Today, after five years of work, the network is launching a little more slowly than expected. The Juan de Fuca plate straddles the Canadian-U.S. border, with a smaller part in Canada. So U.S. partners were to bear the brunt of the cost: 70 percent. But the U.S. funding has stalled, forcing NEPTUNE's proponents to start small.

Instead of lighting up the entire Juan de Fuca plate, NEPTUNE will at first cover just its northern third. What was originally planned as a 3000-km network touching 30 sites of interest will launch as an 800-km loop with instruments at two or three sites along it. Yet even in its truncated state, NEPTUNE poses a daunting technological challenge. "There's significant innovation in almost every facet," says Peter Phibbs, associate director, engineering and operations, for the Canadian team. "Making a communications system work in 3000 meters of water is not trivial." The power engineers, adds Phibbs, had to start from scratch.

Recognizing the complexity and ambition of the task—particularly in light of tight budgets—the NEPTUNE collaborators hatched two preliminary projects, called MARS and VENUS, to test the parts before tackling the whole. MARS stands for the Monterey Accelerated Research System and VENUS for the Victoria Experimental Network Under the Sea. The real action begins this month as these preliminary projects go into the water.

In a sleepy inlet off the southeast shore of Vancouver Island, VENUS engineers from the University of Victoria will test the interface between the power and data infrastructure planned for NEPTUNE and a cluster of scientific instruments—hydrophones, seismometers, and the like. Meanwhile, MARS engineers from the Monterey Bay Aquarium Research Institute, in Moss Landing, Calif., will subject the NEPTUNE infrastructure to crushing pressures in an 800-meter-deep canyon west of the aquarium [see photo, "Coming In for a Landing"].

If these smaller projects work, NEPTUNE itself could hit the water in 2007, setting oceanography on a new course. Scientists around the world are watching its progress and planning their own cabled observatories. Japan's proposed ARENA observatory, for one, is of similar scale to NEPTUNE, and Europe's ESONET would outline the entire continent.

While NEPTUNE is not the first attempt to build a cabled undersea observatory, it is by far the biggest and would deliver the infrastructure and engineering knowledge needed to make such observatories routine. If NEPTUNE fails, it would knock the wind out of the cabled observatory movement, and ocean science could remain stuck in its half-blind state for many years to come.

MARS is a trial run of a science node carrying NEPTUNE's full power and bandwidth in deep water. If all goes according to plan, this month engineers will sink the node and the electric and fiber-optic cable that will link it to shore, turn it on, and hope for the best.

The node is a waterproof titanium vault protected by a truncated pyramid of steel the size of a small car. Packed within the titanium is communications equipment to translate optical signals to electronic ones, route the signals, and then send them out again on copper cables to nearby instruments or via laser along optical fibers to other undersea nodes [see illustration, "Inside a Node"].

But MARS's main purpose is to demonstrate power conversion at the high voltage needed for NEPTUNE. It's not uncommon to transport power over long distances as a very high dc voltage to limit current losses. What is uncommon is to convert that power to lower voltage for use under nearly a kilometer's depth of water. Like NEPTUNE nodes, MARS must step down 10 kilovolts of dc power on the cable from the shore to a more useful 400 volts.

Designing and building a converter to do this fell to Harold Kirkham, a principal engineer at JPL's Center for In-Situ Exploration and Sample Return, and Vatché Vorpérian, JPL senior engineer and an IEEE Fellow [see photo, "Masters of MARS"].

The design constraints were tight. The MARS and NEPTUNE converters must handle 2 to 5 kW each—about what it would take to power a dozen big-screen plasma TVs—and fit in a meter-long pressure tube. JPL's design keeps the converter to a manageable size by stringing together 48 small high-frequency converter modules rather than using one or a few large low-frequency converters.

The modules operate on the same principle as the switched-mode power supply you might find in any modern piece of consumer electronics. Within each module, high-frequency transistors switch the dc to ac, a small transformer steps the voltage down, and a rectifier circuit turns the ac back to dc. Each of the 48 modules performs a 200- to 50-V conversion, and the outputs are tied together in groups of eight to deliver 400 V.

"You bolt together the outputs, so they're forced to be the same voltage, but unfortunately that doesn't mean that the inputs have to be the same," says Kirkham. And the inputs are the real danger: the slightest loss of coordination among the high-frequency switches can expose one module to the entire 10-kV input, which would instantly fry it. Vorpérian found a solution that prevents such an undersea barbecue by using a single controller reading a single output signal to coordinate all 48 modules.

Getting repairs done in 1000 meters of water is not easy—when it is possible at all. So the converter's reliability was a top priority. The converter is designed to be able to lose any one of its 48 modules without failing. And the node carries within its titanium pressure case two complete converters; if one shuts down, the backup automatically kicks in. "We think the failure rate from a converter should be extraordinarily low," says Kirkham. "It might not even be measurable."

That is, once it's ready to launch. As late as this past September, JPL engineers were still troubleshooting start-up problems plaguing the power converter, and Kirkham was grumbling over MARS's oceanography-scale budget, which precluded the full degree of quality engineering and fault analysis that would be lavished on space-bound equipment.

MARS's plan is to start with a technology demonstration and add science instruments to its experimental node in the years to come. In contrast, the other NEPTUNE test bed, VENUS, led by the University of Victoria's Tunnicliffe, will do lots of science right from the start [see photos, "Victorians"]. The point of VENUS is to demonstrate a system for sharing 400 V of dc power and high-speed communications among a bevy of underwater instruments. The number of instruments that can be integrated will determine how much science each of NEPTUNE's nodes will be able to deliver.

VENUS is basically a smaller version of the MARS node that does only data conversion rather than both data and power. It turns electronic data from the science instruments into optical signals for transport to the shore and routes commands coming down from shore to the right instruments. Tunnicliffe's team plans to install VENUS this month in 80 meters of water in the Saanich Inlet, 1.5 kilometers offshore of the southernmost part of Vancouver Island. Beginning this coming winter, a cable carrying an optical fiber and a 400-V dc line will power 15 instruments attached to the VENUS node. A second node planned for the nearby Strait of Georgia should interface with more than 40 instruments by the fall of 2006.

Most of VENUS's complement of scientific equipment is adapted from standard oceanographic fare. However, the project is also financing one novel device specially built to exploit VENUS's high-bandwidth two-way communications: a remarkably versatile array of hydrophones designed to triangulate and track a much broader range of sounds than any previous instrument. VENUS's array will pinpoint sounds ranging from the low grumble of container ships (thought to disturb the local killer whale population) to the high-frequency sound of wind, rain, lightning, and the chirp of herring passing gas.

The VENUS designers initially envisioned wet-mating most instruments directly to the nodes—that is, plugging them in underwater. But that scheme looked unrealistic once they discovered that the required connectors cost roughly US $8500 per pair.

Their solution is a custom-built interface that will sit between the node and the instruments: the Science Instrument Interface Module, or SIIM. Several SIIMs can be wet-mated to a node, and they act as platforms that can serve 10 or more instruments hard-wired to them. So instead of needing one wet-mate connector per instrument, NEPTUNE and VENUS may need only one for every 10. Importantly, converting the node's 400 V to the instruments' 24 V is performed in the SIIM rather than the node, reducing complexity and heat in the node and making it more reliable.