Edge intelligence has emerged as a promising strategy to deliver low-latency and ubiquitous services for mobile devices. Recent advances in fine-tuning mechanisms of foundation models have enabled edge intelligence by integrating low-rank adaptation (LoRA) with federated learning. However, in wireless networks, the device heterogeneity and resource constraints on edge devices pose great threats to the performance of federated fine-tuning. To tackle these issues, we propose to optimize federated fine-tuning in heterogenous wireless networks via online learning. First, the framework of switching-based federated fine-tuning in wireless networks is provided. The edge devices switches to LoRA modules dynamically for federated fine-tuning with base station to jointly mitigate the impact of device heterogeneity and transmission unreliability. Second, a tractable upper bound on the inference risk gap is derived based on theoretical analysis. To improve the generalization capability, we formulate a non-convex mixed-integer programming problem with long-term constraints, and decouple it into model switching, transmit power control, and bandwidth allocation subproblems. An online optimization algorithm is developed to solve the problems with polynomial computational complexity. Finally, the simulation results on the SST-2 and QNLI data sets demonstrate the performance gains in test accuracy and energy efficiency.

The development of 6G networks is driving the need for innovative resource orchestration solutions to meet the diverse and dynamic demands of next-generation applications. As the number of connected devices increases, traditional network management approaches are insufficient to handle the complex and multi-resource requirements of 6G, which include communication, computation, and storage capabilities across multi-domain environments. To address these challenges, we introduce HyOrch, a novel approach for 6G-driven resource orchestration that employs hypergraph theory to model the intricate interactions and dependencies among network elements across multiple domains. HyOrch enables a comprehensive representation of these complexities, facilitating efficient resource orchestration across End Devices (EDs), Edge Servers (ESs), and Cloud Servers (CSs). It employs a hierarchical distributed resource allocation mechanism that dynamically allocates resources based on real-time availability and application-specific requirements, ensuring optimal performance across the entire network. To validate the effectiveness of HyOrch, we conducted evaluations on a real-world testbed with both virtual and physical devices. The results show that HyOrch significantly outperforms existing approaches, improving resource efficiency by 31.54%-42.13% and reducing delay by 9.77%-39.12%, demonstrating its capability to address the evolving challenges of 6G network orchestration.

Artificial intelligence (AI) integration in communication is crucial for 6G. It optimizes terrestrial communication and computing resource usage in the Internet of Things (IoT) using AI techniques, such as supervised learning for data analysis and reinforcement learning for resource allocation. However, in remote areas, i.e., oceans and deserts, IoT devices lose connection due to limited terrestrial coverage. Low Earth Orbit Satellite (LEOS) offers low-latency, high-bandwidth access in these unconnected regions. However, power and computing limitations on both IoT devices and LEOSs present challenges for continuous service. To this end, we present an LEOS-assisted open radio access network (RAN) Architecture (LO-RAN) where an RAN intelligence controller (RIC) is integrated to provide AI abilities. We formulate a joint Offloading decision, Path selection, and Resource allocation problem (OPR) to minimize the weighted energy consumption and latency of LO-RAN. We proposed a Joint Optimization for the Offloading decision, Path selection, and Resource allocation (JOOPR) algorithm. It selects contact and processing LEOSs for path selection, uses proximal policy optimization (PPO) for offloading decisions, and applies Karush-Kuhn–Tucker (KKT) to solve resource allocation. The outputs from path selection and resource allocation contribute to the reward that feeds into the PPO. We conduct numerical simulations to compare the proposed JOOPR with the state-of-the-art approaches. The results show that JOOPR reduces energy consumption and latency by at most 28.75% and 33.01%, respectively.

The integration of autonomous aerial vehicles (UAVs) and low-Earth orbit (LEO) satellites has become attractive for Internet of Things (IoT) task processing, as it can overcome obstacles in terrestrial network coverage, such as those in oceans or desert areas. However, it lacks a collaborative approach for allocating the communication and computing resources among UAVs and LEO satellites and optimizing the hovering point of UAVs to prolong their endurance. In this article, we investigate energy-efficient resource allocation in LEO-assisted UAV networks for the IoT. A novel optimization algorithm, that jointly IoT tasks’ offloading decision, UAVs’ region selection, hovering point chosen, and communication and computing resource allocation (ORHCC), is proposed to optimize UAV trajectories and hovering points, enhancing endurance and minimizing energy consumption. In particular, the UAVs’ region selection and IoT tasks offloading are under the dueling deep Q-network (DuDQN) framework, the hovering point chosen and communication and computing resource allocation via the convex solution. The results show that the proposed ORHCC reduces 12.5% and 20.76% energy consumption compared with the proximal policy optimization and greedy baseline, respectively.

Low Earth Orbit(LEO) is the supplement of the terrestrial network, it has the advantages of low latency and high bandwidth compared to Geostationary Earth Orbit (GEO). LEO can cover and provide the service to the unconnected areas in terrestrial. Limited to the computing capacity of LEO satellites, LEO is not sufficient to process the terrestrial tasks, hence, a LEO-assisted central cloud to provide services is the promising method. The latency-sensitive task has stringent requirements of latency, and LEO has a low attitude and serves the low latency tasks. In this paper, we propose an LEO-assisted terrestrial network architecture and the tasks offloading and resource allocation are also investigated. We present a deep reinforcement learning approach as the solution for task offloading and resource allocation aiming at minimizing energy cost. The results show that our proposed Soft Actor-Critic has a superior performance than other benchmarks and about 51% energy reduction compared to Proximal Policy optimization (PPO).

The evolution of communication architecture shifts towards virtualization and cloud-native network functions, setting the stage for the flexibility and integration of emerging technologies. Artificial Intelligence (AI) and Machine Learning (ML) as intrinsic elements in network design are some of the crucial visions and requirements for 6G. This paper, from the perspective of O-RAN, explores how current network architectures should evolve towards the integration of communication and intelligence in 6G. It begins with a comprehensive analysis and comparison of the AI-related work conducted by various standard organizations. Building on this, an end-to-end AI integration framework is proposed, which leverages AI technologies, data services, and digital twin (DT) technologies to achieve an integrated intelligent 6G communication system. After that, the key enabling technologies for cross-domain AI, service-based RAN, programmable RAN and digital twins are discussed. At last, the paper analyzes the challenges and opportunities for O-RAN evolution.

The demand for higher transmission efficiency and denser user access has been put forth by the next generation of wireless communication systems. To cater to the future communication development, this article focuses on massive random access schemes under the user-centric cell-free massive multiple-input-multiple-output (MIMO) architecture. For uplink transmission, a data frame structure is designed to enable active user detection (AUD), channel estimation (CE), and data transmission. The association between access points (APs) and user equipment (UEs) is presented to facilitate an user-centric cell-free scalable architecture. In this article, a maximum likelihood (ML)-based method is proposed for AUD to obtain the set of active UEs. By setting appropriate thresholds and combining the UE-AP association, accurate active detection results can be obtained. CE can be accomplished with lower computational complexity by utilizing the detected active UE set in AUD module. Specifically, the generalized approximate message passing-based sparse Bayesian learning with Dirichlet process (GAMP-DP-SBL) is adopted as the CE algorithm, leveraging the spatial aggregation and dispersion characteristics of APs to enhance the estimation accuracy. Building upon GAMP-DP-SBL algorithm, a clustered algorithm (GAMP-CDP-SBL) is proposed to reduce the scale of the sensing matrix and improve the accuracy of CE for associated active UEs. Moreover, to enhance system scalability, decentralized AUD and CE algorithms are proposed in this article. Simulation results under various parameter settings and different scenarios exhibit the superior performance of the proposed scheme.

The Internet of Things (IoT) system provides sensing and computing services via terrestrial networks. However, the restricted coverage of terrestrial networks, such as base stations, limits the ubiquitous IoT services. Low-Earth orbit (LEO) satellites are able to provide network coverage for terrestrial IoT devices in unconnected scenarios, e.g., maritime. IoT devices in such scenarios usually have restricted onboard computation and power resources. In this article, we present an LEO-assisted IoT network (L-IoT) architecture where a device splits its task and offloads a portion of its task to the LEO to process within the coverage time. We formulate a task split problem with communication and computation resource allocation (SCC) to minimize the L-IoT energy consumption. We proposed an alternating optimization for split ratio and resource allocation (AOSR) algorithm. In particular, we use the outputs of Karush-Kuhn–Tucker (KKT) for resource allocation as part of the reward that feeds twin-delayed deep deterministic policy gradient. Lastly, the results of numerical simulations show that the proposed AOSR approach reduces 12.7% energy consumption compared to soft actor-critic (SAC) and 15% to deep deterministic policy gradient (DDPG).

Recently, body part as an intuitive movement unit has received increasing attention in skeleton-based action. However, the part-level embedding is hard to be fully exploited, especially for fine-grained actions, as the body joints are aggregated into parts. To address this problem, we propose a novel transformer-based network (IIP-Transformer). Different from previous models that rely on specially designed partition method, our proposed IIPA mechanism which incorporates joint-level (intra-part) and part-level (inter-part) interactions simultaneously is the keypoint for IIP-Transformer to fully exploit part-level data, making considerable improvements in both coarse-grained and fine-grained action recognition. Ablation studies on three typical partition methods show that IIP-Transformer is a relatively general solution for part-level data and thus we could choose the simplest hand-craft partition embedding to significantly reduce computational complexity and model size. Besides, The proposed IIP-Transformer exceeds the state-of-the-art methods with much less computational complexity on NTU RGB+D, NTU RGB+D120 and NW-UCLA datasets.

The diverse performance requirements and distributed deployment needs brought about by new types of services have blurred the boundary between the radio access network (RAN) and core network (CN), necessitating research on 6G network architecture. At the deep edge node (DEN) side, this paper investigates a RAN-CN converged architecture to address the demand for RAN and CN collaboration, as well as the problem of functional redundancy between RAN and sunken CN. Additionally, this paper analyzes and studies the procedures of UE initial access and registration, session management, UE context management, based on the converged architecture. The suggested architecture can eliminate functional redundancy between RAN and CN, simplify the existing signaling and data transmission procedures, and shorten latency and processing complexity.

Edge Computing (EC) has been proposed as a promising approach to fulfil the requirements of high bandwidth and ultra‐latency of mobile applications. However, existing researches on resource allocation only consider the computing resource of mobile devices and EC servers, while ignored the constraint of the networking resources once spreading applications among multiple EC servers via wireless and wired networks. Fibre‐Wireless access networks (FiWi) combine the huge bandwidth of optical fibre networks and flexible access of wireless networks to address the above issue and bridge the coexistence of multiple EC servers. They propose a Virtualisation‐based Architecture converging EC over FiWi (VAECFW) to centralise control and allocate networking and computing resources for serving requested services. In addition, they study the problem of resource allocation of EC over FiWi and propose two algorithms Revenue‐based Virtual Network Embedding (R‐VNE) and Balanced Central Processing Unit Resource Allocation with Virtual Network Embedding (BCRA‐VNE). Simulation experiments show that the services acceptance ratio is increased about 35% under their proposed architecture, and the average service requests bandwidth utilisation of R‐VNE is increasing from 50 to 66%, and the BCRA‐VNE is from 37 to 48%. The two algorithms not only achieve higher revenue but also get better profit rate.

Multipath algorithms except Equal-Cost Multi-Path(ECMP) which has been widely used in networks are difficult to apply, because multipath provisioning is more complex at cross layers and multipath routing need to get all nodes' information. To address the dilemma, this paper proposes a multipath network virtualization implementation scheme with Software Defined Networking (SDN) and Network Function Virtualization (NFV). In this scheme, SDN schedules network resources in a global view for selecting multiple paths and computing weight of each path, and NFV provides computing and storage resources to split flow, add tag, recover flow, to name a few. This paper also proposes a multipath algorithm for elephant flow with network virtualization. Besides, we build an experimental platform based on OPNFV and SDN, and conduct experiments under this experimental platform. The results show that our proposed algorithm applied on multipath network virtualization experimental platform has superior performance than ECMP applied in networks without virtualization.