Diffuse Detection at the AAT: Fabry-Perot Stare Method with Nod+Shuffle

The Sept/Oct 2001 run was the first time that we managed to get a significant amount of photometric data with the nod+shuffle technique. Full operational issues relating to nod+shuffle with TAURUS are to be found here.

This method has now revolutionized the sensitivity which can be reached in emission lines of diffuse extended sources. The night sky emission and absorption features are a worthy adversary which make this sort of work extremely difficult. The nod+shuffle technique (discussed at length by Glazebrook & Bland-Hawthorn 2001, PASP, Feb issue) allows for charge to be shuffled backwards and forwards on the CCD in synchrony with the telescope being nodded between two positions on the sky. As we show below, we are finally able to subtract out even the worst features the atmosphere can throw at us. Note that all of the results shown below were taken in full moon.

Instrument set up

In our experiment, we use the University of Maryland 44um etalon which gives 1.4A resolution at H-alpha. This is placed within the TAURUS focal reducer at the f/8 Cass focus of the AAT. We use a MITLL2 CCD (2Kx4K) in either XTRASLOW or SLOW readout (read noise/gain = 1.0/11, 1.8/2.7 respectively). The high gain and low read noise is essential to the sensitivity of our experiment. The instrument has a 10' field of view sampled at 0.37"/pix.

Here is a summary of the key parameters:

MITLL2 2Kx4K, CCD mask to remove ghosts
XTRASLOW, read=1e, readtime=480s, gain=11 (possible dark time option)
SLOW, read=1.8e, readtime=160s, gain=2.7 (normal observing mode)

The high gain is more important than the read noise. The read time has a major impact on the observing efficiency (see below).

UMd etalon, 1.4A, 54.3A FSR, standard CS100 setup
6570/45A filter, tilt=10deg
Calibrate scale over field with D-alpha (6561)
Optimize for H-alpha only, no tilt, 60000 pixels in line (FWHM), 90000 (EW)
This is full usage since 1600x1600 area, inscribed circle divide by spectral range = 23A.

Important notes on nod + shuffle specific to Fabry-Perot Stare method.

Nod+shuffle, 100 sec on, 20 sec slew (AXES mode), 15% overhead
Axes A/B, A pushed to +240mm X 6.68"/mm X 1.4142 (SW) target
                B pushed to -240mm X 6.68"/mm X 1.4142 (NE) off field

Total nod distance = 1.26 deg (can accomodate 2.5deg cloud size)

This was checked three times by going to object field and ensuring star pattern matched shuffle image; Axes A/B requires that we point at half way point between on and off target. Telescope works in RA,dec (equatorial) so must compute RA,dec offset to middle of field with1/cos(dec) correction. Take dec=-26deg. Offset to middle is (1600",1600") on sky, which is (1780",1600") in time. This is what we give telescope to determine correct displacement. (If at poles, big RA shift corresponds to tiny telescope movement, etc.)

On chip, A is lower part of chip,  B is upper part of chip. Note that the FITS headers will record the wrong RA, dec. Ask the night assistant to record the on-field RA, dec in the log.

Observing efficiency:

XTRASLOW: Main exposure, 8x(100+20)x2 + 480 = 2400s, 1600s on source (35% CCD overhead, 50% total overhead)
SLOW: Main exposure, 8x(100+20)x2 + 160 = 2080s, 1600s on source (15% CCD overhead, 30% total overhead)

It can get to be really complicated figuring out what the RA,dec displacements correspond to on the sky in terms of galactic coordinates. I recommend the calculators on the NED web site and astutil/galactic under IRAF.


The four images below illustrate the method. The first is a sky flat. The same patch of sky was imaged in OFFSET mode (where the offset was zero) and charge shuffled to produce two 10' images side by side. Note that the images are essentially identical showing that the structure is laid down on the chip where the light falls, and is not influenced by the shuffle. The second image is the summed bias frame (10 frames) and reveals high frequency low level structure - this of course must be subtracted from the data. The third field is a 100sec shuffled exposure (1 cycle) on two WHAM survey fields 1.26 deg apart; this image has been flatfielded. The Reynolds layer emission is at the bottom of the geocoronal absorption feature - we were observing in full moon. Note the OH 6553 emission feature at larger radius. (The differenced image is shown below.) The fourth image is one (100sec x 8 cycle) snapshot of the science field TonS210 discussed below (not flatfielded).

The exposure below is the difference of the lower and upper panels in the 3rd image above. Note the residual Reynold's Layer emission of about 200mR as expected. This is quite remarkable given that the exposure is only 100 sec on each field (I have done a 5x5 median smooth to bring out the ring). Note the faint satellite trail which barely shows up in the original data.



We now discuss the TonS210 field. The observing conditions were perfect for the first three nights of a 5 night run. The shuffled image shown above is one exposure of four that were taken. The on-field was positioned on the TonS210 sight line, and the off-field position was 1.26 deg exactly NE from this position. (I have a latex file with complete details.)

We show below the reduced data for the TonS210 field. I have simply bias-subtracted, flatfielded, and azimuthally binned the upper and lower images into independent spectra. The on-field/off-field differences are shown below. The vertical axis shows CCD counts, the horizontal axis is wavelength calibrated.

We are looking for H-alpha emission features at -200 km/s which is about 6558.4A or so. The correction to LSR velocities is negligible along this sight line.

There are basically three different ways to derive azimuthally binned spectra: modes, means and medians. If the noise is truly poissonian, all of these should agree. Below, I show the results for the mean and median analysis. They agree very well although the median spectra are a little cleaner.

Here are the mean spectral differences for the four images. Note the remarkable degree of consistency between them, especially when you consider that the moon was coursing through the sky. The geocoronal H-alpha in emission and absorption subtracts well leaving a weak differential signature from the Reynold's Layer as expected. The very bright OH 6553 feature leaves a residual which can be modelled and removed. The important thing to note is the very flat baseline between 6553 and 6563A.

The median spectra are shown below. These are somewhat clean than above and the Reynold's Layer differential is not as pronounced, although see the combined plots below. Note that we get 23A of spectral coverage. The blue limit corresponds to large radius in the interferograms; the red limit is the centre of the interference pattern.


The plot below compares the summed data for on (upper) and off (lower) field. Note that the curves are doubled up in each case. Here, we are comparing the mean and median for on and off fields. Note the excellent consistency in all cases. This is a major improvement on the previous stare method of getting interleaved images of 10-20 min exposure.

The difference of the on and off fields (mean and median) is shown at the bottom of the plot. The agreement is stunning. I have never seen such a well behaved baseline over 10A before. All you see here are straight differences and nothing more. I can improve things with line scaling as is traditionally required but there is absolutely no need. See the discussion and zoomed plot below.

The noise level from 6553A to 6565A is 1.2 counts at 1-sigma (see the histogram below where the horizontal axis corresponds to counts per pixel). It is easy to show that this is equivalent to 27mR. Thus, we are able to reject a line detection at or close to 6558.4A exceeding 54mR at the 2-sigma level. (Recall that rejections which arise from differences are quoted at the 2-sigma level. Note also that this is a TRUE poisson rejection, something not common to observational astronomy. The numbers quoted here are conservative; the performance may be slightly better than quoted here. I need to recheck the flux calibration.)

The differenced spectrum has a mean level with negative counts since we were observing in full moon. The moon generates a very steep gradient over the sky. Maybe in future we should shuffle along tangent points to the lunar direction.

If the entire CHVC is found to be fainter than 54 mR, in the context of the H-alpha distance model, this puts the cloud beyond about 40 kpc. (I can do a proper estimate and plot if needed.) Note that the TonS210 sight line misses the densest HI regions.

The plot below is a zoomed version of the figure above. Note that the baseline is less than zero simply because the bright moon produces a strong gradient across the entire sky. Here we show a comparison of the differenced mean and median spectra. See below for a better presentation of the final result - this emphasizes the power of the technique.

Below we show the differenced median spectrum. Apart from bias-subtraction and flatfielding, there has been no other processing. The Reynolds Layer residual is at about the 70 mR level. We do not detect H-alpha along this sight line.

In a few dark nights, we could get to a few mR with this technique.

JBH at site, Sept 30, 2001.