Reliable underwater acoustic communication (UWAC) is critical for distributed maritime operations, unmanned systems, and subsea sensor networks. In defence and security contexts, even brief communication failures can cascade into catastrophic outcomes, such as loss of control of unmanned assets, delayed threat detection, or compromised situational awareness across a fleet. Yet, UWAC performance is often degraded by the complex and dynamic nature of the ocean—including turbulence, internal waves, and thermohaline gradients—that distort acoustic signals and create unpredictable channel environments. Traditional approaches to modelling these dynamics rely on computational fluid dynamics (CFD) or full-physics hydrodynamic models, which, while accurate, are computationally intensive and unsuitable for real-time applications. In this context, we propose the TIDE ACOMMs (Turbulence Inference with Data-driven Engine for Acoustic Communications) Systems, a novel machine learning framework that leverages controlled Generative Adversarial Networks (cGANs) to predict high-fidelity, 3D ocean dynamics in real time. Building on our Machine Learning–Accelerated Fluid Dynamics system, TIDE ACOMMs replaces computationally demanding simulations with fast, data-driven surrogate models, delivering actionable environmental intelligence directly into communication protocols. Our methodology is centred on a dual-network cGAN architecture trained on high-fidelity oceanographic data, CFD analyses, and observational data. The first network generates 2D planar slices of ocean dynamics from sparse environmental inputs—including wind, sea state, and limited velocity measurements. These slices are then assembled by the second network into volumetric 3D domains of pressure, salinity, and velocity (u, v, w), enhanced by self-attention and cross-attention mechanisms to maintain physical accuracy. Early results show that this approach can produce high-resolution (>128³) reconstructions at inference rates of 35 kHz with memory footprints (~6 GB) smaller and faster than state-of-the-art generative and diffusion-based methods. By rapidly approximating turbulence and thermohaline structure, TIDE ACOMMs enables real-time optimisation of acoustic signal parameters—including adaptive modulation and coding (AMC), frequency selection, transmission depth, and even dynamic sensor node placement. Embedding environmental intelligence in this way transforms UWAC from reactive to environmentally adaptive, allowing autonomous systems and distributed sensor webs to “sense” their fluid surroundings and adjust communications to maximise clarity, range, and throughput. The novelty of this work lies in addressing whether sparse, noisy environmental inputs can be mapped to reliable 3D ocean states quickly enough to enhance communications. Unlike prior generative models constrained to low-resolution grids, two-dimensional outputs, or high-latency inference, TIDE ACOMMs delivers a scalable solution for operational use. To mitigate risks associated with noisy data or sparse sampling, the project employs physics-informed augmentation, synthetic training data, and lightweight model variants optimised for edge deployment. Ultimately, TIDE ACOMMs aims to transform underwater communications by operationalising the synergy between hydrodynamic prediction and acoustic signal processing, enabling resilient and adaptive underwater acoustic communication.

Key Takeaways:
• Real-Time Ocean Prediction: TIDE ACOMMs uses a cGAN architecture to generate 3D ocean flow field predictions enabling real-time awareness of dynamic underwater conditions.
• Adaptive Acoustic Communication: The system integrates environmental predictions into UWAC protocols, optimising modulation, coding, and node placement to improve signal clarity.
• Operational Impact: Designed for deployment on autonomous platforms, TIDE ACOMMs supports environmentally adaptive communication networks and aligns with strategic priorities in undersea defence and multi-agent maritime coordination.