How CCDs Work

Imagine you have a large paddock and you want to measure the rainfall on it. For the moment we have to assume one thing - rain doesn't fall evenly, and in fact its the pattern of the rainfall we're trying to measure (that's one large paddock!)

One approach would be to divide the paddock up into squares and measure the rainfall in each square. In fact, we could place a very large number of buckets (say I million or so, in a 1000 x 1000 grid) at regular places across the paddock, and measure the amount of rain falling into each.

Of course, I'm not really interested in rainfall (though I do swear occasionally at the clouds!) But the situation I've started to describe above is very similar to the basic idea behind a CCD. Charge coupled devices, or CCDs, are the sensors inside modern digital SLRs which are fast becoming one of the mainstream pieces of amateur astronomical equipment, revolutionising backyard astronomy in a similar way to the advances they help create in professional circles about 25 years ago. They have a number of advantages over film: linear response, high sensitivity to light and digital output being among the more obvious. We'll come to a couple of these soon.

Getting back to our rainy paddock, the CCD is also bathed in a constant drizzle of photons, coming from the faint astronomical objects in the sky. Each photon can interact with the silicon in the CCD sensor to produce a single charge (electron) in the device. This process is not usually very efficient (usually only 50-80% of the photons hitting the silicon produce a charge), but it is far better than the case in photographic film (typically only a few percent efficient). This high quantum efficiency is essential in any device designed for detecting low light levels. Like our paddock, the CCD is divided up into collecting areas, known as pixels, which allow charge to be built up during the exposure. To find out how much charge is in each pixel, a form of readout circuitry, and a way of recording the data is required.

First however, consider the buckets (pixels) themselves. Each has a finite capacity before they overflow (known as, strangely, full well capacity, continuing the liquid analogy). The wider the pixel dimensions, the higher the Capacity. Smaller pixels, while sometimes convenient, will fill sooner. When a pixel overflows, it is said to be saturated, and sometimes the charge in it will overflow into an adjacent bucket in a process known as blooming.

So how do we find out how much 'water is in the bucket', so to speak? If we have a team of people, we could ask each one to look in the bucket, measure the amount, and yell it out. With one million buckets, we'd rapidly get a very confused, noisy paddock! A simpler system is to have just one person doing the counting. If a conveyor belt is installed at one end of the paddock, we could pour the water from the end row of buckets into a row of buckets. The conveyor is then activated, and each bucket on the conveyor is weighed to find out how much water is in it. The conveyor, in the case of the CCD, is known as a shift register, and the person weighing the buckets, the analog to digital converter (ADC). Numbers, (bucket weight) are then stored in a computer memory for later processing.

To synchronise the 'bucket brigade' a CLOCK signal is used throughout the chip to ensure that the charge moves from pixel to pixel at the correct time, and is read out correctly. There is a trade off between the speed of the ADC and the amount of internal readout noise, (readout errors), it produces. This ultimately limits the speed at which the pixel charge is measured, and the amount of time it takes to read the data off the chip (and hence the clock rate).

In addition to readout noise, there are several other forms of noise in a CCD. To return to our paddock, imagine a competing farmer holding a sprinkler nearby, attempting to bias our results.

Water from the sprinkler will fill the buckets the same way our rainfall does, and there is no way to tell them apart. If we waited until it stopped raining, we cold measure the signal due to the sprinkler. Or, we could cover the paddock with a tent to keep the rain off but let the sprinkler continue to fill the buckets. There may also be other ways of affecting the 'rainfall signal', such as trees shielding some of the buckets.

In a CCD, too, noise from the chip itself can contribute unwanted signal. In fact this sometimes dominates the data received on the chip! Such dark current is unavoidable due to the temperature of the chip, but can be reduced by cooling the chip, or by carefully measuring the dark current and subtracting it from the sky signal. By taking dark frames and subtracting the data from such an image from a 'dark + sky' frame, we yield the true sky image. The optics between the sky and the chip (similar to the trees) also make variations in the sensitivity of the chip. Dust can shadow light hitting the chip, or scattered light in the optics can also influence the image. This is removed by taking a 'nonsky' image through the optics, known as a flat field. Again this is later subtracted to provide an image with a suitably uniform background.

Already we can see the importance of the digital data provided by a CCD in processing the data. It is possible to do similar processing with film (e.g. unsharp masking) but the precision of a computer allows the data to be processed such that the image is a true numerical representation of the light hitting the chip. This, together with careful imaging procedures, allows astronomers to make real measurements of object, providing accurate numerical data about an objects brightness and position.