What is 'incompetent digital' ?

[Opus111]

Ethan recently deployed the phrase 'competent digital is audibly transparent' so what are members ideas on how to spot incompetent (i.e. not transparent) digital? Does incompetent digital have any defining traits?

I'll kick off by pointing out a couple of things I see as incompetent in much digital. Firstly S-D DACs which introduce noise modulation when fed with high crest factor (i.e. music-like) signals. Secondly digital filters (known as half-band) which violate the Nyquist criteria by only being -6dB at that frequency. Hence introduce imaging/aliasing artifacts.

Any others?

 

[amirm]

My rule is when they measure poorly, or have been designed without measurements.

 

[Opus111]

Meaning with traditional measurements?

Noise modulation doesn't show up clearly in the traditional measurements. In the communications field it was recognized in testing DACs that THD+N wasn't sufficient to characterize performance so another measurement was introduced using a multitone (i.e. high crest factor) signal. To my knowledge there are no standardized multitone test signals used in audio.

 

[amirm]

Incompetence shows up in classical measurements.

 

[Opus111]

Sure thing, but some incompetence slips by the currently used measurements.

 

[amirm]

I would call that second order incompetence . What do you have in mind on how to measure such?

 

[Blumlein 88]

IMD is multi-tone though I take it not what you have in mind. I guess normally it is dual tone, though it can be multi-tone.

I suppose you are thinking of what is discussed here: http://electronicdesign.com/analog/tips-using-high-speed-dacs-communications-design

Among other things they discuss using an 8 tone signal clustered around 1/4 the sample rate just about halfway down that page. It creates a high crest factor signal. As they are referring to it a single sine wave has a 3 db crest factor. Peak signal to RMS ratio. Conventional twin tones IMD signals would have 6 db crest factor. The 8 tone they describe as having a crest factor of 13.5. They aren't discussing this in relation to audio DACs, but rather radio use. One could easily enough create such a signal for use with an audio DAC.

BTW, Spectral uses a test signal similar to this for their analog preamps and amps. I believe they cluster the tones around 20 khz.

Meaning with traditional measurements?

Noise modulation doesn't show up clearly in the traditional measurements. In the communications field it was recognized in testing DACs that THD+N wasn't sufficient to characterize performance so another measurement was introduced using a multitone (i.e. high crest factor) signal. To my knowledge there are no standardized multitone test signals used in audio.

 

[March Audio]

Meaning with traditional measurements?

Noise modulation doesn't show up clearly in the traditional measurements. In the communications field it was recognized in testing DACs that THD+N wasn't sufficient to characterize performance so another measurement was introduced using a multitone (i.e. high crest factor) signal. To my knowledge there are no standardized multitone test signals used in audio.

Just curious, how do you know you have a noise modulation issue in the systems you have listened to?

 

[Opus111]

I suppose you are thinking of what is discussed here: http://electronicdesign.com/analog/tips-using-high-speed-dacs-communications-design

Yes - notice that they're interested in how DACs perform at low level ('small scale linearity'), not just their high level linearity. Which it seems to me is also relevant to music.

This is a very pertinent point they make:

For reconstructed amplitudes below -9 dBFS, the difference in SFDR performance between the AD9762 and the BiCMOS DAC can be as great as 10 to 15 dB, proving a DAC's resolution, static linearity, and dynamic specifications don't necessarily guarantee dynamic range! Multitone tests performed at similar update rates and output frequencies reveal that the AD9762 consistently shows better SFDR and intermodulation distortion performance.

Yet how many times do I see an analysis of DACs for audio focussing on static transfer functions and INL/DNL? Note that the BiCMOS DAC they're citing here has 2 more bits!

One could easily enough create such a signal for use with an audio DAC.

Yes - AP even provides software to do so, and that's been available for around two decades I believe. Yet there's as yet no standard stimulus tone for audio testing. So why is audio so slow to catch up I wonder?

 

[Opus111]

Just curious, how do you know you have a noise modulation issue in the systems you have listened to?

Well of course I don't know, just suspect the problems are noise modulation. Loss of dynamics - plenty of DACs (S-D ones in the main, but not limited to those) I try out in my system suck at dynamics, yet their traditional measurements (in the case of the S-D ones) are impeccable. So my hypothesis (not knowledge) is this is down to noise modulation.

 

[Blumlein 88]

Well no one test will tell all. However if I could only do two tests, that would be frequency response and IMD test tones of the 20+19khz variety. Rarely if these are good are any of the others worse. You can obviously get a picture of the noise floor somewhat while doing these of course.

Also take a look at these:

http://src.infinitewave.ca/

These are tests of downsampling. The sweep is a log sweep and if done perfectly has a completely black background in that spectrogram view. Some have artefacts there. Now you can use the same sweep and view with other signals without sample rate conversion. I often do it with a twin tone IMD. Very nice for a quick view of how something handles that signal. I have done such tests with 4 and 5 tones. I haven't found they show anything not seen in the two tone view. So one reason such signals haven't found favour may be they simply don't reveal anything extra at the audio DAC level.

 

[Opus111]

I've seen those plots before, yes a very useful resource it is. But that's nothing to do with DACs, just DSP. Other than incorrect dithering its hard to see how artifacts giving rise to non-linearities could be introduced there as these are generally FIR filters.

 

[Blumlein 88]

I've seen those plots before, yes a very useful resource it is. But that's nothing to do with DACs, just DSP. Other than incorrect dithering its hard to see how artifacts giving rise to non-linearities could be introduced there as these are generally FIR filters.

I think you misunderstood me. Yes those were about sample rate conversion which is a form of DSP.

My point is the type of graphs they show would also be excellent for showing the low level hash created with some multi-tone high crest factor test signals. So send the signals through a DAC, monitor with an ADC and such a spectrogram view could pretty cleanly show if various DACs performed better or worse than the norm. In addition rather than just fixed tones you can sweep multiple tones over time or sweep the levels over time to see the effects at various signal levels. That should uncover such noise floor modulation or distortion modulation.

 

[March Audio]

Well of course I don't know, just suspect the problems are noise modulation. Loss of dynamics - plenty of DACs (S-D ones in the main, but not limited to those) I try out in my system suck at dynamics, yet their traditional measurements (in the case of the S-D ones) are impeccable. So my hypothesis (not knowledge) is this is down to noise modulation.

OK, I would like to explore this a little. Probably best if we agree an understanding of "dynamics" first. The obvious literal definition is the difference between loud and quite sounds. Do have have any addition to that definition?

 

[Opus111]

I think you misunderstood me. Yes those were about sample rate conversion which is a form of DSP.

Fair enough, looks as though I did miss your point.

My point is the type of graphs they show would also be excellent for showing the low level hash created with some multi-tone high crest factor test signals. So send the signals through a DAC, monitor with an ADC and such a spectrogram view could pretty cleanly show if various DACs performed better or worse than the norm. In addition rather than just fixed tones you can sweep multiple tones over time or sweep the levels over time to see the effects at various signal levels. That should uncover such noise floor modulation or distortion modulation.

I must confess to being rather skeptical myself - they look to be variations on FFTs which don't seem to suit measurements of noise, at least in my understanding. Added to that the ADC would need to be considerably better than the DAC being tested, or at least produce artifacts orthogonal to these being examined in the DAC. My own hunches lead me to consider that some kind of analog signal processing (passive filtering maybe) might be required prior to digitization for the dynamic range of the ADC not to be the bottleneck in measurement. Certainly this is already the case in very low level THD+N measurements of amplifiers.

 

[Opus111]

OK, I would like to explore this a little. Probably best if we agree an understanding of "dynamics" first. The obvious literal definition is the difference between loud and quite sounds. Do have have any addition to that definition?

Perceived loudness and softness, yes. In terms of listening it also includes the ability to surprise the listener with suddenness in contrasts. -'jump factor'.

 

[Blumlein 88]

Fair enough, looks as though I did miss your point.



I must confess to being rather skeptical myself - they look to be variations on FFTs which don't seem to suit measurements of noise, at least in my understanding. Added to that the ADC would need to be considerably better than the DAC being tested, or at least produce artifacts orthogonal to these being examined in the DAC. My own hunches lead me to consider that some kind of analog signal processing (passive filtering maybe) might be required prior to digitization for the dynamic range of the ADC not to be the bottleneck in measurement. Certainly this is already the case in very low level THD+N measurements of amplifiers.

So is it your opinion noise floor modulation below -110 db is going to be an audible difference?

Further I am not sure why FFTs are unsuited to noise measures. They actually allow noise measures in the presence of considerable signal levels. Otherwise how would you propose to measure noise modulation?

 

[Blumlein 88]

Perceived loudness and softness, yes. In terms of listening it also includes the ability to surprise the listener with suddenness in contrasts. -'jump factor'.

Jump factor can sometimes be exaggerated by bursts of distortion that accompany momentary signal peaks near the limit. Simply listening would lead one to a lower fidelity playback in such a case vs. true dynamics and fidelity. For that matter compression always sounds more dynamic than true wide dynamic range. It is a great tool for that purpose, but judging such by ear is fraught with difficulty and the probability of misleading yourself.

 

[Opus111]

So is it your opinion noise floor modulation below -110 db is going to be an audible difference?

I don't have any experience to know the answer to that. Could you put more meat on the bones of this figure - in what bandwidth and in conjunction with what level of audio signal?

Otherwise how would you propose to measure noise modulation?

I don't have any proposal to make at this time. But I'm keen to make progress towards developing a measurement. Which is why I've raised the topic here in the hope that others might be able to shed some light.

 

[Opus111]

Jump factor can sometimes be exaggerated by bursts of distortion that accompany momentary signal peaks near the limit.

What mechanism would be responsible for such a 'burst of noise'? Oscillation near to clipping? if so that would count as incompetent design.

For that matter compression always sounds more dynamic than true wide dynamic range.

Got an example recording that shows this effect?

There's always the chance one's deceiving oneself, I would agree. One of the unavoidable risks of life.