Master Thread: Are Measurements Everything or Nothing?

[Dan Lavry]

Thank you all for the nice welcome. I am beginning to understand. I completely abandoned the hi-fi market a few years ago. The world seems to have plenty of misinformation without it. The professional community is much better, but it has some snake oil as well. My biggest issue is about sample rate for digital audio.

I wasted a lot of time and energy talking, writing and fighting against 192KHz sampling. My Gold series converters operated at 44.1-96KHz (no 192KHz). It was a costly decision. Some customers insisted on accommodating 192KHz. Many agreed with my arguments, but they needed to also process 192KHz material. I had no choice, I now accommodate 192KHz. I was not happy about it. It was a business decision. It was painful enough and now I see some 384KHz and 768KHz sample rates. I guess people tend to think that “more is better”. The sampling concepts are not intuitive, It leaves a huge opening for snake oil. It is difficult to fight against such crock.

 

[voodooless]

It was painful enough and now I see some 384KHz and 768KHz sample rates

Imagine the square waves out of one of those

 

[j_j]

The early sigma delta I recall used a second order filter. They were “stable”, but the performance was limited (good enough for speech?). Higher order filters came with a new concept- “conditional stability”. The first time I saw higher order filters was the Bob Adams design (mid 80’s).

Did the earlier sigma delta you mention utilize higher than second order?

The original Candy and Condon one used 1 to 5 integrators, not filters. It was certainly primitive. They used it for low-bandwidth exact measurements.

Well, integrators are filters, yes, but the goal was not bandwidth.

 

[Dan Lavry]

You might be surprised how practical this has become. With software like DeltaWave this is really trivial, and takes care of all the things you mention. Just record whatever you like and compare.

Now the real question is how practical this really is as an analysis tool? I think it has it’s uses for sure, but the standardized set of analysis in the frequency domain are far easier to understand and give better, and more specific insights overall.

If you want to test for all possible cases (music can be "anything") I would think that a random signal, say 20-20KHz bandwidth for a few seconds would be ideal. You compare the files and see how good the conversion is. But the signal is random. Random knows nothing about harmonics. You can measure, but it is not easy to relate to the cause the errors. The FFT is not very interesting...

Testing with real music does not guaranty that you cover the whole range. I would trust the random signal to treat everything equally.

The frequency domain is better. It is connected to the physics of sound. It "understands" harmonics. Sound has a complex relationship with sine waves. Steady tone is a sum of harmonically related sine waves (Fourier series). Real music is more complex but has elements of sine waves. So generating and reading in the frequency domain makes sense to me. More important, so far, there is good frequency domain test gear, and it works well.

 

[voodooless]

So generating and reading in the frequency domain makes sense to me

It’s also much closer to how we perceive sound. Frequency domains dominates over time domain by a wide margin.

 

[simon-b]

A question for the engineer types basically hopefully without activating any Luddite triggers.

So if my two DAC's measure exactly the same I.E. basically flat as a pancake and -120db+ distortion why can I easily discern a difference. The premise is that something is not being measured? I take as an example a Topping D90SE and a Gustard X26III. The D90SE is vary harsh in the treble compared to the X26 III. What would be a way of a measured test to confirm that? Thanks for your responses. Chhers.

 

[sigbergaudio]

Nothing measures exactly the same. Even two Gustard X26III won't measure identically.

That being said, the D90SE is not very harsh in the treble, and there's a 99.999% chance they will sound exactly the same to you if you did not know which one was playing.

If there was such a difference that one could generally agree that one of the DACs were "very harsh in the treble" even in a blind test, this would be easy to see in the measurements as well.

 

[JSmith]

why can I easily discern a difference

It's also quite possible you thought you did, but actually didn't... normal human biases and brain, don't feel bad at all.

If you want to go to the trouble of proving it to yourself, you would have to level match them (properly) and see if you can indentify a reliable difference, without peeking or any tells. Without such switching, one can't remember such minor differences... as echoic memory lasts for seconds, not minutes, hours or days.

Once difference is established... is there any actual preference?


JSmith

 

[Beave]

There's also the possibility that your D90SE is defective somehow and really does sound different (and bad). That would *easily* be picked up in measurements.

But it's more likely due to the reasons mentioned in the previous two posts.

 

[simon-b]

@sigbergaudio @JSmith @Beave I would like to believe they would sound exactly the same. But listening either blind ( I forget which one is playing after a while ) or obvious listening for differences I can say without a doubt the Texas Southside album would be easy to distinguish in a blind listening test. But aside from Blind or non blind listening tests the difference is there. My original post is asking how would one go about measuring the difference?

 

[solderdude]

My original post is asking how would one go about measuring the difference?

You will need a good ADC and record the sound of both DACs at 192/24 bits and make sure peaks during the recordings stay below -3dBFS.
Use a 44.1kHz original file to playback.
There must be at least 30 seconds of music in the recordings (no need to record an entire song)

Use an excerpt where you can clearly hear the difference between the DACs.

Post the recordings and original file and let some of the guys here do the analysis.

 

[JSmith]

My original post is asking how would one go about measuring the difference?

Both have been measured here, both are completely transparent.

If you decide to try the ADC route, you can use this too;

DeltaWave Audio Null Comparator | DeltaWave documentation

deltaw.org

JSmith

 

[simon-b]

Both have been measured here, both are completely transparent.

If you decide to try the ADC route, you can use this too;

DeltaWave Audio Null Comparator | DeltaWave documentation

deltaw.org

JSmith

Interesting. Unfortunately I don't have Windows. So the question is: With the known differences in sound between the two DAC's but with the statement "Both have been measured here, both are completely transparent" one can only deduce that some type of missing measurement needs to be added to the tests to match what is heard in the listening tests? It would be superb if that could be done?

 

[j_j]

It’s also much closer to how we perceive sound. Frequency domains dominates over time domain by a wide margin.

Eeeh, depends on frequency. See the talk I gave a few weeks ago, the recording is up on the thread about it.

I use several signals. First, I use a very specific allpass signal. Interestingly (at least to some people) it can be used to measure SNR in the time domain (after handling bandwidth calculation issues), although that's a secondary efffect. What it does is measure, very very clearly, the impulse response of the entire system.
Then I use a very specific "buzz" tone. It's not for listening (FFS, don't even think about it!), but it does capture nonlinearities very, very clearly, as well as showing noise floor, and it can also be arranged to detect interchannel crosstalk OR interchannel nonlinearity. (Yes, that exists, thank you, wimpy power supplies!)

I started to post some stuff on this in a thread on matlab scripts, but people got all bent about not using something they knew, so I stopped.

All you really need to do this is a double-precision FFT and IFFT in any case, really.

 

[Chrispy]

Interesting. Unfortunately I don't have Windows. So the question is: With the known differences in sound between the two DAC's but with the statement "Both have been measured here, both are completely transparent" one can only deduce that some type of missing measurement needs to be added to the tests to match what is heard in the listening tests? It would be superb if that could be done?

So what are the known differences? Got some measurements or something?

 

[JSmith]

missing measurement needs to be added to the tests

No, the measurements are very thorough using a Audio Precision APx555 analyser, which costs around $50K and can measure to levels no human ear could ever reach.

There's no mystery with audio devices like DAC's and amps, they can be completely characterised by measurements, more than any listening tests.

As mentioned, proceed to self testing if your curious though, it can be quite enlightening.... one realises the truth. Usually such "night and day, wife can hear it from the kitchen, veils lifted" differences are non-existent and completely due to things like expectation bias, which we're all inflicted with as humans.

Members here will be happy to assist on this path if you wish... or you can trust the measurements and save yourself a lot of time.

If the difference is very obvious to anyone... then something is wrong, misconfigured or broken in your setup. But then that would not be subtle differences like you describe.


JSmith

 

[RexrothPigeon]

Interesting. Unfortunately I don't have Windows. So the question is: With the known differences in sound between the two DAC's but with the statement "Both have been measured here, both are completely transparent" one can only deduce that some type of missing measurement needs to be added to the tests to match what is heard in the listening tests? It would be superb if that could be done?

The measurements are many many times more sensitive than human hearing. There is no audible difference unless one or both are broken or they are not truly volume matched

 

[simon-b]

So what are the known differences? Got some measurements or something?

Known differences are the harshness in the D90SE compared to the X26III.

 

[voodooless]

Eeeh, depends on frequency. See the talk I gave a few weeks ago, the recording is up on the thread about it.

Yeah, I saw that, just didn't have the time yet to check it out, but definately will

 

[simon-b]

The measurements are many many times more sensitive than human hearing. There is no audible difference unless one or both are broken or they are not truly volume matched

Indeed. That pre-supposes you are actually measuring the right thing. Where you statement fails is you are assuming no audible difference. So berrying your head in the sand is not really what the OP was aimed at.