Given the unpredictable and bursty nature of mission-critical ultra-reliable low-latency communication (URLLC) traffic, collaborative resource allocation across multiple base stations (BSs) enables joint resource pooling and interference management to mitigate instantaneous traffic spikes. Nevertheless, coordinating multi-cell spectrum sharing to achieve sustainable URLLC systems in the long term remains a persistent challenge, largely owing to the prohibitive computational cost of centralized designs and the potentially inefficient interference suppression in overlapping coverage regions. In this paper, we propose a radiation-aware distributed transmission scheme that collaboratively guarantees sustainable URLLC service while mitigating mutual interference. Specifically, a radiation footprint control mechanism is developed to adaptively regulate the radiated power intensity of transmitted wireless signals over the propagation space, preserving the service provisioning efficiency of the distributed network. Based on this mechanism, we formulate a green dynamic resource allocation problem to minimize long-term power consumption while ensuring sustained network stability. To address the nonlinear stochastic optimization challenge, Lyapunov optimization is first employed to transform the long-term problem into sequential short-term online ones. For the deterministic problem in each time slot, the BS selection is first abstracted as a collaborative matching game, after which a distributed scheme with upper/lower-bound approximation algorithms is proposed to solve the coupled beamforming and subcarrier assignment. Simulation results verify the effectiveness of the proposed algorithm in balancing power consumption against queue backlog.

This paper investigates mobile edge computing (MEC)-aided multi-static integrated sensing and communication (ISAC) systems. In the considered system, each device offloads a task to the MEC server for computation. Meanwhile, a sensing receiver (SR) has to detect a target. To enhance the detection capabilities, the devices and base station (BS) collaboratively transmit radar signals, and the SR detects the target based on the received radar echoes. Specifically, the task execution period is divided into two phases. In the first phase, the devices offload part of their tasks while simultaneously transmitting radar signals for target sensing. In the second phase, the MEC server at the BS processes the offloaded tasks, and meanwhile devices along with the BS transmit radar signals for sensing. Thus, the system entities form a multi-static ISAC framework. Radar signal processing schemes are designed for both scenarios—with and without knowledge of the target response amplitudes. The corresponding cooperative detection probabilities under a constant false alarm probability constraint are derived. To ensure high energy efficiency, optimization problem is formulated to minimize energy consumption across all devices, subject to task computation and detection probability constraints. This is achieved by adjusting beamforming at the devices and BS, as well as task partitioning and phase duration allocation. For scenarios with knowledge of target response amplitudes, the problem is efficiently solved by iteratively optimizing beamforming subproblem, time and task division subproblem using weighted minimum mean square error method, successive convex approximation, and golden section search method. For scenarios without knowledge of target response amplitudes, the formulated bi-level optimization problem is first equivalently transformed into a single-level problem, which is then tackled using alternating optimization method, semidefinite programming, and Lagrangian dual method. Simulations validate the benefits of the proposed MEC-aided cooperative sensing scheme compared with numerous benchmarking schemes including traditional non-cooperative sensing scheme.

Split inference divides a global deep neural network (DNN) into several sub-models and assigns them to different computing nodes, thereby leveraging distributed computational resources at network edge for executing complex artificial intelligence (AI) algorithms. As an emerging technique,over-the-air computation(AirComp) turns a multi-access channel into a processor for distributed computing. Specifically, by exploiting the waveform superposition of the multi-access channel, a receiver directly aggregates signals transmitted by different transmitters to obtain their average (or any nomographic function). In order to accelerate inference speed and reduce communication and computation loads, we propose a novelin-network AIframework that realizes inference through communications among devices. This distinctive feature of the framework is to generalize multiple-input-multiple output (MIMO) AirComp to realize over-the-air matrix-vector multiplications (MVM), the most computation intensive operations of a DNN. As a result, the participating devices act like neurons in the DNN but they are now linked through computation capable wireless channels, which gives the name ofOver-the-Air Neural Link(AirNeuralink). The proposed AirNeuralink techniques are associated with different system topologies including relay, star, and distributed topologies. Their design involves jointly optimizing the precoding and post-equalization to minimize the MVM errors of the estimated feature values at the receiver set with respect to their ground truth under the transmit power constraint for each transmitter. The optimal post-equalization is derived in the form of Wiener filter but with an aggregation matrix (weight matrix). Utilizing the property of Schur-concave function and matrix inequality, the optimal structure of precoding in each topology is designed to enable spatial-channel power allocation by efficiently solving a convex problem with scalar variables. Besides, the asymptotic MVM errors are analyzed to show that the errors can be sufficiently small if the rank of model-weight matrix is no larger than that of channel matrix. Simulations demonstrate the superior performance of the proposed framework in transmission latency reduction while maintaining the inference accuracy as compared with traditional digital transmission.

This paper investigates the delay minimization problem for multi-task federated learning (MTFL) systems at the network edge. We develop a novel MTFL framework, based on which the divergence bounds are derived for both task-related and task-unrelated scenarios, systematically quantifying the effects of user importance, user participation, and inter-task correlations on convergence behavior. Building upon these insights, a long-term joint optimization problem is formulated to minimize the overall training delay under the long-term divergence bounds and energy constraints. To address the coupling in multi-slot user scheduling, the optimization problem is decomposed into a single-slot joint resource allocation and task scheduling subproblem and a cross-slot user scheduling subproblem. The former is solved using block coordinate descent (BCD) combined with Johnson’s rule, while the latter is modeled as a constrained Markov decision process (CMDP) and addressed via a dueling double deep Q-network (D3QN) with cost shaping and prioritized experience replay. Numerical results verify the effectiveness of the proposed framework and convergence analysis, demonstrating its significant improvements over baseline schemes in terms of convergence and delay reduction.

Hybrid beamforming (HBF) offers significant advantages for massive multiple-input multiple-output (mMIMO) multi-beam low earth orbit (LEO) satellite systems, due to its ability to reduce hardware complexity and increase spectral efficiency. However, in practical multi-beam LEO systems, angle errors and imperfect channel state information at the transmitter (CSIT) present major challenges for traditional HBF techniques, which assume perfect channel characteristics. To address these challenges, a robust HBF method designed for multi-beam LEO systems is introduced, integrating wide beam (WB)-based analog beamformer with a digital precoder assisted by rate-splitting multiple access (RSMA). Furthermore, to solve the digital precoding problem, we employ a semi-definite programming (SDP) method augmented with penalty functions to transform the original non-convex problem into a convex one and solve it iteratively. Finally, simulation results demonstrate that the proposed HBF method can significantly enhance ergodic sum rate, outperforming traditional HBF methods in terms of rate performance.

Ultra-reliable low-latency communication (URLLC) is a critical type of service that imposes stringent latency requirements. Considering the random and burst URLLC packets arrival characteristics, incorporating spatial frequency reuse into multi-cell networks can significantly improve the system performance. Nevertheless, how to design the frequency refuse strategy for such a multi-cell URLLC system remains technically challenging. In this article, we investigate an online dynamic resource scheduling problem in a multi-cell downlink system with URLLC services. The long-term time-averaged effective throughput is maximized while guaranteeing the instantaneous transmission reliability and prolonged network stability. The formulated problem is a mixed integer nonlinear stochastic optimization problem, in which the Lyapunov optimization is first leveraged to transform the long-term maximization problem into sequential short-term online ones. To tackle the deterministic problem in each time-slot, we further propose a distributed algorithm that delegates computational processes to corresponding base stations for collaborative execution. In this framework, the user association is abstracted as a cooperative game model. Subsequently, each base station exploits alternating optimization and convex optimization approximation algorithms to address the remaining resource allocation problem. Simulation results validate the effectiveness of the proposed algorithm in throughput-backlog trade-off, showcasing that the multi-cell collaborative transmission can attain better performance compared with existing schemes.

This paper investigates the joint optimal allocation of radio and computational resources aiming to minimize global average task offloading age (TOA) over all time slots and mobile devices (MDs) for long-term multi-cell MEC systems with continuous arrival of MDs. TOA represents the total number of offloading time slots, including both transmission and computation. The joint resource allocation problem cannot be solved online because its objective function is long-term average of TOA over all time slots. We transform the long-term resource allocation problem into an online one by the Lyapunov method, then an iterative algorithm is proposed to solve the online problem. The idea of this algorithm is computing iteratively the two sub-problems which optimize sub-channel allocation and offloading power and computational resources joint allocation based on an initial resource allocation scheme. The alternating direction method of multipliers (ADMM) method is employed to solve the first sub-problem. For the second sub-problem, a closed-form expression of optimal power is deduced by solving a convex optimization problem using the Lagrange multiplier method, then the sub-problem is simplified into a linear programming (LP) problem about computational resource allocation. The improved iterative greedy (IIG) algorithm is applied to solve the LP problem. Simulation results demonstrate that the proposed algorithm approaches the performance of the optimal branch-and-bound (BnB) algorithm in the MEC systems with one-time arrival of MDs, and outperforms two benchmark schemes such as first in first out (FIFO) and Chang’s algorithm.

This paper studies the joint on-ground precoding and on-board beamforming design problem in multi-gateway multibeam satellite communications with feeder link interference. Compared with pure on-ground systems, the hybrid on-ground on-board architecture provides a good trade-off between the total throughput and the bandwidth requirements of the feeder link. Moreover, we propose a two-timescale design strategy which significantly reduces the computational burden on the satellite payload. To be specific, the long-term on-board beamforming network (BFN) is designed based on the statistical channel state information (S-CSI), while the short-term on-ground precoding matrices are designed to cater to the instantaneous CSI (I-CSI) with optimized on-board BFN. We formulate an average sum rate maximization problem which incorporates the lossless constraint of the BFN, as well as the average power constraints at the satellite. To tackle this two-timescale non-convex stochastic optimization problem, we propose an algorithm called two-timescale joint on-ground and on-board beamforming (TJGBB). At each iteration, the TJGBB algorithm first solves the short-term on-ground precoding subproblem based on the weighted minimum mean square error (WMMSE) method. Then, it constructs convex surrogate functions for the objective and constraints of the long-term on-board beamforming primary problem, creating an approximate version. Finally, the long-term variables are updated by solving this convex approximation problem. The proposed algorithm is proven to almost surely converge to a stationary solution of the original problem within an acceptable error margin. Simulations show that our proposed method can significantly improve the overall system throughput compared to existing schemes.

This paper investigates a mobile edge computing (MEC) aided multi-static integrated sensing and communication (ISAC) system. In the considered system, each device has a task to be offloaded to MEC server for computation. Meanwhile, a sensing receiver (SR) has to detect a point target. To enhance detection probability, devices and base station (BS) collaboratively transmit radar signals and then SR detects the target of interest based on received radar echoes. Specifically, the whole task execution period is divided into two phases. In the first phase, devices offload partial of their tasks and simultaneously transmit radar signals for target sensing. In the second phase, the MEC server at the BS computes offloaded tasks and meanwhile devices along with BS transmit radar signals for sensing. Therefore, the entities in the system constitute a multi-static ISAC framework. Radar signal processing for multi-static sensing system is designed and the corresponding cooperative detection probability is derived. To ensure high energy efficiency, an optimization problem is formulated to minimize energy consumption of all devices under task computation and detection probability constraints by adjusting devices’ and BS’s beamforming, task partitioning, and phase duration allocation. This problem is efficiently solved by iteratively optimizing beamforming subproblem, time and task division subproblem through weighted minimum mean square error, successive convex approximation and search methods. Simulations verify the effectiveness of the proposed MEC aided cooperative sensing algorithm compared with non-cooperative sensing scheme.

Mobile edge computing (MEC) in the next generation networks will provide computation services at the network edge to enrich the capabilities of mobile devices and lengthen their battery lives. However, the performance of MEC cannot be guaranteed, when large size local tasks are uploaded to the server simultaneously causing network congestion. As a new paradigm that focuses on transmitting the meaning of messages, semantic communications reveals the significant potential to reduce the network traffic. In this paper, we propose a semantic-aware joint communication and computation resource allocation framework for MEC systems. In the considered system, random tasks arrive at each terminal device (TD), which needs to be computed locally or offloaded to the MEC server. To further release the transmission burden, each TD sends the small-size extracted semantic information of tasks to the server instead of the original large-size raw data. An optimization problem of joint semantic-aware division factor, communication and computation resource management is formulated. The problem aims to minimize the energy consumption of the whole system, while satisfying long-term delay and processing rate constraints. To solve this problem, an online low-complexity algorithm is proposed. In particular, Lyapunov optimization is utilized to decompose the original coupled long-term problem into a series of decoupled deterministic problems without requiring the realizations of future task arrivals and channel gains. Then, the block coordinate descent method and successive convex approximation algorithm are adopted to solve the current time slot deterministic problem by observing the current system states. Moreover, the closed-form optimal solution of each optimization variable is provided. Simulation results show that the proposed algorithm yields up to 41.8% energy reduction compared to its counterpart without semantic-aware allocation.

In this paper, the problem of joint user scheduling and computing resource allocation in asynchronous mobile edge computing (MEC) networks is studied. In such networks, edge devices will offload their computational tasks to an MEC server, using the energy they harvest from this server. To get their tasks processed on time using the harvested energy, edge devices will strategically schedule their task offloading, and compete for the computational resource at the MEC server. Then, the MEC server will execute these tasks asynchronously based on the arrival of the tasks. This joint user scheduling, time and computation resource allocation problem is posed as an optimization framework whose goal is to find the optimal scheduling and allocation strategy that minimizes the energy consumption of these mobile computing tasks. To solve this mixed-integer non-linear programming problem, the general benders decomposition method is adopted which decomposes the original problem into a primal problem and a master problem. Specifically, the primal problem is related to computation resource and time slot allocation, of which the optimal closed-form solution is obtained. The master problem regarding discrete user scheduling variables is constructed by adding optimality cuts or feasibility cuts according to whether the primal problem is feasible, which is a standard mixed-integer linear programming problem and can be efficiently solved. By iteratively solving the primal problem and master problem, the optimal scheduling and resource allocation scheme is obtained. Simulation results demonstrate that the proposed asynchronous computing framework reduces 87.17% energy consumption compared with conventional synchronous computing counterpart.

Edgeartificial intelligence(AI) in the sixth-generation networks will provide inference services at the network edge to enrich the capabilities of mobile devices and lengthen their battery lives. As a well-known technique in computing, batching can boost the computation throughput at an edge server by assembling multiple tasks into a batch that is fed into a pre-trained prediction model. This reduces the memory-access frequency and hence accelerates the execution of each task. In this paper, we study joint batching and (task) scheduling to maximise the throughput (i.e., the number of completed tasks) under the practical assumptions of heterogeneous task arrivals and deadlines. The design aims to optimise the number of batches, their starting time instants, and the task-batch association that determines batch sizes. The joint optimisation problem is complex due to multiple coupled variables as mentioned and numerous constraints including heterogeneous tasks arrivals and deadlines, the causality requirements on multi-task execution, and limited radio resources. Our approach of solving the formulated mixed-integer problem is to transform it into a convex problem via integer relaxation method and ℓ0-norm approximation. This results in an efficient alternating optimization algorithm for finding a close-to-optimal solution. Specifically, it iterates between solving two sub-problems, optimal task-batch association and optimal batch starting time. The former is a linear program whose solution can be found using a derived scheme of greedy task selection while that of the latter is derived in closed form. In addition, we also design the optimal algorithm from leveragingspectrum holes, which are caused by fixed bandwidth allocation to devices and their asynchronized multi-batch task execution, to admit unscheduled tasks so as to further enhance throughput. Simulation results demonstrate that the proposed framework of joint batching and resource allocation can substantially enhance the throughput of multiuser edge-AI as opposed to a number of benchmarking schemes.

In this paper, we investigate the energy minimization problem in cell-free mobile edge computing (MEC) networks under dynamic channel conditions and random arrival tasks. We aim to minimize the total energy consumption of user equipment (UEs) and MEC servers (MECSs) by jointly optimizing MECS active/sleep mode selection, UE-MECS association decision (MSAD), and task offloading, communication, and computation resource allocation (TORA) across two different timescales. Considering that the involved optimization variables affect the system performance in different timescales, we decouple the formulated stochastic optimization problem into two subproblems: MSAD operating in the large timescale and TORA in the small timescale. Leveraging the Lyapunov method, the stochastic TORA problem is decoupled into a series of deterministic problems, of which the closed-form solutions are presented. Furthermore, the MSAD problem is reformulated as a constrained Markov decision process (MDP). Then, we propose a double dueling deep Q-network (D3QN) to learn the optimal MSAD based on the TORA results. Numerical results demonstrate that the proposed online TORA-assisted MSAD learning algorithm has effective convergence and achieves substantial energy reductions for the MEC networks compared with the benchmark schemes.

In this paper, we investigate the power allocation problem to maximize the long-term average downlink sum-rate for orthogonal frequency division multiplexing (OFDM) based multi-cell networks. The traditional centralized training and centralized execution (CTCE) scheme is impractical to solve this complex problem, because it is hard for training center to obtain the global information, and the signaling overhead is also unbearable. To this end, a centralized training and distributed execution (CTDE) framework for wireless power control based on deep reinforcement learning (DRL) is proposed. Specifically, we first transform this problem into a sequential decision problem by designing appropriate state, action and reward. Then, a CTDE scheme, where each agent only requires its own local observations for power allocation, is devised through combining QMIX algorithm and Normalized Advantage Functions (NAF) algorithm. In particular, QMIX network is used to fit the total Q-value function and NAF network is deployed to handle continuous power output. Simulation results show that the proposed scheme converges fast and achieves better rate performance compared with conventional CTDE algorithms.

In this paper, the problem of the deployment and resource management for visible light communication (VLC)-enabled, reconfigurable intelligent surfaces (RISs)-assisted unmanned aerial vehicle (UAV) networks is investigated. In the considered model, UAVs provide terrestrial users with wireless services and illumination simultaneously. Moreover, RISs are utilized to further improve the channel quality between UAVs and users. This joint placement and resource management problem is constructed aiming at acquiring the optimal UAV deployment, RISs phase shift, user and RIS association that satisfies the users’ needs with minimum consumption of the UAVs’ energy. An iterative algorithm that alternately optimizes continuous and binary variables is proposed to solve this mixed-integer programming problem. Specifically, RISs phase shift optimization is solved by phases alignment method and semidefinite program algorithm. Next, the successive convex approximation algorithm is proposed to settle the UAV deployment problem. The user and RIS association variables are relaxed to the continuous ones before adopting the dual method to find the optimal solution. Moreover, a greedy algorithm is proposed as an alternative to RIS association optimization with low complexity. Simulation results show that the proposed two schemes harvest the superior performance of 34.85% and 32.11% energy consumption reduction over the case without RIS, respectively.

In this paper, the problem of power allocation is investigated for a multi-user downlink visible light communication (VLC) system with non-orthogonal multiple access (NOMA). In this considered system, not only input-independent Gaussian noise but also input-dependent shot noise are considered due to the properties of realistic VLC channels. This problem is posed as a joint problem of alternating current power and direct current (DC) power under the optical and electrical domain constraints in VLC as well as specific power constraints for NOMA decoding, whose goal is to maximize the minimum signal to interference plus noise ratio (SINR) among all users. Although this problem is non-convex, a geometric programming (GP) based algorithm is proposed to convert the original problem into a convex one, thus the optimal solution is obtained. Furthermore, special cases where lower DC offset is preferred are investigated. Simulation results verify that the DC offset has significant impacts on the performance of NOMA VLC systems with shot noise, and our proposed scheme achieves better SINR performance over the conventional schemes.

The multi-task federated learning (FL) problem in the wireless communication system is investigated in this paper. The base station (BS) and wireless users cooperatively perform a two-task FL algorithm in the established model. Users use their local datasets to train two local models of two different tasks. The trained local model of only one task is transmitted to the BS at each time and the BS aggregates the obtained models to calculate a global model, which will be sent back to all users. Since the resources for wireless transmission, such as transmit power and number of subcarriers are limited, the BS have to allocate resources reasonably to minimize the time consumption of the FL procedure while meeting the required learning performance. On the other hand, users are dynamically arranged to participate in different tasks in each iteration. This resource allocation and users arrangement problem is formulated as an optimization problem which aims to minimize time consumption of the two-task FL procedure. To address this nonconvex problem, we first decompose it into two convex sub-problems. Then we propose an iterative algorithm to solve this problem via iteratively obtaining the optimal solution of the joint power control and communication round optimization subproblem, and user arrangement subproblem. Simulation results of this multi-task FL system show that the proposed algorithm can reduce 7.02% and 9.67% completion time compared to the uniform and random user selection schemes respectively.

This paper investigates the problem of secrecy energy efficiency for an unmanned aerial vehicle (UAV) assisted joint sensing and communication system. In the considered system, there exists one UAV, one legal user, and one eavesdropper. The UAV needs to complete multiple tasks in multiple time cycles. In each time cycle, the UAV first flies to sense one task and then transmits the sensing results to the legal user. To maximize the secrecy energy efficiency of the system, a joint sensing and transmission time and UAV location optimization problem is formulated. To solve this non-convex fractional programming problem, the original problem is first divided into two subproblems. Each sub-problem can be easily transformed to a convex one by using the successive convex approximation (SCA) method and the Dinkelbach’s approach. Then, an iterative algorithm based on the alternating method is proposed. Simulation results reveal that our proposed algorithm is superior to the conventional algorithms in terms of secrecy energy efficiency.

This paper investigates the optimization of physical layer security in multiple-input multiple-output (MIMO) enabled visible light communication (VLC) networks. In the considered model, one transmitter equipped with light-emitting diodes (LEDs) intends to send confidential messages to legitimate users while one eavesdropper attempts to eavesdrop on the communication between the transmitter and legitimate users. This security problem is formulated as an optimization problem whose goal is to minimize the sum mean-square-error (MSE) of all legitimate users while meeting the MSE requirement of the eavesdropper thus ensuring the security. To solve this problem, the original optimization problem is first transformed to a convex problem using successive convex approximation. An iterative algorithm with low complexity is proposed to solve this optimization problem. Simulation results show that the proposed algorithm can reduce the sum MSE of legitimate users by up to 40% compared to a conventional zero forcing scheme.