In terrestrial wireless networks, we safely assume that signals travel instantaneously. Radio waves propagate at the speed of light ($3 \times 10^8$ m/s), meaning the time it takes for a packet to travel from a smartphone to a cell tower is negligible compared to the time it takes to transmit the data packet itself.
Underwater, this assumption completely breaks down. High-frequency electromagnetic waves are rapidly absorbed by saltwater, rendering traditional radio communications useless over long distances, as noted by Mohammad Jahanbakht et al. in their survey on the Internet of Underwater Things. Instead, underwater networks must rely on acoustic waves, as explored by Huimin Lu 0001 et al.. Sound travels through water at roughly $1500$ m/s—five orders of magnitude slower than light.
This sluggish propagation speed introduces a massive temporal challenge: the propagation delay ($t_{\text{prop}}$) becomes equal to or significantly greater than the packet transmission time ($t_{\text{tx}}$). In their foundational work on early acoustic networks, Josko Catipovic et al. pointed out that this inversion of scale fundamentally invalidates standard networking protocols. Managing this extreme propagation delay is the central challenge of underwater network design.
The Breakdown of Carrier Sensing
To understand why underwater networks require specialized architectures, consider the standard Carrier Sense Multiple Access (CSMA) protocol used in terrestrial Wi-Fi. Before a node transmits, it "listens" to the medium. If it senses silence, it assumes the channel is clear and sends its packet.
Terrestrial (Radio):
Node A: |--Packet--|
Node B: |--Senses Clear--||--Packet--| (No Collision)
Underwater (Acoustic):
Node A: |--Packet--| ~ ~ ~ (Sound traveling slowly) ~ ~ ~
Node B: |--Senses Clear (A's sound hasn't arrived yet)--||--Packet--|
▲
COLLISION MID-WATER
In an underwater environment, if Node A begins transmitting a packet to Node B from a kilometer away, the acoustic wave will take nearly $0.67$ seconds just to arrive. If Node B listens to the channel $0.3$ seconds after Node A starts, it will hear absolute silence. Node B will conclude the channel is free and begin its own transmission. The two packets will inevitably collide mid-water, destroying both.
As Zhenghao Xi et al. demonstrate, this latency makes traditional carrier sensing highly unreliable. Nodes cannot immediately detect if the channel is occupied by another transmitter. Consequently, specialized Medium Access Control (MAC) protocols must be engineered from scratch. For example, the SFM-MAC protocol designed by the authors introduces deliberate, intelligent delays before responding to packets, ensuring that closer nodes do not inadvertently drown out transmissions coming from more distant parts of the network. Shiraz Shahabudeen et al. similarly emphasize that these physical constraints require protocols to explicitly calculate and budget for the long delays between signal transmission and reception to maintain network throughput.
Space-Time Coupling and Slotted Collisions
In terrestrial networks, packet collisions are purely a function of time: if two nodes transmit simultaneously, they collide. In underwater acoustic networks, collisions are governed by space-time coupling—they depend on both when a packet is sent and where the sender is located relative to the receiver.
This phenomenon complicates slotted protocols (like Slotted ALOHA), where time is divided into discrete intervals to prevent overlapping transmissions. Quansheng Guan et al. revealed that because sound travels so slowly, a packet transmitted during Slot 1 might still be traversing the water column when Slot 2 or Slot 3 begins, causing complex collision patterns that span multiple time slots at a distant receiver.
Timeline at Receiver:
Slot 1: [--- Packet from Near Node ---]
Slot 2: [--- Empty Slot? ---] ◄─── Packet from Far Node (sent during Slot 1) arrives late!
To counteract space-time coupling, modern underwater MAC protocols often decouple data transmissions from a single shared channel. Fatma Bouabdallah et al. addressed this in their MC-UWMAC protocol by utilizing a grid-based slot assignment on a dedicated control channel alongside a quorum system for data channel allocation. By scheduling transmissions across multiple channels based on spatial geometry, the authors achieved collision-free communication despite the underlying propagation delays.
Decision-Making with Stale Information
Propagation delay does not just disrupt the link layer; it also degrades the performance of routing algorithms and autonomous robotic control loops. When an underwater node receives an update about the state of the network or a neighboring vehicle, that information is already historical data.
This challenge is often framed around the Age of Information (AoI), a metric that quantifies the freshness of received data. Yimian Ding et al. tackled this problem by incorporating AoI directly into the decision-making framework of autonomous underwater vehicles (AUVs) using a Markov Decision Process (AoI-MDP). Instead of assuming that a received observation reflects the real-time state of the environment, the vehicle explicitly calculates how much the data has aged during its acoustic transit. The AUV can then choose optimal actions—such as choosing to "wait" for a fresh packet to arrive—rather than executing maneuvers based on stale, misleading updates.
Similarly, network routing protocols cannot rely on instantaneous topology updates to find the best path to the surface. Yuan Zhou et al. bypassed the need for high-latency global localization by developing an anypath routing protocol (QLFR) driven by Q-learning. Instead of maintaining a rigid map of the network, nodes dynamically learn the best local forwarding paths based on immediate metrics like node depth and residual energy, mitigating end-to-end latency through localized reinforcement learning.
Scaling to Swarms and Field Realities
The compounding effects of acoustic latency become particularly severe when scaling up to multi-agent systems, such as robotic swarms or distributed ocean observatories. Hejun Wu et al. explored this in their work on the TRiForm formation control scheme. They demonstrated that as the number of nodes in an underwater swarm increases, the propagation delays inherent to acoustic coordination hinder real-time synchronization. If a swarm attempts a rapid maneuver, the lag in position updates prevents nodes from maintaining precise formations, potentially leading to coordination failures.
These communication constraints turn the deep ocean into a highly data-isolated environment, a reality documented across various field applications:
- Target Tracking: Tom O’Reilly et al. deployed deep reinforcement learning on AUVs in Monterey Bay to track moving targets using only sparse acoustic range measurements, optimizing trajectories to compensate for frequent signal dropouts and latency.
- Digital Twins: Alexander Barbie et al. utilized digital twin prototypes in the Baltic Sea to model and synchronize heterogeneous robotic networks, using simulation to bridge the gap left by low-bandwidth, high-latency hydroacoustic channels.
- Seafloor Mapping: Even in direct vertical profiling, Weifeng Ding et al. showed that in deep-water environments exceeding 2000 meters, the significant travel time of acoustic signals introduces tracking confusion and data fragmentation over rapidly changing terrain, requiring post-processing fusion with bathymetric data to correct the profile.
Ultimately, whether designing a MAC protocol, a routing algorithm, or a robotic control loop, engineering for the underwater world requires abandoning the terrestrial assumption of instantaneous communication. Successful systems succeed not by trying to eliminate propagation delay, but by embedding an awareness of acoustic latency directly into every layer of the architecture.
Key Foundational Work
Development of Underwater Acoustic Modems and Networks (1993) — contributed 3 concept elements to this direction.
Corrections to “Analysis of a High-Performance MAC Protocol for Underwater Acoustic Networks” [S. Shahabudeen, M. Motani, M. Chitre, IEEE J. Ocean. Eng., [DOI: 10.1109/JOE.2013.2246741] (2014) — contributed 1 concept elements to this direction.