I forgot to mention $250.000 mono-block amplifiers with lots of distortion, 7th even.
Single waveform capture exists on some analog scopes, good luck finding one these days. You pose things that on the surface seem trivial to test. That is things like a slow 16bit LSB staircase examined on a sample by sample basis out of a DS DAC. Last week I ran the test disk waveforms through a cheap Western Digital media server and did not observe anything special.
You understand my trepidation on wasteing my time making pictures.
Scott,
I have a 100MHz analogue Tek one that does. And a digital HP 150MHz one that does. Either one is decades old and was made in large quantities.
But also, forgive me, the lack of tools to observe what I am talking about should not be construed as grounds to dismiss the notion, should it not?
With respect, the "most advanced" gear I have access to is an AP2, the rest are fairly dated analogue and digital generators and 'scopes (though many are of the "very fast" kind) plus a bunch of modern external sound-cards all equipped with DS converters and a wide range of software.
So forgive me, but given the knowledge I gained at university in my EE course at university and in practice since, plus a fairly basic set of tools, which may exceed those available to an average amateur, but not those available to a commensurate professional, I do find them trivial to test.
So I fail to appreciate why anyone would declare them to be difficult to measure. However this failing to understand is no doubt mine. I would appreciate if you could bring some light to this darkness...
Forgive me for asking, what specifically (measured quantity) special did you not observe and how (method) did you not observe it?
And what model of "cheap Western Digital media server" did you employ in your test (as there are several)?
And what test waveforms from which test disk did you employ?
Knowing those minor details may help others to repeat your experiments and find (or find not) "nothing special"...
Past that, my suggestions of experimentation are relatively easy to implement and repeatable and will show the same results.
In fact, I think the observed results which I stipulated have never been called into question as such. The actual debate seems centred around the interpretation of the results more than on the actual test results.
In this whole debate, including all the resulting consequences, like my posts being placed under moderation, I do not remember anyone per se questioning my suggested possible observations or anyone who has been illustrating that what I have asserted to have observed, is in fact false.
Ciao T
I have a 100MHz analogue Tek one that does. And a digital HP 150MHz one that does. Either one is decades old and was made in large quantities.
But also, forgive me, the lack of tools to observe what I am talking about should not be construed as grounds to dismiss the notion, should it not?
With respect, the "most advanced" gear I have access to is an AP2, the rest are fairly dated analogue and digital generators and 'scopes (though many are of the "very fast" kind) plus a bunch of modern external sound-cards all equipped with DS converters and a wide range of software.
So forgive me, but given the knowledge I gained at university in my EE course at university and in practice since, plus a fairly basic set of tools, which may exceed those available to an average amateur, but not those available to a commensurate professional, I do find them trivial to test.
So I fail to appreciate why anyone would declare them to be difficult to measure. However this failing to understand is no doubt mine. I would appreciate if you could bring some light to this darkness...
Forgive me for asking, what specifically (measured quantity) special did you not observe and how (method) did you not observe it?
And what model of "cheap Western Digital media server" did you employ in your test (as there are several)?
And what test waveforms from which test disk did you employ?
Knowing those minor details may help others to repeat your experiments and find (or find not) "nothing special"...
Past that, my suggestions of experimentation are relatively easy to implement and repeatable and will show the same results.
In fact, I think the observed results which I stipulated have never been called into question as such. The actual debate seems centred around the interpretation of the results more than on the actual test results.
In this whole debate, including all the resulting consequences, like my posts being placed under moderation, I do not remember anyone per se questioning my suggested possible observations or anyone who has been illustrating that what I have asserted to have observed, is in fact false.
Ciao T
Last edited:
It is unfortunate, some of us have thicker skin than others. I did not consider what I did as a "test". It was as I posted before, the -80dB sinewave passed through an ideal quantizer as on the CBS/SONY test disk, the horrible one that you said was the "truth". The real time waveform captured looked like the 16bit steps plus noise, a lot less than the 100's or whatever LSB's claimed. Though I still can't quite follow what is claimed. The WD media server I'm sure contains a ubiquitous cheap 24bit DS audio converter. Either I don't understand what you claim or everthing below 10 or 12 bit should be nothing but noise so on a real time basis I should see no 16bit steps, but I do. PMA's picture, which you loved, shows exactly what you would expect 12bit LSB steps plus about an LSB rms noise, it is a 12bit DAC after all.
BTW - I was refering to the old analog scopes with the variable persistence phosphor, I don't collect instruments and we have been forbidden to pay for repair on older scopes.
Last edited:
I've been following this pretty carefully but have apparently missed the suggested experimentation. I have plenty of analog gear fast enough to see the issue if you could point at a *specific* approach.
Thanks,
Chris
I would like to say something here in the spirit of why this thread exists, and why it has lasted so long. Technical details are nice, but they DON'T tell you specifically how a piece of audio equipment sounds, always. That is the problem that I continually try to address here, and virtually 'smirk' when some well meaning, well educated, engineer insists that it is 'all in the measurement' and measurement is based on solid mathematical theory. Therefore it must be true and perfect, and will always make accurate predictions. Hey, I've been there, done that, more that 40 years ago. It was my FAILURE to have my predictions be successful, that I finally saw the 'light'. That is when the Grateful Dead and Mark Levinson became excited with my designs, and it has been that way, more or less, ever since.
I had to go beyond the mathematical correlation between distortions, etc, and just try new ideas, based on NO COMPROMISE, at least at first, low feedback, high open loop, fast designs. It is like exploring uncharted territory. Later, when the territory is established, the road can be cleared to make cheaper, more compromised designs.
This was the essence of the original Levinson JC-2 preamp, and the Blowtorch preamp. Both were all-out efforts in their time (1973) and around the year 2000. Both were made with the same enthusiasm of beating all the competition, sonically, if possible.
We did pretty well with both, in their time and place.
The JC-80, originally a successor to the Levinson JC-2, was a 'mixed bag', with problems that we had to live with, but it still did pretty well, as does the Parasound JC-2, the preamp that has twice the features as the Blowtorch at 1/4 the price. They are not as successful as the Blowtorch, sonically, but it would be almost impossible to tell why by measurement. That is the crux of the matter.
Now how about that $250,000 Japanese power amp? I heard it once, I'm pretty sure that it was that specific example. It was at a CES, maybe 5+ years ago. I was walking down the hall just listening to the competition when I heard a record playing Ella Fitzgerald singing a torch song. I was amazed at the playback, as if Ella herself were singing to me. I will never forget that experience. But I did not know anything specific about the audio playback until after listening. I had never heard of these people, but I sure was surprised with the sound. That is the essence of my experience. If it works, it works, WHY is something I try to understand.
I had to go beyond the mathematical correlation between distortions, etc, and just try new ideas, based on NO COMPROMISE, at least at first, low feedback, high open loop, fast designs. It is like exploring uncharted territory. Later, when the territory is established, the road can be cleared to make cheaper, more compromised designs.
This was the essence of the original Levinson JC-2 preamp, and the Blowtorch preamp. Both were all-out efforts in their time (1973) and around the year 2000. Both were made with the same enthusiasm of beating all the competition, sonically, if possible.
We did pretty well with both, in their time and place.
The JC-80, originally a successor to the Levinson JC-2, was a 'mixed bag', with problems that we had to live with, but it still did pretty well, as does the Parasound JC-2, the preamp that has twice the features as the Blowtorch at 1/4 the price. They are not as successful as the Blowtorch, sonically, but it would be almost impossible to tell why by measurement. That is the crux of the matter.
Now how about that $250,000 Japanese power amp? I heard it once, I'm pretty sure that it was that specific example. It was at a CES, maybe 5+ years ago. I was walking down the hall just listening to the competition when I heard a record playing Ella Fitzgerald singing a torch song. I was amazed at the playback, as if Ella herself were singing to me. I will never forget that experience. But I did not know anything specific about the audio playback until after listening. I had never heard of these people, but I sure was surprised with the sound. That is the essence of my experience. If it works, it works, WHY is something I try to understand.
Now we get to the meat of the argument. Is 1 LSB an absolute limit or a statistical average?
When I have designed I expect the maximum error to be listed, not the RMS value.
If it is RMS then there will be a lot of correlated noise, which for an A/D can mean real problems down the line. For a D/A to me it means it losses a few bits of accuracy.
The AP2 (indeed all traditional analog distortion analyzers) are analog from input to output with the only storage/averaging being the meter. It would be trivial to connect the output of a digital source of the different types (D/S and ladder dac) through the analyzer and look at the distortion residual. Scaled the same with the same bandpass filters the different noise should be obvious on a traditional scope or even one shot on a typical digital scope. Unless there is something I'm missing here, possibly that the noise issue is only present during specific transients, fuzzy distortion/noise should be easily observed.
I don't have access to a ladder dac right now so I can't try this myself.
Scott:
If AD won't repair a scope where do they go? Might there be deals we should know about?
what’s so hard about looking at the datasheet, reading some of the many app notes explaining the terms, conditions for measuring ADCs
the manufacturer's have a huge interest in customers understanding the specs, getting the most from the part, and not being blamed for the customer’s naive personal imagining of what they think/wish the specs mean or misunderstanding the physical world limits of measurement with noise
for PMA’s plot of AD922x output we can just look a “bullet item” DS #
Integral Nonlinearity Error: 0.5 LSB
Differential Nonlinearity Error: 0.3 LSB
Input Referred Noise: 0.09 LSB
That and a look at the plot lead to my comment that the hardware implementation or test setup has a noise issue – way beyond the chip’s performance spec
without more info - what ref V, source noise, Z, filter frequency, ... there's little that can be definitively learned from the plot - or even if it really shows a "problem" that shouldn't be anticipated by the operating conditions, system noise engineering calculations
the manufacturer's have a huge interest in customers understanding the specs, getting the most from the part, and not being blamed for the customer’s naive personal imagining of what they think/wish the specs mean or misunderstanding the physical world limits of measurement with noise
for PMA’s plot of AD922x output we can just look a “bullet item” DS #
Integral Nonlinearity Error: 0.5 LSB
Differential Nonlinearity Error: 0.3 LSB
Input Referred Noise: 0.09 LSB
That and a look at the plot lead to my comment that the hardware implementation or test setup has a noise issue – way beyond the chip’s performance spec
without more info - what ref V, source noise, Z, filter frequency, ... there's little that can be definitively learned from the plot - or even if it really shows a "problem" that shouldn't be anticipated by the operating conditions, system noise engineering calculations
Last edited:
Demian,
The AP system 2 I have uses an A/D to capture the data. I have the optional high resolution converter. I can reduce the accuracy and get higher bandwidth. Most folks are careless in AP2 vs System 2. I gather for some money I can have it turned into a system 2722 or some such. But I suspect a Stanford Research analyzer is a better choice.
Now as tempting as it is to get a low cost oldie scope, I prefer to build my test equipment, or assemble systems out of new core components.
With that in mind I am starting to build a noise analyzer. I am aiming for .1 to 1,000,000 hz. range. I expect actually to only break that into octaves and use a peak detector. I might also consider some sort of slow decay detector. Due to the limited bandwidth I do not expect to consider RMS. The DC level would go to a log converter then into a multiplexor. The output would be a digital scope display.
I can use Biquad filters for all but the highest range. The version that starts with the LPF stage also adding gain is the preferred approach. I suspect I need to play with a biquad or LC for the top octave. I may need to use a discrete front end for the low frequency stages.
I haven't yet come up with a good decision on the input impedance. I expect to be able to get within 3 db of the noise floor.
ES
The AP system 2 I have uses an A/D to capture the data. I have the optional high resolution converter. I can reduce the accuracy and get higher bandwidth. Most folks are careless in AP2 vs System 2. I gather for some money I can have it turned into a system 2722 or some such. But I suspect a Stanford Research analyzer is a better choice.
Now as tempting as it is to get a low cost oldie scope, I prefer to build my test equipment, or assemble systems out of new core components.
With that in mind I am starting to build a noise analyzer. I am aiming for .1 to 1,000,000 hz. range. I expect actually to only break that into octaves and use a peak detector. I might also consider some sort of slow decay detector. Due to the limited bandwidth I do not expect to consider RMS. The DC level would go to a log converter then into a multiplexor. The output would be a digital scope display.
I can use Biquad filters for all but the highest range. The version that starts with the LPF stage also adding gain is the preferred approach. I suspect I need to play with a biquad or LC for the top octave. I may need to use a discrete front end for the low frequency stages.
I haven't yet come up with a good decision on the input impedance. I expect to be able to get within 3 db of the noise floor.
ES
It is when a manufacturer uses a slightly different method to yield the numbers, the devices are new and there are no common standards, or the manufacturing process has change. Not to mention getting clever and using parts in a new way. THEN THERE IS THE DIFFERENCE BETWEEN TYPICAL AND GUARANTEED!
I have never met Pavel, but I am pretty sure he has a scar or two from picking up the wrong end of a soldering iron. He is perfectly capable of evaluating parts and designs. He also shows insight into issues others just don't grasp. If he asks a question it is worth a bit to evaluate it. So if he makes a mistake, he will probably find it.
not so worried about PMA as much as the instant jumping on the plot as "evidence" in the "fuzzy distortion" argument when next to nothing about the conditions of the hardware, test setup are known
there is apparently no recognition of what ADC response can be expected to look like with well understood noise and quantization
whether inband added TPDF dither is audibly modulated by the signal is a psychoacoustic question – but even exaggerated to 8 bit word quantization the dither noise wasn’t perceived as being signal correlated
how does that analog master tape noise sound with 48 dB gain - much signal correlation? (assuming you've had to use Dolby to get acceptable noise floor in the 1st place)
high frequency shaped noise in delta-sigma ADC, DAC does have to be filtered out – but with modulator frequencies 128x fs and higher there is some “room” for the analog filter
there is apparently no recognition of what ADC response can be expected to look like with well understood noise and quantization
whether inband added TPDF dither is audibly modulated by the signal is a psychoacoustic question – but even exaggerated to 8 bit word quantization the dither noise wasn’t perceived as being signal correlated
how does that analog master tape noise sound with 48 dB gain - much signal correlation? (assuming you've had to use Dolby to get acceptable noise floor in the 1st place)
high frequency shaped noise in delta-sigma ADC, DAC does have to be filtered out – but with modulator frequencies 128x fs and higher there is some “room” for the analog filter
Last edited:
The recycle bin non-leadfree waste the same place that the old monitors go.
Mic Article
OK Scott; I am sick of wiping up drool....when is this article coming out in Linear Audio?
Looking forward to learning more...
Howie
Howard Hoyt
CE - WXYC-FM
UNC Chapel Hill, NC
www.wxyc.org
1st on the internet
...though our stream encoder is on it's last legs and soon to be replaced...
OK Scott; I am sick of wiping up drool....when is this article coming out in Linear Audio?
Looking forward to learning more...
Howie
Howard Hoyt
CE - WXYC-FM
UNC Chapel Hill, NC
www.wxyc.org
1st on the internet
...though our stream encoder is on it's last legs and soon to be replaced...
John,
I should think so.
My recommendation would be to employ a sawtooth signal at around -60dBFS (below that you need active probes with gain with most 'scopes) with a frequency is a integer quotient of the sample rate with a reasonably high frequency , so maybe Fs/8 or Fs/16.
This way you can ignore any interference effects from "cycle slips" in the digital sampling.
Setting the scope to infinite persistence (if it is a digital storage scope) will show the resulting error band, if there are random errors they will lead to a widening of the signal lines displayed, the degree of widening allows an estimation of the degree of "fuzzy distortion" or noise present.
Using single shot will show the waveform. Ideally here one overlays the two traces and uses dual channels, one DS DAC the other MB.
Some here have made the point that things should be viewed only after the analogue post filter. THis is true to a degree, HOWEVER it is also quite instructive to look directly at the output of the DAC as post filtering can only attenuate any given artefacts, but not remove them.
Ciao T
I should think so.
My recommendation would be to employ a sawtooth signal at around -60dBFS (below that you need active probes with gain with most 'scopes) with a frequency is a integer quotient of the sample rate with a reasonably high frequency , so maybe Fs/8 or Fs/16.
This way you can ignore any interference effects from "cycle slips" in the digital sampling.
Setting the scope to infinite persistence (if it is a digital storage scope) will show the resulting error band, if there are random errors they will lead to a widening of the signal lines displayed, the degree of widening allows an estimation of the degree of "fuzzy distortion" or noise present.
Using single shot will show the waveform. Ideally here one overlays the two traces and uses dual channels, one DS DAC the other MB.
Some here have made the point that things should be viewed only after the analogue post filter. THis is true to a degree, HOWEVER it is also quite instructive to look directly at the output of the DAC as post filtering can only attenuate any given artefacts, but not remove them.
Ciao T
Excellent point.
I would like to add that some HD recordings being offered as 24-bit PCM are actually SACD recordings that have been converted without the proper band-limiting. Apparently, in the quest to provide 24/96, which SACD claims to match, some engineer hasn't noticed that a lot of shaped noise is still in the data, and that won't be filtered out by a standard 24/96 converter. The SACD players have aggressive low-pass (in that they do not have nearly the same bandwidth as most 24/96 systems), but it's potentially a bad idea to convert these to PCM and listen without the proper signal path.
Personally, I'm not too worried about this. I simply avoid SACD sources that have been converted to PCM.
Well put, but don't forget that audio engineers have learned a thing or two about measuring sound performance in those 40 years. I'm sure we'll never be perfect at measuring sound performance any more than we'll be able to make a perfect amplifier. But a true scientist seeks to improve both the performance of the invention as well as the measurement process used to confirm the performance.
It's possible to go wrong in either direction. Those who give up entirely on measurement just because the technology is not as good as the human hearing system are easily missing out on the full picture. Likewise, those who ignore evaluations from real human beings because of some blind faith that measurements can tell the whole story are missing out on the other parts of the full picture.
Personally, I do not think that either of these two broad categories of measurement are sufficient alone. It is well established that listeners prefer louder sources, that they can ignore vast amounts of distortion of certain types, while other types of distortion can be grating in relatively tiny amount, and that certain aspects of dynamic distortion in particular media recording technologies can fool the ear into thinking one source is 'better' than another when in the final analysis it was merely 'louder' than another and it was the metering of the loudness that failed to represent the human perception. I would prefer that the audio industry seek to improve both the quality of gear (amplifiers) and the quality of testing, with a healthy amount of scientific skepticism for both sides.
It seems to me that understanding measurement processes as well as the particular numerical results can help narrow down the interpretation of the results. Tests can establish certain boundaries even if they cannot entirely predict 'bad' sound versus 'good' sound. If a test shows higher distortion in one design than another, and the test is reasonably precise about which harmonics are louder or softer, then we at least know something objective about the equipment. Perhaps some listeners will prefer one mix of harmonics than another. It's been said many times in this thread that a higher total distortion may sound better than a lower total distortion, mostly because of the relative amplitudes of individual harmonics. Rather than say we should only test with objective equipment or only test with subjective listeners, what's wrong with testing with both techniques?
- Status
- Not open for further replies.