How does a radio wave become a picture?

Planets, stars and nebula’s all emit radio waves, which are a form of invisible light waves. They are detectable with radio telescopes such as the dish telescopes at Westerbork and radio telescope LOFAR. But how do the researchers at ASTRON construct an image out of those invisible waves?

By Mischa Brendel

Radio waves

All matter emits radio waves, but some matter does this at specific wavelengths. Hydrogen, for example, emits radio waves at 1420 MHz. Hydrogen also happens to be one of the most important building blocks in the universe. Stars are born from it. Therefore, measuring the amount of hydrogen in the universe teaches us a great deal about the formation of the universe. This makes radio waves one of the most invaluable resources for understanding our universe.

Radio waves are detectable with radio antennas. This can be done with the traditional dish antennas, like those at Westerbork. The 14 Westerbork parabolic dishes date back to the seventies, but are not as old as they seem: throughout the years the radio telescope has received several updates, of which the most recent in 2018. Each dish antenna carries at its center a receiver at which all the radio waves caught in the dish converge.

Westerbork Synthesis Radio Telescope (WSRT).
(Credit: ASTRON)

Radio waves can also be detected with telescopes like LOFAR, which operates in a different way. This radio telescope basically is a group of many telescopes spread out over several countries, in which each ‘telescope’ is a field full of radio antennas. The dish telescopes at Westerbork look at a single point in the universe simultaneously, a field full of LOFAR-antennas can look at multiple directions simultaneously.

With LOFAR each antenna is connected to its own receiver. Since the radio waves that are caught are very weak, radio telescopes have amplifiers to increase the signal strength. The signals measured are noise-like. However, the amplifiers themselves also add noise to the signal. So how to measure the proper signal and ignore the rest? This is where complicated calculations and computations come in, explains Dr Ir Albert-Jan Boonstra, programme manager technical research at ASTRON. “The signal coming from space has a specific characteristic, while each radio telescope causes its own, unique noise signal. So, the signal which each telescope receives that is identical to the signal received by other telescopes must be the signal from space. We filter out the other noise-like signals.”

Low Frequency Array (LOFAR), with the LOFAR core on the left and antennes for the low frequencies on the right. (Credit: ASTRON)

The radio waves also need to arrive at the receiver simultaneously. All time differences must be corrected for, because in order to receive a strong enough signal so that an object can be properly charted, the received radio waves need to be coherently added up. Again, computational power is key.

By constantly taking measurements the telescopes produce tremendous amounts of data: for LOFAR this adds up to petabytes of data per day. Boonstra: “Ideally we’d store and process all of those data, but we cannot afford to do so. Therefore, we add up the signals from the antennas on a single field and only send through those accumulated data. This reduces the amount of data a hundredfold.”

A central system then correlates the data, after which incorrect data are removed or corrected. Subsequently the data are calibrated. “Then it is up to the astronomers to make proper astronomic charts out of these data,” says Boonstra.

A central system then calibrates the data: incorrect data are removed and corrected. Subsequently the data are correlated, which results in a correlation matrix. “Then it is up to the astronomers to make proper astronomic charts out of these data,” says Boonstra.

A lot of expertise comes into developing these radio telescopes. Not only in the field of astronomy, but also in the fields of antennas, electronics, algorithms, and scientific computing. ASTRON cannot outsource the development of all these systems, many are too specific and push the boundaries of current technologies. Therefore, it develops these systems in-house, but in close collaboration with other knowledge institutes and industry. And that makes ASTRON much more than just a research institute.

Compact receivers

“A receiver picks up the radio signal through an antenna. The (very weak) signal is amplified and is then digitized,” prof. dr. Mark Bentum explains. Bentum is one of the researchers of the radio group at ASTRON working on the development of compact receivers.

Westerbork Synthesis Radio Telescope, one of the dishes viewed from above. (Credit: ASTRON)

In radio telescopes, receivers consist of an antenna and several electronic components to transform the electrical signal. In a compact receiver these various components are integrated. But why make the effort to put all those separate components together in one single part? There are several reasons for that, Bentum and his colleague André Gunst, system architect of the digital & signal producing group, explain.

First and foremost is that with a compact receiver far less of the captured signal is lost and that more data can be collected. Gunst: “Currently we capture a radio signal and send this through a piece of coax cable of a few hundred meters before we digitize it, which is not without risk. For instance, a difference in temperature between several of these cables can cause a variation in how quickly it transports the signal.” The researchers want to receive the various signals of all antennae at exactly the same time to make certain that the data they receive from each individual antenna is from the exact same moment and location of looking into space. By digitizing the signal close to the antenna and putting a time stamp on it there, the signal is not sensitive for delay variations and loss anymore.

A compact receiver. (Credit: ASTRON)

Gunst: “Digitizing the signal much closer to the antenna is exactly what we plan to do, but this is also our biggest challenge. This is because we need to boost the signals received by the antenna greatly before we can digitize them. That process creates noise which interferes with the measurements. Because of this issue the housing for the compact receivers needs to be designed in such a way that this interference is limited.” A second challenge to overcome is to distribute a clock signal to the antennas for the exact timing.

All in all this sounds like a rather costly affair, but that is not so. In fact, cost reduction is another important reason for the development of the compact receiver. Bentum: “When we have developed the integrated compact receiver, we plan to have a chip manufacturer to actually build the chips. The first one will be very costly, but the second one less so, and the third one will be even cheaper, and so forth. And these compact systems will function better and will be more stable than with separate components.” Another advantage: when a malfunction occurs, a compact receiver can be easily replaced, whereas a separate part in a larger receiver system needs to be disconnected, replaced and then reconnected to all the other parts.

Doing maintenance on LOFAR. (Credit: ASTRON)

Several parts needed to construct a compact receiver are already available on the market. Gunst: “Analog to digital converters are already for sale for instance.” The compact receiver will eventually be available to the mass market as well, although both researchers suspect the system won’t cause any instantaneous huge impacts for the average consumer. Bentum: “But the technologies we develop here can have a major impact in the daily lives of people. WiFi for example was developed because we needed algorithms for the quick transportation of data.”

So far, Bentum and Gunst don’t foresee any problems with the development of the compact receivers and the two gentlemen expect that ASTRON – despite numerous challenges that still need to be overcome in its development – will be presenting the compact receiver in a couple of years.

Accelerators

A radio telescope like LOFAR consists out of several fields filled with antennae, roughly a hundred thousand in total. That means one hundred thousand data streams that need to be processed in parallel. As such, LOFAR produces terabytes of data per second.

To process all those data, ASTRON uses supercomputers. But not the well-known huge computers which occupy half the room. No, it uses clusters of server machines that are interconnected, and equipped with accelerators. They are called accelerators because they accelerate the calculations performed on the data. The data need to be correlated and for that GPUs, Graphics Processor Units, are used. The general public knows GPUs as graphics cards, used in computers and gaming consoles. “CPUs, Central Processor Units, better known as the computer processing core, are designed to perform all sorts of tasks fairly quickly. A GPU is designed to perform a very specific task very quickly and to do that in parallel with other GPUs,” says John Romein, system researcher at ASTRON.

But why doesn’t ASTRON develop its own system for such a specific task? Romein: “The development of GPUs is driven by a vast market. Therefore, several manufacturers have been able to invest heavily into the development of GPUs. That is a development we cannot hope to match, but we really don’t need to.” There are two kinds of GPUs available on the market: the cheaper GPUs meant for the consumer market, and the GPUs developed especially for data centers. Those are considerably less cheap.

Energy efficient

Near the telescopes in the field there are special boxes with in them self-built electronics. In those boxes the data of the antennae converge and are subsequently sent along to the GPU server machines. “Next, those data are correlated which, given de amount of data and the speed with which it is collected, requires a lot of computational power,” Romein says. A correlation is a statistical connection between – in this case – separately measured radio waves. Although the calculation of a single correlation doesn’t require that much computational power, it becomes a different matter when there is a constant stream of data coming in. That is where the computational power of GPUs proves itself. But there is also a second reason why ASTRON has chosen to use GPUs, Romein says. “Energy efficiency. The GPUs work more efficiently en use less energy. And that benefits the environment as well as our wallet.

“A recent expansion on GPUs is a special piece of hardware that is able to do matrix multiplications,” Romein says. Matrix multiplication is also a mathematical operation which requires the necessary computational power. Romein continues: “The funny thing is that this piece of hardware was developed for an entirely different purpose, namely deep learning. But it turns out that it can also perform correlations very well, leading to a 5 to 20-fold reduction in energy use.”

When the data have been correlated, they need to be turned into an image. That process also utilizes GPU’s. More on that, in the next part of this series.

Calibration and imaging

The data of a telescope are not only formed by the sky brightness but are distorted in many ways. To ensure that the data that a radio telescope delivers are reliable, it is crucial that the antennae that collect the data are properly calibrated. The calibration encompasses removing unwanted data (noise and erroneous data) and correcting for distortions in the data where possible.

Known sources of noise can be corrected for in the data in advance. “From certain radio stations and wind turbines in the surroundings of the radio telescopes we know exactly which disturbance in our data they cause”, says Stefan Wijnholds, who is a senior researcher R&D at ASTRON.

The same goes for imperfections which are already known. Every antenna in LOFAR station has its own coaxial cable through which the antenna sends the data it collects. All these cables converge in an electronics box, but the cables vary in length, meaning that not all data measured at the exact same moment arrive at the electronics box at the exact same moment. However, since the exact length of all cables is known, the ASTRON scientists can correct for these variations.

But there is also noise which cannot be predicted in advance. However, even that noise can be corrected for in the data quite successfully, Maaijke Mevius explains. “You have to make your filtering techniques smart.” Mevius is a system researcher R&D at ASTRON. “We have a fairly detailed idea of what the signal from the universe looks like; everything that deviates from that, has to be corrected for.” Although you cannot filter away that noise too rigorously, as you may then filter out inconsistencies that you actually want to measure.

But even with all these steps, there is another matter to consider.

Ionospheric effects

One of the harder things to correct for are the ionospheric effects. Without getting into too much detail: radio waves slow down when they encounter the ionospheric layer, which surrounds our planet. The problem is that this layer varies in thickness which makes it impossible to predict how much distortion it causes on the data. Mevius: “There is no other way than actually looking closely at your own data.” By looking at the same piece of the universe at the exact same moment from different locations, you can measure the thickness of the ionospheric layer, since you are actually looking at one single point in the sky. Mevius: “A side effect is that, in this way, you can actually make a detailed map of the ionosphere.” Radio astronomers therefore share much of their data with ionospheric physicists.

How accurate does the calibration need to be? “To within a nanosecond,” Wijnholds says. That seems a bit extreme, but when you realize that the individual antennae stare at the same point in the universe into the infinite and measure the differences in arrival times of radio waves moving at 300,000 km/s, it becomes clear that this extreme accuracy is necessary: otherwise you run the risk that each telescope is looking at a different moment, producing a distorted visual of the universe.

Putting it all together

The electronics box sends its collected data to a computer center, where they are correlated: separate measurements are put together. And then these data, which are the correlations, are converted into an actual image. Through a set of mathematical formulas, the data of in total 6 to 12 hours of measurements are integrated and converted to a visual.

Essentially, a radio signal is measured at two different stations (radio interferometry). There is a minor difference in the arrival time of the signal between the two stations. This difference depends on the location of the radio source, whereas the amplitude of the signal is related to its brightness. By combining these complex data of many station pairs, the locations and brightness of many sources in the sky can be reconstructed into an image. In the case of radio telescope LOFAR, a measuring station consists of a field of antennae.

Wijnholds: “Every measuring point is the sum of what was registered at that moment by a pair of stations (antennae groups). From that sum, we then need to determine where the measured sources, which sent the radio waves, were at that time and how bright they were. A stronger source is registered as a higher correlation.” This shows why it is so important to correct for the noise from the measurements in advance. But if you do this properly, as described above, you finally end up with a very detailed map of the universe. Well, a very tiny bit of the universe.

This animation also explains the basics of converting radio signals into an image: