why don't we graph DACs ? ( i did !)

[amanieux]

i never see graphs for dacs, why isn't it done like it is systematically done for headphones and iems ?

here are 2 graphs of my iem (letshuoer s12) playing pink noise from 2 different dacs :
* ifi hip2 (burr-brown) - smooth warm

* moondrop dawn 4.4 {Non-oversampling filter mode} (cs43131) - energetic bright

they sound different and graphs indeed looks different (notice the +6db@50hz on hip2 vs dawn 4.4)

is there something i am doing wrong and dacs cannot/should not be graphed ?

 

[solderdude]

i never see graphs for dacs, why isn't it done like it is systematically done

It is done and the plots are shown. Just not with music but with test tones.
Well... Serge does as one of the few.
Another way, using music, is nulling (PK metric)

It seems that the bottom one is high pass filtered and a different reconstruction filter is used (or none at all, filterless NOS or 'slow' filter).
Could also be a result of not selecting the same length of data in a file.

Try to record pink noise for at least a few minutes and see if the FR differs in the same way.

FR is shown here:

and the usually problematic part is minutely shown here:

 

[amanieux]

It is done and the plots are shown. Just not with music but with test tones.
Well... Serge does as one of the few.
Another way, using music, is nulling (PK metric)

It seems that the bottom one is high pass filtered and a different reconstruction filter is used (or none at all, filterless NOS or 'slow' filter).
Could also be a result of not selecting the same length of data in a file.

Try to record pink noise for at least a few minutes and see if the FR differs in the same way.

FR is shown here:

and the usually problematic part is minutely shown here:

this was pink noise and dawn 4.4 was in Non-oversampling filter mode and hip2 has no filtering options, why do i need to sample for a few minutes ? my samples where about 3 seconds long (+-1 second), do samples must be of exactly the same lenght to show their graph ? i redid 3 takes and it looked similar even if recording length varied from 3 to 4 seconds

 

[solderdude]

Looks like a poor quality white noise was used or some 'correction' was applied afterwards.
Pink noise looks very different.
White noise should be an almost straight line on the ifi when run with a proper file and at least 2 minutes.

It is like comparing apples to oranges.
The differences above 5kHz is solely due to incorrect reconstruction filters being applied.
The Moondrop appears to have a high-pass filter (coupling cap somewhere that is not properly dimensioned)

 

[amanieux]

Looks like a poor quality white noise was used or some 'correction' was applied afterwards.
Pink noise looks very different.
White noise should be an almost straight line on the ifi when run with a proper file and at least 2 minutes.

It is like comparing apples to oranges.
The differences above 5kHz is solely due to incorrect reconstruction filters being applied.
The Moondrop appears to have a high-pass filter (coupling cap somewhere that is not properly dimensioned)

this was the pink noise source :

and here is the dawn 4.4 with slow roll off phase compensated filter graph (initial graph was NOS), it does change the graph in the high frequencies you were right (but i cannot hear any difference between filters modes unlike changing from dawn 4.4 to hip2 where sound difference is noticeable when doing a/b testing , must be the difference in low frequencies )

 

[JSmith]

this was the pink noise source :

High Quality Pink Noise | wav mp3 Audio Files Download

What is Pink Noise and how to use it.

www.audiocheck.net

Signal Generator

JSmith

 

[amanieux]

Looks like a poor quality white noise was used or some 'correction' was applied afterwards.
Pink noise looks very different.
White noise should be an almost straight line on the ifi when run with a proper file and at least 2 minutes.

It is like comparing apples to oranges.
The differences above 5kHz is solely due to incorrect reconstruction filters being applied.
The Moondrop appears to have a high-pass filter (coupling cap somewhere that is not properly dimensioned)

here is a30 seconds samples instead of 3 seconds long sample (dawn 4.4 slow rolloff), curve smoothes out but don't tend to go to a flat curve, why should it be at least 2 min long ?

 

[amanieux]

High Quality Pink Noise | wav mp3 Audio Files Download

What is Pink Noise and how to use it.

www.audiocheck.net

Signal Generator

JSmith

here is the result with your mp3 pink noise (sampled for 3 seconds) , it looks the same as the pink noise from youtube i think it is because we see the response of my iem (s12), i do not sample the voltage out of the dac but the sound coming out of my iem (i try to visualise what i hear)

 

[thulle]

curve smoothes out but don't tend to go to a flat curve, why should it be at least 2 min long ?

My intuition would be because pink noise should approach a straight line on a log/log plot ( https://en.wikipedia.org/wiki/Pink_noise#/media/File:Pink_noise_spectrum.svg ), but we need enough of it to make this happen. Compared to that these graphs look totally wack, going up 6dB between ~1600 and 3200 Hz? huh?

 

[amanieux]

My intuition would be because pink noise should approach a straight line on a log/log plot ( https://en.wikipedia.org/wiki/Pink_noise#/media/File:Pink_noise_spectrum.svg ), but we need enough of it to make this happen. Compared to that these graphs look totally wack, going up 6dB between ~1600 and 3200 Hz? huh?

i sample the sound from my iem through the dac and not the voltage out of the dac, the graph must be the combination of response of dac+iem and not dac alone i guess

 

[thulle]

i sample the sound from my iem

ah, missed that part as I skimmed this

so, the signal path is pc -> dac -> iem -> unknown microphone -> unknown adc -> pc?

 

[amanieux]

ah, missed that part as I skimmed this

so, the signal path is pc -> dac -> iem -> unknown microphone -> unknown adc -> pc?

yes exactly, the unknowns are [macbook white internal mic and internal adc] so you realise i am not a sound engineer but only an average user using available average gear trying to understand the difference he ears with his gear when changing dacs but even with this basic process and basic gear, graphs do show a clear difference that confirms the subjective impression i had of a dac sounding smooth and the other sounding a bit harsh

 

[solderdude]

i sample the sound from my iem through the dac and not the voltage out of the dac, the graph must be the combination of response of dac+iem and not dac alone i guess

Explains a lot.
When you measure a DAC, just measure the DAC electrically.
When you measure an amplifier (section) measure it without a load (electrically) and with a load (also electrically)

Measuring a DAC output from a driver is like measuring a distance with a piece of rubber band.

 

[solderdude]

here is a30 seconds samples instead of 3 seconds long sample (dawn 4.4 slow rolloff), curve smoothes out but don't tend to go to a flat curve, why should it be at least 2 min long ?

Noise is random so especially the lower frequencies it takes time for all random frequencies and amplitudes (so you only end up with peak values) to have been present.
This is why it needs to be recorded for quite some time to be sure.

When using an IEM the output resistance of the amplifier(section) also plays a sound determining role.

When you only would be replacing a DAC, use long enough test period and level match the output you could actually see differences in the DAC.
Alas, when that DAC has its own headphone out all bets are off.

 

[thulle]

yes exactly, the unknowns are [macbook white internal mic and internal adc] so you realise i am not a sound engineer but only an average user using available average gear trying to understand the difference he ears with his gear when changing dacs but even with this basic process and basic gear, graphs do show a clear difference that confirms the subjective impression i had of a dac sounding smooth and the other sounding a bit harsh

You have so many unknowns that the only thing other people can read from this is another example of confirmation bias. The graphs aren't normalized/overlaid or something that makes them easy to interpret, they're even different scale, so maybe work a bit on the presentation too. And, as solderdude shows, all the tests include frequency response already, making the thread topic question erroneous.

The outlook to measure to verify the difference you're hearing is good one though, and I encourage you to continue, but you gotta think it through and be aware of the unknowns and what you're actually measuring. As an exercise you could try thinking through the whole signal path and what might affect it, ie. where your sources of errors might be.

 

[Matias]

i never see graphs for dacs, why isn't it done like it is systematically done for headphones and iems ?

here are 2 graphs of my iem (letshuoer s12) playing pink noise from 2 different dacs :
* ifi hip2 (burr-brown) - smooth warm View attachment 322912
* moondrop dawn 4.4 {Non-oversampling filter mode} (cs43131) - energetic bright View attachment 322913
they sound very different and graphs indeed looks different,

is there something i am doing wrong and dacs cannot/should not be graphed ?

Very interesting how bass around 50Hz is different between them, probably because of different output impedances. The highs is more likely due to filters.

 

[amanieux]

Explains a lot.
When you measure a DAC, just measure the DAC electrically.
When you measure an amplifier (section) measure it without a load (electrically) and with a load (also electrically)

Measuring a DAC output from a driver is like measuring a distance with a piece of rubber band.

makes sense but a dac in a vaccum is less usefull than the combination/synergy of dac+driver you are using, what is important is what we hear at the end of the audio chain and not a chain link on it's own

 

[thulle]

what is important is what we hear at the end of the audio chain and not a chain link on it's own

The issue still remains though, too many unknowns. How is the IEM fixed to the laptop?

 

[solderdude]

makes sense but a dac in a vaccum is less usefull than the combination/synergy of dac+driver you are using, what is important is what we hear at the end of the audio chain and not a chain link on it's own

The reason no one measures a DAC/amp with an earphone using a mic in a tube as there are too many variables here 'measured' at the same time.
Of course from earphone to plot will be the same path but what you call a DAC is actually a DAC+headphone amp.
There could be a different output resistance in the amp which can change the response of the earphone but is not part of the DAC.

So the: why don't we graph DACs ? ( i did !) should say:
Why aren't combinations of DAC/amps and headphones measured on acoustical output (I attempted this)

The reason is there are too many variables. There are literally hundreds of head/earphones and thousands of possible combinations between DAC/amp/headphones possible so picking just one combination is pointless other than to the owner.
On top of that sticking an earphone in a tube and some mic on the other side is not very scientific and explains why your result is what it is and does not look like pink noise.

 

[amanieux]

The reason no one measures a DAC/amp with an earphone using a mic in a tube as there are too many variables here 'measured' at the same time.
Of course from earphone to plot will be the same path but what you call a DAC is actually a DAC+headphone amp.
There could be a different output resistance in the amp which can change the response of the earphone but is not part of the DAC.

So the: why don't we graph DACs ? ( i did !) should say:
Why aren't combinations of DAC/amps and headphones measured on acoustical output (I attempted this)

The reason is there are too many variables. There are literally hundreds of head/earphones and thousands of possible combinations between DAC/amp/headphones possible so picking just one combination is pointless other than to the owner.
On top of that sticking an earphone in a tube and some mic on the other side is not very scientific and explains why your result is what it is and does not look like pink noise.

you are right i was looking for a way to get objective data to confirm my subjective impression when switching dacs on my iem, when things sounds different we must always try to get rid of the placebo effect that is always present.