Why 12-bit delays sound like that

Ask why an early-80s delay line sounds the way it does and you’ll get answers about “warmth” and “analog glue.” Neither is quite it. The character is arithmetic.

A 12-bit converter has 4,096 codes to describe a signal that a modern 24-bit path describes with over sixteen million. Every sample gets rounded to the nearest of those 4,096 rungs. The rounding error — quantization noise — isn’t uniform hiss; its spectrum tilts with the signal, so it rides on the sound rather than under it. On a repeating delay, each pass re-quantizes, so the error compounds tail over tail. That’s the grit that “warmth” is standing in for.

The companding schemes of the era complicate it further: the signal is compressed before conversion and expanded after, which moves the quantization floor around with level. Quiet passages get more effective resolution; loud transients get less, and the top end frays first. This is why the runaway on a self-oscillating delay doesn’t just get louder — it gets dirtier in a specific, level-dependent way.

A capture that “fixes” any of this is no longer the instrument. The whole reason to model the converter path rather than sample a clean impulse is that the error is the signature. Preserve the rounding, preserve the companding curve, preserve the way the tail degrades — and the box sounds like itself. Average it away and you’ve built a reverb that happens to repeat.

Field notes are where the technical detail lives. The registry pages stay in the liner-note register; here we can say “quantization noise” out loud.