How Does Subtractive Dither Work

[bhobba]

Hii Guys

I am not an engineer, my background is applied math and computer science, and I normally post over in the QM sub-forum. I have an interest in audio, and know Shannon's sampling theorem and what not. I know what dither is and how it works, but in investigating MQA coding I came across something I have I ever heard about before - subtractive dither. As far as I can see one adds a psudo-random dither prior to digitization then subtracts it after digitization before playback. All I can think of in how to subtract it is start with the same pseudo random seed. But what I don't get is how it improves things - you add and subtract the same thing so you really haven't done anything.

Anyone know what the go is - MQA looks very interesting technically and does sound fantastic, but this one has me a bit confused.

I did a listening session with a very experienced audiophile on a (very) high end reference system yesterday with MQA from Tidal (the album Blue which we both know well) and using just the first unfolding from the Tidal app we were gob smacked and I ordered a full MQA DAC to see what full unfolding does. Its the real deal - many things in audio aren't.

Thanks
Bill

 

[jedishrfu]

I found this article but it may be too simple:

http://electronicdesign.com/analog/dither-can-boost-sampled-data-system-performance-least-10-db

 

[bhobba]

http://electronicdesign.com/analog/dither-can-boost-sampled-data-system-performance-least-10-db

Ahhhh. It helped. What's seems to be going on is its subtracted at analogue or much higher res digital. Then later if you down-convert and or reduce bit depth then add the dither back at the new resolution it will be a little different to if you didn't do it. From what I read elsewhere it is more efficient in apparent depth reduction than normal dithering.

Thanks
Bill

 

[analogdesign]

Usually subtractive dither is done using a single-bit DAC. That way it is inherently linear (when implemented in a differential way) and can be extracted after conversion with much higher precision.

The main problem with subtractive dither is that it uses up some of the dynamic range of the converter. This puts more pressure on the thermal noise performance of the converter but does a lot to relax linearity issues.

Interestingly, subtractive dither can also be used (and often is) as a test signal for self-calibration in the background). The dynamic range issue remains, however.