• 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!

DAC types and their sonic signature

zalive

Active Member
Joined
Sep 4, 2018
Messages
263
Likes
38
No that is not my understanding... it's what you think is my understanding of it.
It's the consequences of the waveform in the amplitude domain over time.
Have a look at the amplitude of the 10 and 16kHz signal in post #310 and look at the amplitude which varies and it should not do this.
the amplitude thus drops and rises (varies) constantly depending on synchronization of the frequency in question with the sample frequency.
That dropping of amplitude is the 'roll-off cutting in the audible band'

The 12kHz square and sinewave comparison is not what this is all about.
That has to do with related harmonics.
What if the amplitude of that 12kHz, that should be constant, varies in amplitude between nominal and way below nominal constantly and the average of it is lower than what it should be ?

Waveform shows various frequencies at the same time.
You see a fluctuation of the amplitude but this fluctuation doesn't exist at the frequency of the fundamental tone. It graphically shows when you add oranges and apples, different frequencies. But different frequencies don't add in nature, they simply coexist.

Also it's irrelevant whether if it's harmonics or some other frequencies which show in the NOS waveform. All frequencies which show in a waveform are higher than fundamental frequency. And the only thing that matters is whether any of them (aside of fundamental frequency) belong to the audible range. If not...you won't hear them, you will only hear a fundamental frequency sine tone.

This is why you can't hear a difference between a 12kHz tones. First harmonic is 24kHz, you can't hear it.
If you look carefully graphs of NOS waveforms, you'll understand that those frequencies that change the shape of the fundamental tone sine all belong to >20kHz range. If you filter them out above Nyquist, of course you'll get a (almost) perfect sine, because this garbage frequencies are all above Nyquist.
 
Last edited:

zalive

Active Member
Joined
Sep 4, 2018
Messages
263
Likes
38
Yes it does with an illegal test signal.

http://archimago.blogspot.com/2013/06/measurements-pulse-response-5khz-10khz.html

What you need to understand is that (on my request) Archimago did that test but... and here is the kicker... is only true for test signals that start and stop on the 0 line.
By definition that signal is illegal and contains a frequency content (at the start and stop) that goes far beyond Nyquist and can not exist in a recorded music signal as that would have been filtered out by the anti-alias filter in the ADC chain.
You can 'see' this when you look at the waveform and zoom in around the 0V line.
When you do, you will see the signal rises and stops instantly which is squarewave and thus a signal that is not recorded.

This is where my local tech guys who think ringing does happen in the music should kick in and explain why ringing happens with real music :D
All I can do is to try collect a little info from them.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
Waveform shows various frequencies at the same time.
You see a fluctuation of the amplitude but this fluctuation doesn't exist at the frequency of the fundamental tone. It graphically shows when you add oranges and apples, different frequencies. But they don't add in nature, they simply coexist.

giphy.gif


Also it's irrelevant whether if it's harmonics or some other frequencies which show in the NOS waveform. All frequencies which show in a waveform are higher than fundamental frequency. And the only thing that matters is whether any of them (aside of fundamental frequency) belong to the audible range. If not...you won't hear them, you will only hear a fundamental frequency sine tone.

giphy.gif


its about the AMPLITUDE varying....

If you look carefully graphs of NOS waveforms, you'll understand that those frequencies that change the shape of the fundamental tone sine all belong to >20kHz range. If you filter them out above Nyquist, of course you'll get a (almost) perfect sine, because this garbage frequencies are all above Nyquist.

Yes ... I understand that ... you do realize you just used the words
If you filter them out above Nyquist, of course you'll get a (almost) perfect sine
and you are correct ... that happens WHEN you filter them.
Here is the funny bit... that does NOT happen when you do NOT filter them.
You see ... that's what the filter is for and DOES for you ... unless you leave it out ... then it doesn't and the amplitude will vary which it doesn't when run through a filter.
 
Last edited:

zalive

Active Member
Joined
Sep 4, 2018
Messages
263
Likes
38
its about the AMPLITUDE varying....

Amplitude doesn't exist alone, it belongs to a frequency. And you see a graph which sums amplitude of various frequencies and conclude that amplitude has changed. Amplitude of which frequency has changed?

If you have a 12kHz tone with some higher frequency content which makes its waveform looks funny, do you conclude that those added frequencies changed (increased or decreased, fluctiations of course) the amplitude of a 12kHz tone?

Yes ... I understand that ... you do realize you just used the words

and you are correct ... that happens WHEN you filter them.
Here is the funny bit... that does NOT happen when you do NOT filter them.

Nothing wrong with the words, I'm fully aware what I have written. But it's just waveforms looks if you can't hear those ultrasonic additions.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
bla bla ... nonsense ... blla.

tenor.gif


Amplitude doesn't exist alone, it belongs to a frequency. And you see a graph which sums amplitude of various frequencies and conclude that amplitude has changed. Amplitude of which frequency has changed?

What you see is NOT a summation of frequencies. What you see is a sample-and-hold representation of measurement points in time of a pure sinewave frequency that does NOT have 'various other frequencies' in it.
It is an incorrectly reproduced signal that SHOULD have represented a nice sinewave HAD it been run through a reconstruction filter.
Hence the word ... RECONSTRUCTION in that word.
Leave out the reconstruction bit (the filter) and that wave form is what you get.
The ears hear a sinewave (because your speakers and ears turn it into one) but at the incorrect level (the roll-off) because it fluctuates in amplitude and NOT in frequency.

Again .... the stairsteps are NOT harmonics, they are NOT related to the signal... They are quantization errors because of a lack of reconstruction filter.
 
Last edited:

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,524
Likes
37,057
Filters with post ringing sound different because almost always they slightly roll off the response. Unfiltered NOS more than slightly rolls off the response. You could use one of the horrid ringing DACs and roll off the response. It is the response roll off you are hearing.

The reason filterless NOS works at all is your speakers and ears become the filters. I've said up-thread, a DAC without a filter is a broken by design DAC. What is the benefit in that? Some ringing that only happens with illegal signals is different?
 

zalive

Active Member
Joined
Sep 4, 2018
Messages
263
Likes
38
What you see is NOT a summation of frequencies. What you see is a sample-and-hold representation of measurement points in time of a pure sinewave frequency that does NOT have various other frequencies in it.
It is an incorrectly reproduced signal that SHOULD have represented a nice sinewave HAD it been run through a reconstruction filter.
Hence the word ... RECONSTRUCTION in that word.
Leave out the reconstruction bit (the filter) and that wave form is what you get.
The ears hear a sinewave (because your speakers and ears turn it into one) but at the incorrect level (the roll-off) because it fluctuates in amplitude and NOT in frequency.

But those are frequencies, above the Nyquist. 'Reconstruction filter' is a high pass filter designed to attenuate frequencies higher than Nyquist - you said it yourself. Those are likely burst tones all over the place but they're nothing but frequencies and the sine wave of the frequency which was sampled is there. A single frequency can only be a sine wave, if you see anything else it's a complex tone which consists of various frequency. And there's no fluctuation in its amplitude because this additions never touch the sampled frequency, so they cannot change its amplitude. As I said, you can't sum oranges and apples.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
CrazySimpleCattle.webp


I did not say that at all.. if you misinterpret the info we and your local designers give you no wonder you are still confused.

The reconstruction filter is not just a high pass filter to attenuate frequencies above Nyquist.
It is a RECONSTRUCTION filter.
It correctly fills-in (hence the word reconstruction) the spaces, in this case stairsteps, between the dots (bars in this case) and it does so BECAUSE it 'rings'.
It ALSO filters the US crap above Nyquist that you really don't want because in some circumstances that might bite you in the a...

A single frequency can only be a sine wave, if you see anything else it's a complex tone which consists of various frequency. And there's no fluctuation in its amplitude because this additions never touch the sampled frequency, so they cannot change its amplitude. As I said, you can't sum oranges and apples.

You are correct when you MIX tones, then you don't see a pure tone.
BUT as I explained a thousand times already in this case there are no 'mixed frequencies'.
And no.... it also isn't a single frequency 'mixed' with an inaudible 44.1kHz square wave.
That would LOOK different on the scope screen.

Here you are... I made a scope plot of a pure 16kHz sinewave MIXED with a 44.1kHz squarewave, of which you insist the sample-and-hold picture consists of. As you can see it does not look like the sample-and-hold picture nor does it sound like one either.
scope.png


Again you are looking at a sample-and-hold signal of a pure sine wave that NEEDS to pass through a RECONSTRUCTION filter.
That filter reconstructs the original sine wave AND filters the quantization errors out as well.
Leave the filter out and it is not reconstructed.
 
Last edited:

zalive

Active Member
Joined
Sep 4, 2018
Messages
263
Likes
38
That's a pretty genius argument :) hats off @Blumlein 88

That's actually an argument which validates filterless NOS :D it says you can safely use it as you won't hear it anyway even if you don't filter it out, because it gets filtered at another place anyway.
 

Veri

Master Contributor
Joined
Feb 6, 2018
Messages
9,596
Likes
12,036
That's actually an argument which validates filterless NOS :D it says you can safely use it as you won't hear it anyway even if you don't filter it out, because it gets filtered at another place anyway.
I disagree with that logic, "thank God our ears are ok with a flawed unfiltered signal" does not validate NOS to me :) only that we have very tolerant ears...
 
Last edited:

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
What @Blumlein 88 meant with this is that the ultrasonic crap is filtered out by the speakers and ears.
He did not say that the output signal gets properly reconstructed by the ears and speakers. ;)
He also mentions that a filterless NOS DAC is broken by design... that is hardly 'confirming' reconstruction filters are not needed.
That is only you (@zalive ) drawing the wrong conclusions ... yet again. :rolleyes:
Saying something 'works at all' is not the same as 'something works properly'.

The ears and speakers do NOT constitute a proper reconstruction filter. They merely remove ultrasonic sounds. How much a speaker removes differs. Assuming they don't have too much IMD....
How much the ears remove is evident.
Because of this aspect we do not hear the 44.1kHz sampling frequency and the images.

You still are confusing reconstruction filters with gentle low pass filters and do not seem to understand that the sudden drop of HF response in ears is not the same as a steep reconstruction filter.

There is NO JUSTIFICATION for omitting a reconstruction filter (be it in software, firmware in the digital domain or as hardware in the analog domain) in the DA conversion process.
Regardles if you and some others think/reason/assume/hope/theorize/hear/know there is.
 
Last edited:

zalive

Active Member
Joined
Sep 4, 2018
Messages
263
Likes
38
I did not say that at all.. if you misinterpret the info we and your local designers give you no wonder you are still confused.

The reconstruction filter is not just a high pass filter to attenuate frequencies above Nyquist.
It is a RECONSTRUCTION filter.
It correctly fills-in (hence the word reconstruction) the spaces, in this case stairsteps, between the dots (bars in this case) and it does so BECAUSE it 'rings'.
It ALSO filters the US crap above Nyquist that you really don't want because in some circumstances that might bite you in the a...

Solderdude, I trust you are a fine audio designer and an audio engineering expert. It's obvious you have knowledge there.
However it seems you lack understanding in physics beneath.

If you use a steep low pass filter designed to block content (aliasing garbage) above Nyquist and pass through the audible band, as a result you will get a nice sine wave of a sampled frequency. Nothing else is needed and this reconstruction filter does nothing else. It's just a picturesque name which describes the end result of filtering out ultrasonic garbage.

The part you don't understand so far is that any waveform consists of frequencies. While this ultrasonic garbage obviously consists of various spurious pulses , it's still frequencies. Everything shown on graph has its frequency, regardless of how short a pulse this was. But the sampled tone frequency is pure sine. There is no error in sampling the sine tone, it will be sampled perfectly. But graph shows superposition of sampled frequency plus ultrasonic garbage. Sampled tone doesn't need reconstruction itself, it's fine right from the conversion.

You need to understand this superposition concept of various frequencies, to forget the notion that a broken tone is being generated. All belonging to the audible range generated by DA conversion is fine, those are not broken audible tones. Its just that garbage above Nyquist makes it look so, because it's what sample and hold will show. It shows a superposition, a sum.

You are correct when you MIX tones, then you don't see a pure tone.
BUT as I explained a thousand times already in this case there are no 'mixed frequencies'.
And no.... it also isn't a single frequency 'mixed' with an inaudible 44.1kHz square wave.
That would LOOK different on the scope screen.

Square wave itself is a complex tone consisting of various harmonics with various amplitudes made in a way they will sum to a square waveform.
Ultrasonic garbage consists obviously of various pulses, but pulse nature doesn't change the fact that those are frequencies as well.
The fact that they don't make a stable identical waveform in each sampled graph doesn't change a thing, the principle of superposition remains identical. You need to understand this.

In the end you can break everything down to sine waves of different duration - continuous or pulses. As I said but you don't accept it - a single frequency can only be sine. It's the nature of waves. If you look at waveform or a sample of anything different than a sine wave, you're looking at the mix of various frequencies present at the moment of sampling. There's no other option, Dude :D

I studied this once in the past, you know :D I don't work in this line, but I have a degree from the Faculty of electrical engineering and computer science, and we used to have lot of physics there, so I understand well things like Fourier series/analysis.
 

zalive

Active Member
Joined
Sep 4, 2018
Messages
263
Likes
38
I disagree with that logic, "thank God our ears are ok with a flawed unfiltered signal" does not validate NOS to me :) only that we have very tolerant ears...

Ears at first place are the premise, not a circumstance. Everything being designed in hifi audio world is being designed for human ears.
So no matter how more or less tolerant ears are their frequency bandwidth will be finite and the audio equipment would be designed for a given audible band.

No one mentioned pets. They might be a valid argument for implementing filtering. However my cats don't mind the sound from the filterless NOS (even when bit louder), so I guess at least speakers attenuated it sufficiently.
 

zalive

Active Member
Joined
Sep 4, 2018
Messages
263
Likes
38
You still are confusing reconstruction filters with gentle low pass filters and do not seem to understand that the sudden drop of HF response in ears is not the same as a steep reconstruction filter.

Reconstruction filter is nothing but a steep low pass filter.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
If you use a steep low pass filter designed to block content (aliasing garbage) above Nyquist and pass through the audible band, as a result you will get a nice sine wave of a sampled frequency. Nothing else is needed and this reconstruction filter does nothing else. It's just a picturesque name which describes the end result of filtering out ultrasonic garbage.

Totally correct, that's what I am trying to tell you all this time but you seem to insist that it is not needed.
That is the reconstruction part of it. use no filter or a gentle one and you do NOT get the perfect sine NOR will you hear it at the correct level.

The part you don't understand so far is that any waveform consists of frequencies.

What is it I do not understand about this ?

Sampled tone doesn't need reconstruction itself, it's fine right from the conversion.

tenor.gif


You need to brush up on the sampling theorem.... It is fine right after conversion WHEN A STEEP FILTER is used.
The filter is essential, not optional.

As I said but you don't accept it - a single frequency can only be sine.


Oh I accept it, it is you that doesn't accept that a sample-and-hold signal is NOT a pure sine at all !
Nor is it a pure sinewave with an added inaudible squarewave.

I have a degree from the Faculty of electrical engineering and computer science, and we used to have lot of physics there, so I understand well things like Fourier series/analysis.

giphy.gif


To quote @SIY : Get a refund You obviously did not get what you were taught.
Contact your professors, present them this thread and ask them who did not get it. (hint: don't tell 'm you are Zalive)
 
Last edited:

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
Reconstruction filter is nothing but a steep low pass filter.

Now you get it... you NEED it and can't leave it out.
And a steep filter is not the same as a gentle filter nor the same as no filter.
 
Last edited:

zalive

Active Member
Joined
Sep 4, 2018
Messages
263
Likes
38
Totally correct, that's what I am trying to tell you all this time but you seem to insist that it is not needed.
That is the reconstruction part of it. use no filter or a gentle one and you do NOT get the perfect sine NOR will you hear it at the correct level.

Then why do we waste time discussing if we agree on this?

Oh I accept it, it is you that doesn't accept that a sample-and-hold signal is NOT a pure sine at all !

It's because it is a mix of frequencies which sample-and-hold will show. It's not a broken tone, it's a mix of a perfect tone plus garbage. Right?
Perfect sine of the tone is there! You may not see it because of garbage being superponed, but it's there!

Now you get it... that's why you NEED it and can't leave it out.

If ears filter it out anyway you can. Ears are a steep enough filter themselves.
From my perspective, it's more about hurting engineers' feelings about how a proper design should look like.
Yes it is a compromise, but the valid question is, what's the gain?

Now, the benefit of filterless vs filtered in NOS: apply a filter and you'll get not only some ringing, but as well an additional roll off in highs because of filter. And the same dilemma: steeper filter with less roll off and more ringing vs milder filter with more roll off and less ringing (and less filtering).

In this respect filterless will give less roll off than filtered and no ringing. Win-win. Trade off is carrying ultrasonic garbage. Which you won't hear anyway as ears will filter it out.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
Then why do we waste time discussing if we agree on this?

Because this is the part we agree on... the part we don't agree on is the fact that you claim such a filter is NOT needed.
That filter is an obligated and integrated part of the DA process and can not and should not be left out.

It's because it is a mix of frequencies which sample-and-hold will show. It's not a broken tone, it's a mix of a perfect tone plus garbage. Right?
Perfect sine of the tone is there! You may not see it because of garbage being superponed, but it's there!

tenor.gif



Not right.. the file represents sample POINTS in time at a certain interval. It does NOT contain the whole sine.
The space in between those samples is not present in that file. It relies on a steep filter to 'interpolate' ... to 'connect the dots'.
An R2R DAC chip (without OS filters) can only convert these 'points' by holding those points for the same time interval period.
The majority of the time that sample level is erronious/wrong/incorrect/broken.
This SHOULD be repeaired and that's what the STEEP filter is for... to correct the errors... the broken part of it, the incorrect part.

If ears filter it out anyway you can. Ears are a steep enough filter themselves.
From my perspective, it's more about hurting engineers' feelings about how a proper design should look like.
Yes it is a compromise, but the valid question is, what's the gain?

The ear ONLY filters out the part that you cannot hear. That steep drop is NOT the same as the steep drop from a filter. It is based on a completely different mechanism. If it weren't and the 'steep filter' in the ear would work the same it would always be ringing at the highest frequency you can hear. That doesn't happen.

And your persepective is incorrect. It has nothing to do with hurting engineers feelings (I am not hurt), annoyed .. yes.

Now, the benefit of filterless vs filtered in NOS: apply a filter and you'll get not only some ringing, but as well an additional roll off in highs because of filter. And the same dilemma: steeper filter with less roll off and more ringing vs milder filter with more roll off and less ringing (and less filtering).

In this respect filterless will give less roll off than filtered and no ringing. Win-win. Trade off is carrying ultrasonic garbage. Which you won't hear anyway as ears will filter it out.

tenor.gif


This is completely false... If that were the case it would clealy show no roll-off in the measurements of a filterless NOS DAC.
When a NOS R2R DAC shows no audible roll-off it has a filter in it.. be it switched capacitor or otherwise and would show no stairsteps.

I am not running out of NO-gifs yet so you can continue with your incorrect assumptions for quite a while (untill the sheriff locks it ?).

Hoping at least this thread clarifies some of the workings of a DAC to some readers. It seems hopeless in your case :)
 
Last edited:
Top Bottom