Cellular traffic uncertainty caused by spatiotemporal volatility limits prediction accuracy and undermines proactive optimization. To address this issue, we propose a self-tuning deep Gaussian mixture model (ST-DGMM), which replaces the traditional expectation-maximization (EM) algorithm with a deep neural network (DNN) to efficiently estimate Gaussian mixture parameters. A reinforcement learning-based self-tuning mechanism further adapts the network architecture to the historical traffic characteristics of each cell. Experiments show that ST-DGMM consistently outperforms baseline methods in prediction accuracy while providing reliable uncertainty quantification through a 95% confidence interval (CI) derived from the estimated Gaussian mixture distribution. A multi-unmanned aerial vehicle (UAV)-assisted edge offloading case study demonstrates that such uncertainty estimates improve proactive resource allocation by encouraging conservative decisions under high uncertainty and aggressive decisions under low uncertainty.

Graph neural networks (GNNs) have been widely applied in software-defined network (SDN) to enhance network modeling and performance forecasting. However, the closed-box nature of deep learning makes GNNs difficult to interpret, hindering their broad use and the application of GNN-based SDN systems in engineering. In this paper, we propose a novel interpretation framework named GEEK-Explainer, designed to efficiently provide instance-level interpretation of GNNs in SDN. Specifically, we introduce a KernelSHAP-based scoring module to generate intuitive and human-friendly explanations for each performance prediction. To address conflicts in computation cost, we propose a soft discrete mask matrix that identifies a critical set of important nodes. Extensive experiments demonstrate that the RouteNet model can effectively learn the relationships among features, which can provide a better understanding of the prediction process with less computation cost. These findings improve the transparency and robustness of the model and promote the application of GNN-based SDN systems in engineering practice.

The future of UAV interaction systems is evolving from engineer-driven to user-driven, aiming to replace traditional predefined Human-UAV Interaction designs. This shift focuses on enabling more personalized task planning and design, thereby achieving a higher quality of interaction experience and greater flexibility, which can be used in many fields, such as agriculture, aerial photography, logistics, and environmental monitoring. However, due to the lack of a common language between users and the UAVs, such interactions are often difficult to be achieved. The developments of Large Language Models possess the ability to understand natural languages and Robots’ (UAVs’) behaviors, marking the possibility of personalized Human-UAV Interaction. Recently, some HUI frameworks based on LLMs have been proposed, but they commonly suffer from difficulties in mixed task planning and execution, leading to low adaptability in complex scenarios. In this paper, we propose a novel dual-agent HUI framework. This framework constructs two independent LLM agents (a task planning agent, and an execution agent) and applies different Prompt Engineering to separately handle the understanding, planning, and execution of tasks. To verify the effectiveness and performance of the framework, we have built a task database covering four typical application scenarios of UAVs and quantified the performance of the HUI framework using three independent metrics. Meanwhile, different LLM models are selected to control the UAVs with compared performance. Our user study experimental results demonstrate that the framework improves the smoothness of HUI and the flexibility of task execution in the tasks scenario we set up, effectively meeting users’ personalized needs.

We propose a Reinforced Meta-learning-based Traffic Prediction (RML-TP) method for a cellular mobile network. We characterize real-world cellular traffic data using Fast Fourier Transform, offering a significant improvement over traditional methods. Given the varying feature spaces of different cellular traffic data, we demonstrate the necessity of adapting the Deep Neural Network (DNN) structure accordingly, supported by mathematical proof. To achieve this, RML-TP is proposed to capture the intrinsic relationship between the feature spaces and the corresponding optimal network structures. In RML-TP, the underlying DNN is used to predict cellular traffic data, while the upper layer employs value-based RML to adjust the structure of the DNN. Numerical results demonstrate the superiority and generalization of the proposed RML-TP compared to other algorithms, such as Fixed-layer and Rand-layer and applying different networks based on different prediction algorithms, e.g., Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and Transformer. Finally, we demonstrate the advantages of RML-TP through a case study in which, we deploy RML-TP on a Unmanned Aerial Vehicle (UAV) for data offloading tasks, showcasing the superiority of our method compared to other methods.

Spatial-temporal graph neural networks have demonstrated remarkable performance in time series forecasting. However, their inherent opacity poses a significant challenge, limiting their adoption in high-stakes applications. Existing post-hoc explainers are fundamentally limited, as they operate on static physical topologies and isolated time slices, failing to capture essential semantic relationships and long-range temporal dynamics. To address this, we propose FusionSHAP, a novel model-agnostic explanation framework. FusionSHAP makes two primary contributions. First, it introduces Semantic-Physical Graph Fusion, a process that constructs a hybrid graph by augmenting the conventional physical topology with a semantic graph generated by a Large Language Model. This fusion enables the discovery of non-obvious, functionally-driven relationships between nodes. Second, the framework employs Window-Level Attribution with Causal Lag Alignment, a method that adapts KernelSHAP to a node-timestep feature space. This provides temporally-precise explanations that pinpoint not only which nodes were influential but also when. Experiments on real-world traffic forecasting datasets demonstrate that FusionSHAP generates explanations with demonstrably higher fidelity and sparsity than existing methods, paving a new path toward uncovering the complex, non-local, and temporally-specific dependencies learned by STGNNs.

With the advancement of Unmanned Aerial Vehicle (UAV) technology, its role in disaster recovery communication has become increasingly prominent. This paper addresses the challenge of rapidly deploying UAVs in disaster scenarios to ensure timely and energy-efficient communication for affected areas. We propose a disaster recovery communication system comprising two types of UAVs: communication UAVs, which provide coverage for ground users, and charging UAVs, which supply energy to sustain the network. To optimize system performance, we formulate a multi-objective optimization problem that jointly allocates power, bandwidth, and flight trajectories while minimizing total energy consumption, subject to constraints on minimum communication rates and user data demands. By employing the block coordinate descent method and continuous convex approximation techniques, we decompose the problem into three subproblems: communication resource allocation, trajectory optimization, and time allocation. Each subproblem is effectively resolved via convex optimization, ensuring the convergence and feasibility of the solutions. Simulation results confirm that the proposed approach significantly reduces overall system energy consumption and ensures robust performance under diverse mission conditions.

This paper exploits an Unmanned Aerial Vehicle assisted Mobile Edge Computing (UAV-MEC) system to meet the computation demand of User Equipments (UEs). However, in conventional UAV-MEC architecture, a UAV is dedicated only to one single service, which will inevitably limit overall system performance. To address the above issue of the conventional UAVMEC architecture, we propose a hybrid UAV-MEC model. In this model, a UAV is treated as an agent to provide either relay or computing services. Moreover, we propose an MEC Multiple Agent UAV Trajectory and User Association Policy Gradient (MEC-MAUTUA-PG) algorithm to optimize offloading strategies of the UEs as well as the trajectory design of UAV. We show the superiority of the proposed algorithm in terms of delay and the safety of trajectory design of UAV in the environment.

We investigate the minimum-delay multicast scheduling problem for mmWave networks by fully leveraging reflections. An analog beamforming system is considered, where both the orientation and beamwidth of the transmission beam can be adjusted to serve multiple receivers simultaneously. We formulate a combinatorial optimization problem that is difficult to solve directly. To address this challenge, we propose a polynomial-time algorithm that heuristically selects multicast groups based on the distances between receivers. Simulation results demonstrate that, despite lower complexity, the proposed algorithm achieves significant delay reduction compared to baseline approaches and performs close to optimal, with performance gains that increase with the number of reflected paths, showing good scalability.

The computing power network (CPN) offers exceptional computational capabilities and reliable network services, with significant potential for future applications. To achieve ubiquitous coverage and efficient computational resource allocation, CPN can be seamlessly coordinated with low-cost autonomous aerial vehicle (AAV)-based mobile computing platforms. This article investigates an efficient low-altitude AAV-assisted computing power and resource allocation mechanism tailored for urban environments. The aim is to ensure seamless scheduling and efficient processing of computational tasks across various computing devices at different layers of the CPN, while minimizing AAV energy consumption and ensuring flight safety. First, this article proposes an Urban AAV-assisted CPN task-allocation and AAV trajectory-management decision-making problem. The AAV works until it safely lands, aiming to minimize overall task processing delay and AAV energy consumption while ensuring fairness in task allocation. Then, a novel AAV-protection-based multiagent deep deterministic policy gradient (UP-MADDPG) algorithm is introduced. It offers dynamic management of secure computing and communication flight paths when facing building blockages. Finally, we compared the proposed algorithm with baseline algorithms across various metrics. Experimental results demonstrate that the proposed algorithm achieves lower and more balanced task execution delay and AAV energy consumption while also improving fairness.

To address the modal adaptation and modal control issues in polymorphic network (PINet) for software for open networking in the cloud (SONiC) switch operating system, a SONiC network element control channel container called p4runtime-pins based on P4Runtime was proposed. It enabled multi-modal network element devices to support configuration of various network modal flow tables. In the p4runtime-pins container, a connection with the controller through the gRPC service module was established, a neighboring network element discovery algorithm for controller discovery of links was utilized, and a network element port update algorithm to resolve the issue of port changes in practical application environments was designed. Furthermore, to address the issue of poor support for flow table variations in the hardware forwarding processing unit of SONiC network element switches, the internal flow table dumping and gRPC network element proxy functionality was introduced in p4runtime-pins to facilitate the deployment of different network modal flow tables. Experimental results demonstrate that the p4runtime-pins container has low resource consumption, occupying only 1.70% of CPU and 0.45% of memory. Moreover, SONiC network element devices deployed with the p4runtime-pins container can accurately receive and configure flow table rules issued by the controller, with an only 0.027~0.037 s flow table configuration delay.

Most existing research on autonomous vehicle groups focuses on utilizing networks to achieve semi-centralized control of leaders/sub-leaders. However, these approaches encounter difficulties when striving to attain precise cooperative environmental awareness in open scenes with inherent interference. To address this problem, we propose a systematic model for distributed autonomous vehicle groups. It incorporates contributed perception among autonomous vehicles. Firstly, this work leverages edge computing to enable a cooperative interaction among autonomous vehicles, thus improving precision of environmental awareness when individual sensing is limited. Then, it introduces a transformer-based prediction method to analyze influencing factors of contributed perception. Finally, it constructs an autonomous vehicle group model and solves it by using a multi-objective optimization method. The simulation results demonstrate that the proposed prediction method has lower mean-square error than existing prediction methods, and the proposed autonomous vehicle group model outperforms existing ones in terms of average group contribution, accessibility, persistence, and timeliness.

In this letter, we propose an accurate and efficient heterogeneous graph neural network-powered ray tracing (HGNN-RT) framework to predict path loss for urban scenarios. The proposed HGNN-RT effectively captures both global and local features, which include broad path attributes, such as total path length and line-of-sight conditions, as well as specific interactions, such as reflection angles and obstacles. Moreover, due to the fast information flow of the graph neural networks, the HGNN-RT achieves efficiency-improved path loss prediction in complex multireflection and multipath propagation channels. Experimental results demonstrate that the proposed approach achieves errors within 2 dB of ray tracing benchmarks, while also being time-efficient and reliably generalizing to unseen environments for channel modeling.

Federated learning (FL) in resource-constrained wireless networks faces the challenge of long training delays. In this work, we explore delay-efficient FL by leveraging device-to-device (D2D) communications to accelerate the uploading of local models. We formulate a joint problem of upload mode selection and bandwidth allocation, which is a mixed-integer nonlinear programming (MINLP) problem and difficult to solve directly. To address this, we propose a low-complexity two-step algorithm: the first step determines the upload modes for edge devices, while the second step optimally allocates the bandwidth. Simulation results show that our algorithm outperforms baseline schemes, with delay reduction becoming more pronounced as the number of edge devices increases.

In the contemporary landscape of computationally intensive applications, Computing Power Network (CPN) offers a solution to enhance computational efficiency and cost-effectiveness by integrating and sharing computing resources. However, with the surge in task volume within multi-user environments, effectively scheduling these tasks to optimize system profit and delay presents a significant challenge. This article introduces an optimization approach leveraging Deep Deterministic Policy Gradient (DDPG) to enhance CPN performance through task decomposition and computing path optimization. We initially construct a multi-layer CPN system model encompassing cloud computing, edge computing, and terminal device layers. Subsequently, we integrate a novel mechanism for convex optimization-based task decomposition, enabling intelligent subdivision of tasks into sub-tasks and dynamic allocation to suitable nodes within the network. Furthermore, we devise a Convex Optimization Task Decomposition-based Multi-Agent Deep Deterministic Policy Gradient (CO-MADDPG) algorithm, empowering multiple computing tasks as independent agents to learn and identify optimal offloading paths and computing nodes, thereby minimizing delay and maximizing system profit. A series of simulation experiments validate the effectiveness of the CO-MADDPG algorithm in handling concurrent tasks, demonstrating its capability to reduce task completion times, enhance system revenue, and maintain adaptability and stability across varying task demands.

The Computing Power Network (CPN) offers exceptional computational capabilities and reliable network services, with significant potential for future applications. To achieve ubiquitous coverage and efficient computational resource allocation, CPN can be seamlessly coordinated with low-cost Unmanned Aerial Vehicle (UAV)-based mobile computing platforms. This paper investigates an efficient low-altitude UAV-assisted computing power and resource allocation mechanism tailored for urban environments. The aim is to ensure seamless scheduling and efficient processing of computational tasks across various computing devices at different layers of the CPN, while minimizing UAV energy consumption and ensuring flight safety. Firstly, this paper proposes an Urban UAV-assisted CPN task-allocation and UAV trajectory-management decision-making problem. The UAV works until it safely lands, aiming to minimize overall task processing delay and UAV energy consumption while ensuring fairness in task allocation. Then, a novel UAV-Protection based Multi-Agent Deep Deterministic Policy Gradient (UP-MADDPG) algorithm is introduced. It offers dynamic management of secure computing and communication flight paths when facing building blockages. Finally, we compared the proposed algorithm with baseline algorithms across various metrics. Experimental results demonstrate that the proposed algorithm achieves lower and more balanced task execution delay and UAV energy consumption while also improving fairness.

We study efficient federated learning (FL) using random pruning in resource-constrained edge intelligence networks. We propose an edge device selection strategy to identify appropriate edge devices for participating in FL at the beginning of each training iteration. We then formulate an optimization problem that jointly optimizes the pruning ratio, CPU frequency, uplink power, and bandwidth allocation for the selected edge devices. Since the optimization problem is non-convex and challenging to solve directly, we decompose it into three subproblems and propose efficient algorithms or closed-form solutions for each subproblem. Based on the solutions to the subproblems, an alternating optimization algorithm is constructed to solve the original problem. Simulation results demonstrate that our scheme outperforms baseline schemes in terms of both learning accuracy and energy consumption.

Traditional ground-based illumination equipment is limited in mobility and light source height, making it difficult to adapt to diverse living scenarios such as camping that require quick and flexible illumination solutions. With the rapid development of Unmanned Aerial Vehicle (UAV) technology, particularly in illumination services, UAVs have demonstrated unique advantages. Addressing the inadequacies of conventional illumination, this study proposes a prototype of an autonomously deployed illumination system based on the RoboMaster Tello Talent (Tello) UAV, designed to provide quick and flexible on-site illumination solutions. The system’s design encompasses three complementary modules to enhance its overall functionality and efficiency. Firstly, the illumination module equips the Tello UAV with a specialized illumination extension, ensuring flight stability and effective illumination. Secondly, the addressing module employs iterative algorithms to identify optimal UAV deployment locations and precisely plan flight paths. Lastly, the flight control module, guided by the results from the addressing module, scripts flight commands, integrates with the Tello UAV’s Application Programming Interface (API), and executes flight plans optimized for path efficiency, ensuring the UAV quickly and accurately reaches designated locations, coordinating with the illumination module to deliver effective illumination. Experimental results demonstrate that the proposed illumination system can swiftly respond to various user demands, autonomously deploy UAVs to optimal illumination positions, and provide high-quality service.

In the current era of mobile edge networks, a significant challenge lies in overcoming the limitations posed by limited edge storage and computational resources. To address these issues, accurate network traffic prediction has emerged as a promising solution. However, due to the intricate spatial and temporal dependencies inherent in mobile edge network traffic, the prediction task remains highly challenging. Recent spatio-temporal neural network algorithms based on graph convolution have shown promising results, but they often rely on pre-defined graph structures or learned parameters. This approach neglects the dynamic nature of short-term relationships, leading to limitations in prediction accuracy. To address these limitations, we introduce Ada-ASTGCN, an innovative attention-based adaptive spatio-temporal graph convolutional network. Ada-ASTGCN dynamically derives an optimal graph structure, considering both the long-term stability and short-term bursty evolution. This allows for more precise spatio-temporal network traffic prediction. In addition, we employ an alternative training approach during optimization, replacing the traditional end-to-end training method. This alternative training approach better guides the learning direction of the model, leading to improved prediction performance. To validate the effectiveness of Ada-ASTGCN, we conducted extensive traffic prediction experiments on real-world datasets. The results demonstrate the superior performance of Ada-ASTGCN compared to existing methods, highlighting its ability to accurately predict network traffic in mobile edge networks.

We jointly consider full-duplex operation and network coding in two-hop relay networks to enhance the throughput of the block transmission of packets over erasure channels. Two coded transmission schemes, termed Fewest Broadcast Packet First (FBPF) and Buffer Contents-based Coded Transmission (BCCT), are proposed, where random linear network coding is employed at the Base Station (BS) and the Relay Station (RS), respectively. Both schemes do not rely on users’ Channel State Information (CSI), buffer status, channel parameters, etc., and hence are practically viable. We derive closed-form upper bounds on the throughput of both schemes. We prove that both schemes achieve the optimal throughput when the BS-to-RS channel is perfect. Through extensive simulations, we demonstrate that both schemes incur substantially higher throughput than the traditional uncoded Automatic Repeat-reQuest (ARQ) scheme and perform close to a general upper bound on the system throughput. Furthermore, even with imperfect Self-Interference Cancellation (SIC) at the full-duplex RS, our schemes are shown to be superior to state-of-the-art coded transmission schemes designed for half-duplex relay networks, given that the impact of imperfect SIC on the BS-to-RS channel quality is not high.