In this paper, we propose a novel physics-informed generative learning approach, named RadioDiff-k2, for accurate and efficient multipath-aware radio map (RM) construction. As future wireless communication evolves towards environment-aware paradigms, the accurate construction of RMs becomes crucial yet highly challenging. Conventional electromagnetic (EM)-based methods, such as full-wave solvers and ray-tracing approaches, exhibit substantial computational overhead and limited adaptability to dynamic scenarios. Although existing neural network (NN) approaches have efficient inferencing speed, they lack sufficient consideration of the underlying physics of EM wave propagation, limiting their effectiveness in accurately modeling critical EM singularities induced by complex multipath environments. To address these fundamental limitations, we propose a novel physics-inspired RM construction method guided explicitly by the Helmholtz equation, which inherently governs EM wave propagation. Specifically, based on the analysis of partial differential equations (PDEs), we theoretically establish a direct correspondence between EM singularities, which correspond to the critical spatial features influencing wireless propagation, and regions defined by negative wave numbers in the Helmholtz equation. We then design an innovative dual diffusion model (DM)-based large artificial intelligence framework comprising one DM dedicated to accurately inferring EM singularities and another DM responsible for reconstructing the complete RM using these singularities along with environmental contextual information. Experimental results demonstrate that the proposed RadioDiff-k2framework achieves state-of-the-art (SOTA) performance in both image-level RM construction and localization tasks, while maintaining inference latency within a few hundred milliseconds. Code is available at https://github.com/UNIC-Lab/RadioDiff-k.

The increase of bandwidth-intensive applications in sixth-generation (6 G) wireless networks, such as real-time volumetric streaming, and multi-sensory extended reality, demands intelligent multicast routing solutions capable of delivering differentiated quality-of-service (QoS) at scale. Traditional shortest-path and multicast routing algorithms are either computationally prohibitive or structurally rigid, and they often fail to support heterogeneous user demands, leading to suboptimal resource utilization. Neural network-based approaches, while offering improved inference speed, typically lack topological generalization and scalability. To address these limitations, this paper presents a graph neural network (GNN)-based multicast routing framework that jointly minimizes total transmission cost and supports user-specific video quality requirements. The routing problem is formulated as a constrained minimum-flow optimization task, and a reinforcement learning algorithm is developed to sequentially construct efficient multicast trees by reusing paths and adapting to network dynamics. A graph attention network (GAT) is employed as the encoder to extract context-aware node embeddings, while a long short-term memory (LSTM) module models the sequential dependencies in routing decisions. Extensive simulations demonstrate that the proposed method closely approximates optimal dynamic programming-based solutions while significantly reducing computational complexity. The results also confirm strong generalization to large-scale and dynamic network topologies, highlighting the method's potential for real-time deployment in 6 G multimedia delivery scenarios. Code is available athttps://github.com/UNIC-Lab/GNN-Routing.

Radio maps (RMs) are essential for environment-aware communication and sensing, providing location-specific wireless channel information. Existing RM construction methods often rely on precise environmental data and base station (BS) locations, which are not always available in dynamic or privacy-sensitive environments. While sparse measurement techniques reduce data collection, the impact of noise in sparse data on RM accuracy is not well understood. This paper addresses these challenges by formulating RM construction as a Bayesian inverse problem under coarse environmental knowledge and noisy sparse measurements. Although maximum a posteriori (MAP) filtering offers an optimal solution, it requires a precise prior distribution of the RM, which is typically unavailable. To solve this, we propose RadioDiff-Inverse, a diffusion-enhanced Bayesian inverse estimation framework that uses an unconditional generative diffusion model to learn the RM prior. This approach not only reconstructs the spatial distribution of wireless channel features but also enables environmental building outlines perception, just relying on pathloss, through integrated sensing and communication (ISAC). The proposed method operates on routine communication measurements, without new waveforms, specialized feedback, or protocol changes, thereby enabling a plug-and-play ISAC capability. Remarkably, RadioDiff-Inverse is training-free, leveraging a pre-trained model from Imagenet without task-specific fine-tuning, which significantly reduces the training cost of using a generative large model in wireless networks. Experimental results demonstrate that RadioDiff-Inverse achieves state-of-the-art performance in accuracy of RM construction and environmental reconstruction, and robustness against noisy sparse sampling.

Sixth generation (6G) systems require environment-aware communication, driven by native artificial intelligence (AI) and integrated sensing and communication (ISAC). Radio maps (RMs), providing spatially continuous channel information, are key enablers. However, generating high-fidelity RM ground truth via electromagnetic (EM) simulations is computationally intensive, motivating machine learning (ML)-based RM construction. The effectiveness of these data-driven methods depends on large-scale, high-quality training data. Current public datasets often focus on single-input single-output (SISO) and limited information, such as path loss, which is insufficient for advanced multi-input multi-output (MIMO) systems requiring detailed channel state information (CSI). To address this gap, this paper presents UrbanMIMOMap, a novel large-scale urban MIMO CSI dataset generated using high-precision ray tracing. UrbanMIMOMap offers comprehensive complex CSI matrices across a dense spatial grid, going beyond traditional path loss data. This rich CSI is vital for constructing high-fidelity RMs and serves as a fundamental resource for data-driven RM generation, including deep learning. We demonstrate the dataset’s utility through baseline performance evaluations of representative ML methods for RM construction. This work provides a crucial dataset and reference for research in high-precision RM generation, MIMO spatial performance, and ML for 6G environment awareness. The code and data for this work are available at: https://github.com/UNIC-Lab/UrbanMIMOMap.

This paper investigates the secrecy performance optimization of an uplink satellite communication system exposed to dual-domain eavesdropping threats. Specifically, a cooperative relaying architecture is considered, where a ground user (GU) transmits confidential information to a target satellite (TS), assisted by an amplify-and-forward (AF) relay satellite (RS). Simultaneously, a ground-based malicious node acts as an AF relay to enhance the interception capability of a satellite eavesdropper (SE). To improve secure transmission, a secrecy rate maximization problem is formulated by jointly optimizing the transmit powers of the GU and RS, subject to quality-of-service constraints at the TS. The resulting non-convex problem is solved efficiently using a successive convex approximation algorithm. Simulation results demonstrate that the proposed cooperative relaying scheme significantly enhances secrecy performance compared to traditional non-cooperative baselines. These findings highlight the potential of cooperative multi-satellite relaying as an effective approach to securing next-generation satellite communication networks against sophisticated eavesdropping threats.

We introduce a novel Multi-modal Guided Real-World Face Restoration (MGFR) technique designed to improve the quality of facial image restoration from low-quality inputs. Leveraging a blend of attribute text prompts, high-quality reference images, and identity information, MGFR can mitigate the generation of false facial attributes and identities often associated with generative face restoration methods. By incorporating a dual-control adapter and a two-stage training strategy, our method effectively utilizes multi-modal prior information for targeted restoration tasks. We also present the Reface-HQ dataset, comprising over 21,000 high-resolution facial images across 4800 identities, to address the need for reference face training images. Our approach achieves superior visual quality in restoring facial details under severe degradation and allows for controlled restoration processes, enhancing the accuracy of identity preservation and attribute correction. Including negative quality samples and attribute prompts in the training further refines the model's ability to generate detailed and perceptually accurate images.

Deep learning plays increasingly important role in future wireless network management and optimization. Existing training methods such as label-based supervised learning and label-free learning have inherent limitations. The performance of supervised learning is limited by labels, while label-free training methods require extensive exploration. To address these limitations, this paper proposes a novel mixture of gradients (MoG) method, which integrates gradients from different sources within the training process in order to improve the convergence performance of neural networks (NNs). Particularly, MoG is a modular, plug-and-play solution requiring no structural modifications to existing NNs. Its implementation necessitates only minor modifications to the loss function, where the label-based supervised loss is combined with a label-free loss through weighted summation. The label-free loss can be either unsupervised loss or reinforcement learning loss. This flexibility allows seamless integration into nearly all NN-based methods, making it applicable to a wide range of wireless optimization problems with minimal implementation cost. Extensive simulations across multiple classic wireless scenarios demonstrate that MoG can significantly enhance the performance of NN decision-making, leading to higher transmission rates.

Event cameras, which are highly sensitive to light intensity changes, often generate substantial noise during imaging. Existing denoising methods either lack the speed for real-time processing or struggle with dynamic scenes, mistakenly discarding valid events. To address these issues, we propose a novel dual-stage polarity-focused denoising (PFD) method that leverages the consistency of polarity and its changes within local pixel areas. Whether due to camera motion or dynamic scene changes, the polarity and its changes in triggered events are highly correlated with these movements, allowing for effective noise handling. We introduce two versions: PFD-A, which excels at reducing background activity (BA) noise, and PFD-B, which is designed to address both BA and flicker noise. Both versions are lightweight and computationally efficient. The experimental results show that PFD outperforms benchmark methods in terms of the SNR and ESR metrics, achieving state-of-the-art performance across various datasets. Additionally, we propose an FPGA implementation of PFD processes that handles each event in just 7 clock cycles, ensuring real-time performance. The codes are available athttps://github.com/shicy17/PFD.

Mobile edge computing (MEC) enhances data processing by enabling users to offload tasks to edge servers with enough computation resource. In multi-user and multi-server scenario, the offloading scheduling is overwhelming complex and significantly influences the processing delay, which makes deep learning (DL) become an appealing approach. Yet, prior DL-based methods often overlook dynamic topology challenges due to the inflexibility of fixed neural network structures, leading to constrained performance. To tackle this challenge, a novel reinforcement learning framework named dynamic topology adaptive reinforcement learning (DTA-RL) is proposed in this paper. The MEC network is modeled as a graph based on the communication relationships between users and servers, and the offloading process is formulated as a Markov decision process (MDP). Building on the graph model and MDP, DTA-RL leverages graph attention networks to handle dynamic observation spaces and incorporates an attention mechanism for decision-making in environments with evolving action spaces. Simulation results illustrate that DTA-RL effectively reduces task processing delays and offloading failure rates within the MEC system. Furthermore, the pre-trained model can be seamlessly implemented in networks with new topology without experiencing significant performance degradation. The code is available at https://github.com/UNIC-Lab/DTA-RL.

Automatic Modulation Classification (AMC) plays a crucial role in non-cooperative communication systems and is an essential component of blind signal processing. The application of deep learning methods in modulation classification has shown tremendous potential, surpassing the performance of traditional methods by a large margin. However, the high storage and computational requirements of existing deep learning methods limit their practical applications. In this paper, we propose an AMC technique using an Adaptive Lightweight Wavelet Neural Network (ALWNN) that features a streamlined design and lower computational demands. This innovative model introduces an adaptive wavelet-based feature extraction method that effectively captures information at different frequencies in the input data, ensuring classification accuracy. Additionally, the model incorpo-rates depthwise separable convolution techniques, transforming traditional convolutions into depthwise convolutions and point-wise convolutions, Substantially diminishing the count of the model’s parameters and the complexity of its computations. The proposed ALWNN model strikes a balance between efficiency and accuracy. Simulation results demonstrate that with only 9899 and 9700 parameters, it achieves accuracies of 62.14% and 63.93% on the datasets known as RML2016.10a and RML2016.10b, respectively. Furthermore, we evaluate the model in terms of Floating Point Operations Per Second (FLOPS) and Normalized Multiply-Accumulate Complexity (NMACC) to provide a more comprehensive measure of computational complexity. Compared to existing methods, ALWNN reduces FLOPS by 1.25 to 1.91 orders of magnitude and NMACC by 0.81 to 1.6 orders of magnitude.

The advent of deep learning (DL) technology has increasingly captivated the research community’s interest in harnessing DL to enhance data transmission efficiency. Notwithstanding, prevalent methodologies often overlook the computation delay of DL processing data during the inferencing procedure, and fail to adjust intelligent algorithm complexity based on user features. To bridge this gap, we introduce $\mathbf{C o m}^{2}$ (Communication-Computation) framework, to synergize the optimization of communication and computational delays. $\mathrm{Com}^{2}$ adeptly navigates the trade-offs between communication and computation delays, facilitated by autoencoders of varying depths, thus one user can reduce communication delay through more computation latency, and vice versa. Further enhancing this framework, we present an optimization algorithm that marries QMIX with a cascaded graph neural network (GNN), designed to select the optimal autoencoder depth and optimize transmission resources in a distributed manner. This algorithm pioneers a label-free training regime, employing reinforcement learning and unsupervised learning to adaptively improve without the need for high-quality labels. Simulation results show that $\mathrm{Com}^{2}$, alongside the proposed optimization algorithm, maximizes the utility of users’ computing and transmission resources, significantly curtailing the overall data transmission delay by intelligently managing delay trade-offs.

Unmanned aerial vehicles (UAVs), due to their low cost and high flexibility, have been widely used in various scenarios to enhance network performance. However, the optimization of UAV trajectories in unknown areas or areas without sufficient prior information still faces challenges related to poor planning performance and low distributed execution. These challenges arise when UAVs rely solely on their own observation information and the information from other UAVs within their communicable range, without access to global information. To address these challenges, this article proposes the Qedgix framework, which combines graph neural networks (GNNs) and the QMIX algorithm to achieve distributed optimization of the Age of Information (AoI) for users in unknown scenarios. The framework utilizes GNNs to extract information from UAVs, users within the observable range, and other UAVs within the communicable range, thereby enabling effective UAV trajectory planning. Due to the discretization and temporal features of AoI indicators, the Qedgix framework employs QMIX to optimize decentralized partially observable Markov decision processes (Dec-POMDP) based on centralized training and distributed execution (CTDE) with respect to mean AoI values of users. By modeling the UAV network optimization problem in terms of AoI and applying the Kolmogorov-Arnold representation theorem, the Qedgix framework achieves efficient neural network training through parameter sharing based on permutation invariance. Simulation results demonstrate that the proposed algorithm significantly improves convergence speed while reducing the mean AoI values of users. The code is available athttps://github.com/UNIC-Lab/Qedgix.

Federated learning (FL) has gained increasing attention due to its ability to collaboratively train while protecting client data privacy. However, vanilla FL cannot adapt to client heterogeneity, leading to a degradation in training efficiency due to stragglers, and is still vulnerable to privacy leakage. To address these issues, this paper proposes RingSFL, a novel distributed learning scheme that integrates FL with a model split mechanism to adapt to client heterogeneity while maintaining data privacy. In RingSFL, all clients form a ring topology. For each client, instead of training the model locally, the model is split and trained among all clients along the ring through a pre-defined direction. By properly setting the propagation lengths of heterogeneous clients, the straggler effect is mitigated, and the training efficiency of the system is significantly enhanced. Additionally, since the local models are blended, it is less likely for an eavesdropper to obtain the complete model and recover the raw data, thus improving data privacy. The experimental results on both simulation and prototype systems show that RingSFL can achieve better convergence performance than benchmark methods on independently identically distributed (IID) and non-IID datasets, while effectively preventing eavesdroppers from recovering training data.

Federated Learning (FL) is a promising paradigm widely used in privacy-preserving machine learning. It enables distributed devices to collaboratively train a model while avoiding data transfer between clients. Nevertheless, FL suffers from bottlenecks in training speed due to client heterogeneity, resulting in increased training latency and server aggregation lagging. To address this issue, a novel Split Federated Learning (SFL) framework is proposed. It pairs clients with different computational resources based on their computational resources and inter-client communication rates. The neural network model is split into two parts at the logical level, and each client computes only its assigned part using Split Learning (SL) to accomplish forward inference and backward training. Besides, a heuristic greedy algorithm is proposed to effectively deal with the client pairing problem by reconstructing the training latency optimization as a graph edge selection problem. Simulation results show that the proposed method can significantly improve the FL training speed and achieve high performance in both independent identical distribution (IID) and Non-IID data distribution.

In Space-air-ground integrated networks (SAGIN), the inherent openness and extensive broadcast coverage expose these networks to significant eavesdropping threats. Considering the inherent co-channel interference due to spectrum sharing among multi-tier access networks in SAGIN, it can be leveraged to assist the physical layer security among heterogeneous transmissions. However, it is challenging to conduct a secrecy-oriented access strategy due to both heterogeneous resources and different eavesdropping models. In this paper, we explore secure access selection for a scenario involving multi-mode users capable of accessing satellites, unmanned aerial vehicles, or base stations in the presence of eavesdroppers. Particularly, we propose a Q-network approximation based deep learning approach for selecting the optimal access strategy for maximizing the sum secrecy rate. Meanwhile, the power optimization is also carried out by an unsupervised learning approach to improve the secrecy performance. Remarkably, two neural networks are trained by unsupervised learning and Q-network approximation which are both label-free methods without knowing the optimal solution as labels. Numerical results verify the efficiency of our proposed power optimization approach and access strategy, leading to enhanced secure transmission performance.

Deep learning has been successfully adopted in mobile edge computing (MEC) to optimize task offloading and resource allocation. However, the dynamics of edge networks raise two challenges in neural network (NN)-based optimization methods: low scalability and high training costs. Although conventional node-output graph neural networks (GNN) can extract features of edge nodes when the network scales, they fail to handle a new scalability issue whereas the dimension of the decision space may change as the network scales. To address the issue, in this paper, a novel link-output GNN (LOGNN)-based resource management approach is proposed to flexibly optimize the resource allocation in MEC for an arbitrary number of edge nodes with extremely low algorithm inference delay. Moreover, a label-free unsupervised method is applied to train the LOGNN efficiently, where the gradient of edge tasks processing delay with respect to the LOGNN parameters is derived explicitly. In addition, a theoretical analysis of the scalability of the node-output GNN and link-output GNN is performed. Simulation results show that the proposed LOGNN can efficiently optimize the MEC resource allocation problem in a scalable way, with an arbitrary number of servers and users. In addition, the proposed unsupervised training method has better convergence performance and speed than supervised learning and reinforcement learning-based training methods. The code is available at https://github.com/UNIC-Lab/LOGNN.

As a fundamental problem, many studies are dedicated to the optimization of signal-to-interference-plus-noise ratio (SINR), in a multi-user setting. Although traditional model-based optimization methods achieve strong performance, it has high complexity. To fully leverage the high performance of traditional methods and the low complexity of the neural network (NN) based method, a knowledge distillation (KD) based algorithm distillation (AD) method is proposed in this paper, where traditional optimization methods serve as "teachers" for NN "students", improving unsupervised and reinforcement learning. This approach tackles common issues: unattainable optimal labels, overfitting, and inefficient training. Simulations confirm the advantages of AD, paving the way for traditional optimization integration with NNs in wireless communication.

By offloading computation-intensive tasks of vehicles to roadside units (RSUs), mobile edge computing (MEC) in the Internet of Vehicles (IoV) can relieve the onboard computation burden. However, existing model-based task offloading methods suffer from heavy computational complexity with the increase of vehicles and data-driven methods lack interpretability. To address these challenges, in this paper, we propose a knowledge-driven multi-agent reinforcement learning (KMARL) approach to reduce the latency of task offloading in cybertwin-enabled IoV. Specifically, in the considered scenario, the cybertwin serves as a communication agent for each vehicle to exchange information and make offloading decisions in the virtual space. To reduce the latency of task offloading, a KMARL approach is proposed to select the optimal offloading option for each vehicle, where graph neural networks are employed by leveraging domain knowledge concerning graph-structure communication topology and permutation invariance into neural networks. Numerical results show that our proposed KMARL yields higher rewards and demonstrates improved scalability compared with other methods, benefitting from the integration of domain knowledge.

With the rapid development of the Internet of Things (IoT), there are a dramatically increasing number of devices, leading to the fact that only using terrestrial infrastructure can hardly provide high-quality services to all devices. Due to their flexibility, maneuverability, and economy, unmanned aerial vehicles (UAVs) are widely used to improve the performance of IoT networks. UAVs can not only provide wireless access to IoT devices in the absence of a terrestrial network but can also perform rich IoT services and applications such as video surveillance, cargo transportation, pesticide spraying, and so forth. However, due to the high complexity, dynamics, and heterogeneity of the UAV-assisted IoT networks, growing attention has focused on using artificial intelligence (AI)-based methods to optimize, schedule, and orchestrate UAV-assisted IoT networks. In this article, we comprehensively analyze the impact of applying advanced AI architectures, models, and methods to different aspects of UAV-assisted IoT networks, including key IoT technologies, tasks, and applications. In addition, this article also explores challenges and discusses potential research directions of AI-enabled UAV-assisted IoT networks.

Task scheduling is a critical problem when one user offloads multiple different tasks to the edge server. When a user has multiple tasks to offload and only one task can be transmitted to server at a time, while server processes tasks according to the transmission order, the problem is NP-hard. However, it is difficult for traditional optimization methods to quickly obtain the optimal solution, while approaches based on reinforcement learning face with the challenge of excessively large action space and slow convergence. In this paper, we propose a Digital Twin (DT)-assisted RL-based task scheduling method in order to improve the performance and convergence of the RL. We use DT to simulate the results of different decisions made by the agent, so that one agent can try multiple actions at a time, or, similarly, multiple agents can interact with environment in parallel in DT. In this way, the exploration efficiency of RL can be significantly improved via DT, and thus RL can converges faster and local optimality is less likely to happen. Particularly, two algorithms are designed to made task scheduling decisions, i.e., DT-assisted asynchronous Q-learning (DTAQL) and DT-assisted exploring Q-learning (DTEQL). Simulation results show that both algorithms significantly improve the convergence speed of Q-learning by increasing the exploration efficiency.

Swarm robotics is envisioned to automate a large number of dirty, dangerous, and dull tasks. Robots have limited energy, computation capability, and communication resources. Therefore, current swarm robotics have a small number of robots, which can only provide limited spatio-temporal information. In this paper, we propose to leverage the mobile edge computing to alleviate the computation burden. We develop an effective solution based on a mobility-aware deep reinforcement learning model at the edge server side for computing scheduling and resource. Our results show that the proposed approach can meet delay requirements and guarantee computation precision by using minimum robot energy.