Hi Mark,
Again. The equipment isn't at fault. You need to use the appropriate equipment for the task, and know the limitations. That means you must understand what you are doing and what you are testing for.
Mark, you really need to roll up your sleeves and do some real testing with real equipment to understand. You can't just read about it. You have to actually do it, and more than a workshop event. You must gain an understanding or you will forever be at odds with what we do.
and the quote ...
and now the rest of what Demian said ...
You basically edited what was said to suit. The equipment measuring the result wasn't an AP or anything else commonly used for measuring distortion. That equipment commonly reaches about 90 KHz, human hearing is much lower in frequency. Some audio analyzers (AP, Keysight) can reach over 1 MHz. Note that the bandwidth is needed to be up over 1 MHz or even 5 MHz as Demian said. You can see the distortion with our normal equipment, but not what caused it. For that, you need a spectrum analyzer reaching several MHz.
Again. The equipment isn't at fault. You need to use the appropriate equipment for the task, and know the limitations. That means you must understand what you are doing and what you are testing for.
Mark, you really need to roll up your sleeves and do some real testing with real equipment to understand. You can't just read about it. You have to actually do it, and more than a workshop event. You must gain an understanding or you will forever be at odds with what we do.
For years I known about the RF noise and the effect it has on opamps. Other people have known too. If you can reduce the RF noise before the opamp then the sound can subjectively get better. Its because the opamps respond to it in ways that are non-PSS that the measurements don't look much different.
But the claim must be subjective because standard measurements don't show it well. Nothing new there.
EDIT: I have looked at dac output noise with 600MHz scope sampling at 2.5GHz. I have measured square waves at half the BLCK frequency buried in the noise at the dac analog outputs before the I/V opamp. I measured that same square wave at the HPA output of a Neurochome HP-1 set to maximum gain and attached to he dac output. I made sure it wasn't some stray coupling or ground coupling of the clock signal too. I am not new to difficult measurements at all. But I am also not new a deducing what must be true from available data when the measurements I would like to have are not possible.
But the claim must be subjective because standard measurements don't show it well. Nothing new there.
EDIT: I have looked at dac output noise with 600MHz scope sampling at 2.5GHz. I have measured square waves at half the BLCK frequency buried in the noise at the dac analog outputs before the I/V opamp. I measured that same square wave at the HPA output of a Neurochome HP-1 set to maximum gain and attached to he dac output. I made sure it wasn't some stray coupling or ground coupling of the clock signal too. I am not new to difficult measurements at all. But I am also not new a deducing what must be true from available data when the measurements I would like to have are not possible.
Last edited:
When you send signals to an op amp it can't respond to, it goes open loop, the feedback loop is broken. The op amp will then behave in ways not intended. This is no surprise. The filters before the op amp in the example reduced the HF energy to levels the op amp could handle.
The sampling isn't noise. It is high amplitude high frequency energy. The people who designed the circuit in question should know this and have filtered the input of the op amp in the initial design. Nothing here is unknown or a surprise.
The sampling isn't noise. It is high amplitude high frequency energy. The people who designed the circuit in question should know this and have filtered the input of the op amp in the initial design. Nothing here is unknown or a surprise.
Yes, I know. The usual problem is that modern low output Z dacs such as those made by ESS need the analog output node held tightly at virtual offset ground by the I/V opamp or else measured distortion increases (and its audible too). Allo made output stages for Katana dac in two versions, one with a pre-filter before the I/V opamp, and one without the filter. The one without the filter measured better. Some people thought the one with the filter sounded better, so they named the SQ version. I didn't like the distortion in that version so I stuck with the non-filtered version.
Also, I was in private discussions with the chief designer at Allo. We both agreed more passive filtering was needed before the usual balanced MFB filter stage because remaining RF after the I/V stage would still cause problems with the next opamp stage.
Also there is a problem with adding input capacitance in front of an I/V opamp. It causes frequency peaking which can be compensated for, but it kind of ends up being a wash then.
So, I would be interested to know how @1audio got a big improvement without paying the usual cost of the pre-filter.
Last edited:
I'm curious how you measured a square wave that was buried in the noise.
But even if you did find traces of the BCLK in the analog output - and that wouldn't surprise me, especially if the I/V or post-filtering was sub-optimal - it would be at 2*24*fs or 2*32*fs, depending on the bit depth. So for 24-bit stereo at 44.1 kHz, BCLK would be 2.1 MHz. Isn't that an example of what @1audio said and that you argued against in Post 219?
Great! Link? Are we allowed to critique your experiment and conclusions?
Tom
Well, a current output DAC is like that.
Eval kits are not optimized, they are not ready for production in a product. It's easy to add filtering, you decouple first. A simple resistor is the easiest way.
Eval kits are not optimized, they are not ready for production in a product. It's easy to add filtering, you decouple first. A simple resistor is the easiest way.
I trigged off the clock going into the dac. Also had to put scope into high resolution mode where it oversamples 2x at each sample point.
Have to see if I can find it. You know how the search engine here works as well as I do.
BTW, it wasn't at BCLK frequency; IIRC it was half that.
Just so I understand you correctly. Did you mean post-filter? I.e., the filter that follows the DAC. Also known as the reconstruction filter.
Tom
Hi Mark,
A lock-in amplifier is what you needed. I have one.
Looking at signals buried in the noise floor is tricky. Also, be aware that scopes can lie to you, especially digital scopes. Even expensive ones.
A lock-in amplifier is what you needed. I have one.
Looking at signals buried in the noise floor is tricky. Also, be aware that scopes can lie to you, especially digital scopes. Even expensive ones.
In #219, the point I was trying to make is that sometimes we have listen to figure things out. That includes for 1audio. There should be no problem with that if we do it well enough. Nobody I know of in industry today (which includes ESS and Purifi) uses formal, publication quality, formal listening tests. Some people, including me, have done some blind testing, but it not usually full blown DBT. Still, within our available resources, sometimes we have to listen because we are usually not equipped to measure everything. Just measuring the combined phase noise of a clock and dac together is still not possible to practically do. Just measuring the phase noise contribution of the dac chip itself requires using dual synthesized clocks to null out other effects, so that the real crystal clocks can't be included in the measurement.
I built one and used it in the project I based my master's thesis on. They're useful. 🙂
My fancy analog scope does a wonderful job of receiving FM radio, so it probably wouldn't be my first choice for digging things out of the noise floor. Then again, one does not need a fancy scope to measure 2.1 MHz.
Tom
Yeah, I know about those. You have an integrating voltmeter too? NMR Gaussmeter? (just curious about your collection)
That's why we listen to things Mark. Whatever you hear is visible with test equipment if you're using the right equipment. A good FFT analyzer will show you everything you can hear, the details may be low amplitude, but they are there. They will show you things you can't possibly hear as well.
These days I listen just as a sanity test more than anything. A scratchy control may not show up readily as it is transient in nature. I can hear it, but I cannot quantify it. A 'scope can quantify the noise level. So human hearing is an okay detection method. Forget quantifying anything though. Then you drag out the appropriate piece(s) of test gear and get busy.
The art of being good is to take in all the information you can from reliable sources, and think about it, then act accordingly. That means your ears, whatever test gear you have. The intelligence to follow the right path is a learned thing. There are no mysteries, only things you don't understand but others do.
These days I listen just as a sanity test more than anything. A scratchy control may not show up readily as it is transient in nature. I can hear it, but I cannot quantify it. A 'scope can quantify the noise level. So human hearing is an okay detection method. Forget quantifying anything though. Then you drag out the appropriate piece(s) of test gear and get busy.
The art of being good is to take in all the information you can from reliable sources, and think about it, then act accordingly. That means your ears, whatever test gear you have. The intelligence to follow the right path is a learned thing. There are no mysteries, only things you don't understand but others do.
Any needle movement integrates.
So what are you referring to specifically? I use peak responding meters, average responding and true RMS responding meters. I have more meters on hand and in tolerance than you can imagine.
So what are you referring to specifically? I use peak responding meters, average responding and true RMS responding meters. I have more meters on hand and in tolerance than you can imagine.
Actually, ESS says not true with dacs, and I believe they are right. State variable settling after transients is one such thing.
Not necessarily true with Benchmark AHB2 either. Its a rail switching amp, and if it is quirky into some non-resistive loads, its probably right around in the switching region. I guess if the test setup was arranged to look just for that type of effect under specific load conditions then it should in principle be measurable. Makes it easier than the ESS example.
Aside: An integrating voltmeter reads out in units of volt-seconds.
Last edited:
I think that's the best summary of where you're coming from that I've seen from you. Thank you.
I disagree that we have to listen to figure things out. I disagree with that simply because any test that involves humans as the "analyzer" are riddled with sources of error. I can list a good handful of psychological effects that come into play. These effects are well documented. You can get around some of the sources of error with a double-blind test, but running those is pretty expensive and a double-blind test can introduce errors of its own. I spent quite a while looking at these things during my psychology degree.
I am truly amazed that all science goes out the window in audio. I would never fly in an airplane designed by engineers who based their design choices on listening to the materials for example. Or tasting them. Or how they felt about them. So why would I buy audio gear designed that way?
And, again: This thread is about (lemme check the title) FFT and Multi-Tone testing. This is not the place to get into yet another fruitless subjectivist vs objectivist discussion.
I suggest you rent an Agilent/Keysight E5052B phase noise analyzer and learn how to use it. I spent seven years of my career with National/TI designing crystal oscillators and characterizing them on the 5052. You wanna talk about pulling things out of the noise floor? That thing can measure the grass grow. It's amazing.
If you measure the phase noise of the crystal oscillator and then measure the phase noise of the clocks within the DAC (which any competent IC designer would make available as a test mode that's probably not disclosed to the public) you would easily be able to tell which of the two is the dominant noise contributor. It's not difficult to design a crystal oscillator with low phase noise. The extremely high Q of the crystal helps a lot here. So I'm willing to bet that the noise measured from the DAC will be dominated by the DAC.
Further, you can't have a DAC without providing a clock, so isn't it a moot point to try to separate the two?
To quote my advisor on my master's thesis: "It's only black magic until someone turns on the light".
Tom
Sean Olive apparently disagrees with you. JBL uses qualified listeners in blind tests of its speakers. They also find that expert listeners reduce the number test subjects needed and reduce testing time for equal accuracy.
Purifi has reported using human listening tests, some of which were done in collaboration with a university, but not to publication quality.
ESS reportedly trained all of its executive team except one guy to hear state variable settling.
Thorsten Loach used his employees in China as test subjects to do blind testing of a number of audio designs to see what differences were audible.
In both the Purifi and ESS listening tests, test subjects needed training to notice unusual sound effects that most people don't normally notice.
Moreover, most of what we know about masking and about the sound of HD partials comes from listening tests, as do all published thresholds of audibility and published absolute limits of hearing, not to mention the range of normal FR audibility.
Listening tests were done to determine how much MP3 sound could be compressed before the sound because objectionable to most people.
Last edited:
Hi Mark,
There is no end to people who may disagree. You also need to look at the date they made those statements (as I touched on earlier). Just because someone disagrees doesn't mean they are correct either.
To further what Tom stated, a standard crystal oscillator (not even ovenized or anything) is more stable and lower noise than anything else in that system. That's assuming the drive circuit for it isn't botched, or the power supply. You'll have more issues with the DAC or PCB layout, certainly the analogue sections that follow.
The rules of physics are not suspended or changed just because it's audio. In fact, other industries that use electronics require more precision and less noise. By comparison, the ability of any human to hear is a joke compared to what is routinely measured.
Also, why would I use an integrating voltmeter? That measurement doesn't apply to audio electronics. My power supplies and meters will log current vs voltage if I need anything like that. Upload into excel and we can bend it to whatever we want.
There is no end to people who may disagree. You also need to look at the date they made those statements (as I touched on earlier). Just because someone disagrees doesn't mean they are correct either.
To further what Tom stated, a standard crystal oscillator (not even ovenized or anything) is more stable and lower noise than anything else in that system. That's assuming the drive circuit for it isn't botched, or the power supply. You'll have more issues with the DAC or PCB layout, certainly the analogue sections that follow.
The rules of physics are not suspended or changed just because it's audio. In fact, other industries that use electronics require more precision and less noise. By comparison, the ability of any human to hear is a joke compared to what is routinely measured.
Also, why would I use an integrating voltmeter? That measurement doesn't apply to audio electronics. My power supplies and meters will log current vs voltage if I need anything like that. Upload into excel and we can bend it to whatever we want.
I couldn't agree more Tom!