• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Any correlation between measurements and perceived sound quality?

Blumlein 88

Major Contributor
Joined
Feb 23, 2016
Messages
6,743
Likes
6,844
and this is the point. If we have aliasing within the normal audio band then we can measure it. Any competent ADC should be correctly input filtered, not just for anti aliasing but also for RFI.
Well they should be, but sometimes they aren't.

I need to run it again to see if some of my ADCs have half band filters. Seems as if one of them did.
 

March Audio

Major Contributor
Manufacturer
Patreon Donor
Joined
Mar 1, 2016
Messages
3,965
Likes
3,864
Location
Perth Western Australia
Well, yes, it must be. My goodness. And DAC outputs that are delta-sigma should have something to filter out all that RF before it hits your equipment, too.

Aliasing can indeed be audible to some few people even at 20kHz. We have a couple of listeners (who are young) who can detect aliasing when a redbook is upsampled by 2 without a filter.

Of course, no filtering means that you also make your time resolution worse. I know that sounds odd, but it's a fact.
Its one reason I object to amps with excessive wide open bandwidth. Its just asking for trouble even if its more subtle than directly demodulating your local AM station :)
 

j_j

Senior Member
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
357
Likes
431
Location
My dining room.
Ah yes, half band filters. No, half band filters are not good, even if they cut down the cost. There has to be at least ONE filter that has a zero at the final, ultimate 'pi', i.e. FS/2.
 
Joined
Feb 21, 2019
Messages
37
Likes
17
Location
Singapore
What could be the difference of perceived sound quality between these devices?

Sorry for the different in frequency range, the tests were done in different time, so not exactly matching in the frequency range. But let say we just observe the 20Hz - 20kHz range. Both done in -12 dBFS of the output level, at 32 ohm load. Both measured on RME Babyface Pro Input 3-4 (TS unbalanced) with level set at -10 dBV (the input has 2 level setting, -10 dBV and +4 dBu). So measured using the same audio interface input.

When we see the 2 test results like the graphs below, how do we interpret the results into perceived sound quality?


RME Babyface Pro AN34 3.5mm 32ohm 10-40kHz Multitone.png


Tempotec Sonata HD - Multitone 20Hz-20kHz.png
 

SIY

Technical Expert
Technical Expert
Joined
Apr 6, 2018
Messages
3,009
Likes
5,365
Location
Phoenix, AZ
The second spectrum is definitely a measurement error. It's very distinctive. Windowing was my first guess (I don't use REW so don't know how reliable the window setting is), other possibilities include a signal interruption during acquisition. You should re-run that.
 

KSTR

Active Member
Joined
Sep 6, 2018
Messages
268
Likes
633
Location
Berlin, Germany
Yes, measurement error.
Multitone in REW requires sample-synchronous record-while-playback, the FFT size set equal to the generator sequence length, and rectangular (==no) window.
Looks like the first condition wasn't met, hence the leakage skirts on the bottom of the needles.
Both devices may operate at the same nominal sample rate, but the clocks are not phase-locked (drifting apart by some amount) and can't be locked, with both devices being USB-driven with no way to sync them.

Workaround: use a way larger FFT size with the Dolph-Chebychev-200 window. Note that this still will probably favor the RME because it is sample-synced loopback,.
 

scott wurcer

Active Member
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
299
Likes
528
Yes, measurement error.
Multitone in REW requires sample-synchronous record-while-playback, the FFT size set equal to the generator sequence length, and rectangular (==no) window.
Tangential question, do any of the built in multitone tests allow variation in the crest factor?
 

KSTR

Active Member
Joined
Sep 6, 2018
Messages
268
Likes
633
Location
Berlin, Germany
REW help says:
"The phases of the tones in the multitone sequences are adjusted to minimise the crest factor of the signal. For linearly spaced tones that will typically produce sequences with crest factors below 5 dB. Log spaced and NID tones may have crest factors of 12 dB or more. The crest factor is shown on the panel. Note that this is the crest factor of the signal as generated, the crest factor may increase during D to A conversion. The maximum RMS level before clipping is 3 minus the crest factor, e.g. -6 dBFS for a 9 dB crest factor (with the View option Full scale sine rms is 0 dBFS selected, 3 dB lower otherwise) but a lower setting may be required to avoid clipping on the receiving device."


So, no.
I have my own code for multitones where I can adjust magnitudes and phases of the compenents at will.
 
Joined
Feb 21, 2019
Messages
37
Likes
17
Location
Singapore
The second spectrum is definitely a measurement error. It's very distinctive. Windowing was my first guess (I don't use REW so don't know how reliable the window setting is), other possibilities include a signal interruption during acquisition. You should re-run that.
Yes, measurement error.
Multitone in REW requires sample-synchronous record-while-playback, the FFT size set equal to the generator sequence length, and rectangular (==no) window.
Looks like the first condition wasn't met, hence the leakage skirts on the bottom of the needles.
Both devices may operate at the same nominal sample rate, but the clocks are not phase-locked (drifting apart by some amount) and can't be locked, with both devices being USB-driven with no way to sync them.

Workaround: use a way larger FFT size with the Dolph-Chebychev-200 window. Note that this still will probably favor the RME because it is sample-synced loopback,.
Yes, my mistake sorry, REW set to Rectangular window on multitone, not Flat Top as I mentioned earlier. I forgot.
Next time I tried your advice with larger FFT size and Dolph-Chebychev-200 window. Thanks for the tips!

It seems that the implementation still a bit buggy and not consistent.
For the same device, once I got this:
Tempotec Sonata HD 3.5mm Out -6 dBFS to Input 3-4 0 dB gain - Multitone.png


On another measurement I got this:
Tempotec Sonata HD 3.5mm Out -15 dBFS to Input 3-4 0 dB gain - 32ohm - Multitone - Java.png


I did several other measurements and most of the time I got the second graph which is more consistent with rectangular window.

So I tested both devices, RME Babyface Pro and the cheap Tempotec USB C DAC with QuantAsylum QA401, this time I match the level, and I got this with Hann Window (all at 24bit - 192kHz with 32 ohm load):

RME Babyface Pro (192kHz):
RME Babyface Pro 3.5mm Out -7 dBFS to Input 3-4 0 dB gain - 32ohm - Multitone - QA401 192k.png


Tempotec Sonata HD:
Tempotec Sonata HD 3.5mm Out -15 dBFS to Input 3-4 0 dB gain - 32ohm - Multitone - QA401 192k.png


The Tempotec FR doesn't go beyond 20kHz despite operating at 24/192k. Seems same low pass filter for all sampling rate.
 
Last edited:

scott wurcer

Active Member
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
299
Likes
528
So, no.
I have my own code for multitones where I can adjust magnitudes and phases of the compenents at will.
We had to do that for testing 256 QUAM. There is some good literature on minimizing it, not surprising the resulting waveform starts looking strange and chirp-like.
 

KSTR

Active Member
Joined
Sep 6, 2018
Messages
268
Likes
633
Location
Berlin, Germany
So I tested both devices, RME Babyface Pro and the cheap Tempotec USB C DAC with QuantAsylum QA401, this time I match the level, and I got this with Hann Window (all at 24bit - 192kHz with 32 ohm load):
Looks good now. The Tempotec surely produced more high frequency hash and considerable amounts of mirrored/spurious frequencies. Will this be audible in a comparison? Even with specific test signals I would doubt that, let alone with normal music... and even when it is, there might be other cues that allow A/B discrimination...
 

j_j

Senior Member
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
357
Likes
431
Location
My dining room.
It's way too vague a plot (sorry) to say anything for sure, but the first of those spectra in that last bunch looks like it was run through some kind of a codec system.
 

Juhazi

Addicted to Fun and Learning
Joined
Sep 15, 2018
Messages
522
Likes
438
Location
Finland
Back to basics... The German DIN 45-500 standard was (and to me still is) considered as limit of hifi sound, it was published in 1966. Then it was really tough to find appliances that would pass these figures. The limits are based on audibility somehow, but most likely it was just an opinion of a group of engineers.

http://www.hifimuseum.de/45-500-die-englische-version.html

At the outset it must be understood that the standard is based on what DIN considers to be minimum requirements for hi-fi. Probably the nearest British document to the DIN is that evolved in 1960 by the Audio Group of BREMA {British Radio Equipment Manufacturers' Association) and finally published as British Standard 3860 in 1965, but this is concerned essentially with measuring and expressing performance of amplifiers.

Distortion factor (measured according to DIN 45-403) is not to exceed 1% from preamplifiers at full output from 40Hz to 4kHz, and not to exceed 1% from power amplifiers and integrated amplifiers from full power down to -20dB over a power bandwidth of 40Hz to 12,5kHz

Signal-to-noise ratios on preamplifiers should be better 50dB at the nominal input signal amplitude, and on power amplifiers and integrated amplifiers up to 20W rating the figure is 50dB for 100mW output. Power output specified is at least 10W mono and at least 2 x 6W stereo, and the amplifier must be capable of producing sine-wave signals at 1kHz for a period of 10 minutes.
 
Joined
Feb 21, 2019
Messages
37
Likes
17
Location
Singapore
Looks good now. The Tempotec surely produced more high frequency hash and considerable amounts of mirrored/spurious frequencies. Will this be audible in a comparison? Even with specific test signals I would doubt that, let alone with normal music... and even when it is, there might be other cues that allow A/B discrimination...
For sure I don't know how that Tempotec spurious frequencies beyond the 20kHz sounds like... but just comparing the 3.5mm headphone output of RME Babyface Pro and the $25 Tempotec Sonata HD USB C DAC (dongle DAC), 'I think' the most audible differences to my ears are the attack and transient. The RMA attack and transient are a lot faster and impactful. I compared using IEM, listening to Chesky records (24/192k), foobar with ASIO or WASAPI driver to avoid OS resampling. Yes, this $25 DAC comes with ASIO driver :D. Both sounds good, and the Tempotec is surprisingly sounds pretty good for casual listening considering the price, more lively than my HP laptop headphone output. A nice and simple upgrade for just $25. But in comparison to the RME, the RME wins especially in perceived transient, attack, and dynamic. Therefore perceived resolution is also better on the RME. The Tempotec sounds more mellow on the dynamic and attack. But the Tempotec has a nice tonal density around the midrange, while the RME may sounds a tad leaner in comparison. How do we translate midrange tonal density in measurement? Is it the even order harmonic like the tube distortion? I haven't check the Tempotec harmonic distortion yet. Anyway, the differences is a lot less than the price different between the 2.
 
Last edited:

KSTR

Active Member
Joined
Sep 6, 2018
Messages
268
Likes
633
Location
Berlin, Germany
Yeah, that's the situation I was expecting. I'm convinced to get some grip on the question one has to make time-domain meassurements (recordings with an independant device) with actual music tracks and compare (subtract) them "directly", after de-embedding the sample-rate mismatch and drift as well as simple frequency response differences (linear error). @pkane's DeltaWave software simplifies this big time.
 

j_j

Senior Member
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
357
Likes
431
Location
My dining room.
Well, the problem with all of the DIN specs given above is that they fail to consider error spectra.

It should be possible to develop better standards now, but I greatly fear that commercial interests would be involved to an unhealthy degree
 
Top Bottom