In the lobby of Astron, the Netherlands Foundation for Research in Astronomy, a clear plastic column almost full of dirt commemorates an unusual aspect of the institution's struggle to study the cosmos. Over the past three years, the institute's astronomers have been negotiating with 42 Dutch farmers to buy up the 400 hectares of land they need for the first stage of a remarkable new radio telescope — the low frequency array, or LOFAR, a €148-million (US$188-million) attempt to open up a new part of the spectrum to astronomical inspection. Every time a stretch of land was acquired, a bit of soil was added to the column in the lobby, making a layer cake of peaty soil and sand.


Out in the fields, the business end of LOFAR consists of sites dotted with low-slung, leggy wire antennas liberally spattered with the droppings of the birds that perch on them and a few grey metal Portaloo-like cabins. But the fields aren't the right place to look for LOFAR's grandeur: the telescope's impressive heart is being built in cyberspace. It is there that the faint noise of the Universe may eventually be separated from the background roar of Holland's air-waves and the Milky Way.

'Software telescopes', which couple a vast capacity for absorbing information with a matched capacity for analysing and reassembling it, are widely seen as the future of radio astronomy. Such technologies will be crucial to the Square Kilometre Array (SKA) with which the world's radio astronomers hope to revolutionize their field by the 2010s. At LOFAR, though, the technology faces an extra problem. To secure funding, the array's core had to be built in the Netherlands. Nowhere on Earth is completely free of radio signals, but the Netherlands, the most densely populated country in one of the most built-up regions of the world, is quite remarkably noisy.

To tune in to the early history of the Universe, LOFAR must studiously ignore TV shows, FM radio stations, air traffic control and local taxi drivers, not to mention purely accidental emissions. Heino Falcke, LOFAR's international project scientist, gestures to a tractor trundling past one of what will be 77 fields of antennas. “We would probably measure that as a series of sparks going past our telescope.” LOFAR's rivals, cloistered in a remote part of China and the Australian desert, enjoy far quieter environments. Whether LOFAR's software cleverness and number-crunching prowess will allow it to overcome the disadvantage remains to be seen.

Get close enough to the 'Portaloos' in the field and you can hear them hum. Unlock their doors and you find tangles of cable and racks of electronics and blinking lights. This is where the wire roots of the antennas end up, and where the software telescope begins to take shape.

Although the antennas may look simple, they can produce data at an alarming rate. One antenna site, consisting of 96 spindly antennas collecting low-frequency waves and 1,536 squat structures detecting higher frequencies, will produce about 500 gigabits of data every second — far more than even the most sophisticated optical telescopes.

Cyber optics

The humming Portaloos are there to make sure that most of the data are left in the field, reducing a flood to a mere torrent. But that torrent still outdoes the data rate for the detectors on the LHC, the vast particle collider currently being completed at the European Centre for Particle Physics (CERN) near Geneva — instruments whose vast data flows have led to a world-wide effort in GRID computing. The torrent from the fields, channelled through optical fibres, ends up in a windowless strip-lit room at the University of Groningen. Here, an IBM Blue Gene/L supercomputer called Stella compares the data from the 50,000 channels measured at each station with the equivalent data from all of the other 76.

At Astron, they think that this 34-trillion-operations-a-second wonder could restore them to the glories of the 1950s, when the 25-metre Dwingeloo dish, now rusting away behind their institute, was briefly the largest steerable radio telescope in the world. In the 1990s, the institute decided to follow a path that concentrated less on dishes and more on digital. “We thought that Moore's law would be a good thing to hook onto,” recalls Harvey Butcher, Astron's director since 1991. To maximize the effort that could be put into data processing meant that the cost of picking up the signals in the first place had to be minimized, leading to the use of cheap dipole antennas.

LOFAR's dipoles pick up long-wavelength signals at frequencies from 30 to 240 MHz. This long-wavelength radio is a little-explored part of the spectrum, not only because it is widely used on Earth, but also because it is affected by the ionosphere — another source of noise that has to be removed with the help of computers. What's more, the long-wavelength sky is filled with the synchrotron emission given off by electrons being whipped around by magnetic fields in the Galaxy, the glare from which masks intriguing cosmological signals some 10,000 times as faint.

In 2001, the Astron team calculated the computing requirements for a long-wavelength system big enough to find those faint signals. Then they had a look at the Top 500 list that ranks the world's quickest supercomputers. It looked as though, given the way that performance was improving over time, the power they needed might be purchasable in the near future. And so it was. The Blue Gene/L in Groningen is three times as powerful as the most powerful machine in the world was back then — and doesn't quite make it into the Top 500's top ten today.

Perhaps the most interesting of the feeble signals that the computer will be sifting out are those emitted in the dark ages of the first few hundred million years after the Big Bang, a chapter in the Universe's history that remains obstinately closed. The dark ages' darkness stems from the fact that the glowing plasma of the Big Bang, details of which are still visible in the cosmic microwave background, had cooled down into neutral atomic hydrogen. It was only after the stars and galaxies had come to life that their light burnt off this hydrogen mist, reionizing the gas into a transparent plasma.

I don't think you can do the measurement there. Jacqueline Hewitt, Kavli Center for Astrophysics and Space Research

Huge sums of money are being spent on various telescopes that promise to peer back at this epoch of reionization. The $650-million Atacama Large Millimetre Array, under construction in the Chilean Andes, aims to detect the infrared glow of hot dust around the very first galaxies, radiation that will have been stretched, or 'redshifted', by the expansion of the Universe to millimetre wavelengths. NASA's $5-billion James Webb Space Telescope, scheduled for launch in 2013, will look for the ultraviolet light of these first galaxies, now redshifted into the infrared wavelengths that the telescope is optimized for.

Shedding light

LOFAR's approach, shared by the Mileura Widefield Array (MWA), planned for the outback of western Australia, and the 21CMA installation in China, is to look at the neutral hydrogen itself. Neutral hydrogen emits radiation with a wavelength of 21 cm — hence the name of the Chinese array. That 21-cm radiation from the epoch of reionization is redshifted down into the metre-long wavelengths the new telescopes can detect.

What the arrays will actually be looking for are the holes in the 21-cm radiation caused as the reionization centred on stars and galaxies took hold, creating a Swiss-cheese effect. In principle, different stages in the growth of these holes could be teased out by finding the 21-cm hydrogen line at different redshifts — the longer the wavelength, the older the radiation. By moving up and down the frequency dial, it would be possible to watch these bubbles expanding and merging as the Universe opened itself up to inspection. Simulations of the early Universe produce beautiful pictures of this process.

A nine-step simulation shows 21-cm emissions being eaten away as reionization spreads through the young Universe. Credit: B. CIARDI

But images from this first generation of telescopes will be, at best, very fuzzy. The head of LOFAR's reionization project, Ger de Bruyn, demonstrates what might be seen by bringing up the theorist's beautiful simulations on his laptop and then removing his glasses, scooting his chair to the far side of his office, and squinting: “This is more likely.” The first detection of reionization is more likely to be statistical, giving a measurement of how the brightness of the glow varied from one time to another, than a resolvable picture of specific bubbles. And even a statistical measurement is going to be tricky. Xiang-Ping Wu, chief scientist for 21CMA, says that it could take five years to be confident of seeing a reionization signal.

With 21CMA already taking data, construction on LOFAR started this summer and the first antennas of Australia's MWA expected to be installed next year, it could, says Butcher, “be a really interesting race”. The Chinese array, built for just US$3 million, is the smallest and cheapest. It is located in the west of the country, in a remote valley in Xin Jiang province. When the project was approved in August 2004, “we pitched a tent in the Ulastai valley and started the construction on the same day”, says Wu. The completed array consists of 10,287 log-periodic dipoles — television antennas, more or less — arranged along two perpendicular arms roughly four and six kilometres long.

Unlike LOFAR and the WMA, which have other scientific fish to fry, 21CMA is essentially interested only in the reionization signal. Its field of view is restricted to a patch of sky around the North Celestial Pole and its resolution rather coarse. Although it has been collecting data for three months, no money has yet been allocated to continue operations. But Wu isn't worried. The Chinese Academy of Sciences and National Astronomical Observatories of China, which funded the array's construction, are likely to continue to support the project, he says, and concentrating on a small patch speeds up data acquisition.

Universal challenge

Close behind is LOFAR, which will use two types of antenna, one gathering low-frequency radiation, between 30 and 80 MHz, the other higher frequencies, between 110 and 240 MHz. Because of its geographical situation, LOFAR has a whole suite of systems to deal with radio interference. It will use software filters to flag and discard signals that are too strong to be coming from space. Some noise will be fleeting, and so average out over a long observing time, or be at a single frequency that can be filtered out. This will leave gaps in the data, but the team estimates that they can use about 90% of the spectrum at any one time. The exception is the commercial FM band, which LOFAR will simply discard. This is why there's a gap between the ranges of the two antennas. And that could pose a problem for the study of reionization.

Current models predict reionization would have been getting into its stride by the time the Universe was 500 million years old. If they're correct, “the frequency will be right in the middle of the LOFAR band. Then LOFAR can really have a go at it,” says de Bruyn. But if galaxies formed a bit more quickly than predicted, the hydrogen would have burnt up a few million years earlier. In that case, its redshifted signal will lie in the impenetrable realm of Exxact FM and Radio Veronica.

Three years ago this blank spot, and other concerns about the effects of noise, led to a split in the international collaboration of astronomers planning the telescope. The US and Australian members left after the Astron team and their Dutch collaborators committed themselves to a site in the Netherlands. Some of the Americans opted for a Long Wavelength Array in New Mexico, which is not pursuing the 21-cm signal from reionization, the others for the MWA in Australia. “We didn't agree with the decision to build in northern Europe. We didn't think it was wise,” says Jacqueline Hewitt, director of the Kavli Center for Astrophysics and Space Research at Massachusetts Institute of Technology in Cambridge, now part of the collaboration working on the MWA in Australia. “I don't personally think you can do the epoch of reionization measurement there — I might be wrong, but I wasn't prepared to spend the next ten years of my life trying.” The site chosen for the MWA — a 12-hour truck drive from Perth, the nearest big city — is beset by very little interference.

But the Dutch part of the collaboration was tied to building their telescope in the Netherlands by a generous dollop of funding. LOFAR was granted €52 million in 2003 from the national Fund for Economic Structure, which reinvests money made from sales of the Netherlands' natural gas. This money has to be used to strengthen the national economy, and so couldn't be spent on building a telescope in another country. The Dutch contingent wasn't going to walk away from this kind of cash (see 'Making it pay').

Space race

The MWA sees itself more as an experiment — and as a forerunner to the SKA — than as a dedicated piece of infrastructure for many users, which is what LOFAR aspires to be. It has fewer antennas, of only one type, all in one place, and a much smaller project team relying on postdocs and graduate students for a lot of labour. And there are far fewer farmers to negotiate with. This makes it much cheaper than LOFAR, with an expected initial cost of just US$10 million — although Rachel Webster, a member of at the MWA team at the University of Melbourne, says she expects the team will actually spend much more. The initial money is being provided by the US National Science Foundation, the Australian Research Council and the government of Western Australia, among others, but the project is still at the prototype stage. The first 32 of its 500 antenna elements are due to be installed in the outback in 2007.

Hewitt notes that, regardless of who sees the reionization signal first, it will be good for the measurement to be repeated. “We're going to need to compare our results. If one group sees it, I don't think people will believe it until another group has too.” Butcher, while regretting the break-up of the project team, is matter-of-fact about its consequences. “On the one hand, there are some bits of the spectrum that we can't use. On the other hand, we have a telescope and they don't.”

LOFAR and the MWA will use their all-sky views to study more than just the epoch of reionization. “The epoch of reionization is the lighthouse. It is the project that most people are interested in,” says Falcke. But the fact that they will be looking through a new window of the spectrum offers a prospect that he sees as just as tantalizing. “There's at least a chance that there is something out there that we haven't found yet.”

We have a telescope and they don't. Harvey Butcher, Astron

Estimates based on the number of objects detected in surveys at higher frequencies suggest that LOFAR's surveys will turn up some 100 million radio-emitting objects. The view at the lowest frequencies should include bright radio galaxies more distant than any seen before. The data stored in LOFAR's buffers will let it zoom in on transient flashes in the radio sky, which could be the radio counterparts of poorly understood γ-ray bursts, or even bursts of radio emission from extrasolar planets, like those that come from Jupiter. Searches for transient events will 'piggyback' on all the other projects, commandeering the telescope if something particularly spectacular shows up.

Long wavelengths also have implications for the search for extraterrestrial intelligence — a topic that stirred up significant interest among the farmers selling Astron their land. Working at wavelengths where the earthly background noise makes things difficult becomes a plus when looking for radiation leaking out from civilizations around other stars — at least it does if they have FM radio stations and the like1.

Network building

Closer to home, flashes of lightning and radio showers from cosmic rays will also be picked up by the array. Falcke's pet project looks at both of these, testing theoretical suggestions that cosmic rays might sometimes trigger lightning. A prototype LOFAR station in Germany, called LOPES, has already managed to image cosmic-ray showers with a time resolution of 30 nanoseconds2.

The sharp resolution needed to study individual sources such as galaxies and even planets makes the LOFAR team keen to spread their antennas beyond Holland's borders; whereas the core set of elements doing the reionization studies will be clumped near the Dutch town of Exloo (see map), other antenna elements could be hundreds of kilometres away. Universities in Germany have expressed interest in at least six stations, and a UK group is interested in three. There are also discussions going on with astronomers in Sweden, France and Italy. Falcke hopes that, as the project gets under way, a me-too effect will lead to interest snowballing and the array spreading further, expanding from the initial 77 stations to more than 100.

Antennas of the the Dutch LOFAR telescope will be centred on the town of Exloo.

Long baselines, though, exacerabate a problem of calibration. Incoming radio waves are distorted on their way through Earth's ionosphere. LOFAR will have to correct the signal from every station separately, using the results from a model of the ionosphere running on a Linux cluster. This model will be updated every ten seconds by comparing the apparent positions of 200 sources on the sky with their known positions, with the results being fed into the Blue Gene/L. There are now so many cables plugging into the computer's vitals that the doors of its case barely close.

Nor do the challenges end with the inputs. While the various filters and calculations compress the immense amount of data the system gathers, the Blue Gene/L will still be outputting correlated data at an astonishing 39 gigabits per second. That can't be stored for long in its unprocessed form, and the team plans to keep the data for only a week or so. After that, the data will only be available in the form of star maps from which quite a lot will have been lost. Astronomers who want to preserve the primary correlated data will have to take them away while they are still fresh, either through high-capacity fibre or saved to high-density storage media that can be shipped.

And LOFAR is relatively simple compared to the demands of the SKA, with its square kilometre of collecting area and far greater frequency range. When Kjeld van der Schaaf, who is overseeing development of LOFAR's computer systems, makes presentations on the SKA, he includes a slide in which the trends of the Top 500 computer list are extrapolated through to 2015. The SKA's correlating system, which needs to be 3,000 times faster than LOFAR's Blue Gene/L, sits at the top of this ranking.

While the participants in the SKA wait for computing power to catch up with them, design their technology, plead for their funding and choose their site (it will be a radio-quiet spot in South Africa or Australia), LOFAR will forge ahead with its ground-breaking work. And although winning the race to detect the 21-cm signal from reionization is not the project's be-all and end-all, it does have a certain sentimental significance. The very existence of the 21-cm hydrogen line was first predicted by a Dutch astronomer in 1944, but US astronomers beat his compatriots in the race to find it. This time round, they want to get there first.