Quote:
Originally Posted by UnderTow  I am still not convinced this is true. I still believe this has more to do with the accumulators in the processors in the same way that 32 bit float operations go to 80 bit accumulators. Saying a 32 bit float DAW is 80 bit isn't really true.
I read a lot of the comments on the the SAW forum and I found the developer deliberately vague about all this so I don't trust it. I could be entirely wrong of course.
Alistair |
Allistair -
I've never seen Bob Lentini deliberately vague about this. In response to inquiries on his forum he's spelled out explicitly what it does in some posts, and in direct phone conversation with him he's also confirmed and clarified some points for me.
to wit:
* Multiples and divides are in fact done using 64bit double precision fixed point math.
This returns a full 32bit DWORD (truncated, not dithered, from the 64bit) before being handed off to the next signal process.
* The built in eq's use 64bit floating point math however. Originally fixed point math was used for these as well but then he found it was causing some DC offset in some operations with this method so he changed the algorithm for these.
* When a VST plugin is placed in line the 32bit integer figure has to be converted to 32bit float first, and then the plugin handles the math at whatever maximum allowable bit depth it has been coded at. The returned figure needs to be converted back again from floating point to integer.
* SAW native plugins from RML Labs all use fixed point math except for the Reverberator which uses 32bit floating point as it is legacy code created by another developer
* SAW native plugins from JMS Audioware afaik 64bit floating point math.
* SAW native plugins from Sonoris use 64bit double precision fixed point math (except for some filters which use 64bit floating point) and return a full 32bit DWORD with TPDF dither used when requantizing this from the 64bit figure.
* Unless the native dither is activated or a dither plugin is placed just prior to output then the 32bit fixed point figure will be truncated to 24bit prior to hand off to the DAC.
So I actually think Bob Lentini has been more forth coming about what's actually going "under the hood" with his DAW app than nearly any other developer with the exception of Digidesign, who went to the length of issuing a white paper.
Anyway - SAWStudio is definitely not for everyone as it commands are a little different than other DAW's, and it's price is in the higher range for DAW apps - but I've been using some version of it since 1994 so that I'm super quick on it, it's incredibly stable, the support is excellent, and do find that the results after processing is done with it remain very high quality.
An older article of definite interest to this thread discussing advantages of fixed point and floating point math is at ->
Fixed-Point vs. Floating-Point DSP for Superior Audio
Best regards,
Steve Berson