John
Forum Host
- Total Posts : 30467
- Joined: 2003/11/06 11:53:17
- Status: offline
RE: A question about the 64bit engine
2009/05/21 10:36:05
(permalink)
Just to try and make it absolutely clear, the question of clipping in either 32 or 64 bit floats is a complete red herring! Both offer such a vast range, the real world of audio simply cannot fill it. The real problem is not the absolute range, defined by the exponent (8 bits for 32 bit floats and 12 bits for 64) but the resolution of the mantissa (25 bits effective for 32 bit and 53 bits for 64 bit floats).
Where clipping rears it's head is with the 24 or 16 bit integer's from the A-D, D-A process. D-A is not a problem as the 24 or 16 bit stream to the D-A can be easily scaled by the software and is an issue that can be addressed in comfort at the mastering phase. Input clipping is the real issue where the use of 24 bit converters can be used to advantage by lowering the input levels with no consequent loss of resolution (provided it is not overdone!). It is only input clipping that cannot be fixed in the mix!
With respect to the OP's question, I think I have already answered this. Although 32 bit floats are adequate for say 99%+ of our mixes (I'm talking time line here), the advantage given by 64 bit floats easily takes care of the last little bit. With regard to the accumulation of errors, even simple mixes are subject to thousands of calculations ALL of which contribute to the error build up (not primarily summing as was suggested earlier). The accumulation of errors will certainly affect 24 bit usage, for DVD soundtracks say, but can also often accumulate to affect even 16 bit output files. Certainly if your target is 16 bit CD quality, you are less likely to ever notice a difference between 32 or 64 bit floats than with 24 bit media such as DVD or SACD.
This is great theory but I am talking about experience. When Sonar was not using a FP audio engine (Sonar 3) it clipped and it was noticeable. One track clipping even a little was sonically apparent. When CW went to a floating point audio engine clipping became sonically unnoticeable. This is not a guess on my part nor is it theory but based on real world experience. You can test this out yourself if you do not believe me. The clipping is due only to the internal signal path being incapable of expressing a level that is beyond its ability to handle. It sounds the same as a hardware clip of an AD or DA converter.
|
Mr. Ease
Max Output Level: -71 dBFS
- Total Posts : 960
- Joined: 2003/11/24 18:44:01
- Location: West Sussex, UK
- Status: offline
RE: A question about the 64bit engine
2009/05/21 12:24:53
(permalink)
ORIGINAL: John
Just to try and make it absolutely clear, the question of clipping in either 32 or 64 bit floats is a complete red herring! Both offer such a vast range, the real world of audio simply cannot fill it. The real problem is not the absolute range, defined by the exponent (8 bits for 32 bit floats and 12 bits for 64) but the resolution of the mantissa (25 bits effective for 32 bit and 53 bits for 64 bit floats).
Where clipping rears it's head is with the 24 or 16 bit integer's from the A-D, D-A process. D-A is not a problem as the 24 or 16 bit stream to the D-A can be easily scaled by the software and is an issue that can be addressed in comfort at the mastering phase. Input clipping is the real issue where the use of 24 bit converters can be used to advantage by lowering the input levels with no consequent loss of resolution (provided it is not overdone!). It is only input clipping that cannot be fixed in the mix!
With respect to the OP's question, I think I have already answered this. Although 32 bit floats are adequate for say 99%+ of our mixes (I'm talking time line here), the advantage given by 64 bit floats easily takes care of the last little bit. With regard to the accumulation of errors, even simple mixes are subject to thousands of calculations ALL of which contribute to the error build up (not primarily summing as was suggested earlier). The accumulation of errors will certainly affect 24 bit usage, for DVD soundtracks say, but can also often accumulate to affect even 16 bit output files. Certainly if your target is 16 bit CD quality, you are less likely to ever notice a difference between 32 or 64 bit floats than with 24 bit media such as DVD or SACD.
This is great theory but I am talking about experience. When Sonar was not using a FP audio engine (Sonar 3) it clipped and it was noticeable. One track clipping even a little was sonically apparent. When CW went to a floating point audio engine clipping became sonically unnoticeable. This is not a guess on my part nor is it theory but based on real world experience.
You can test this out yourself if you do not believe me. The clipping is due only to the internal signal path being incapable of expressing a level that is beyond its ability to handle. It sounds the same as a hardware clip of an AD or DA converter.
Nothing to test John, whatever your experience, as I never made any comment about non floating point calculations. If you look in my earlier posts here I did mention 48 bit fixed point but related to another DAW. Where did you get the information to say Cake ever used fixed point maths for audio? If ever, I think you will find that it is much earlier, Pro audio days perhaps, if and when Cake used any fixed point maths. There would be little point in using custom fixed point 32 bit maths when CPU's have had inbuilt 32 bit floating point units since the i486 days and I'm pretty sure they predate Pro-audio! EDIT: Even if they did use fixed point, common sense would have dictated that they could easily have used a higher resolution than 32 bit floats - say a 26 or 27 bit mantissa and 5 or 6 bit exponent (this would be achieved with the necessary scaling of the fixed point maths) and even then it should not have suffered from clipping. Now integer maths is another matter entirely!
post edited by Mr. Ease - 2009/05/21 12:44:45
|
Mr. Ease
Max Output Level: -71 dBFS
- Total Posts : 960
- Joined: 2003/11/24 18:44:01
- Location: West Sussex, UK
- Status: offline
RE: A question about the 64bit engine
2009/05/21 12:39:10
(permalink)
A bit more! Something we should not discount is that there may have been a problem with how they handled overs in the track or master output and not scaled the entire signal appropriately. Are you saying you had this problem if your master output ever exceeded 0dBFS? It may be that they handle this more elegantly now and limit overs to max and min values.
post edited by Mr. Ease - 2009/05/21 12:53:29
|
Mr. Ease
Max Output Level: -71 dBFS
- Total Posts : 960
- Joined: 2003/11/24 18:44:01
- Location: West Sussex, UK
- Status: offline
RE: A question about the 64bit engine
2009/05/21 13:10:49
(permalink)
From my Sonar (1!) manual, Page 1-3. "All audio effects are 32 bit floating point for faster processing and high quality sound reproduction" I guess your problem with Sonar 3 was not related to this then. EDIT: To add - Why did you present Sonar 3 as not using floating point as some sort of fact? It clearly isn't. Someone of your experience should really know not to present false "facts".
post edited by Mr. Ease - 2009/05/21 13:25:22
|
UnderTow
Max Output Level: -37 dBFS
- Total Posts : 3848
- Joined: 2004/01/06 12:13:49
- Status: offline
RE: A question about the 64bit engine
2009/05/21 15:05:38
(permalink)
ORIGINAL: dbmusic
If our calculation engine is 6 bits, the result is truncated to a value of 156.17. Final exporting of our file will truncate to 5 bits, which would be 156.1.
Yes but if that 156.17 value gets added to the output of another calculation (let's say another bus) that has a value of 23.83 the result would be 180. If instead you were working with less precise values like 156.1 and 23.8 the sum would be 179.9. That is a difference at the final precision in your example. So it is all about small errors accumulating to the point that they affect the digits that are not being truncated. Of course this is all theory. Even if the 64 bit engine makes an audible difference, it is probably the absolute least important thing to worry about in the whole recording, mixing and mastering chain. (I leave it on because the option is there...) UnderTow
|
UnderTow
Max Output Level: -37 dBFS
- Total Posts : 3848
- Joined: 2004/01/06 12:13:49
- Status: offline
RE: A question about the 64bit engine
2009/05/21 15:11:34
(permalink)
ORIGINAL: John
When Sonar was not using a FP audio engine (Sonar 3) it clipped and it was noticeable. One track clipping even a little was sonically apparent. When CW went to a floating point audio engine clipping became sonically unnoticeable. This is not a guess on my part nor is it theory but based on real world experience.
Sonar has ALWAYS used a floating point engine. You must have been clipping the outputs or clipping a plugin that used fixed point maths (or plain broken). UnderTow
|
UnderTow
Max Output Level: -37 dBFS
- Total Posts : 3848
- Joined: 2004/01/06 12:13:49
- Status: offline
RE: A question about the 64bit engine
2009/05/21 15:17:49
(permalink)
ORIGINAL: Fog
I always found "odd" (my poor joke) with this whole computer and music thing.. yep cards are 24 bit conversion wise.. while from programming background stuff is normally in multiples of 8 or 16..
ROFL! 24 is also a multiple of 8! Anyway, the reason we don't have more than 24 bit converters is because we can't even use the full 24 bits. No converter achieves better than 21 or 22 bits in practise. The laws of physics don't allow it. The extra bits are just random noise generated by the components. UnderTow
|
UnderTow
Max Output Level: -37 dBFS
- Total Posts : 3848
- Joined: 2004/01/06 12:13:49
- Status: offline
RE: A question about the 64bit engine
2009/05/21 15:21:54
(permalink)
ORIGINAL: dbmusic
What I'm not seeing is the usefulness of truncating at the 64th bit versus the 32nd bit in the mixing process when everything is truncated to the 24th bit in rendering.
You have to remember that 32 bit floating point gives 25 bits of precision. Not much above 24 bits fixed point. Going to 63 bits gives 53 bits of precision. UnderTow
|
UnderTow
Max Output Level: -37 dBFS
- Total Posts : 3848
- Joined: 2004/01/06 12:13:49
- Status: offline
RE: A question about the 64bit engine
2009/05/21 15:29:15
(permalink)
ORIGINAL: Mr. Ease
 YAWN! I (personally) think the 20 -40% jokes are well past their sell by date.
I think it is literally years since I saw an original and funny take on this! IMHO of course...
And don't expect any soon. Due to the credit crisis the humour budget has gone down 20 to 40% this year. UnderTow
|
Mr. Ease
Max Output Level: -71 dBFS
- Total Posts : 960
- Joined: 2003/11/24 18:44:01
- Location: West Sussex, UK
- Status: offline
RE: A question about the 64bit engine
2009/05/21 15:47:47
(permalink)
ORIGINAL: John
This is great theory but I am talking about experience. When Sonar was not using a FP audio engine (Sonar 3) it clipped and it was noticeable. One track clipping even a little was sonically apparent. When CW went to a floating point audio engine clipping became sonically unnoticeable. This is not a guess on my part nor is it theory but based on real world experience.
You can test this out yourself if you do not believe me. The clipping is due only to the internal signal path being incapable of expressing a level that is beyond its ability to handle. It sounds the same as a hardware clip of an AD or DA converter.
I apologise to other forum members but I find this post has me quite riled and I would like to see John's response to this. Firstly, I have already disproved your claim that Sonar 3 did NOT have a floating point audio engine. The problem you ascribed to Fixed versus Floating point technology is completely incorrect. This was in response to my post although I had not actually made any comment regarding fixed point maths. Additionally, although you say "this is not a guess" it most certainly is. It is also based on false assumptions and is just plain wrong to boot. If you find that this is fine in theory, let's also talk experience. I find that I resent the tone and implication that I only know the theory and that your experience is somehow more telling. I have been an electronic design consultant since 1986 and a UK Chartered Engineer since 1979. My first job in audio design was in 1973 (whilst still a student) for Rank Leak Wharfedale designing audio amplifier test equipment. My experience of design in my mid teens directly led to this regular vacation post. After graduation I then worked worked for seven years as an electronic design engineer for Decca. I think you may well have heard of both of these companies. As a consultant for rapidly approaching 25 years, I have worked on many audio projects for many prestigious companies including reference audio designs for large semiconductor manufacturers, military secure digital audio communications and many others. Of course you seem to assume that I only know the theory. Personally, I think I also have the experience! What say you?
post edited by Mr. Ease - 2009/05/21 15:58:37
|
John
Forum Host
- Total Posts : 30467
- Joined: 2003/11/06 11:53:17
- Status: offline
RE: A question about the 64bit engine
2009/05/21 19:36:59
(permalink)
|
dbmusic
Max Output Level: -68 dBFS
- Total Posts : 1150
- Joined: 2005/07/04 12:52:46
- Location: Illinois
- Status: offline
RE: A question about the 64bit engine
2009/05/21 21:59:28
(permalink)
No need for a pissing match here. There's a lot of fact, opinions, and conjecture surrounding this subject. I, for one, am thankful of everyone's input. I now understand the cumulative impact of rounding errors and how mixing at 32-bit increases the probability that this will affect some degree of mathematical imprecision in a 24-bit rendering. I see now that mixing at 64-bit will significantly insulate the final output from those errors.
Now whether any of this can actually be heard in our final output medium is another matter entirely. My guess is that for most of us it will never make a difference. That being said, I would still opt for the 64-bit mixing engine simply because there is no good reason not to.
Practically speaking, I'm much more concerned about matters like proper gain staging in the analog realm and the affect this has on my noise floor (which I never can seem to get below about -65db).
Best regards,
DB
|
RTGraham
Max Output Level: -57 dBFS
- Total Posts : 1824
- Joined: 2004/03/29 20:17:13
- Location: New York
- Status: offline
RE: A question about the 64bit engine
2009/05/21 22:35:56
(permalink)
ORIGINAL: dbmusic
No need for a pissing match here. There's a lot of fact, opinions, and conjecture surrounding this subject. I, for one, am thankful of everyone's input. I now understand the cumulative impact of rounding errors and how mixing at 32-bit increases the probability that this will affect some degree of mathematical imprecision in a 24-bit rendering. I see now that mixing at 64-bit will significantly insulate the final output from those errors.
Now whether any of this can actually be heard in our final output medium is another matter entirely. My guess is that for most of us it will never make a difference. That being said, I would still opt for the 64-bit mixing engine simply because there is no good reason not to.
Practically speaking, I'm much more concerned about matters like proper gain staging in the analog realm and the affect this has on my noise floor (which I never can seem to get below about -65db).
Best regards,
DB
I'll just throw this in the mix: I've done reference bounces at 32-bit, and then final bounces at 64-bit; and just out of curiosity, I've occasionally phase-nulled them against each other. What's left is an extremely quiet, but audible, ghost of a sound; I can only describe it as low-level reverb tails coupled with very slight noise. I originally tried this because I thought I perceived a bit more depth and clarity to mixes with the 64-bit engine enabled, especially when using high-quality reverbs, and I was curious to see what the measureable difference really was.
~~~~~~~~~~ Russell T. Graham Keys, Vocals, Songwriting, Production russell DOT graham AT rtgproductions DOT com www DOT myspace DOT com SLASH russelltgraham
|
UnderTow
Max Output Level: -37 dBFS
- Total Posts : 3848
- Joined: 2004/01/06 12:13:49
- Status: offline
RE: A question about the 64bit engine
2009/05/22 08:26:56
(permalink)
ORIGINAL: John
Try it and see!
Do you really think I'm going to install Sonar 3 (or earlier) to prove a point?  Cakewalk Pro Audio 9 features: ORIGINAL: http://www.etcetera.co.uk/products/TWVPA9.shtml
<snipped for brevity>
Audio Features
* 32-bit Audio Mix Engine
* Real-time 32-bit, floating-point, stereo effects processing
<snipped for brevity>
UnderTow
|
Mr. Ease
Max Output Level: -71 dBFS
- Total Posts : 960
- Joined: 2003/11/24 18:44:01
- Location: West Sussex, UK
- Status: offline
RE: A question about the 64bit engine
2009/05/22 13:01:46
(permalink)
ORIGINAL: John
Try it and see!
What exactly would that prove then John? Certainly not that there is any truth whatsoever in your assertions? Quite likely that there may be some problem/bug in an older version of Sonar that is clearly unrelated to the topic of discussion. I think I have very clearly disproved your assertions, complete with relevant evidence. While I could (if I had any doubt whatsoever) install Sonar 3 and get the results you cite it most certainly would not prove you right, whatever your experience. All it would do is prove that there was some problem with Sonar 3 and yield absolutely no discernable information about the topic in discussion. You claimed Sonar 3 does/did not use floating point maths. WRONG - Proven with evidence. Any subsequent argument and speculative guessing on your part is based on that wrong information and therefore totally futile. I am happy to state, quite categorically, that you are wrong in your assertions regarding the maths of mixing. I think others will be able to clearly see the gaping holes in your logic and evidence. Why can't you - or are you just not prepared to admit it? EDIT: To dbmusic. I am sorry for this appearing on your thread but I found the comments from John to be:- 1. Factually wrong and obviously flawed. Thus they were likely to misinform others (including you) who do not deserve to be misinformed and 2. Personally offensive in the tone in which they sought to discredit the information I gave in good faith. I reserve my right to defend myself and apologise if this encroaches on your thread. I will happily start a new thread and reference this one if you wish.
post edited by Mr. Ease - 2009/05/22 13:48:11
|
Mr. Ease
Max Output Level: -71 dBFS
- Total Posts : 960
- Joined: 2003/11/24 18:44:01
- Location: West Sussex, UK
- Status: offline
RE: A question about the 64bit engine
2009/05/22 13:31:25
(permalink)
ORIGINAL: dbmusic
Practically speaking, I'm much more concerned about matters like proper gain staging in the analog realm and the affect this has on my noise floor (which I never can seem to get below about -65db).
Best regards,
DB
In atonement for hijacking your thread I hope this will help you. The dynamic range you can achieve really depends on what you are recording and your environment. If for instance you are recording vocals or acoustic guitar, you are unlikely to be recording anything much over 100 dB SPL. A recording environment of 35 dB SPL is quite good if you have not gone to great length to isolate the room from external noise (and I'm not talking about room treatment for monitoring). This will result in a dynamic range of (just!) 65 dB. So the answer is that it is most likely not your gain staging that is the problem. Unfortunately it is your environment which alas is very expensive to fix!
post edited by Mr. Ease - 2009/05/22 13:50:47
|