data,making it intractable for cryptography-based data encryption or other security means in the application layer.In addition during a relay attack,both authentic devices are unaware of the attacker,making detection of such attacks intractable.Due to widespread and extensive NFC applications such as mobile payment or PKES,successfully conducted relay attacks will result in property loss.To defend relay attacks,the mainstream countermeasures proposed in previous works focused on distance bounding protocols and ambient-based methods.Radio frequency fingerprinting,which identifies specific transmitting devices by utilizing their unique hardware imperfections,is a promising method for detecting and identifying relay attacks since additional relaying devices are used.Recent studies have seen the negative impact of channels on wireless RF fingerprinting that result in generalization problems,drawing practicability concerns of deep learning models.This paper focuses on the NFC protocol of the ISO/IEC 14443-A standard and proposes an NFC relay attack detection and identification method based on machine learning and wireless RF fingerprinting.The main contributions of this paper are as follows:1.This study proposes a method for effectively detecting and identifying NFC relay attacks based on machine learning and RF fingerprinting by analyzing the waveform of transmitted wireless signals.Due to the lack of required standard open datasets in the current research field,this thesis work designed and implemented the devices for simulating and conducting wired and wireless NFC relay attacks,created the data set for relay attack detection and identification and finally,made the data set public.2.This thesis work designed and trained a convolutional neural network(CNN)classifier to identify and classify the waveforms of physical layer transmission signals.The model achieved a high accuracy of 99%,and the thesis demonstrated that when using relay devices other than real NFC tags,the CNN model can effectively distinguish between normal and relayed NFC signals and thus detect relay attacks.3.To address the negative impact of channel on deep-learning-based wireless signal fingerprinting found in recent studies which have resulted in model performance degradation,this paper leverages federated learning and constructs a training framework to train federated models on diverse data from multiple channel conditions and improved the accuracy and generalization ability of the model.In addition,quantization operation is used to reduce the communication overhead,making the proposed relay attack detection scheme more suitable for NFC application scenarios.