Federated learning (FL) has emerged as a promising paradigm for privacy-preserving model training across distributed edge devices, enabling local data utilization without explicit sharing. However, in edge computing environments characterized by heterogeneous resources and intermittent connectivity, FL remains vulnerable to gradient leakage attacks (GLA), where adversaries reconstruct private data from shared model updates. Although the existing defenses, such as differential privacy (DP) and gradient compression, offer partial mitigation, they often result in significant performance degradation or increased communication overhead. In this paper, we analyze that the risk of privacy leakage is highly sensitive to the client-side training configurations and gradient magnitudes. Based on this, we propose a risk-aware FL framework tailored for the edge scenarios, which not only performs per-device privacy risk assessment but also introduces subtractive dithering quantization to the inject controllable Gaussian noise into local models. Additionally, a noise-aware aggregation strategy is presented by adjusting each client’s contribution to preserve the global model utility. Experimental results on FashionMNIST and CIFAR-10 demonstrate that the proposed framework achieves strong defense against the GLA, reduces the communication costs by over 50%, and maintains the competitive accuracy.

Abstract Edge computing presents a promising approach for achieving communication Quality of Service (QoS) by employing a task offloading strategy to transfer latency-sensitive tasks into edge servers. Considering the offload equalization challenge, in this paper, we propose a novel task offloading optimization method based on a Hybrid Whale Genetic Algorithm (HWGA) with Reinforcement Learning (RL) to optimize the task offloading decisions within a tri-layer edge computing architecture comprising edge, fog, and cloud layers. Due to the expansive dimensionality of the action space from the increasing number of devices, we adapt the RL into a multi-layer architecture. In this framework, multi-layer RL techniques are first utilized to determine which layer should handle the task offloading. Subsequently, the HWGA is applied to guide the task offloading decisions for devices within each layer. Simulation results demonstrate that, when compared to baseline methods, our HWGA-based approach significantly reduces task completion time and energy consumption, while improving the task success rate, particularly in high-device-density scenarios.

Low peak-to-average power ratio (PAPR) orthogonal frequency division multiplexing (OFDM) waveform design is a crucial issue in integrated sensing and communications (ISAC). This paper introduces an OFDM-ISAC waveform design that utilizes the entire spectrum simultaneously for both communication and sensing by leveraging a novel degree of freedom (DoF): the frequency-domain phase difference (PD). Based on this concept, we develop a novel PD-based OFDM-ISAC waveform structure and utilize it to design a PD-based Low-PAPR OFDM-ISAC (PLPOI) waveform. The design is formulated as an optimization problem incorporating four key constraints: the time-frequency relationship equation, frequency-domain unimodular constraints, PD constraints, and time-domain low PAPR requirements. To solve this challenging non-convex problem, we develop an efficient algorithm, ADMM-PLPOI, based on the alternating direction method of multipliers (ADMM) framework. Extensive simulation results demonstrate that the proposed PLPOI waveform achieves significant improvements in both PAPR and bit error rate (BER) performance compared to conventional OFDM-ISAC waveforms.

The widespread adoption of computation- and communication-intensive applications, such as object detection, VR/AR, and telemedicine, has significantly alleviated transmission pressure on backbone networks and improved user experience. However, efficiently managing and computing these tasks on user sides remains a significant challenge, particularly under resource-constrained conditions. To address this problem, we propose a new service caching decision method based on deep dueling double Q-network (D3QN) by employing a learnable policy to handle the unknown task requests and determine the optimal caching strategies. Additionally, the limited storage capacity of edge servers (ES) is mitigated by forwarding the resource-intensive or infrequently requested tasks to the cloud data centers (CDC). The channel selection problem is modeled as a multiuser game and a distributed method is developed to achieve the Nash Equilibrium (NE). Simulation results demonstrate that the proposed method outperforms the existing benchmarks, showcasing its effectiveness in managing complex, dynamic environments.

Computing offloading optimization for energy saving is becoming increasingly important in low-Earth orbit (LEO) satellite-terrestrial integrated networks (STINs) since battery techniques have not kept up with the demand of ground terminal devices. In this paper, we design a delay-based deep reinforcement learning (DRL) framework specifically for computation offloading decisions, which can effectively reduce the energy consumption. Additionally, we develop a multi-level feedback queue for computing allocation (RAMLFQ), which can effectively enhance the CPU’s efficiency in task scheduling. We initially formulate the computation offloading problem with the system delay as Delay Markov Decision Processes (DMDPs), and then transform them into the equivalent standard Markov Decision Processes (MDPs). To solve the optimization problem effectively, we employ a double deep Q-network (DDQN) method, enhancing it with an augmented state space to better handle the unique challenges posed by system delays. Simulation results demonstrate that the proposed learning-based computing offloading algorithm achieves high levels of performance efficiency and attains a lower total cost compared to other existing offloading methods.

Satellite communication networks with the characteristics of wide coverage, high deployment flexibility, and seamless communication services can provide communication services to users who don’t communicate with ground networks but directly communicate with satellites. In response to the increasing demand for user services, this paper proposes a collaborative computing offloading scheme for satellite edge computing networks with a four-layer architecture. By utilizing collaborative computing between ground users and three layers of satellites (low-orbit satellites, edge, and cloud data centers), the service quality for ground users is improved. Considering the mobility of vehicles and satellite nodes, the frequent changes in link states further complicate the design and implementation of such systems, leading to increased latency and energy consumption. This paper proposes to optimize the computation offloading decision while satisfying the constraint of satellite computing capabilities, aiming to improve the success rate of tasks and minimize the overall cost of the system. However, with the increase in the number of ground users and satellites, the formulated problem becomes a mixed-integer nonlinear programming (MINLP) problem, which is difficult to solve with general optimization algorithms. To address this issue, this paper proposes a dynamic distributed learning offloading (DDLDO) algorithm based on distributed deep learning. The algorithm utilizes multiple parallel deep neural networks (DNN) to dynamically learn computation offloading strategies. Simulation results demonstrate that the algorithm outperforms other benchmark algorithms in terms of latency, energy consumption, and successful execution efficiency.

As the demand for global Internet connectivity continues to grow, the satellite-terrestrial integrated networks (STINs) have become more and more crucial for expanding the service coverage and enhancing the network performance. However, the task offloading problem in STINs faces many significant challenges, such as high processing latency and energy consumption. The current intelligent offloading strategies often rely on the real-time interactions with the environments which not only consume valuable satellite resources but also cause irreversible damage to the satellite equipment due to some operational errors. To address these issues, in this article, we propose an offline deep reinforcement learning (offline DRL) approach to learn and optimize the task offloading decisions by leveraging the stored historical decision data and employing the soft actor-critic (SAC) algorithm specifically. Experimental results show that the proposed strategy outperforms most of the existing methods in terms of latency and energy consumption and effectively reduces the direct interactions with STINs.

Unmanned aerial vehicles (UAVs)-assisted mobile-edge computing (MEC) communication system has recently gained increasing attention. In this article, we investigate a 3-D multi-UAV trajectory optimization based on ground devices (GDs) selecting the target UAV for task computing. Specifically, we first design a 3-D dynamic multi-UAV-assisted MEC system in which GDs have real-time mobility and task update. Next, we formulate the system communication, computation, and flight energy consumption as objective functions based on fairness among UAVs. Then, to pursue fairness among UAVs, we theoretically deduce and mathematically prove the optimal GDs’ selectivity and offloading strategy, that is, how GDs select the optimal UAV for task offloading and how much to offload. While ensuring the optimal offloading strategy and GDs’ selectivity between UAVs and GDs at each step, we model UAV trajectories as a sequence of location updates of all UAVs and apply a multiagent deep deterministic policy gradient (MADDPG) algorithm to find the optimal solution. Simulation results demonstrate that we achieve the minimum energy consumption under the premise of fairness and the efficiency of model processing tasks.

In this article, we investigate the average transmission delay performance for the distributed multiple access cooperative communication systems with splitting opportunistic relay selection in the presence of strong co-channel interference (CCI). We consider a distributed wireless network with random access and analyze the impact of imperfect channel state information (CSI) due to CCI on latency. In general, the splitting relay selection algorithm is provably fast and scalable in distributed wireless networks. Exact closed-form performance expression is derived for the end-to-end average delay with the range of both transmission phase and relay selection phase. Numerical and simulation results are all provided to show that when the CSI is outdated or error estimation, the splitting opportunistic relay system has a performance degradation compared to the perfect CSI situation, while it is better than other opportunistic relay selection schemes with outdated or error CSI. Furthermore, the explanation for this performance loss is provided along with theoretical results to support the analytical study.

This study presents a ‘subblock tracking’ based equalisation technique for orthogonal frequency division multiplexing (OFDM) based device‐to‐device communications over high mobility environment. This technique is to use a rectangular‐shaped receive window to partition the OFDM block into subblocks rendering the channel response of each subblock to be time‐invariant, thereby allowing one to equalise the channel frequency response on each subcarrier of the subblocks simply by a single tap. The equalised signals are then combined to give the final output, where the combination weights are designed to minimise detection errors. The authors mathematically characterise the detection performance by deriving the signal‐to‐interference ratio as a function of channel‐time‐variation and the numbers of subblocks. To further enhance the proposed design without imposing computational consumption, a subblock tracking scheme with partially overlapped partition is presented and theoretically analysed. Simulation results demonstrate the advantages this method over the conventional schemes in high mobility environment.

ABSTRACT An approach of superimposed training (ST)‐aided time‐varying (TV) channel estimation for multiple‐input multiple‐output orthogonal frequency division multiplexing systems is presented. By modeling the TV channel with the truncated discrete basis expansion model, a two‐step approach is adopted to estimate the TV channel. In addition, the mean square error (MSE) of the proposed channel estimation is analyzed, and its closed‐form expression is derived, which is a function of the data‐to‐ST power ratio. Using the developed channel MSE, we case the problem of ST power‐allocation by minimizing the lower bound on the average channel capacity. To enhance the performance of channel estimation, a low‐complexity decision feedback mechanism is introduced to iteratively mitigate the unknown data interference. Numerical results verify the performances of the proposed approach. Copyright © 2012 John Wiley & Sons, Ltd.