My Research
I am conducting research in Underwater Acoustic Networks (UANs) for Autonomous Underwater Vehicles (AUVs), part of the Computer Network Research Group at UNH. My advisor is Dr. Radim Bartoš.A Little Background
Trying to form wireless networks underwater is by no means a new concept. Back in the 1980s, folks at the Marine Systems Engineering Laboratory (now AUSI) were using CB radios on underwater vehicles to send data. But trying to transmit radio signals in a conductor (i.e., water) is difficult, and only works over very short distances. Optical links are not much better.
On the other hand, acoustic signals travel great distances underwater (this was the basis
for SOSUS). So, for untethered submersible
vehicles or sensors, acoustic modems dominate the underwater networking field. However, they suffer
from noisy and unreliable channels, asymmetric links (I can hear you, you can't hear me), and
unpredictable "shadow zones" (locations in which signal reception is nearly impossible due to echoes).
Perhaps the greatest barrier to achieving this mission is that, while each of these illustrated
communication links exist in practice, each uses a different networking model.
- Gliders usually communicate directly with a "Dock Server" over an RF Iridium or Freewave modem while on the surface.
- Sensor networks usually employ some form of shortest-path or lowest-energy routing protocol, since they are stationary.
- Multiple Cooperating AUVs, such as the SAUV, move around too much for simple routing to converge, but still need to communicate frequently while underwater. (These are the vehicles I work with — we use a simple flooding protocol called COFSNET right now, which performs surprisingly well in this environment.)
The development of this model is the goal of my work.
Why is it Difficult?
To understand the difficulty of creating networking protocols that will work in this domain, it is important to recognize the impact of loss and latency on the more traditional approach to networking (or even the more modern approach to mobile ad-hoc wireless networking).Latency
This graph illustrates the effects of latency (our AUVs typically operate between 1000m and 2000m apart), on RTS/CTS; random access is unaffected.
Loss
This graph illustrates the effects of loss on Dynamic Source Routing (DSR), LASR (Location-Aware Source Routing) and simple flooding. DSR is far more efficient than flooding — if there is no loss. However, small amounts of packet loss effectively cripple this protocol. LASR performs slightly better, but notice what happens when packet loss approaches 20%: it too becomes less efficient than flooding.
20% packet loss is not unheard of — bear in mind that turbulent currents, high wind, shallow water, thermoclines, haloclines, and mechanical noise can all cause packet losses. To be considered useful, protocols must degrade gracefully under these conditions.
My Efforts
In a multi-hop network, the saturation point is dependent mostly on the node density. However, in a single hop network it is the MAC protocol that ultimately indicates whether the network is saturated. Each protocol has different tolerances for the number of nodes and the latency between them, dividing the region slightly differently.
Although the exact shape of the curves shown here will depend on the traffic characteristics and protocol implementation, we can say in general that no single MAC protocol is best in all cases.
- TDMA is limited by acoustic range and number of nodes in the cluster
- RTS/CTS is limited by the number of exchanges multiplied by the distance between the nodes, but may be more efficient than TDMA over short distances if not all nodes need to send data
- Random access is not affected by the limits of acoustic range, although it may be difficult to implicitly determine who a node's neighbors are
For a detailed explanation of how this works, download my UUST my paper entitled "A Delay-Tolerant Networking
Framework for Mobile Underwater Acoustic Networks," available from this page.
| Acceptability of Loss | |||
|---|---|---|---|
| Low | High | ||
| Expiration | Short | UDP | |
| Long | TCP | ||
As an example of the "short expiration, low acceptability of loss" category, consider a status message that is broadcast from a node every 60 seconds. Because of the inherent value of the message to the neighboring nodes, it might be worth waiting several tens of seconds for a moment where measured environmental conditions suggest that the message will be received successfully. However, there will be a newer, more accurate status packet available after 60 seconds — so the first packet should now be discarded if it has not yet been sent.