Fixed Point vs. Floating Point

[Oblivion777]

Why are some of you so much against using "floating point"? I understand the theoretical difference, but not so much the practical difference. I had a discussion half a year ago with a very capable sound engineer and he too was against using floating point, but at that time I had no idea what "floating point" is, so naturally I could not understand his arguments (or any other for that matter).

So - assuming you are against - do you then think that HD system will be "better" (sorry I don't have the proper word here apart that one) than HDX since HD uses fixed points whereas HDX using floating?

[lennieh]

I may be wrong, but I think the argument goes like this:

Converters all operate at 24-bit so at the point were you receive audio from the outside world, or present it to the outside world you are limited to 24 bit resolution.

24 bit resolution is more than adequate for human hearing, the range of values you can represent is much greater than our ability to distinguish between them.

Using 32-bit float file formats allow you store an even greater range of values, so you can effectively have files that contain peaks higher than the max allowed by 24 bit resolution.

The antis say that if you need to use the extra range you get from 32-bit file formats then your gain structure is wrong. End of story. You can't directly present that file to your DA converter, so you have to dither it or apply gain reduction to it, so what's the point, you are just introducing more scope for error.

At the really nerdy end there is the fact that floating point arithmetic isn't precise. You can actually get floating point processing to give seriously different answers on different chips, I work as a programmer in my day job mainly on Alpha/VMS machines and have seen simple sums like 100.0 / 2.0 give a result of 49.9999999999999999*

The pros would say it gives you added flexibility when mixing.

At the end of the day, do some blind listening tests and trust your ears

[Bob Olhsson]

There's no reason to record more than 24 bits from analog.

Where 32 float comes in is when you want to create intermediate files using DSP that you are going to be applying additional processing to later on. Rendering a virtual instrument using Logic and then importing it into Pro Tools is a good example of when 32 bit float file support would be useful.

[jjnssn]

some information I've gathered around this subject...

http://www.rane.com/pdf/old/note153.pdf

http://www.jamminpower.com/PDF/48-bit%20Audio.pdf

http://akmedia.digidesign.com/suppor...ixer_26688.pdf

[dubpimpjuice]

It could be me, but lately the floating point/fixed point issue has been confused on the arrival of the new HDX systems.
The replies have been correct, but what I think the OP is talking about is the summing depth. There are entire threads where people ask about this and replies will be talking fixed point summing, when the OP was talking floating point files and vice versa...
My understanding is that as far as summing (which was the big change IMO for protools), the old HD's and MIX systems used 48bit fixed for summing and the new HDX hardware will use 64 bit floating point summing. There is a lot of material on the web about this difference, but my final conclusion was that 64 bit float will be sonically adequate, if not superior to 48 bit fixed point.... (we are talking summing here)
As far as the 32 bit float file option, it's great in that it gives you the power to save the file exactly how the cpu sees it (without all of the usual truncation/ and or dithering)
Thoughts?

[jjnssn]

Originally Posted by MADIrouting

some information I've gathered around this subject...

http://www.rane.com/pdf/old/note153.pdf

http://www.jamminpower.com/PDF/48-bit%20Audio.pdf

http://akmedia.digidesign.com/suppor...ixer_26688.pdf

Originally Posted by dubpimpjuice

It could be me, but lately the floating point/fixed point issue has been confused on the arrival of the new HDX systems.
The replies have been correct, but what I think the OP is talking about is the summing depth. There are entire threads where people ask about this and replies will be talking fixed point summing, when the OP was talking floating point files and vice versa...
My understanding is that as far as summing (which was the big change IMO for protools), the old HD's and MIX systems used 48bit fixed for summing and the new HDX hardware will use 64 bit floating point summing. There is a lot of material on the web about this difference, but my final conclusion was that 64 bit float will be sonically adequate, if not superior to 48 bit fixed point.... (we are talking summing here)
As far as the 32 bit float file option, it's great in that it gives you the power to save the file exactly how the cpu sees it (without all of the usual truncation/ and or dithering)
Thoughts?

I have the same understanding... both Native and HDX use 64-bit FP for the mix engine... I agree 64-bit FP will have advantages over the 48-bit mix engine (summing) when making use of the additional headroom for processing (overs and calculations)...

As for the OP question...It isn't clear to me that a properly processed HD mix will be theoretically better or worse (or sonically better or worse) than a properly processed HDX mix as long as clipping isn't occurring at some level within the processing/summing. The benefit of the 48-bit mix engine is described pretty good in the attached white paper.. it is similar to the head and foot room with FP. The dither that is used on the 48-bit TDM bus is down around -140db..... so I believe it is practically insignificant.

I can't say that I understand the 32-bit file process other than it is 24-bit with the 8-bits of headroom when saved.

It isn't clear to me if I have 24-bit converters and I have an over going in from the mic-pre because the gain is greater than the input does the software somehow fixing this or allow for the over... it doesn't make sense to me that it could so I get the impression that what gets put to file is only the 24-bit + 8 bits and thus can be processed and saved the same without the dither and truncation as you've described. I wonder if there is also some efficiency gained from start with a 32-bit FP file and the processing that ensues in the DAW...

[jjnssn]

Here is another link comparing Fixed and Floating point DSP (closer to the HD vs HDX question)... written by TI.... I believe AVID is using TI DSP for the HDX...

This paper gets down to the nuts of what is going on and what can be changing with TDM vs AAX..... assembly vs C and the subsequent libraries/code used to do the processing within the architecture of the system...

http://www.ti.com/lit/wp/spry061/spry061.pdf

[Firechild]

Originally Posted by Bob Olhsson

There's no reason to record more than 24 bits from analog.

Where 32 float comes in is when you want to create intermediate files using DSP that you are going to be applying additional processing to later on. Rendering a virtual instrument using Logic and then importing it into Pro Tools is a good example of when 32 bit float file support would be useful.

+1

[SimonN]

Maybe somewhone who has stepped over to PTX can pull up a mix done all ITB on PTHD and post the two mixes here?

[corp]

Wasn't there a time/argument when Avid/Digidesign conveyed that "fixed" was superior to "floating"....?