Physical layer key generation technique extracts shared keys from observed signals of legitimate communication parties by exploiting the randomness, reciprocity and spatial decorrelation of wireless channels. Most existing key generation methods rely only on instantaneous sampling values, failing to effectively extract temporal dynamic features and adapt to slow-varying channel scenarios. In addition, the observation discrepancy of channel states at both ends caused by non-ideal reciprocal channels leads to an increased key disagreement rate after quantization. To address these issues, a physical layer key generation method based on a joint temporal prediction network architecture is proposed. First, a prediction network based on a gated recurrent unit encoder was constructed and trained with the channel states of both the transmitter and receiver at future moments as the prediction target, yielding a base model capable of characterizing the common channel evolution law of both parties. On this basis, the feature discrepancy between the two parties was further introduced as the loss function to fine-tune the network, guiding the features of both sides toward consistency. Finally, to reduce the key disagreement caused by quantization, the quantization error was incorporated as a new loss function for a second round of fine-tuning, so that the network takes subsequent quantization into account during feature extraction, thereby lowering the key disagreement rate. Simulation results show that over the entire SNR range, the proposed method reduces the key disagreement rate by more than 58% compared with amplitude/phase quantization-based physical layer key generation methods, and by more than 23% compared with existing deep learning-based physical layer key generation methods.
Physical layer key generation technique extracts shared keys from observed signals of legitimate communication parties by exploiting the randomness, reciprocity and spatial decorrelation of wireless channels. Most existing key generation methods rely only on instantaneous sampling values, failing to effectively extract temporal dynamic features and adapt to slow-varying channel scenarios. In addition, the observation discrepancy of channel states at both ends caused by non-ideal reciprocal channels leads to an increased key disagreement rate after quantization. To address these issues, a physical layer key generation method based on a joint temporal prediction network architecture is proposed. First, a prediction network based on a gated recurrent unit encoder was constructed and trained with the channel states of both the transmitter and receiver at future moments as the prediction target, yielding a base model capable of characterizing the common channel evolution law of both parties. On this basis, the feature discrepancy between the two parties was further introduced as the loss function to fine-tune the network, guiding the features of both sides toward consistency. Finally, to reduce the key disagreement caused by quantization, the quantization error was incorporated as a new loss function for a second round of fine-tuning, so that the network takes subsequent quantization into account during feature extraction, thereby lowering the key disagreement rate. Simulation results show that over the entire SNR range, the proposed method reduces the key disagreement rate by more than 58% compared with amplitude/phase quantization-based physical layer key generation methods, and by more than 23% compared with existing deep learning-based physical layer key generation methods.
Physical layer encryption is a promising solution to provide transmission security for Internet-of-Things (IoT). To realize physical layer encryption, channel state information (CSI) as a random source to generate secret keys should be exploited. The reciprocity of wireless channel between legitimate communication nodes which is typically assumed for time division duplex (TDD) systems, is of critical importance to ensure the consistency in generated keys at two nodes. However, in practical TDD systems, due to the asynchronous measurement of CSI, the receiver noise and the the difference in hardware, etc., the reciprocity of channel measurements between legitimate parties is hard to be guaranteed, which makes physical layer key generation approach infeasible. To address this issue, we propose a wireless key generation scheme based on deep learning techniques. Firstly, a feature extraction network, which consists of two auto-encoders trained jointly, is developed to extract the common feature from the different yet correlated channel observations at two legitimate nodes, which will be used as the source to generate keys. Then a quantizer with an adaptive guard band width is devised to output a binary secret-key sequence from the feature vector. Numerical results show that the proposed scheme achieves remarkable improvement in terms of key disagreement rate (KDR) and key generation rate (KGR) compared existing counterparts in the literature.
Physical-layer secret key generation (PLKG) is a new research area that has emerged in recent years. It is aimed at scenarios where legitimate IoT devices communicate directly, interacting with confidential information for lower overhead and higher security by using wireless channel. When applying it to wireless feature extraction, noise removal is not taken into account in current deep learning networks. To address these problems, the PLKG scheme based on bidirectional convergence feature learning convolutional network (BCFL-based scheme) is proposed, which consists of neural network called BCFL and a new quantization method to achieve better secret key generation. Unlike existing PLKG schemes that enabling both parties to communicate for obtaining higher channel feature similarities, when training it, channel state information (CSI) obtained by channel estimation for two legitimate devices during coherent time is used as inputs; and mean square error (MSE) between two outputs is used as a result of loss function for iterative training. Thus, it can obtain better denoising ability with guaranteed low computational resource consumption, and two legitimate devices can obtain highly correlated channel features. Multiple quantization method is also proposed to address low secret key generation rate (KGR) and low-secret key randomness (KR). The results show that the proposed BCFL-based scheme has a lower MSE than other schemes in different scenarios, indicating that it has better capability to learn channel reciprocity; and secret key error rate (KER) and time consumption are only about 50% of other schemes, which is a significant performance improvement.
Physical layer encryption is an emerging security paradigm which supplements the higher-layer encryption solutions. To realize physical layer encryption, channel measurement is typically utilized as a randomness source to generate secret keys. Due to non-perfect reciprocity of the channels and the inevitable channel estimation errors resulting from noises and interferences, the channel measurements at the legitimate transceivers are not the same, yielding a mismatch in the generated keys. To address this issue, a distributed physical layer key generation scheme based on deep learning is proposed in this paper. A channel state information (CSI) learning neural network (CLNet) based on the autoencoder is deployed at both the transmitter and the receiver side to refine the initial channel estimation results. The CLNet takes the least square (LS) channel estimations as input and outputs the channel estimations more similar to the real channel. To train this CLNet, a channel equalization module together with a pre-trained maximum-likelihood (ML) detection network is deployed, which is used to make a decision on the transmitted pilot signal. Since the pilot is known as a priori, the cross entropy loss between the known pilot and the recovered one can be calculated, which implicitly indicates the accuracy of the refined CSI. Further, the weighted sum of the aforementioned cross entropy loss and the mean square loss between the original and the refined CSI estimates is utilized for training, which not only improves the CSI estimation quality but also avoids the over-fitting effect. Extensive simulation results show that compared with benchmark schemes, the proposed scheme improves the key agreement rate and the key generation rate at the legitimate users. Moreover, our method ensures a higher key disagreement rate between the legitimate user and the eavesdropper as well, especially in the low signal-to-noise ratio (SNR) regime.
Using the physical layer channel information of wireless devices to establish the highly consistent secret keys is a promising technology for improving the security of wireless networks. Nevertheless, in the time division duplex system, the reciprocity of the wireless channel that is the basic principle of key generation is impaired by nonsimultaneous sampling and noise factors. Existing physical layer key generation approaches rely on hand-crafted feature extraction algorithms, which have high overhead or security issues and are impractical in real-world situations. This paper presents a novel physical layer key generation method to extract highly consistent keys from imperfect channel responses, which exploits channel reciprocity through deep learning. Specifically, we first design the Channel Reciprocity Learning Net (CRLNet), a neural network for efficiently learning channel reciprocity features from the wireless channel in TDD OFDM systems. Later, a new key generation scheme based on CRLNet is developed that can achieve a high key agreement rate. Experiments indicate that the CRLNet-based key generation scheme performs excellently in terms of key generation rate, key error rate, and randomness, confirming that our method has better performance and lower overhead than existing methods.
Physical-layer key generation exploits the randomness and unpredictability of wireless channels to enhance the security of wireless communications. However, factors such as asynchronous measurements, channel noise, and hardware impairments undermine the ideal reciprocity of channel state information (CSI) observed by legitimate users, thereby limiting key generation performance. To address this issue, this paper proposes a jointly optimized contrastive learning network (JOCLNet), which consists of a joint learning model and a contrastive learning model. The joint learning model, built upon a convolutional neural network (CNN) and a feedforward neural network (FNN), separates non-reciprocal components such as noise and measurement errors from the raw CSI of legitimate users, while extracting reciprocal components for key generation, thus improving robustness in noisy environments. To overcome the inability of existing deep-learning-based SKG schemes to resist passive eavesdropping attacks, the contrastive learning model introduces a contrastive loss on top of the joint learning model. This enables a dual objective: enhancing reciprocity between the processed CSI of legitimate users while ensuring that the processed data of an eavesdropper remains highly uncorrelated with that of legitimate parties. Furthermore, a complete key generation scheme is designed based on the proposed JOCLNet. Experimental results demonstrate that the proposed scheme achieves strong resistance to eavesdropping and robust adaptability to noise.
Channel reciprocity is the foundation for physical layer key generation, which is influenced by noise, hardware impairments, and synchronization offsets. Weak channel reciprocity will result in a high key disagreement rate (KDR). The existing solutions for improving channel reciprocity cannot achieve satisfactory performance improvements. Furthermore, the existing quantization algorithms generally use one-dimensional channel features to quantize and generate secret keys, which cannot fully utilize channel information. The multidimensional vector quantization technique also needs to improve in terms of randomness and time complexity. This paper proposes a physical layer key generation scheme based on deep learning and balanced vector quantization. Specifically, we build a channel reciprocity compensation network (CRCNet) to learn the mapping relationship between Alice and Bob’s channel measurements. Alice compensates for channel measurements via a trained CRCNet to reduce channel measurement errors between legitimate users and enhance channel reciprocity. We also propose a balanced vector quantization algorithm based on integer linear programming (ILP-BVQ). ILP-BVQ reduces the time complexity of quantization on the basis of ensuring key randomness and a low KDR. Simulation results showed that the proposed CRCNet performs better in terms of channel reciprocity and KDR, while the proposed ILP-BVQ algorithm improves time consumption and key randomness.