High Communication Latency Research Articles

An edge computing and distributed ledger technology architecture for secure and efficient transportation

Intelligent Transportation Systems (ITS) faces significant challenges in achieving its goal of sustainable and efficient transportation. These challenges include real-time data processing bottlenecks caused by high communication latency and security vulnerabilities related to centralized data storage. We propose a novel architecture that leverages Edge Computing and Distributed Ledger Technology (DLT) to address these concerns. Edge computing pushes cloud services, such as vehicles and roadside units, closer to the data source. This strategy reduces latency and network congestion. DLT provides a secure, decentralized platform for storing and sharing ITS data. Its tamper-proof nature ensures data integrity and prevents unauthorized access. Our architecture utilizes these technologies to create a decentralized platform for ITS data management. This platform facilitates secure processing, storage, and data exchange from various sources in the transportation network. This paper delves deeper into the architecture, explaining its essential components and functionalities. Additionally, we explore its potential applications and benefits for ITS. We describe a case study focusing on a data marketplace system for connected vehicles to assess the architecture’s effectiveness. The simulation results show an average latency reduction of 83.35% for publishing and 87.57% for purchasing datasets compared to the cloud architecture. Additionally, transaction processing speed improved by 18.73% and network usage decreased by 96.67%. The proposed architecture also achieves up to 99.61% reduction in mining centralization.

Scalable and Low-Latency Federated Learning With Cooperative Mobile Edge Networking

Federated learning (FL) enables collaborative model training without centralizing data. However, the traditional FL framework is cloud-based and suffers from high communication latency. On the other hand, the edge-based FL framework that relies on an edge server co-located with mobile base station for model aggregation has low communication latency but suffers from degraded model accuracy due to the limited coverage of edge server. In light of high-accuracy but high-latency cloud-based FL and low-latency but low-accuracy edge-based FL, this paper proposes a new FL framework based on cooperative mobile edge networking called cooperative federated edge learning (CFEL) to enable both high-accuracy and low-latency distributed intelligence at mobile edge networks. Considering the unique two-tier network architecture of CFEL, a novel federated optimization method dubbed cooperative edge-based federated averaging (CE-FedAvg) is further developed, wherein each edge server both coordinates collaborative model training among the devices within its own coverage and cooperates with other edge servers to learn a shared global model through decentralized consensus. Experimental results based on benchmark datasets show that CFEL can largely reduce the training time to achieve a target model accuracy compared with prior FL frameworks.

A New Method of Fuzzy Multicriteria Routing in Vehicle Ad Hoc Network

The internet of vehicles (IoVs) provides delay-sensitive services, but high-latency communication with roadside units causes service failures and high costs. Mobile edge computing (EC) is migrating cloud computing platforms from the core network to the edge of mobile networks, giving vehicles local access to numerous computing resources. The classic greedy boundary stateless routing (GPSR) method is widely used to meet the communication requirements of vehicle self-organizing networks. In order to solve the problems of GPSR&#x2019;s neighbor node position acquisition lag and single routing criterion, a new routing method based on the fuzzy logic system Geographic Routing method based on Velocity, Angle, and Density (GRVAD) is proposed. This method gets the relative velocity between the nodes and the angle between the current node, the neighbor node and the target node, and the node density of the neighbor node as the input of fuzzy logic, and the unscented Kalman filter is used to predict the location of the neighbor node to obtain more accurate location information of neighbor nodes. Simulation results show that the routing method compensates for some of the shortcomings of the GPSR method and reasonably considers the delivery rate and end-to-end delay of data packets and is more in line with the communication requirements of the vehicle <i>ad hoc</i> network.

FedSSC: Joint client selection and resource management for communication-efficient federated vehicular networks

As a promising distributed technology, federated learning (FL) has been widely used in vehicular networks involving large amounts of IoT-enabled sensor data, which derives federated vehicular networks (FVNs). However, the efficiency of FVN is generally limited by vehicle selection policy and communication conditions, which leads to high communication costs and latency. The original FVN transmits model parameters from a random subset of vehicles to the roadside unit (RSU) and ignores the diversity of learning quality among vehicles. In addition, a few vehicles with poor wireless channel conditions may prolong communication latency. To address these two issues, we propose a communication-efficient federated learning approach, composed of Vehicle Selection, Student–Project Allocation (SPA) matching model, and Convex Optimization, called FedSSC, to improve the communication efficiency in FVN. The parameter variations for the same vehicle in two consecutive rounds are used to quantify the quality of the learning results by cosine distance and Affinity Propagation (AP) clustering. Moreover, a subchannel allocation algorithm based on the SPA matching model, as well as a convex optimal power allocation solution are integrated to minimize the communication latency of each training round. According to extensive experiments, the proposed FedSSC reduces the communication overhead by 26.32% on average compared with the benchmarks, whereas the communication latency decreases by 21.84%.

Propose technology and management model of mobile cloud computing for the Vietnam national digital transformation program

Cloud computing is considered a digital platform to develop mobile applications for the National Digital Transformation Program 2025 with a vision to 2030. There are many solutions to build a cloud computing platform, but those solutions consume hardware resources, high communication latency and lack of mobility, thus increasing the cost of investment in building a cloud computing platform. To address the disadvantages of the above solutions, this study proposes a mobile cloud computing model aimed at providing services that only implement on mobile devices, enhancing the calculation capacity on mobile devices, minimising resource shortages and performing efficiently for resource-intensive mobile applications. To do that, the solution proposed in this study is to enable the use of the nearest resource-rich mobile devices that can provide computing services to mobile users in the vicinity. Utilising leisure resources to serve resource-intensive mobile applications helps the government to reduce hardware investment for building cloud computing platforms. In addition, this study also proposes a management model for mobile applications through the App Store that is owned and managed by the Vietnamese government. From there, the government will monitor mobile applications and not let them develop spontaneously, causing difficulties and loss of trust for users.

DECCo-A Dynamic Task Scheduling Framework for Heterogeneous Drone Edge Cluster

The heterogeneity of unmanned aerial vehicle (UAV) nodes and the dynamic service demands make task scheduling particularly complex in the drone edge cluster (DEC) scenario. In this paper, we provide a universal intelligent collaborative task scheduling framework, named DECCo, which schedules dynamically changing task requests for the heterogeneous DEC. Benefiting from the latest advances in deep reinforcement learning (DRL), DECCo autonomously learns task scheduling strategies with high response rates and low communication latency through a collaborative Advantage Actor–Critic algorithm, which avoids the interference of resource overload and local downtime while ensuring load balancing. To better adapt to the real drone collaborative scheduling scenario, DECCo switches between heuristic and DRL-based scheduling solutions based on real-time scheduling performance, thus avoiding suboptimal decisions that severely affect Quality of Service (QoS) and Quality of Experience (QoE). With flexible parameter control, DECCo can adapt to various task requests on drone edge clusters. Google Cluster Usage Traces are used to verify the effectiveness of DECCo. Therefore, our work represents a state-of-the-art method for task scheduling in the heterogeneous DEC.

Fog-ROCL: A Fog based RSU Optimum Configuration and Localization in VANETs

Intelligent Transportation Systems (ITS) are one of the pillars of smart cities that enable smart traffic and smart mobility. Vehicular Ad-hoc Networks (VANETs) are utilized as platforms for ITS applications. In VANETs, vehicles collect relevant information from sensors and exchange information about road conditions and traffic status with each other and with roadside units (RSUs). Roadside units facilitate reliable communication among the vehicles and perform real-time processing for the sensed data before sending to the cloud. Although cloud computing offers high performance computational and storage resources, it does not conform to the real-time nature of the ITS applications and the massive amount of data exchange and generation rate due to its centralized nature and high communication latency. In this paper we propose a cost-effective strategy to solve the configuration and localization problem of fog-based RSU deployment in VANETs. The proposed strategy is able to assign the computational capacity of each fog node based on the amount of computational demand inside its coverage region. The problem is formulated as a Satisfiability Modulo Theories (SMT) problem. Our proposed strategy is more efficient than other strategies in terms of the total deployment cost and the overall satisfied computational demand.

MultiFed: A fast converging federated learning framework for services QoS prediction via cloud–edge collaboration mechanism

Federated learning (FL) has become a common approach for distributed training of quality of service (QoS) prediction tasks in smart city solutions. However, FL is vulnerable to heterogeneity. Existing FL-based QoS prediction methods usually update the global model by directly aggregating the divergent local gradients computed on heterogeneous datasets, which does not effectively capture the heterogeneity of users’ QoS data distribution and requires many rounds of training and much communication time to achieve satisfactory accuracy. In addition, the high communication latency between the central server and the client further slows down the convergence process. To address the above issues, we propose a multi-centre federated learning framework for services QoS prediction via a cloud–edge collaboration mechanism, namely MultiFed. In this framework, we sink the central server in traditional FL from cloud to edge and train a global model in each edge region. The edge server performs two gradient aggregation strategies to update the global model in each training round. One is an internal aggregation strategy for regional users; the other is an external aggregation among edge servers with the assistance of a cloud server. Extensive experiments have shown that 9.52%∼40.00% reduces the communication delay, and 1.91%∼38.89% reduces the communication rounds required to reach the target prediction in this method compared to existing methods, validating the effectiveness of MultiFed.

Offloading Using Traditional Optimization and Machine Learning in Federated Cloud–Edge–Fog Systems: A Survey

The huge amount of data generated by the Internet of things (IoT) devices needs the computational power and storage capacity provided by cloud, edge, and fog computing paradigms. Each of these computing paradigms has its own pros and cons. Cloud computing provides enhanced data storage and computing power but causes high communication latency. Edge and fog computing provide similar services with lower latency but limited capacity, capability, and coverage. A single computing paradigm cannot fulfill all the requirements of IoT devices and a federation between them is needed to extend their capacity, capability, and services. This federation is beneficial to both subscribers and providers and also reveals research issues in traffic offloading between clouds, edges, and fogs. Optimization has traditionally been used to solve the problem of traffic offloading. However, in such a complex federated system, traditional optimization cannot keep up with the strict latency requirements of decision-making, ranging from milliseconds to sub-seconds. Machine learning approaches, especially reinforcement learning, are consequently becoming popular because they could quickly solve offloading problems in dynamic environments with some unknown information. This study provides a novel federal classification between cloud, edge, and fog and presents a comprehensive research roadmap on offloading for different federated scenarios. We survey the relevant literature on the various optimization approaches used to solve this offloading problem and compare their salient features. We then provide a comprehensive survey on offloading in federated systems with machine learning approaches and the lessons learned as a result of these surveys. Finally, we outline several directions for future research and challenges that have to be faced in order to achieve such a federation.

Learned Gradient Compression for Distributed Deep Learning.

Training deep neural networks on large datasets containing high-dimensional data requires a large amount of computation. A solution to this problem is data-parallel distributed training, where a model is replicated into several computational nodes that have access to different chunks of the data. This approach, however, entails high communication rates and latency because of the computed gradients that need to be shared among nodes at every iteration. The problem becomes more pronounced in the case that there is wireless communication between the nodes (i.e., due to the limited network bandwidth). To address this problem, various compression methods have been proposed, including sparsification, quantization, and entropy encoding of the gradients. Existing methods leverage the intra-node information redundancy, that is, they compress gradients at each node independently. In contrast, we advocate that the gradients across the nodes are correlated and propose methods to leverage this inter-node redundancy to improve compression efficiency. Depending on the node communication protocol (parameter server or ring-allreduce), we propose two instances for the gradient compression that we coin Learned Gradient Compression (LGC). Our methods exploit an autoencoder (i.e., trained during the first stages of the distributed training) to capture the common information that exists in the gradients of the distributed nodes. To constrain the nodes' computational complexity, the autoencoder is realized with a lightweight neural network. We have tested our LGC methods on the image classification and semantic segmentation tasks using different convolutional neural networks (CNNs) [ResNet50, ResNet101, and pyramid scene parsing network (PSPNet)] and multiple datasets (ImageNet, Cifar10, and CamVid). The ResNet101 model trained for image classification on Cifar10 achieved significant compression rate reductions with the accuracy of 93.57%, which is lower than the baseline distributed training with uncompressed gradients only by 0.18%. The rate of the model is reduced by 8095× and 8× compared with the baseline and the state-of-the-art deep gradient compression (DGC) method, respectively.

ERDAC: Efficient and Reliable Remote Direct Access and Control for Embedded Systems

Emerging embedded systems, such as autonomous vehicles, demand highly efficient remote data transfer, whereas existing networking hardware and protocols cause high communication latency and CPU consumption. In this article, we propose embedded RDAC (eRDAC), an efficient and reliable remote direct access and control solution for embedded systems. The proposed remote access controller in eRDAC has a two-layer protocol offload engine that employs the command/response protocol on UDP to ensure the data reliability and security, and a multichannel DMA controller with configurable priority to improve the efficiency. Besides, a reusable hardware Ethernet MAC is implemented to support not only remote access commands but also standard Ethernet communication. We implement eRDAC on FPGA and the corresponding software in the Linux system. Experimental results show that eRDAC can reduce the latency of remote I/O reading/writing by 74.3%/74.9% ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$3.76\times /3.98\times $ </tex-math></inline-formula> performance improvement) and reduce the latency of remote memory reading and writing with 1024B by 54.2% compared to the socket-based communication. Meanwhile, eRDAC can cut off the consumption of the remote processor and achieve 0.250mJ/Mb energy consumption with only 25-mW power.

Distributed Information Exchange With Low Latency for Decision Making in Vehicular Fog Computing

Traditional decision making in a vehicle network includes uploading vehicle sensing data to faraway cloud platforms and then returning correlated results to the vehicles. The data have features of large quantity and high redundancy, which causes high communication latency and vehicle applications to deteriorate. Vehicular fog computing (VFC) is a new network paradigm that uses local fog nodes for decision making. However, how to achieve distributed information exchange with low latency is a challenging issue because the connectivity of the vehicle network is low due to vehicle mobility. In this article, a distributed information exchange scheme with low latency in VFC is proposed. First, considering the frequent changes in vehicle positions and the randomness in driving routes, public transportation facilities with a wider driving range such as buses and taxis are used as fog nodes to increase the probability of uploading data. Then, the fog nodes should dynamically adjust the data sampling frequency according to the time-space correlation of the data to ensure that only nonredundant data are received. To minimize the interruption latency caused by accidents during an exchange, the fog nodes evaluate and predict connection states among them and their neighboring vehicles when establishing exchanges. If a fog node finds that a vehicle cannot complete information exchange because the vehicle may move outside its communication range in a future period, it will recalculate an optimized relay route for the vehicle by using mixed integer programming. Theoretical analysis and simulation results show that compared with the existing work, the proposed scheme can completely exchange all vehicle data with lower latency.

Enabling Energy Trading in Cooperative Microgrids: A Scalable Blockchain-Based Approach With Redundant Data Exchange

Blockchain has recently been regarded as an important enabler for building secure energy trading in microgrid systems because of its inherent features of distributively providing immutable data record, storage, and sharing across networks in a peer-to-peer (P2P) manner. However, designing highly efficient and scalable blockchain-enabled energy trading mechanisms is extremely challenging because of the unique features of microgrid systems, e.g., bandwidth-constrained and high-latency communications and large-scale renewable energy source (RES) integration. To address this challenge, in this article, we propose a novel scalable blockchain-based energy trading framework for cooperative microgrid systems, which include four planes, i.e., data plane, consensus plane, smart plane, and application plane. Different from the existing solutions without consideration of network transmission, these four planes are designed with the capability of perceiving the status of block generation and transmission over interrupted P2P networks, and thus proactively improving the consensus process to guarantee the reliability of energy trading in cooperative microgrids. Meanwhile, built on this framework, a novel redundant data exchange strategy is proposed to improve the scalability of block creation with the presence of large-scale RES penetration and interrupted and dynamic communication links. Simulation results show that the proposed system framework outperforms the benchmark blockchain solutions. Furthermore, we investigate the potential applications of the proposed solutions in the practical microgrid systems to facilitate a clear understanding of the mechanisms of the proposed solutions.

Civil Aircraft Assisted Space-Air-Ground Integrated Networks: An Innovative NTN of 5G and Beyond

To realize global coverage of 5G, non-terrestrial networks (NTN) is proposed as an vital solution. Satellite based NTN has high communication latency and limited performance. Aerial NTN needs to launch too many airborne communication platforms, which is expensive. Aiming at the problems existing in the classic NTN architecture, civil aircraft assisted space-air-ground integrated network (CAA-SAGIN) is proposed. CAA-SAGIN consists of civil aircrafts and satellites. Civil aircrafts carry airborne base stations, which provide services for users along the route while carrying out tasks. CAA-SAGIN ensures the global coverage of the network, and uses civil aircrafts to provide high throughput and low latency services for hot spot areas. Through the coverage analysis and link budget, the full coverage of CAA-SAGIN for hot spot areas and the reduction of transmission loss are determined. Finally, as an inspiration, the open issues and challenges in the new architecture are summarized.

Secure Protocol for Resource-Constrained IoT Device Authentication

Wireless sensor networks (WSNs) are crucial components of internet of things (IoT) and have been deployed in numerous fields such as battlefield surveillance. The exploitation of broadcasts in WSNs renders these networks susceptible to numerous attacks. Consequently, to boost security, reliability, and successful cooperation, trust must be established among the sensor nodes. Unfortunately, the current authentication and authorization approaches exhibit high key management overheads, depend on static digital signatures or trusted third parties, and have both high communication latencies and computational complexity that render them inefficient. In this paper, challenge-response mutual authentication protocol is proposed for enhancing security in WSN-based IoT environment. The simulation results showed that the proposed protocol has the least transaction costs, time complexity, end to end delays and energy consumptions. It is also resilient against dictionary, side channel, cloning, man-in-the-middle (MitM, denial of service (DoS) and next password prediction attacks.

Implementation and Evaluation of HTTP/3 Connectivity Check Using Happy Eyeballs Algorithm

The HTTP Alternative Services (Alt-Svc) method is defined as an application to check connectivity in HTTP/3. This method is designed based on the fact that communication with old HTTP is guaranteed and the HTTP/3 adoption rate is not necessarily dominant, and it is considered effective in the early stages of transition. However, once HTTP/3 has reached its peak and the transitional period has passed, the uncertainty and redundancy of the Alt-Svc procedure become detrimental. In Alt-Svc, the procedure involves first completing the old HTTP connection to use HTTP/3, and then migrating to HTTP/3 if possible; however, because HTTP/3 is a protocol that eliminates the waste of the old HTTP handshake (TCP handshake followed by TLS handshake), HTTP/3 does not fully benefit from the rapid connection establishment of HTTP/3. Therefore, we propose a method to apply the Happy Eyeballs algorithm, which is used for IPv4 and IPv6 connectivity checks, to the old HTTP and HTTP/3 connectivity checks. The Happy Eyeballs algorithm performs the two selections in parallel to eliminate the delay that occurs in sequential processing, but the proposed method differs from the conventional Happy Eyeballs algorithm in that, even if the old HTTP is adopted once, it switches to the HTTP/3 connection if it is possible to connect using HTTP/3. The proposed method differs from the conventional Happy Eyeballs algorithm by introducing a mechanism to switch to HTTP/3 connections when HTTP/3 connections are available, even when the old HTTP is adopted. Results of the evaluation experiments demonstrated that the adoption rate of HTTP/3 increases in environments with high communication latency because the old HTTP performs the TLS handshake after the TCP handshake, but with this improvement, HTTP/3 is preferentially selected even in low latency environments when it is selectable.

Exploring the Effectiveness of Service Decomposition in Fog Computing Architecture for the Internet of Things

The Internet of Things (IoT) aims to connect everyday physical objects to the internet. These objects will produce a significant amount of data. The traditional cloud computing architecture aims to process data in the cloud. As a result, a significant amount of data needs to be communicated to the cloud. This creates a number of challenges, such as high communication latency between the devices and the cloud, increased energy consumption of devices during frequent data upload to the cloud, high bandwidth consumption, while making the network busy by sending the data continuously, and less privacy because of less control on the transmitted data to the server. Fog computing has been proposed to counter these weaknesses. Fog computing aims to process data at the edge and substantially eliminate the necessity of sending data to the cloud. However, combining the Service Oriented Architecture (SOA) with the fog computing architecture is still an open challenge. In this paper, we propose to decompose services to create <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">linked-microservices</i> (LMS). <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Linked-microservices</i> are services that run on multiple nodes but closely linked to their linked-partners. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Linked-microservices</i> allow distributing the computation across different computing nodes in the IoT architecture. Using four different types of architectures namely cloud, fog, hybrid, and fog+cloud, we explore and demonstrate the effectiveness of service decomposition by applying four experiments to three different type of datasets. Evaluation of the four architectures shows that decomposing services into nodes reduce the data consumption over the network by 10 - 70 percent. Overall, these results indicate that the importance of decomposing services in the context of fog computing for enhancing the quality of service.

Communication-Oriented Model Fine-Tuning for Packet-Loss Resilient Distributed Inference Under Highly Lossy IoT Networks

The distributed inference (DI) framework has gained traction as a technique for real-time applications empowered by cutting-edge deep machine learning (ML) on resource-constrained Internet of things (IoT) devices. In DI, computational tasks are offloaded from the IoT device to the edge server via lossy IoT networks. However, generally, there is a communication system-level trade-off between communication latency and reliability; thus, to provide accurate DI results, a reliable and high-latency communication system is required to be adapted, which results in non-negligible end-to-end latency of the DI. This motivated us to improve the trade-off between the communication latency and accuracy by efforts on ML techniques. Specifically, we have proposed a communication-oriented model tuning (COMtune), which aims to achieve highly accurate DI with low-latency but unreliable communication links. In COMtune, the key idea is to fine-tune the ML model by emulating the effect of unreliable communication links through the application of the dropout technique. This enables the DI system to obtain robustness against unreliable communication links. Our ML experiments revealed that COMtune enables accurate predictions with low latency and under lossy networks.

Cross-Cluster Federated Learning and Blockchain for Internet of Medical Things

Federated learning (FL) has been gaining popularity as a way to provide privacy-preserving data sharing for the Internet of Medical Things (IoMT). As a complementary, blockchain technology is used in recent literature to make FL secure. However, existing blockchain-based FL (BFL) solutions do not perform well when data in a BFL cluster are sparse. A direct solution is to collect as many devices as possible to establish a large BFL cluster. However, these devices may locate in geographically distant areas and be separated by great distance, which further results in high communication latency. The high latency will lead to BFL’s low system efficiency due to frequent communications in the blockchain consensus. In this article, we propose that the large cluster should be divided into multiple smaller clusters, each in its own geographical area and organized with a BFL. In this context, we propose CFL, a cross-cluster FL system facilitated by the cross-chain technique. CFL connects multiple BFL clusters, where only a few aggregated updates are transmitted over long distances across clusters, thus improving the system efficiency. The design of CFL focuses on a cross-chain consensus protocol, which guarantees the model updates to be exchanged securely across clusters. We carry out extensive experiments to evaluate CFL in comparison with BFL, and show both CFL’s feasibility and efficiency.

Reconfigurable Intelligent Surface Enabled Federated Learning: A Unified Communication-Learning Design Approach

To exploit massive amounts of data generated at mobile edge networks, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">federated learning</i> (FL) has been proposed as an attractive substitute for centralized machine learning (ML). By collaboratively training a shared learning model at edge devices, FL avoids direct data transmission and thus overcomes high communication latency and privacy issues as compared to centralized ML. To improve the communication efficiency in FL model aggregation, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">over-the-air computation</i> has been introduced to support a large number of simultaneous local model uploading by exploiting the inherent superposition property of wireless channels. However, due to the heterogeneity of communication capacities among edge devices, over-the-air FL suffers from the straggler issue in which the device with the weakest channel acts as a bottleneck of the model aggregation performance. This issue can be alleviated by device selection to some extent, but the latter still suffers from a tradeoff between data exploitation and model communication. In this paper, we leverage the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">reconfigurable intelligent surface</i> (RIS) technology to relieve the straggler issue in over-the-air FL. Specifically, we develop a learning analysis framework to quantitatively characterize the impact of device selection and model aggregation error on the convergence of over-the-air FL. Then, we formulate a unified communication-learning optimization problem to jointly optimize device selection, over-the-air transceiver design, and RIS configuration. Numerical experiments show that the proposed design achieves substantial learning accuracy improvement compared with the state-of-the-art approaches, especially when channel conditions vary dramatically across edge devices.