Underwater acoustic communication channel modeling: methods, tools, and development trends

Underwater acoustic communication algorithms require simulation-based testing and performance evaluation under a variety of propagation conditions. However, the high cost of sea trials and the limited availability of measured data constrain the evaluation of these algorithms across diverse propagation scenarios. This review classifies channel modeling methods into deterministic and stochastic approaches based on their underlying modeling mechanisms and treatment of channel uncertainty. The inputs, outputs, and applicability of these methods are compared, and their applications and limitations in communication performance evaluation are examined in conjunction with the associated simulation tools. Deterministic models use specified environmental parameters or measured channel records to calculate acoustic propagation characteristics or construct channel responses. Direct replay preserves the recorded multipath structure and temporal variations, enabling communication schemes to be compared repeatedly under identical measured conditions. However, its applicability is limited by the measurement bandwidth, recording duration, and range of scenarios covered. Empirical models estimate propagation loss using a small number of parameters and are suitable for link analysis, but cannot independently provide the multipath delays and complex gains required for waveform-level simulation. Propagation-theory-based models construct channel responses by calculating propagation paths or complex acoustic fields, allowing the effects of changes in environmental conditions and transceiver positions to be investigated. Their predictive accuracy depends on the accuracy of the environmental inputs, while constructing dynamic, broadband channel responses typically requires calculations across multiple frequencies and time instants, increasing computational costs. Stochastic models generate different channel realizations for repeated simulations. Geometry-based stochastic models combine random scattering with propagation geometry to investigate the effects of platform motion and array configurations, but the large number of geometric and scattering parameters can make model configuration complex. Statistical models characterize channel variations using probability distributions and correlation structures. Stochastic replay is one approach in this category, using statistics derived from measurements to generate new channel responses. Learning-based generative models learn the joint characteristics of channel responses from training samples, reducing reliance on manually specified statistical distributions. These approaches generate additional channel realizations for repeated simulations, enabling the variability of communication performance and algorithm robustness to be evaluated under the modeled conditions. However, their applicability to previously unseen propagation scenarios requires independent validation. Model selection should be guided by the level of channel detail required for the evaluation task and the available environmental information and measurement data. Propagation quantities, such as propagation loss, are sufficient for link analysis, whereas waveform-level testing requires channel responses that capture the relevant multipath structure and its variations. Specific models can then be selected based on bandwidth, platform motion, and array configuration requirements, while accounting for acceptable modeling errors and computational costs. Tool selection should likewise be guided by the required outputs and subsequent processing needs. Arrival information or complex acoustic fields obtained from propagation modeling tools must be converted into channel responses and processed at the waveform level for receiver testing. Network simulation tools, by contrast, translate propagation results into link-level information to evaluate packet transmission and network performance. A signal-level simulation and measured-channel replay framework, termed the hydro-acoustic replay benchmarking and operational repository (HARBOR), is proposed for evaluating underwater acoustic communication performance. It integrates model-generated and measured channel responses into a unified waveform-processing workflow, with dedicated modules for motion-induced Doppler effects, noise, and interference. The framework outputs received signals and records the corresponding channel responses and associated features. The current implementation supports single-input single-output (SISO) communication simulations. Future work should prioritize measurements of long-range propagation in deep-sea environments and highly dynamic platform motion to broaden test-set coverage and use independent measurement records to assess model applicability. Where corresponding communication records are available, simulation results can be compared with field measurements to evaluate the effects of channel modeling errors on performance predictions. Models with well-established ranges of applicability can then support the standardization of reference channel models for representative propagation conditions. This process should specify parameter ranges and channel response generation procedures, together with reference cases for verifying consistency across implementations. Communication tests should define power reference levels and signal-to-noise ratio (SNR) conventions and maintain consistent noise settings, while stochastic models should undergo statistical consistency checks across repeated realizations. As more independent measurements become available, reference scenarios can be expanded and model parameter ranges refined.