Filters & CCDs.


Picture UBVRI Filters UBVRI filters (the U and I are essentially opaque).


Here I will attempt to tell the whys, whats and hows of colour filters. The whys are easy - photometry and tri-colour imaging. The whats are also easy - bits of coloured glass. Hows? I discuss tri-colour imaging elsewhere, but photometry will have to wait.



A Potted History of Astronomical Photometry.

Most knowledge gained in astronomy is based on spectroscopic observations of astronomical objects. Photometry is really very low resolution spectroscopy and was practiced before true spectroscopy took over. Instead of using a prism or diffraction grating to produce a spectrum, special filters isolate sections of the spectrum allowing some knowledge of the objects colours to be observed.

Ever since photometry became an accurate science, standard bands have been used. Originally defined by photographic plates and filters it was supplanted by photoelectric photometry which was done with (mostly) blue sensitive photomultiplier tubes (PMT). Three bands became standard which corresponded to the UV, blue and yellow-green regions of the spectrum, and called U, B, & V. Their passbands were defined both by the filters and the PMT which was used. The PMT was the 1P21 which had good UV and blue response and trailing off into the red. The filters which defined the bands were standard coloured glass. 1mm UG1 for the U band; 2mm GG385 plus 1mm BG12 for the V band; and 2mm GG495 for the V band (the filters had different names then, but these names are the current equivalents). The atmosphere provides the UV cut-off for the U filter, while the PMT produces the red cut-off of the V filter. Only the B filter required 2 pieces of glass, one to define its blueward edge and the other to restrict its redward side.

Later, red sensitive PMTs were produced and 2 more passbands were added - R and I - in the red and infrared. But this introduced problems for the U & B filters in that they had red leaks (i.e. the filters allowed red and infrared light through) and so extra filters had to be introduced to stop this. Additionally, the V filter had its redward side delimited by the PMT, and so another filter had to be added to produce the same cut-off.

The original UBV passbands were defined by Johnson, while the RI passbands now used are the work of Cousins. The systems are usually referred to as such (e.g. Johnson UBV and Cousins RI). There are other passbands available and some have the same names; e.g. Johnson also developed R and I passbands but they are less used by professionals than is the Cousins R and I. Also there is an Eggen R and I system; there is the Stromgren uvby system; Thuan and Gunn developed another system using only interference filters. The list goes on.

Whatever system you use, you must make sure you specify it when reporting magnitude estimates. When doing photometry, it is essential to match these standard passbands as closely as possible. While it is possible to transform magnitudes from one system to another, this is only really valid for "normal" stars - objects with strong emission lines (such as supernovae and peculiar stars like eta Carina) are almost impossible to transform.

CCDs have even more IR sensitivity and so test the red-leak problems of UBVR filters even more than did the red PMTs. It also adds another problem for the I band filter. Just as the Johnson V band was delimited by the sensitivity cut-off of the PMT, so was the Cousins I limited by the IR cut-off of the S20 and GaAs PMTs. The extended IR response of many CCDs means that the I band also needs to be forcibly delimited to properly conform. While it appears that the problem isn't too bad, especially for some amateur CCDs (like the TC-245) it does present a problem for professionals as no suitable glasses exist to do the job. The only solution is to add an interference coating to the glass to provide the necessary cut-off - an option not usually available to amateurs.


Coloured Glass

The original photometric filters were made from coloured glass. Such coloured glass is made in two ways - ionically coloured or colloidally coloured. Without getting too deeply into the mechanisms involved (another way of saying I don't understand it), ionically coloured means that ions of transition elements or rare earths are placed in solution within the base glass and basically absorb (or more precisely scatter) certain wavelength. Colloidally coloured is a more difficult mechanism whereby initially clear glass with certain colourants added is given a second heat treatment to activate the colourant; it produces different characteristics to ionically coloured glass.

Schott is one of the leading makers of coloured glass (and many other types of glass, too!) and has become the standard for photometric filters. Their filters are usually specified by a two character code and then some numbers. The letters relate to the general colour of the glass, e.g. BG, UG, RG, etc. One or two numbers following are often just a code for the particular design e.g. BG39, KG4; while three digits are usually the wavelength cut-on of a filter, e.g. GG495 has a sharp cut-on wavelength of 495 nm.

Schott glass grouping.
Code

Description

UG

Black and blue glasses, Ultraviolet transmitting

BG

Blue, blue-green and bandpass glasses

VG

Green glasses

GG

Yellow glasses

OG

Orange glasses

RG

Red and black glasses, IR transmitting

NG

Neutral density glasses

WG

Colourless, ultraviolet absorbing glasses

KG

Virtually colourless glasses, IR absorbing

FG

Blue and brown colour temperature conversion glasses

Sometime near the beginning of 1995, Schott (in Germany) had a major review of its coloured filter operations. The end result was that they have deleted some 19 glasses from their catalogue (some of interest to astronomers for making photometric filters) and their prices have increased significantly. This affects Schott factories throughout the world, not just in Germany.


Assembling Filters.

If you've made your own telescope, then assembling filters will prove to be very easy. The technique that I have used is as follows. (The same basic technique can be used to cement refractor objectives, or any lens for that matter.)

Equipment

You'll need the following equipment:

I use Schott coloured glass as defined in the tables at the end of this page. The filters for my home CCD camera are 25mm diameter, 4mm thick.

The "glue" is supposed to have the same refractive index as the pieces of glass so that it becomes invisible when applied to the glass. Don't try to use "super-glue" or some other ordinary glues as they will not have the right refractive index and introduce unwanted effects.

There are various types of optical glues, the simplest of which is Canada Balsam. It is cheap, readily available, and non-permanent; the latter point being very important if you're experimenting. You'll need an oven of some sort to heat the Canada Balsam to make it flow more readily - but only low heat is used, about 70°C. (I also warm the pieces of glass before use to aid in keeping the Canada Balsam fluid when it is placed on the glass.) Xylene is used to thin the Balsam should it have dried out and become hard, and as a solvent to clean up the excess Balsam squeezed out when the filters are assembled. I don't know how easy Xylene is to obtain and may prove to be a stumbling block when using Canada Balsam. (Xylene is a nasty substance and care should be exercised when using it.) I think that you could survive without Xylene but it might take a little more effort to clean up afterwards. There are possibly other solvents but I know nothing about them. (Suggestions for alternatives are welcome.)

There are two other types of glues that are of interest; epoxy glues, and ultraviolet (UV) setting glues. The UV setting glues are the easiest to use for cementing "clear" glasses together. The glue is a liquid that is easy to work with and doesn't harden until intense UV light is applied. This can prove difficult to use for coloured glass filters as they usually block out the UV and so you can't harden the glue. (I'm told that with an intense enough source then what little is transmitted through the filters will be enough to make the glue harden, but I've not tried it.) The epoxy glues are the best - if you know what you're doing. They are permanent glues and so no mistakes are allowed, but they offer the best transmission characteristics. The trouble is that they typically set in only a few minutes after mixing so you have to be quick in applying them and cleaning up any spills before the mistakes are permanent. (Now you know why I still use Canada Balsam!)

I have access to a nice oven which allows me to control the temperature of the Balsam and glass, and has a suitable work area around. I guess a kitchen would work equally well.

Finally, a few blocks of clean wood or steel are used as stays to keep the filters from slipping about while the glue is setting (cooling in my case).

Technique

While the oven is warming up, I lay out all the equipment for easy access. I lay the pieces of glass on some soft tissue. (I use facial tissue which is very bad practice because it leaves lots of lint behind when used - but at least it won't scratch the glass. I have filtered, dry air on tap so I can just blow the lint off when necessary.)

Only one drop of glue is used per piece of glass so you won't need much. The Canada Balsam that I have comes as a fairly viscous fluid and I find that heating it up to about 70°C makes it easier to use. Decant a small amount of the Balsam into a clean container (one that can take the heat) and place it in the oven. Make sure that it doesn't stay there too long (like an hour) as it will dry out and be very difficult to use. A few drops of Xylene will help if it has dried out too much.

While it is warming up I clean the filters with iso-propyl alcohol to remove any trace of finger marks etc. It is only necessary to do one side as the other side is likely to be well fingered by the end of the proceedings. Lay the glass on something in the oven with the clean side up (and lint free) to warm. Some glasses can change their characteristics if heated too much so make sure that the oven doesn't get too hot.

The glue (in whatever form you use) must be applied in one clean drop in the centre of the filter. When the second piece of glass is applied the drop will spread out in a circle until it reaches the edge. If there is too much glue then it will just run over the edge and create a mess. This isn't disastrous, although a lot too much will take a while to clean off. But if there isn't enough glue then it won't reach the edge and you'll have to clean the whole lot off and start again. With Canada Balsam (and UV setting glues) this isn't a problem, but with epoxy glues you're in trouble. The trick is to not have any air bubbles in the drop else they will never squeeze out and the join will be flawed. (It's not that the join will be weak, rather the tiny bubbles will allow spurious reflections from the uncoated interface - remember that the glue is essentially invisible).

I find that one drop is easy to gauge for the 25mm diameter filters that I usually use, but I've also done 2-inch squares and they are considerably harder. I've never tried anything larger and so I don't have any idea of the necessary technique (but I may soon have to find out as I'm looking at making 160mm square filters for a new prime focus camera to go with a proposed 8k × 8k CCD mosaic.)

To apply the Canada Balsam I use a clean screwdriver with about a 3mm blade. I find that this delivers a nicely sized drop. Dip the end into the balsam and pick up a suitable drop - but don't get any air bubbles caught in it. Lift it out and stop any streamers from getting too long, then quickly transfer it over the centre of the glass and allow one drop to fall into the centre. When it looks like the right amount quickly remove the applicator, not allowing any streamers to fall onto the glass. Quickly place the other piece of glass centrally on the drop and squeeze down, making sure that there are no air bubbles trapped between the glass. Continue squeezing - and you may have to squeeze quite hard - to get the balsam to reach the edge. Reject any thoughts of laterally moving the glass to help the glue along as this will do more harm than good. Just squeeze.

You should be able to see if the glue has spread completely over the area and there are no entrapped air bubbles. Inspection over a bright light will help. I've only ever had difficulty seeing through the UV filters (particularly a Washington C filter which was 8mm thick when assembled), all the other filters are easy. If free of bubbles and the glue has uniformly reached the edge, I then place a weight on the filter and place v-blocks around the filter to keep it perfectly centred. It is then allowed to cool.

I don't believe that Canada Balsam ever fully sets, at least not when it's sandwiched between two pieces of glass well away from the atmosphere. It will cure very slowly from the outside inwards, but if the filter gets hot you may find that the pieces can move. Eventually, perhaps a few years, it will cure enough that this won't happen, but take care until then. I've sometimes tried to help this process along by heating the completed filter in the oven and allowing it to cool again. I do this with the weight on and the filter restrained by blocks to stop it from moving. This process certainly squeezes some more of the balsam out and does seem to help.

Once cool - a few hours or perhaps the next day - and it's time to clean the excess balsam off. You'll find that the new filter is quite a mess, both top and bottom (and also your fingers if you're like me). Xylene and a bit of rubbing will remove it all, but take care and have patience. You may find that for the first few months the balsam will leak out a bit and you'll have to clean it off. Despite this slightly worrying aspect of Canada Balsam, none of my filters have ever shown signs of separating.

The B filter is made from 3 pieces of glass so you have to go through this process again, but I find it necessary to completely clean the filter before adding the next piece of glass.

The only difference between using Canada Balsam and UV setting glue is that UV setting glue is a lot easier to use. Its consistency is more like water and so can be applied from a dropper, and you've got lots of time to work the glue around and spread it over the glass. You can then clean up any spills before blocking the glass in place. Once ready you hit it with a dose of UV light (while wearing your UV safety glasses, of course). It is much cleaner and a lot less hassle than Canada Balsam, but it is essentially permanent - no mistakes allowed.

I've never used any epoxy glues for filters so I can't comment.


Designing Filters.

Warning! Some of the following information is WRONG, but I've left it here for demonstration purposes. The error is due to the published QE curve for the TC-245 chip being totally wrong. It would appear that the QE of a TC-211 and TC-245 (and a few other chips) is exactly the same as for a TC-241. So the design examples tweaking the filters to work better for the TC-245 chip, while correct for the published TC-245 QE curve, are in reality pointless. The "standard" filters work well for the TC-245 CCD.

Designing filters is a simple matter of mixing and matching different filter glass of known properties until you get the desired result. When I started doing this many years ago I wrote a little computer program to help me along, using the Schott filter catalogue as my set of glasses. This program has produced the graphs which follow. (If anybody is interested in trying out this program then contact me.) The only controls you have when designing filters are the type of glass and its thickness. The range of glasses usually means that by juggling the two parameters you can get close to what you need - although there are always some aspects which are impossible. I'll try to show how I designed my standard UBVRI filters for use with CCDs.

Below is a diagram showing the relative wavelengths of the Johnson U, B and V and the Cousins R and I wavebands (data from Dr. M.S. Bessell of MSSSO, from the Publications of the Astronomical Society of the Pacific, Volume 102, pages 1181-1199). These are what we would like our filters to produce.

Unfortunately, it's not that simple. The above curves are what the original filters plus detectors produced. While we can use exactly the same glasses that defined the original wavebands, they won't reproduce the same effective passbands with different detectors. To reproduce the desired UBVRI wavebands on our CCD, we must know how it responds to light of different colours and modify the filter glass combinations to suit.

Most CCDs readily available to amateurs have a different response to light than those available to professionals. Below are QE curves (quantum efficiency - the detector's response to light of different colours) for 2 CCDs. The upper curve is for the first CCD we bought at the AAT, an RCA device (512 × 320 pixels of 30 microns); while the lower curve is a Kodak KAF-0400 (768 × 512 pixels of 9 microns) now gaining popularity amongst amateurs. My standard filters were originally designed for the RCA device but had proved perfectly interchangeable for many other CCDs with which I had worked.

There are very significant differences between the CCDs. The RCA device has obviously better QE than the other device and its response also varies quite smoothly. The KAF-0400 has dramatically less response - essentially zero in the UV, very poor in the blue, and marginal in the yellow, green and red region. In fact, it isn't until IR wavelengths that the two curves come together. The KAF-0400 also has a very uneven response curve showing many sharp rises and falls in sensitivity. (NOTE: my curves are on a linear scale whereas most manufacturers curves are logarithmic.)

I had thought that all CCDs were similar to the RCA device and it came as a shock to discover that they were not. What is worse, there are differences amongst all the devices that are typically available to amateurs. Here are the 3 major devices from Texas Instruments of interest to amateurs, the TC-211 (SBIG ST-4, Cookbook 211, and many others), TC-241 (SBIG ST-6), and TC-245 (Cookbook 245 - of most interest to me!).

(As a side note, the published QE curve for the TC-211 chip is wrongly labeled. Where the graph says 50% it should read 70% - a significant change on a logarithmic scale.)

While it would be nice to understand why such significant differences exist, it is beyond the scope of this note to pursue it. I will make a point here that may be of interest to users of the TC-241 device. The good blue response of this device, I am told, is because of the layout of the gate structures on the chip. Basically, some portion of each pixel is free of the usual metal grids which cross the surface of a CCD. This stops the blue-ish light from being absorbed as in many other devices (e.g. TC-245, KAF-0400, etc). The net result is that the response over the physical pixel varies according to where on the pixel you look. While this is possibly of no concern for the camcorder manufacturers that use the chip, it might produce strange effects if you try to do photometry with this chip on a telescope which under-samples stellar images (like most amateur telescopes). I'd be interested to hear from anybody who uses this chip for photometry whether they have seen this effect.

Not to digress any further, we will look at how the differences between CCDs affects designing filters. For a start, here's a diagram of the ideal V filter and my "standard" filter combination (2mm GG495 plus 2mm BG39). The filter combination has been normalised so that the comparison can be done on a uniform scale.

A pretty good match. But what happens when we convolve it with some CCDs. The following plot results.

The RCA and TC-241 are acceptable, but the TC-245 is somewhat messy and by no means ideal. Can anything be done about it? Yes, but not a lot, mainly because it is the lumpiness of the CCD's response that is the cause. There are only two ways open for the filter designer to tune a design - use different glasses, or change the thickness of those used. Just by changing the thickness of the existing glasses in this case we can get closer to the ideal. Here is 3mm GG495 plus 4mm BG39 with the TC-245, plotted with the ideal case.

Apart from the obvious effect that the new filter is no longer the same thickness as our others, the other problem is that the filter throughput is a little lower than the standard filter. It depends on how close you demand to be and whether you are prepared to pay the price. However, we can remedy the first problem by using different glass in the design. Here is 0·5mm of OG515 plus 3·5mm of GG495; convolved with the TC-245 and compared to the ideal V curve.

Of course this new design is even less efficient, but by less than 2%. It is also a poor design in that it is difficult to work with 0·5mm thin pieces of glass. None-the-less, it does show what is possible by both techniques. The same procedure is followed for the other filters.

Conclusion

See the warning at the start of this section.

To cut a long story short, the TC-245 is such a dreadful chip that only the I filter doesn't need to be changed from my standard designs. The chip has absolutely no UV response so there is no point in getting a U filter; the B band can't properly be done because of the lack of blue-UV response; I've just shown how the V band is distorted by the chips uneven response, and the same happens in the R band.

While the TC-241 has some lumps and bumps, it isn't affected as seriously and the standard filters are fine for it. So too is the KAF-0400, although its UV response is just as bad as the TC-245 and so the U band is out, and the B filter should be the design for the TC-245 and not the standard.

So, having gone through the exercise of proving that my standard filter designs work with most CCDs, I managed to discover that they fail for my own personal Cookbook camera CCD. While I knew already that for peak performance each filter combination had to be worked out for each new CCD, I am amazed at how much the standard design had to be changed to make it work with one chip, while they are perfectly adequate for all other CCDs.


Filter Designs.

As I've just shown (I hope to your satisfaction) the following designs will produce very good matches to the standard photometric passbands on almost all CCDs. The only exception is the TC-245 chip as used in the Cookbook camera, and the B band for the Kodak CCD chips. While it is possible to design better filters for any CCD (particularly the B filter), these are a good compromise for ease of making and using.

This table details the standard filters for most CCDs. I wouldn't recommend attempting the U band unless you have lots of patience and a good high-altitude location, or a UV enhanced CCD.

Recipe for standard photometric filters.
Passband

glass #1

glass #2

glass #3

U

1mm UG1 3mm BG40

B

2mm GG385 1mm BG12 1mm BG39

V

2mm GG495 2mm BG39

R

2mm OG570 2mm KG3

I

4mm RG9

These are the filter recipes that I would recommend for the TC-245. The U filter has been removed as this chip has no UV response. The same comment applies to the Kodak chips (KAF-0400 and KAF-1600), and also this B filter is the one to use for these chips instead of the standard B.

Recipe for photometric filters for TC-245.
Passband

glass #1

glass #2

B

1mm BG1 3mm BG39

V

0·5mm GG515 3·5mm BG39

R

2mm OG570 2mm KG5

I

4mm RG9

Here are the above filters convolved with the TC-245. Not perfect, but good enough.


Some Professional CCDs.

And for those with aspirations for better CCDs, here is the QE curve for the SITe 512 CCD. There are 3 curves shown. The lower plot, labeled frontside is quite similar to the CCDs typically used by amateurs - i.e. it has no UV response and its overall QE is little better than 40%. This is because the light has to go through the electronic structures that make up the CCD chip, thus causing much light to be lost. But there are ways to improve the situation.

To start, turn the chip over and look through its back where there is just the silicon. The overall QE is significantly improved and there is less lumpiness. To improve the blue response the chip's substrate can be thinned, and a final boost is achieved by applying an anti-reflection coating. That's what the upper 2 curves show. Thinned and back illuminated with 2 different AR coatings, a standard and a UV enhanced one. Of course it's not that simple else we would all have one. There are problems thinning the chip, and when it is successfully thinned you have to support it so that it doesn't bend.

I'm told a thinned, back illuminated and standard AR coated chip can be had for about US$1000. And then you'll need the support electronics...

Contact SITe for further information.


home back to ATM


Page last updated 1996/07/11
Steven Lee