The seawater itself takes on complex characteristics; temperature, salinity and density differentials create refractive surfaces and subduction zones, which serve to confuse sonar by generating reflections and distortions.
A project at NERC has developed a new side-scan sonar fish, called GLORIA. Looking like a 30' yellow hot-dog, this is a wide-aperture acoustic array that is towed behind a ship and sends signals down to bounce off the seafloor many miles below. Receivers on the ship pick up the echos and can calculate the reflectivity of the seafloor from the intensity of the echo. Recently, the phase information of the signal has been captured, since from calculating the phase angle difference from echos the topography of the undersea landscape can be directly measured rather than inferred from reflectivity profiles. However, subduction and refraction, coupled with surface reflection and ship-created noise, means that the signal received is very weak and hugely distorted.
We are collaborating with NERC on this project, and the work has four phases. We first have to develop a real-time system that accepts the 64k/s that the PC dumps out. This signal represents the received signal having been processed by dedicated electronics and a DSP chip. This has to be sent via ethernet onto a multiprocessor Sun system. We need fault-tolerant, fail-safe communications protocols, since the data costs in excess of $10,000 per day to collect.
The next phase involved devising algorithms to reduce the noise in the signal, reimplementing FORTRAN algorithms in C, and doing the mathematics necessary to derive the topology from the phase angle information. Having done this, we then have to process the raw data, including correcting it for ship speed variation, correcting the side-to-side motion that the fish follows, and normalising the data to fit into conventional lat./long. coordinate systems (using a Mercator-derived projection). Once this has been done we provide an image display interface that presents this data in a sensible manner, including full image manipulation and false colouring functions. This interface has to be designed to be used by computer-illiterate users, and there are a number of complex interface issues to be decided.
The final phase of the work is still under development, and that is the implementation of adaptive signal processing filters, based on neural networks, which learn to remove the noise from the signal. The difficulty in such an approach is that non-linear signal processing is inherently difficult, even with good signal-to-noise ratios, whilst we have real-time data with a terrible s.n.r.
I was in the S.E. Pacific for approximately 3 1/2 months, from Feb. to mid-May. Routine is approximately 16 hours on, 8 off, 7 days a week, for 2 shifts of 40 days at a time, working a rota of nights since the data collection is continuous. And the ship is `dry' (no alcohol at all)! The project is run by NERC in conjunction with the Scripps Institute of Oceanography and represents a cutting-edge research project that has great oceanographic impact.
It wasn't all work.....
Return to main map