Chaofeng Wang

Underwater acoustic (UWA) networks suffer from long propagation delays and highly dynamic channels, making transmission scheduling difficult. While cooperative multi‐agent reinforcement learning (MARL) can improve performance, its conventional learning framework requires a central scheduler (CS) to collect full state information from all agents, incurring heavy communication overhead especially costly in UWA environments. In this study, we propose a cooperative MARL framework with compressed state sharing. Each agent encodes its local history, that is, queue length, CSI, and past transmission strategy, using a deep learning model to generate a compact state representation, which is then sent to the CS. The CS uses the aggregated compressed state to guide the joint scheduling and power allocation of all the agents. Simulation results confirm that the proposed method significantly reduces communication overhead while preserving high scheduling performance.

The transportation of nuclear radioactive materials is critical to the nuclear industry, yet any leakage during transit is unacceptable due to its severe consequences. To enable rapid localization and identification of a leaking container during transportation, we propose a deep learning (DL)-based method utilizing a sensor array system. A major challenge in deploying DL for real-world nuclear radiation monitoring during the transportation is the distribution shift between training and deployment environments, which often leads to overfitting and undermines model effectiveness. To simulate this discrepancy, we intentionally introduce distribution inconsistency between the training and testing datasets. To improve generalization and reduce overfitting under such mismatches, we adopt a multi-task (MT) learning framework in which several localization tasks are trained jointly. The localization results of these tasks are fused using a probability-informed neural network (PrINN), where the task-specific predictions are integrated based on their probabilistic relationships. Simulation results demonstrate that the proposed approach maintains strong generalization performance, even in the presence of significant discrepancies between the training phase and real-world deployment.

This study addresses transmission scheduling in underwater acoustic (UWA) networks by optimizing transmission schedule and power allocation according to time availability and channel variations. The objective is to maximize network energy efficiency (EE) while minimizing transmission latency (TL). To capture transmission and interference relationships among links, a spatial-temporal (ST) graph is designed to represent the system state. Transmission scheduling is then formulated as a sequential decision-making process (SDP), where the optimal strategy is adaptively determined based on the graph-based system representation. Network performance is optimized using graph reinforcement learning (GRL), specifically through a graph policy gradient (GPG) approach. Simulation results demonstrate that the proposed method achieves higher EE and lower TL compared to benchmark methods.

The feedback of downlink channel state information (CSI) is essential for adaptive resource allocation to improve performance of orthogonal frequency-division multiplexing (OFDM) systems. However, excessive feedback incurs significant communication overhead. In this study, we propose UWAcsiNet, a deep learning (DL)-based model that encodes the downlink CSI feedback directly into compact codewords while preserving critical channel characteristics for efficient and accurate recovery at the transmitter. Specifically, a data processing method is designed to guide the focus of model towards sparsely distributed channel features. Furthermore, a binarization layer is employed to dynamically allocate information bits within the codeword, enabling efficient bit-level compression. Simulation results show that the proposed model achieves lower reconstruction error with limited feedback, even under extremely high compression ratios, while maintaining decent computational complexity.

Electrocardiogram (ECG) signals are crucial in diagnosing cardiovascular diseases (CVDs). While wavelet-based feature extraction has demonstrated effectiveness in deep learning (DL)-based ECG diagnosis, selecting the optimal wavelet base poses a significant challenge, as it directly influences feature quality and diagnostic accuracy. Traditional methods typically rely on fixed wavelet bases chosen heuristically or through trial-and-error, which can fail to cover the distinct characteristics of individual ECG signals, leading to suboptimal performance. To address this limitation, we propose a reinforcement learning-based wavelet base selection (RLWBS) framework that dynamically customizes the wavelet base for each ECG signal. In this framework, a reinforcement learning (RL) agent iteratively optimizes its wavelet base selection (WBS) strategy based on successive feedback of classification performance, aiming to achieve progressively optimized feature extraction. Experiments conducted on the clinically collected PTB-XL dataset for ECG abnormality classification show that the proposed RLWBS framework could obtain more detailed time-frequency representation of ECG signals, yielding enhanced diagnostic performance compared to traditional WBS approaches.

In exploiting large propagation delays in underwater acoustic (UWA) networks, the time-domain interference alignment (TDIA) mechanism aligns interference signals through delay-aware slot scheduling, creating additional idle time for improved transmission at the medium access control (MAC) layer. However, perfect alignment remains challenging due to arbitrary delays. This study enhances TDIA by incorporating power allocation into its transmission scheduling framework across the physical and MAC layers, following the cross-layer design principle. The proposed quasi-interference alignment (QIA) mechanism enables controlled interference on useful signals by jointly optimizing the transmission schedule and power. The formulated optimization problem to maximize network throughput is divided into two sub-problems: one for coarse slot scheduling and another for refining both scheduling and power allocation. The simulation results validate the QIA framework’s superiority over the traditional TDIA and genetic algorithm benchmarks.

Underwater acoustic communication networks can satisfy the need for long-distance reliable communication for underwater node, serving as a key technology for the informatization and intelligentization of such communication. However, due to the use of sound waves as a carrier for information, underwater acoustic networks face complex challenges such as long propagation delays, severe channel attenuation, multipath effects, and environmental noise. To enhance the communication efficiency and reduce network congestion of underwater acoustic networks, this study proposes a MAC protocol based on deep reinforcement learning to adaptively adjust the transmission strategy of network nodes. Each network node collaboratively decides the optimal transmission slot based on its own data buffer status and the history of transmission conflicts, to suppress transmission conflicts and thereby maximize transmission rates. Simulation results show that the designed protocol significantly improves the network throughput while ensuring the fairness of transmission among nodes, providing an important foundation for the rapid exchange of information between underwater nodes.