In this paper, we propose a digital agent (DA)-assisted resource management scheme for enhanced user quality of experience (QoE) in integrated sensing and communication (ISAC) networks. Particularly, user QoE is a comprehensive metric that integrates quality of service (QoS), user behavioral dynamics, and environmental complexity. The novel DA module includes a user status prediction model, a QoS factor selection model, and a QoE fitting model, which analyzes historical user status data to construct and update user-specific QoE models. Users are clustered into different groups based on their QoE models. A Cramér-Rao bound (CRB) model is utilized to quantify the impact of allocated communication resources on sensing accuracy. A joint optimization problem of communication and computing resource management is formulated to maximize long-term user QoE while satisfying CRB and resource constraints. A two-layer data-model-driven algorithm is developed to solve the formulated problem, where the top layer utilizes an advanced deep reinforcement learning algorithm to make group-level decisions, and the bottom layer uses convex optimization techniques to make user-level decisions. Simulation results based on a real-world dataset demonstrate that the proposed DA-assisted resource management scheme outperforms benchmark schemes in terms of user QoE.

The rapid growth of connected vehicles creates new opportunities to exploit internet of vehicles (IoV) data for traffic state estimation (TSE), which is a key enabler of intelligent transportation systems (ITS). In this paper, we propose a cost-effective TSE framework that leverages sparse IoV data, which significantly reducing the data collection overhead associated with large-scale IoV datasets. We further analyze the impact of data sparsification and show that the induced estimation errors can be well approximated by Gaussian noise, thereby reformulating sparse IoV-based TSE as a denoising problem. To enhance estimation accuracy, we develop a spatial-temporal attention model, termed the convolutional retentive network (CRNet), which integrates convolutional neural networks (CNNs) for spatial correlation learning with a retentive network (RetNet) for temporal dependency modeling. Extensive experiments conducted on a large-scale real-world IoV dataset validate the feasibility of TSE under sparse IoV sensing conditions. Notably, even when only 5% of the data is available, CRNet achieves a mean absolute error (MAE) below 5 km/h, demonstrating both the high accuracy of the proposed approach and its practical applicability in real-world scenarios.

Achieving the stringent requirements of ultra reliable low latency communication (URLLC) in vehicular networks, characterized by high mobility, presents significant challenges. Orthogonal time frequency space (OTFS) modulation has surfaced as a promising solution to tackle Doppler shifts and delay spreads in high-mobility wireless channels by mapping symbols into the delay-Doppler (DD) domain. This paper introduces a novel OTFS-enabled transceiver framework for predictive beamforming in vehicular networks. The framework employs frequency division duplex (FDD) to diminish latency and enhance flexibility, with the predictive beamforming technique implemented at the transmitter to improve the received signal strength (RSS). Owing to the latency constraints of URLLC, predictive beamforming necessitates leveraging historical channel state information (CSI) in the DD domain. In this context, a deep learning (DL) approach using the ConvLSTM network is harnessed for predictive beamforming, concentrating on capturing spatial-temporal correlations of high-mobility wireless channels from historical DD domain CSI. Extensive simulations validate the efficacy of our proposed DL-based predictive beamforming for OTFS-enabled URLLC in vehicular networks.

In this paper, we propose a novel service continuityaware Service Function Chain (SFC) embedding scheme for dynamic large-scale LEO satellite networks, where service disruptions occur when satellites hosting virtual network functions of an SFC move out of the service region. Particularly, we define a new metric, i.e., the Remaining Time to Migration (RTTM), which indicates the remaining functional time of an SFC before SFC reconfiguration is needed. We then formulate a service continuity-aware SFC embedding problem with the objective of maximizing the long-term acceptance ratio while increasing the normalized RTTM of accepted SFCs. We propose a scalable graph neural network-assisted deep reinforcement learning (DRL) approach to solve the embedding problem. By employing a differentiable pooling technique, we condense the feature representation of large-scale LEO satellite networks, thereby reducing the computational complexity of the down-stream DRL-based decision-making. Simulation results show that our approach reduces the proportion of reconfigured SFCs by 60 % compared to the benchmark, indicating its effectiveness in enhancing service continuity.

In satellite-terrestrial links, the high-speed movement of low earth orbit (LEO) satellites induces significant doppler effects. By employing orthogonal time frequency space (OTFS) modulation technology, information symbols are effectively mapped into the delay-doppler domain, significantly mitigating the adverse impacts of doppler effects on transmissions. Furthermore, the short packet is introduced into satelliteterrestrial communications to meet diverse latency requirements. Due to the limited channel coding capacity of short packets, adjusting the channel coding rate and symbol modulation order according to the channel conditions is crucial in satellite communications. However, real-time channel state information feedback is precluded by the rapid variations in satellite-terrestrial channels, the large feedback latency, and the limited resources for frequent feedback. Therefore, implementing a feedback-free transmission approach is particularly essential. In this paper, we design a feedback-free OTFS-based adaptive modulation and coding (AMC) scheme to accommodate varying communication requirements and conditions. Extensive simulations with different orbital altitudes, carrier frequencies, and packet lengths have shown that the proposed approach can be effectively adapted to LEO satellite channels, enabling a feedback-free AMC in the LEO satellite-terrestrial communications.

Deep learning methods for accurately predicting data center status, which are essential for addressing the exponential growth of energy consumption, have gained significant attention, driven by the vast amounts of data collected through the advancement of Internet of Things (IoT) technologies. However, conventional small models often face data scarcity issues in practical deployment. While large models show promise in addressing this challenge, they encounter obstacles, such as multivariate tasks, computational intensity, and ineffective information capture. Moreover, their applications in data centers remain largely unexplored. In this article, we investigate local semantic capture empowered large model for multivariate time series forecasting in IoT-enabled data centers. We first introduce time series tasks within data centers and propose the Point Lag (Plag)-Llama framework with the Lag-Llama backbone to support zero-shot forecasting and fine-tuning for multivariate point time series forecasting. To address computational intensity and enhance the capabilities of multivariate forecasting, we propose the local semantic capture (LOSEC) for adapter fine-tuning, which captures local semantic information across time and channel dimensions alternately with low-complexity. Specifically, time series are patched into tokens, and channels are clustered together, forming local semantic information that can be captured more effectively. Extensive experiments demonstrate that Plag-Llama exhibits superior zero-shot capability and that the LOSEC empowered adapter fine-tuning achieves state-of-the-art performance on real-world datasets collected from data centers, with ablation studies further validating the effectiveness of each module within the proposed models.

Recently, mobile crowdsensing (MCS) has emerged as a promising solution for traffic state estimation (TSE), which provides real-time and accurate traffic information for supporting diversified intelligent transportation systems (ITS) applications. However, the prohibitive overhead of collecting massive data in vehicular networks limits the available data amount, while the sparsification of MCS data incurs instability and degrades TSE accuracy. To this end, this paper proposes a novel sparse MCS framework to facilitate cost-effective TSE, which utilizes a small number of vehicular MCS participants distributed across all regions as data sources. By utilizing spatial and temporal correlations of traffic flow, an innovative spatiotemporal deep learning model, namely Transformer Graph Attentional Sample and Aggregate neural network (TGASA), is proposed to improve the TSE accuracy with sparse MCS data. Specifically, we design an incorporated graph neural network (GNN) to aggregate the spatial correlation by taking both node features and edge properties into account. And, the transformer neural network architecture is applied to capture the temporal correlation. Extensive simulation results based on real-world datasets demonstrate that the proposed framework can significantly address the instability incurred by the sparsification of MCS data and effectively achieve a more accurate TSE.

Ultra reliable low latency communication (URLLC) in vehicular networks is crucial for safety-related vehicular applications. Mini-slot with a short packet that carries only a few symbols is used to reduce the transmission time interval and enable quick scheduling for URLLC that requires extremely low latency. However, a single air interface transmission of URLLC packets may fail due to the high mobility of vehicles. Leveraging multi-connectivity technologies, the real-time reliability of URLLC can be greatly enhanced without relying on packet retransmission. In this paper, we propose a multi-connectivity URLLC downlink transmission scheme for vehicular networks, where the URLLC packet is duplicated and transmitted over multiple independent wireless links to improve packet reliability. Specifically, we design a multi-agent cooperative deep reinforcement learning algorithm, called transformer associated proximal policy optimization (TAPPO), to achieve real-time robust power allocation for multi-connectivity URLLC with imperfect channel state information (CSI). The transformer neural network architecture is employed to share the information among multiple links serving the same URLLC user and choose appropriate transmit powers, enabling cooperation to ensure reliability while minimizing inter-cell interference and energy consumption. Extensive simulation results validate the effectiveness of multi-connectivity packet duplication for URLLC and proposed TAPPO for power allocation.

The increasingly severe energy consumption caused by exploding wireless demands attracts considerable research. Remarkably, base station (BS) sleeping is a promising technique to enable the green network. A disruptive and original fully-decoupled radio access network (FD-RAN) architecture aiming at the next-generation mobile communication networks is developed, which removes the obstacles to achieving BS sleeping, i.e. deficient cooperation between BSs, coupled data-control transmission and coupled uplink-downlink transmission. In this paper, we investigate the joint user association and uplink BS sleeping considering power control in the FD-RAN with the superiority of fully decoupled architectures. Specifically, we propose an energy consumption model for the uplink FD-RAN and tackle the mixed-integer second-order cone problem to minimize the whole network energy consumption by leveraging the many-to-many swap matching theory. Extensive simulation results validate a higher energy efficiency of the uplink FD-RAN compared to the traditional cellular network and cell-free networks and demonstrate the effectiveness of our proposed algorithm.

Cellular vehicle-to-everything (C-V2X) network is a promising solution to support on road diverse quality of services (QoS) such as ultra reliable low latency communication (URLLC) and enhanced mobile broadband (eMBB). However, satisfying the stringent QoS requirements in high-dynamic C-V2X environment is very challenge. In this paper, we leverage the multi-connectivity technology to enhance the reliability of downlink URLLC in C-V2X. Specifically, with the aid of the cloud radio access network (C-RAN), the network controller duplicates each URLLC packet and transmits its replicas over multiple independent wireless links. To ensure the reliability of URLLC links while maximizing the average rate of eMBB links, we design a coordinated multi-agent deep reinforcement learning algorithm for real-time power allocation of multi-connectivity URLLC links. Each URLLC link is treated as an agent here, and its transmit power is its action. The multiple links serving the same URLLC user are coordinated with a three-layer neural network for information sharing, allowing them to cooperatively choose transmit powers in terms of ensuring reliability while minimizing inter-cell interference and energy consumption. Extensive simulation results validate the effectiveness of the proposed power allocation algorithm for multi-connectivity downlink URLLC.

We consider a fully-decoupled radio access network (FD-RAN), where base stations (BSs) are physically decoupled into control BSs, uplink BSs and downlink BSs, and multi-connectivity becomes the default user equipment (UE) association mode. Specifically, we study the inter-frequency multi-connectivity in downlink of FD-RAN and present a deep reinforcement learning based online multi-connectivity mobility management scheme. We formulate a UE dynamic multiple access problem and transform it into a handover decision problem, then apply the double deep Q-network (DDQN) algorithm to make real time mobility management decisions. Simulation results show that the proposed scheme outperforms benchmarks in terms of handover frequency and quality of service, while ensuring real-time performance.

Traffic flow estimation (TFE) plays an important role in intelligent transportation systems (ITS). Considering the prohibitive overhead of collecting and processing massive data in vehicular networks, it is more practical to use the sparse vehicular big data. In this paper, we focus on accurately estimating the urban traffic flow of vehicular networks only using a small portion of vehicular data. A new spatiotemporal machine learning model, named Graph Sampling and Aggregate Transformer (GSAT), is developed to improve the estimation accuracy by leveraging the inner correlation of traffic data. Specifically, the GSAT uses the graph sample and aggregate (GraphSAGE) model, a variety of graph neural network (GNN), to aggregate the spatial correlation and applies the Transformer model to capture the temporal correlation. We evaluate GSAT at multiple sparsity on the real world dataset of vehicular network, and it is demonstrated that GSAT can achieve accurate estimation with sparse big data.