Is the dynamic range of CD sufficient?

[Don Hills]

... my observations are based upon observing the digital waveform from the 16-bit digital file in SoundForge ...

Mark,
I suggest you find 25 uninterrupted minutes and watch this video:
http://www.youtube.com/watch?v=cIQ9IXSUzuM

Some people disagree with the presenter's opinions on the use of dither in some cases, but the technical content is correct.

 

[j_j]

However, my observations are based upon observing the digital waveform from the 16-bit digital file in SoundForge, with the vertical axis magnified to allow inspection down to the LSB.

What you see in SoundForge does not in any fashion resemble what should come out of a DAC. Sorry, I missed that before. You'd need to convolve that stairstep waveform with the impulse response of an anti-imaging filter to get a sense of what you really would see.

 

[Orb]

I have no idea if I've already replied. Let us remember a few things.

1) The actual noise level at your eardrum due to the molecular nature of air is between 6dB SPL and 8dB SPL white noise. The only way (which is obviously unadvisable to say the least) to be rid of this is to put the listener in a vaccum so that there is no air on either side of his or her eardrum. Don't do that.

2) Going up 96dB from that puts one at 102dB SPL as a peak level. This is below what most speakers can reproduce, so it is possible that in a perfectly quiet listening room this could create a problem

BUT

3) Most listening rooms are in the 20-30dBa SPL range, or even worse in DBc. projecting up from even the 20dB level puts us at 116dB SPL peak. This is now reaching the range of most common, good, hifi speakers.

Now, room noise does not have flat frequency content. A reasonable estimate for a really good listening room under perfect conditions with very clean, very powerful speakers, suggests that 18 bits is enough.

For most rooms in a home, 16 bits is probably more than enough.

This, of course (note: BIG (*&(*&*( IF HERE), IF the CD is well-recorded.

Now, for extraordinary signals (fireworks, space shuttle takeoffs) reproduced at moderately close to realistic levels, using sound reinforcement systems in order to achieve the necessary levels, 16 bits is not enough.

For recording, 16 bits is not enough. End of discussion. It lacks overhead, noise floor, and ability to EQ, mix, and process. At least 20 bits is the minimum for a recording, and there's nothing wrong with as close to 24 bits as you can manage.

Although JJ just to expand as a bit of the puzzle is missing; the actual digital signal level must also be considered as this has no relation to physics and room noise/maximum dB in a room.
Mentioning this because some recordings will be based closer to dbfs (and usually sound pretty crud coming back to loudness wars) while others can be as low down as -80dBFS on average for complex sound related harmonics at say 5-to-10khz.
HiFiNews has a pretty complex analysis for hirez download music, and it is very interesting to see where the signal level is against frequency/plotted time of a track for the best recorded and mastered tracks, and also for those that are done poorly or with issues.
The best they have reviewed do reach down to -120db (but these are rare), the other consideration is the nature of harmonics-notes and how it is the higher frequencies that are substantially lower in energy-amplitude, although percussive intruments also go way beyond 15khz as well; however it is this information I assume that is possible (and in other cases not be) to be masked by various factors.
Not disagreeing just added another important angle to this that relates to this discussion.
Cheers
Orb

 

[Julf]

The best they have reviewed do reach down to -120db (but these are rare)

And how do we know if that is actual source audio content or just noise?

 

[Orb]

Let's test your assertion which says that something with only 2 steps must be bad. Ok.

For instance, SACD home systems use a 1-bit playback DAC. Yes, they oversample, but in that revelation lies your failure to understand how it all works. SACD uses oversampled, noise-shaped PCM. Yes, boys and women, that's all SACD is, an inefficient form of PCM.

When you have no oversampling, it is still entirely possible to switch that 1 bit (2 level) system so that the resulting sound is pure white noise. That's what dither is supposed to do, and what it does perfectly well when it's done properly.

As to your illusion of steps, don't forget that there is a reconstruction filter in any DAC, so you don't actually see the steps in any case.

You are operating from a point of view that has been intentionally propagated by some completely irresponsible authors and bloggers, which I am not blaming you for, HOWEVER, I must advise you that you are flat-out wrong in your assertion that you MUST have granularity at low levels in PCM.

SACD is a perfect example of how that has to be wrong, since it's a 1-bit system that is oversampled, yet lacks granularity. And, yes, it's oversampled PCM with quantization noise shaping, nothing whatsoever more or less, despite all of the hype and claims about it.

Sorry to be picky but not sure SACD is a good example JJ; mainly because it is closer to PWM (being PDM if really being picky) than PCM.
This is an important differentiation between PCM and PWM as they behave differently in terms of use.

Yeah it is unfortunate the concept of understanding the steps in digital is subtly misrepresented by some sites, but then I feel Monty and others do something similar as well to create the narrative that there are no steps (whether physical or theory functional) or how sampling rate has no effect on resolution; both camps are missing the finer details as each are trying to present a narrative.

Mark,
as JJ says you will never see actual steps at the final output stage due to filters and all music is dithered, however there is one example where you will see steps and that is with NOS DACs (this is in reality alias images but will look like steps), but importantly it is interesting no-one has ever mentioned fatigue/graining/etc with any of the NOS DACs they either own or when auditioning.
That said due to the nature of NOS DACs, they are limited to roughly 17bit resolution at very best and more average examples can be as low as 14/15bit.
IF music or sounds were done without dither (remember all studios will use a dither process-stage) then at lower amplitude (when very quiet tone) you would see the step-squares as you say; this is why from a technical point calculating distortion/noise/etc can be volatile between magazines as there will be subtle differences between using a dithered tone and one without.
The transfer funtion/quantization does create errors/distortion but this is masked with dither; so in theory one "might" say digital is not truly perfect because it requires a further artificial source of noise (dither) combined with the actual original signal.
This is where another debate will be between various camps; how much influence if any does dither and the types used/if applied frequently/incorrectly by studio/etc have on the music listened to - such discussions have already been covered by everyone here so no need to go over old ground I say.

Anyway worth remembering;
Greater sampling rates allow greater flexibility and more ideal filters and provides greater FR for complex musical sounds - greater sampling rate does not improve resolution per se.
Greater bit depth lowers quantisation errors-distortion even further at lower signal levels (closer to -96dBFS rather than 0dBFS) and provides greater dynamic range - very subtle distortion-noise difference between 16-bit and 24-bit at say -60dbfs and lower but mostly academic due to dither.
Look at graph 13,14,15 for an example of bit depth with undithered tone : http://www.stereophile.com/content/dcs-vivaldi-digital-playback-system-measurements
Another good example would be the HiFiNews measurement that is only in their physical publication showing subtle difference between 16-bit and 24-bit signal relating to noise-distortion against signal level (0 dBFS loud to -120dBFS incredibly quiet).

Cheers
Orb

 

[Orb]

And how do we know if that is actual source audio content or just noise?

Because Keith Howard/Paul Miller has created their own tools (Paul Miller has had a professional measurement product on sale for over a decade now) and both came from scientific research backgrounds academically and post graduate; bear in mind both these guys have made many software analysis tools over the past.
That said and more relevant to your question noise has a particular structure with this analysis tool as is easily identifiable including the structure of DSD related noise-etc (need to read HiFiNews if you are doubtful - but bear in mind a lot of the investigation and what the tool shows was provided in greater detail back in 2012 I think), furthermore in the past Keith Howard did a separate investigation into whether hirez beyond 16bits was music or noise and again used specific tools to prove this.
BTW their tool identified a lot of issues with hirez in the early days Julf, and still does with on average 2 in 6 of the hirez reviews that are also measured with some kind of digital recording-mastering issue.
They can tell if the recording is upsampled/downsampled/bit depth change/transcoded from-to PCM or DSD/issues with filter used/etc.
TBH no-one else I know has done anything like they have to analyse-measure hirez music.
Cheers
Orb

 

[Julf]

Sorry to be picky but not sure SACD is a good example JJ; mainly because it is closer to PWM (being PDM if really being picky) than PCM.

DSD can definitely be viewed as 1-bit, oversampled, noise-shaped PCM.

I feel Monty and others do something similar as well to create the narrative that there are no steps (whether physical or theory functional) or how sampling rate has no effect on resolution

And what, in your view, does Monty get wrong?

as JJ says you will never see actual steps at the final output stage due to filters and all music is dithered, however there is one example where you will see steps and that is with NOS DACs (this is in reality alias images but will look like steps), but importantly it is interesting no-one has ever mentioned fatigue/graining/etc with any of the NOS DACs they either own or when auditioning.
That said due to the nature of NOS DACs, they are limited to roughly 17bit resolution at very best and more average examples can be as low as 14/15bit.
IF music or sounds were done without dither (remember all studios will use a dither process-stage) then at lower amplitude (when very quiet tone) you would see the step-squares as you say; this is why from a technical point calculating distortion/noise/etc can be volatile between magazines as there will be subtle differences between using a dithered tone and one without.
The transfer funtion/quantization does create errors/distortion but this is masked with dither; so in theory one "might" say digital is not truly perfect because it requires a further artificial source of noise (dither) combined with the actual original signal.
This is where another debate will be between various camps; how much influence if any does dither and the types used/if applied frequently/incorrectly by studio/etc have on the music listened to - such discussions have already been covered by everyone here so no need to go over old ground I say.

I think you are confusing dither and reconstruction filter/function. But yes, NOS filterless designs will show steps caused by aliasing that should have been removed by the (missing) reconstruction filter.

 

[Orb]

DSD can definitely be viewed as 1-bit, oversampled, noise-shaped PCM.



And what, in your view, does Monty get wrong?



I think you are confusing dither and reconstruction filter/function. But yes, NOS filterless designs will show steps caused by aliasing that should have been removed by the (missing) reconstruction filter.

Err no I am not confusing dither and reconstruction filter/function....
NOS DACs cause alias artifacts, which when analysed looks like steps.
What way do you think I misconstrued dither considering what I wrote?

Ok SACD, is it closer to PWM or PCM?
Look up PDM.
I have worked in digital engineering for quite a few years in the past, anyway everyone I know gets picky when SACD is said to be PCM like.

regarding Monty,
well he argues there is no stair whether in theory or actual digital structure, even though the transfer function is a stair and reason it needs noise and that without the filters you do have a stair - look this is oversimplifying it and something I have gone over at length with Monty in his own threads and also when discussing engineering of NOS/quantisation/etc in one of the other threads.
BTW I am trying to stay away from using specific technical terminology here as it is just going to confuse the heck out of most, as said we all have debated this at length in some other technical threads.
Cheers
Orb

 

[Julf]

Because Keith Howard/Paul Miller has created their own tools (Paul Miller has had a professional measurement product on sale for over a decade now) and both came from scientific research backgrounds academically and post graduate; bear in mind both these guys have made many software analysis tools over the past.
That said and more relevant to your question noise has a particular structure with this analysis tool as is easily identifiable including the structure of DSD related noise-etc (need to read HiFiNews if you are doubtful - but bear in mind a lot of the investigation and what the tool shows was provided in greater detail back in 2012 I think), furthermore in the past Keith Howard did a separate investigation into whether hirez beyond 16bits was music or noise and again used specific tools to prove this.
BTW their tool identified a lot of issues with hirez in the early days Julf, and still does with on average 2 in 6 of the hirez reviews that are also measured with some kind of digital recording-mastering issue.
They can tell if the recording is upsampled/downsampled/bit depth change/transcoded from-to PCM or DSD/issues with filter used/etc.
TBH no-one else I know has done anything like they have to analyse-measure hirez music.

I do read HiFi News regularly. Just stating that Paul Miller and Keith Howard come from a scientific research background doesn't really answer my question, so let me restate it - how, specifically, can you detect what part of a low-level signal comes from the original audio signal, and what part is noise?

Yes, I appreciate that they post spectrograms of the hi-res recordings they review, but they are not at all alone in doing that. I used to post a bunch of similar spectrograms on CA way back, so did many others, and hydrogen audio has had some good ones. The HiFi News guys don't have any magical measurement tools that can tell for sure if a recording is upsampled or resampled - they use the same tools as the rest of us, and make educated guessed based on that.

 

[Julf]

Err no I am not confusing dither and reconstruction filter/function....
NOS DACs cause alias artifacts, which when analysed looks like steps.
What way do you think I misconstrued dither considering what I wrote?

This part:

IF music or sounds were done without dither (remember all studios will use a dither process-stage) then at lower amplitude (when very quiet tone) you would see the step-squares as you say

The reason you would see step-squares is lack of reconstruction filtering, not dither.

Ok SACD, is it closer to PWM or PCM?
Look up PDM.

One can be seen as a variation of the other depending on the constraints. I am somewhat familiar with PDM, having worked with it, off and on, for 35 years.

I have worked in digital engineering for quite a few years in the past

So have many of us.

well he argues there is no stair whether in theory or actual digital structure, even though the transfer function is a stair

No. It is a set of discrete points in time.

I am trying to stay away from using specific technical terminology here as it is just going to confuse the heck out of most, as said we all have debated this at length in some other technical threads.

I am afraid that trying to avoid the applicable technical terminology only leads to more confusion.

 

[Orb]

You do agree noise has a particular structure yes that does not conform to music?

BTW you will see I also said in very next sentence "That said and more relevant to your question noise has a particular structure with this analysis tool as is easily identifiable including the structure of DSD related noise-etc (need to read HiFiNews if you are doubtful - but bear in mind a lot of the investigation and what the tool shows was provided in greater detail back in 2012 I think), furthermore in the past Keith Howard did a separate investigation into whether hirez beyond 16bits was music or noise and again used specific tools to prove this."

So what do you take from the hirez measurement graph as you read it and Paul Miller comments about the noise floor being high and therefore poor due to probably from the analogue master or when he comments about the noise floor being exceptionally low and excellent/etc?
Anyway as I keep saying, the noise has a particular structure, same way filters or other anomalies can show up on their analysis.
Cheers
Orb

 

[Orb]

This part:



The reason you would see step-squares is lack of reconstruction filtering, not dither.



One can be seen as a variation of the other depending on the constraints. I am somewhat familiar with PDM, having worked with it, off and on, for 35 years.



So have many of us.



No. It is a set of discrete points in time.



I am afraid that trying to avoid the applicable technical terminology only leads to more confusion.

Sigh last response as we are screwing this thread up with your approach Julf - not sure if you deliberately being unhelpful to this thread or not.
1. You will see stair step due to lack of reconstruction filter (if look back I said NOS DACs create a stair step due to no filter), HOWEVER you will also see step without dither IF the signal is low and say using 16-bit (I provided a good example with the Stereophile link where I emphasised it is an undithered tone at -90dBFS), this is a step.

2. Great so you work with PDM, so is SACD PDM or PCM?
You know the answer is PDM, and what is PDM closer to; PCM or PWM.
Again you know the answer is PWM, so why say I was wrong initially in pointing out to JJ that in reality SACD is closer to PWM due to it being PDM.....

3. Transfer Function/Quantisation ; you complain about my responses being too simple and then go with a 1 liner It is a set of discrete points in time.
However the Transfer function IS associated as a "stair" process, but this is such a complex topic we covered over 10 pages just on this in the past from a technical and semantic context, including university lecture papers.
Look I am not saying it is an actual stair (look back I say there is theory-process that this fits with and the "physical" aspect that the alias image step fits with due to no filters).

Cheers
Orb

 

[Julf]

You do agree noise has a particular structure yes that does not conform to music?

Random noise does not conform to music. Intermodulation can be viewed as distortion or noise - SNR measurements tend to clump both random noise and distortion in "unwanted information" or noise vs. the original signal.

BTW you will see I also said in very next sentence "That said and more relevant to your question noise has a particular structure with this analysis tool as is easily identifiable including the structure of DSD related noise-etc (need to read HiFiNews if you are doubtful - but bear in mind a lot of the investigation and what the tool shows was provided in greater detail back in 2012 I think), furthermore in the past Keith Howard did a separate investigation into whether hirez beyond 16bits was music or noise and again used specific tools to prove this."

Yes, and their "specific tool" only works in some very limited scenarios (where there has been zero-padding without subsequent filtering or gain adjustment). The same result can be accomplished using the flac encoder and looking at "wasted bits".

So what do you take from the hirez measurement graph as you read it and Paul Miller comments about the noise floor being high and therefore poor due to probably from the analogue master or when he comments about the noise floor being exceptionally low and excellent/etc?

I have seen quite a number of spectrograms showing a pretty high noise level that then drops to "absolute black" in between - a result of processing 16-bit material using 24-bit software.

Anyway as I keep saying, the noise has a particular structure, same way filters or other anomalies can show up on their analysis.

And I keep asking what that particular structure is.

 

[Orb]

Random noise does not conform to music. Intermodulation can be viewed as distortion or noise - SNR measurements tend to clump both random noise and distortion in "unwanted information" or noise vs. the original signal.



Yes, and their "specific tool" only works in some very limited scenarios (where there has been zero-padding without subsequent filtering or gain adjustment). The same result can be accomplished using the flac encoder and looking at "wasted bits".



I have seen quite a number of spectrograms showing a pretty high noise level that then drops to "absolute black" in between - a result of processing 16-bit material using 24-bit software.



And I keep asking what that particular structure is.

I have to disagree with you here Julf, their tool is nothing like normal spectrograms or your example at the end.
Anyway my last response on this as we are not going to agree on anything it seems.
Orb

 

[Orb]

And I keep asking what that particular structure is.

Sorry then obviously you have not read HiFiNews that often or have forgotten Keith Howard investigation into this back 2012.
Maybe you should write to Paul Miller and notify him that when he mentions what is noise or a good recording/issues he is in fact wrong as the tool only works in a very basic way that can be fooled and is no better than basic spectrogram.
He does respond as I have had quite a bit of correspondence with both him and Keith Howard.

Cheers
Orb

 

[j_j]

Sorry to be picky but not sure SACD is a good example JJ; mainly because it is closer to PWM (being PDM if really being picky) than PCM.

SACD is 1 bit oversampled noise-shaped PCM, guy. No more, no less. Not PWM at all, the feedback method is different in very important ways.

It is of course a (*&(*&* bear to work with. That's a different problem, and I think that's why it exists, security by annoyance.

 

[Julf]

Maybe you should write to Paul Miller and notify him that when he mentions what is noise or a good recording/issues he is in fact wrong as the tool only works in a very basic way that can be fooled and is no better than basic spectrogram.

He is not wrong. His tool (that looks at transitions) only works in limited circumstances - something he admitted to when I pointed out the similar functioning of the flac encoder "wasted bits" function.

 

[Orb]

SACD is 1 bit oversampled noise-shaped PCM, guy. No more, no less. Not PWM at all, the feedback method is different in very important ways.

It is of course a (*&(*&* bear to work with. That's a different problem, and I think that's why it exists, security by annoyance.

JJ so why do all technical/engineering articles say SACD uses PDM, which is related to PWM?
Thanks
Orb

 

[j_j]

Noise can be signal-correlated (i.e. a random process that is modulated by some characteristic of the signal), or signal-independent.

The term "distortion" is poorly defined and has a very limited mathematical basis. Once you have more than one frequency present in a signal, the sum and difference terms become plentiful.

So "noise characteristics" for signal-independent noise are very simple, frequency spectrum, level, PDF.

For signal-dependent noise, you must consider the cause of the noise. Much messier.

Of course, taking a white noise source and putting it through a filter and a time-domain window is a whole lot like some kinds of percussion, or even some kinds of woodwind, so that confuses the issue even further.

 

[Orb]

He is not wrong. His tool (that looks at transitions) only works in limited circumstances - something he admitted to when I pointed out the similar functioning of the flac encoder "wasted bits" function.

Can you provide a real world example how this happens with actual music released?
Thanks
Orb