In order to take a measurement with a PC or other microprocessor based system, the measurement must be converted into a digital value (since computers only operate with digital signals), by using an analog to digital converter (ADC). Similarly to output a value from a PC requires a digital to Analog converter (DAC). Converting an analog signal to digital value is also called sampling, and multiple samples are needed to represent a variable signal over time, such as sound.

How well the signal is represented by the computer depends on a number of parameters that must be chosen when selecting the converter. The two most important parameters are resolution and sampling speed. The resolution typically decreases as the sample speed increases (for a given price). The most common forms of ADC are Sigma-Delta and successive approximation.

The resolution of an ADC determines how many steps it takes between the minimum and the maximum input value. This is specified by number of bits resolution, such as 8-bits or 12-bits. An 8-bit converter takes 256 steps from minimum to maximum

2⁸ = 256

If you wish to measure a 0-10V signal and you require an accuracy of 0.25%, we will initially consider an 8-bit analog to digital converter

0-10V divide by 256 = 0.03906V steps

Accuracy required 0.25% = 0.025V. Therefore we can see that the size of steps is larger that the accuracy required so it is not possible ensure the accuracy with 8-bits resolution

click image to enlarge

A 12-bit ADC may enable the desired accuracy to be met. 2¹² = 4096 steps.

When we divide 10V by 4096, it equals 0.0024V, which is 10 times smaller that the required accuracy.

A 16-bit ADC will provide 2¹⁶ steps = 65,536, so on a 10V signal we can see steps of 0.153mV.

Please note - these calculations only look at resolution of analog to digital conversion, and do not include errors caused by component tolerance, temperature drift and other adjustment errors that make up measurement accuracy.

Nyquist theorem says we should use a sampling speed of at least two times the frequency of the waveform to be sampled. However by using only two samples per cycle, the only measurement we can make is the frequency of the waveform. We will not be able to analyse signal amplitude, see spikes or even the waveform shape.

Ideally we should sample at least 10 times faster than the waveform to enable a good representation to be recreated by the computer.

click image to enlarge