Wide-field CCD Imaging

HET609, Project 101, Pamela Gay supervisor

Semester 1/2003

Jose Ribeiro

Abstract

CCDs, due to their limited size, cannot by themselves achieve images of wide-fields of the sky. This work aims the study of the methods to overcome this limitation, and experiment the efficacy of some of them.

Keywords: CCD, wide-field sky imaging

Introduction

A great improvement in imaging devices happened in the decade of 70 in the 20th century. A solid-state chip was developed as a new kind of computer memory. It was then found that this chip is photosensitive and that it has the capability to produce electrical charges proportional to the number of photons that hit it. The Charge Coupled Device ( CCD ) was born as an electronic imaging device. In less than three decades CCDs substituted almost completely the cathodic tubes, less photosensitive, bigger and heavier, becoming the most used electronic imaging device.

CCDs consist on two-dimensional arrays of pixels, each pixel corresponding to a dot in the image. In practice, each pixel works as a charge well, that is filled every time a photon hits it. At the end of the image exposure the charge of the pixels is read in a sequential way, where each line of pixels is shifted into a register and then the charge of each pixel of that line is measured and digitised.

Due to their high photosensitivity and linearity with exposure time, CCDs conquered their royal seat in scientific as well as artistic astro-imaging, pushing photography to a second plane.

There is however an important issue: big CCDs are very expensive, and there are technological obstacles to the increase of their number of pixels. This problem tends to limit the field-of-view ( fov ) covered by each CCD image.

The fov of an image is given by

fov = 2 arctg (w/2f)

where w is the dimension of the sensor and f is the focal length of the objective in the same units as w. For small angles, where tgq »q, this formula may be simplified to [1]

fov(arcsec) = 206265 w/f

At title of example, comparing a 35mm photographic film, which is 36 x 24 mm sized, with an average size CCD with 8 mm diagonal, (6.4 x 4.8 mm), for the same objective the fov of the film is 5.6 bigger than that of the CCD. Saying this in another manner, a 35mm film at the focus of a 1m focal telescope has approximately the same fov as an amateur sized CCD with a 180mm photographic objective.

Wide-Field Techniques

In order to obtain wide-fields of the sky with CCDs some measures may be taken. The scientific institutions which are economically supported, have the means to get large CCDs, and combine them in mosaics. At this level, 4096 x 4096 pixel CCDs may be found and the mosaic of CCDs may be positioned at the telescope’s focus. Another solution, consists in the adaptation of photographic objectives, with shorter focal lengths to the CCDs. Some groups have used for years arrays of such systems in the search of the gamma-ray burst’s optical afterglows (LOTIS and ROTSE), where wide-field search was imperative due to the GRB’s position uncertainty [2].

At amateur level, wide-field CCD imaging is also possible. The chips are smaller, of the order of 512 x 512 pixels, and mosaics of chips are not affordable at this level. The solution here passes by the acquisition of adjacent images at

the field, that will be assembled in a mosaic at home. Another solution is the use of short focus photographic objectives.

The major issue in using short focus objectives is the risk of undersampling the images. For example, star may fall in just a part of a pixel. As a measure of size, one uses the full-width at half-maximum (FWHM) which is the width of the object at a point that is half the object’s pick intensity value. An object is considered to be well-sampled if its FWHM attains at least 2.5 pixels in the sensor. Less, will be considered undersampled.

In order to solve the undersampling problem some techniques are applied, such as shift-and-add, interlacing and variable-pixel linear reconstruction (Drizzle). The drizzle algorithm was created in order to solve the undersampling problem in the WFPC2 camera of the Hubble Space Telescope, a mosaic of three large wide-field CCDs plus one smaller planetary camera. This algorithm is based on the fact that small shifts in consecutive images makes the undersampled object fall in different positions at the detector (a phenomena known as dither). A finer output grid of pixels is created and overlaid on the original one, in order for the undersampled object to occupy more than 1 pixel of this new grid. The amount of each part of the original pixel in the new grid is then calculated and averaged [3](Fig.2). This method may be used either by a professional or by an amateur astronomer.

In this project I have experimented the use of short focus objective. I have done some mosaics with them and the major mosaic with the telescope. I have tried the drizzle algorithm, but without any good results. On what concerns data reduction, I have experimented either dome flat-fields as well as twilight sky flats.

Material and Related Issues

In the practical part of this work, the following material was used:

- one video security camera Mintron 12V1C-EX (www.nitehawk.com/rasmit/low_light.html)

- one German equatorial mount Meade LXD55 with Autostar

- one old wooden tripod from Carl Zeiss Jena

- one telescope Schmidt-Newton Meade LXD55, 8”, f/4

- one photographic teleobjective 135mm, f/2.8

- one photographic objective, 25mm, f/1.5

- Pentax and Minolta to C adapters from Soligor

- one notebook Toshiba Satellite

- one PCMCIA card ImperX video-grabber, and respective software

- Iris data reduction software

- Corel Photopaint

Caixa de texto: Fig.3 In field sometimes one must improvise

One must say that this will be rather a video-astronomy work than a CCD work, but at the end, the principles of actuation are similar.

The video camera is equipped with a Sony CCD type ICX248AL, with 768 (H) x 494 (V) effective pixels, and a pixel size of 8.4m x 9.8m. The image area of the chip is 6.45 mm x 4.84 mm. This camera allows up to 2 s integration, attaining 0.0001 Lux.

The usual procedure with this camera for deep-sky imaging is to grab, correct, register and add the amount of frames necessary to obtain a good signal-to-noise ratio, and then do the final touches with a photopaint software.

If in a conventional CCD camera one must wait for the end of the exposure to see the results, with the above mentioned set-up, in elusive objects one only knows that they are there after having reduced the images. Sometimes, an object is missed and that is only detected after reducing the data.

The major problem in video-astronomy is the fact that the camera output is an analogic video-compost signal that must be reconverted into digital data again. For this job a video grabber card is used (the issue here is the fact that we are converting and image from a 768 x 494 rectangular pixels into an image of 640 x 480 square pixels). The card used was the only one that I found in PCMCIA connection. The major problem with this card is the fact that it digitises only in 8 bits, 256 brightness levels and it is notorious that the digital frame quality in what concerns contrast is worse than the video frame as seen in a video monitor. Another issue is the video lines that risk to be visible in the digitised frame. The best way to resolve this problem is to interlace the even with the odd lines. The card’s software allows this to be done and in fact ii strongly improves the image quality.

The obtained images were stocked in the computer’s hard disk, in a sequential numeration with the bitmap format. Each frame occupied 900 kilobyte of information, because the video grabber software only accepted the 3 x 256 colour mode and not only the 256 grey-scale, something that should be improved. Due to this fact, the memory capacity of the hard disk became a problem in the field. This was solved by compressing the files while acquiring new images.

The software used for the data reduction was Iris software from www.astrosurf.com/buil/us/iris/iris.htm The reason for the choice is that it is a complete package that works under Windows, and it is freeware. Unfortunately, I never used UNIX nor Linux and so I could not use that fascinating program IRAF.

Concerning the mount, an improvement was mandatory: the change of the tripod of origin by a more stable one. For this job I elected an old wooden tripod used in geodesy. Some images were taken at the focus of the telescope, others with photographic objectives adapted to the camera, where the system was fitted to the mount through a dove-tail.

Field Work

First Night

It was only in the 2nd May that I could make the first experiments. Before that date it was raining all the time. By then I was uncertain about what objects could be appropriate for my limited set-up.

As I was using a 135 mm objective I opted by a dome flat field due to the wide field of the system, 2.7º x 2º . For this I painted a painting canvas in a flat white colour. Here, I am not sure if I have done it correctly, but I profited the sky natural illumination, that seemed to me homogeneous, with the canvas positioned horizontally. I have taken several (30) 2 s exposure images with the lens focused at infinity. This was my first 135 mm flat field. After that I took 30 2 s dark frames. A word must be said on the difficulty of making dome flat fields without an observatory, specially when one is using a telescope. The fact that the camera must be in the same relative position in the flats and in the images of interest, makes it difficult to do a dome flat and then transport everything into the field. At least, as I am not using filters, I do not have a system to position my camera in the telescope always with the same orientation ... something that I must improve in the future.

My first object was a galactic field in Coma Bernices. I have taken four contiguous images just to see if it worked, with a set of 2 s dark frames between them. As my camera allows only 2 s integration, I decided to take 225 frames of each image (7.5 minutes).

When I finished this object, Cygnus was there and I tried to image the North-America Nebula. This object is very elusive to my camera, so I had to guide myself and the system through the star field, with the help of the Sky Atlas 2000 [4]. I have made a line connecting x Cygnus to Deneb. Two 225 frame images were enough to cover that path. However, I took a third image to East, just in case. Sets of 2 s dark frames were taken between images (Fig.5).

Dawn was approaching, but I still had time to explore NGC6992, just some frames to see if it was at the possibility of my equipment. Later, I realised that this magnificent object was at my range of operation, which made me taking the risk of making a telescopic mosaic of it.

Second night

This night had the purpose to make the necessary images of the mosaic of NGC6992. I chose the darkest possible site, one of the darkest in Western Europe, in the Southern side of the border between Portugal and Spain (named Pulo do Lobo). It was on the 9th May and the moon was at the first quarter. That would not be a problem because the moon would set when Cygnus is at a good position. As I was using the telescope with a fov of 27’ x 20’ I opted to experiment a twilight flat field. Howell [5] writes in his book that “one good sky region for twilight flats has been determined to be an area 13º East of Zenith just after sunset”. So, be it. Telescope aimed to that region, focused, and drives off. When I began to see the star trails I began shooting. After data reduction, I realised that my first twilight flat was OK, and that the median process makes the stars disappear by magic.

I had then plenty of time to dinner while I reviewed the sky maps that I had printed in order to aim the telescope correctly to NGC6992.

An accurate polar alignment was performed in order to have a great steadiness in the 339 frames of each image, corresponding to 11.3 minutes of exposure. The night was a bit windy, and fortunately I had the wooden tripod.

As soon as the left wing of Cygnus rose to 30º above the horizon I aimed the telescope to the Northern region of the Veil. I then realised that none of my maps had enough accuracy to orient myself in that part of the sky. The number of stars was extremely high and nothing in the image corresponded to nothing in the maps. I tried a photograph that I had in Burnham’s Celestial Handbook [6], but it was overexposed, without stars ‘ contrast. Fortunately, an observer companion had the “The Night Sky Observer’s Guide” [7], where I found a detailed photography of NGC6992. I must confess that I wasn’t expecting such a difficulty and by then I understood that a photographic atlas is a fundamental tool for sky orientation when using the set-up I was using. Perhaps this work will be the beginning of such an atlas.

The first asterism that was found oriented me on the place of the Veil I was and gave birth to the mosaic’s image #1. During that night six images were taken with dark frames in between, apparently covering the whole object. The image #6 was taken at dawn. After having reduced the images, I realised that image #5 had missed completely the object. Fatigue was the winner that night forcing me to a new night session in order to complete the mosaic.

Third night

For schedule reasons, I could not displace myself to the same dark place, far from home and decided to finish the mosaic at my group’s usual observing place, Atalaia, (www.atalaia.org) some fifteen linear miles from Lisbon, a one-million people town. Fifteen days had passed from the last session, the moon was in the last quarter and Cygnus was rising one hour earlier. My sky orientation was now very good at the place of interest. Furthermore, that would not be my first twilight flat field! Now I knew that the star trails would be removed by the median of the images.

Three more images were taken in order to cover the whole object. My question now was whether the difference of skies would or not be compensated by the data reduction.

That night I managed to go to bed just before dawn.

The Drizzle Experiment

I used an image of Monoceros region taken in January 2003 with a 25 mm f/1.5 objective, composed by 20 frames. The image composed by drizzling seems to be defocused, relatively to the image obtained by adding frames (Fig.6) Either the algorithm is not correct in Iris or I have done something wrong, which is the most probable. I think that the failure of this experiment is due to the fact that I did not have enough dithering between frames in order to the algorithm to work. More work must be done in the future.

Data Reduction

Data reduction is essential in order to calibrate the images, i.e. images of the same object taken at different times and with different equipment must give the same scientific information. In order to reduce data, a set of procedures must be done. For each frame of interest, one dark frame, one flat field frame, and one flat field flat frame are associated. In very well refrigerated CCDs, where liquid nitrogen is used, the thermal noise (dark current) is nearly null, being the dark frame substituted by a bias frame. On the other hand, a dark frame already contains the bias information. The basic data reduction is as follows:

Reduced frame = (raw object – dark frame) / (flat field – flat field dark frame)

The bias frame is used to determinate the noise due to the on-chip amplifiers. In order to obtain it just a mere read out of the CCD without exposure is required.

The dark frame consists in an exposure in the dark with a duration equivalent to the exposure of the object of interest. It determinate the thermal noise of the CCD, which as in all the semiconductors is extremely sensitive to the temperature.

Flat fielding is by far the most complex procedure in the image calibration process. The flat fielding is necessary to correct two factors in the image. One, is related to the fact that each pixel in a CCD has its own sensitivity to light (quantum efficiency), i.e. for the same number of photons each pixel generates its own electric charge that is different from the other pixels. An homogeneously illuminated, not saturated CCD will give different values for each pixel. Flat field corrects that anomaly as if all the pixels in the CCD had the same quantum efficiency. The other correction that the flat field makes is the “cleaning” of the artefacts due to dust (donuts and spots) and scratches in the optics. The procedure consists in illuminating homogeneously the CCD in order to stay about 70% below saturation. This can be done either by aiming the telescope to an illuminated surface at the observatory’s dome or directly by aiming the telescope to the twilight sky. In the infrared, the night sky is also suitable. Several flat fields should be taken and averaged for each filter used. In the case of the twilight flats, a median must be used instead of averaging in order to remove the star trails. A dark frame for the flat field must be taken and subtracted to it.

Practical Notes in the Reduction of NGC6992 Mosaic

At title of example I describe here the procedure for reducing one of the nine images done to complete the Veil’s mosaic. The Iris program was used. It has the advantage of repeating the same instruction for a set of frames.

Object Dark Frame: ten 2 s dark frames were taken for each image of interest.

1 – Bitmap conversion into fit.

2 – The ten frames are added

3 – The resulting image is divided by ten (average)

4 – Save dark frame

Flat Field Dark Frame: the same procedure as above.

Flat Field Image: thirty 2 s twilight images were taken.

1 – Bitmap conversion into fit.

2 - Subtraction of the flat field dark frame to the 30 images.

3 – Determination of the median of the 30 flat field frames

4 – Save the flat field

Image of Interest: for each image in the mosaic, 339 2 s frames were taken

1 – Bitmap conversion into fit.

2 - Subtraction of the dark frame to the 339 images.

3 – In “statistics”, measure the maximum pixel intensity of the image 1/339. In this case, a value of 238 out of 255 was obtained.

4 – Load the flat field image

5 – Measure the maximum pixel of the flat: 172. The maximum pixel value allowed by Iris is 32727. The maximum value of the flat must be corrected to one which is equivalent to multiply the image by 172 and divide it by the flat. The issue here is that 172 x 238 = 40936 > 32727. So, I opted to multiply the flat by a factor of 0.799 in order to avoid saturation. The maximum pixel of the flat became 137.

6 – The object frames were multiplied by 137 and divided by the corrected flat field. Iris has an automatic routine for this.

7 – Register the resulting images. This procedure is also automatic, being necessary to mark one star.

8 – Subtracting the sky. Here a small area of the sky must be chosen. For this job it is better to add a few registered images in order to see the object of interest and so elect a dark area of sky. The command also acts automatically in all the 339 frames (NOFFSET2 A B 0 339).

9 – With the function “statistics” measure the value of the brightest pixel in frame 1. As a stack of 339 images is about to be added, one must be sure that the maximum pixel x 339 will not attain the saturation value of 32727. The value was 267 (I feel that this value should be less or equal to 255. Perhaps due to rounding errors, or the algorithm used by Iris is not accurate?). An attenuation factor was calculated: 32700/267/339 = 0.361.

10 - Multiplication of the 339 frames by 0.361.

11 – Add the 339 frames.

12 – Save the result.

After having repeated this procedure for the work of the second night, it was time to assemble the mosaic.

Iris has a function “mosaic” where by hand one can orient the several images in order to overlay the stars common in adjacent images. Some software packages based by professional astronomers based on star triangulation do this job automatically. The position map of stars in dithered images (as when drizzling) may also provide useful information for automatic mosaicing. [16]

When I finished the part of the mosaic I noticed that image #5 was out of the target, and that some parts missed the object. Three more images were necessary to accomplish the job. When the final draft mosaic was assembled, I calculated what portion of each image could be trimmed, and then I reassembled the final mosaic that was then ready for the final touches.

A logarithm applied to the mosaic in order to enhance the middle tones did not improve significantly the image and so I opted by not using it. Small correction in contrast and brightness were sufficient to enhance the object. A final touch of a very discrete unsharp masking and I gave the job as concluded.

The Object

Fig.8 The complete mosaic of the Northern part of the Veil nebula, NGC6992, a supernova remnant. Its progenitor might be a O8 or O9 type massive star that blasted away some 10000 years ago at a distance of 440 pc.

The image shows the Northern part of the Cygnus Loop, NGC6992 and NGC6995. The Cygnus Loop also known as the Veil Nebula due to its shape of bridal veil, is a supernova remnant thought to be 770 pc from us [8]. However, recent studies reveal that this distance may be about 440 pc [9].

It is thought that the progenitor was a main sequence of 15 solar masses star with an O8 or O9 spectral class, derived by the distribution of the metal abundance, that increases toward the inner region [10][11]. “The low abundance of Fe relative to Si and S suggests a type II S.N.” [14]. Due to its total dimension, the Veil is at least ten times the age of the Crab Nebula that was registered in 1054 AD [12]. The resulting neutron star was not yet detected but the discovery of a compact X-ray source inside the Cygnus Loop may point to the possibility of the existence of a rotating neutron star [13]. The image clearly shows the filamentary structure of the Nebula. This structure is due to shock waves generated by the interaction of the expanding matter with the interstellar medium. The radiation is mainly due to the deceleration of electrons in huge magnetic fields, synchrotron radiation.

A recent work refers that the Cygnus Loop could be two supernova remnants interacting with each other [15].

Conclusion

Wide field of views of the sky are important for the study of large structures in our neighbourhood. These structures, due to the proximity to the Earth, are easy to study allowing the construction of models applicable in the more distant Universe.

With this work I managed to prove that CCDs, even small ones, are apt to image wide fields of sky. However, this a time consuming task, leading to the eternal dilemma “time-for-money-for-time”.

In a practical work like this one it is easy to make mistakes. The important thing is to learn from them. My major error was going to the field trusting in my skills and passion of visually observing the sky. Big mistake! The sky through a CCD is so rich that the reference points are difficult to get. No atlas of the sky, as far as I know, reaches the magnitudes in question.

A lot of things must be improved concerning flat field acquisition, mainly in what concerns dome flats set-up. I was impressed with the twilight flat technique, despite its limitations due to the quick changing of the sky illumination.

I feel that my knowledge on the sky subtraction techniques must still be worked. I live in a town and should like to develop/find a technique to subtract an unhomogeneous light polluted sky in order to get wide fov of sky from home.

Bad pixel removal is also an issue that must be improved.

I learnt a lot from this work. No doubt that when applying the theory to practice one realises all the weak points and the doubts one has on the subject itself. Blood, sweat and tears, but also the satisfaction of the accomplishment of a difficult but rewarding task.

Acknowledgements

I thank to my project supervisor Pamela Gay for her prompt orientation whenever asked for.

I thank to my observing companion Alberto Fernando, who accompanied and assisted me in the second and third nights. Thanks to his book I managed to get the Veil’s mosaic.

Finally, I thank to my observing group from Atalaia, for the spirit of discussion (and beer) that makes amateur astronomy so pleasant to me.