Recording & signals / A listening lesson

Bit depth measures amplitude choices

Understand quantization without confusing it with sample rate or loudness.

A finite set of codes

An integer PCM representation with b bits has 2ᵇ possible codes per sample. Eight bits provide 256 codes; sixteen provide 65,536. The representation maps those codes to amplitude values over a defined range. More codes permit finer amplitude steps within that range, but do not by themselves change how often samples are taken.

Evidence: 1

Rounding introduces an error

Quantization replaces a value with a nearby representable one. The difference between the original and represented value is quantization error. For a uniform nearest-level quantizer away from overload, its magnitude is bounded by half a step. With undithered simple signals the error can be correlated with the signal rather than behaving like independent random noise.

Evidence: 1

Headroom and noise are different concerns

A signal too large for a bounded output range can clip. Increasing storage precision within the same normalized range makes amplitude steps finer but does not authorize arbitrary peaks beyond that range. Conversely, a quiet signal can occupy few codes in a coarse representation without clipping at all. Bit depth, gain and headroom should therefore be discussed separately.

Evidence: 2

Use the site’s models for the right question

The aliasing lab changes sample timing and frequency, not integer bit depth. Compare its sampling result with this numerical example: if a normalized interval is divided into 16 equal bins, the bin width is sixteen times larger than if it is divided into 256. That calculation explains amplitude resolution without pretending the lab provides an unimplemented bit-depth processor. Browser synthesis commonly uses floating-point processing, which has different storage behaviour from integer PCM.

Evidence: 3

An idealized noise estimate

For a full-scale sine and an ideal uniform quantizer under standard uncorrelated-error assumptions, signal-to-quantization-noise ratio is approximately 6.02b + 1.76 dB. It is not a universal noise specification for every signal or converter.

Keep this in mind

Bit depth controls a representation’s amplitude precision; sample rate controls its timing grid.

Make the connection

When samples tell another story

Use the sampling lab to identify the temporal variable, then contrast it with amplitude-code counts in this lesson.

Open the experiment ↗

Useful words

Bit depth

The number of bits used to encode each amplitude sample in a specified representation. It is distinct from samples per second.

See in glossary ↗

Sources & further reading

These sources support the explanations and model assumptions. Recording credits are listed in the listening panels and the source register.

  1. Quantization and decibels ↗Chris Donahue / Carnegie Mellon University · 2026
  2. Clipping ↗Chris Donahue / Carnegie Mellon University · 2026
  3. Sampling and the frequency domain ↗Chris Donahue / Carnegie Mellon University · 2026
Full source and rights register ↗