High performance or "scientific grade" CCD cameras are designed to operate under extremely low-light conditions with high linearity and accuracy. Currently CCD sensors, rather than CMOS sensors, are used due to their higher performance. This performance difference is partly due to the relative maturity of the sensors, which have been on the market for many more years.
The biggest limitation to the performance of low-light cameras is ”dark current.” It turns out that light is not the only thing that can excite electrons in the detector. Heat – which is simply the vibration of individual atoms – can occasionally bump an electron up into the conduction band. The warmer the array is, the more likely this is to happen; dark current typically doubles with every 5 to 8 degrees Celsius increase in temperature.
This causes two problems. First of all, if the dark current production rate is high enough, it will swamp the detector. Each pixel has only a limited capacity to store electrons (the ”well capacity”). For many sensors, dark current can fill the well in a matter of seconds at room temperature.
A second problem arises because of the randomness of the dark current. Although it accumulates at a steady average rate, the dark current is quite random. The more dark current that accumulates, the more random noise it contributes to the image. Noise makes it difficult or impossible to detect the signal – photoelectrons produced by photons. In order to detect an object reliably, you need at least three times as much signal as noise. For low-light levels and long exposures, some means of reducing dark current is essential.
Since dark current increases with temperature, the obvious solution is to cool the sensor down. Scientific-grade cameras typically operate at freezing temperatures – typically anywhere from 0º C to -70º C. Some CCD sensors are specially designed to minimize dark current and can operate near the upper part of the range; others require more cooling. The cooling may be provided by liquid nitrogen, or by more a convenient but less powerful thermoelectric cooler (TEC).
Most cameras are not cooled enough to completely eliminate dark current. Pixel-to-pixel variation in dark current can cause a large amount of degradation to the image. Correcting this effect accurately requires a precisely regulated CCD temperature; therefore high performance cameras are not only cooled but precisely temperature regulated.
Read noise is another important performance specification. Every time a pixel is read, a small amount of random noise is added to the signal. Assuming quiet electronics, the majority of this noise originates in the sensor itself. There is often a compromise between readout speed and read noise; as signal is clocked out of the CCD faster, the bandwidth of the amplifiers must be expanded. The wider bandwidth allows more noise through. As a result, some high performance cameras are relatively slow to read out. Some models have multiple readout speeds, and in some cases the user can select between different readout electronics, depending on whether they need more speed or lower noise.
Large well depth is also important for good performance. For highly accurate measurements, very good signal-to-noise ratio (SNR) is required, meaning that more photoelectrons must be detected. A larger well allows more electrons to be collected, resulting in more accurate measurements. Large well depth also results in greater dynamic range; that is, the range of brightness over which the sensor will operate. At the bottom end is the noise floor, below which the SNR is inadequate for detection. At the top end is ”saturation”, where the well fills up and may even start to bloom (bleed into adjacent pixels). A high dynamic range is extremely important for imaging astronomical targets, which have extreme differences in brightness levels.
Recording high dynamic range data also requires an A/D converter that produces a large range of numbers. Typical scientific-grade cameras produce 16 bits of data, or 0 to 65535 ADUs (Analog to Digital Units). A camera with very small pixels and therefore small wells might only be able to produce 8 bits worth of data, even if sampled with a high resolution A/D converter. Many cameras with 16 bit converters can actually only produce 11 to 13 bits of "real" data; the remaining bits are simply sampled noise. That said, stacking images can give you additional bits of information; every time you quadruple the number of images stacked you add another bit worth of real data.
The images produced by CCD cameras are not perfect. Each pixel in each camera tends to produce a particular voltage offset, rate of dark current production, and sensitivity to light. These flaws can be very accurately corrected using calibration frames, one each for offset (the bias frame), dark current (the dark frame), and sensitivity variations (the flat-field frame). Calibration makes a huge difference to the accuracy and sensitivity of the camera; therefore calibration is also one of the most important functions of MaxIm DL.