Next-generation (NextG) wireless networks are expected to transform connectivity by supporting diverse services such as enhanced mobile broadband (eMBB) for high data rates and ultra-reliable low-latency communication (URLLC) for mission-critical applications. A key enabler of these advancements is multiple-input multiple-output (MIMO) technology, which significantly boosts spectral efficiency and network robustness. However, efficiently managing URLLC traffic while preserving the performance of eMBB services remains a major challenge. Conventional solutions, such as resource superposition or puncturing of eMBB traffic, ensure URLLC’s strict latency and reliability requirements but often degrade the overall throughput of eMBB users. In this context, unmanned aerial vehicles (UAVs), or drones, have emerged as promising elements in wireless networks due to their flexibility, rapid deployment, and ability to provide dynamic coverage, particularly in dense or hard-to-reach areas. In this work, we propose a novel puncturing scheme for URLLC traffic that leverages channel quality estimates of eMBB users located at the cell edge or coverage boundary. These users, often requiring fewer resources due to lower throughput demands, are identified through channel quality assessment. To overcome this challenge, we propose a novel quantum deep reinforcement learning (QDRL) approach that harnesses the benefits of quantum computing to improve learning efficiency and solution quality. The QDRL framework employs quantum neural networks (QNN) to efficiently process complex channel state information and perform optimal puncturing decisions within high-dimensional state spaces. Simulation results substantiate the efficacy and resilience of the proposed framework in concurrently maximizing throughput for eMBB services and minimizing latency for URLLC traffic. In particular, the proposed approach improves the eMBB throughput by 7%-19% over Deep Deterministic Policy Gradient (DDPG) and by 15%-37% over Multi-Agent Deep Reinforcement Learning (MADRL) for URLLC arrival rates between 0 and 35.
Wireless-powered mobile-edge computing (MEC) has emerged as a promising paradigm to enhance the data processing capabilities of energy-constrained networks. However, in IoT-based remote monitoring, data-intensive applications still encounter significant challenges in mitigating performance degradation caused by the double near-far effect, a phenomenon induced by dual signal propagation losses occurring simultaneously in both energy harvesting (EH) and communication stages. To overcome this, we first propose a novel dual fixed relay backscatter communication (BackCom)-assisted laser wireless power transfer (LWPT)-enabled MEC network. This architecture uses LWPT for stable energy supply and dual fixed relay base stations to balance energy distribution and relay data, mitigating the double near-far effect. Secondly, to address the coupling between backscatter reflection and energy harvesting (EH), we formulate a non-convex computational energy efficiency (CEE) maximization problem by jointly optimizing multi-domain parameters, including the backscatter coefficient, local and remote computation durations, offloading power, and processor frequencies. We further propose a hierarchical decoupled and coordinated multi-layer resource optimization (HDCO-MLRO) framework. By fully exploiting the structural characteristics of the original problem, the proposed framework enables joint optimization of multi-dimensional variables through hierarchical decoupling and coordinated update mechanisms. Simulation results demonstrate that the proposed LWPT-assisted dual fixed-relay architecture effectively mitigates the double near–far effect under various network conditions. Moreover, the HDCO-MLRO framework significantly outperforms other algorithms and schemes in terms of system computational energy efficiency and energy consumption.
Existing particle-based Bayesian federated learning algorithms face challenges in substantial maintenance expenses in terms of computation, storage, and communication of particles, as well as the oversight of inherent correlation among particles in the lower layers of network parameters. To address the aforementioned issues, we propose a federated low-rank Bayesian approximation learning algorithm via Stein variational gradient descent (FedLoBA) in this paper. Specifically, we verify the occurrence of low-rank subspace among parameters in the same layer of particles through experiment and then design a low-rank Bayesian approximation module to decrease the maintenance costs and improve the shareability among particles. The lower and upper bounds of the discrepancy between the original and approximate posterior distributions are established theoretically. Furthermore, an aggregation refinement strategy has been developed to eliminate extra cross terms during the particle aggregation stage. The experimental results demonstrate that the proposed FedLoBA algorithm exhibits superior prediction accuracy compared to other benchmark algorithms, as well as significant efficiency advantages in terms of computation, storage, and communication over its full-dimensional competitor.
Space-air-ground integrated networks (SAGINs) extend vehicular edge computing (VEC) beyond terrestrial coverage, but task scheduling becomes challenging when task queues, unmanned aerial vehicle (UAV) topology, satellite visibility, and heterogeneous service loads evolve over time. This paper studies quality-of-service (QoS)-constrained queue-aware task mode scheduling, where each UAV jointly selects a pending task and its execution mode. We formulate the problem as a constrained cooperative Decentralized Partially Observable Markov Decision Process (Dec-POMDP) that maximizes long-term priority-weighted successful-service utility subject to average delay and task-service energy constraints. To solve this problem, we propose a unified Feasibility-Guided Multi-Agent Proximal Policy Optimization (FG-MAPPO) framework that enables feasible decentralized scheduling under dynamic task queues and long-term QoS constraints. Simulation results demonstrate that FG-MAPPO improves deadline satisfaction while reducing average delay and task-service energy under dynamic SAGIN-assisted VEC scenarios.
Emerging 6G sensing-based applications rely on ultra-reliable target-aware actuation, where timely and accurate sensing information triggers critical actions. Achieving this requires tightly integrated sensing and communication (ISAC) under stringent reliability and latency constraints. This paper investigates ISAC in a downlink cell-free massive multiple-input multiple-output (CF-mMIMO) system supporting multi-static sensing and ultra-reliable low-latency communications (URLLC). We propose a joint power and blocklength allocation algorithm to minimize total energy consumption, accounting for both radio-site and cloud-side processing energy for communication and sensing, while meeting communication and sensing requirements. The non-convex optimization problem is solved using a combination of feasible point pursuit-successive convex approximation (FPP-SCA), concave-convex programming (CCP), and fractional programming techniques. We consider two types of target detectors: clutter-aware and clutter-unaware, each with distinct complexity and performance trade-offs. A computational complexity analysis based on giga-operations per second (GOPS) is conducted to quantify the processing requirements of communication and sensing tasks. We also introduce the refreshing rate for sensing information and derive a closed-form expression that accounts for both observation and processing delays. Simulation results show that the proposed algorithm achieves up to 34% energy reduction compared to schemes using the maximum allowable blocklength and enhances detection capability while consuming less total energy. Clutter-aware detectors, despite higher complexity, require fewer antennas and sensing receive APs, yielding up to 40% energy savings.
