IT lexicon › Hardware › Noise shaping

Noise shaping

Hardware På svenska → Updated: 2026-10-27

Pushing quantisation noise into the frequencies the ear cares least about, so fewer bits sound like more.

When a 24-bit mix is reduced to the 16 bits of a CD, every sample is rounded and the rounding error becomes noise. Dither turns that error into smooth, harmless hiss instead of distortion that tracks the music. Noise shaping goes a step further: each sample's error is fed back through a filter and subtracted from the next, which shoves the noise towards the top of the spectrum. There is more noise in total, but it sits around 15–20 kHz where hearing is indifferent, and the midrange ends up quieter than plain rounding would allow.

In the 1990s it turned into a small arms race among mastering houses, with Sony's Super Bit Mapping and POW-r as branded curves. The same principle is how sigma-delta converters work: a single bit at a very high rate, with all the noise kicked far above the audible band.

← Back to the lexicon