Maybe read the thread (and understand aliasing) first:
Best No oversampling dac to buy??
would always have been 0dB in a properly filtered DAC It’s not a thing these days.
audiosciencereview.com
-
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!
Best No oversampling dac to buy??
- Thread starter KehaDNb
- Start date
audiosciencereview.com
solderdude
Grand Contributor
Nothing destructive about a proper reconstruction filter ...
There is something destructive about not using a proper reconstruction filter.
Of course, when using 88.2 or higher sample rates the 'destructive' part (of waveform fidelity) is above the audible range so becomes moot other than possible interference with say the sample frequency of a class-D amp which could cause IM products or tweeters not liking HF components that really should not be there in the first place but are always present in filter less (or slow filter) DACs.
There is something destructive about not using a proper reconstruction filter.
Of course, when using 88.2 or higher sample rates the 'destructive' part (of waveform fidelity) is above the audible range so becomes moot other than possible interference with say the sample frequency of a class-D amp which could cause IM products or tweeters not liking HF components that really should not be there in the first place but are always present in filter less (or slow filter) DACs.
KSTR
Major Contributor
@solderdude , when you're having and IMD problem with a frequency and its image (say, 19kHz and it 25.1kHz image, @44.1kHz) then you already have a much more relevant IMD problem with in-band signals, like 19kHz + 20kHz (or a regular 19kHz plus 25.1kHz signal at a higher sample rate)
The additional error from the images is effectively negligible. Same goes for the alleged risk of damaging tweeters from the additional signal. This risk simply does not exist.
The alleged detrimental effects of image frequencies are way way overrated. The images themselves are normally inaudible except for younger listeners and the IMD problem does not exist without a much stronger in-band IMD problem.
The only thing that's really relevant is the drooping frequency response but that can be corrected for.
The additional error from the images is effectively negligible. Same goes for the alleged risk of damaging tweeters from the additional signal. This risk simply does not exist.
The alleged detrimental effects of image frequencies are way way overrated. The images themselves are normally inaudible except for younger listeners and the IMD problem does not exist without a much stronger in-band IMD problem.
The only thing that's really relevant is the drooping frequency response but that can be corrected for.
KSTR
Major Contributor
That is not correct. The reconstruction filter does reconstruct the original waveform, it does not matter whether it is analog or digital. Actually, almost all reconstruction filters today are digital. That's the whole idea of up-/oversampling (no matter if it's D/S-DAC or not), let's do the filtering in the digital domain because it is so much easier and better to do this way. You still need need an analog filter at the output but that has only to to cope with the remaining images at the upsampled rate.
solderdude
Grand Contributor
No it can't because the droop is dependent on the moment of sampling.
At one moment there may not be any droop (when the peak of the analog waveform happens to be sampled at that peak) and a few ms later it can be 6dB down.
You could 'correct' for the average droop and that might sound less rollled off and closer to a filtered one but that would include a filter and that word seems to be a big issue in the world of filterless believers.
Of course a smart salesman would call it 'droop compensation' ?
Yep, but one should consider that in music the treble is already much lower in level than mids and bass and the IM products will be lower in level and in real music very likely masked by the rest of the music. It looks terrible in measurements with near 0dBFS signals but music don't look like it.
Aside from the droop interference between the not filtered (and inaudible) HF crap coming from such a filterless R2R DAC (which in practice all have a low order low pass in there anyway) may well IM with cheaper class-D amps with a poor input filter.
May..... have never seen any tests for this.
A digital filter does not reconstruct the continuous signal, nor its frequency content. I mean, this goes without saying—as digital samples, it's PCM. The Pulse Code Modulated signal, not the signal. It's the signal with sidebands to infinity. We haven't recovered—or reconstructed—the signal yet.
Of course sample rate conversion in the digital domain mimics the components of ADC and DAC. That is, to sample rate convert from 48k to 96k, we could output to a DAC at 48k, then run its analog output into an ADC at 96k. But by inspection we can skip the domain (analog/digital) conversions and just up the rate (include the virtual zeros lying between sample impulses). That leaves us with the two bandlimiting lowpass filters (DAC, ADC). Also by inspection, we only need the lowest lowpass filter, which happens to be the one on the DAC when upsampling. Yes, that is the bandlimiting filter lowpass often called the "reconstruction filter" because it's what gets us from the discrete domain back to the continuous domain where we originally recorded.
But, for this use it's not reconstructing anything, it's just bandlimiting in the digital domain. If we move an exit gate to an entrance, we don't call it an exit gate anymore. A reconstruction filter is just a filter, like an exit gate is just a gate. "Reconstruction" is just a label to describe its job. But if we move it to the digital domain, it's not doing that job of reconstructing analog. No sense in calling it a reconstruction filter.
If that weren't inconvenient enough, when downsampling we actually keep the other filter—the one in the ADC that's never ever called a reconstruction filter—because it's the lowest.
Look, this is a trivial point for this thread. If you look at my website, you'll see I prefer to just call it a bandlimiting filter, allowing it to be seen as symmetrical with the ADC. But, you're telling me I'm wrong (I'm not), so I have to defend my position.
KSTR
Major Contributor
That's an ADC issue. When you sample a frequency exactly at fs/2 and the ADC has an non-perfect anti-imaging filter the the output values depend on the start time (phase offset).
- Joined
- Mar 29, 2021
- Messages
- 2,515
- Likes
- 4,552
What do you gentlemen mean with "drooping frequency response"?
KSTR
Major Contributor
So you are saying all the scientific / engineering textbooks are wrong?
Reconstruction filters can be realized with filters from any domain, be it digital, time-discrete analog or time-continuous analog.
source: https://inst.eecs.berkeley.edu/~ee247/fa05/lectures/L17_2_f05.pdf
Blumlein 88
Grand Contributor
- Joined
- Feb 23, 2016
- Messages
- 23,002
- Likes
- 42,729
With no filtering of a dac output there is a droop at the upper end. Seems it is 3 db down at half of the sample rate if I remember correctly.
KSTR
Major Contributor
The red curve, for NOS filter, starts to drop in level at 10kHz and is already 3dB down at 18kHz. This is a lot of change and (slightly) audible when your ears are still good to ~15kHz or so. When compensated for, you may or may not hear any difference to this and a "normal" DAC filter but when you do, it's not coming from the frequency response.
You are showing me a slide that for "reconstruction filter options" has multiple blocks that result in analog output. In other words, the reconstructed output. There's a digital filter at the start of the block. How does that justify calling a digital filter a reconstruction filter?
>So you are saying all the scientific / engineering textbooks are wrong?
^ This is incredibly annoying. Did you read what I wrote? No evidence of it. Oh, sure, I'm saying all scientific and engineering textbooks are wrong. Brilliant deduction.
- Joined
- Mar 29, 2021
- Messages
- 2,515
- Likes
- 4,552
Thank you both.
Based on the previous block diagram, found this on Wiki, which made it much clearer.
The fact that practical digital-to-analog converters (DAC) do not output a sequence of dirac impulses, xs(t) (that, if ideally low-pass filtered, would result in the unique underlying bandlimited signal before sampling), but instead output a sequence of rectangular pulses, xZOH(t) (a piecewise constant function), means that there is an inherent effect of the ZOH on the effective frequency response of the DAC, resulting in a mild roll-off of gain at the higher frequencies (a 3.9224 dB loss at the Nyquist frequency, corresponding to a gain of sinc(1/2) = 2/π). This drop is a consequence of the hold property of a conventional DAC, and is not due to the sample and hold that might precede a conventional analog-to-digital converter (ADC).
So "good filters" like fast linear etc would have compensation for this HF droop caused by ZOH in their design?
I do this every day and I don't hear any particular fault whether it's an AKM or ESS DAC which is responsible for the conversion...
But I hear a lot of piano records where the instrument is poorly recorded or poorly adjusted... And I listen to it every day for work reasons.
And I also listen to a lot of pianists in concerts; I'm even currently in Dubai for an international piano competition and I can tell you that live depending on the pianist who plays the one and only instrument on stage... I can hear very well those who master the sound they produce and those which give a harsh, dirty sound in the treble of the piano...
KSTR
Major Contributor
That process is exactly step 1 in any oversampling DAC. The second is to remove the images (all final DAC outputs are "NOS" -- Zero Order Hold) with analog filters (SC or contiuous). The relaxes the requirement for those final filter big time because you have much more bandwidth in which the signal can be rolled off, that is, a much gentler filter slope starting right above the audio band (or 100kHz, if you will) is sufficient and readily realized as a circuit.
Well, it's not a product of digital to analog conversion, it's a product of a possible implementation of digital to analog conversion. But that possible implementation is also the most practical one. Samples are impulses, so the straight forward way to make a DAC is convert each digital sample to an impulse of the proportional height, then lowpass filter to get rid of the modulation images. But impulses have little area under the curve, hard to make with good SNR, and other drawbacks. Dragging them out with zero order hold (stair steps) is more practical. But the impulse response of the step is that of a lowpass filter with the shape of the sinc function. That's where you get the droop with 3 dB at half the sample rate. It's far easier to just fix the droop than to make impulses.
Last edited:
Yeah, please read what I wrote through in one piece. I get the feeling you're seeing phrases and reacting.
It's just a lowpass filter. One whose purpose is to bandlimit. A DAC first converts from digital (inherently discrete time) to discrete-time analog to analog. The part that converts from discrete analog (ideally impulses, but for practical purposes ZOH) to continuous analog is that final analog lowpass filter. In honor of its special place in the chain and therefore purpose, it's dubbed the "reconstruction" filter.
Sure, in practical modern implementations, it's not the only thing doing the filtering, it's usually helped by digital processing (well before that final analog filter, because it has to be ahead of the discrete analog conversion). And of course, some people might say the digital filter is part of the reconstruction, and that's true if the analog filter isn't compensating for the ZOH sinc rolloff.
That doesn't mean that it makes any sense to call a digital filter on its own in an SRC algorithm a "reconstruction filter". There is no reconstruction. (There is no ZOH in the SRC either, because samples are impulses. We might describe the concept of SRC as emulating a DAC/ADC pair, but we don't actually.)
I'm not demanding that you don't call it that, I don't care what you call it. But you are saying I'm wrong, that I'm giving incorrect information (that goes against "all" scientific and electronic textbooks)—I'm not. (Wanting for a "shrug" emoji.)
This is simply not correct to state that ringing does not manifest itself in any normal audio signal. The Dac does not differ between audio signals and sinus waves.
A lot of people here reference to Archimago and he shows it in this test:
Audiophile Myth #260: The Detestable Digital Filter Ringing and Real Music...
A blog for audiophiles about more objective topics. Measurements of audio gear. Reasonable, realistic, no snakeoil assessment of sound, and equipment.
archimago.blogspot.com
If it is audible on complex music is another discussion, but of course it is manifested in an audio signal.
solderdude
Grand Contributor
And that's why you can't correct for that afterwards.
The only way to get it right is to use a good reconstruction filter and is the main reason why filterless 44.1 and 48kHz is 'broken'.
With 88.2kHz and up it isn't audibly as broken but the unwanted HF garbage is still undesirable but is most likely, in most situations harmless.
Yes, ringing occurs whenever there is something in the audio signal that is not correctly band-limited like you see when it clips. I'd say: you'll have other things to worry about than inaudible pre-rining in those instances, and NOS will not solve any of them.This is simply not correct to state that ringing does not manifest itself in any normal audio signal. The Dac does not differ between audio signals and sinus waves.
A lot of people here reference to Archimago and he shows it in this test:
Audiophile Myth #260: The Detestable Digital Filter Ringing and Real Music...
A blog for audiophiles about more objective topics. Measurements of audio gear. Reasonable, realistic, no snakeoil assessment of sound, and equipment.
If it is audible on complex music is another discussion, but of course it is manifested in an audio signal.
Similar threads
