It is the tiny hiss caused by rounding each sound snapshot to the nearest available value.
Each sample rounds to the nearest step; the leftover gap (+/- half a step) stacks up as a constant hiss whose level drops ~6 dB per bit.
What it is
Faint background fuzz caused by rounding each digital sample to the nearest available step value.
Key facts
Quantization = snapping each amplitude sample to the nearest level on a fixed staircase of steps.
Number of steps = 2 to the power of bit depth. 16-bit = 65,536 levels. 24-bit = 16,777,216 levels.
Dynamic range rule: about 6.02 dB per bit. 16-bit ~ 96 dB, 24-bit ~ 144 dB.
Exact formula: SQNR (dB) = 6.02 x N + 1.76, where N = number of bits per sample.
Quantization error per sample is bounded at +/- half a step (+/- 0.5 LSB); RMS noise = step / sqrt(12) ~ 0.29 LSB.
Each extra bit halves the step size and lowers the quantization noise floor by ~6 dB.
8-bit ~ 48 dB (audibly grainy). CD = 16-bit. Pro recording/mixing = 24-bit.
Dither = tiny added noise (~1 LSB) that randomises rounding so quiet detail stays clean instead of distorting.
32-bit float (DAW internal, ~1500 dB range) makes quantization noise effectively inaudible.
0 dBFS = loudest storable level; noise is constant so it bites hardest on quiet fades and reverb tails.
How it works
A mic signal is sampled thousands of times per second (e.g. 44,100 Hz).
Each sample's loudness is measured against a fixed ladder of allowed values.
The true value almost never lands on a step, so it gets rounded to the nearest one.
That rounding gap (+/- 0.5 step) is the quantization error for that sample.
Millions of tiny errors stack up over time and are heard as a faint constant hiss.
More bits = many more steps = smaller gaps = much quieter, less audible noise.
Real examples
Old 8-bit game/sampler sound: audibly gritty fuzz on sustained notes.
CD audio (16-bit): noise floor ~96 dB down, inaudible on normal material.
A quiet acoustic guitar fade-out where the last reverb tail reveals faint hiss.
A poorly dithered 16-bit master showing crackly distortion on a quiet vocal breath.
Recording a whisper-quiet room tone at 24-bit so the floor stays well below the signal.
How it helps in live sound
Run your console/recorder at 24-bit, not 16-bit, for ~48 dB more headroom against the noise floor.
Set gain so peaks hit around -18 to -12 dBFS: signal stays far above the noise without clipping.
Don't over-attenuate a hot source then boost in the digital domain; you waste bits and lift the hiss.
For multitrack capture (Dante/MADI/USB), confirm the stream is true 24-bit, not down-converted to 16.
Apply dither only on the FINAL bounce/export to a lower bit depth, never between processing stages.
Mix internally at 32-bit float in the DAW so intermediate quantization noise never accumulates.
Everyday analogy
Like a ruler marked only in whole centimetres: every measurement gets nudged to the nearest line, so it is never truly exact and the little errors pile up.
Watch out
Myth: 'more bits = higher resolution / better highs.' Truth: bit depth only sets the noise floor and dynamic range; sample rate sets frequency range. Extra bits lower hiss, they don't add highs.
Fun fact
Counterintuitively, adding a tiny bit of random noise (dither) BEFORE rounding makes the result sound cleaner: it trades harsh quantization distortion for a smooth, low hiss your ear ignores.
Key takeaways
Quantization noise = rounding error from snapping samples to fixed steps.
Each extra bit lowers the noise floor by ~6 dB (SQNR = 6.02N + 1.76 dB).
16-bit ~ 96 dB range, 24-bit ~ 144 dB; capture at 24-bit.
Worst on quiet, exposed passages; buried under loud music.
Dither hides it as smooth hiss instead of ugly distortion.
Aim peaks at -18 to -12 dBFS to stay well above the noise.