The rapid adoption of 5G has boosted e-commerce live streaming, raising expectations for real-time performance, but traditional Adaptive Bitrate (ABR) algorithms designed for Video-on-Demand (VoD) struggle with bandwidth estimation, short-term prediction, and buffer management in low-latency live streaming (LLL) scenarios, often leading to stalls and latency due to idle- time-induced bandwidth inaccuracies, unstable network fluctuations, and small buffer sizes; to address these issues, this paper proposes LLL-ABR, which combines a CMAF-based bandwidth measurement strategy to filter idle time, a hybrid prediction model for second-level bandwidth forecasting, and progressive bitrate adaptation with nonlinear speed control to optimize Quality of Experience (QoE), with experimental results on public datasets confirming its superiority in maintaining stable QoE for LLL scenarios.

Mobile Edge Computing (MEC) offers a promising solution to reduce latency in video streaming by bringing compute and storage closer to end users. However, dynamic content popularity, multi-bitrate demand, and limited edge resources pose significant challenges for joint transcoding and resource allocation. In this paper, we formulate the transcoding decision and resource-scheduling problem as a Markov Decision Process whose state encodes bandwidth and compute availability, request context, and workload features. We design a composite reward that balances end-to-end delay, backhaul bandwidth use, and compute cost. Building on this formulation, we propose DRRT-PPO, a single-agent Proximal Policy Optimization framework that jointly decides when to transcode and how to allocate bandwidth and CPU resources. By leveraging prioritized experience replay, entropy-regularized exploration, and clipped policy updates, DRRT-PPO learns robust, adaptive strategies under fluctuating workloads. Extensive simulations with Zipf-distributed requests demonstrate that DRRT-PPO reduces average latency by 6.4% compared to Greedy, No-Transcoding, and heuristic baselines, while achieving lower backhaul usage, minimal compute cost, high cache hit rates, and low timeout probabilities. These results confirm DRRT-PPO’s effectiveness for efficient and reliable edge-based video delivery.

Unmanned aerial vehicles (UAVs) have emerged as a promising solution to enhance communication networks, especially in scenarios where traditional infrastructure is impractical or unavailable. However, the reliance on line-of-sight channels in UAV-assisted relay networks introduces significant security vulnerabilities. This paper proposes a novel framework that jointly optimizes transmit power and UAV trajectory to achieve secure communications in such networks. To enhance system performance against environmental uncertainties, we employ a model predictive control (MPC)-based approach, which allows for real-time adaptive control of the UAV’s trajectory by considering future states and disturbances. This approach significantly improves the network’s resilience to dynamic environments and potential eavesdropping threats. Simulation results show that the proposed joint optimization of power allocation and trajectory design not only enhances communication security but also demonstrates the effectiveness of the MPC framework in real-time trajectory tracking, ensuring robust performance under varying environmental conditions.

The vehicle edge network leverages mobile edge computing (MEC) to provide low-latency data and intelligent services. A key challenge is how to efficiently utilize limited resources to disseminate data and promptly respond to service requests. Existing methods, which cache data at the edge to reduce access delay, only consider single-source data distribution. However, applications like object detection and trajectory prediction involve multiple data copies across edges and vehicles. The destination vehicle can retrieve data from multiple sources, requiring optimal source selection. To tackle these challenges, we formulate a joint optimization problem for data dissemination and caching from multiple sources to multiple destinations, considering data timeliness and request deadlines to maximize request success rates. Specifically, we transform the problem into a minimum Steiner tree problem and propose a multi-agent reinforcement learning algorithm to determine source selection and caching strategies. Extensive experiments demonstrate that our approach significantly improves request success rates compared to benchmarks.

Vehicular edge computing (VEC) has emerged as a cutting-edge distributed computing paradigm capable of addressing network congestion and excessive energy use in vehicular systems. To enhance VEC performance, we examined the energy-latency tradeoff for partial tasks offloading in end-VEC-cloud orchestrated networks. We formulated a joint computation offloading and resource allocation problem aimed at minimizing latency and energy consumption. To address the underlined problem, we proposed a collaborative task splitting and resource allocation optimization (CTSRAO) algorithm. We initially decoupled the problem into two convex sub-problems and then applied the Lagrangian and simplex methods for joint optimization of computation resources and task splitting ratio. Furthermore, we investigated the criteria for determining whether a task should be offloaded to the VEC or cloud. Simulation results showed that our algorithm significantly enhances systems performance, achieving lower latency and energy consumption than the benchmark and state-of-the-art methods.

By integrating cloud, edge, and device resources, the computing power networks provide integrated services such as data sensing, transmission, and computation for the digital economy. However, their rapid development is accompanied by pressing challenges of high energy consumption. The task offloading technology is an important solution that allocates computing tasks properly, improves user experience, reduces transmission latency and energy consumption, making it a crucial solution. In order to reduce the overall energy consumption of computing power networks and achieve green, sustainable development, an intelligent matching task unloading scheme based on matching mechanism was proposed. By matching tasks and node resources in the computing power network, the scheme minimized energy consumption caused by inefficient task offloading and enhances overall network performance. Furthermore, a deep learning approach combining reinforcement learning with neural networks was employed to further optimize the offloading strategy, significantly reducing network energy consumption. Simulation experiments demonstrate that the proposed method is effective and reliable.

The emergence of the Fifth Generation (5G) era has ushered in a new era of diverse business scenarios, primarily characterized by data-intensive and latency-sensitive applications. Edge computing technology integrates the information services environment with cloud computing capabilities at the edge of the network. However, the evolving landscape of business models necessitates a unified edge architecture capable of accommodating diverse requirements, posing substantial challenges for service providers in meeting Service-Level Agreements (SLAs).In response to these challenges, we introduce SLA-ORECS. This innovative framework dynamically allocates dedicated and shared resources within the edge-cloud system to cater to service requests with varying SLAs, thereby facilitating performance isolation. Furthermore, we have developed an optimization algorithm to enhance the efficiency of SLA assurance during request dispatch.The evaluation of SLA-ORECS highlights its noteworthy performance improvements, particularly in terms of system throughput and average time consumption.

Edge computing enhances task reliability by employing redundant task executions across edge nodes. Conventional decentralized task offloading strategies, based on heuristics and game theory, either focus on optimization or are based on unrealistic assumptions. Moreover, existing Deep Reinforcement Learning (DRL) task offloading approaches underperform due to discrepancies between simulated environments and real systems, and overlook task redundancy, thus failing to meet reliability requirements. This paper proposes a DRL-based approach to offload distributed redundant tasks, namely DR-DRL (Decentralized and Redundant DRL), to solve the redundancy problem of distributed task offloading under edge computing. Experimental results show that DR-DRL has about 8% higher task success rate than other benchmark methods.

The rise of video content has significantly driven the growth of mobile data traffic, placing substantial pressure on core networks and adversely affecting user experience. To address these challenges, this paper proposes a Multi-Access Edge Computing (MEC) framework based on Deep Reinforcement Learning (DRL). By leveraging edge servers for video transmission, caching, and transcoding, this framework aims to reduce network load and latency. Our framework particularly considers the variability in video popularity and optimizes caching and transcoding strategies for different bitrate video streams. Using the Dueling Double DQN (D3QN) algorithm, it intelligently makes content caching decisions in scenarios with unknown content popularity, achieving optimal resource utilization. Experimental results demonstrate that the proposed framework enhances caching efficiency and user experience, exhibiting higher cache hit rates and lower caching costs in both the short and long term compared to traditional caching strategies and the Wolpertinger architecture.

As the rising of the Internet of Things (IoT), edge computing is widely adopted in numerous applications. However, current autoscaling tools are not designed for edge applications and can not utilize the heterogeneous resources of edge nodes efficiently. In this paper, we propose a proactive hybrid autoscaler specifically optimized for edge computing scenario. With the Bidirectional Long Short Term Memory (Bi-LSTM) based load prediction model, the proposed autoscaler is able to predict the future workload and perform scaling operation before it arrives. In addition, a overload compensation algorithm is implemented to mitigate the Quality of Service (QoS) decreasing due to under-prediction. Then, a hybrid scaling method is applied to simultaneously modify the number of pods and their resource quotas without restarting. Experimental results with a real-world workload dataset shows the proposed load prediction model has better accuracy compared with the Long Short Term Memory model and the state-of-the-art statistical analysis model, Autoregressive Integrated Moving Average (ARIMA), which is also more than 350 times slower than our model in prediction speed. Finally, evaluation in a real Kubernetes cluster shows that the proposed proactive hybrid autoscaler outperforms the default Horizontal Pod Autoscaler (HPA) of Kubernetes in terms of both QoS and resource utilization efficiency.

With the development of deep learning technology, the detection and classification of distracted driving behaviour requires higher accuracy. Existing deep learning-based methods are computationally intensive and parameter redundant, limiting the efficiency and accuracy in practical applications. To solve this problem, this study proposes an improved YOLOv8 detection method based on the original YOLOv8 model by integrating the BoTNet module, GAM attention mechanism and EIoU loss function. By optimising the feature extraction and multi-scale feature fusion strategies, the training and inference processes are simplified, and the detection accuracy and efficiency are significantly improved. Experimental results show that the improved model performs well in both detection speed and accuracy, with an accuracy rate of 99.4%, and the model is smaller and easy to deploy, which is able to identify and classify distracted driving behaviours in real time, provide timely warnings, and enhance driving safety.

The emergence of Intelligent Connected Vehicles (ICVs) heralds a transformation in modes of transportation. This intricate system capitalizes on a wealth of data generated by various terminals, enabling AI models to streamline numerous applications, ranging from travel assistance and urban road monitoring to navigation path planning. Despite this, the data collected from these disparate terminals frequently encapsulates sensitive information pertaining to the entire ICV ecosystem. This can raise concerns about privacy breaches, especially with regard to geographical locations and personal images. Therefore, it is imperative to devise strategies that not only enable intelligent learning and data sharing across the whole scenario but also preserve privacy. In this paper, we introduce SemanticICV, an endogenous secure full scenario learning framework designed to formulate the semantic logic of ICVs at a foundational level. This semantic abstraction inherently imbues the framework with privacy-preserving attributes. Building upon this, we propose a horizontal semantic sharing method (using federated learning) and a vertical semantic cross-layer linkage method (employing knowledge distillation) integrated within the full scenario learning. These methods serve to enhance the data sharing and innate privacy protection of the full-scenario. We substantiate the efficacy of our proposed solution through meticulous simulation experiments.

Balancing video quality and latency in fluctuating network environments is a challenging task for live streaming. Existing Adaptive Bitrate (ABR) algorithms fail to accurately predict throughput, which impedes their ability to fully utilize bandwidth in dynamic network conditions. Additionally, most ABR algorithms inadequately consider the impact of latency on the Quality of Experience (QoE), leading to poor performance in live streaming scenarios. To address these issues, this paper proposes a new algorithm, DRALVS, which integrates the N-BEATS and deep reinforcement learning to jointly optimize video bitrate selection and playback speed control. The idea of DRALVS is to use the N-BEATS to accurately predict future available throughput and use this prediction as the state input for the next deep reinforcement learning module. This module then selects the appropriate video quality and playback speed based on the current state. Simulation results based on real network traces show that DRALVS outperforms existing rule-based and learning-based ABR algorithms in terms of average QoE and latency.

Service Function Chains (SFCs) forward data traffic through a series of Virtual Network Functions (VNFs) to enhance the flexibility of network services. One of the key challenges in SFCs is the deployment of VNFs and scheduling incoming requests across computational nodes to ensure low latency and high reliability. However, existing studies primarily focus on static networks and assume that all SFC requests are predetermined, which is impractical. In this paper, we consider the high-reliability scheduling of SFCs in a dynamic network environment, specifically within Directed Acyclic Graphs (DAGs), to maximize the number of requests that can be successfully processed while satisfying latency and reliability constraints. To tackle this issue, we propose an efficient approach to determine the redundancy of VNFs under the constraint of latency while maximizing success. Experimental results demonstrate that our proposed solution significantly enhances the success rate and effectively conserves computational resources.

卫星通信可以在光纤或者物理线路无法到达的特殊环境提供稳定的传输通道,在大型自然灾害或者公共事故等场景中受到了广泛应用.运营商承担着国家应急通信保障的重要使命,因此在卫星应急通信系统与设备上投入了巨大的财力人力.文章通过介绍运营商的卫星应急通信系统与设备的相关建设、运行、维护、使用情况,总结相关应急通信设备的使用经验,为运营商应急通信保障提供借鉴.

Network security is the cornerstone of computing power network. It is necessary to improve network security awareness, monitoring, early warning, disposal and evaluation capabilities of computing power network in all aspects. It makes a comprehensive analysis on network security issues in computing power network from dimensions of computing facility security, network facility security, combination and scheduling security, operation service security, data security, etc. It is set up gradually evolving atomic power security capabilities for building a ubiquitous security network computing brain. It identifies data assets through active and passive methods, sorts out data assets through in-depth scanning and information completion, supports preset templates formation according to AI (artificial intelligence) models, regular expression matching, keywords, combination rules, etc. It classifies data according to information sensitivity, and visually displays data in charts. In view of serious security risks faced by computing networking services such as network attacks and data privacy leaks, it creatively proposes introduction of privacy computing, data tagging, full process trust, audit traceability, endogenous security and other technologies to achieve security and credibility of computing networking services.

Various metaverse applications have entered our daily life and show a promising trend that will occupy people's attention in the era of Web3. This makes interoperability across metaverses become one of the fundamental technologies in the context of multiple metaverse platforms. The aim of interoperability is to provide a seamless service for users when their requests interact with multiple metaverses. However, the development of cross-metaverse interoperability is still in its initial stage in both industry and academia. In this article, we review the state- of-the-art cross-metaverse interoperability solutions, which are designed for a dedicated purpose but do not apply to all metaverse platforms. To this end, we propose MetaOpera, a generalized cross-metaverse interoperability protocol. Connecting to MetaOpera by means of wireless communication, users and digital objects across different metaverses that rely on centralized servers or decentralized blockchains are capable of interacting with each other. We also implement a proof-of-concept mechanism for Meta- Opera, aiming at evaluating its performance with a state-of-the-art cross-metaverse solution based on the Sidechains technique. Simulation results demonstrate that the size of cross-metaverse proof and the average latency of cross-metaverse transactions using the proposed solution are about eight to three times smaller, respectively, than those of the Sidechains solution. This article also suggests a number of open issues and challenges faced by cross-metaverse interoperability that may inspire future research.

Natural language understanding (NLU) is the key part of task-oriented dialogue systems. Nowadays, most existing task-oriented NLU models use pre-trained models (PTMs) for semantic encoding, but those PTMs often perform poorly on specific task-oriented dialogue data due to small data volume and lack of domain-specific knowledge. Besides that, most joint modeling models of slot filling and intention detection only use a joint loss function, or only provides a one-way semantic connection, which fails to achieve the interaction of information between the two tasks at a deep level. In this paper, we propose a Domain Augmentation and Bidirectional Stack Propagation (DABP) model for NLU. In the proposed model, we use the masked language model (MLM) task and the proposed part-of-speech tagging task to enhance PTMs with domain-specific knowledge include both implicit and explicit. Besides that, we propose a bidirectional stack-propagation mechanism to propagate the information between the two tasks. Experimental results show that the proposed model can achieve better performance than the state-of-the-art models on the ATIS and SNIPS datasets.