Quantization of an analog signal produces a step-shaped signal whose number of steps depends on the number of samples. The levels of the digitized signal steps differ from the original signal, which is expressed in quantization noise.
These signal differences between the original analog signal and the digital signal are expressed as noise and are referred to as quantization noise or digital noise. They are smaller the more accurately the original analog signal is reproduced. This means that the closer the approximation to the original signal, the lower the quantization noise.
The quantization noise depends on the sampling rate and the quantization. At high sampling rate and quantization with high resolution, the quantization noise is lower than at low sampling rate and low digitization. This is also reflected in the signal-to-noise ratio( SNR). The quantization noise is calculated from the logarithm of the quantization levels: 20log quantization levels and reduces by approximately 6 dB per bit. According to this, an AD converter with 10-bit quantization and 1,024 quantization levels has a signal-to-noise ratio of 54 dB; with 16-bit quantization and 65,536 quantization levels, the SNR value is 36 dB higher. It thus amounts to 90 dB. The quantization noise is in a certain way signal-dependent. If there is no signal, then there are no quantization errors and also no quantization noise.
AD converters often have nonlinear deviations in the quantization, which show up in the differential nonlinearity( DNL). The DNL value expresses the positive and negative deviations in the individual quantization stages. When considering the quantization noise exactly, the deviations by the DNL value from the ideal stage width must be taken into account.