- Data distribution
- Using the Archive
- Gemini Home
- Telescopes and Sites
- Science Visitors at Gemini
- Observing With Gemini
- Retired Instruments
- Visiting Instrument Policy
- Visiting Instrument Telescope Interfaces
- DSSI Speckle Camera
- TEXES (North)
- Integration Time Calculators
- Magnitudes and Fluxes
- Near-IR Resources
- Mid-IR Resources
- Observing Condition Constraints
- Performance Monitoring
- SV/Demo Science
- Future Instrumentation & Current Development
- Queue and Schedules
- Data and Results
- Calibration Data Retrieval
- Past/Future Development
Change page style:
Introduction to Ground-based Mid-IR Observing
Compared to observing from space platforms such as Spitzer, observing at mid-IR wavelengths from large ground-based telescopes on good sites brings two principal benefits. First the image quality from such sites is usually diffraction limited (thus scaling with the telescope diameter) and therefore on Gemini, for example, is higher by more than an order of magnitude than on mid-IR space telescopes such as Spitzer. Second, ground-based telescopes, and Gemini North in particular, offer the possibility of observing at higher spectral resolution than do mid-IR instruments on current space telescopes. However, because of the bright thermal background, ground-based mid-IR observations are less sensitive than those from cooled space-borne telescopes, and in fact pose great challenges. This page outlines the principal techniques employed to alleviate the problems that arise in carrying out ground-based mid-IR imaging/photometry and spectroscopy. This presentation, given by Pat Roche of Oxford University at the 2005 UK Gemini Support Group mid-IR workshop, contains much useful background information on mid-IR observing.
Dealing with the thermal background: chopping and nodding
All Michelle, T-ReCS and TEXES observations are made in the thermal infrared, where the atmosphere and telescope are bright sources of radiation. Performing mid-IR observations from the ground is analogous to making optical observations during daytime on a telescope with luminous (although also reflecting) surfaces. The sky background in the best parts of the N band window (near 11.5 microns) is roughly 0 mag per square arcsec and it is brighter still in the Q band (17 to 25 microns).
The sky brightness changes on short timescales and over small distances on the sky, and often these changes are much larger than those due to Poisson statistics (i.e., >> sqrt(N), where N is the background in photons/sec). To alleviate the effects of this when performing imaging and low-resolution spectroscopy, a nearby position on the sky is observed frequently by moving the secondary mirror at a frequency of a few Hz (the telescope is said to chop between the target position and an adjacent sky position), and the pairs of images are subtracted. Moving the secondary mirror results in the telescope being seen by the detector in slightly different ways in the two secondary mirror positions. Since the telescope glows strongly at mid-IR wavelengths the subtraction leaves a residual radiative offset, which is much less than the telescope and sky brightnesses but is still significant. To get rid of (most of) this offset the entire telescope is moved or nodded typically about twice per minute. Normally the nod is set to be the same amplitude and direction as the chop, so that the science target switches chop positions between the two nod positions. Thus a pair of subtracted chop/nod observations produces a positive image of the target in the center which corresponds to half of the exposure time, and two displaced negative images of the target, each of which corresponds to 1/4 of the exposure time. If the chop and nod are large enough the negative images are off of the array and are not seen. Currently the maximum chop throw at Gemini is 15 arcsec.
The level of background radiation sets the frequency at which the detector is read out to avoid saturation (typically about every 20 milli-seconds) and the frequency at which chopping is needed (typically around 3 Hz); chopping needs to be faster when using filters that are more sensitive to atmospheric changes (see below). The result of the chop/nod procedure is to remove most of the background and leave the residual astronomical signal. Further details can be found at the bottom of this page.
In stable observing conditions, the background can be removed remarkably well. If, however, the atmosphere is changing over the course of a few minutes (e.g., when there are clouds) then the cancellation of a chop/nod cycle will be poor. This usually results in an offset in the sky background away from zero, which can be either positive or negative depending on how the sky is varying. The Gemini IRAF midir package provides tools to identify and reject frames with high residual background that would otherwise compromise "good" observations in the same sequence.
Currently Gemini only uses one of its peripheral wavefront sensors for tip/tilt guiding. Because the sensor has a very small field of view, autoguiding is only active at one of the chop positions. The image quality in the unguided position is almost always much worse than in the guided position and the unguided images should generally not be considered useful for science. When the nod distance is equal to the chop throw, half of the measurement time is spent on the single guided image and 1/4 of the time on each of the two unguided images. Compared to an ideal chop system with autoguiding at both chop positions, Gemini measurements are lower S/N by a factor of sqrt(3/2).
In most chop systems there is an additional factor of ~1.5 loss in S/N from the ideal due to overheads taking up roughly the same amount of time as the exposures. The overheads are associated with motion and settling of the secondary mirror, reacquiring guiding after each chop cycle, and reading out the array. See each instrument's Overheads page for more details.
Overall, compared to staring and nodding (as for typical near-IR observations) the loss in S/N due to the production of three images, single beam guiding, and overheads is more than a factor of two. We chop only because we have to.
Stare/nod medium- and high-resolution spectroscopy
In spectroscopic mode the background per pixel on the detector is much lower, both because the light is being dispersed and because the slit limits the field of view. One often can chop and nod more slowly than for imaging observations, but tests have shown that the same chop/nod method used for imaging must be used for low resolution spectroscopy. At the higher spectral resolutions available with Michelle (but not T-ReCS) sky fluctuations are low enough that stare/nod, with its inherently higher efficiency, may be used to advantage. At these resolutions and for short exposures Michelle is read-noise limited at wavelengths where the earth's atmosphere is transparent, and there is a considerable advantage in taking as long exposures as possible at those wavelengths. However, if portions of the spectral intervals to be observed contain telluric absorption lines, the sky emission in those lines may saturate the array if the exposure is too long, resulting in no useful information close to the line wavelengths (while improved S/N is obtained at other wavelengths in the intervals). Those who wish to propose to use Michelle at the higher resolutions are encouraged to get in touch with Gemini staff to discuss this issue.
Chopping and nodding: an example
Consider the case of the Gemini telescope with aperture diameter = 8.1 m, temperature = 273 K, and emissivity = 0.03. At the focal plane, at 10 µm wavelength, the thermal background flux density from the telescope itself will be about 144 Jy per square arcsecond. (1 Jansky = 10-26 W/m2/Hz.) For simplicity we will ignore the background flux from the sky — in the most transparent parts of the 8–13 µm window this could be below 20% of the telescope flux, but in the 20 µm window where the transmission is very low it could be several times higher. The flux density of the bright mid-infrared standard Alpha Boo, for comparison, is about 600 Jy at 10 µm. If 50% of the starlight is concentrated in a 0.7 arcsec diameter circle, then only 300/(300+147) = 0.67 of the total flux in that circle will be from the star.
The accuracy to which background flux can be subtracted is limited by several sources of noise: shot noise of the photon flux itself, "1/f" noise due to variations in the background caused by temperature drifts and clouds, and read noise and 1/f noise in the detector. The principal goal is to reach background-limited performance, where the photon noise dominates the other sources.
In our example case, 144 Jy/arcsec2 corresponds to 2.2e8 photons/m 2 /s/µm/arcsec2 at 10.0 µm. With a 1 micron bandpass filter, 0.084" × 0.084" pixels, and an instrumental throughput of 0.7, the photon flux at the detector is 5.5e7 photons/sec/pixel. With a quantum efficiency of 0.25, the rate that electrons are generated in the detector is 1.4e7 e-/sec/pixel. In a 10 second integration (which would be composed of several hundred ~20 msec frames), the number of accumulated electrons is 1.4e8 and the shot noise is the square root: 11,800. Therefore, the other noise components must be < 11,800 e-, or 1 part in 11,800 of the absolute signal, for this 10 second integration to be background limited.
1/f sky noise is suppressed with a technique called "chopping," in which the telescope's secondary mirror is oscillated in a square-wave pattern at a frequency of several Hz. The detector alternately views two fields or "beams" on the sky; we will call these beams A and B. At the start of an integration, beam A is "on-source," containing the object under observation, and beam B is "off- source", containing blank sky. Thus we have two measurements: (signal1 = source + skyA), and (signal2 = skyB). Computing (signal1 - signal2) cancels most, but not all, of the sky emission. In practice, because the optical path through the telescope optics is different for the two chop positions, the background level is also slightly different (skyA != skyB). To completely remove the background, therefore, the telescope is "nodded" periodically (~ 2 - 4 times per minute) to move the source from Beam A to Beam B, and the quantities (signal3 = skyA) and (signal4 = source + skyB) are recorded. By computing:
( signal1 - signal2) + ( signal4 - signal3) = ((source + skyA) - skyB ) + ((source + skyB) - skyA ) = 2 * source,
the background is properly subtracted. Usually the nodding is executed in a sequence with the source in Beam A, B, B, A to minimize nodding overheads and cancel the linear component of any temporal background variations. A detailed discussion how the chop/nod mode of observation removes the sky and telescope background where there are gradients with time and position, and of the difference between the results of this background subtraction for an ABBA nod pattern and that for an ABAB nod pattern, is given in in this PDF document.
This scenario is often complicated by the presence of astronomical sources in the "off-source" beams, so that the signal level of "skyA", for example, changes when the telescope nods. A judicious choice of chopping amplitude and angle is often necessary to avoid background-subtraction problems in crowded fields.
With modern imaging detectors, one might think that nodding alone may be sufficient to subtract the background, especially in the case of imaging a nearly point-like object where the surrounding field can be used to zero the sky. Several groups have attempted this, but each have found that the background varies so irregularly over the field after just a few seconds that it cannot be modeled and subtracted to better than 1 part in 10,000. So, all groups with mid-IR imaging systems continue to chop and nod. High resolution spectroscopic work, where the contrast between an emission line and the background can be much higher, is often done without chopping.
Example OSCIR images taken in the chop & nod mode at IRTF (provided by C. Telesco, R. Pina, and R.S. Fisher, then at the University of Florida.)
(Top) Four raw signal images which are coadditions of ~15,000 20 msec frames. The integration time for each image is ~ 5 minutes. The mean signal level is ~ 2.2e11 e-.
(Middle) "Chopped-Difference" images constructed from (signal1 - signal2) and (signal4 - signal3). The peak-valley variation is ~ 2e9 e-, only 1% of the absolute signal but still a few orders of magnitude above the background limit of 4.7e5.
(Bottom) The final "Net Source Signal" image showing NGC 253. The signal at the peak of the galaxy's nucleus is 6.0e7 e-. The noise level is about 6.6e5 e-/pixel, in good agreement with the expected background-limited performance.
Last update 24 Aug, 2006; T. Geballe and R. Mason, based on previous work by K. Volk and T. Hayward.