Dither

Dither

Dither is a technique used in digital signal processing, particularly in digital audio and digital imaging, to reduce the perceptibility of quantization errors. Quantization is the process of mapping a large set of input values to a smaller set, such as rounding a continuous range of values to a discrete representation. This process inevitably introduces some error, known as quantization error. Dither involves adding a small amount of random noise to the signal before quantization. This noise is carefully controlled and has a specific statistical distribution. The added noise helps to randomize the quantization error, effectively spreading it across the frequency spectrum. By doing so, dither reduces the audibility or visibility of the quantization error, as the human auditory and visual systems are less sensitive to noise that is spread out over a wide frequency range. Dither is particularly useful when reducing the bit depth of audio or color depth of images, as it helps to maintain a higher perceived quality despite the reduced resolution. Different types of dither exist, such as rectangular probability density function (RPDF) dither, triangular probability density function (TPDF) dither, and shaped dither, each with its own characteristics and applications. The choice of dither type depends on factors such as the desired noise floor, the distribution of quantization error, and the target medium. Dither has played a crucial role in the development of digital audio and imaging, enabling high-quality reproduction despite the limitations of digital quantization.

noise quantization error digital audio signal processing bit depth imaging frequency spectrum shaped dither TPDF RPDF

— John Armstrong