All infrared astronomy is dependent to a greater or lesser degree on the atmosphere. It also depends greatly on the amount of infrared radiation that comes from other sources than the objects that we wish to study. Everything that is warm emits infrared radiation: the sky, the telescope, its mirrors, etc.
 

The atmosphere and its influence

Infrared astronomy is carried out in a series of windows in the atmosphere that allow us to see out into space. In between, the atmosphere is largely or totally opaque. Infrared radiation is absorbed by various atmospheric components, in particular, water vapour, carbon dioxide, ozone and methane. The very same gases that make life on our planet possible by causing a natural greenhouse effect block the radiation that we wish to detect from space. The amount of carbon dioxide, ozone and methane are quite stable all over the Earth and from night, but the water vapour is extremely variable, both from place to place and from moment to moment.

CanariCam will be mounted on the 10-m Gran Telescopio Canarias (GTC) in the Roque de los Muchachos Observatory in La Palma (Canary Islands, Spain). This is a medium altitude site (about 2300-m), but its latitude and the fact that the island of La Palma is extremely steep (La Palma is a small island) means that it is as dry or drier than many much higher observatories.

In the near infrared the H and K windows (1.65 and 2.2 microns respectively) are broad and clean. The J window is somewhat narrower and affected by water, but still easy to use even from sea level. Other atmospheric windows though are neither clean, nor stable.

CanariCam will work in the 10 and 20-micron windows. If we look at the transmission of the atmosphere over this wavelength range we see how important the atmosphere is. The plots shown here are calculated for Mauna Kea when there is 1.2-mm of water vapour above the observatory. In the Roque de los Muchachos Observatory around 10-15% of the time the water vapour over the observatory is as good as this or better, corresponding to what is meteorologically an extremely dry atmosphere - at sea level the amount of water vapour even on a cloudless day is usually some 30 to 50-mm (this means that if all the water were suddenly to rain out it would cover the ground with a layer of water this deep).

Around 11 microns there is a region where the atmosphere is almost completely transparent and a narrower one at 8 microns - in between atmospheric ozone makes this window almost totally opaque. Around 15 microns Carbon dioxide completely blocks out infrared radiation from space. We see that the window at 20 microns has some good, very transparent regions, but these are narrow and separated by deep chasms of absorption by water vapour.

In the following plots we can see these regions in detail. CanariCam will work from about 7.5 microns to about 24 microns (these limits depend on the atmosphere at the short wavelength end and on the detector at the long wavelength limit).

As it is quite transparent in its best regions the 10-micron window has been quite widely used in the past from the Earth's surface.

CanariCam will use a broad filter to exploit this transparency, plus a series of narrow filters in this window to look at individual lines and emissions from astronomical objects.

Deep water vapour lines cross the 20-micron window. In some regions there is still 95% transmission by the atmosphere, but these are narrow and separated by highly opaque areas. CanariCam will use both a broad filter that covers the entire window, and a series of very narrow filters that take in the best of the individual regions of this window.

Here we see why beyond 24-microns science returns are not guaranteed. The atmospheric transparency drops off rapidly and the windows become ever fewer and more opaque. As the sensitivity of the detector will probably drop off sharply from 24 microns, we can see that this region of the spectrum is a very poor one for ground-based science. CanariCam will have a narrow filter at 24.5 microns to exploit the best of these windows that, in very dry conditions could give extremely important results. This filter will allow astronomers to examine cool objects in space, possibly even Kuiper Belt comets that cannot be detected at shorter wavelengths and planets orbiting around other stars at about the distance of Uranus or Neptune.

Click here to see more about atmospheric conditions at the GTC site.
 

The influence of the warm background

Mid-infrared astronomy is greatly influenced by the fact that everything emits infrared radiation due to the very fact of it being warm. An object at room temperature emits radiation best and in largest quantity around 10 microns. Cooler objects emit at longer wavelengths.

What this means in practice is that the part of the light that the mirror of the telescope does not reflect is reemited as heat and detected by CanariCam and the parts of the telescope structure that the mirror sees also emit infrared radiation. In the same way, the part of the light that the atmosphere absorbs is also reemited as heat and detected by the instrument.

In other words, any star or galaxy that we wish to measure is bathed in a haze of radiation from the instrument's surroundings. Often the galaxy is many millions of times fainter than this background radiation.

Here we see the relative contributions of background light from the sky (green line) and a perfect telescope (red line). Here we assume that there is 2-mm of water vapour and that the telescope reflects 98% of the light that reaches its mirror (an optical telescope may not pass 80% unless it has recently been cleaned and 95% is very difficult to attain). By making the optics as clean and reflective as possible (this includes both the telescope mirrors and the mirrors inside CanariCam) we can make the radiation from the atmosphere the deciding factor in many cases, but in the best parts of the 10-micron window what limits us is the optical path.

The contribution to the background from the telescope is minimised by making the telescope structure as light as possible (in that way we "see" the minimum of support structure) and by taking careful measures to maintain the mirrors very clean. This includes removing segments from the GTC mirror every week to clean them on a cyclical basis so that every six months approximately the entire mirror surface has been renewed. Giving the telescope a gold-coated secondary mirror may make another improvement in the future, as gold reflects infrared light very efficiently - much more so than aluminium or silver. These factors are beyond the control of CanariCam.

The CanariCam team can control what happens inside the instrument. Gold mirrors, each 99% reflective for infrared light are used. The detector and mirrors are cooled to below 30K (-245ºC approximately) and a special shield (a light baffle) cooled to the same temperature blocks out all light that does not come from the telescope mirror.

We can also improve the instrument's sensitivity significantly by observing in cold and very dry conditions that will reduce the background still further. By scheduling the most critical observations in queue observing mode we can take advantage of the very best conditions in the observatory, whenever they happen.
 

How does CanariCam detect stars one million times fainter than the sky?

We use a special technique to detect faint sources. We do not "stare" at a fixed point (our star) but rather flip between different positions to compare the signals received. Two movements are involved: a fast chopping movement and a slow nod.

If we are observing a point source (a star), we chop between two positions on the sky separated by about 10 arcseconds 5 times each second. Making the GTC's secondary mirror oscillate quickly in a very precisely controlled manner does this. Our star, which is probably completely invisible, changes between two positions (termed "on" and "off") on the detector. We subtract one exposure from the other and remove the sky background and most of the noise in the image and are left with a positive and a negative image of our star. However, there may be gradients and slow variations of the sky across the field of view so, to eliminate these, once or twice a minute we move the telescope to a new position (termed a "nod") and integrate again in exactly the same way as before.

We end up with two positive and two negative images of the star as in the OSCIR data shown below.

These are then combined to give the final summed image.

This is a very simple case. We can even see the star in the unprocessed images (which will rarely be the case on the telescope with real sources). How though do we manage to detect a star that has 2000 counts above a sky that may have one million?

The answer is quite simple. Although the sky has one million counts, the noise in the sky background - the random fluctuations from pixel to pixel, or Poisson noise - is the square root of this... 1000 counts. In other words, we have twice as much signal from the star as noise from the sky and our object is easy to detect by the process above.

What about much fainter objects? As we integrate for longer, the noise in the sky goes up as the square root of the total exposure time, but the signal from the star increases with the exposure time. In other words, every time we multiply the exposure time by a factor of 4 we get twice as much signal to noise and detect objects 0.8 magnitudes fainter.

One big advantage of CanariCam is its small pixels and exceptional image quality. The image of a source is effectively limited to a diameter of two pixels and because the pixels are small, the sky background in each pixel is minimised. This maximises the sensitivity.

CanariCam should be able to detect magnitude 11.5 at 10 microns in a 100s exposure with a signal to noise ratio of 5 (this is much better than IRAS and a little better than ISO and some 4-5 magnitudes better than even the 4-m telescope at CTIO) and has the huge advantage that it can make very long, deep exposures. In ten thousand seconds (a fairly standard deep exposure for astronomers), CanariCam will give a good detection at magnitude 14 at 10 microns.

Here is a case of real OSCIR data of a source (the central region of the Starburst galaxy NGC 253) that is both invisible in the raw frames and extended so that the chop must reach out to blank sky to one side of the source, outside the field of view.

The top row of four images shows the raw data frames from the two secondary mirror positions at each of the two nod positions (called "Nod Position A" and "Nod Position B") of the telescope. Fixed-pattern offsets due to pixel-to-pixel variations and offsets between the 16 channels of the acquisition electronics dominate these images. The background counts in these raw images correspond to 7.4x 10⁸ e^-/s. Each raw image consists of ~5 minutes of total integration time (i.e. 15,000 frames co-added using a 20 ms frame integration time and the N-band filter) obtained in the chop/nod sequence as described above.

The second row of two images shows the "chop-differenced" data derived from the subtraction of the on-source and off-source data in the two nod positions of the telescope. Note that the dominant pattern (principally a gradient along the diagonal connecting the lower-left to upper-right corners of the images) has changed sign between the two chop-differenced frames. However, since the subtraction is always done as "on-source minus off-source", the source signal remains positive in both chop-differenced frames. The signal levels in these differenced frames range ±3.2x10⁶ e^-/s, which is ~0.4% of the raw background signal. Finally, the bottom row shows the net signal obtained by adding together the two chop-differenced frames shown in the middle row. No other processing has been done to the data other than the additions and subtractions as described above.

The net signal is the result of a total exposure of ~20 minutes in which half that time is actually spent imaging the off-source "reference" position. The signal level at the "tail" of this source near the middle of the frame is ~6.4´ 10⁴ e-/s. This is about four orders of magnitude below the background level shown in the raw frame. In fact, the signal-to-noise ratio at this level in each pixel is about seven, so that the effective background subtraction is more nearly five orders of magnitude below the background!