CHANGSTAR: Audiophile Headphone Reviews and Early 90s Style BBS
Lobby => Amp and DAC Measurements => Topic started by: Marvey on July 29, 2015, 07:51:28 PM
-
I figure this might be a good spot for unified measurements for amps or DACs. I'd like to propose a set of measurements standards so we all don't do slightly different things. This keep in mind that let's not go overboard to keep number of measurements small. More measurements, unless well presented = core dump - that even makes my head spin around. Also, like to have meaningful measurements... not a lot of useless random measurements. Presenting noise is just that. Presenting noise.
Please do not comment unless you are one of the regulars who post measurements. In other words, no laymen.
Standard 1kHz distortion spectrum
I suggest loads near 33 ohms and 330 ohm loans.
Any suggestion for Vrms at these two loads? I think we should do a look at volume knob position with real headphones during real listening and start from there. Maybe push up a bit from that position.
I really don't care for this measurement because it seems most gear does well with this test, but it does provide a good baseline.
20Hz single tone distortion spectrum
I suggest loads near 33 ohms and 330 ohm loans.
Any suggestion for Vrms at these two loads?
60Hz single tone distortion spectrum
I suggest loads near 33 ohms and 330 ohm loans.
Any suggestion for Vrms at these two loads?
Two tone 600Hz and 1700Hz distortion spectrum
I suggest loads near 33 ohms and 330 ohm loans.
Any suggestion for Vrms at these two loads? I think we should keep both tones at the same level 1:1 ratio.
Two tone 60Hz and 7000Hz distortion spectrum (a la SMPTE)
I suggest loads near 33 ohms and 330 ohm loans.
Any suggestion for Vrms at these two loads?
Tones at 2:1 ratio for 60hz and 7kHz respectively. (7kHz signal -6db down)
FR
Audio band
Wide band
As far out as possible (limited by sampling rate of ADC and signal generators)
I'd like a three to five tone, but we need to figure that one out. Need a fancy signal generator or wave file.
-
For the single tone distortion measurements I would try 550mVrms which represents ~1mW into a 300 ohm load. Many dynamic headphones like the HD800 and DT880 250 ohm, for example, are 95-100dB efficient with 1mW of input power. This will be loud on rock music and provide comfortable transient peaks for classical. I would then level match for a popular lower efficiency planar and note that voltage for the 33ohm test. This could be a good starting point.
I also liked your previous choice to test at 40Hz which is a reasonable bass cutoff for a lot of music.
-
I've done 1 Vrms and 0.4 Vrms mainly because other folks have done it that way, but 0.55 Vrms is a good and reasonable number going forward.
We could do the classic 19 and 20 kHz CCIF to see abnormal out of band distortion artifacts due to amp near instability or so.
What I do for the voltage on two tones is that I calibrate voltage with a 1 KHz signal (which I also use for THD). Then I do 2 tone while keeping the level constant from where I had it with the THD measurement.
Note the nice thing about SMPTE is that 7000/60 = irrational number meaning that one can separate IMD from THD. So I would suggest something like 600Hz and 1700Hz or similar instead of 500Hz and 1800Hz.
-
We could do the classic 19 and 20 kHz CCIF to see abnormal out of band distortion artifacts due to amp near instability or so.
I second CCIF or something similar.
-
For the 60Hz/7kHz SMPTE test, shouldn't the ratio be 4:1?
About the 19+20kHz CCIF test, how meaningful is this towards the perceived listening experience? It is interesting from an academic perspective, but does not seem to me to reflect as much a real-world condition as does the 60/7000Hz SMPTE test.
-
I can see CCIF being a good measurement for comparative purposes with other published measurements, but it's useless for the following reasons:
1) Oftentimes we don't know at what load and what voltage/power with other measurements.
2) What Donald said - it's more for academic reasons. There was probably a better reason for this test in 1937 or 1927.
---
600/1700 instead of 500/1800 is fine for the alternate 1:1 two tone test.
---
60Hz / 7kHz... prefer 2:1 ratio instead of standard SMPTE of 4:1 because 2:1 better reflects spectrum of popular music.
---
40Hz is good instead of 60Hz. What Donald said.
However, should we still bother with 20Hz?
-
I would bother with 20 Hz. Music with content there may go down south if the amp goes non-linear distortion wild into the more audible bands.
-
What sort of FFT windowing should be used for consistency? I suggest Kaiser 5 with a larger window for looking at distortion spikes
-
What sort of FFT windowing should be used for consistency? I suggest Kaiser 5 with a larger window for looking at distortion spikes
I recommend using a common window which is supported on a variety of hardware platforms.
-
What sort of FFT windowing should be used for consistency? I suggest Kaiser 5 with a larger window for looking at distortion spikes
Well, TBH, the one that looks the best. :-) The QA400 doesn't have that many options as ARTA. I used Hann because it had the least amount of spectral leakage. In terms of FFT window size, the higher the better, until your CPU barfs. Kaiser 5 was good on ARTA, but not supported on the QA400
Rect was the best for obtaining FR based on averaged PN White Noise, all the others have ripples.
-
Well it matters if we're talking about getting comparable measurements, especially for LF distortion tests.
-
Examples:
(http://i.imgur.com/PGmg1Mp.png)
(http://i.imgur.com/jhAdMPN.png)
(http://i.imgur.com/4YiE7QR.png)
(http://i.imgur.com/22BGzhj.png)
-
You are making this much more complex than it needs to be. Simply use the skinniest longest [redickted] shape. I don't know why ARTA Hanning sucks.
See here for examples from QA400: http://www.changstar.com/index.php/topic,2617.0.html (http://www.changstar.com/index.php/topic,2617.0.html)
-
You are making this much more complex than it needs to be. Simply use the skinniest longest [redickted] shape. I don't know why ARTA Hanning sucks.
See here for examples from QA400: http://www.changstar.com/index.php/topic,2617.0.html (http://www.changstar.com/index.php/topic,2617.0.html)
Ok, so let's say people should stick to Kaiser5 with 131072 FFT if they're going to use ARTA for measurements. Just want to clear things up so people don't post random shit and ask what's wrong.
-
Exp. averaging? The window size seems rather large... might barf on older laptops and PCs.
-
We're talking about FR now? Need different settings. I propose periodic white noise (PN white), 32768 FFT and Uniform window, linear averaging
Increase FFT size to capture more subsonics, increase sample rate to get more ultrasonics, with obvious trade off between the two
-
We're talking about FR now? Need different settings. I propose periodic white noise (PN white), 32768 FFT and Uniform window, linear averaging
Increase FFT size to capture more subsonics, increase sample rate to get more ultrasonics, with obvious trade off between the two
Essentially the standard will be "don't be a retard on FFT Windows". We are not writing ISO or ANSI. It's going to be obvious on the plots when someone is retard.
-
Ok, fine by me.
I think we should expand the tests to include wide-band FR. If there are any strange subsonic (weird DC servo activity, too small coupling cap) or ultrasonic (oscillation caused by limited open-loop gain, peaking caused by signal transformer) then that should be revealed. I don't think I am alone in thinking that out-of-band activity can have an effect on sound quality.
-
Agreed. Wide-band FR as far out as possible (limited by sampling rate of ADC and signal generators)
at spec'd power and loads.
Going to try Donald's recommended .550Vrms and see if its a good start.
-
Thoughts on including square wave tests?
-
Square waves are good for a starting point. But not that good. I would go then for impulse instead of squarewaves.
The only reason to use squarewaves IMO is that you really wanted to do a step response but your oscilloscope could not latch to it or something like that. And that's because the step response is the integral of the impulse response.
One could capture a time domain sinusoid for visualization purposes however. It could also help discriminate cross-over distortion on class A/B like Atomic Bomb did. Or perhaps where an amp stops being class A. I think a 1 kHz tone would suffice if this is of interest.
-
Square waves are good for a starting point. But not that good. I would go then for impulse instead of squarewaves.
The only reason to use squarewaves IMO is that you really wanted to do a step response but your oscilloscope could not latch to it or something like that. And that's because the step response is the integral of the impulse response.
One could capture a time domain sinusoid for visualization purposes however. It could also help discriminate cross-over distortion on class A/B like Atomic Bomb did. Or perhaps where an amp stops being class A. I think a 1 kHz tone would suffice if this is of interest.
I do like the idea of zooming in on x-over distortion. Although I'm guessing it would be very difficult to resolve in most cases
-
If it's difficult to resolve best guess is that it proly is not of concern. But some of the plots that Atomic Bob showed made it seem like it was not too difficult to see. It may depend on the interface. If it's obvious I would post it. If it's not, perhaps not worth it. Time will tell.
-
I'm still fond of the ARTA/STEPs type distortion plot, especially for getting a fix on low-end tube amp distortion. (As long as the D6+ element is included).
(http://i.imgur.com/cW6kyOF.gif)
Looking at a few single-frequency ffts seems limiting, and that why I like the above style plot: it's a whole picture. The limitation of the above style, compared to fft, is that you can't get a good look at any higher order distortion products.
I've been toying with the idea of combining the above style distortion plot with the fft presentation into a video that decomposes a long sweep (or a series of 'steps') with a tracking window. With a good video player that can 'scrub' to individual frames, it's possible to take a close look anywhere one chooses, bypassing the limitation of just a couple fft plots.
But that's just a future possibility. The above STEPS type distortion graph seems doable, and I think should be considered for a standard, at least for tube/simple-discrete-SS amps. That is, for amps where the basic HD may be audible, I'd like to know what it is and what its character is.
-
I'm still fond of the ARTA/STEPs type distortion plot, especially for getting a fix on low-end tube amp distortion. (As long as the D6+ element is included).
Good idea. A concern with ARTA/STEPS is that the DAC will have to have sufficiently good measurements. In the case of a tube amp, shouldn't be a problem since the tube amp distortion will likely be a magnitude or more higher. The key would be visualization and presentation. I already hate presenting the Magni2, O2, Vali, and Studio measurements on a scale where -150dbFs is the bottom. People get weird about 0.0001 being better than 0.01 distortion when it really doesn't matter at either of those numbers.
The QA400 software can actually script this process with MS development tools for its tone generator (which has cleaner output than practically all DACs out there.)
-
Agree. One could do that and call it THD+Noise. Could do dB or %. I think one can convert between % and dB as long as one has the fundamental (frequency response) as reference for dB. We could start with 0.55Vrms 33/300 ohms as we seem to be converging with the THD and other measurements.
-
The key would be visualization and presentation.
Something I always though would be cool is a 3D distortion graph with frequency, distortion, and output power.
Would probably be a bitch to set up though...
-
Also, with sample rate going through the roof these days it might be interesting to do something like a 39+40KHz IMD test and see if any garbage shows up in the audible range.
-
Also, with sample rate going through the roof these days it might be interesting to do something like a 39+40KHz IMD test and see if any garbage shows up in the audible range.
That reminds me of some stuff that I think the Vorbis from Xiph.org dude did. Still, I think must stuff doesn't have much signal above 20 kHz to start with, even if recorded at overkill sampling rates. But I could be wrong...
-
That reminds me of some stuff that I think the Vorbis from Xiph.org dude did. Still, I think must stuff doesn't have much signal above 20 kHz to start with, even if recorded at overkill sampling rates. But I could be wrong...
There shouldn't be much, but since nobody can hear it anyway who knows what kind of weird noise or interference is actually hiding up there.
Considering there are SACDs that are just upsampled redbook I would be surprised if there were plenty of recordings with lots of weird ultrasonic noise that nobody bothered to do anything about.
-
There shouldn't be much, but since nobody can hear it anyway who knows what kind of weird noise or interference is actually hiding up there.
Considering there are SACDs that are just upsampled redbook I would be surprised if there were plenty of recordings with lots of weird ultrasonic noise that nobody bothered to do anything about.
But can't that ultrasonic garble lead to IMD in the audible band?
-
But can't that ultrasonic garble lead to IMD in the audible band?
Well yes. That's what I think the Vorbis dude showed. We can try it. We would have to agree on what represents worst case and typical ultrasonic levels and frequencies.
Some, if not most, amps do not filter input ultrasonics coming from DACs. Proly not even the mighty Hoe-2.
-
Well yes. That's what I think the Vorbis dude showed. We can try it. We would have to agree on what represents worst case and typical ultrasonic levels and frequencies.
Some, if not most, amps do not filter input ultrasonics coming from DACs. Proly not even the mighty Hoe-2.
(making crap up because I'm only wikipedia smart about this...)
Maybe something like 1000 Hz, 82000 Hz and 85500Hz
produce a 3.5kHz blippityboo, see if that interacts with the 1kHz tone to make something at 2.5 kHz.
or whatever values work to create spikes that won't overlap with the natural harmonics of the test tones, or even the first order of imd products