Ultra-cheap astronomical CCD imagery

- Modifying Logitech's Quick Cam 3000 Pro modified for long-exposure operation

Inspiration

I was curious.

Once the province of mountain-top astronomers only, CCD imaging sensors are now commonplace in better "web cameras" one attaches to your PC. Some talented enthusiasts banded together to form the QuickCam Unorthodox Astronomical Imaging Group ( QCUAIG), and started tinkering with pressing these inexpensive devices into service for astronomical imaging. This amounted to making modifications to the control circuitry of the cameras to enable longer exposures, and writing custom control software. One in particular, Steve Chambers in the UK, is chiefly responsible for the modification for the "Vesta Pro", "ToUCam", and "QuickCam 3000 Pro" - type web cameras built around SONY's HAD-type interline ICX098BQ CCD. Two others, Keith Wiley, and in Switzerland, Martin Burri, then performed Steven's modification to great success. There are likely many more.

Caveat Emptor

PC webcams are not substitutes for scientific CCDs - at least, where "scientific" means the ability to perform reproducible photometry. The reasons for this are several, and are chiefly due to the fact that webcams were never designed to do photometry - but simply to produce pictures pleasing to the human eye. The human eye is much more sensitive to brightness variations (Y) than colour variations (UV space), and hence - as in the world of colour television - webcam signals are usually represented in YUV-space. For current USB1 webcams, each quantity (Y, U and V) is represented by only 8-bits which are sent sequentially (this is to get the 640x480 pixels in an image through a 1.1Mbit/s Universal Serial Bus port fast enough for several frames per second). It should also be mentioned that right in the physical CCD array itself, individual pixels are sensitive to only one of the red, green or blue bandpasses courtesy of built-in primary colour filters in a Bayer matrix (R, Gr, Gb, and B; there are twice as many green because this is the colour component to which human eyes are most-sensitive). Each pixel of the 640x480 output has a red, green and blue component where only 1/3 of the physical pixels are sensitive to any one component, because the on-board digital signal processor (the large SAA811x IC) interpolates 3x3 pixel neighborhoods in obtaining YUV signals for individual output pixels. Because of this, post-processing sharpening filters are usually required to remove the consequent "blurring" - more transformations. Noise is still an issue, but noise analysis (relatively straight-forward for photometry) is almost hopelessly complex in this instance. Of course, the PC software that drives the whole process usually involves all sorts of automatic gain controls, colour balance controls, and so on. Webcams are not for photometry.

However...

The effective 8-bit-ish output (as the Y or magnitude component is encoded to 8 bits - for arguments' sake we can think of starlight or galaxy light as being chiefly white) can be effectively extended to 12 bits by co-adding 32 images, for example, or 14 bits by co-adding 128 images. "Stacking" many images is the only way to operate webcams so that they're sensitive to very weak signals. To recover weak signal "lost in the noise" of single frames does requires one thing: that the individual frames be recorded at "full gain", so that the noise is preserved. One therefore would want to turn off everything-automatic in the controlling softare, and crank the gain. Additionally, I'd turn off any colour-balance adjustments (gamma, correction for "incandescent vs. white light", etc). Add together enough images that don't inadavertantly "clip" signals weaker than a certain threshold, and the (random) noise should average-out, leaving the sought-after signals. There is an explosion of deep-sky webcam imagery out there proving just how surprisingly-well this method works.

And in principle, any imaging device can be used e.g., to conduct an automated supernova survey where all you're interested in are relative changes within a given image from night to night.

With these ideas in mind, I thought I'd give it a shot [sic]. I decided to pick up a cheap (~$130 CDN in Aug.'02) QuickCam 3000 Pro, and modify it more or less following M. Burri's instructions. Mostly because I was curious, and because this might make a nice, informal platform for planetary imagery with the option for "one-shot color" bright deep-sky. Or as a guide camera, since the pixels are quite small (5.6 microns square). Or as the detector in a ludicrously-cheap robotic telescope.

Modification

A few photos of the modification process. Essentially, the mod amounts to interupting two connections on the QC3k's circuit board with bi-switches (the 4066 IC) that are controlled by pins 1&2 of a PC parallel port. Pull-up resistors (I used 10k) keep the switches closed when the parallel lines are not connected to anything, so the camera can still be used as intended by the manufactuer. If you wish to try this yourself, be aware that this is delicate and precise work at sub-millimeter scales. Sharpen your soldering iron and avoid the coffee.

Results

First light photo, taken using M. Burri's "Cap4Astro" software; bare camera with stock lens and near-IR filter, literally laid on our back deck, facing up more or less at zenith, three 15 second exposures at ambient temperature (roughly 8 degrees C), co-added with Robert J. Stekelenburg's "AstroStack". As this was a test, focus is crude. No attempt was made to subtract a dark or bias frame. Note that the Milky Way is apparent behind Cassiopeia, and the Andromeda Galaxy in the lower right-hand corner (also note the blue on-chip amplifier glow in the upper left-hand corner):

Now that the camera is working, efforts (as time allows) to supress the amplifier glow and the read noise (faintly visible here as horizontal banding) will be implemented. The driver chips and CCD will also be cooled to supress dark current and thermal noise. The aim is to see how far this can be taken, of course.