Physical layer secret key generation (PLKG) exploits wireless channel reciprocity to enable two legitimate parties to establish shared secret keys without relying on trusted third parties. However, most existing deep-learning-based PLKG schemes are designed or trained for specific channel environments, and their reciprocity enhancement capability can degrade significantly when deployed in unseen environments with different channel statistics, leading to severe performance loss under out-of-distribution (OOD) conditions. To address this issue, this paper proposes a PLKG scheme based on a mutual-information-driven mixture-of-experts autoencoder (MIMoAE). First, we design a mixture-of-experts encoder with specialized sub-networks and dynamic routing, enabling flexible adaptation to diverse channel environments. Then, a routing consistency loss is introduced to explicitly align the routing behavior between the two legitimate parties under distribution shift, preventing routing inconsistency. Finally, a nonlinear projection head is incorporated to stabilize mutual information estimation and provide more reliable gradient guidance. Compared with the state-of-the-art baseline, MIMoAE reduces the average key error rate by approximately 67% and achieves a raw key generation rate of about 3.85 bits/channel use on OOD test scenarios.
Physical layer secret key generation (PLKG) exploits wireless channel reciprocity to enable two legitimate parties to establish shared secret keys without relying on trusted third parties. However, most existing deep-learning-based PLKG schemes are designed or trained for specific channel environments, and their reciprocity enhancement capability can degrade significantly when deployed in unseen environments with different channel statistics, leading to severe performance loss under out-of-distribution (OOD) conditions. To address this issue, this paper proposes a PLKG scheme based on a mutual-information-driven mixture-of-experts autoencoder (MIMoAE). First, we design a mixture-of-experts encoder with specialized sub-networks and dynamic routing, enabling flexible adaptation to diverse channel environments. Then, a routing consistency loss is introduced to explicitly align the routing behavior between the two legitimate parties under distribution shift, preventing routing inconsistency. Finally, a nonlinear projection head is incorporated to stabilize mutual information estimation and provide more reliable gradient guidance. Compared with the state-of-the-art baseline, MIMoAE reduces the average key error rate by approximately 67% and achieves a raw key generation rate of about 3.85 bits/channel use on OOD test scenarios.
The reciprocity of wireless channels is a prerequisite for physical layer secret key generation (SKG). However, inherent factors, such as noise, asynchronous observations, and hardware impairments, disrupt the ideal reciprocity in the channel state information (CSI) observed by the two legitimate parties. To address this issue, we propose a mutual information-driven autoencoder (MIAE) architecture to extract reciprocal channel features from the non-ideal channel observations of legitimate parties. MIAE is constructed with an AutoEncoder Network (AENet) and a mutual information neural estimator (MINE). Specifically, AENet employs a structure with dual encoders and a shared decoder. The two encoders, integrated with convolutional block attention modules (CBAMs), are designed to focus on reciprocal features within CSI observations and to compensate for temporal variations caused by asynchronous observations. The shared encoder and the correspondingly designed loss function further concentrate the autoencoder’s reciprocity enhancement capability into the encoders. MINE is integrated into the proposed MIAE to estimate the mutual information between the channel features of the two parties. This estimation is then used to formulate a mutual information loss, which guides the encoders to learn channel features that closely match the optimal distribution, thereby boosting the key generation rate. Furthermore, a complete SKG scheme is designed based on the proposed channel feature extractor, MIAE. Simulation results show that our proposed MIAE can extract channel features with strong reciprocity and thus achieve excellent SKG performance based on the extracted features. The generalization performance of the proposed MIAE architecture for communication scenarios of different scales has also been examined.
The reciprocity of the wireless channel is the basis of the secret key generation on the physical layer as an emerging method for information security enhancement. Due to the duplex mode of communication systems, the channel responses detected by two connected legitimate nodes are not completely reciprocal. In this paper, an efficient physical layer secret key generation scheme is proposed based on an autoencoder to extract reciprocal features from weakly correlated channel estimates. The encoder part after training is equipped at both nodes to generate features for quantization in various channel environments. Simulation results show lower mean squared error (MSE) and better mutual information of the extracted features comparing to the most efficient linear generation method, the principal component analysis (PCA) based one. It reveals the great potential to generate symmetric secret keys with better performance, particularly in low SNR scenarios.
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.
The diverse application scenarios envisioned for sixth generation (6G) are characterized by the deep integration of sensing and ubiquitous connectivity, which imposes unprecedentedly stringent security requirements. However, due to the open nature of wireless channels, mobile communications systems always face severe information security threats such as falsification, spoofing, interception, and repudiation. Cryptography-based symmetric and asymmetric encryption techniques remain mainstream solutions for information protection. Symmetric encryption is efficient and secure for legitimate users but suffers from key-distribution difficulties over open wireless channels, whereas asymmetric encryption resolves this problem but faces increasing risks from quantum computing due to its reliance on structured mathematical hardness assumptions. In response to these limitations, physical layer security (PLS) has gained a great deal of research attention as a powerful security component that leverages the features of varying wireless channels. Existing PLS schemes can be broadly classified into two categories. The first one takes advantage of the legitimate link’s opportunistic channel-quality superiority over or different spatial-domain directions from the attacking link, which still faces many practical implementation difficulties. The second category is termed physical-layer key generation (PLKG). It extracts the unique features of the legitimate link’s channel variation, which is often reciprocal, as the source of secret key generation and therefore can naturally implement secure key distribution tasks. This advantage no doubt injects new vigor to symmetric encryption as a stronger protection approach. Following this trend, we in this paper concentrate on the PLKG techniques. Specifically, we present a comprehensive overview on existing PLKG schemes, discussing diverse secret key generation and reconciliation methods. We further investigate the model-driven and deep-learning-based approaches tailored for the scenario with imperfect channel reciprocity between the sender and receiver. After comprehensively reviewing major existing schemes, we further discuss a recently proposed PLKG design based on codeword reconstruction, which makes use of the strong error-correcting capability of the forward-error-correction (FEC) codes to effectively implement secure and consistent secret key generation between the legitimate sender and receiver. Finally, we share our opinions on the unsolved challenges and potential research directions dedicated to PLKG toward meeting the security requirements of 6G.
In this paper, we propose a denoising autoencoder (DAE)-assisted secret key generation (SKG) scheme, designed to improve the reliability and efficiency of physical-layer secret key generation. Our approach leverages the capability of DAEs to learn denoised channel representations while suppressing additive noise, thus reconstructing channel features that are more strongly correlated between the legitimate nodes. By mitigating reciprocity imperfections, the proposed scheme reduces the key disagreement rate (KDR) and improves the key generation rate (KGR). Furthermore, our extensive simulation results confirm that the DAE-assisted SKG outperforms the transform-based and learning-based baselines, including discrete wavelet transform, principal component analysis, discrete cosine transform, traditional autoencoder (AE)-based approaches, convolutional neural network (CNN)-based approaches, convergence-learning fully connected neural network, and mutual-information-driven AE. The NIST SP 800-22 results confirm that the generated keys successfully pass the suite of statistical randomness tests after privacy amplification. These results indicate that the proposed DAE-assisted framework is an effective preprocessing approach for empirical SKG under the adopted threat model and assumptions, with moderate online complexity relative to the learning-based baselines.
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.