Boswell

Moderator
Joined
Apr 19, 2006
Location
UK
The only place that an inter-sample over-limit can occur is in the analogue reconstruction filter in a DAC. Early in the advent of CD players, there were some models that could overload because the designer had not left enough headroom on the output amplifiers. This applied not only to the Walkman type, where battery consumption and maximising output sound level were both primary concerns, but also (embarrassingly) in certain hi-fi component CD players, where there was no excuse. I was involved at the time in generating "problem" CDs and testing them on commercial players, looking for waveform flat tops on an oscilloscope and corresponding abnormalities on a distortion measurement system.

When it comes to DAW processing at a fixed sample rate, it will not happen, since inter-sample levels never exist inside a DAW. However, you can route tracks that get near 0dB to effect modules that may then give overrange outputs, but in fixed-wordlength coding, it's the job of the module to report that samples in the effected result cannot be represented in the selected wordlength.
 

pcrecord

Quality recording seeker !
Well-Known Member
Joined
Feb 21, 2013
Location
Quebec, Canada
The only place that an inter-sample over-limit can occur is in the analogue reconstruction filter in a DAC. Early in the advent of CD players, there were some models that could overload because the designer had not left enough headroom on the output amplifiers. This applied not only to the Walkman type, where battery consumption and maximising output sound level were both primary concerns, but also (embarrassingly) in certain hi-fi component CD players, where there was no excuse. I was involved at the time in generating "problem" CDs and testing them on commercial players, looking for waveform flat tops on an oscilloscope and corresponding abnormalities on a distortion measurement system.

When it comes to DAW processing at a fixed sample rate, it will not happen, since inter-sample levels never exist inside a DAW. However, you can route tracks that get near 0dB to effect modules that may then give overrange outputs, but in fixed-wordlength coding, it's the job of the module to report that samples in the effected result cannot be represented in the selected wordlength.
So do you think this can still happens to modern CD players ?
Do you think it's still relevant to lower the final ouput half or 1 db quieter to be safe ?
 

Boswell

Moderator
Joined
Apr 19, 2006
Location
UK
I think that, many years ago, when we exposed this problem, the perpetrators took notice, so that modern designs are proof against it. In addition, most modern DACs perform their reconstruction and sinc (sin(x)/x) compensation in the digital domain and require only low-pass filtering in the analogue domain.

It's not something I bother to check in my own mixes these days, but then I never let peak values get to 0dBFS. Depending on the material and when the peaks happen, I am usually happy with 0.5dBFS or even closer. I've also not seen any studies showing what happens to occasional full-scale peaks in mixes that are subsequently processed to LUFS standards. They are supposed to be preserved if the average level is -14dBFS or below.
 

bouldersound

Real guitars are for old people.
Well-Known Member
Joined
Jan 23, 2010
Location
Boulder, Colorado
What I read, and this may not be correct, is that having headroom beyond 0dBFS was actually a violation of the Red Book specification. Not that they wanted to cause a problem but that it was never considered at the time that signals would be processed as they are these days. The output voltages for given dBFS levels were specified in a way that didn't account for clipped digital signals that would be reconstructed with peaks above the originally specified peak voltage.
 

Boswell

Moderator
Joined
Apr 19, 2006
Location
UK
For "Red Book Specification" you could say "the laws of physics", since there is no way of storing over-range numbers on a digital medium. However, that assumes that all the numbers are independent, which is not the case when representing digitised sounds, leaving room for some exploitation.

There are two ways this anamoly can show up: one intentional and the other unintentional. To take the latter first, there is no specification in the Red Book about the exact form the DAC reconstruction filter should take, as long as it meets certain requirements. Different implementation of these filters can lead to subtly different characteristics while still meeting the prescribed requirements, and one of these differences is overshoot. It's not really relevant as long as the reproduction system can correctly handle the result, which was one of the characteristics I was looking into all those years ago in the early CD players.

The second aspect is the intentional one, and that is deliberately tweaking the recorded samples such that any type of reconstruction filter would cause the output to go over range. There were several rock bands that made use of this technique to gain even more points in the loudness wars. For obvious reasons I won't go into the details on a public forum of how it can be done, but the difference was very audible.
 

bouldersound

Real guitars are for old people.
Well-Known Member
Joined
Jan 23, 2010
Location
Boulder, Colorado
I was referring to what dBFS values were specified to correspond to what voltages, which is arbitrary. If the specification includes something like "0dBFS shall produce 1.414v" and something like "the output shall not exceed 1.414v" then it wouldn't be surprising to have a some DACs that don't behave well with digital signals that reconstruct to more than 1.414v. But that's just speculation on my part.

My habit used to be to master things to -.3dBFS peak, but I found that peaks could rise when a file was compressed so I've shifted to a full -1dBTP ("true peak"), which happens to be the EBU R128 specification.
 

audiokid

Chris
Moderator
Joined
Mar 20, 2000
Location
Nanaimo BC, Canada
I was referring to what dBFS values were specified to correspond to what voltages, which is arbitrary. If the specification includes something like "0dBFS shall produce 1.414v" and something like "the output shall not exceed 1.414v" then it wouldn't be surprising to have a some DACs that don't behave well with digital signals that reconstruct to more than 1.414v. But that's just speculation on my part.

My habit used to be to master things to -.3dBFS peak, but I found that peaks could rise when a file was compressed so I've shifted to a full -1dBTP ("true peak"), which happens to be the EBU R128 specification.
I've found this as well. I always put a limiter on to stop this. Sequoia's or Fabfilter's limiter work well for me.
 

kmetal

Kyle P. Gushue
Well-Known Member
Joined
Jul 21, 2009
Location
Boston, Massachusetts
If I understand it correctly, the EBU R128 standard specifies -23dBLUFS, which is much lower than the usual for music streaming. It also specifies -1dBTP ("true peak"), which might seem a bit overly cautious until you see what data compression can do to an audio file. I generally master to -14dBLUFS and -1dBTP.



That's the thing, it may or may not have an audible effect depending on the DAC in question and how far above 0dBFS the intersample peaking goes. Some DACs have headroom to accommodate ISP and some don't, which I think is a legacy of the Red Book standard and early digital formats. You probably aren't likely to get ISP from a properly digitized analog signal, but you can get it from a digitally processed signal. Or something like that. I suspect ISP sounds different on different DACs. Most likely it sounds like some variation on clipping.

My approach is to be a bit overcautious and leave a full dB of true peak headroom on my uncompressed file. Then it should be nearly impossible for data compression (e.g. mp3 encoding) to cause it to go above 0dBTP.


This is interesting stuff. It makes me wonder how much dimminishing returns there are when making different masters optimized for different delivery formats.

Theres EBU, digital delivery in flac, wav, and mp3, vinyl, cd, hi res digital, then streaming, and probably more i cant think of.

Im not sure there has ever been more delivery formats at once in history. Maybe the closest was vinyl, 8 track, cassette, and radio mixes/masters at one very short crossover period?

These types of things like ISP facinate me since they can be one of those "cant put my finger on it, but something aint right here" kind of things. I bet its also something a dedicated mastering engineer is alot more cognizant of than an average joe like me.

(Edit i posted this before refreshing my browser and seeing the 7 replies i havent read yet)
 

Smashh

Well-Known Member
Joined
Oct 12, 2012
Location
Australia
Great stuff guys . I struggle to get clarity and I am starting to think that my DA converter is not helping.
I really need to experiment more with the plug ins that I have .

Can I compensate by using plug ins to create an exceptable clarity ?
Or . will I never get clarity with this DA ( studio live 16.0.2 )
Excuse my ignorance , but Im trying to understand the process better .
Thanks for the great insights
 

kmetal

Kyle P. Gushue
Well-Known Member
Joined
Jul 21, 2009
Location
Boston, Massachusetts
Generally using less processing is a way to maintain clarity. The most transparent 3rd party pluggins ive used are fabfilter and ozone, they may be worth a try. Since the DA is partly responisible what your hearing from your daw/plugins, no pluggins can compensate for the DA, everything passes thru the DA.

As far as the DA, the question becomes which of these is your weakest link- the room, the monitors, or the DA? Id classify the Studiolive as 'decent', and my general guess without knowing what your using is the room or speakers are probably lying to you more than your DA.
 

bouldersound

Real guitars are for old people.
Well-Known Member
Joined
Jan 23, 2010
Location
Boulder, Colorado
I struggle to get clarity and I am starting to think that my DA converter is not helping.

I don't think it's your converter. I had to be pretty aggressive with my eq to clear up the mix and what I ended up with wasn't as bright as some of the other mixes.
Sex and Candy kick eq.jpg
 
Top