In Automatic Modulation Classification (AMC), deep learning methods have shown remarkable performance, offering significant advantages over traditional approaches and demonstrating their vast potential. Nevertheless, notable drawbacks, particularly in their high demands for storage, computational resources, and large-scale labeled data, limit their practical application in real-world scenarios. To tackle this issue, this paper proposes an automatic modulation classification model based on the Adaptive Lightweight Wavelet Neural Network (ALWNN) and the few-shot framework (MALWNN). The ALWNN model, by integrating the adaptive wavelet neural network and depth separable convolution, reduces the number of model parameters and computational complexity. The MALWNN framework, using ALWNN as an encoder and incorporating prototype network technology, decreases the model's dependence on the quantity of samples. Simulation results indicate that this model performs remarkably well on mainstream datasets. Moreover, in terms of floating point operations per second (FLOPS) and normalized multiply-accumulate complexity (NMACC), ALWNN significantly reduces computational complexity compared to existing methods. This is further validated by real-world system tests on Universal Software Radio Peripheral (USRP) and Raspberry Pi platforms. Experiments show MALWNN's superior performance in few-shot learning scenarios compared to other algorithms.

Fine-tuning large language models (LLMs) is critical for adapting pretrained models to specialized downstream tasks. Federated LLM fine-tuning enables privacy-aware model updates by allowing data owners (DOs) to contribute a global LLM without exposing local data. However, full-parameter fine-tuning in federated settings incurs significant computational and communication overhead, while frequent gradient exchanges increase the risk of privacy leakage, such as memorized data inference. Parameter-efficient fine-tuning (PEFT) with differential privacy (DP) offers a low-overhead alternative with formal privacy guarantees, but fails to strike privacy-utility tradeoff under heterogeneous privacy preferences: individual DOs may inject excessive DP noise to maximize privacy, whereas the curator aims to minimize noise to preserve model quality. In this paper, we present an innovative game-theoretical framework that enables dynamic privacy trading within differentially private federated LLM fine-tuning. In the game, DOs strategically adjust their local DP noise levels in exchange for customized incentives from the curator, thereby balancing privacy and utility. We begin by establishing a theoretical convergence bound that quantifies the influence of locally injected noise on the global model utility. Under this bound, we analytically characterize the pure-strategy Nash equilibrium of the game, accounting for DO heterogeneity, curator budget constraints, and noise estimation errors. For mixed-strategy settings with incomplete information, we design a hierarchical reinforcement learning algorithm that jointly learns DOs’ optimal noise-saving strategies and the curator’s optimal pricing policy without presupposing their private information. Experiments on real-world datasets demonstrate that the proposed scheme improves DO utility, reduces curator cost, mitigates free-riding, and accelerates convergence compared to existing methods.

The open and zero-trust nature of the heterogeneous low-altitude intelligence network requires more stringent secure authentication that cannot be meet with conventional schemes, due to the static authorization misalignment, long-validity token infiltration risk, and single-factor credential ossification. To address these challenges, this study proposes a blockchain-based cross-domain authentication scheme. We first develop a blockchain-enabled secure cross-domain registration and information management architecture incorporating a dual-index data structure for efficient historical query operations. Unmanned aerial vehicles (UAVs) achieve cross-domain registration through blockchain-based secure interactions with target domain trusted authorities (TAs). A cross-domain authentication protocol integrating physical unclonable function (PUF) and hash-based signature technique is designed, for mutual authentication. The TA generates time-limited cross-domain tokens with restricted communication attempts for UAVs, which subsequently establish negotiated session keys with base stations for secure resource sharing. To enhance security dynamics, both parties update temporary identity information and prepare fresh authentication keys during each token request cycle. The TA delegates token-updating random factors to base stations to ensure secure token renewal. Additionally, as the blockchain records the hash values of each token round, TA can detect if internal attackers have tampered with the token state. The security analysis and experiments demonstrate the advantages of our scheme.

Effective recognition of jamming in a communication system is essential to maintain the integrity of the electromagnetic spectrum space. In this paper, a novel feature-enhanced open-set jamming pattern recognition method (FOSR) is proposed. First, an in-phase and quadrature (I/Q) data feature enhancement module is designed based on a complex-valued autoencoder to capture the interaction features between the I and Q channels. Then, a jamming feature extraction module is designed to extract jamming characteristics for known patterns by integrating the raw I/Q data with their interaction features. Subsequently, an adaptive threshold open-set classification module is proposed to recognize both known and unknown patterns. Finally, to address the domain shift problem, we extend FOSR with a domain adaptation (DA) module based on distribution alignment and classifier calibration, referred to as FOSR-DA. Simulation results show that the proposed method achieves superior recognition accuracy and exhibits strong robustness when dealing with the domain shift problem.

Distributed spectrum allocation for large-scale UAV swarm remains a challenging issue, due to spectrum allocation collisions and the high communication overhead required to reach consensus. To address these challenges, we propose a lightweight consensus protocol for distributed collision-free spectrum allocation (LCCFSA), where UAV nodes in the swarm form a blockchain and spectrum allocation consensus is reached on the chain. Specifically, a fast low-complexity allocation scheme is developed for each UAV based on an interference graph, where each UAV adaptively adjusts its occupancy area to avoid mutual interference. To further reduce the consensus overhead, we design a lightweight consensus protocol with a transaction-based blockchain ledger and provide a formal security analysis of the proposed protocol. A prototype is built to validate the feasibility of the proposed scheme. Simulation results show that the average consensus latency can be reduced by more than 20% in scenarios with 100 consensus nodes.

Satellite communications have been considered a key part of global connectivity, effectively supporting diverse applications such as the Internet of Things (IoT) and real-time communication services. However, security-sensitive devices face significant challenges due to the threat of eavesdropping satellites, which compromise data confidentiality. Existing approaches often rely on deterministic models and fail to account for the stochastic nature of eavesdropping threats and the dynamic demands of satellite networks, limiting their applicability in practical scenarios. To address these challenges, this work proposes a novel secure access strategy for satellite mobile edge computing (SatMEC) systems, integrating a stochastic risk assessment model and an evolutionary game-theoretic framework. The proposed solution leverages a probabilistic model to evaluate the spatial distribution of eavesdropping satellites, quantifies the eavesdropping risk via the concept of eavesdropping capacity, and incorporates a dynamic service selection strategy that balances secrecy capacity and queuing delay.Furthermore, a distributed algorithm is developed to enable IoT devices to select service satellites based on real-time utility optimization adaptively. Extensive simulation experiments validate the effectiveness of the proposed strategy, demonstrating its ability to improve system security, balance the network load, and enhance overall performance in large-scale and dynamic satellite network environments. The results highlight the reliability and scalability of the proposed solution, making it a practical approach for secure and efficient access in LEO satellite networks.

To address the issues in spectrum sharing systems, such as unreliable sensing results because of the data falsification and unfair access allocation, a spectrum sharing model based on the sensing quality was proposed. By jointly optimizing sensing quality and access resource allocation, this model improved efficiency and fairness. In the sensing phase, a leader-follower game modeled the interaction between sensing users and demanders: the sensing users optimized sensing quality to maximize their benefits, while the demanders determined data purchases based on sensing quality, ensuring efficient and reliable data collection. In the access phase, spectrum resources were dynamically allocated based on the sensing quality, with the VCG (Vickrey-Clarke-Groves) mechanism ensuring truthful reporting and preventing data falsification and fraud. This led to fair and optimized spectrum allocation, enhancing the overall system efficiency. Simulation results show that the proposed mechanism effectively incentivizes users to report the true sensing quality and protects the honest users.

With the rapid development of low-altitude economy, the low-altitude intelligent network (LAIN) serves as a critical infrastructure for supporting diversified aerial activities. However, current LAIN lacks a physical-network synchronization mechanism and fails to handle heterogeneity at the architecture level, which leads to difficulties in real-time adaptive adjustments and efficient unified coordination. To address these issues, this paper proposes the networked embodied intelligence (NEI) paradigm, with a network-level sensing-decision-action-feedback (SDAF) closed-loop mechanism to synchronize physical and network states. Building on this paradigm, we functionally reconfigure LAIN into four collaborative subnetworks: sensing, computing, communication and navigation. As a further step, we propose the NEI-LAIN architecture, where these subnetworks collaborate via the SDAF loop to achieve global collaboration and continuous evolution. Simulation results demonstrate that the proposed NEI-LAIN can significantly enhance the performance of communication robustness, resource utilization and task responsiveness in highly dynamic scenarios. Finally, we discuss its implementation challenges and future research directions.

The emerging increasingly sophisticated, intelligent, and stealthy network attacks pose more severe security threats to the edge network. In particular, the emergence of novel intelligent attacks makes it a challenging issue to obtain sufficient attack samples, and thus classical deep learning-driven intrusion detection frameworks (IDSs) become ineffective. To tackle this issue, we introduce a novel intrusion detection framework lever aging few-shot class-incremental learning (FSCIL) capabilities to achieve robust detection of emerging threats with few samples. This approach pre-trains a backbone traffic classification model and employs few-shot training with prototypical networks. To further reduce catastrophic forgetting while improving both accuracy and system robustness, we incrementally fine-tune the classification model with supervised contrastive learning, and also realize rapid adaptation to new attacks. Evaluations on the intrusion detection datasets CIC-IDS2017 and USTC-TF2016 demonstrate that our framework consistently outperforms base line models for emerging attacks detection with few attack samples while preserving effective recognition of known threats

With the advancement of low-altitude intelligent networks, low-altitude networks (LANs) centered on low-altitude vehicles (LAVs) play a pivotal role in scenarios such as emergency communication and low-altitude inspection. However, the high dynamics of LAVs induce drastic topological changes and link instability, making traditional routing protocols struggle to satisfy the stringent requirements for low latency and highly reliable communication. Furthermore, existing reinforcement learning approaches are frequently confined to single-objective optimization, rendering them ineffective in addressing these challenges. To address these issues, we construct a highly dynamic time-varying LAN model oriented towards efficient data transmission. Building upon this foundation, we propose an adaptive routing algorithm based on multi-agent proximal policy optimization (MAPPO). First, we formulate the adaptive routing problem as a multi-agent partially observable Markov decision process. Second, the reward function is designed to comprehensively consider node congestion, link stability, and routing overhead, thereby avoiding the local optimality often associated with single-metric optimization. In addition, we enhance the MAPPO framework by integrating gated recurrent units (GRUs), which encode historical local observations into recurrent hidden states, thereby enhancing routing decisions under time-varying topologies. Simulation results demonstrate that the proposed algorithm outperforms the improved Q-learning multi-hop routing (IQMR) algorithm in low-dynamic scenarios. Specifically, it improves the data delivery ratio by approximately 15% and reduces end-to-end communication delay by 33.1%, fully validating the proposed algorithm's capability to achieve efficient routing in LANs.

With next-generation mobile networks evolving toward intelligent wireless infrastructures, wireless nodes require robust and lightweight spectrum perception in dynamic electromagnetic environments. ADS-B monitoring is a representative aerial mobile networking scenario, where specific emitter identification (SEI) based on radio frequency fingerprints (RFFs) provides transmitter-level perception for trustworthy spectrum awareness and autonomous operation. However, existing deep-learning-based SEI methods often involve complex structures and large parameter scales, limiting their use in resource-constrained embodied wireless agents. This paper formulates ADS-B SEI as a lightweight spectrum perception task for electromagnetic spectrum embodied intelligence and proposes QSENN, a hybrid quantum-classical neural network. QSENN integrates convolutional feature extraction, quantum-enhanced channel recalibration, temporal modeling, and attention-based aggregation. A quantum squeeze-and-excitation (QSE) module with dual-axis rotation encoding is designed to enhance channel-wise feature interaction under limited qubit resources. Experiments on a real-world ADS-B dataset with 20 emitter classes show that QSENN achieves the highest average accuracy of 74.95% with only 88.95K parameters and strong low- and medium-SNR robustness. These results indicate that QSENN is suitable as a lightweight spectrum perception engine for electromagnetic spectrum embodied intelligence.

To address weak target energy, missed detection, and network redundancy in unmanned aerial vehicle (UAV) radio-frequency (RF) time-frequency map detection under low signal-to-noise ratio (SNR) conditions, a joint method combining time-frequency enhancement and lightweight detection was proposed. The proposed method selectively enhanced target-related regions using an adaptive foreground mask, multi-scale energy density, and connected-component interior weighting. RF-YOLO-Lite was then constructed through P2/P5 scale reconstruction, a Ghost-based backbone, SlimNeck feature fusion, and a lightweight shared decoupled detection head. Experimental results show that the proposed method achieves 94.85% mAP@0.5 and 79.86% mAP@0.5:0.95 on the simulated dataset, with 1.39×106 parameters and 5.4 GFLOPs. Experiments on the noisy DroneRFa measured dataset further demonstrate its robustness under low-SNR degradation. The method provides a favorable trade-off between detection accuracy and computational complexity.

The Internet of Vehicles (IoV) is undergoing a transformative evolution, enabled by advancements in future 6G network technologies, to support intelligent, highly reliable, and low-latency vehicular services. However, the enhanced capabilities of loV have heightened the demands for efficient network resource allocation while simultaneously giving rise to diverse vehicular service requirements. For network service providers (NSPs), meeting the customized resource-slicing requirements of vehicle service providers (VSPs) while maximizing social welfare has become a significant challenge. This paper proposes an innovative solution by integrating a mean-field multi-agent reinforcement learning (MFMARL) framework with an enhanced Vickrey-Clarke-Groves (VCG) auction mechanism to address the problem of social welfare maximization under the condition of unknown VSP utility functions. The core of this solution is introducing the “value of information" as a novel monetary metric to estimate the expected benefits of VSPs, thereby ensuring the effective execution of the VCG auction mechanism. MFMARL is employed to optimize resource allocation for social welfare maximization while adapting to the intelligent and dynamic requirements of IoV. The proposed enhanced VCG auction mechanism not only protects the privacy of VSPs but also reduces the likelihood of collusion among VSPs, and it is theoretically proven to be dominant-strategy incentive compatible (DSIC). The simulation results demonstrate that, compared to the VCG mechanism implemented using quantization methods, the proposed mechanism exhibits significant advantages in convergence speed, social welfare maximization, and resistance to collusion, providing new insights into resource allocation in intelligent 6G networks.

In this paper, we propose a deep learning (DL)-based task-driven spectrum prediction framework, named DeepSPred. The DeepSPred comprises a feature encoder and a task predictor, where the encoder extracts spectrum usage pattern features, and the predictor configures different networks according to the task requirements to predict future spectrum. Based on the DeepSPred, we first propose a novel 3D spectrum prediction method combining a flow processing strategy with 3D vision Transformer (ViT, i.e., Swin) and a pyramid to serve possible applications such as spectrum monitoring task, named 3D-SwinSTB. 3D-SwinSTB unique 3D Patch Merging ViT-to-3D ViT Patch Expanding and pyramid designs help the model accurately learn the potential correlation of the evolution of the spectrogram over time. Then, we propose a novel spectrum occupancy rate (SOR) method by redesigning a predictor consisting exclusively of 3D convolutional and linear layers to serve possible applications such as dynamic spectrum access (DSA) task, named 3D-SwinLinear. Unlike the 3D-SwinSTB output spectrogram, 3D-SwinLinear projects the spectrogram directly as the SOR. Finally, we employ transfer learning (TL) to ensure the applicability of our two methods to diverse spectrum services. The results show that our 3D-SwinSTB outperforms recent benchmarks by more than 5%, while our 3D-SwinLinear achieves a 90% accuracy, with a performance improvement exceeding 10%.

Due to the fixed bidding and matching process in traditional sealed-bid spectrum auctions, participants with overlarge demands may not be matched, resulting in suboptimal total social welfare and low spectrum utilization. To address this problem, we propose a flexible bidding double spectrum auction scheme, where each buyer can submit both a base bid and an additional bid based on their basic spectrum demand and additional spectrum needs. Then we propose a two-step spectrum auction mechanism: the Sort-based matChing And vickRey Pricing (SCARP) mechanism for base bids in the first step, and the Fairness-based aLlocation And Pricing (FLAP) mechanism for additional bids in the second step. Furthermore, we prove that the proposed mechanism satisfies the truthfulness, budget balance, and individual rationality properties. Simulation results demonstrate that the proposed flexible bidding scheme outperforms the benchmark scheme, significantly improving the social welfare and spectrum utilization.

Due to the high flexibility and versatility, uncrewed aerial vehicles (UAVs) are leveraged in various fields including surveillance and disaster rescue. However, in UAV networks, routing is vulnerable to malicious damage due to distributed topologies and high dynamics. Hence, ensuring the routing security of UAV networks is challenging. In this paper, we characterize the routing process in a time-varying UAV network with malicious nodes. Specifically, we formulate the routing problem to minimize the total delay, which is an integer linear programming and intractable to solve. Then, to tackle the network security issue, a blockchain-based trust management mechanism (BTMM) is designed to dynamically evaluate trust values and identify low-trust UAVs. To improve traditional practical Byzantine fault tolerance algorithms in the blockchain, we propose a consensus UAV update mechanism. Besides, considering the local observability, the routing problem is reformulated into a decentralized partially observable Markov decision process. Further, a multi-agent double deep Q-network based routing algorithm is designed to minimize the total delay. Finally, simulations are conducted with attacked UAVs and numerical results show that the delay of the proposed mechanism decreases by 13.39%, 12.74%, and 16.6% than multi-agent proximal policy optimal algorithms, multi-agent deep Q-network algorithms, and methods without BTMM, respectively.

In this paper, we propose a simultaneous secrecy and covert communications (SSACC) scheme in a reconfigurable intelligent surface (RIS)-aided network with a cooperative jammer. The scheme enhances communication security by maximizing the secrecy capacity and the detection error probability (DEP). Under a worst-case scenario for covert communications, we consider that the eavesdropper can optimally adjust the detection threshold to minimize the DEP. Accordingly, we derive closedform expressions for both average minimum DEP (AMDEP) and average secrecy capacity (ASC). To balance AMDEP and ASC, we propose a new performance metric and design an algorithm based on generative diffusion models (GDM) and deep reinforcement learning (DRL). The algorithm maximizes data rates under user mobility while ensuring high AMDEP and ASC by optimizing power allocation. Simulation results demonstrate that the proposed algorithm achieves faster convergence and superior performance compared to conventional deep deterministic policy gradient (DDPG) methods, thereby validating its effectiveness in balancing security and capacity performance.

As a malicious attack targeting on the GPS receiver, GPS spoofing attack interferes the normal received satellite signal by reproducing or relaying the signal, resulting in severe position deviation. Such attack has posed significant security threat to unmanned-aerial-vehicles (UAVs), especially in the era of low-altitude economics. However, due to the similarity of the spoofing and intended signal, and the presence of noise, accurate detection and recognition of GPS spoofing attack still remains a challenging issue, particularly in the case of limited samples. In this article, we apply the AdaBoost-CNN algorithm, which combines multiple weak convolutional neural network (CNN) classifiers into a strong classification model, to achieve GPS spoofing attack recognition. To further improve the recognition accuracy when there are very limited samples, we improve the AdaBoost-CNN algorithm by transferring previous network parameters to subsequent CNN. Both simulated and real measurement data are employed to verify the effectiveness of the proposed scheme. It is shown that the recognition accuracy can reach up to 93.75% and 95.83% with 160 simulated samples and 120 measured samples, respectively.

Objective The opening of low-altitude airspace and the widespread deployment of Unmanned Aerial Vehicles(UAVs)have significantly increased low-altitude flight activities.Trajectory planning is essential for ensuring UAVs operate safely in complex environments.However,wireless remote control links are vulnerable to interference and spoofing attacks,leading to deviations from planned trajectories and posing serious safety risks.To mitigate these risks,UAV position parameters can be predicted and used to replace erroneous navigation system values,thereby correcting abnormal trajectories.Existing prediction-based correction methods,however,exhibit low efficiency and error accumulation over long-term predictions,limiting their practical application.To address these limitations,this study proposes a multi-model fusion method to improve the efficiency and accuracy of abnormal trajectory correction,providing a robust solution for real-world UAV operations. Methods An Long Short-Term Memory(LSTM)-Transformer prediction model,integrating LSTM and Transformer,is proposed to exploit the strengths of both architectures in time series forecasting.LSTM efficiently captures short-term dependencies in sequential data,whereas Transformer is well-suited for modeling long-term dependencies.By combining these architectures,the proposed model enhances the capture of both short-term and long-term dependencies,reducing prediction errors.The overall framework of the LSTM-Transformer prediction model is illustrated in(Fig.3).The input time series data undergoes preprocessing before being fed into the LSTM and Transformer sub-models,each generating a corresponding feature vector.These feature vectors are concatenated and further processed by a fully connected layer to extract intrinsic data features,ultimately producing the prediction results.To further optimize the model,a blockwise attention strategy is proposed.The detailed computation process is shown in(Fig.4).During self-attention calculations in the Transformer sub-model,the input sequence is divided into multiple sub-blocks,allowing for parallel computation.The results are then concatenated to obtain the final output.This approach effectively reduces the computational complexity of the Transformer sub-model while improving the efficiency of abnormal trajectory correction.The blockwise attention strategy not only enhances computational efficiency but also maintains prediction accuracy,making it a crucial component of the proposed method. Results and Discussions Experiments are conducted using a public dataset to predict UAV positional parameters,including longitude,latitude,and altitude.The dataset's feature parameters are presented in(Table 1).The trajectory correction performance of the proposed method is evaluated and compared with other correction methods using Root Mean Square Error(RMSE),Mean Absolute Error(MAE),and Mean Absolute Percentage Error(MAPE).(Fig.5)and(Fig.6)present the error metrics of the proposed method in comparison with Support Vector Regression(SVR),CNN-LSTM,and LSTM-RF under different prediction step sizes and measurement noise standard deviation conditions.The results indicate that the proposed method achieves the lowest correction errors.At a prediction step size of 20 and a measurement noise standard deviation of 0.19,the proposed method achieves RMSE,MAE,and MAPE values of 0.297 1,0.220 8,and 21.688%,respectively.Compared with SVR,CNN-LSTM,and LSTM-RF,the RMSE is reduced by 39.52%,6.22%,and 20.65%,the MAE by 45.5%,8.46%,and 20.52%,and the MAPE by 8.955%,2.03%,and 3.532%,respectively.(Fig.7)and(Fig.8)compare the proposed method with the original LSTM-Transformer,the Transformer with the blockwise attention optimization strategy,and individual LSTM and Transformer models in terms of error metrics under different prediction steps and measurement noise standard deviation conditions.When the prediction step is 20 and the measurement noise standard deviation is 0.19,the proposed method achieves RMSE reductions of 12.23%,4.07%,1.36%,and 3.48%,MAE reductions of 19.36%,6.76%,3.83%,and 4.21%,and MAPE reductions of 3.84%,3.616%,2.075%,and 2.087%,compared to the other four correction methods.These findings demonstrate the superior performance of the proposed method in reducing trajectory correction errors.The runtime efficiency of the proposed method under different prediction steps is evaluated,as shown in(Fig.9).With a prediction step size of 20,the proposed method completes the prediction in 0.699 s,which is 35.87%faster than the original LSTM-Transformer model.This confirms that the blockwise attention optimization strategy enhances correction efficiency.Finally,(Fig.10)presents trajectory comparisons,illustrating the accuracy of the proposed method.The predicted trajectories closely align with actual trajectories,outperforming baseline methods in correcting UAV abnormal trajectories under various conditions. Conclusions The proposed multi-model fusion method for UAV abnormal trajectory correction enhances correction efficiency and reduces errors more effectively than benchmark methods.The results demonstrate that the method achieves accurate and reliable trajectory correction,making it suitable for practical UAV applications.