Last edited:
I wrote an article for AudioXpress several years ago comparing my DIY distortion setup using an Emu sound card, voltage divider interface and software on a Windows PC compared to a professional piece of gear. I used a Standord Research audio analyzer. I spoke with Audio Precision several times, but they would not lend me a piece of gear. Maybe they were scared of the result or maybe they were not interested. Point is, you can do this on the cheap and get close to professional results.
Well, my professional experience with spectrum analysis outside the audio realm is that you really can find these issues if you perform the right tests. That is, use peak capture and let the system sweep for a significantly long time. This isn't the way measurements are usually made, of course, but the facilities are there in most test instruments at any price and most software, too.
I was easily able to track MER defects in an RF system that got completely washed out by the normal averaging functions used for the test, but still showed in BER measurements over some time in the decoded digital domain.
You don't spectral lines - just bumps in the apparent noise floor.
I was easily able to track MER defects in an RF system that got completely washed out by the normal averaging functions used for the test, but still showed in BER measurements over some time in the decoded digital domain.
You don't spectral lines - just bumps in the apparent noise floor.
Obviously you are chasing narrow rare glitches.
This maybe a rabbit hole as I would be not so confident these are related to poor PSSR of op-amps at high frequencies.
There are other ways you may capture these, common mode noise between measurement and DUT,
magnetic fields inducing spikes in small wire loops etc.
A viable option to avoid possible interference are 1uF-MLCC blocking caps between each op-amp supply pin and gnd plane of pcb.
Generally a low inductive pcb design with short traces is the goal.
Prefer SMD parts except film caps and other bulky stuff.
And than testing immunity against cell-phone noise.
This maybe a rabbit hole as I would be not so confident these are related to poor PSSR of op-amps at high frequencies.
There are other ways you may capture these, common mode noise between measurement and DUT,
magnetic fields inducing spikes in small wire loops etc.
A viable option to avoid possible interference are 1uF-MLCC blocking caps between each op-amp supply pin and gnd plane of pcb.
Generally a low inductive pcb design with short traces is the goal.
Prefer SMD parts except film caps and other bulky stuff.
And than testing immunity against cell-phone noise.
Last edited:
Who is? Not necessarily me. Consider spectral line noise skirts from a dac, for example. They consist of amplitude noise mostly from Vref (AN), and phase noise from the clocking system and from the dac itself (PN). AN and PN have a little bit differently shaped noise skirts, but they can be hard to disentangle.
What does it mean to have a wide noise skirt due to PN? It means the frequency of the signal is changing. The skirt is higher towards the center of the spectral line because clocks do not spend much time widely deviated from their nominal operating frequency. In any case, the spectral line at full 0dBFS level may be shifting around in frequency. The height of the skirt is a measure of the percentage of time the full scale test signal is at the wrong frequency and by how much it is off from nominal.
In the case of AN, we also have the full 0dBFS spectral line from the test signal that is moving a little bit up and down in amplitude around that 0dBFS average level.
because the AN and PN are multiplied by the dac analog output (they are convolved in the frequency domain view), sideband frequencies are created from the test tone signal being amplitude and frequency modulated.
Moreover, such modulation is NOT a merely weak nonlinearity. It is full blown modulation. Thus one cannot just add up frequency partials and analyze them as though the frequency domain display is a precise representation of the time-domain signal. The validity of that assumption rests on no more than weak nonlinearity being present. IOW, since FT is linear transformation from one domain to another it is only strictly true when dealing with linear systems. Okay, we can assume weak-nonlinearity is close enough. Strongly nonlinear systems are another matter. Of course, we can run the transform math on a strongly nonlinear system and get some numbers, but doesn't mean the numbers we get are still reliable indicators of the behavior of the TD system.
Mark,
With the right equipment we can see everything, even buried in the noise floor. When you get rid of the problems, the reported effects you are talking about disappear. That is unless those effects are entirely mental.
I can't tell you enough times, or strongly enough, ad hoc testing doesn't lead anywhere. If you can't test to the levels others do, or know how to interpret the results, you'll be stuck not knowing. At that point, conjecture runs the show.
With the right equipment we can see everything, even buried in the noise floor. When you get rid of the problems, the reported effects you are talking about disappear. That is unless those effects are entirely mental.
I can't tell you enough times, or strongly enough, ad hoc testing doesn't lead anywhere. If you can't test to the levels others do, or know how to interpret the results, you'll be stuck not knowing. At that point, conjecture runs the show.
You can get rid of non-PSS signals by averaging FFT data before discarding phase information. That's all. For the third time recently, that's how the math works. So if you want to get rid of all non-PSS signals you can. But you will be throwing away things humans can hear. Again, humans can hear more than just clear, distinct spectral lines.
If you want to know how to disentangle noise skirts into AN and PN components, then I would be happy to post some recent research on the subject. Since the tech is new, and because the PN of a dac chip has been measured and presented for the first time ever, pretty sure no one here has measured it before.
If you want to know how to disentangle noise skirts into AN and PN components, then I would be happy to post some recent research on the subject. Since the tech is new, and because the PN of a dac chip has been measured and presented for the first time ever, pretty sure no one here has measured it before.
Regarding clocks and voltage references. If they are poor enough quality, you will have problems. However, the poor equipment has nothing to do with equipment designed competently. We measure peak aberrations in spectral purity with ease. With RF, like cell phones for example, it is very important as it affects data errors. We use a constellation display or diagram to express these things. For radar, it will affect ranging and bearing depending on the measurement.
Anyway, beating audio over the head with RF issues is great for an argument, but ultimately entirely pointless. What you need to do is look at real effects at audio frequencies, then what the human body is mechanically capable of perceiving. Nothing to do with how we process, simply the limits of our senses. As soon as you attempt to introduce how our brains might process it, the doors are wide open and anything goes. This serves zero purpose except to maybe show how much you might know about psychology and maybe win an academic argument that doesn't matter. Besides, we cannot directly control how a person's brain will interpret anything at any specific point, so why even bring it up?
Anyway, beating audio over the head with RF issues is great for an argument, but ultimately entirely pointless. What you need to do is look at real effects at audio frequencies, then what the human body is mechanically capable of perceiving. Nothing to do with how we process, simply the limits of our senses. As soon as you attempt to introduce how our brains might process it, the doors are wide open and anything goes. This serves zero purpose except to maybe show how much you might know about psychology and maybe win an academic argument that doesn't matter. Besides, we cannot directly control how a person's brain will interpret anything at any specific point, so why even bring it up?
Very good question. One reason is because what we are not good at measuring is where our designs are not optimized. Some devices that measure quite well on an AP machine can have obvious audible flaws. I have two such devices here that I am keeping as evidence in case anyone wants to come see for themselves. The devices were both designed by a member of this forum who has very good test equipment. Because the devices are from a fellow forum member I would prefer to keep the problems between that individual and myself for now.
I can se a lot of word salad in some posts, without any proof, any facts, any real theory supporting. Just opinions, nothing else. Mostly pointless. Exercise in writing.
How about this basic fact: An FFT (or DFT) gives a measure of signal correlation with bin frequencies? That's true whether a signal is steady-state or not.
A lot of consequences fall out from that, at least if one understands the mathematical concept of correlation and how it is calculated in discrete systems.
A lot of consequences fall out from that, at least if one understands the mathematical concept of correlation and how it is calculated in discrete systems.
Obviously you did not get my point.
Your decades of working as a repair guy do to not entitle you to lecture me with that arrogant attitude you do.
And no, I will not boast with my merits of the past here.
So what? Bin width tells you about maximum frequency resolution only. Are you afraid of something hidden between the bins? 😉
And you always may use both time domain view and frequency domain view.
I would appreciate a suggestion what is "unmeasured" recently, except for human emotions.
Not of that nor of you. What is between bins has partial correlation with the two adjacent bin frequencies. The reason why should be obvious, I would think. Have you ever calculated a DFT graphically? It might help lend some further and deeper insight.
Regarding unmeasured, up until recently the contribution of a dac chip to phase noise in the analog output signal was never measured.
Nor have I seen meaningful measurements of lateral localization cues and depth cues in stereo audio reproduction systems. The cues are well known and given in perceptual science literature.
https://en.wikipedia.org/wiki/Sound_localization
That said, I have copies of a few general scientific articles on characterization of multi-channel sound field cues. Don't recall offhand any good tests that can be done on audio electronics with equipment like an AP machine, or in something like a computer app.
Last edited:
I dug out my original (classic?) E1DA Cosmos ADC that has no isolation built in and just made some measurements that I hope are relevant.
The first is a Topping E50 DAC, my usual source, powered by battery. This is followed by an Autoranger into the ADC. Unbalanced output from the DAC. The DAC was run at -10 dBFS. 997 Hz tone, to avoid the USB glitch potential.
Now, add a Topping HS02 USB isolator, which I also had to dig out. All else the same.
OK, here is the balanced output of the DAC. No USB isolator.
Now, with the USB isolator.
My previous measurement DAC was a Topping D10s. It only has unbalanced output and is powered from the USB Vbus. Otherwise, the same. No USB isolator first.
With the USB isolator.
Finally, just for reference, this is my current test set-up. The ADC is an E1DA Cosmos ADCiso, which is also a higher grade ("0") than the classic ADC above.
I guess my only real point here is that different pieces of equipment have different susceptibilities with regard to garbage currents running around. There are known ways to improve performance, both for test systems and for the audio system in your listening room (just for me, the test equipment is a tool to get the best sound in the living room - other people have other goals, which is really just a different part of the hobby). Some of the imperfections shown above would probably be audible in an actual sound reproduction system. Yet, I rarely see any of these characterized.
Edit: Just now I realized that the input sensitivity for the Cosmos ADC Classic was set to 2.7 Vrms, while my ADCiso is set for 4.5 Vrms. That plus the gain difference between the Autoranger and the Cosmos Scaler explain why the tone level in the last plot is lower than in all the rest. Sorry about that...
The first is a Topping E50 DAC, my usual source, powered by battery. This is followed by an Autoranger into the ADC. Unbalanced output from the DAC. The DAC was run at -10 dBFS. 997 Hz tone, to avoid the USB glitch potential.
Now, add a Topping HS02 USB isolator, which I also had to dig out. All else the same.
OK, here is the balanced output of the DAC. No USB isolator.
Now, with the USB isolator.
My previous measurement DAC was a Topping D10s. It only has unbalanced output and is powered from the USB Vbus. Otherwise, the same. No USB isolator first.
With the USB isolator.
Finally, just for reference, this is my current test set-up. The ADC is an E1DA Cosmos ADCiso, which is also a higher grade ("0") than the classic ADC above.
I guess my only real point here is that different pieces of equipment have different susceptibilities with regard to garbage currents running around. There are known ways to improve performance, both for test systems and for the audio system in your listening room (just for me, the test equipment is a tool to get the best sound in the living room - other people have other goals, which is really just a different part of the hobby). Some of the imperfections shown above would probably be audible in an actual sound reproduction system. Yet, I rarely see any of these characterized.
Edit: Just now I realized that the input sensitivity for the Cosmos ADC Classic was set to 2.7 Vrms, while my ADCiso is set for 4.5 Vrms. That plus the gain difference between the Autoranger and the Cosmos Scaler explain why the tone level in the last plot is lower than in all the rest. Sorry about that...
Last edited:
Same issue below. I can see lot of failed measurements here. Worse, people have their systems wired wrong, like 2 class I devices with SE link interconnection. These are real issues, non-esoteric and easily measurable (though the frequencies are non-harmonic, FFT will find them, even as discrete or noise modulation if they are traveling), before they become audible.
However, the stories sell. Remember BT line of threads.
Hi bucks bunny,
I am not lecturing anyone. I'm also not trying to be arrogant in any way. Mind you, much training I had was from an Austrian, maybe that comes through. Never did I address you, I was responding to Mark.
I was trained in the test and measurement field and worked in a calibration lab, and in telecommunications. Many related things to audio. So I do some design, lots of service on audio equipment. I'm not a better engineer than most, but I'm not too bad at applying things to reality.
Just a comment on clock stability and phase. Look at the clock frequency and sample rates. If clock stability causes errors in data, we have a problem. Even if such errors occur, what is the time difference? Sample time difference? Output filter slope and passband? Okay, can we hear that much jitter? I know we can't. Now, master clock errors are common to both channels. Humans detect phase how? Not an internal clock, we detect phase difference between our ears. How fine a difference can we detect? Right, for normal jitter we can't if you look at the frequencies involved. They would be common to both our ears anyway.
So for normal systems that operate well enough (not junk), can we safely ignore this? We can run around the bush forever, but apply this to reality.
I am not lecturing anyone. I'm also not trying to be arrogant in any way. Mind you, much training I had was from an Austrian, maybe that comes through. Never did I address you, I was responding to Mark.
I was trained in the test and measurement field and worked in a calibration lab, and in telecommunications. Many related things to audio. So I do some design, lots of service on audio equipment. I'm not a better engineer than most, but I'm not too bad at applying things to reality.
Just a comment on clock stability and phase. Look at the clock frequency and sample rates. If clock stability causes errors in data, we have a problem. Even if such errors occur, what is the time difference? Sample time difference? Output filter slope and passband? Okay, can we hear that much jitter? I know we can't. Now, master clock errors are common to both channels. Humans detect phase how? Not an internal clock, we detect phase difference between our ears. How fine a difference can we detect? Right, for normal jitter we can't if you look at the frequencies involved. They would be common to both our ears anyway.
So for normal systems that operate well enough (not junk), can we safely ignore this? We can run around the bush forever, but apply this to reality.