Vehicular edge intelligence (VEI) is a promising paradigm for enabling future intelligent transportation systems by accommodating artificial intelligence (AI) at the vehicular edge computing (VEC) system. Federated learning (FL) stands as one of the fundamental technologies facilitating collaborative model training locally and aggregation, while safeguarding the privacy of vehicle data in VEI. However, traditional FL faces challenges in adapting to vehicle heterogeneity, training large models on resource-constrained vehicles, and remaining susceptible to model weight privacy leakage. Meanwhile, split learning (SL) is proposed as a promising collaborative learning framework which can mitigate the risk of model wights leakage, and release the training workload on vehicles. SL sequentially trains a model between a vehicle and an edge-cloud (EC) by dividing the entire model into a vehicle-side model and an EC-side model at a given cut layer. In this work, we combine the advantages of SL and FL to develop an adaptive split FL scheme for VEC (ASFV). The ASFV scheme adaptively splits the model and parallelizes the training process, taking into account mobile vehicle selection and resource allocation. Our extensive simulations, conducted on nonindependent and identically distributed data, demonstrate that the proposed ASFV solution significantly reduces training latency compared to existing benchmarks, while adapting to network dynamics and vehicles’ mobility.

Mobile Edge Computing (MEC) is considered as a promising solution for augmenting the computational capabilities of Internet of Things (IoT) devices by offloading tasks to nearby edge servers (ESs). However, finite communication and computational resources at both IoT devices and ESs, coupled with escalating congestion as more devices connect, present critical challenges in maintaining low latency and high efficiency. In this work, we jointly design the task sensing, communication, and resource allocation for MEC with Non-Orthogonal Multiple Access (NOMA). To address the urgent need for timely data processing in IoT applications, we utilize the Age of Information (AoI) metric as a measure of data freshness. With the objective to minimize the system cost, we propose to jointly optimize sensing sampling intervals, sensing frequencies, offloading decision, and power allocation. Recognizing that this problem is NP-hard, we decompose it incrementally and propose a High-Dimensional Progressive Cost Optimization (HDPCO) algorithm to reduce overall system cost. Simulation results confirm the effectiveness of HDPCO, showing significant improvements in minimizing overall system cost compared to other proposed schemes.

Virtual Reality (VR) applications delivered over wireless networks have attracted interest from academia and industry. The delay of VR applications is mainly composed of computing delay and communication delay. Although cloud computing centers have adequate computing power, accessing them requires long communication delay. Mobile edge computing (MEC), which offloads the computing power from the cloud computing center to the edge, is regarded as a feasible way to alleviate communication delay. However, due to the differences in the capability and location of MEC nodes, the selection of MEC nodes will affect both the computing delay and communication delay. In this paper, we focus on the joint representation of computing and communication resources and the selection of the optimal MEC node. First, we adopt graph-based joint computing and communication resources (GCC) model for VR applications routing and formulate the VR routing problem as an ILP problem. Then we design a Computing Nodes Expanded (CNE) algorithm, which allows us to use the Dijkstra algorithm to quickly obtain the optimal computing node and the path of shortest total delay. Finally, we run numerical experiments to evaluate the performance of the proposal algorithm. Simulation shows that the CNE algorithm can reduce the total delay by 42.9% and increase the delay satisfaction ratio by 23.3% compared to other benchmark algorithms.

In this paper, we consider an Internet of Things (IoT) system where the employed unmanned aerial vehicle (UAV) carries edge computing server to perform data collection and execution for multiple IoT nodes (INs). For such a network, UAV trajectory and uplink transmission power optimization are integrated to minimize the Age of Information (AoI) of the results returned to all ground INs subject to energy consumption limitations. Due to the non-convex nature of the formulated problem, it is divided into two subproblems, which are respectively solved by Lagrangian dual and convex optimization methods, and block coordinate descent method is applied to solve the overall problem. Simulation results show that the proposed algorithm achieves the lowest average AoI of all INs compared with other schemes. The results also reveal the relationship between the average AoI and the number of INs and advantages of the proposed scheme.

Predictive autoscaling (autoscaling with workload forecasting) is an important mechanism that supports autonomous adjustment of computing resources in accordance with fluctuating workload demands in the Cloud. In recent works, Reinforcement Learning (RL) has been introduced as a promising approach to learn the resource management policies to guide the scaling actions under the dynamic and uncertain cloud environment. However, RL methods face the following challenges in steering predictive autoscaling, such as lack of accuracy in decision-making, inefficient sampling and significant variability in workload patterns that may cause policies to fail at test time. To this end, we propose an end-to-end predictive meta model-based RL algorithm, aiming to optimally allocate resource to maintain a stable CPU utilization level, which incorporates a specially-designed deep periodic workload prediction model as the input and embeds the Neural Process [11, 16] to guide the learning of the optimal scaling actions over numerous application services in the Cloud. Our algorithm not only ensures the predictability and accuracy of the scaling strategy, but also enables the scaling decisions to adapt to the changing workloads with high sample efficiency. Our method has achieved significant performance improvement compared to the existing algorithms and has been deployed online at Alipay, supporting the autoscaling of applications for the world-leading payment platform.

In addition to comforting passengers' journey, the modern railway system is responsible to support a variety of on-board Internet services to meet the passenger's demands on seamless service provisioning. In order to provide wireless access to the train, one idea attracting increasing attention is to deploy a series of track-side access points (TAPs) with high-speed data rates along the rail lines dedicated to the broadband mobile service provisioning on board. Due to the heavy data traffic flushing into the base stations (BSs) of the cellular networks, TAPs act as a complement to the BSs in data delivery. In this paper, we focus on the TAP association problem for service provisioning in a heterogeneous wireless railway network, where the TAP and BS coexist by applying a queueing game theoretic approach. Specifically, we present comprehensive theoretical analysis of the delay performance on the circumstances of partially observed, totally unobserved, and totally observed state of the system. Moreover, based on the considered payoff model and the derived association delay time, the passenger's equilibrium strategies on association behaviors, i.e., whether to associate with a TAP or not, are studied. Finally, performance evaluations and discussions are provided to illustrate our proposed passenger-TAP association scheme for the heterogeneous wireless railway communication system.

Vehicular ad hoc network (VANET) is able to facilitate data exchange among vehicles and provides diverse data services. Intuitively, end-to-end backlog and delay bounds are considered significant metrics to evaluate the quality of service in VANETs. In order to analyze how the multi-hop transmission impacts the delay performance, we model the multi-hop service process into a virtualized single service in a min-plus convolution form. To obtain multi-hop end-to-end backlog and delay bound, we consider the stochastic network traffic characteristics and the highly dynamic channel environment under the static priority, first in first out, and earliest deadline first scheduling policies by applying the martingale theory. The IEEE 802.11p enhanced distributed channel access mechanism is also adopted to analyze the access performance in the MAC sub-layer. With three kinds of real wireless data traces, i.e., VoIP, gaming, and UDP, we verify our algorithm by considering the double Nakagami-m fading channel model among vehicles. From the simulation results, we can see that the supermartingale end-to-end backlog and delay bound are remarkably tight to the real simulation results when compared with the existing standard bounds. The effect of the number of vehicles on the highway on the end-to-end backlog and delay performance is also investigated.

Recently, high-speed train is rapidly developed as a popular public transportation to carry passengers and goods with low cost and energy consumption. How to provide passengers's broadband mobile communication services efficiently, such as voice over IP (VoIP) or other multimedia services, is receiving more and more attention nowadays. To fulfill the passenger's demand, we consider a heterogeneous network (HetNet) structure consisting track-side access points (TAPs) and cellular networks for the high-speed rail communication system (HRCS). End-to-end delay is one of the most important quality of service (QoS) indicators to evaluate the HetNet performance. Therefore, this paper investigates the joint end-to-end delay of VoIP and multimedia services in this HetNet architecture. Intermittent connectivity of TAPs and scheduling of multiple on-demand services are considered. In order to obtain the theoretic value of queueing delay bounds, we utilize the martingale theory by analyzing the Markov arrival processes. By combing the arrival-martingale and service- martingale concepts, the theoretic delay bounds under the first in first out (FIFO) scheduling scenario are obtained. For the simulation, we use three kinds of real wireless data traces, VoIP, gaming and UDP to evaluate our algorithm by using Nakagami fading channel and LTE fading channel. From the results we can see that the martingale end-to- end delay bounds are tight to the real data trace simulation results.

This study analyses the performance of a roadside two‐way relaying system in which a roadside access point (AP) and a vehicle exchange messages with the aid of amplify‐and‐forward mobile relay (MR) based on partial relay selection. It has been shown that AP‐MR‐vehicle communications may experience severer channel fading than conventional cellular communications. Mixed Nakagami ‐m and ‘double’ Nakagami ‐m fading is adopted to provide a realistic description of the involved AP‐MR‐vehicle channels. In this scenario, a tight closed‐form lower bound and high signal‐to‐noise ratio approximate expression for the system outage probability are derived. By applying these results, the authors obtain the diversity and the coding gains and the average symbol error rate for the considered system. In particular, the optimum number of relays is provided, providing valuable guidelines for practical system design. It is shown that when the number of relays is greater than the optimum number, no performance gain would be further achieved. The simulation results highlight the authors theoretical analysis.

Energy efficiency (EE) is among the main considerations in the design of modern wireless networks. In this paper, we investigate the EE enhancement for asymmetric analog network coding (ANC) protocol of a two-way relay system based on statistical channel information. A power allocation problem is formulated as the system EE maximization problem with objective function quantified by Goodbit-per-Energy (GPE). Importantly, the EE optimization problem may not be convex and can be categorized into a nonlinear fractional programming problem. Therefore, to solve the problem, a nonlinear fractional programming based algorithm is proposed and closed-form solution is obtained, providing valuable insights into practical system designs. Simulation results highlight the effect of the proposed power allocation.