UDDPG-Net: Energy- and Fairness-Aware Deep Reinforcement Learning for Resource Allocation in Underwater Communication Networks

Abstract

This dataset supports the study “UDDPG-Net: Energy- and Fairness-Aware Deep Reinforcement Learning for Resource Allocation in Underwater Communication Networks.” It contains the simulation data, experimental results, and lake-based acoustic measurement records used for evaluating UDDPG-Net. The simulation data cover underwater acoustic propagation, mobility, interference, energy states, resource-allocation decisions, training configurations, benchmark comparisons with Random, Greedy, PPO, TD3, SAC, MADDPG, and DDPG-GA, as well as scalability, ablation, exploration, SHAP, computational-efficiency, and robustness analyses. The field-test component corresponds to six campaigns conducted in Fengquan Lake, Xinxiang, China, and includes environmental, motion, acoustic-channel, and packet-level variables such as water temperature, salinity, pH, conductivity, sound speed, ambient-noise PSD, link distance, subband SNR, propagation delay, Doppler shift, and packet reception ratio. Reconstructed raw-format field logs are additionally provided to reproduce the organization and statistical characteristics reported in the manuscript; these files are clearly identified as reconstructed records and should not be interpreted as original instrument acquisition files. README files and structured metadata are included to describe the variables, units, experimental conditions, and correspondence between the released data and the figures and tables in the paper.