Analog-to-Digital Converter (ADC)

Each pixel's photoelectrons become a voltage at an amplifier, and the ADC turns that voltage into a whole number: a count, also called a gray level or ADU. In an sCMOS camera the conversion runs in parallel. Every column has its own amplifiers and converters, so many pixels are digitized at once, which is a large part of why these sensors are fast. A CCD, by contrast, passes its pixels one after another through a readout amplifier and converter.

Two settings decide what a count means. The bit count N sets how many levels exist, 2^N: 4,096 at 12 bits and 65,536 at 16. The gain, in electrons per count, sets how much charge one step represents. Scientific cameras typically offer 12- and 16-bit modes, and many add an 8-bit speed mode. A common rule, from Pawley, sets the gain so that one count equals between half and all of the read noise, a kind of Nyquist sampling in intensity. Steps finer than that describe the noise in more detail without adding information; steps much coarser start to merge values the sensor could tell apart.

The ADC then needs enough levels to span the full well at that gain: roughly one to two levels for every read-noise unit, or one to two times the dynamic rangeLoading... (full well divided by read noiseLoading...). With fewer levels, either the top of the well is cut off or each count spans more than the read noise, and the converter rather than the sensor becomes the limit. Rounding to whole counts adds a small noise of its own, one step divided by √12 (rms); at the half-to-all gain that is at most about 0.29 of the read noise.

A worked example. Take an sCMOS sensor with a 30,000-electron full well and 1.5 electrons of read noise. Its dynamic range is 30,000 / 1.5 = 20,000 to 1, or log₂(20,000) = 14.3 bits. A 14-bit converter spread over that well gives 30,000 / 16,384 = 1.83 electrons per count, more than the read noise, so it is coarser than the rule. A 15-bit converter gives 0.92 electrons per count, 0.61 of the read noise, which is inside the rule; a 16-bit mode gives 0.46 and leaves room. These inputs are close to a real camera: one current 6.5 µm sCMOS lists a 30,000-electron full well, 1.4 to 1.6 electrons rms read noise, and a 37,000:1 dynamic range (computed with its median read noise, 0.8 electrons).

So the bit count on the box does not tell you the range you get. The sensor sets the useful range, and the ADC's job is to record it without throwing any of it away. Digitizing in more levels than the rule calls for costs nothing in accuracy and adds nothing either; see bit depthLoading... and quantizationLoading....

Many sCMOS cameras read every pixel through two amplifiers at once: a high-gain channel with the lowest read noise, for dim signal, and a low-gain channel that reaches the full well, for bright signal. Each channel has its own converter; on one manufacturer's camera these are 11-bit converters, and a single channel is offered as 12-bit data. In the 16-bit high-dynamic-range mode, the camera takes each pixel's value from the channel suited to its signal level and combines the results into one 16-bit image that keeps both the low noise and the full well. On cameras built this way, the 16-bit image is a merge of two lower-bit readings, not the output of one 16-bit converter.

The high-sensitivity channel is the high-gain one, not the low-gain one. A single-channel 12-bit mode gives smaller files and, on some cameras and interfaces, faster frame rates. Lower-bit modes can be much faster still: one current camera reaches about 500 frames per second in an 8-bit mode against about 90 in a 12-bit mode, which is why the frame rateLoading... you get depends on the mode you choose. Choose the 16-bit merged mode when bright and dim structures must be measured in the same frame; a 12-bit mode is enough when the signal stays within one channel's range.

Sources: Pawley, Handbook of Biological Confocal Microscopy, 3rd ed., ch. 4 and ch. 10; Vetterli, Kovačević and Goyal, Foundations of Signal Processing, ch. 6; Andor Zyla manual, Dual Amplifier Dynamic Range; Hamamatsu, sCMOS camera synchronization technical note; Hamamatsu ORCA-Flash4.0 V3 specifications; Teledyne Kinetix specifications.