Live video streaming has grown significantly, especially on networks with limited bandwidth. Traditional video streaming methods, which drop less important frames, often struggle with high latency. We propose ReparoV2, a novel approach designed specifically for live streaming environments. On the client side, ReparoV2 selectively omits video frames at the source, which substantially reduces bandwidth usage without significantly degrading the QoE. ReparoV2 implements a real time Video Frame Discarding (VFD) algorithm, which categorizes even-indexed frames into three types: preserving, recovering using neural networks and substituting with the previous frame, which is corresponding to the high, medium and low degree of difference between two adjacent frames. Additionally, to gain a better QoE, ReparoV2 integrates an adaptive bitrate strategy and introduces multiple discrete low-frame-rate encoding modes, balancing video quality and bandwidth reduction when the network bandwidth is limited. On the server side, ReparoV2 uses a Video Frame Interpolation Deep Neural Network (VFI-DNN) and frame-copying to reconstruct dropped frames, ensuring a smooth viewing experience. It updates the VFD model based on received video chunks and sends the updated model back to the client. ReparoV2 outperforms conventional Dynamic Adaptive Streaming over HTTP (DASH), achieving higher Structural Similarity Index Measure (SSIM) (+0.024), lower bandwidth usage (-23.19%), and improved QoE (+26.66%) on 25fps videos under 0.974Mbps average bandwidth.

High-definition (HD) maps generated from autonomous vehicle (AV) sensor data are essential for enabling high levels of driving automation. However, offloading large volumes of raw sensory data to edge servers in dense vehicular ad hoc networks (VANETs) introduces significant latency due to network congestion and packet collisions. Existing solutions primarily focus on dynamically adjusting the minimum contention window (CWmin), while additional MAC-layer parameters — including the maximum contention window (CWmax) and interframe space number (IFSn) — remain largely underexplored. To address this, we propose a cross-layer multi-agent reinforcement learning (MARL) framework that jointly optimises CWmin–CWmax, IFSn, and transmission waiting time within IEEE 802.11p-compliant bounds. The proposed multi-task mixed-dependence (MTMD) framework decomposes the optimisation problem into specialised subtasks handled by selectively coupled agents, balancing coordination and scalability while avoiding the overhead of fully symmetric MARL or centralised hierarchical controllers. A lightweight orchestration layer coordinates agent interaction with the simulation environment via secure message exchange. Evaluated against standard EDCA and representative RL baselines, MTMD achieves latency reductions of 31%, 49%, 87.3%, and 64% for Voice, Video, HD Map, and Best-Effort traffic, respectively, confirming the effectiveness of structured multi-parameter optimisation for latency-critical vehicular applications.

The dynamic nature of Low Earth Orbit (LEO) satellite networks introduces unique challenges to satellite edge computing (SEC), particularly regarding handover (HO) strategies for computational task continuity. Conventional HO strategies fail to account for computational task urgency, leading to task timeouts and reduced performance. This paper presents a Risk Predictive Proactive Handover (RPP-HO) mechanism tailored for SEC environments. Unlike conventional reactive approaches, RPP-HO enables satellites to proactively initiate handovers by predicting task timeout risks based on queue congestion and deadline constraints. A multi-step methodology incorporating task-level risk factor evaluation, greedy user selection, and adaptive locking mechanisms ensures task success without excessive handover frequency. Simulation results using real Starlink constellation data demonstrate that RPP-HO improves task success rates by effectively offloading at-risk tasks while maintaining system stability under varying workloads. The proposed strategy offers a lightweight, computation-aware enhancement to standardized Conditional Handover procedures in 6G Non-Terrestrial Networks (NTN).

Predicting vehicle trajectories at roundabouts is crucial for road safety, as it enables advanced driver-assistance systems (ADAS) and autonomous vehicles to anticipate and respond to other drivers' intentions effectively. This capability enhances situational awareness, reduces the risk of collisions, and contributes to smoother traffic flow. This paper presents an open-source dataset designed to predict vehicle turning intentions at roundabouts, integrating YOLOv8 for object detection and DeepSORT for multi-target tracking. The dataset includes vehicle timestamps, pixel coordinates and heading angles, utilizing monocular ranging to map vehicles to an actual coordinate system. It supports real-time collision prediction, driver alerts, and the detection of abnormal behaviors such as sudden lane changes or harsh braking. This dataset contributes to the development of safer and more efficient traffic management and autonomous driving systems. The dataset is available for public access on GitHub11Details of the roundabout video dataset can be found in: https://github.com/zhoudashi2016/Roundabout-Video-Dataset-for-ITS..

Roundabouts present unique challenges in intelligent transportation systems due to their complex geometry, dynamic interactions, and the limited availability of high-quality datasets. Existing methodologies typically lack contextual richness and perform inadequately when addressing the non-linear nature of roundabout behavior. This paper introduces a data augmentation framework that leverages large language models (LLMs), specifically a fine-tuned GPT-2, to enrich roundabout datasets with semantically meaningful behavioral patterns. Our approach initiates with extracting vehicle features from real-world video using YOLOv8 and DeepSORT, subsequently using a feedforward neural network (FNN) to extract three latent feature, and training GPT-2 on this corpus to generate high-level behavioral labels, thereby semantically enriching the dataset's diversity and semantic depth. To validate the effectiveness of our framework, we train a long short-term memory (LSTM) model for trajectory prediction. Evaluation with an LSTM predictor shows that models trained on synthetic GPT-generated data outperform those using original data, with lower average relative error (1.46% vs. 1.72%) and improved accuracy at the 100th percentile (62% vs. 54%). Experimental results show that our augmented synthetic dataset is capable of improving the robustness of prediction, establishing a new paradigm for intelligent traffic modeling through the integration of foundational models. Our code is available at https://github.com/Rebecca689/llm-roundabout.

Viewport in immersive media corresponds to the field of view (FoV), playing a critical role in both data transmission volume and user experience. However, instantaneous and highly dynamic interactions often conflict with segment-based transmission modes, resulting in substantial redundant data transmission and wastage of valuable resources. In this paper, we analyze data from an open-source dataset and our self-collected records to investigate the interactive characteristics of viewers in immersive scenes, including focus time, viewing area scope, movement direction, and tile access probability. Based on empirical statistical inference, we innovatively introduce the concept of an irregular, expandable, and directional extended field of view (EoV) to describe the dynamically variable area mimicking human visual motion. Furthermore, we propose a motion-aware tile-based adaptive control scheme for viewing areas, named VAAC-IM, designed to enable flexible transmission of immersive media. Specifically, we developed an FoV prediction model based on ConvLSTM, leveraging spatiotemporal features from historical viewing records to provide advanced predictions of visual motion preferences. Subsequently, we model the viewing area control process as a constrained submodular minimization problem, dynamically managing irregular EoV area using marginal effects. Finally, we perform a comprehensive validation. The results demonstrate that VAAC-IM significantly enhances performance in terms of reducing black edge coverage, minimizing data volume, lowering latency, and improving overall user experience.

The Domain Name System (DNS) is a growing center of cyber attacks, including both volumetric and non-volumetric attacks. Programmable switches provide a new opportunity for more efficient defense against DNS attacks since they can offer better cost, performance, and flexibility trade-offs compared to traditional defense systems. However, programmable switches have strict limitations on the operations and storage space supported to ensure line-speed packet processing. In this paper, we propose DNSGuard, an intelligent in-network defense framework that can handle volumetric and non-volumetric DNS attacks on programmable switches. We propose a recursive incremental parsing algorithm that can effectively extract variable-length domain names. To achieve real-time and accurate detection against two types of DNS attacks, we design a switch-optimized and resource-efficient algorithm to extract both independent features of each packet and domain-based cumulative features. Then, we propose a multi-phase hybrid model architecture to perform dynamic packet analysis at different time phases of a domain. Further, we design efficient model representation mechanisms to deploy tree-based ensemble models in the data plane. Experimental results show that DNSGuard can defend against diverse DNS attacks at the line rate. In addition, DNSGuard introduces a minimal nanosecond latency to normal traffic in heavily loaded networks.

The video analytics system utilizes deep learning models (DNN) to perform inference on the videos captured by cameras. Continuous learning algorithms are used to address the data drift problem in video analytics systems. However, uploading images from deployment environments and processing on the cloud carry the risk of privacy leakage. In this paper, we have designed a system called CL-Shield to protect user’s privacy. First, we review the causes of privacy leakage in a continuous learning system and propose the objective of full privacy protection. Second, we design an online training mechanism based on a scene library to avoid direct uploading of user’s frames to the cloud server. Lastly, we design a fast training set search algorithm based on a novel Ebv-List, which effectively improves the speed of model updates. We collect various real-world scenario data to build our scene library and validate our system on a dataset of over 10 hours. The experiments demonstrate that our privacy-aware continuous learning system achieves an F1-score of over 92% compared to the conventional systems without protecting privacy and has long-term stability in analytic F1-score.

Deploying deep neural network (DNN) models on resource-constrained edge devices for real-time video analytics poses significant challenges due to the high resource demands of these models. Current edge-based video analytics approaches often overlook optimizing deep learning models and GPU resource allocations in multi-tenant scenarios. In this paper, we present JSAS-MTMGS, a collaborative video analytics system employing three innovative design strategies. First, we propose a novel video configuration optimization space based on a joint DNN model sharing and splitting scheme to balance computational loads for collaborative processing. This approach reduces network transmission data volume and alleviates resource contention. Second, we design a GPU resource allocation scheme that combines GPU batching with spatial sharing to optimize GPU utilization and increase system throughput, all without relying on costly offline latency collection. Finally, we define the configuration optimization problem alongside GPU allocation as a convex problem and apply convex optimization to make scheduling decisions dynamically. Our experiments demonstrate that JSAS-MTMGS has the best service quality among all compared algorithms.

Offloading machine learning models for network classification on high-throughput programmable switches is a promising technology, enabling line-speed in-network classification. Existing solutions are centralized, deploying a complete but heavy model on a single switch with limited hardware resources, causing unsatisfactory accuracy, network-wide resource wastage, and non-generic single-task classification. Therefore, we propose In-Forest-M, a general distributed multi-task in-network classification framework. Firstly, we develop a Lightweight Ensemble Generic Optional Model (LEGO), which can be transformed into base models with full functionality. Each switch only needs to deploy lightweight base models rather than complete ensemble models. The significant reduction in resource consumption allows the deployment of larger models with higher accuracy and more models that support diverse tasks. We employ a fine-grained enhancement mechanism to enhance the classification performance of base models. As traffic traverses different switches, In-Forest-M aggregates the classification results of multiple enhanced base models to improve accuracy further. Secondly, we introduce a two-phase resource-aware model allocation strategy that assigns different task-specific enhanced base models to switches under resource constraints and task requirements. To respond to dynamic traffic changes, we design an optimization-driven reinforcement learning algorithm. Moreover, we propose a lightweight update mechanism for flexible model scaling. Comprehensive experiments reveal that, compared with state-of-the-art in-network classification solutions in three real network topologies, In-Forest-M achieves increased accuracy and reduced switch rules while exhibiting great generality in multi-task classification.

Automatic modulation classification (AMC) plays an indispensable role in wireless communication systems. Deep learning-based AMC has become the mainstream solution due to its high accuracy and no need for manual feature engineering. However, every coin has two sides. DL-based AMC is susceptible to adversarial perturbations, which are carefully crafted to be superimposed on the transmitted signals in an iteratively try-and-error manner, resulting in incorrect classification. In this paper, we propose a model diversity-based moving target defense mechanism (MD-MTD), which employs multiple classifiers and switches periodically, preventing intelligent attackers from deducing universal adversarial perturbations (UAP). Besides, to jointly optimize the robustness and accuracy of different AMC models to be trained, we design a novel multi-agent reinforcement learning (MARL) module. It is worth mentioning that the proposed algorithm significantly mitigates the curse of dimensionality during the large-scale training process via integrating value-decomposition networks and illegal action masking, improving the feasibility of our solution in real-world wireless communication systems. Experimental results on the GNU radio dataset also exhibit the remarkable advantages of our method in terms of convergence and defense performance.

Deep learning (DL) has been recently used for malicious traffic detection. However, DL models are often faced with a dilemma between model size and performance: larger models have better accuracy, but suffer from high detection latency, which severely impacts realtime traffic performance, while lightweight models have low detection latencies, but sacrifice accuracy. In this paper, we introduce Proteus, a swift and precise attack detection framework that adaptively adjusts DL models in real-time based on sample detection difficulty. To address diverse detection difficulties in traffic data, we devise a Double Dynamic Convolutional Neural Network (DDCN) with two pivotal modules: the Dynamic Feature Campaign (DFC) and the Tailor Module (TM). DFC enables the model to discern and accentuate the most influential features, while TM autonomously gauges sample difficulty, cropping the overall model. We further design an auxiliary detection module to streamline the detection, especially for network devices like routers lacking GPUs but equipped with multiple CPU cores. Experiments on different network devices show that Proteus completes the detection of each flow within 0.6 ms, and achieves$\mathbf{9 9. 3 4 \%}$detection accuracy, outperforming other solutions.

The adoption of High-Definition (HD) mapping applications represents a critical step towards achieving Level-5 autonomous driving, revolutionizing road safety and paving the way for unprecedented advancements in transportation technology. However, HD mapping imposes significant computational demands in processing the raw data generated by autonomous vehicle sensors. To mitigate this issue, researchers have opted to offload the data reducing the processing time. Unfortunately, the current de-facto standard IEEE802.11p in Vehicular Ad-hoc Network (VANET) does not provide the best latency or throughput for applications with low latency and heavy data transfer requirements. This is because of the fixed Contention Window (CW). To address this problem, solutions have been developed to dynamically allocate the CW nowadays with the help of Machine Learning (ML) paradigms. Nonetheless, these solutions do not include a strategy to dynamically allocate an optimal CW per service type. Instead, they focus on sharing the wireless channel fairly. In this paper, we have developed a cross-layer Reinforcement Learning (RL) algorithm between the application and Medium Access Control (MAC) layer that allocates CW per service type. Results showed improvement with a different gap in the latency Cumulative Distribution Function (CDF) of 181%, 120%, 107%, and 119% for Voice, Video, HD Map, and Best-effort respectively compared to other different approaches.

Video analytics systems use deep learning models to perform inference on videos and are widely applied in fields such as smart cities and robotics. However, in order to improve inference speed, models with fewer parameters are often chosen to deploy on edge devices, thereby sacrificing the detection accuracy of the task. Continuous learning is an emerging approach to improve the accuracy of lightweight models deployed on the edge for video inference. However, most solutions constantly upload data from edge to the cloud without considering the limited resources and latency of system modules, resulting in the unreliability of the system and loss of model accuracy. This paper investigates the impact of retraining frequency and video encoding configurations on continuous learning. We then design an adaptive streaming continuous learning algorithm (ASCL) with the aim of achieving the desired level of accuracy while minimizing bandwidth resources as much as possible. First, ASCL can adaptively start up according to the needs of users. Second, during the retraining, an adaptive profiling method is designed to select the appropriate encoding configurations to ensure high profiling accuracy. Third, we perform a layer-wise downloading streaming algorithm to ensure secure and smooth transmission. Real-world network traces are driven to the overall evaluation of ASCL. The results of a multitude of videos show the advantages of ASCL over traditional baselines.

Non-terrestrial networks are regarded as crucial infrastructure in forthcoming 6G networks. Edge caching services can be deployed on satellites to optimize the transmission delay of non-terrestrial networks. The dynamically changing user requests and the limited cache space challenge the decision-making process for satellite caching. This paper proposes a Handover-Aware Cache Replacement (HACR) strategy to dynamically replace cache content with satellite mobility. The strategy incorporates the deterministic mobility of satellites to make optimal caching decisions over time and utilizes deep reinforcement learning to implement the strategy. We develop a simulation platform based on the configuration of the Starlink constellation and compare our proposed strategy with conventional methods. The results demonstrate that HACR achieves a 16.74% improvement in cache hit rate and a 48.9 % improvement in stability of cache hit rate compared to the state-of-the-art cache strategy.

One effective way to optimize the offloading process is by minimizing the transmission time. This is particularly true in a Vehicular Adhoc Network (VANET) where vehicles frequently download and upload High-definition (HD) map data which requires constant updates. This implies that latency and throughput requirements must be guaranteed by the wireless system. To achieve this, adjustable contention windows (CW) allocation strategies in the standard IEEE802.11p have been explored by numerous researchers. Nevertheless, their implementations demand alterations to the existing standard which is not always desirable. To address this issue, we proposed a Q- Learning algorithm that operates at the application layer. Moreover, it could be deployed in any wireless network thereby mitigating the compatibility issues. The solution has demonstrated a better network performance with relatively fewer optimization requirements as compared to the Deep Q Network (DQN) and Actor-Critic algorithms. The same is observed while evaluating the model in a multi-agent setup showing higher performance compared to the single-agent setup.

Drones connected via the web are increasingly being used for crowd anomaly detection (CAD). Existing solutions, however, face many challenges, such as low accuracy and high latency due to drones' dynamic shooting distances and angles as well as limited computing and networking capabilities. In this paper, we propose Air-CAD, an edge-assisted multi-drone network that uses air-ground cooperation to achieve fast and accurate CAD. Air-CAD consists of two stages: person detection and multi-feature analysis. To improve CAD accuracy, Air-CAD dynamically adjusts the inference of person detection model based on drones' shooting distances and assigns appropriate feature analysis tasks to drones shooting at variable angles. To achieve fast CAD, edge devices connected to drones are deployed to offload assigned feature analysis tasks from drones. Air-CAD schedules the connection between each drone and edge to accelerate processing based on drone's assigned task and the computing/network resources of the edge device. To validate the performance of Air-CAD, we generate a new simulated human stampede dataset captured from various drone-view recordings. We deploy and evaluate Air-CAD in both simulation and real-world testbed. Experimental results show that Air-CAD achieves 95.33% AUROC and real-time inference latency within 0.47 seconds.

With the rapid popularity of short video applications, a large number of short video transmissions occupy the bandwidth, placing a heavy load on the Internet. Due to the extensive number of short videos and the predominant service for mobile users, traditional approaches (e.g., CDN delivery, edge caching) struggle to achieve the expected performance, leading to a significant number of redundant transmissions. In order to reduce the amount of traffic, we design a Novel Coded Transmission Mechanism (NCTM), which transmits XOR-coded data instead of the original video content. NCTM caches the short videos that users have already watched in user devices, and encodes, multicasts, and decodes XOR-coded files separately at the server, edge nodes, and clients, with the assistance of cached content. This approach enables NCTM to deliver more short video data given the limited bandwidth. Our extensive trace-driven simulations show how NCTM reduces network load by 3.02%-14.75%, cuts peak traffic by 23.01%, and decreases rebuffering events by 43%-85% in comparison to a CDN-supported scheme and a naive edge caching scheme. Additionally, NCTM also increases the user's buffered video duration by 1.21x-13.53x, ensuring improved playback smoothness.

Unmanned Surface Vehicles (USVs) in inland waterways have drawn increasing attention for their excellent capability to serve maritime time-consuming missions such as autonomous navigation and intelligent monitoring. However, USVs struggle to accomplish emerging computation-intensive tasks (e.g., sensor, telemetry, etc) timely due to the limited on-board resources. This paper proposes a novel reconfigurable intelligent surface (RIS)-assisted unmanned aerial vehicle (UAV) multi-access edge computing (MEC) network architecture to support low-latency USVs data computation with time window. Aiming to enhance USVs task processing efficiency, the minimization of USVs task processing time is formulated by jointly considering UAVs flight route selection, USVs execution mode selection, UAVs hovering coordinates and RIS phase shift vector. A heuristic solution is proposed to tackle the formulated challenging problem iteratively. The original problem is decoupled into three subproblems: an enhanced deferred acceptance algorithm is proposed to solve UAVs flight route selection subproblem; an enhanced Lagrangian relaxation method is proposed to solve USVs execution mode selection subproblem; a joint alternating direction method of multipliers (ADMM)-successive convex approximation (SCA)-based algorithm is proposed to solve UAVs hovering coordinates subproblem. Experiment results demonstrate that the proposed solution can decrease task processing time by approximately 54% compared with numerous selected advanced algorithms. Moreover, the performance of the proposed solution under typical UAVs caching capability and the number of UAVs has been investigated.

5G and Beyond 5G communication technologies have promoted the architectural innovation of the Industrial Internet of Things (IIoT) and the wide application of edge learning. As Beyond 5G technologies enhance wireless communication within IIoT, the demand for efficient, secure data management becomes paramount. Edge learning emerges as a solution for localized model training, reducing the necessity for extensive data transmission. However, this decentralization introduces vulnerabilities, particularly in data security during transmission and efficient resource utilization. To address the challenges of data scheduling for edge learning in the Wireless IIoT (WIIoT), we propose a novel architecture that leverages blockchain for secure, decentralized data scheduling and employs physically unclonable functions (PUFs)-based algorithm to ensure data integrity and confidentiality. The primary contributions consist of a task scheduling model based on blockchain, along with a data compression scheme in multiple stages combined with a data scheduling algorithm that is optimized for energy efficiency in edge learning environments. Experiments conducted on a simulated WIIoT platform comprising embedded devices validate our approach, demonstrating enhanced data security and learning efficiency which can reduce 40% in the training stage and 70% in the inference stage. Our findings contribute to the advancement of security and efficient edge learning frameworks in the context of WIIoT, addressing the intricate balance between security, efficiency, and decentralized trust.

Fairness-aware federated learning (FFL) plays a crucial role in mitigating bias against specific demographic groups (e.g., gender, race, occupation) during collaborative training. Along with the ever-emerging new attack paradigms like gradient leakage and model poisoning, the reliability of FFL also obtains lots of research attention. Either UAV nodes or FFL aggregators could be untrusted adversaries. Although multiple security mechanisms involving encryption, obfuscation, Byzantine-robustness, and detection have been proposed, concrete to UAV networks, the majority of existing solutions are unfeasible due to high heterogeneity and limited resources among participants. Hence, in this paper, we propose mutually reliable FFL (MR-FFL), a stratified community-based framework to facilitate privacy protection (FFL aggregator’s reliability) and poisoning elimination (client nodes’ reliability) jointly for FFL in heterogeneous UAV networks. We first divide UAV nodes into both peer communities (PC) and colleague communities (CC) according to cross-participant similarity and task-oriented fitness, respectively. Thus, the arbitrarily settled learning tasks following fair principles can be efficiently completed by fine-tuned colleague communities, even in the presence of a large degree of heterogeneity among peer communities. Then, we integrate community-specific differential privacy into the MR-FFL process, to achieve privacy amplification as well as efficient and personal collaborative training at the same time. More importantly, we proposed a community-based credit evaluation to resist poisoning attacks in heterogeneous environments. The results on several standard datasets also highlight the performance of MR-Fed in terms of fairness, accuracy, and integrity jointly.

Neural networks (NNs) are widely used in classification-based networking analysis to help traffic transmission and system security. However, there are heterogeneous network devices (e.g., switches and routers) in a network. Manually customizing NNs with specific device requirements (e.g., max allowed running latency) can be time-consuming and labor-intensive. Furthermore, the diverse data characteristics of different networking classification tasks add to the burden of NN customization. This paper introduces Loong, a neural architecture search (NAS) based system that automatically generates NNs for various networking tasks and devices. Loong includes a neural operation embedding module, which embeds candidate neural operations into the layer to be designed. Then, the layer-wise training is used to generate a task-specific NN layer by layer. This layer-wise scheme simultaneously trains and selects candidate neural operations using gradient feedback. Finally, only the important operations are selected to form the layer, maximizing accuracy. By incorporating multiple objectives, including deployment memory and running latency of devices, into the training and selection of NNs, Loong is able to customize NNs for heterogeneous network devices. Experiments show that Loong's NNs outperform 13 manual-designed and NAS-based NNs, with a 4.11% improvement in F1-score. Additionally, Loong's NNs achieve faster (7.92X) speeds on commodity devices.

Real-time video analytics on edge devices has gained increasing attention across a wide range of business areas. However, edge devices usually have limited computing resources. Consequently, conventional approaches to video analytics either deploy simplified models on the edge (resulting in low accuracy) or transmit video content to the cloud (resulting in high latency and network overheads) to enable deep learning inference (e.g., object detection). In this paper, we introduce ParaLoupe, a novel real-time video analytics system that parallelizes deep learning inference in the edge cluster with task-oriented mini models. These mini models do not attain State-of-the-Art accuracy individually, but collectively can achieve much better accuracy-latency tradeoff than State-of-the-Art models. To achieve this, ParaLoupe crops multiple single-object patches from a given video frame. These single-object patches are then sent to multiple edge devices for parallel inference with specifically designed mini models. A patch-based task scheduling algorithm is further proposed to leverage the computing resources of the edge cluster to meet the service-level objectives. Our experimental results on real-world datasets show that ParaLoupe significantly outperforms baseline methods, achieving up to 14.1× inference speedup with accuracy on par with state-of-the-art models, or improving accuracy up to 45.1% under the same latency constraints.

The optimal read strategy for strong consistent key-value applications is to enable the per-replica local reads that each replica has the ability to serve reads locally. Unfortunately, current schemes for the per-replica local reads are perplexed by two issues. First, some schemes have to violate the per-replica local reads when the workload is skewed, degrading the throughput. Second, most of current schemes rely on leases or a specialized hardware to guarantee the linearizability, bringing difficulties to the deployment.

SummaryThe mobile video quality can be improved by video super-resolution (SR) especially when bandwidth is limited. To achieve real-time SR, the latest work, ClassSR (CVPR 19), divides frames into equal-size image blocks (IBs), and different-complexity SR models are used respectively to reduce the computational burden.

A variety of model representation methods have been used in recent works to translate machine learning models into programmable switch rules to address network classification tasks at line-speed, i.e., in-network classification. These works generally deploy a complete but heavy model on a switch with limited hardware resources, causing both network-wide waste of resources and unsatisfactory accuracy. Therefore, we propose In-Forest, a general distributed in-network classification framework. Firstly, to improve accuracy with limited resources, we develop a Lightweight Ensemble Generic Optional Model (LEGO), which can be further enhanced into multiple enhanced base models with full functionality. Each switch only needs to deploy a simple base model, rather than the complete ensemble model. Thus, hardware resources required for both switches and the entire network can be significantly reduced. Secondly, as traffic traverses multiple switches, In-Forest aggregates the classification results from different enhanced base models for higher accuracy. Furthermore, we design a two-phase resource-aware model allocation strategy that assigns enhanced base models to switches under different scenarios. We use stable deep reinforcement learning to respond to dynamic traffic changes. Experimental results show that when compared to SwitchTree, Planter, and Netbeacon in two real network topologies, In-Forest can increase accuracy by up to 19.31%, while reducing the number of switch rules by 89.98%.

Decision trees (DT) have been used for high-speed networking classification on programmable switches. Most DT solutions, however, are static and cannot be deployed once the switch resource changes. In this paper, we propose Dryad to fast reprogram tree models when resource budgets change. In Dryad, we first develop a large and accurate “one-training-for-all“ DT (ODT) that can be quickly resized without computational retraining. ODTs are deployed in switches using a progressive search algorithm that searches the adaptations according to their resources. To achieve high accuracy and low packet latency, the adaptation leverages 1) innovative hard and soft pruning methods to compress the ODT rapidly with minimal performance loss; and 2) P4 scaling operations of match-action table arrangement and joint range-ternary match, which allow the switch to accommodate a larger (i.e., more accurate) ODT. Finally, an ODTCompiler is proposed to automatically convert the adapted ODT into a P4 program and then install it. Experimental results on three commodity switches under different resource scenarios show that Dryad achieves a higher classification F1-score (3.78 % higher), and completes the adaptation 161 × faster than other solutions.

High-definition (HD) map is a key enabler to achieving fully autonomous driving. Unlike traditional multimedia data, HD map consists of hybrid data types including 3D point clouds, images, GPS, etc. However, transporting HD map data to and from autonomous vehicles is challenging. IEEE 802.11p Vehicular ad-hoc networks (VANETs) are the de facto standard to establish short-range communications for vehicle-to-everything (V2X). One of the key limitations of 802.11p is ensuring quality of service for HD map traffic, since it might be categorised as best-effort (low priority) between all four access categories (AC) best-effort, background, voice, and video. We proposed a new AC for HD map traffic in this paper to address the aforementioned limitation. We also demonstrate the benefits of using the new AC, as well as the importance of selecting the appropriate channel control parameters within the AC, such as contention window (CW) and Arbitrary Inter-Frame Space (AIFS). Various values of CW and AIFS were examined under dynamic vehicular density and mobility to investigate end-to-end latency and throughput. Experimental results reveal that both, the average delay and throughput of HD map traffic improved by 80%, with the new AC. Furthermore, the delay manifested a steady behavior of 2.3 seconds for thirty, forty, and fifty vehicles. For the selection of AFS and CW parameters, a correlation is observed between vehicular mobility and density.

The user experience of mobile web video streaming is often impacted by insufficient and dynamic network bandwidth. In this paper, we design Bidirectionally Optimized Super-Resolution (BiSR) to improve the quality of experience (QoE) for mobile web users under limited bandwidth. BiSR exploits a deep neural network (DNN)-based model to super-resolve key frames efficiently without changing the inter-frame spatial-temporal information. We then propose a downscaling DNN and a mobile-specific optimized lightweight super-resolution DNN to enhance the performance. Finally, a novel reinforcement learning-based adaptive bitrate (ABR) algorithm is proposed to verify the performance of BiSR on real network traces. Our evaluation, using a full system implementation, shows that BiSR saves 26% of bitrate compared to the traditional H.264 codec and improves the SSIM of video by 3.7% compared to the prior state-of-the-art. Overall, BiSR enhances the user-perceived quality of experience by up to 30.6%.

Edge-cloud collaborative video analytics is transforming the way data is being handled, processed, and transmitted from the ever-growing number of surveillance cameras around the world. To avoid wasting limited bandwidth on unrelated content transmission, existing video analytics solutions usually perform temporal or spatial filtering to realize aggressive compression of irrelevant pixels. However, most of them work in a context-agnostic way while being oblivious to the circumstances where the video content is happening and the context-dependent characteristics under the hood. In this work, we propose VaBUS, a real-time video analytics system that leverages the rich contextual information of surveillance cameras to reduce bandwidth consumption for semantic compression. As a task-oriented communication system, VaBUS dynamically maintains the background image of the video on the edge with minimal system overhead and sends only highly confident Region of Interests (RoIs) to the cloud through adaptive weighting and encoding. With a lightweight experience-driven learning module, VaBUS is able to achieve high offline inference accuracy even when network congestion occurs. Experimental results show that VaBUS reduces bandwidth consumption by 25.0%-76.9% while achieving 90.7% accuracy for both the object detection and human keypoint detection tasks.

The Multi-access Edge Computing (MEC) paradigm offers cloud-computing support to rich media applications, including Dynamic Adaptive Streaming over HTTP (DASH)-based ones at the edge of the network, close to mobile users. MEC servers, typically deployed at base stations (BS), help reduce latency and improve quality of experience (QoE) of video streaming. Unfortunately the communications involving mobile users require handovers between BSs and these influence both transmission efficiency because of the relative position of the MEC servers and transit cost. At the same time, serving MEC for a mobile user should not necessarily be changed when handover occurs. This paper introduces QoE Ready to Respond (QoE-R2R), a QoE-aware MEC Selection scheme for DASH-based mobile adaptive video streaming for optimizing video transmission in a MEC-supported network environment. Simulation-based testing shows that the proposed (QoE-R2R) scheme outperforms some traditional alternative solutions. Compared to hit rate and delay-based schemes, QoE-R2R reduces by 27.6% transmission time and improves with 6.2% QoE.

With the growing popularity of live streaming, high video quality and low latency with limited uplink bandwidth have become a significant challenge. In this study, we propose Live Super-Resolution Based Video Coding (LiveSRVC), a novel video uploading framework that improves the quality of live streaming with low latency under limited uplink bandwidth. We design a new super-resolution-based key frame coding module to improve the coding compression efficiency. LiveSRVC dynamically selects the bitrate and the compression ratio of key frames, mitigating the influence of uplink bandwidth capacity on live streaming quality. Trace-driven emulations verify that LiveSRVC can provide the same quality while reducing up to 50% of the required bandwidth compared to the original encoding method (H.264). LiveSRVC consumes at least 10X less GPU occupation time compared to the method of reconstructing all frames with super-resolution.

Multimedia synchronization involves a temporal relationship between audio and visual media components. The presentation of “in-sync” data streams is essential to achieve a natural impression, as “out-of-sync” effects are often associated with user quality of experience (QoE) decrease . Recently , multi-sensory media (mulsemedia) has been demonstrated to provide a highly immersive experience for its users. Unlike traditional multimedia, mulsemedia consists of other media types (i.e., haptic, olfaction, taste, etc.) in addition to audio and visual content. Therefore, the goal of achieving high quality mulsemedia transmission is to present no or little synchronization errors between the multiple media components. In order to achieve this ideal synchronization, there is a need for comprehensive knowledge of the synchronization requirements at the user interface. This paper presents the results of a subjective study carried out to explore the temporal boundaries within which haptic and air-flow media objects can be successfully synchronized with video media. Results show that skews between sensorial media and multimedia might still give the effect that the mulsemedia sequence is “in-sync” and provide certain constraints under which synchronization errors might be tolerated. The outcomes of the paper are used to provide recommendations for mulsemedia service providers in order for their services to be associated with acceptable user experience levels, e.g. haptic media could be presented with a delay of up to 1 s behind video content, while air-flow media could be released either 5 s ahead of or 3 s behind video content.

Multiple sensorial media (mulsemedia) combines multiple media elements which engage three or more of human senses, and as most other media content, requires support for delivery over the existing networks. This paper proposes an adaptive mulsemedia framework (ADAMS) for delivering scalable video and sensorial data to users. Unlike existing two-dimensional joint source-channel adaptation solutions for video streaming, the ADAMS framework includes three joint adaptation dimensions: video source, sensorial source, and network optimization. Using an MPEG-7 description scheme, ADAMS recommends the integration of multiple sensorial effects (i.e., haptic, olfaction, air motion, etc.) as metadata into multimedia streams. ADAMS design includes both coarse- and fine-grained adaptation modules on the server side: mulsemedia flow adaptation and packet priority scheduling. Feedback from subjective quality evaluation and network conditions is used to develop the two modules. Subjective evaluation investigated users' enjoyment levels when exposed to mulsemedia and multimedia sequences, respectively and to study users' preference levels of some sensorial effects in the context of mulsemedia sequences with video components at different quality levels. Results of the subjective study inform guidelines for an adaptive strategy that selects the optimal combination for video segments and sensorial data for a given bandwidth constraint and user requirement. User perceptual tests show how ADAMS outperforms existing multimedia delivery solutions in terms of both user perceived quality and user enjoyment during adaptive streaming of various mulsemedia content. In doing so, it highlights the case for tailored, adaptive mulsemedia delivery over traditional multimedia adaptive transport mechanisms.

Multipath TCP (MPTCP) is an evolution of the regular TCP that allows multiple radio interfaces to be used simultaneously by a single connection while presenting regular TCP interface to applications. Although its benefits include better resource utilization, higher throughput and smoother reaction to connection failures, MPTCP does not take energy consumption into account, especially important when using wireless mobile devices with limited power resources. In this paper, we demonstrate that smartphones with MPTCP support consume more energy than those with regular TCP when using the same service and the same network interface. Additionally, novel energy consumption models are developed based on real life measurements on a real life smartphone. The proposed energy consumption models consider four different multimedia-based services (i.e. video streaming, voice over IP, web-browsing and file download) in 3G and WiFi networks when MPTCP or regular TCP are used, respectively.

Energy consumption is a key issue that impacts on user quality of experience when delivering rich media services to smartphones via heterogeneous wireless networks. Previous research works have studied the smartphone energy consumption in a broad manner only. This paper focuses on comparative energy consumption investigation of rich media transmissions over 3G and WiFi networks involving a real life smartphone device. In particular, the energy consumption of the CPU and radio interfaces (i.e. HSDPA and WiFi) for different rich media services (i.e. video streaming, interactive video call, file download, web-browsing) is recorded. The results obtained show how deliveries over the WiFi interface are more energy efficient than those over the 3G interface (i.e. up to 36.5% for downloading service). Additionally, the difference between the energy consumption when employing WiFi and 3G is the lowest and highest for web-browsing and file downloading services, respectively. Outcome of the investigation can provide beneficial input for smartphone energy optimization solutions.

Traditional video sequences make use of both visual images and audio tracks which are perceived by human eyes and ears, respectively. In order to present better ultra-reality virtual experience, the comprehensive human sensations (e.g. olfaction, haptic, gustatory, etc) needed to be exploited. In this paper, a multiple sensorial media (mulsemedia) delivery system is introduced to deliver multimedia sequences integrated with multiple media components which engage three or more of human senses such as sight, hearing, olfaction, haptic, gustatory, etc. Three sensorial effects (i.e. haptic, olfaction, and air-flowing) are selected for the purpose of demonstration. Subjective test is conducted to analyze the user perceived quality of experience of the mulsemedia service. It is concluded that the mulsemedia sequences can partly mask the decreased movie quality. Additionally the most preferable sensorial effect is haptic, followed by air-flowing and olfaction.

User Quality of Experience (QoE) is of fundamental importance in multimedia applications and has been extensively studied for decades. However, user QoE in the context of the emerging multiple-sensorial media (mulsemedia) services, which involve different media components than the traditional multimedia applications, have not been comprehensively studied. This article presents the results of subjective tests which have investigated user perception of mulsemedia content. In particular, the impact of intensity of certain mulsemedia components including haptic and airflow on user-perceived experience are studied. Results demonstrate that by making use of mulsemedia the overall user enjoyment levels increased by up to 77%.

IETF-proposed Multipath TCP (MPTCP) extends the standard TCP and allows data streams to be delivered across multiple simultaneous connections and consequently paths. The multipath capability of MPTCP provides increased bandwidth for applications in comparison with the classic single-path TCP, which makes it highly attractive for the current consumer mobile devices that support more than one radio interfaces (e.g. 3G, WiFi, Bluetooth, etc.). However, MPTCP does not consider energy consumption aspects which are highly important for these devices. This paper proposes eMTCP, a novel energy-aware MPTCP-based content delivery scheme which balances support for increased throughput with energy consumption awareness. eMTCP is located at upper transport layer in mobile devices and requires no additional modifications of the MPTCP-enabled server. eMTCP increases the energy efficiency of mobile devices by offloading traffic from the more energy-consuming interfaces to others. Simulation-based experiments employing an eMTCP model which sends data streams via the 3GPP Long Term Evolution (LTE) and IEEE 802.11 (WiFi) interfaces show an increase of up to 14% in energy efficiency when using eMTCP in comparison with MPTCP and of up to 66% in terms of quality in comparison with single-path TCP.

IEEE 802.11e protocol enables QoS differentiation between different traffic types, but requires MAC layer support and assigns traffic with static priority. This paper proposes an intelligent Prioritized Adaptive Scheme (iPAS) to provide QoS differentiation for heterogeneous multimedia delivery over wireless networks. iPAS assigns dynamic priorities to various streams and determines their bandwidth share by employing a probabilistic approach-which makes use of stereotypes. Unlike existing QoS differentiation solutions, the priority level of individual streams in iPAS is variable and considers service types and network delivery QoS parameters (i.e. delay, jitter, and packet loss rate). A bandwidth estimation technique is adopted to provide network conditions and the IEEE 802.21 framework is used to enable control information exchange between network components without modifying existing MAC protocol. Simulations and real life tests demonstrate how better results are obtained when employing iPAS than when either IEEE 802.11 DCF or 802.11e EDCA mechanisms are used. The iPAS key performance benefits are as follows: 1) better fairness in bandwidth allocation; 2) higher throughput than 802.11 DCF and 802.11e EDCA with up to 38% and 20%, respectively; 3) enables definite throughput and delay differentiation between streams.