Misleading Measurements

[lucretius]

2 hours ago, bluesman said:

A digital signal file is indeed a discrete entity that exists "in isolation",  despite your erroneous insistence a few posts above that "[d]igital signals are turned into analogue signals. They don't exist in isolation".  SInce you don't seem to understand what a digital signal is, here's one you can see in isolation - it's a snippet of an mp3 (shown in hex format):

 

 

 

 

To be clear, the signal is the representation (just the information*, an abstraction), not the physical manifestation.  Then, signals do not exist in the same sense as physical objects.

 

*For example, the same signal can be physically represented in many ways.

[lucretius]

4 hours ago, lucretius said:

*For example, the same signal can be physically represented in many ways.

 

Indeed, the same signal can be physically represented in both a "digital" and in an "analog" way.  A DAC converts one physical manifestation of a signal into yet a different physical manifestation of the same signal. The resulting signal (in its physical analog representation) coming from the DAC is not a "de novo" analog signal (that would suggest it was created from nothing).

[lucretius]

17 hours ago, bluesman said:

You can’t listen directly to a digital signal.

 

The same is true of an "analog" (representation of a) signal coming from a DAC (or from a turntable for that matter) -- it's just a pattern of voltages, not sound waves.

[Summit]

19 hours ago, bluesman said:

And with digital signals, there is neither headroom nor clipping as we understand them in the analog world.  Analog devices are generally tolerant of minor level excursions over whatever 0 dB reference is used.  Distortion, compression etc are progressive in analog circuitry and increase proportionately (but not linearly) as the level goes further above 0.  But when a digital signal level hits 0 dBFS, the maximum # of 1s has been reached for the word size being used and further increases just discard the additional information.  There is no soft clipping in digital audio, and there are no innocuous or potentially pleasurable effects of digital clipping.

 

DAWs using 32 bit floating point resolution can keep adding more, so they’re theoretically immune to overload during the recording process.  But when you convert back to fixed point to edit, mix, master and/or otherwise turn the capture into a playable and archivable file, every 1 in the code to the right of the truncation point (16 or 24 bits with few exceptions) is simply lost as is the signal information encoded in it.

 

This is not correct.

[bluesman]

5 hours ago, Summit said:

 

This is not correct.

PLease tell us what you think is incorrect.  If you're disputing that clipping and headroom manifest entirely differently (and thus need to be treated differently) in analog and digital signals, here's just one of an endless stream of factual support for what I said, which was that "...with digital signals, there is neither headroom nor clipping as we understand them in the analog world".  You may have missed the bold, underlined, italicized phrase that ends this sentence. 

 

This quote comes from modernmixing.com:

 

"With floating point processing, even though the clipping light is red, the audio file is actually not clipping.  Floating point processing essentially gives you an infinite amount of headroom, as long as the audio stays inside the digital realm. [highlighting added by me to make sure you see and read it] That means it’s impossible to clip your audio files inside Pro Tools (or whatever DAW you use). But if you aren’t careful, once the “clipped” audio leaves your DAW, you will hear the nasty effects of digital distortion."

 

From here, we get this explanation supporting my statement that 0 dBFS is an upper limit for digital audio file signal levels in recordings:

 

'FS' stands for 'Full Scale' and 0 dBFS is the highest signal level achievable in a digital audio WAV file. Higher levels are possible inside digital audio workstation software, but in the files that are recorded on disk, 0 dBFS is the highest level.......All other levels can be measured and described with respect to 0 dBFS. So for example a signal that is 10 decibels lower than the maximum possible level is -10 dBFS. A signal inside the digital audio workstation could be +10 dBFS, but it would need to be lowered in level for output as a WAV file, otherwise it would be clipped, meaning that the tips of the waveform would be squared off at 0 dBFS."

 

From sources like this, this, and this, here are two of many descriptions identical to mine of the reason 32 bit float point is functionally immune to overload and clipping, plus a clear and simple example:

 

"Each audio sample for 32-bit float files consumes 32 bits of space on a hard disk or memory, and for a 48 kHz sampling rate, this means that 32 x 48,000 = 1,536,000 bits per second are needed for 32-bit, 48 kHz files. So for 33% more storage space compared to 24-bit files, the dynamic range captured goes from 144 dB up to, essentially, infinite (over 1500 dB). [highlighting added by me to make sure you see and read it] But more importantly, audio signals above 0 dBFS are preserved in the file, rendering clipped audio a thing of the past."

 

"32-bit float recording technology allows what seems to be impossible: Recording at a level so low that it is inaudible when monitoring at normal levels and not even visible on the meters, and also allows recording at levels so high they are impossible to achieve. This inaudible, invisible, low-level recording or incredibly high-level recording can then be normalized to full reasonable, workable level (<0 dBfs) with no additional amplifier noise, distortion, or artifacts."

 

"These samples were recorded on a MixPre-3 II to demonstrate the advantage of using 32-bit float WAV files for recording. We split a signal from a MKH40 mic into two MixPre-3 IIs, one recording 24-bit fixed WAVs, and the other 32-bit float WAVs. We then applied too much gain to loud dialogue. Low-cut was set to 80 Hz, and no limiters were active. Both the original 32-bit float and 24-bit fixed file are heavily distorted and unusable.  These files were imported into iZotope RX7, and -30 dB of gain was applied to each file. As you can see and hear, the 32-bit float files scale back perfectly, and the 24-bit files do not."  [Download the files HERE and see the explanatory web page HERE.]

 

[yamamoto2002]

17 hours ago, bluesman said:

"Each audio sample for 32-bit float files consumes 32 bits of space on a hard disk or memory, and for a 48 kHz sampling rate, this means that 32 x 48,000 = 1,536,000 bits per second are needed for 32-bit, 48 kHz files. So for 33% more storage space compared to 24-bit files, the dynamic range captured goes from 144 dB up to, essentially, infinite (over 1500 dB). [highlighting added by me to make sure you see and read it] But more importantly, audio signals above 0 dBFS are preserved in the file, rendering clipped audio a thing of the past."

 

When FTZ CPU flag is not set, On 32bit float values that contain 24bit integer PCM values,

• amplify it by +765dB and attenuate it by -765dB → original 24bit value is fully recovered  but
• amplify it by +771dB and attenuate it by -771dB → least significant bit is smashed and bit precision is reduced to 23bit
• attenuate it by -759dB and amplify it by +759dB → original 24bit value is fully recovered but
• attenuate it by -765dB and amplify it by +765dB → least significant bit is smashed and bit precision is reduced to 23bit

 

765 + 759 = 1524 dB. I think the above `over 1500 dB'  is came from here.

 

But typical old DAWs use 32bit float with FTZ CPU flag is set (to avoid very slow denormalized number computation that hampers realtime delivery of data)

and 32bit float dynamic range is reduced to 1385 dB.

 

Newer DAWs use 64bit float as a internal representation of PCM and it has about 12000 dB of dynamic range.

[Summit]

18 hours ago, bluesman said:

PLease tell us what you think is incorrect.  If you're disputing that clipping and headroom manifest entirely differently (and thus need to be treated differently) in analog and digital signals, here's just one of an endless stream of factual support for what I said, which was that "...with digital signals, there is neither headroom nor clipping as we understand them in the analog world".  You may have missed the bold, underlined, italicized phrase that ends this sentence. 

 

This quote comes from modernmixing.com:

 

"With floating point processing, even though the clipping light is red, the audio file is actually not clipping.  Floating point processing essentially gives you an infinite amount of headroom, as long as the audio stays inside the digital realm. [highlighting added by me to make sure you see and read it] That means it’s impossible to clip your audio files inside Pro Tools (or whatever DAW you use). But if you aren’t careful, once the “clipped” audio leaves your DAW, you will hear the nasty effects of digital distortion."

 

From here, we get this explanation supporting my statement that 0 dBFS is an upper limit for digital audio file signal levels in recordings:

 

'FS' stands for 'Full Scale' and 0 dBFS is the highest signal level achievable in a digital audio WAV file. Higher levels are possible inside digital audio workstation software, but in the files that are recorded on disk, 0 dBFS is the highest level.......All other levels can be measured and described with respect to 0 dBFS. So for example a signal that is 10 decibels lower than the maximum possible level is -10 dBFS. A signal inside the digital audio workstation could be +10 dBFS, but it would need to be lowered in level for output as a WAV file, otherwise it would be clipped, meaning that the tips of the waveform would be squared off at 0 dBFS."

 

From sources like this, this, and this, here are two of many descriptions identical to mine of the reason 32 bit float point is functionally immune to overload and clipping, plus a clear and simple example:

 

"Each audio sample for 32-bit float files consumes 32 bits of space on a hard disk or memory, and for a 48 kHz sampling rate, this means that 32 x 48,000 = 1,536,000 bits per second are needed for 32-bit, 48 kHz files. So for 33% more storage space compared to 24-bit files, the dynamic range captured goes from 144 dB up to, essentially, infinite (over 1500 dB). [highlighting added by me to make sure you see and read it] But more importantly, audio signals above 0 dBFS are preserved in the file, rendering clipped audio a thing of the past."

 

"32-bit float recording technology allows what seems to be impossible: Recording at a level so low that it is inaudible when monitoring at normal levels and not even visible on the meters, and also allows recording at levels so high they are impossible to achieve. This inaudible, invisible, low-level recording or incredibly high-level recording can then be normalized to full reasonable, workable level (<0 dBfs) with no additional amplifier noise, distortion, or artifacts."

 

"These samples were recorded on a MixPre-3 II to demonstrate the advantage of using 32-bit float WAV files for recording. We split a signal from a MKH40 mic into two MixPre-3 IIs, one recording 24-bit fixed WAVs, and the other 32-bit float WAVs. We then applied too much gain to loud dialogue. Low-cut was set to 80 Hz, and no limiters were active. Both the original 32-bit float and 24-bit fixed file are heavily distorted and unusable.  These files were imported into iZotope RX7, and -30 dB of gain was applied to each file. As you can see and hear, the 32-bit float files scale back perfectly, and the 24-bit files do not."  [Download the files HERE and see the explanatory web page HERE.]

 

 

I have already clearly marked what was wrong. I'm not fooled by your attempt to change the theme. We did not discuss digital mixing in a studios, but digital signals that are sent between different devices and their gain.

 

Clipping audio is when an audio signal is amplified past the maximum allowed limit, whether in a digital or analog system.

 

A digital signal carries the data in the form of binary. The digital signal just like the analog has amplitude, frequency and phase.

 

Digital electronics that send and receive digital signals actually handle (physically) analog signals which are digitally coded and they can easily be over loaded if the gain stage output more gain (v) than the receiving device can handle, and will start clipping.

 

A common situation where this occurs is when someone uses EQ and raises the base (for example) without also reducing the overall output. 

 

[stereo coffee]

On 4/30/2021 at 1:06 AM, March Audio said:

With respect this is simply incorrect.

 

Well I can only suggest you take a test signal at 0dBfs and measure the voltage at the RCA output of a range of dacs to test if your opinion is correct.

 

You will find they are nearly all around 2 volts rms.

 

If you don't believe me please take a visit to Audio Science Review where there are tests of probably a hundred plus dacs that show this to be the case.

 

Can I ask if you have a dac or CD player and if so what make and model it is

 

There are two realities and one fantasy here, as I see it. 

: 

Reality 1 is what is at the output of your RCA plugs on your CD player DVD player and is arranged to provide the D/a converter

in your player from a level that is already arranged that way  on the CD's you buy from your record store or online. Yes you can choose to use DAC's that internally amplify that further - but that is your choice, departing from what the industry sensibly has provided which is -10 dbv . the fact you can do it, does not mean everyone else is doing that too.  

 

Here are some articles discussing what is at the output of CD players not when being tested with sine waves, but when playing actual CD's.  https://www.sweetwater.com/insync/understanding-signal-levels-audio-gear/

           https://en.wikipedia.org/wiki/Line_level

           

 

 

 

It infers when a CD, lets say Fleetwood Mac Rumours is manufactured, the CD is held back in level so digitally the information only allows at the analog output stage of the player conversion to  0.316v RMS, 11.7825dBu because the CD manufacturer complies with rules concerning consumer line level. This is imminently sensible, as the CD label wants consumers to enjoy the CD in many forms of consumer equipment, none of which they have control over directly, 

 

The scope image level provided earlier by March Audio, which differed so much to my old school scope image  I suspect  shows his DAC is being switched for higher output,  which is a function usually of adding different resistance settings allowing gain to a inverting opamp expressed as R2/R1, Although his is the gustard X16 we can see the X22 model has feature of adjusting level.   In my test I ran

the output from my sony CD player via a RCA to BNC lead of very short length. 

 

But normal CD players do not have such a feature to increase level.   

 

The question to ask is,  can you presently buy Rumours on CD with 0DbFs   level available anywhere, and the answer is No because

the record labels are sensible with providing their product to match to consumer equipment. In my promised separate thread I have begun explanation of each type of equipment, but it is a work in progress for the moment.

 

Reality 2 ( A fantasy world, enticing some &  playing with mainly lots of visual treats, just look at those colours !  ) 

This is where forums like ASR are at presently, the reality see's pretension that when you play your CD at home via its RCA socket  it somehow gets close to the players 0DbFs level - reread Reality 1 , it is just not happening. At ASR review after review presenting sine waves,  as justifying music, and using Odbfs. Yes for hopefully the last time, players can reach such level, but its not being provided via a normal CD players RCA outputs, for very good reason, so as to reserve the player both digitally and via its analog output to have ability for dynamics in music which is the peak level. For consumer line level its 0.894V peak to peak   

 

Reality 3  ( very Real ) 

This is the levels used in recording studios, the levels here are very impressive but totally unrelated to consumer line level. The future of recorded sounds is in their hands, many fine engineers , and professionals getting it generally right, every day.   

 

Most of us live  with consumer equipment in Reality 1, this is for very good reason, as escaping or trying to escape  ( bypassing the fantasy of reality 2 )  to Reality 3 is full of pitfalls namely eternal frustration not being able to change the master tapes. A quick diversion to MQA - it will never be right as the original companding is not being matched.

 

Reality 1 is eminently sensible, and we can be thankful the CD industry chose this level as it enable the standard player we use to have reserve ability for dynamics, more on this in the promised thread. 

 

Hopefully that clears things up.       

[March Audio]

52 minutes ago, stereo coffee said:

 

There are two realities and one fantasy here, as I see it. 

: 

Reality 1 is what is at the output of your RCA plugs on your CD player DVD player and is arranged to provide the D/a converter

in your player from a level that is already arranged that way  on the CD's you buy from your record store or online. Yes you can choose to use DAC's that internally amplify that further - but that is your choice, departing from what the industry sensibly has provided which is -10 dbv . the fact you can do it, does not mean everyone else is doing that too.  

 

Here are some articles discussing what is at the output of CD players not when being tested with sine waves, but when playing actual CD's.  https://www.sweetwater.com/insync/understanding-signal-levels-audio-gear/

           https://en.wikipedia.org/wiki/Line_level

           

 

 

 

It infers when a CD, lets say Fleetwood Mac Rumours is manufactured, the CD is held back in level so digitally the information only allows at the analog output stage of the player conversion to  0.316v RMS, 11.7825dBu because the CD manufacturer complies with rules concerning consumer line level. This is imminently sensible, as the CD label wants consumers to enjoy the CD in many forms of consumer equipment, none of which they have control over directly, 

 

The scope image level provided earlier by March Audio, which differed so much to my old school scope image  I suspect  shows his DAC is being switched for higher output,  which is a function usually of adding different resistance settings allowing gain to a inverting opamp expressed as R2/R1, Although his is the gustard X16 we can see the X22 model has feature of adjusting level.   In my test I ran

the output from my sony CD player via a RCA to BNC lead of very short length. 

 

But normal CD players do not have such a feature to increase level.   

 

The question to ask is,  can you presently buy Rumours on CD with 0DbFs   level available anywhere, and the answer is No because

the record labels are sensible with providing their product to match to consumer equipment. In my promised separate thread I have begun explanation of each type of equipment, but it is a work in progress for the moment.

 

Reality 2 ( A fantasy world, enticing some &  playing with mainly lots of visual treats, just look at those colours !  ) 

This is where forums like ASR are at presently, the reality see's pretension that when you play your CD at home via its RCA socket  it somehow gets close to the players 0DbFs level - reread Reality 1 , it is just not happening. At ASR review after review presenting sine waves,  as justifying music, and using Odbfs. Yes for hopefully the last time, players can reach such level, but its not being provided via a normal CD players RCA outputs, for very good reason, so as to reserve the player both digitally and via its analog output to have ability for dynamics in music which is the peak level. For consumer line level its 0.894V peak to peak   

 

Reality 3  ( very Real ) 

This is the levels used in recording studios, the levels here are very impressive but totally unrelated to consumer line level. The future of recorded sounds is in their hands, many fine engineers , and professionals getting it generally right, every day.   

 

Most of us live  with consumer equipment in Reality 1, this is for very good reason, as escaping or trying to escape  ( bypassing the fantasy of reality 2 )  to Reality 3 is full of pitfalls namely eternal frustration not being able to change the master tapes. A quick diversion to MQA - it will never be right as the original companding is not being matched.

 

Reality 1 is eminently sensible, and we can be thankful the CD industry chose this level as it enable the standard player we use to have reserve ability for dynamics, more on this in the promised thread. 

 

Hopefully that clears things up.       

 

None of this is correct.  Im not sure I have the energy to go through it in detail.

 

This is a rip of a Rumours CD ( a very old one, I think I bought it in the late 80's).  I looked at one track, Go Your Own Way and the peak value is -2.35dBFS.  So I dont know where you got -11dB from.

 

Most recordings are normalised to peak close to 0dBFS to maximise SNR.

 

You have been shown data that confirms literally hundreds of DACs conform to the industry standard of 2 v rms out at 0dBFS.  The X16 conforms to this standard.  There are no modern dacs or cd players that conform to -10dBv.

 

Again can you tell us which model of Sony CD player you have and we can look up the specs and check what output voltage it has?

 

Also could you please provide a list of albums and I will post the the amplitude stats for them like above.

[SoundAndMotion]

44 minutes ago, stereo coffee said:

But normal CD players do not have such a feature to increase level.   

 

Out of curiosity, how many CD players have you tested/measured? Do you think Stereophile has perhaps tested more than you?

See what they say here (link):

Quote

The standard CD-player output voltage is 2V RMS, with units varying between 1.74V on the low side (the Audio Research DAC1) and a whopping 7.2V on the high side (the Theta DS Pro Basic). Most CD players and processors put out between 2.2V and 3.5V. Note that this value is the highest RMS output voltage possible from the player—there's no digital signal greater in amplitude than 0dBFS (see the "Decibels" sidebar).

 

[March Audio]

Just for giggles this is the 24/96 remaster of the same track.

 

Rammed up to -0.1dBFS!  So @stereo coffee, not 0dBFS but near as damn it. So it will be peaking 2.8 volts. The old Rumours cd at -2.35dB will peak 2.1 volts.

 

[Jud]

1 hour ago, SoundAndMotion said:

 

Out of curiosity, how many CD players have you tested/measured? Do you think Stereophile has perhaps tested more than you?

See what they say here (link):

 

 

@stereo coffee, like the Hedgehog, knows One Great Thing. (Apologies to Archilocus and Isaiah Berlin.)

[fas42]

Just bumped into this, on ASR ... https://www.audio “science” review/forum/index.php?threads/things-that-cannot-be-measured.20808/post-690068

 

By member Sawdust123 there, who just happens to be this chap,

 

 

Considering he's in the AP world, I suspect he knows a thing or two about the subject ... bits of a couple of recent posts,

 

Quote

The point of mentioning all this is that we already have the science that explains why transient responses will be different than steady state responses yet the vast majority of our testing is steady state.

 

and

 

Quote

I sold audio test gear for 15 years and sat on standards committees too. In a nutshell, we measure what we do because we can do so easily. It is my opinion that we can get A LOT more from instrumentation but choose not to because the financial reward for doing so is not there.

 

[March Audio]

2 hours ago, fas42 said:

Just bumped into this, on ASR ... https://www.audio “science” review/forum/index.php?threads/things-that-cannot-be-measured.20808/post-690068

 

By member Sawdust123 there, who just happens to be this chap,

 

 

Considering he's in the AP world, I suspect he knows a thing or two about the subject ... bits of a couple of recent posts,

 

 

and

 

 

 

 

Its a good video and raises some great points.  A few comments though.

 

He says towards the beginning that measurements are used to infer a subjective performance level.  Well yes, in simplistic marketing maybe, however thats not what measurements are for.  They are there to objectively assess various technical performance parameters.  The correlation of that data to subjective performance is very much in the domain of psychoacoustics.  As he pointed out, different types of distortion have different levels of subjective impact.

 

Secondly, again whilst in simplistic product specs you may only see a small number of test results, the tests any good designer will be making are much wider than the "single sine wave".  So I think thats a bit of a mis-characterisation.  Some of the more complex test he mentions are often used - you will even see them on ASR, multitone for example.

 

On the other side of things we have significant problems with casual subjective comparisons.  Audiophiles will perform comparisons without any controls in place.  They will see the products which will induce inevitable bias.  They wont accurately match volume which will lead to erroneous conclusions.  Every time I have tested audiophiles and put basic controls in place the previously heard significant differences in sound either evaporate or at least become much, much less significant.  Now dont take that to mean I think everything sounds the same - it absolutely does not, however I take peoples subjective opinions with a very large pinch of salt.

 

Finally one other thing to note is that we really can measure way beyond levels of audibility.  The video is really about types of tests and their interpretation rather than measurements being insensitive of ineffective.

[opus101]

28 minutes ago, March Audio said:

On the other side of things we have significant problems with casual subjective comparisons. 

 

Who is the 'we' here? You?

[Summit]

On 5/2/2021 at 3:24 PM, SoundAndMotion said:

 

Out of curiosity, how many CD players have you tested/measured? Do you think Stereophile has perhaps tested more than you?

See what they say here (link):

 

 

The link is to a page written by Harley 1995. Today many DAC can output higher than 2V RMS, especially in balanced, so I would not say that it is the "standard" any more.

 

That there's no digital signal greater in amplitude than 0dBFS is correct.  

[lucretius]

On 4/30/2021 at 11:31 AM, bluesman said:

And with digital signals, there is neither headroom nor clipping as we understand them in the analog world.

 

Some observations:

1. The AES standard (AES17) defines 0 dBFS as the RMS value of a full-scale sine wave.  Because the definition of full scale is based on a sine wave, it will be possible with square-wave test signals to read as much as + 3.01 dBFS.
    
2. It's true that the value in dBFS does not relate directly to the original absolute sound pressure level of the audio measured in dB. Nonetheless, we do convert between digital and analog levels, although no single standard/convention exists. Thus, a potential for ambiguity exists when assigning a level on the dBFS scale to a waveform; some engineers (and software programs) use a conventional RMS calculation leading to a full scale sine wave reading -3.01 dBFS (which is incorrect per the AES standard), while other engineers choose the reference level so that RMS and peak measurements of a sine wave produce the same result.
    
3. Peak meters in some DAWs and digital mixing systems indicate the peak among the sampling points, not the resulting waveforms. The resulting waveforms represented by the samples may exceed full scale. That is to say, a digital signal that does not contain any samples at 0 dBFS can still clip when converted to analog form due to the signal reconstruction process interpolating between samples.

[March Audio]

14 hours ago, lucretius said:

 

Some observations:

1. The AES standard (AES17) defines 0 dBFS as the RMS value of a full-scale sine wave.  Because the definition of full scale is based on a sine wave, it will be possible with square-wave test signals to read as much as + 3.01 dBFS.
    
2. It's true that the value in dBFS does not relate directly to the original absolute sound pressure level of the audio measured in dB. Nonetheless, we do convert between digital and analog levels, although no single standard/convention exists. Thus, a potential for ambiguity exists when assigning a level on the dBFS scale to a waveform; some engineers (and software programs) use a conventional RMS calculation leading to a full scale sine wave reading -3.01 dBFS (which is incorrect per the AES standard), while other engineers choose the reference level so that RMS and peak measurements of a sine wave produce the same result.
    
3. Peak meters in some DAWs and digital mixing systems indicate the peak among the sampling points, not the resulting waveforms. The resulting waveforms represented by the samples may exceed full scale. That is to say, a digital signal that does not contain any samples at 0 dBFS can still clip when converted to analog form due to the signal reconstruction process interpolating between samples.

Hi

 

Just on point 1, its not quite correct, sorry :).  The waveform shape makes no difference as it is the peak voltage level that is 0dBFS.  With a sinewave the RMS level will be - 3.01dB

 

You can see in the statistics how the peak sample value is +- 32768, the maximum possible 16 bit value. This is 0dBFS.  You cant exceed this.

 

 

 

A square wave will peak at 0dB also. In the picture below you can see ripples because we dont have the infinite bandwidth necessary to correctly reproduce a square wave. The ripples are hard clipped at 0dB.

 

 

 

When signal processing/upsampling yes the interpolation can create inter-sample overs.  However these are still clipped, they wont in practice be reproduced over 0dB.

 

So as you never know what the replay might end up doing in that processing you should never really record/normalise to 0dBFS.  With 24 bit recording there is never any real need to either.

[danadam]

1 hour ago, March Audio said:

Just on point 1, its not quite correct, sorry :).  The waveform shape makes no difference as it is the peak voltage level that is 0dBFS.

Are you saying that @lucretius is not correct or that the AES17 standard is not correct? :-)

The standard clearly says:

Quote

6.3 Output amplitude at full scale

In systems where the input is accessible in the digital domain, the output amplitude at full scale shall be the
r.m.s. voltage that results from a sine wave whose positive peak value reaches the positive digital full scale
under normal settings of gain controls.

 

[March Audio]

27 minutes ago, danadam said:

Are you saying that @lucretius is not correct or that the AES17 standard is not correct? :-)

The standard clearly says:

 

Its the interpretation of the words in the standard.  The standard is correct in what it says.

 

What its saying is that the analogue output is "full scale" (maximum output voltage measured in volts RMS) when the peak level of the sine wave reaches the maximum digital level (+ 32768 for 16 bit).  This is 0dBFS.

 

For a sinewave the relationship between the peak value and the RMS value is 1.414.  This is equal to 3.01dB.

 

The values cant go above 0dBFS

 

You can see this on the pictures I posted above.

[opus101]

44 minutes ago, March Audio said:

(+ 32768 for 16 bit).  

 

Point of pedantry - 16bit twos-complement signed numbers are limited to +32767 (or 0x7FFF hex).

[Bill Brown]

This is getting bizarre, heck, borderline ridiculous.  @March Audio and @bluesman are absolutely correct.

 

There is a conversion of an analog signal to digital (ADC).  You CANNOT encode a music signal with one and zeroes above 0 dbFS, it will all be (terrible) noise, there is no headroom recording digitally.  In fact, you should not approach 0 dbFS, as they said, to avoid intersample overs when the ones and zeroes are converted back to analog (DAC).

 

When a DAC chip converts the digital signal back to analog (in a DAC designed for pros or consumers), it produces either a signal that could be described as current or voltage.  If the former, a current to voltage converter is the next stage (traditional), in the latter, this is done on the chip (modern).  This voltage then goes through additional gain stages in the DAC to either produce -10dbV from consumer DACs (typically, but not always), or +4dbU in professional DACs (more closely adhered to).  These levels are important to interface with subsequent/corresponding consumer or professional gear that is designed to accept the different levels for proper gainstaging (enough to drive to full output levels in preamps, then amps, but not overload them).  It really isn't complicated.

 

Bill

[March Audio]

38 minutes ago, Bill Brown said:

This is getting bizarre, heck, borderline ridiculous.  @March Audio and @bluesman are absolutely correct.

 

There is a conversion of an analog signal to digital (ADC).  You CANNOT encode a music signal with one and zeroes above 0 dbFS, it will all be (terrible) noise, there is no headroom recording digitally.  In fact, you should not approach 0 dbFS, as they said, to avoid intersample overs when the ones and zeroes are converted back to analog (DAC).

 

When a DAC chip converts the digital signal back to analog (in a DAC designed for pros or consumers), it produces either a signal that could be described as current or voltage.  If the former, a current to voltage converter is the next stage (traditional), in the latter, this is done on the chip (modern).  This voltage then goes through additional gain stages in the DAC to either produce -10dbV from consumer DACs (typically, but not always), or +4dbU in professional DACs (more closely adhered to).  These levels are important to interface with subsequent/corresponding consumer or professional gear that is designed to accept the different levels for proper gainstaging (enough to drive to full output levels in preamps, then amps, but not overload them).  It really isn't complicated.

 

Bill

Yes, but (sorry) as discussed at length earlier in the thread there are no domestic dacs that output at -10dBV.  Most are close to 2 v rms at 0dBFS. This is +6dBV. (RCA output)

[bluesman]

18 hours ago, lucretius said:

 

Some observations:

1. The AES standard (AES17) defines 0 dBFS as the RMS value of a full-scale sine wave.  Because the definition of full scale is based on a sine wave, it will be possible with square-wave test signals to read as much as + 3.01 dBFS.
    
2. It's true that the value in dBFS does not relate directly to the original absolute sound pressure level of the audio measured in dB. Nonetheless, we do convert between digital and analog levels, although no single standard/convention exists. Thus, a potential for ambiguity exists when assigning a level on the dBFS scale to a waveform; some engineers (and software programs) use a conventional RMS calculation leading to a full scale sine wave reading -3.01 dBFS (which is incorrect per the AES standard), while other engineers choose the reference level so that RMS and peak measurements of a sine wave produce the same result.
    
3. Peak meters in some DAWs and digital mixing systems indicate the peak among the sampling points, not the resulting waveforms. The resulting waveforms represented by the samples may exceed full scale. That is to say, a digital signal that does not contain any samples at 0 dBFS can still clip when converted to analog form due to the signal reconstruction process interpolating between samples.

We need to be honest and clear when discussing things based on a well defined standard.  First, #1 is not your observation - you're quoting definition 3.4 in the very AES standard under discussion.  Here's a link to the actual document, so everyone can read it.

 

Second, while it's true that a square wave contains more energy than a sine wave of the same frequency and amplitude, the standard specifically tells us in the ssentence after the one you offered as your observation that only a sine wave is to be used, specifically:  "Square-wave signals at this level are not recommended because tilt or overshoot introduced by any filtering operations will cause clipping of the signal".

 

Third, "full scale amplitude" is defined clearly as the "amplitude of a 997-Hz sine wave whose positive peak value reaches the positive digital full scale, leaving the negative maximum code unused". It isnnot defined as the RMS value of that signal - it's defined on the basis of peak value.

 

Fourth, DB FS is "amplitude expressed as a level in decibels relative to full-scale amplitude (20 times the common logarithm of the amplitude over the full-scale amplitude).  

 

Fifth, "The existence of an AES standard does not in any respect preclude anyone, whether or not he or she has approved the document, from manufacturing, marketing, purchasing, or using products, processes, or procedures not conforming to the standard."

 

And finally, none of this standard addresses the use of 0 dB FS as the maximum level of a digital audio signal in the digital domain - but that's where it's used. Further, the standard does not address the bit depth or the issue of fixed vs floating point.  It simply requires that the digital format being used be specified when reporting the value determined for the standard in the equiment under test.

[Bill Brown]

1 hour ago, March Audio said:

This voltage then goes through additional gain stages in the DAC to either produce -10dbV from consumer DACs (typically, but not always)

 

Yes, my mistake, you are correct.  Sorry.  2V is typical (what I meant to write first and noted before instead referencing the standards, oy), but there is more variation than seen in professional DACs.

 

Bill