As factories in the Industrial Internet of Things (IIoT) scale up and complexity increases, distributed Large Language Models (LLMs) enabled Digital Twin Network (DTN) is required to enhance real-time mapping capabilities. However, current wired communication processes in distributed LLMs fine-tuning cannot meet the demands of mobile IIoT scenarios. Confronting this thorny problem, we propose a wireless communication strategy for gradient synchronization to enhance the efficiency of distributed LLM-enabled DTN fine-tuning. Specifically, leveraging mobile edge intelligence, we jointly optimize central node closeness centrality and computational capacity, thereby enhancing the efficiency of distributed LLM fine-tuning. Subsequently, we design the Dueling Double Deep Q-Network (D3QN) enabled parameter fine-tuning gradient synchronization (D3QN-REFRESH) algorithm and analyze its complexity. Extensive simulations using real factory data demonstrate the superior performance and generalization capabilities of D3QN-REFRESH.

Physical layer authentication (PLA) as a promising solution has gained widespread attention due to its high security and lightweight deployment. A fixed authentication center is vulnerable to attacks, making both centralized PLA (CPLA) with a single authentication center and collaborative PLA with multiple collaboration nodes at risk of a single point of failure. In this paper, we propose a leader election algorithm for distributed physical layer authentication, where multiple receivers are col-laboratively trained under a leader and complete authentication independently. Specifically, the one with the strongest model generalization ability is elected as the leader by individual receivers voting based on data quality. The role of the leader is to filter out the underperforming receivers, assign reasonable weights to local models, and construct an authentication white list to achieve better authentication performance. Simulation results show that the proposed scheme outperforms the randomly selected leader and the traditional PLA schemes.

In the context of the industrial Internet of Things (IIoT), communication devices are typically mobile, increasing the complexity and diversity of channels due to metal device occlusion. A crucial aspect of this intricate environment is the development of an authentication scheme based on physical layer channel characteristics. One approach to achieving this is through deep learning, which is a hot topic in physical layer authentication. However, designing a network that is suitable for channel classification tasks and establishing a reasonable training procedure that leads to high authentication accuracy can be challenging. To address the physical layer authentication of mobile devices in IIoT, we implement ResNet to extract channel features of Channel State Information (CSI) from different transmitters and classify them at the network output layer, enabling authentication decisions based on classification results. To improve accuracy and speed up network convergence, we utilize the exponentially averaging data augmentation algorithm and parameter-based transfer learning strategy during the training procedure. Simulation results demonstrate that multi-user physical layer authentication based on ResNet can achieve higher authentication accuracy as the number of network layers increases. The data augmentation and transfer learning are proved to improve the authentication accuracy. Numerical results on NIST industrial datasets reveal that the authentication scheme based on ResNet50 can achieve 99.64% authentication accuracy in scenarios with four users present, which is 32.68% higher than existing algorithm.

Collaborative physical layer authentication (CPLA) is a promising alternative, addressing common single-point failure issues in centralized authentication systems through its unique architecture. However, the necessary involvement of multiple parties increases the risk to collaborative systems, particularly from hostile cooperators, significantly impacting the performance of CPLA. In existing CPLA approaches, the most common strategy to combat malicious cooperators attacks is to select the best collaborative combination. This strategy achieves the customization goal by excluding hostile-minded devices. However, processing a non-fixed search space typically demands a substantial investment of time and resources. As a remedy, we propose a decision-level-based CPLA scheme with a weighted voting mechanism. Our scheme aims to implement streamlined and effective dynamic management of cooperators to ensure that multi-directional information provides positive effects on authentication. Specifically, we conduct a two-stage performance appraisal of all cooperators. To measure the trustworthiness of cooperators, an impression-driven reliability evaluation scheme is developed. We analyze the riskiness of individual cooperators to prevent centers from falling into cognitive blind spots. Finally, we validate the feasibility of the scheme. The results demonstrate that, in a scenario where 50% of participants are malicious, our approach achieves an accuracy improvement of 2.96% to 3% compared to other dynamic weighted voting schemes. The robustness and stability of the proposed CPLA scheme outperform the benchmark schemes.

Collaborative physical layer authentication (CPLA), which leverages spatial diversity, holds promise for enhancing the performance of feature-based physical layer authentication. However, some existing CPLA schemes simply aggregate the local information of collaborators to make final judgments and rarely consider the involvement of malicious collaborators. In this paper, we propose an impression-weighted based local decision aggregation scheme for detecting spoofing attacks in the presence of malicious collaborators. Specifically, the authenticator continually evaluates the authentication capabilities of collaborators by verifying the accuracy of local decisions and then synthesizes their long-term capabilities into impressions using a fuzzy membership function. These impression values will be dynamically updated upon completion of each authentication task. Moreover, a reinforcement learning scheme is employed to find the optimal threshold for authentication in a dynamic environment. Simulation results validate the high robustness and effectiveness of our proposed approach, guaranteeing the CPLA system's reliable operation.

Machine learning (ML)-based physical layer authentication (PLA) has attracted much attention since neural networks can be constructed to identify channel characteristics in complex wireless environments. This enables high-authentication performance and lightweight deployment in the Internet of Things (IoTs). Due to the booming growth of IoT connections, the workload of the central authenticator increases significantly. As a result, resource-constrained terminals are unable to independently handle the computationally complex ML task. Therefore, cooperative PLA (CoPLA), which introduces multiple supervised nodes as task-sharing entities, is emerged as a promising solution to address this concern. However, in existing CoPLA studies, some critical issues have been overlooked. For example, the consideration of which collaborative nodes are eligible or best suited for cooperation to maximize the authentication gains. Moreover, the security threats posed by untrusted collaborators are equally challenging. In this article, we propose a federated learning (FL)-based CoPLA scheme that utilizes a group of edge devices to jointly build an authenticator. This ensures privacy preservation and higher robustness. To figure out the optimal collaborator selection in CoPLA, an adaptive search procedure via reinforcement learning (RL) is customized. Furthermore, we introduce a lightweight reputation estimation method to evaluate each collaborator’s credibility, thereby uncovering underperforming devices or hidden internal attackers. Finally, simulations and real-world experiments are carried out. The results show that the authentication accuracy of our scheme is 9.52% higher than that of blind cooperation. And, it outperforms other existing CoPLA schemes in terms of time efficiency and robustness.

Physical layer authentication as a promising complement for upper layer authentication is the first line of defense against malicious attacks in wireless communication. However, the smart spoofer can learn the rules from receiver’s authentication process and dynamically choose the proper time sending spoofing signal which poses a severe threat to wireless communications. According to this, the Stackelberg game-based physical layer authentication strategy is proposed in this paper to model the interactions between the receiver and the smart spoofer. We first consider the static game-based authentication under the worse condition that the smart spoofer acts as the leader with privilege over the receiver. Moreover the Stackelberg equilibrium of static authentication strategy is derived. Then, we propose a dynamic game-based strategy according to reinforcement learning technique named Policy Hill Climbing, in which the spoofer always choose equilibrium solution and the receiver is unaware of the system parameters, such as the channel timevarying coefficient. Simulation results are presented to validate the effectiveness of the proposed authentication strategy, and the Policy Hill Climbing algorithm improves the utility compared with Q-learning-based algorithm.