• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Differences in DAC's...

SIY

Grand Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
10,383
Likes
24,749
Location
Alfred, NY
So I don't really understand why my assertion that such filters are used to tame high frequencies is being contested.

Not so much to "tame high frequencies" as "a tone control that can't be twiddled." So not high fidelity in the sense of trying to accurately recreate the original (band-limited) signal with nothing added or subtracted. The actual needed reconstruction filter would remove essentially all content above fs/2 while not affecting the amplitude below fs/2.
 

andreasmaaan

Master Contributor
Forum Donor
Joined
Jun 19, 2018
Messages
6,652
Likes
9,399
So such filters remove spurious noise above the Nyquist 22kHz limit and may depending on the type of filter used attenuate other frequencies within the audible band. In the case above the filters are active in the range of 5K and above which while not ultrasonic are high frequencies. So I don't really understand why my assertion that such filters are used to tame high frequencies is being contested. As an explantaion for the technically disinterested it would seem that is exactly what they do. Of course other filters can be applied in other ranges of the audio band.

There might be someone out there using a NOS-type filter as a tone control, but that is not the doctrine behind the existence of such filters.
 

bravomail

Addicted to Fun and Learning
Joined
Oct 19, 2018
Messages
817
Likes
461
... Darkvoice tube amp. I love the sound but at what point do you stop buying stuff hoping to get that last ounce of perfection?

You can try with non-tube amp like JDS Atom or Schiit Magni3. You will notice much bigger difference than from swapping the DAC.
I don't think you can talk about "sound signature" of DAC (except for those ugly filter, solderdude brings).
But you can talk about resolving power, muddy vs detailed.
 

Shadrach

Addicted to Fun and Learning
Joined
Feb 24, 2019
Messages
662
Likes
947
Not so much to "tame high frequencies" as "a tone control that can't be twiddled." So not high fidelity in the sense of trying to accurately recreate the original (band-limited) signal with nothing added or subtracted. The actual needed reconstruction filter would remove essentially all content above fs/2 while not affecting the amplitude below fs/2.
Oh good. I thought I had misunderstood something but it seems it's more of a case of semantics.:)
 
  • Like
Reactions: SIY

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
A lack of reconstruction filter is NOT used to 'tame' high frequencies.
That is what the '5kHz' filter you mention has as an effect... again it is NOT a filter it is caused by a lack of reconstruction filter.
Therefore it cannot be used as a high filter.

Of course there do exist low pass filters that start to roll-off at 5kHz with a similar slope that can be used as a treble filter.
The one in the plot isn't a filter.

Reconstruction filters pass all signals below 20kHz but some go as far as 22kHz

One can filter any frequency in all kinds of bandwidths, notch, bandpass, shelve, low pass and high pass in all kinds of slopes.
Reconstruction filters are only used in DAC's.
 
Last edited:

filo97s

Active Member
Joined
Feb 17, 2019
Messages
120
Likes
278
Location
Sestri Levante
A lack of reconstruction filter is NOT used to 'tame' high frequencies.
That is what the '5kHz' filter has as an effect... again it is NOT a filter it is cause by a lack of reconstruction filter.
Therefore it cannot be used as a high filter.

Of course there do exist low pass filters that start to roll-off at 5kHz with a similar slope that can be used as a treble filter.
The one in the plot isn't a filter.

Reconstruction filters pass all signals below 20kHz but some go as far as 22kHz

One can filter any frequency in all kinds of bandwidths, notch, bandpass, shelve, low pass and high pass in all kinds of slopes.
Reconstruction filters are only used in DAC's.
i have some questions about the filter in a DAC...
The reconstruction filter is an analog filter after the DAC, used to smoothen the output of the dac chip. DACs have also digital filters placed before the DAC, that limits the bandwith to the half of Nyquist rate before feeding the signal into the chip. Why would anyone do that? I mean, bandlimiting should happen even before the ADC, once you have recorded a signal that has alias, there's no filtering technique that can return the original signal. Am i wrong? I really can't understand why you should select a digital filter in a DAC, indeed to my ears there's no change in sound when you turn it on or off (my dac is an UD-503, and has the option to turn every filter off).
I must admit however that i struggle quite a lot to hear differences between my Teac and a FiiO X3 MK III used as a DAC so i really don't know how some audiophiles can spot differences between them or even between a single cable. This goes beyond my comprehension, maybe my ears are faulty.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
In CD players, where a fixed frequency (be it oversampled or not), is present the reconstruction filter is pure analog or hybrid.
When the CDP does oversampling part of the reconstruction filter is in the oversampler. The calculated 'in between values' thus are smaller in amplitude as well as in time.
The analog post filtering can be less steep in this case which is easier to build as it always can be the same frequency and does not have to change.
Only the very first CDP's had no oversampling and very steep and complicated analog filters that showed lots of ringing and ripple in the upper band.
Later on everybody used 4, 6, 8 or even more during the 'sample race' period. It made the analog post filter much cheaper and easier to construct (tolerances of capacitors was less important) so a win-win situation.

Then came 1 bit, MASH, SD etc.
When Sony made their first 1 bit DAC players they said nothing about it as they could not explain why they would part with the increased bit and sample rates and suddenly come with 1 bit. Later on they started to come out with the info.
Funnily enough nobody had noticed that there was a 1 bit DAC in there. Plots looked like any other oversampling DAC.

USB DACs need to be able to handle different bitrates (not a fixed one) so reconstruction is done in the digital domain and the spurious coming out of the DAC are 'smoothed' with a not so steep analog post filter that is fixed for the highest bitrate the DAC can handle.

The difference between most of the used filtertypes is inaudible.
In case of the filterless NOS things differ as headphone or speaker drivers that are limited to around 20kHz will do the 'smoothing' or 'reconstruction' part for you.
How audible this is thus depends on: Music content, used transducers, listening experience etc.
The 'seemingly audible' roll-off can be not present and very present a few ms later.
One would have to see FR plots, squarewave response or scope plots of the output signal in the 'filter off' position to know how the signal is reproduced. I have only seen UD501 plots. I assume the idea is the same in the UD503 but they don't call the filter 'NOS' specifically.

The FiiO has a correct filter. Most of the UD503 filters also do a good job. I believe the NOS setting is an 'emulation' where the DAC chip is fed a 'sample and hold' signal rather than properly filtered.


In an ADC there is analog filtering (as well as digital in the chip or switched capacitor filters) which is to prevent aliasing.
This is not the same as reconstruction.
Filters, however, work at the same frequency and steepness but serve another purpose.

There are members here that know a LOT more about filters, ADC and DAC here than I do though.
I know nothing about the used calculations or theory, just practical knowledge.
 
Last edited:

filo97s

Active Member
Joined
Feb 17, 2019
Messages
120
Likes
278
Location
Sestri Levante
In CD players, where a fixed frequency (be it oversampled or not), is present the reconstruction filter is pure analog or hybrid.
When the CDP does oversampling part of the reconstruction filter is in the oversampler. The calculated 'in between values' thus are smaller in amplitude as well as in time.
The analog post filtering can be less steep in this case which is easier to build as it always can be the same frequency and does not have to change.
Only the very first CDP's had no oversampling and very steep and complicated analog filters that showed lots of ringing and ripple in the upper band.
Later on everybody used 4, 6, 8 or even more during the 'sample race' period. It made the analog post filter much cheaper and easier to construct (tolerances of capacitors was less important) so a win-win situation.

Then came 1 bit, MASH, SD etc.
When Sony made their first 1 bit DAC players they said nothing about it as they could not explain why they would part with the increased bit and sample rates and suddenly come with 1 bit. Later on they started to come out with the info.
Funnily enough nobody had noticed that there was a 1 bit DAC in there. Plots looked like any other oversampling DAC.

USB DACs need to be able to handle different bitrates (not a fixed one) so reconstruction is done in the digital domain and the spurious coming out of the DAC are 'smoothed' with a not so steep analog post filter that is fixed for the highest bitrate the DAC can handle.

The difference between most of the used filtertypes is inaudible.
In case of the filterless NOS things differ as headphone or speaker drivers that are limited to around 20kHz will do the 'smoothing' or 'reconstruction' part for you.
How audible this is thus depends on: Music content, used transducers, listening experience etc.
The 'seemingly audible' roll-off can be not present and very present a few ms later.
One would have to see FR plots, squarewave response or scope plots of the output signal in the 'filter off' position to know how the signal is reproduced. I have only seen UD501 plots. I assume the idea is the same in the UD503 but they don't call the filter 'NOS' specifically.

The FiiO has a correct filter. Most of the UD503 filters also do a good job. I believe the NOS setting is an 'emulation' where the DAC chip is fed a 'sample and hold' signal rather than properly filtered.


In an ADC there is analog filtering (as well as digital in the chip or switched capacitor filters) which is to prevent aliasing.
This is not the same as reconstruction.
Filters, however, work at the same frequency and steepness but serve another purpose.

There are members here that know a LOT more about filters, ADC and DAC here than I do though.
I know nothing about the used calculations or theory, just practical knowledge.
ok so if i understand correctly, digital filtering is used for obvious differences in sampling frequencies where an analog filter cannot change its cutoff frequency (well, a switched cap can... but maybe is not used because of noise and DC offset), in order to maintain the same signal bandwith during the oversampling, is it correct?
Just finished an exam on digital and analog filtering last week, but even my professor was not sure on why they use a digital filter. He's not an audiophile so maybe he's not aware of oversampling in hifi dacs, but his words were "well maybe it's used as a precaution, maybe to avoid digital noise entering in the D/A process" not sure however if he was 100% right.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
ok so if i understand correctly, digital filtering is used for obvious differences in sampling frequencies where an analog filter cannot change its cutoff frequency (well, a switched cap can...

Correct.

The various types of digital filters exist because consumers 'demand' these options and because manufacturers either come up with different filters themselves to be 'different'. Arguably a steep linear phase filter is the most correct one that fits the sampling theorem best.
One can also create steep filters with post ringing and enough attenuation but with lmore post ringing.
Even here there are different implementations with different amounts of taps.
Drawbacks of linear phase are latency (not an issue with listening to music) and the fear of the dreaded 'pre-ringing'.
This pre-ringing happens outside of the audible range. Most people believe all frequencies 'pre-ring' and this 'degrades' sound and they would rather see digital versions of the first CD Player filters (with no pre-ringing but massive post ringing.
Then there are the 'I want square-waves' reproduced correctly believers and the 'impulse' guys that base their belief on the reproduction of an illegal pulse (one that does not exist in music and is never recorded).
Usually these filters have a slow roll-off, so filter less outside of the pass-band and thus have more ultrasonic crap.
There are ones with and without (very short) pre-ringing and post ringing. Looks nice with square waves and impulse plots but alas these signals do not exist in any recordings and are for tech guys that want to evaluate filter aspects.
Some folks believe the stored bit levels should be reproduced and that this represents the actual audio info. They forget what the actual samples are and that samples are points in time but are reproduced in sample and hold (which differs) and that these filters do not comply to the theorem.

There are plenty of plots found on the web that show the plots @Calexico wants to see. Google 'imaging test'.
There you can see which type of filter produces how much 'garbage'.
Archimago often shows these plots.
They are enlightening and is what @Calexico 'demands' to see from Amir otherwise he cannot see how the DACs perform in the 20kHz to 100kHz range which he (and some others) believes is important for perceived quality.

Whether or not the present ultrasonic 'garbage' may or may not result in sound degradation (never in improved fidelity) depends on many factors.
In most cases, using analog amplifiers there is no penalty and the ultrasonic garbage is simply there but not reproduced by transducers (headphones/speakers) and not heard by humans either.
 
Last edited:

Calexico

Senior Member
Joined
May 21, 2019
Messages
358
Likes
72
Asr is here to measure and compare dac? How can i compare with one measurement found on google?
Also bad power can add noise. If switching freq of psu put ultrasonic at 30khz it's better to know it.
That's why if find ASR tests to be incomplete and also asr compare dacs only at cd samplerate.
 

Veri

Master Contributor
Joined
Feb 6, 2018
Messages
9,596
Likes
12,036
Asr is here to measure and compare dac? How can i compare with one measurement found on google?
Also bad power can add noise. If switching freq of psu put ultrasonic at 30khz it's better to know it.
That's why if find ASR tests to be incomplete and also asr compare dacs only at cd samplerate.
Getting tired of this trolling behaviour of you. You seem to be begging for an ignore :confused:

Multi-tone is always 192Khz
index.php


Amir thus notices when there are problems at higher sample rate https://www.audiosciencereview.com/...iew-and-measurements-of-smsl-vmv-d1-dac.4375/

Also, as has already been told very explicitly, Amir is not going to do every measurement 3-4 times just to appease your, honestly completely unnecessary, request to perform more tests in ultrasonic ranges.

It's not going to happen. Stop complaining, you're not donating via paypal or patreon or in any way contributing to the site in a positive way.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
@Calexico wants to see the 192 plots all the way up there and not limited to 22kHz.
For example the difference between a filterless NOS DAC and cheap FiiO X3-II player (functioning as an USB DAC) below
metrum.png


Fiio.png


And he does have a point because the plots do not show what is up there.
Whether or not this has audible consequences is another matter.
I don't think Amir is going to do much more plots just to please some readers.
I suggest, if Calexico wants to see more plots he visits RAA and scrutinizes the full measurement suite there instead of demanding Amir does the same suite (which takes an awful lot of time to do)

But to discuss the above.. the Metrum has shitloads of shit above nyquist which can certainly become audible under certain specific circumstances but in most cases will be benign. This DAC is subjectively well regarded amongst NOS R2R addicinado's (see upper treble roll-off)

The cheap SD DAC from the FiiO X3-II is often said to sound boring and dull and not highly regarded amongst audiophiles with a huge budget (it is to me regarded high enough) and as one can see is a LOT cleaner than R2R.

This totally wrecks Calexico's theory by the way but he will now start looking for another way to explain to us why his subjective preference for R2R DACs is valid after all.
I suggest he looks at the Impulse Response Envelope (ETC - Energy Time Curve) plots
 
Last edited:

Calexico

Senior Member
Joined
May 21, 2019
Messages
358
Likes
72
@solderdude
I don't want to be pleased i want readers to have the full information.
Otherwise before review it should be explained that the measurements validate only 44.1/48khz samplerate and that they don't show if the psu or the design leads to ultrasonic noise on the output.
 

suttondesign

Addicted to Fun and Learning
Forum Donor
Joined
Feb 4, 2019
Messages
732
Likes
1,310
Location
Bellingham, WA
You can spend $50 or $5000 and the chances of you being able to tell the difference in an ABX are marginal. Once you have accepted this, that is the point you stop buying stuff hoping to get that last (ounce?) of perfection.

No doubt, I am the only one who sighs in mild regret each time I buy a DAC or amp which gets shown up by another, newer one on ASR. Acceptance is hard.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
I don't want to be pleased i want readers to have the full information.
Otherwise before review it should be explained that the measurements validate only 44.1/48khz samplerate and that they don't show if the psu or the design leads to ultrasonic noise on the output.

ASR may not be what you expect/want it to be.
 

Jangle

New Member
Joined
Jul 4, 2019
Messages
1
Likes
0
So you have a 0.5dB roll off at 10K but nothing to show what happens at say 15K. No you can't extrapolate.

It seems you're confused both about interpolation/extrapolation as well as how to read a log graph. You can easily figure out the value for anything below 20kHz on this graph, which is the farthest right grey line. For example, to figure out where 15kHz lies, you can calculate (log(15000)-log(10000))/(log(20000-log(10000)) which gives you 0.585, so you know 15kHz is 58.5% of the way between the 10kHz and 20kHz line (which gives you something like -1.6dBs). Another example, 13kHz is 37.9% of the way from 10kHz to 20kHz on the log graph, so it is around -1.3dB. This isn't interpolation or extrapolation, it's just reading the graph at known points. (technically, you're interpolating the point at which you're reading the graph from, but you're not interpolating/extrapolating the value on the graph).
As for not knowing enough about DACs to read a graph... well...
 

daftcombo

Major Contributor
Forum Donor
Joined
Feb 5, 2019
Messages
3,687
Likes
4,068
Anecdote I've told before, but bears repeating. For AX's review of the RME ADI-2, someone else did the "subjective" part, then sent me the unit for measurement. I asked him before he sent it to me, "How did it sound?" as a hint of where to dig in with the measurements. He answered, "Clean, but dull. I mean, really really dull."

I received the unit and worked through the menu. I noticed that the "NOS" filter was chosen, which rolls things off similarly to the graphs above. I asked the subjective reviewer, "What filter did you use for evaluation?" and he responded, "I didn't know that was selectable, whatever it was already set to."

Mystery solved.

(The only bad news was that I couldn't afford to keep it after the review)

Do you correlate early rolloff to 'dull sound" then?
Would the guy enjoy the other filters more?
 

SIY

Grand Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
10,383
Likes
24,749
Location
Alfred, NY
Do you correlate early rolloff to 'dull sound" then?
Would the guy enjoy the other filters more?

In order:

Absolutely.

And I would guess yes for him, and in my case, definitively yes (I tried it with most of the filter options, and that was the one that sounded significantly different). The NOS filter is not for those of us who want the DAC to seamlessly recreate the encoded waveform. The other reviewer expressed a bit of annoyance in his writeup that switchable filters were even an option.
 

daftcombo

Major Contributor
Forum Donor
Joined
Feb 5, 2019
Messages
3,687
Likes
4,068
In order:

Absolutely.

And I would guess yes for him, and in my case, definitively yes (I tried it with most of the filter options, and that was the one that sounded significantly different). The NOS filter is not for those of us who want the DAC to seamlessly recreate the encoded waveform. The other reviewer expressed a bit of annoyance in his writeup that switchable filters were even an option.

I understand but -1dB at 10kHz isn't really that much of a difference.
People with non-calibrated mic will know.
 

SIY

Grand Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
10,383
Likes
24,749
Location
Alfred, NY
I understand but -1dB at 10kHz isn't really that much of a difference.

Nonetheless, I believe it was audible to my aging ears. The subjective reviewer is 20 or 30 years younger than I am.

Disclaimer: I did not ABX since my portion of the review focused on measurement, not the sound.
 
Top Bottom