Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Lee Buckalew]

Arnold,

Quote:

Marketing blurbs, I fear. People seem to be running around casting stones in ignorance. The whole fixed point/floating point digital audio battle has been fought once already in the realm of DAW software. The outcome was that different developers make different choices, but all competent developers end up making stuff that works very well, thank you!

I am not relying on marketing but on my own listening experiences in the recording and live sound world.  I also listen to the opinions of other engineers/producers.
I do a lot of classical recording work, some which was a part of the grammy process this year.  One of the great things about audio is the diversity of people, equipment, technique, etc. and that we can agree to disagree and still have sounds that our audiences like.
In my experience and the experiance of many other recording engineers (certainly not all), floating point systems sound better for some reason.  I suspect, but can not empirically prove, that it has to do with the additional mathmatical processes and scaling that they must use.  I do not believe that it has to do, at least not entirely, with the bit depth capability of the processor.
This does not mean that fixed point systems don't or can't sound good, just that, in my opinion, floating point systems are more transparent and sound better.

Quote:

I understand that under the covers, Yamaha uses 40 bit fixed point data paths, and 56 bit accumulators. That's more than enough to get the24 bit job done accurately and transparently.

The Yamaha architecture in the consoles that I am familiar with (if my memory serves me correctly) is 32 bit fixed point processing with a 58 bit accumulator internal processing.  I can't say that all of their consoles use the same architecture since I am not intimately familiar with all of them.
The Venue uses dual fixed point 24 bit processors, don't know the accumulator depth.  
I am not certain on the absolute specifics of other consoles off the top of my head.

Quote:

BTW for the record, my DAW software of choice is 32 bit floating point under the covers, but that had nothing to do with my choice.

I would agree that most decisions regarding a specific piece of equipment are not made based solely upon the processor type or any single factor but also on the interface, available peripherals, plug ins, etc.  

Quote:

People need to consider what the consequences are, of running out of CPU power. When a computer array (and that is what a modern digital console is) runs out of CPU power, the results are pretty catastrophic and non-subtle. There are major drop-outs, clicks and pops in the output signal.

I was not talking about the processor running out of power and dropping out, etc.  I was talking about some audible sonic differences between fixed point and floating point based products which seem to hold true, at least predominantly, across the board.

Quote:

Also, converters in good modern equipment don't sound dramatically different. Virtually ever converter chip used today implements the same basic technology, delta-sigma. As a rule their outputs are all fixed-point. The good ones all have very low noise and very good frequency response compared to other audio components. Nobody is going to use too-cheap converters in something as complex and expensive as a digital console, because they are a relatively minor expense. With most digital consoles the greater part of the cost goes into the User Interface, and the remaining analog components.

Now we have changed topics to Digital Audio Converters.  
Yes, most today are Delta Sigma rather than some version of a ladder type converter, that does not mean they all sound the same or we would see loads of Grace, Avalon, Benchmark, Gordon, etc. (fill in your favorite top of the line pre-amps here) connected to digital consoles or DAW's through Alesis, Behringer, Peavey, Presonus, M-Audio or Mackie DAC's.  This is not typically the case or manufacturer's of high end, better sounding DAC's would be out of business (unless they put pretty blue lights on them, but that's another story:).
Delta Sigma converters sound better than ladders but add latency.  Most designers today choose additional latency when they "choose" better sound quality because they can't have both.  They compromise.  We can make up for some latency issues by utilizing a higher sample rate, this not only decreases latency but, given quality equipment, also improves the high frequency response and "opens up" the sound.  This has benefits within the reproducable audio range even if the system being used for the live or recorded music can not reproduce the full bandwidth of what was captured by the mic and processed by the console/DAW.  But, again, slightly different topic.

Again, everything is about how the whole system (console from input to output or PA from microphone to speaker) works together as a cohesive unit.  All components within any system have had some compromises made for some reasons either in the equipment or system design stages.  The original question was a comparison between two very different consoles in terms of price point and complexity.

"nuff said.  I have hijacked the thread for too long.  Wish I could learn to be much more succinct like a few of you other guys.

His,
Lee Buckalew
Pro Sound Advice, Inc.

[Andy Peters]

Lee Buckalew wrote on Wed, 07 January 2009 09:50

Delta Sigma converters sound better than ladders but add latency.  Most designers today choose additional latency when they "choose" better sound quality because they can't have both.  They compromise.  We can make up for some latency issues by utilizing a higher sample rate, this not only decreases latency but, given quality equipment, also improves the high frequency response and "opens up" the sound.

The latency is a consequence of the modulator and the antialiasing filter, and this latency is usually on the order of a dozen or so samples. Which is to say, not much.

-a

[Andy Peters]

Arnold B. Krueger wrote on Wed, 07 January 2009 05:40

Me As for the second point about "more channels and processing you utilize, taxing the processor" -- the people who do the actual coding recognize that they have a "CPU budget," meaning that they know how much processing they can do within a sample period. The processing doesn't "degrade." You can either do what is called for, or you can't. And any reasonable system will not give you the option to do what it can't do. People need to consider what the consequences are, of running out of CPU power. When a computer array (and that is what a modern digital console is) runs out of CPU power, the results are pretty catastrophic and non-subtle. There are major drop-outs, clicks and pops in the output signal.

Um, you might want to read a post before you respond to it.

A modern digital console is designed such that it cannot run out of CPU power. As I've already said, the DSP engineer knows how much processing is available, and will not allow for plug-ins or whatever to exceed that limit. I mean, this is a solved problem. If more CPU horsepower is required, then additional processors are added.

Put simply: you will never hear clicks, pops and drop-outs as a result of running out of DSP horsepower.

-a

[Lee Buckalew]

Andy,
As I am understanding it, but I am not a digital designer/programmer, one ended Sigma Delta creates, in general, around 1 millisecond of latency.  Having Sigma Delta conversion for both input and output adds about 2 milliseconds.  This depends upon sample rate so since a typical digital console has about a 2.4 millisecond latency or so at 44.1/48kHz and 1.2 millisecond latency at 88.2/96kHz that sounded about right.

His,
Lee Buckalew
Pro Sound Advice, Inc.

[Andy Peters]

Lee Buckalew wrote on Wed, 07 January 2009 16:20

Andy, As I am understanding it, but I am not a digital designer/programmer, one ended Sigma Delta creates, in general, around 1 millisecond of latency.  Having Sigma Delta conversion for both input and output adds about 2 milliseconds.  This depends upon sample rate so since a typical digital console has about a 2.4 millisecond latency or so at 44.1/48kHz and 1.2 millisecond latency at 88.2/96kHz that sounded about right. His, Lee Buckalew Pro Sound Advice, Inc.

RTFDS. Just this past week I've been looking some Cirrus converters and they have a latency of 12 samples or less. At 48 kHz, twelve samples is 250 microseconds. Most converters have group delay of less than a couple dozen samples. So don't worry about it.

There is additional, unavoidable, latency in the console due to the actual processing.

Hers,
Andy Peters

[Arnold B. Krueger]

Lee Buckalew wrote on Wed, 07 January 2009 16:50

Arnold, Quote: Marketing blurbs, I fear. People seem to be running around casting stones in ignorance. The whole fixed point/floating point digital audio battle has been fought once already in the realm of DAW software. The outcome was that different developers make different choices, but all competent developers end up making stuff that works very well, thank you! I am not relying on marketing but on my own listening experiences in the recording and live sound world.  I also listen to the opinions of other engineers/producers.

People say the darnedest things. I can provide dozens quotes of engineers and producers unknowingly claiming to have suspended the laws of physics. Do you think that people become infallible when they take on the job title of engineer or producer? How many "hits" does one need to produce to be granted the ability to suspend or countermand the laws of physics? No such thing!

Something as seemingly simple as a listening test is something that like everything else has to be done with knowledge of the pitfalls, and with craft and precision, if its results are to have broader meaning. People are forever doing alleged listening tests that fail on any number of scientific grounds.

How many live sound and audio production people are familiar with the principles and practice of experimental psychology as applied to audio listening tests?

If you're going to say "floating point systems are more transparent and sound better", you are, whether you know it or not, making a highly exceptional claim. Go to an IEEE or AES developer's forum, where the *real* experts are, and try to say it. The question will then be whether anybody will take you seriously enough to bother to answer you, given that you just said something so silly.

Quote:

I do a lot of classical recording work, some which was a part of the grammy process this year.  One of the great things about audio is the diversity of people, equipment, technique, etc. and that we can agree to disagree and still have sounds that our audiences like.

That's all fine and good, but when you're making exceptional claims, it is not about Grammy's, it is about Science.

Quote:

In my experience and the experiance of many other recording engineers (certainly not all), floating point systems sound better for some reason.

There is no scientific reason why a properly designed fixed or floating point system needs to be less than perfectly sonically transparent.

It is likely that most if not all of the music that we listen to whether recorded or live, has been processed *both* ways (however many instances of fixed-point processing, and however many instances of floating point processing) by the time we hear it.

For example, the digital filters on modern ADCs and DACs for audio (themselves small DSPs) are generally fixed point. I don't know of any that are floating point. Do you?  

Quote:

I suspect, but can not empirically prove, that it has to do with the additional mathmatical processes and scaling that they must use.

Anybody who is familiar with numerical methods, particularly anybody who has written assembly-level code to implement common numerical methods used in audio, will be shocked by this *revelation*.  

Floating point processing requires *more* mathematical processing if for no other reason due to the need to take the standard fixed point audio data, and convert it to and from floating point. In fact this conversion can be done quickly and transparently, but it does amount to being an example of the dreaded (by some) "additional mathematical processes and scaling".

You do know that all common audio transmission schemes (AES/EBU, AES3, SP/DIF, TOSLINK, ADAT, etc.) are all fixed point, right?  

You do know that while floating point .wav files are possible, and used by many (including me on occasion), in fact most WAV files are fixed point?

You do know that all widely-distributed digital media (CD, DAT, SACD, DVD-A, MP3, AC/3, AAC, etc.) are fixed-point?

The bottom line is that if fixed point processing causes audible difficulties, then we are all in a world of hurt because it is endemic in the audio world.

Among truely knowledgeable people (e.g. people who tech people to write DSP code), either fixed point or floating point processing are completely acceptable for professional purposes, and can be chosen based on other considerations than sound quality.

If someone believes they heard a difference solely due to data being in fixed point or floating point, either their test has serious problems (all too common), or the equipment they are testing is poorly-designed (not likely in a modern digital console) or broken (possible).

[Jeff Ekstrand]

Also, to say that one console sounds better/less digital than another simply because of floating-point or fixed point is to say that is the only variable that has changed.

I don't believe, and correct me if I'm wrong, that anyone here has ever run the same console, the EXACT same console, in both floating and fixed-point. Mind your, that would need to be in the exact same room, with the exact same PA, with the exact same number of people, the exact same DSP (is that fixed or floating???), etc.

This is to piggyback Arnold's statements that there is no scientific proof for fixed or floating sounding better.

I'm still intrigued to hear what "so many people/experts/engineers" are hearing that makes these floating-point systems so sonically superior to their fixed-point brethren. Or maybe it's something that only "the experts/engineers" can hear, and the rest of us are doomed to listen with our puny, undeveloped, less-than-adequate heads. Which then begs the question: Is it worth it for all "the experts" to go to these lengths if "the masses" aren't going to hear the difference?

*end sarcasm*

[Lee Buckalew]

Jeff,
I was making the same point, in response to Arnold, that you just made about consoles but when I went to post it said I had been logged out.

Yes, I think the only way to test this whole idea, if we were talking about the processors, would be for two identical systems to be run, with identical consoles whose "only" difference was the processing type.

I am not saying that there is something scientific about sonic differences between processors.  I very clearly stated that it was my impression (and the impression of a number of others but certainly not all) that consoles and DAWS utilizing floating point processing had sonic qualities that they preffered over those with fixed point.  I also said that this was rarely the reason that these devices were chosen and that many very good sounding devices are out there in both processor types.  Choices are generally made for other reasons.
It is quite possible that the differences being heard are not due to the processors but that they have to due with other design or component differences.  Since there are no identical systems with only processor differences that I am aware of we can not test this.  I have had a consistent preference for floating point based systems but, there are many possible reasons for this.  

I certainly should have put "Best" in quotes in my comment about the best sounding...as that is my opinion, shared by others but my opinion in that post that many of the best sounding examples utilize floating point processing.  
I think I also clearly stated that there are many very good sounding devices out there that utilize fixed point processing.

Going back to my original point, I will be very interested to hear the SD8 and until then withold comment on how it sounds.

Now, on to an important discussion.  Which green markers make CD's sound the best and, is light a wave or particle phenomenon

His,
Lee Buckalew
Pro Sound Advice, Inc.

[John Fiorello]

Lee Buckalew wrote on Fri, 09 January 2009 10:39

Which green markers make CD's sound the best and, is light a wave or particle phenomenon

Marker?  Crayola.


Light?  Both. or neither.





JF

[Mike Galica]

I'm sorry but I have to ...

Light is both a particle and wave, except when you want it to be one of the options and then it is the other just to spite you.

[ProSoundWeb Community]