To understand why underwater communication is one of the most punishing domains in engineering, consider the physics of the ocean. Terrestrial networks rely on electromagnetic waves (like radio or Wi-Fi), which propagate through the air at the speed of light. In seawater, however, high-frequency electromagnetic waves are absorbed almost instantly.
As a result, subsea networks must rely on acoustic waves—sound—to transmit information over significant distances. But sound travels roughly five orders of magnitude slower than light, and the physics of underwater sound propagation inherently restricts the usable frequency spectrum. While a standard home Wi-Fi router operates over bandwidths of tens or hundreds of megahertz, underwater acoustic channels are often restricted to frequencies below 30 kHz, yielding throughputs measured in kilobits or even single bytes per second.
To build reliable networks in this environment, researchers cannot simply adapt terrestrial protocols. They must fundamentally redesign how data is compressed, how packets are scheduled, and how networks are structured.
Extreme Data Reduction at the Edge
When the communication pipe is narrow, the most effective strategy is to avoid sending raw data altogether. If a sensor node can process information locally and transmit only the absolute essence of its findings, it bypasses the bandwidth bottleneck.
In Cross-domain fusion in smart seafloor sensor networks, Zainab et al. propose shifting the burden from the communication channel to edge computation. By deploying resource-efficient AI models directly on seafloor nodes, raw, high-volume sensor streams are compressed into compact "patterns of interest." Instead of streaming continuous acoustic or seismic waveforms, the node transmits a tiny semantic summary when an event occurs.
For applications like target tracking, where continuous state updates are required, quantization offers a mathematical approach to minimizing data footprints. Zhang et al. demonstrate this in Optimal Quantization Scheme for Data-Efficient Target Tracking via UWSNs Using Quantized Measurements. Instead of transmitting high-precision floating-point coordinates, their scheme optimizes how tracking measurements are quantized into ultra-low-bit representations. The system maintains tracking accuracy not by sending more data, but by maximizing the information density of every transmitted bit.
Raw Data Stream (High Volume) ──> [ Edge Intelligence / Quantization ] ──> Compact Semantic Patterns (Low Bits) ──> Narrow Acoustic Channel
When visual data is mandatory, standard compression algorithms like JPEG or VVC fail because they do not account for the specific constraints of underwater imagery or the extreme limits of a ~20 kbps acoustic link. To solve this, Li et al. introduced RFD-ECNet in RFD-ECNet: Extreme Underwater Image Compression with Reference to Feature Dictionary.
The intuition behind their approach is elegant: rather than compressing an image from scratch, both the sender and receiver share a pre-constructed "feature dictionary" of common marine objects and textures. The underwater node only needs to transmit highly compressed structural features and dictionary indices, allowing the receiver to reconstruct high-quality visual details. This domain-specific knowledge yields massive bitrate savings over general-purpose compression standards. To evaluate how these heavily compressed streams perform under real-world constraints, Moreno-Roldan et al. developed a no-reference video quality assessment model in A No-Reference Video Quality Assessment Model for Underwater Networks, allowing networks to dynamically balance compression depth with the perceptual needs of ocean scientists.
Redesigning the Network Stack for High Latency
Terrestrial networking protocols rely heavily on chatty, conversational mechanisms. For instance, TCP ensures reliability by requiring the receiver to send an acknowledgment (ACK) for every packet; if an ACK doesn't arrive quickly, the sender retransmits the data. In an underwater environment where propagation delays are measured in seconds rather than milliseconds, this constant back-and-forth destroys throughput and rapidly drains node batteries.
To achieve reliability without the overhead of conversational handshakes, Liu et al. designed the Adaptive Redundancy Transport Protocol (ARRTP) in Internode Distance-Based Redundancy Reliable Transport in Underwater Sensor Networks. Instead of waiting for ACKs and retransmitting lost packets, ARRTP uses forward error correction (FEC). It proactively injects mathematical redundancy into the data stream so the receiver can reconstruct missing packets on the fly. Crucially, the protocol is distance-aware: it dynamically shifts between bit-level and packet-level redundancy based on the physical distance between nodes, ensuring that energy is not wasted on excessive redundancy over short hops.
Terrestrial (Chatty): [Sender] ───Packet───> [Receiver] ───ACK───> [Sender] (High overhead)
Underwater (Proactive): [Sender] ───[Data + Distance-Aware Redundancy]───> [Receiver] (Reconstructs locally)
At the Media Access Control (MAC) layer, managing channel access is equally challenging. If two nodes transmit simultaneously, their signals collide, destroying the data and wasting precious energy. In Collision Avoidance Energy Efficient Multi-Channel MAC Protocol for UnderWater Acoustic Sensor Networks, Bouabdallah et al. (and similarly Zidi et al. in MC-UWMAC: A multi-channel MAC protocol for underwater sensor networks) propose splitting the narrow acoustic spectrum into a single control channel and multiple data channels. By using a specialized quorum system to coordinate data channel assignments on the control channel, they isolate handshaking from data transfers, achieving collision-free communication despite long propagation delays.
This shift from bursty, high-throughput paradigms to highly structured, low-overhead messaging is critical for complex robotic coordination. In Developing an Underwater Network of Ocean Observation Systems With Digital Twin Prototypes - A Field Report From the Baltic Sea, Pech documents a field deployment where hydroacoustic channels were limited to an extreme bottleneck of 64 bytes per second. To maintain synchronization between physical subsea crawlers and their surface digital twins, the architecture relies on small, fixed-size packets and dynamic serialization schemes rather than high-throughput burst modes. This microservice-driven, highly serialized approach is expanded upon by Barbie et al. in Prototyping Autonomous Robotic Networks on Different Layers of RAMI 4.0 with Digital Twins to safely simulate and execute autonomous missions within these rigid communication boundaries.
Shifting to Optical and Hybrid Architectures
When applications demand multi-gigabit throughput—such as real-time high-definition video streaming or massive data offloading from seafloor observatories—acoustic communication hits a hard physical ceiling. To bridge this gap, researchers are turning to Underwater Optical Wireless Communications (UOWC).
As detailed by Saeed et al. in Underwater Optical Wireless Communications, Networking, and Localization: A Survey, light propagates through water at much higher frequencies than sound, enabling high-speed data transfer. For example, Oubei et al. demonstrated a record-breaking 4.8 Gbit/s link using a compact 450-nm blue laser in 48 Gbit/s 16-QAM-OFDM transmission based on compact 450-nm laser for underwater wireless optical communication, proving that optical links can handle high-volume data.
However, light introduces a severe trade-off: it suffers from intense scattering and absorption in turbid water, restricting its range to tens of meters, and it requires highly precise, line-of-sight alignment between nodes.
┌─────────────────────────────────────────────────────────────────────────┐
│ THE UNDERWATER COMMUNICATION TRADE-OFF │
├────────────────────────────────────────┬────────────────────────────────┤
│ ACOUSTIC │ OPTICAL │
├────────────────────────────────────────┼────────────────────────────────┤
│ • Long Range (Kilometers) │ • Short Range (Meters) │
│ • Omnidirectional / Robust │ • Strict Line-of-Sight │
│ • Severe Bandwidth Bottleneck (<30kHz)│ • Multi-Gigabit Throughput │
└────────────────────────────────────────┴────────────────────────────────┘
Because neither modality is perfect, modern subsea networks increasingly rely on hybrid architectures that combine the strengths of both. In AUV-Aided Optical—Acoustic Hybrid Data Collection Based on Deep Reinforcement Learning, Bu et al. propose a system where Autonomous Underwater Vehicles (AUVs) act as mobile data mules.
The AUV uses long-range, low-bandwidth acoustic signaling to coordinate its path and locate stationary seafloor sensors. Once the AUV maneuvers directly above a sensor node into perfect line-of-sight alignment, it switches to a high-speed optical link to rapidly harvest large volumes of data. By using deep reinforcement learning to optimize the AUV's trajectory and communication scheduling, the system minimizes data staleness while respecting the physical limitations of both light and sound in the ocean.
Key Foundational Work
Energy Balanced Strategies for Maximizing the Lifetime of Sparsely Deployed Underwater Acoustic Sensor Networks (2009) — contributed 1 concept elements to this direction.
Internode Distance-Based Redundancy Reliable Transport in Underwater Sensor Networks (2010) — contributed 1 concept elements to this direction.