bandlimited bitcrusher

[itoa]

Any suggestions on how band limited bit-reduction can be done besides heavy oversampling?

[FabienTDR]

When reducing word-length, all you need to do is to dither properly. Oversampling is pointless, at least as long you don't oversample by crazy amounts (64x and more) and go the pulse-width modulation/DSD way (which really isn't reasonable in DSP).

On a side note, how does white-noise look like when it aliases (in the Nyquist sense)? The result is white-noise again.

[itoa]

Thanks I gonna take a look at the dithering then. I like this effect, but want to try a cleaner version.

[FabienTDR]

Properly dithered truncation has no side-effect beside an increased noise floor.

No matter if you're looking at bit-crushers or sample-rate decimators, the "creative" artefacts are always a product of a "wrong idea".

Maybe you want to try little more than no dithering (which would be still technically "wrong" and thus create these "creative" effects, but at a lower level).

[itoa]

Theoretically I can perform interpolation to determine precise "bit crossing point" and then use band limited stairs.

[earlevel]

Thanks I gonna take a look at the dithering then. I like this effect, but want to try a cleaner version.

This explains it:

https://www.youtube.com/watch?v=zWpWIQw7HWU

You'd want TPDF dither in this case. That's adding white noise (random value) at plus/minus half the level of the lowest bit that you intend to keep, then do it again (with another random value), before rounding or truncation.

Alternatively, you can just generate one random number per sample, and it to you sample, and subtract the previous random number (give a high pass-filtered noise, slightly faster).

Dithering will mainly save you from gritty artifacts when fading the levels when using limited resolution.

Some more articles:

http://www.earlevel.com/main/category/d ... tal-audio/