Fig 1.
UWSN communication scenario with concurrent communication.
Fig 2.
Mobility model of underwater sensor nodes.
Fig 3.
Workflow diagram of MARL.
Fig 4.
Training and distributed execution architecture.
Fig 5.
UWSN communication scenarios including node failures.
Fig 6.
Process of semi-cooperative power allocation mechanism.
Table 1.
Parameter settings.
Fig 7.
Network performance during the training phase.
Table 2.
Performance of the MARL model and other comparison methods.
Fig 8.
Training curves under different network densities in the HomNet scenario.
Fig 9.
Training curves under different network densities in the HetNet scenario.
Table 3.
Average latency of MARL model and baseline methods under different network densities.
Table 4.
Simulation results under different synchronization error levels (Shallow sea complex channel, D = 4).
Fig 10.
The influence of SINR thresholds on network performance in various scenarios.
Fig 11.
The influence of water flow velocity on network performance in different scenarios.
Fig 12.
The impact of mobile node interference on network performance.
Table 5.
Comparison of inference overhead among different micro-controller platforms.
Table 6.
Performance comparison statistics results with modern MARL baselines (HetNet scenario, n = 10).
Table 7.
Results of ablation experiment.