Friday, April 24, 2009

Underwater sensor networks

T-Lohi: A New Class of MAC Protocols for Underwater Acoustic Sensor Networks
A. Syed, W. Ye, J. Heidemann

Data Collection, Storage, and Retrieval with an Underwater Sensor Network
I. Vasilescu, K. Kotay, D. Rus
M. Dunbabin, P. Corke

We wrapped up the semester with two papers on underwater sensor networks. As you would imagine, being underwater poses a number of new challenges in the operation of the network. Radio signals don't propagate very far, so optical or acoustic systems instead must be used for communication.

We started with a discussion the data collection paper. This paper presents an underwater sensor network with fixed sensor nodes resting on the bottom of a body of water, while rechargeable mobile "data mules" periodically come by, collecting data from each node. To collect the data, the nodes communicate via line of sight using timed pulses from an LED on one end, and a photo diode on the other end. This requires the mobile node to get fairly close to the stationary node and stay relatively stationary, which could be very difficult with strong currents. Despite some micro-benchmarks performed in the Charles river, the class had its doubts that the system would not work robustly under murky ocean conditions. We talked about how this system could be made to be more robust. Maybe by slowing down the pulses of light, the system would have more time to detect a weak signal, but this comes at the cost of burning more energy on the fixed nodes. Some of the class was disappointed that the authors didn't do more with trying to encode the signals more robustly.

The other way the nodes can communicate with the data mule is acoustically. The nodes were equipped inexpensive audio transducers. But aside from one simple benchmark, it wasn't really used in their system. That was one of the frustrating thing about this paper. The introduction has this vision of many many nodes scattered across the ocean bottom. What they did was far more modest. They were able to collect data in a large swimming pool. We were also left wanting more of an evaluation. For example, how long did it take to collect data from the nodes? Throughput? Latency? In general there is a fine line between explaining your vision of the system vs. the risk of overselling your work and loosing the trust of the reader.

Someone commented that the paper read more like a technical report then a research paper. Lots of mundane details about the system. We liked that they were able to build the system out as much as they did, and the paper does a good job of motivating why this approach makes the most sense in terms of energy, but some of the class felt the cost of moving the mobile node should have been mentioned. Overall, it was pretty impressive that they got things working as well as they did.

The T-Lohi paper focuses on MAC protocols for UWSNs. They use acoustic modems, which have 5 orders of magnitude greater latencies than radio and cost 100 times more to transmit than to receive. T-Lohi uses a short reservation tones and a contention round (CR) before transmitting. The tones are short enough that nodes can often hear other reservation tones. This can give the nodes a count of how many other nodes are wishing to transmit and the backoff timer can be set accordingly for the next CR. When there is only contender in the CR, that node begins transmitting.

They compare 3 variants of T-Lohi: First, synchronized T-Lohi (ST-Lohi) were all nodes are time synced and time is divided up into slots and nodes only transmit at the start of a slot of a CR. Next was cUT-Lohi which is unsynchronized. Every node transmits when it wants, but must listen for twice the propagation delay plus twice the tone length. No collisions are possible. Teh last protocol was aUT-Lohi. Same as above, but only wait the propagation time plus the tone time, but with a chance of collision.

We talked about the problem they were trying to solve, routing packets back to the base station. They were only sending 8 packets per second, in the worst case. That's a pretty low data rate. Is it worth all this extra overhead of the contention rounds? The point of the CR is to reduce collisions. But collisions aren't the end of the world. It's probably the case that since transmitting costs so much, collisions really do hurt you and should be avoided at all costs, but they could have spelled this out more clearly. It would have been great to compare to a strawman of where there was no protocol.

Their use of "optimal" in the graphs was confusing some of us. It seems like a practical upper bounds. It's T-Lohi optimal. Then the discussion turned to whether CSMA schemes are the right way to go with these networks. TDMA would make more sense, but the time synchronization overhead might not make it worthwhile. What about more of a z-mac hybrid approach? Would that be better?