Yong Ren

With the integrated development of global marine exploitation and 6G technology, building an all-domain marine wireless network has become crucial for supporting marine activities. However, the unique communication environment, varying collaboration of heterogeneous devices, and dynamic network changes pose technical bottlenecks for balancing real-time and efficient resource competition. To overcome those challenges, this paper proposes a novel integrated marine wireless network with multi-type unmanned device clusters across space-surface-submarine media. To address heterogeneous resource allocation, we consider channel capacity and device connection, modeling it as a three-sided matching framework with size constraints and cyclic preferences (TMSC). Building on this, we propose the satellite-prioritized restricted double-TMSC (SPR-DT) algorithm to solve optimal matching in quasi-static networks, aiming to maximize total backhaul revenue. To handle rapid dynamic network changes, we initialize the proximal policy optimization (PPO) with the stable solution of SPR-DT, thus addressing the challenge of acquiring real training data while accelerating algorithm convergence. Then, we propose a PPO-assisted multi-slot matching algorithm to enhance solution efficiency in large-scale dynamic scenarios. The simulation results show that the proposed algorithm achieves an optimal effect of 94.6% in quasistatic scenarios, with a complexity reduced to 3.2%. In dynamic scenarios, the results are 87.2% and 28.7%, respectively.

The unmanned aerial vehicle (UAV) plays an indispensable role in catalyzing the advancement of the low-altitude economy (LAE). However, the existing UAV motion control strategies are vulnerable and susceptible to specified time constraints, unknown environmental interferences, dynamic topology and sensor malfunctions. To address the aforementioned issues, this paper proposes a deep reinforcement learning (DRL)-based fixed-time formation control strategy for networked UAVs with topology reconfiguration capability in the presence of wind disturbances. Firstly, a DRL-driven topology reconfiguration algorithm is designed, leveraging a deep deterministic policy gradient, to address communication channel loss in UAV formation networks caused by sensor faults, which can be achieved by determining the optimal topology from a set of directed graphs. Furthermore, based on a simple filtering operation on state variables and invariant manifolds, an unknown system dynamics estimator (USDE) is developed for UAVs to counteract external wind disturbances effectively. Finally, a fixed-time UAV formation controller with disturbance rejection is elaborated to propel UAVs to converge to the desired trajectory within an explicit time, independent of their initial conditions. Simulation results validate the effectiveness of the proposed method.

Integrated sensing and communication (ISAC) technology is crucial for next-generation underwater networks. However, covering multiple users and targets and balancing sensing and communication performance in complex underwater acoustic (UWA) environments remains challenging. This paper proposes an interleaved orthogonal frequency division multiplexing-based MIMO UWA-ISAC system, which employs a horizontal array to simultaneously transmit adaptive waveforms for downlink multi-user communication and omnidirectional target sensing. A multi-objective optimization framework is formulated to maximize the product of communication rate and range (PRR) while ensuring sensing performance and peak-to-average power ratio (PAPR) constraints. To solve this mixed-integer nonconvex problem, a two-dimensional grouped random search algorithm is developed, efficiently exploring subcarrier interleaved patterns and resource allocation schemes. Numerical simulations under real-world UWA channels demonstrate the designed system's superiority and effectiveness: our algorithm achieves 90% faster convergence than conventional exhaustive search with only a marginal 0.5 kbps km PRR degradation. Furthermore, the proposed resource allocation scheme maintains robustness beyond the baseline allocation schemes under stringent PRR and PAPR constraints.

Network optimization remains fundamental in wireless communications, with Artificial Intelligence (AI)-based solutions gaining widespread adoption. As Sixth-Generation (6G) communication networks pursue full-scenario coverage, optimization in complex extreme environments presents unprecedented challenges. The dynamic nature of these environments, combined with physical constraints, makes it difficult for AI solutions such as Deep Reinforcement Learning (DRL) to obtain effective reward feedback for the training process. However, many existing DRL-based network optimization studies overlook this challenge through idealized environment settings. Inspired by the powerful capabilities of Generative AI (GenAI), especially diffusion models, in capturing complex latent distributions, we introduce a novel Diffusion Model-based Reward Shaping Scheme (DRESS) to achieve robust network optimization. By conditioning on observed environmental states and executed actions, DRESS leverages diffusion models’ multi-step denoising process to refine latent representations progressively, generating meaningful auxiliary reward signals that capture patterns of network systems. Moreover, DRESS is designed for seamless integration with any DRL framework, allowing DRESS-aided DRL (DRESSed-DRL) to enable stable and efficient DRL training even under extreme network environments. Experimental results demonstrate that DRESSed-DRL achieves about 1.5× times faster convergence than its original version in sparse-reward wireless environments and significant performance improvements in multiple general DRL benchmark environments compared to baseline methods. The code of DRESS is available at https://github.com/NICE-HKU/DRESS.

It is significant to employ multiple autonomous underwater vehicles (AUVs) to execute the underwater target tracking task collaboratively. However, it’s pretty challenging to meet various prerequisites utilizing traditional control methods. Therefore, we propose an effective two-stage learning from demonstrations training framework, FISHER, to highlight the adaptability of reinforcement learning (RL) methods in the multi-AUV underwater target tracking task, while addressing its limitations such as extensive requirements for environmental interactions and the challenges in designing reward functions. The first stage utilizes imitation learning (IL) to realize policy improvement and generate offline datasets. To be specific, we introduce multi-agent discriminator-actor-critic based on improvements of the generative adversarial IL algorithm and multi-agent IL optimization objective derived from the Nash equilibrium condition. Then in the second stage, we develop multi-agent independent generalized decision transformer, which analyzes the latent representation to match the future states of high-quality samples rather than reward function, attaining further enhanced policies capable of handling various scenarios. Besides, we propose a simulation to simulation demonstration generation procedure to facilitate the generation of expert demonstrations in underwater environments, which capitalizes on traditional control methods and can easily accomplish the domain transfer to obtain demonstrations. Extensive simulation experiments from multiple scenarios showcase that FISHER possesses strong stability, multi-task performance and capability of generalization.

Emerging real-time computer vision (CV) applications on wireless edge devices demand energy-efficient and privacy-preserving learning. Federated learning (FL) enables on-device training without raw data sharing, yet remains challenging in resource-constrained environments due to energy-intensive computation and communication, as well as limited and non-i.i.d. local data. We propose FedDPQ, an ultra energy-efficient FL framework for real-time CV over unreliable wireless networks. FedDPQ integrates diffusion-based data augmentation, model pruning, communication quantization, and transmission power control to enhance training efficiency. It expands local datasets using synthetic data, reduces computation through pruning, compresses updates via quantization, and mitigates transmission outages with adaptive power control. We further derive a closed-form energy-convergence model capturing the coupled impact of these components, and develop a Bayesian optimization(BO)-based algorithm to jointly tune data augmentation strategy, pruning ratio, quantization level, and power control. To the best of our knowledge, this is the first work to jointly optimize FL performance from the perspectives of data, computation, and communication under unreliable wireless conditions. Experiments on representative CV tasks show that FedDPQ achieves superior convergence speed and energy efficiency.

Autonomous agents, including unmanned aerial vehicles (UAVs), unmanned ground vehicles (UGVs), unmanned surface vessels (USVs), and unmanned underwater vehicles (UUVs), are widely applied across diverse fields, such as environmental monitoring, logistics, exploration, and military applications, due to their ability to operate autonomously in complex environments. As tasks grow in complexity, there is a shift from individual autonomous agents to collaborative autonomous agent swarms (AASs). These swarms, connected through wireless links, leverage collective intelligence to perform more sophisticated tasks than individual autonomous agents can achieve alone. Deploying AASs introduces challenges in coordination, resource utilization, and adaptability, which traditional model-based methods struggle to address. Recent advancements in machine learning (ML), particularly distributed machine learning (DML) techniques such as federated learning (FL) and multi-agent reinforcement learning (MARL), offer promising solutions. These techniques enable autonomous agents to learn collaboratively without centralizing data, thereby preserving efficiency while adapting to dynamic environments. However, applying DML in AASs presents unique challenges due to autonomous agents’ dynamic movement, unreliable communication links, harsh operating conditions, heterogeneity, and resource constraints. Therefore, this survey provides a comprehensive review of integrating DML into AASs. Specifically, we analyze four representative types of autonomous agents in a unified perspective examining their fundamental characteristics, communication models, and dynamic behaviors. Then, we discuss the limitations of basic ML models in AASs, highlighting the need for DML and the key requirements and metrics for its successful implementation. Furthermore, we explore recent advancements in FL and MARL as applied to AASs, highlighting use cases and key techniques. By identifying current technological progress and gaps in the literature, this survey offers valuable insights for researchers and practitioners and outlines potential directions for future research to enhance the capabilities of AASs through DML.

With the exponential growth of smart devices connected to wireless networks, data production is increasing rapidly, requiring machine learning (ML) techniques to unlock its value. However, the centralized ML paradigm raises concerns over communication overhead and privacy. Federated learning (FL) offers an alternative at the network edge, but practical deployment in wireless networks remains challenging. This paper proposes a lightweight FL (LTFL) framework integrating wireless transmission power control, model pruning, and gradient quantization. We derive a closed-form expression of the FL convergence gap, considering transmission error, model pruning error, and gradient quantization error. Based on these insights, we formulate an optimization problem to minimize the convergence gap while meeting delay and energy constraints. To solve the non-convex problem efficiently, we derive closed-form solutions for the optimal model pruning ratio and gradient quantization level, and employ Bayesian optimization for transmission power control. Extensive experiments on real-world datasets show that LTFL outperforms state-of-the-art schemes.

In constrained signal processing, which encompasses areas such as compressed sensing, noisy signal recovery, and matrix completion, the communication overhead of gradients, both inter- and intra-process, often emerges as a substantial bottleneck. Gradient compression significantly alleviates this problem by sending low-bit gradients while ensuring gradient descent convergence. Although such low-bit technique is effective in application, the theoretical basis still remains largely unexplored. In this work, we establish a unified framework for the convergence analysis of projected optimization methods with low-bit gradients, especially from the perspective of continuous-time nonsmooth dynamical systems. Moreover, we propose a provably convergent distributed gradient compression scheme for constrained low-bit signal processing applications. Numerical experiments are conducted to confirm the validility of the theoretical analysis, showing that our distributed algorithm effectively transmits low-bit gradients with negligible effect on the convergence rate for constrained nonconvex optimization.