Serious Question: How can DAC's have a SOUND SIGNATURE if they measure as transparent? Are that many confused?

[abm0]

It’s like seeing a UFO once and getting a clear picture of it. We want more photos and data, but let’s not ignore the available data that we do have.

I'm trying to tell you: it's not a one-off, it's been under our noses for years, usually it's a simple rolloff in the highs like Solderdude immediately correctly assumed, but it seems manufacturers have been getting more creative with it recently, trying to come up with different "flavors of neutral". )

As for replication of my UFO, for what it's worth I did multiple re-plots at different gain levels and with SE vs. BAL-out, and after a firmware reflash.

 

[solderdude]

It needs to be replicated with a different method... say white noise and a simple capture of that (at least 1 minute) and then using something like audacity to show the spectrum.

As stated... I have never seen any DAC do anything like this so the most likely thing is measurement error or a defective device (PEQ stuck in some mode or so)
Passband ripples...sure... roll-off due to poor filtering ... sure.
I don't believe FiiO would intentionally 'tune' a DAC this way to be different and then fake AP measurements to show it is 'flat'.
There is a reason you measure what you measure and is interesting to find out why.
One thing I am sure of... it is not intentional to create a 'sound signature'.

 

[abm0]

It needs to be replicated with a different method... say white noise and a simple capture of that (at least 1 minute) and then using something like audacity to show the spectrum.

OK, I may try that tomorrow, I haven't sold the thing yet. I tried to go higher in number of tones but pkane's tool crashed on me at 400, so I did all of the subsequent ones at 300.

As stated... I have never seen any DAC do anything like this so the most likely thing is measurement error or a defective device (PEQ stuck in some mode or so)

PEQ was off always. Keep in mind this is a DAC-amp, not a pure DAC. I don't imagine an in-house designed amp would be as hard to shape the response of as a 3rd-party DAC chip, which doesn't even show any audio-band rolloff for 99% of its selectable filters. The curious thing to me there would be how can you put a 0.4 dB dip in the mid-treble and still have your amp be THX-AAA certified, maybe the THX design criteria just aren't very strict about this kind of thing?

 

[olieb]

I'm trying to tell you: it's not a one-off, it's been under our noses for years, usually it's a simple rolloff in the highs like Solderdude immediately correctly assumed, but it seems manufacturers have been getting more creative with it recently, trying to come up with different "flavors of neutral". )

That, to me, seems like a tough claim based on your measurement of a single specimen of one DAC model. Some more evidence is needed here.

Competition and continuous sales wouldn't work nearly as well if it was all placebo.

My guess is that placebo is a MUCH stronger effect than a tiny dip in FR, at the threshold of audibility with critical listening, direct switching and the right material. If that dip were actually there with other DACs, which would hardly be considered an advantage for most people.

 

[GXAlan]

I'm trying to tell you: it's not a one-off, it's been under our noses for years,

Oh, I fully believe your work. See my TA-ZH1ES review posted here. I was using that example for solderdude’s implication that your data is meaningless without “more.” If I misinterpreted solderdude’s intent, that’s a different story.

 

[solderdude]

OK, I may try that tomorrow, I haven't sold the thing yet. I tried to go higher in number of tones but pkane's tool crashed on me at 400, so I did all of the subsequent ones at 300.


PEQ was off always. Keep in mind this is a DAC-amp, not a pure DAC. I don't imagine an in-house designed amp would be as hard to shape the response of as a 3rd-party DAC chip, which doesn't even show any audio-band rolloff for 99% of its selectable filters. The curious thing to me there would be how can you put a 0.4 dB dip in the mid-treble and still have your amp be THX-AAA certified, maybe the THX design criteria just aren't very strict about this kind of thing?

The most likely reasons are measurement error, a load that is doing this, a defective device.
The least likely reason is a 'deliberate' change in that part of the frequency range. It does not make sense nor would that be easy to do.

Perhaps look at the actual output (using a scope) and sweep or the white noise method ?

An example of a DAC with a really obvious tonal character:

Measurements of a $5 DAC

Hi, this is a review of digital to analog converter which I bought on a local (Polish) e-commerce platform. Since it is a no-name company, I called it $5 DAC which also indicates its price. You can easily find similar DACs on Amazon or Aliexpress but the price can vary from $5 to $20. The...

www.audiosciencereview.com

I reckon something went wrong with assembly of that device and they were thrown out (to cheap to be reworked) or they did not catch the error in QC or they did and decided to sell them anyway.

 

[abm0]

My guess is that placebo is a MUCH stronger effect than a tiny dip in FR, at the threshold of audibility with critical listening, direct switching and the right material.

What threshold of audibility? It's FOUR TIMES the threshold, and manifests over a narrower band than 0.1 dB/octave. I heard it with slow switching requiring cable and adaptor unplugging and replugging and re-checking in the software that the DAC was being detected and under exclusive control every time, and I still heard the difference. This was all before I had even installed pkane's multitone tool. ) Like I said at #2: people don't know/accept what the thresholds of audibility of spectral tilt really are.

If that dip were actually there with other DACs, which would hardly be considered an advantage for most people.

LOL, have you read any (small-ish, affordable-ish) DAC-amp reviews in the last 5 years? 80-90% of reviewers praise DACs for how they manage to roll off the highs just right (OK, they don't describe it that way) in order to sound "more analog / natural / musical" as opposed to perfectly neutral DACs that to them sound "clinical / sterile / harsh / fatiguing / too digital". I think it's even more pronounced since the spread of CS43131 and 43198 chips, which include early-rolloff filter options that cut into the audio band. Messed up FR is considered an advantage these days, absolutely, this is part of why I wanted to make more noise about this finding.

 

[Newman]

Other one: Questyle M15, highly praised, glowing reviews across the board. Does it have a signature? Yes. Very close to inaudible, but not close enough, depending on listener health.

https://cdn.l7audiolab.com/wp-content/uploads/2022/07/Relative-Level-1.jpg

Even if it were to be detectable in a DBT, which I would not assume, the Sighted Listening Effect will swamp such marginal tweaks, and it is safe, until completely proven otherwise, to assume that the opinions that are being reached when doing sighted listening of these devices are dominated by SLE.

 

[abm0]

OK, I fiddled with this some more and figured out that my limitation in going up in number of tones for the multitone test was the FFT size. The highest I could use repeatedly without the app crashing was 400 with the max FFT of 4 million (vs. FFT=1M in all previous tests). What shows up with FFT 4M and with all multitone numbers from 200 up is a stable picture of some 0.7-0.8 dB dips at 2k, 6k, 10k, 13k and 17k.
MT200:

, MT300:

, MT400:

Lower numbers of multitones start to not reflect this as well going down from 100, and lower FFT sizes also start to smooth this out and make them look like fewer and fewer dips as I go down. I got it to work once with 500 multitones but it threw an error during the run and then didn't compute the FR once it was done, but it did give me the base spectral plot, and if you look closely at the spike tips the same dips are there in the same spots:

They don't quite look sinusoidal, but I would still bet on this being based on the DAC's in-band ripples (one of the options suggested by @solderdude), just made a lot worse by the amp section, because the DAC's datasheet says these ripples should all be sub-0.01 dB no matter what filter you choose.

This doesn't look like an artifact created by the software itself or by pkane's choice of how to space his multitones apart, because the biggest differences are seen when I test different DAC/amps:

Cheapo dongle JCally JM6 Pro (CX31993):

Half decibel wobbles, heard them as treble grain/mush in the original listening test.

Also cheap AkLiam PD4 (CS43131, discontinued, maybe their PD4 Plus, PD5, PD6 etc. perform better):

Interestingly, a similar response shows up for the 2015 iFi Micro iDSD (first edition with the silver housing, Burr-Brown chip):

A bit fewer wobbles and somewhat lower in amplitude, should sound better but still has that "inflatable mattress response".

I'd say that was simply the state of the art at the time, but there did seem to be other devices around in 2015 that went for more of a KA17 approach with far fewer dips but more audible in amplitude, e.g. for the Oppo HA-2 (ES9018Q2M, line-out, MT400, FFT=4M):

And of course the star of the show so far, the HiBy FC3 (ES9281pro) with its pristine response, probably because it's such a simple device, almost just a "DAC chip with cable connectors", which I'm betting is why it can perform so close to the chip's datasheet specs:

(Thanks again to @Serge Smirnoff for pointing out this beautiful performer via his Df-metric results page.)

And since I actually started yesterday's tests with Solderdude's suggestion to put white noise through the KA17, I can now show it because we know better what to look for:

Audacity is dumb and doesn't let us zoom as much as we'd like on the Plot Spectrum graph, I had to drag the window partially off-screen to enlarge it a bit, and I got 1-dB gridlines. Even at this resolution you can still see the dips are there, maybe not the 2k one but there's a 5-6k, a 10k, a 13k and a 17-ish k. Much subdued, probably within 0.2 dB of amplitude delta, but still there, and could be still audible.

I think these ripple effects are not revealed as well by white noise as the randomness of the signal probably doesn't allow the oscillatory phenomena that must be the root cause of such behavior to ramp up over time as they would with a stable input. But even that doesn't completely hide the problem. With music I think the effect comes out far closer to what we see with multitone.

The next thing I want to try is to measure with all the different KA17 filter options, see if that's really where this whole thing starts. Everything up to now has been with the linear-phase fast-rolloff aka. "FAST" in the menu. (What this also makes me wonder is what filter is pre-selected in the HiBy FC3 - maybe that superb flatness in the audio band comes at the cost of subpar attenuation in the stopband, and of higher aliasing problems, though I'd bet that's far less audible than 0.2-0.8 dB passband ripples and might have been a smart trade-off.)

 

[Blumlein 88]

Why don't you use frequency sweeps in Multitone?

Another option is to do single tones around where you see the dips and where you don't. Just a few would either confirm or not the results with the multiple tones.

I'm still not convinced the output is what you are measuring. Seems likely there is some issue somewhere.

With Audacity, due to its poor display of the data, you are better off Exporting the FFT into a spreadsheet where you can look at the value of each FFT bin. You also could use it by running single tones around the dips to see if they show up to the same degree to crosscheck what you are doing.

 

[solderdude]

Firstly I highly appreciate the effort you put in it.

However, using multitones to determine frequency response is not they way to do this.
Sweep or white noise is the way to go.
As you can see your initial conclusion was based on an incorrect measurement (method) and it also shows how confirmation bias works (blaming a measured response for something you perceived)
The 'wiggles' in the noise is caused by randomness of the signal. If you were to record 15 minutes it would show smoother. You can also change the dB scale in Audicity b.t.w.

Perhaps use REW or RAA as frequency response measurements... not mutitone in Paul's excellent program.

What your plots are showing is NOT the actual frequency response.

 

[abm0]

As you can see your initial conclusion was based on an incorrect measurement (method)

No, I don't see that. The biggest differences appear when I apply the method to different devices. That looks to me like hundred-tone multitone reveals things other methods don't, and that would be a worthwhile addition to the classic/usual battery of tests. Since the problem seems to be based on unwanted oscillations, white noise doesn't look like a better method to test for this, more like a method that hides the problem better.

 

[KSTR]

Looks a bit like DAC pass-band ripple (actually, the combined passband ripple of the DAC Under Test and the ADC used to capture the results.
Use a regular Farina-style swept sine measurement to confirm, or periodic white noise (the latter only works when DAC and ADC run on the same clock).

As for the nature of test signals, many people are not aware that there is a lot of overlap or even identities.
A multitone with all FFT bins populated with random phases is periodic white noise, and a multitone with all bins populated with so-called Newman phases creates a periodic sine sweep, and a multitone with all bins populated with zero cosine phase gives an impulse, etc etc. When not all bins are populated, you get holes in the spectrum but the time-domain signal still is quite similar.

As for random-phase multitones (as designed to give lowest crest-factor but still be random enough -- all frequencies present at all times), I prefer sparsely popoluated ones, like below 10 or so, and occupying specific FFT bins (all multiples of the same base number) so that the spectrum's noise floor is not cluttered all the way with IMD products. That really allows some insights from one single measurement that other test signals don't offer.

 

[antcollinet]

Like I said at #2: people don't know/accept what the thresholds of audibility of spectral tilt really are.

Bear in mind thresholds of audibility are exactly that - a limit beyond which no-one (even under ideal conditions) can hear them.

They are not levels which can be heard by everyone under all conditions. If you are measuring something close to the limit, then it is pretty unlikely that any one individual will hear that in real world listening conditions.

 

[abm0]

Why don't you use frequency sweeps in Multitone?

I didn't see anything like a sweep in the test signals drop-down, I think I may need to switch some other setting, I saw some measurement mode(?) somewhere that was preset to "Spectrum" and could be set to "Sweeps". I'll look into it next time. I'm usually on my Linux and I need to boot into Win10 to do all this.

Bear in mind thresholds of audibility are exactly that - a limit beyond which no-one (even under ideal conditions) can hear them.

They are not levels which can be heard by everyone under all conditions. If you are measuring something close to the limit, then it is pretty unlikely that any one individual will hear that in real world listening conditions.

I mean sure, but I've been looking to upgrade my portable DAC with something I can keep for years without worry or FOMO, so for that purpose I'd really like to see FRs with sub 0.1 dB ripples if possible within my budget. Other than that, my direct complaints about specific devices have been only at (what look so far to be) 0.4-0.8 dB ripples, after having heard weak treble in A/B comparisons. I'm not going around declaring devices with 0.1-0.2 dB ripples or rolloff to be a scam or not worth buying.

 

[ObiWanKenobi]

I know there are people out there that think cables affect sound, which is much worse, but there really is no response to something like that, but just to smile and nod. But what about people who talk about DACS as if they were headphone drivers or speakers, and talk about the SOUNDSTAGE, IMAGING, and MIDRANGE of a DAC? I actually don't know what to say to people to not be rude. If you try explaining that a DAC isn't something that actually changes the sound, they accuse you of having "a hard-on for measurements", as if it were the measurements themselves that tell you that DACs don't have a sound. What they don't get though, is that even if we had no equipment to measure distortion or other aspects of sound, still would not have a sound to them. So you try explaining by telling them that when you listen to different DACs using the same headphone and amp, that you cannot tell the difference. "You can't tell the difference between DACS????" "There must be something wrong with your system. You don't have revealing enough upstream and downstream equipment. Either that you haven't "learned" to tell the difference between them." Then you explain that in double-blind studies people are not able to tell the difference between Dacs any better than someone picking random answers. And their response is that the differences are "subtle", and them and other audiophiles who have spent time practicing and learning how to listen properly can hear a difference. "That doesn't sound like a very good way of testing that. Just taking a random group of people who know nothing about audio equipment and asking them to try to find the difference between DACs? Those people haven't yet learned to know the difference!" Then you ask them how they know that they actually hear the difference and it isn't just placebo. ETC.

The problem is that this isn't even an uncommon view. I would say that people who understand there isn't a difference between decently engineered dac (except perhaps small amounts of distortion in the lower end ones that may or may not be audible). Most audiophiles think there is at least a subtle difference between DACS and don't realize that saying the DACS sound different is like saying the portion of a DVD player that takes the 0s and 1s that are read off the disk and converts them into video can make the same DVD "look different" on the same exact TV. It's incredible, but if you want to be friends with audiophiles or even post on an audiophile board, you either have to pretend you agree or somehow remain silent when people talk about this stuff. Like "ohh have you heard the utopias in the chord hugo?? it really makes the mids stand out, but its a warmer dac". The main problem is actually that there is a confusion. They think that we mean that what makes a DAC "objectively good" is a TRANSPARENT DAC, and that we first define a good dac as a transparent DAC and then say that the measurements prove that the DAC is transparent, and therefore it is the better DAC. They think there are other dacs that are not transparent, but rather, color the sound in a good way, and therefore "measure worse" but sound better. This is nothing but a huge confusion. If that were how dacs worked, then I would actually agree with them. What matters most is how something sounds. However that is literally not what DACS do. DACs by nature do not have a sound signature. Saying a DAC has a sound signature is like saying a cable has a sound signature (well I guess if it is a really ****** dac it can have a sound signature of "fuzzy" or whatever dac distortion is, but you get the picture). Problem is, I don't think there will ever be an easy way to educate audiophiles about this, and so the only remedy will be like who the hell knows?

You’re spot on.
Amplifiers, Loudspeakers, TVs and everything in between sound and output the same picture quality.
After all, they are all fed from the same source, don’t they?
50 or 60Hz is the same, right?
I would have quote Einstein but you won’t get it.

 

[Svend P]

I would have quote Einstein but you won’t get it.

No, you don't get it. You are replying to a 5 year old post, and consensus after more than 500 pages of discussion is that your 6400 euro HoloAudio May DAC KTE DAC does not sound any better than a 100 euro Topping DAC.

 

[KSTR]

I'm usually on my Linux and I need to boot into Win10 to do all this.

Use REW (runs natively on Linux) rather than MT ;-)

 

[solderdude]

I didn't see anything like a sweep in the test signals drop-down, I think I may need to switch some other setting, I saw some measurement mode(?) somewhere that was preset to "Spectrum" and could be set to "Sweeps". I'll look into it next time. I'm usually on my Linux and I need to boot into Win10 to do all this.


I mean sure, but I've been looking to upgrade my portable DAC with something I can keep for years without worry or FOMO, so for that purpose I'd really like to see FRs with sub 0.1 dB ripples if possible within my budget. Other than that, my direct complaints about specific devices have been only at (what look so far to be) 0.4-0.8 dB ripples, after having heard weak treble in A/B comparisons. I'm not going around declaring devices with 0.1-0.2 dB ripples or rolloff to be a scam or not worth buying.

Tell you what .... the fun part of Paul's software is that you can null an original music file to the reproduced file and even listen to the null.
When you want to prove it sounds different (do the same with other DACs you have) using music this is the best method to prove that.
Tried any of the other 7 available filters in the KA17 which should have different passband ripples ?

Also use REW (also runs fine on Linux). There you can easily see sub 0.1dB variations. Now you are faffing about in ways the software is not meant to be used.

 

[ObiWanKenobi]

No, you don't get it. You are replying to a 5 year old post, and consensus after more than 500 pages of discussion is that your 6400 euro HoloAudio May DAC KTE DAC does not sound any better than a 100 euro Topping DAC.

Funny statement as you’re reading my reply to this 5 years old statement