Dithering effect on sine wave

[onlyoneme]

Is it normal that adding 32 bits dithering to the sine wave saved as 24 bit PCM and played over spdif gives different measurements than 26 bits dithering or no dithering at all? All in the digital domain.

 

[John Mulcahy]

Yes. REW generates signals using 32-bit floats. Their integer-equivalent precision is 25 bits but lower level dither could have an effect on small amplitude values. The dither selection needs to correspond to the sample width to have the desired effect of linearising the effects of quantisation, 24-bit data needs 24-bit dither.

 

[onlyoneme]

I've always used 24 bit dither for 24 data and I've been surprised today when I've tried to use 32 bit dithering because I've expected measurements equal to non-dithered signal. And I got a result like 24 bit dither but with lowered noise level and slightly better distortion calculations instead.

 

[onlyoneme]

I guess I do not understand what "The level at which the dither is added is controlled by the sample width selector to the right of the check box." means exactly considering 32-bit float internal REW precision and 24 bit final file sample rate for example.

 

[John Mulcahy]

When dither is applied to a 24 or 32-bit output format REW does two kinds of dithering. The main one is TPDF dither at 2 lsb, but because the source data is float some float dither is also applied as otherwise artefacts can remain. Dithering signals below the lsb will have variable results depending on the extent to which the signals self-dither, FFT-locked tones or tone frequencies with non-integer sample sequence periods will be much better than sequences with small sets of sample values. Completely removing dither artefacts for signals with few distinct sample values requires one step higher dither than would otherwise be used.