Rounding errors in DSP

[kylen]

Here's an interesting thread on the topic: follow all the links too, especially to the DAW SUMM project CD:

http://forum.nuendo.com/forum/Forum12/HTML/001138.html

Cool ! Thanx it looks good - might be a 3 cup of coffee read - at least

 

[Bulls Hit]

What is the mathematical difference between rounding error and dither, except for the position of the least significant bit?

I think it has to do with the rounding errors accumulating with repeated operations, and the distortion they introduce is inharmonic, whereas with dither it's only done at conversion time, and even though bits are turned on randomly, the 'randomness' can be controlled to minimise the effect on frequencies the ear is sensitive at

 

[masteringhouse]

I think it has to do with the rounding errors accumulating with repeated operations, and the distortion they introduce is inharmonic, whereas with dither it's only done at conversion time, and even though bits are turned on randomly, the 'randomness' can be controlled to minimise the effect on frequencies the ear is sensitive at

Dithering may be done other than conversion time. It should be done whenever there is a reduction in bit depth. For example when going from a 32 bit float or 48 bit fixed buss to a 24 bit plugin or external device.

That's one of the reasons (I assume) that the Waves plugins have a dithering option on their mastering plugins and why Pro Tools created the dithered mixer.

 

[darnold]

I dont see how the bit will change lower any other time than on conversion at the end. If your running at 24 bit in nuendo 2 its going to be 32 bit float on all the plugins, fader movements, etc. I dont think it ever really changes until you change it.

I guess the question is do you need dithering when converting up? Cause if so then that would mean you would need dithering at every single step in the chain.

Danny

 

[masteringhouse]

I dont see how the bit will change lower any other time than on conversion at the end. If your running at 24 bit in nuendo 2 its going to be 32 bit float on all the plugins, fader movements, etc. I dont think it ever really changes until you change it.

I guess the question is do you need dithering when converting up? Cause if so then that would mean you would need dithering at every single step in the chain.

Danny

Here is one example. Going from the DAW to an external 24 bit digital processor.

Many plugins also convert to 24 bit on output even though internally it is processing at a higher depth. When going from the internal depth to 24 bit some plugins may be dithering for you automatically without you even know ing it, or not dithering at all which is not good.

No you do not need dithering when converting up, only when going to a lower bit depth.

 

[kylen]

Here is one example. Going from the DAW to an external 24 bit digital processor.

Many plugins also convert to 24 bit on output even though internally it is processing at a higher depth. When going from the internal depth to 24 bit some plugins may be dithering for you automatically without you even know ing it, or not dithering at all which is not good.

No you do not need dithering when converting up, only when going to a lower bit depth.

I was worried about this at one point so I inserted a bitscope at various places just to see what was happening. Here's a free VST (shareware) one called bitviewer:

http://www.tobybear.de/p_utilbag.html

Any others that go past 24 bits?

 

[darnold]

oh i see what you were saying. I guess i just misunderstood before. That makes sense.

danny

 

[masteringhouse]

I was worried about this at one point so I inserted a bitscope at various places just to see what was happening. Here's a free VST (shareware) one called bitviewer:

http://www.tobybear.de/p_utilbag.html

Any others that go past 24 bits?

Nice link Kylen!

What were the results of your testing?

 

[kylen]

Nice link Kylen!

What were the results of your testing?

Well I can sleep better!

I have it plugged in as the last plugin insert in Sonar4 right now and I can see that if all the effects in that track are turned off the audio stream is 24bit (the wave file is 24bit). When I begin switching on the 32bit/64 bit plugs then the bits start to fill up the rest of the 32 bits so any calculation made is now at 32bit [float] or whatever the plug resolution is.

I've also used it to make sure the dither switch actually worked on some plugs and I guess there are other things you could do with it to judge loudness ( ).

It's for those of use who like to see the meter swing!