DacMagicPlus Filter Mirror Images - Short Study

[bennetng]

The worse situation is with Fs = 44.1 kHz and the mirror images are same as shown here, so the issue is not limited to 23kHz/48kHz. The cure is proper filtering or higher sampling rate. I think that manufacturers want for their products look better so they extend the passband as high as possible at the expense of mirror images. Marketing vs. proper technical solution, as always.

It depends on how you define "marketing". In this case I would say your DacMagic Plus is also "marketing" their filters as well. If there is only one correct way to resample, then I don't know why they also offer filter selection and explain things in this way.

From the manual:

We encourage you to experiment with the filters to determine which sound best to your ears and
using your source equipment/programme material.

A bandlimited, alias-free square wave can be decomposed into a series of tones with decaying amplitude towards Nyquist, which is at least a slightly more practical signal than yours.

My Realtek on the other hand uses a filter which is very similar to Mola Mola Tambaqui but I can't find any marketing stuff from Realtek about this particular behaviour.

Mola Mola @ 44k

Realtek ALC892 @ 48k

At the end, you still evaded my linked discussion thread about someone listened to audio files with 24kHz sample rate but still prefer the sound of imaging. If using your near Nyquist high amplitude tone as example then the tone would be at 11.5kHz, so tell me what is the safe level to listen to this signal in order to make the IMD products audible?

The reality is that one cannot completely ignore psychoacoustics when dealing with audio signals. It is fine to "request" a desired filter type, but claiming other people's preferences wrong is completely ridiculous.

 

[solderdude]

Do you reckon at 48kHz the filter frequency also shifts upwards 8.9% ?
The filter does fit perfectly for 48kHz.

Obviously it would be just as easy to make correct filters for 44.1 and multiples. The question is why some chip manufacturers don't.
Too much hassle to use 2 different clock signals ?
Do they want to boast over a 'wider' frequency range (22kHz for CD) ?
Do manufacturers feel they should filter higher as at the ADC stage steep filters are already used so there is no content above 20kHz to begin with ?
Is the issue only present with artificial test signals and not with actual recordings (which are down-sampled anyway)?

When comparing recorded music and nulling it with the DAC output does one see IM products when too high reconstruction filters are used ?
When we have a measurement file, run it through a proper anti-alias filter and then look at the signal from a incorrectly filtered DAC at 44.1Hz do we still get IM products ?
All assuming steep filters are used of course.

 

[bennetng]

Do you reckon at 48kHz the filter frequency also shifts upwards 8.9% ?
The filter does fit perfectly for 48kHz.

It is a limitation of the ALC892 codec. It runs in 48kHz based rates internally, tests with 44.1k will introduce additional artifacts due to non-integer resampling ratio like this:
https://www.audiosciencereview.com/...-sample-rate-is-enough.4037/page-3#post-95240

As you can see, at 44k, the filter completely fades into noise at about 26kHz, so a bit softer than CS43198 and AK449x, but very similar to the Mola Mola. The additional distortions in the 44.1k graph is caused by the aforementioned limitation, and irrelevant to the filter cutoff point/steepness. The Realtek's 48kHz plot fades into the noise floor at about 28kHz, so it is also relative to sample rate as well.

 

[bennetng]

To avoid confusion, here is the Realtek filter at 44.1k. Near Nyquist tone not shown due to irrelevant non-integer resampling artifacts.

 

[ElNino]

This is a great thread (all your threads are awesome, btw).

The parameter that counts is the attenuation below Fs/2, in our case 20-24kHz band and the steep filter is much better here, as we can see from mirror images. The absolute attenuation level in the rejection area is not that much of interest. Just race of numbers. On the other hand, attenuation at Fs/2 matters, even if some designers are not willing to admit.

I draw the opposite conclusions from your results here. Your results show that, at worst, the amount of IMD caused by aliases in the passband with the "slow" filters is more than about -125 dB down -- i.e., definitely inaudible. You don't show the time domain measurements, but the steep filter necessarily has about double the amount of pre-ringing. There is debate about how much pre-ringing is audible, but to the extent it may be audible, it's more likely to be audible than IMD at -125 dB down. (When you hear increased presence with the sharp filter, I'd argue what you're hearing is a psychoacoustic effect of the pre-ringing.)

A lot of people on this forum (not you) don't seem to understand that you can't have good attenuation at Fs/2 on 44.1kHz sampled data while preserving 20kHz bandwidth without also having considerably more pre-ringing. You can't have your cake and eat it too because of the way the math works. The reason DAC designers from all of the major IC vendors have generally settled on filters with a 0.54Fs stopband isn't because they're incompetent -- it's because they realize the inherent tradeoff in time-domain performance. For telecom applications or higher sample rates, Fs/2 stopband is obviously ideal, but at 44.1Khz for audio applications there is no ideal solution, and I'd even go so far as to say that Fs/2 is more "wrong".

 

[Veri]

A lot of people on this forum (not you) don't seem to understand that you can't have good attenuation at Fs/2 on 44.1kHz sampled data while preserving 20kHz bandwidth without also having considerably more pre-ringing. You can't have your cake and eat it too because of the way the math works. The reason DAC designers from all of the major IC vendors have generally settled on filters with a 0.54Fs stopband isn't because they're incompetent -- it's because they realize the inherent tradeoff in time-domain performance. For telecom applications or higher sample rates, Fs/2 stopband is obviously ideal, but at 44.1Khz for audio applications there is no ideal solution, and I'd even go so far as to say that Fs/2 is more "wrong".

I do wonder...

 

[solderdude]

It is a limitation of the ALC892 codec. It runs in 48kHz based rates internally, tests with 44.1k will introduce additional artifacts due to non-integer resampling ratio like this:

Yes, I get that.
In these tests a tone sweeps above 20kHz.
What if one (at the recording stage) filters out all signals above 20kHz sharply.
This already happens at the downsampling in when creating a 44.1kHz file.
There will be no content above 20kHz so the signal stops before 1/2fs.

So repeating the measurement but attenuate the sweep to 20kHz (or at least using an anti-alias filter intended for 44.1kHz)

A bit like 320kbs MP3 which doesn't go beyond 18kHz. When playing back an MP3 at max. resolution (24/44.1) there will never be IM products even when using an 48kHz reconstruction filter.

I have not seen redbook recordings with content above 20kHz (due to the anti-alias filter) so why would a 48kHz reconstruction be bad thing ?

 

[bennetng]

Obviously it would be just as easy to make correct filters for 44.1 and multiples. The question is why some chip manufacturers don't.
Too much hassle to use 2 different clock signals ?
Do they want to boast over a 'wider' frequency range (22kHz for CD) ?

No and no because the filter cutoff is proportional to sample rate.
https://www.audiosciencereview.com/...ds/jds-labs-atom-dac-review.14002/post-427203
https://www.audiosciencereview.com/...u-m4-audio-interface-review.15757/post-532975

Even for old chips without selectable filters.

so why would a 48kHz reconstruction be bad thing ?

A slightly late reply after reading your new post, hope you understand what I meant. No such things as "filters are designed for 48kHz".

 

[solderdude]

I mean with that a filter that is designed for 48kHz would have full attenuation at 24kHz as all these built-in reconstruction filters from some DAC chip manufacturers can do.

When one uses a 48kHz test signal does the filter frequency also shift upwards in the same percentage ?
Why does no one test at 48kHz ?
Are filters at 96kHz correct ?
When one would test at 88.2kHz is the filter set way too high as well ?

Why would a music signal that has no contents above 20kHz have aliasing ?

the filter cutoff is proportional to sample rate.

This is what one would expect (would be the easiest to build). The question remains why chip some manufacturers still create incorrect filters not fitting to the sample rate. Is that lazyness ? Is that difficult because one cannot use 50% of the sample rate that easily but has to have a certain percentage of the samplerate ? In that case is it related to the master clock frequency ? Why can ESS make correct filters but other manufacturers can (or choose) not to ?

 

[bennetng]

What if one (at the recording stage) filters out all signals above 20kHz sharply.
This already happens at the downsampling in when creating a 44.1kHz file.
There will be no content above 20kHz so the signal stops before 1/2fs.

So repeating the measurement but attenuate the sweep to 20kHz (or at least using an anti-alias filter intended for 44.1kHz)

Because published recordings are most likely not as simple as ADC -> DAC. They can be combined with different materials with different sample rates, or from samplers/synthesizers, and many other different processes which involve different kinds nonlinear processing.
Just a random example of these things released on a CD. Music, not test signal. Download the file to see frequencies up to fs/2.
https://www.audiosciencereview.com/...e-rate-and-audible-frequency.9411/post-304560

As mentioned in another post, the ESS filter with full fs/2 attenuation exhibits other kinds of artifacts, so there is a trade off.
https://www.audiosciencereview.com/...-d30pro-review-balanced-dac.20259/post-668061

 

[solderdude]

Because published recordings are most likely not as simple as ADC -> DAC. They can be combined with different materials with different sample rates, or from samplers/synthesizers, and many other different processes which involve different kinds nonlinear processing.
Just a random example of these things released on a CD. Music, not test signal.

I know of very little recording (studio's) working on 44.1kHz so yes, during the mastering for CD there will always be an anti-alias filter in the down sample process.
I expect there to be good and poorer ones.

My specific question is IF one has a properly produced 44.1 file, with proper filtering starting at 20kHz and down minus a lot at 22.05kHz and thus no signals present above 22.05kHz will there still be 'crap' above 24kHz on the output that could then IM with the audible frequencies below ?
My question is will there be a difference between a proper reconstruction filter (- a lot dB) at 22kHz and - the same lot dB at 24kHz ?

I understand that not all CD's are properly mastered (anti-alias filtered).

When one would only play 44.1kHz MP3 (limited to 18-19kHz or so) and one uses a reconstruction filter at 20kHz or 22kHz then would the result be the same ? I mean the nearest alias would be at 25khz which both reconstruction filters would filter with say -140dB or so.

 

[bennetng]

To avoid confusion as we are talking about both ADC/DAC and recording/playback, aliasing and imaging are two things.
https://www.audiosciencereview.com/...u-m4-audio-interface-review.15757/post-504008
If we assume all ADC/DAC/resamplers etc have a standardized, identical set of filters, then everything must be perfect. In reality it is not. One may use a filter starting from 0.45fs to 0.55fs, others may use 0.4 to 0.51, and perhaps with different level of attenuation and such, so there is no way guarantee anything. The CDDA format, or the file/data format does not have these forced specs, so in reality there will be some measurable artifacts in some cases. However I am not the one who worry about these things. As mentioned in the first page of this thread, I cannot hear differences between typical filters (which some people considered as "wrong") at 44.1kHz, as well as the "correct" ones, which I can always simulate them with software resamplers.

 

[AnalogSteph]

I know of very little recording (studio's) working on 44.1kHz so yes, during the mastering for CD there will always be an anti-alias filter in the down sample process.
I expect there to be good and poorer ones.

https://src.infinitewave.ca/ is a good source for these things. Some big-name DAWs are surprisingly poor, while the SoX resamplers built into Audacity and available for Foobar2000 are as good as anything and free.

 

[solderdude]

To avoid confusion as we are talking about both ADC/DAC and recording/playback, aliasing and imaging are two things.
https://www.audiosciencereview.com/...u-m4-audio-interface-review.15757/post-504008
If we assume all ADC/DAC/resamplers etc have a standardized, identical set of filters, then everything must be perfect. In reality it is not. One may use a filter starting from 0.45fs to 0.55fs, others may use 0.4 to 0.51, and perhaps with different level of attenuation and such, so there is no way guarantee anything. The CDDA format, or the file/data format does not have these forced specs, so in reality there will be some measurable artifacts in some cases. However I am not the one who worry about these things. As mentioned in the first page of this thread, I cannot hear differences between typical filters (which some people considered as "wrong") at 44.1kHz, as well as the "correct" ones, which I can always simulate them with software resamplers.

I'm not worried either. Just was wondering if at the recording side something was 'properly' filtered then the higher reconstruction filter would not be as 'problematic' as some make it out to be. I have looked at quite a few recordings in the past and noticed the vast majority of recordings was clipped at 20kHz. These, would not give IM products not even with the filters that are tuned a bit higher.

Yes, I understand not all productions are the same and not all resamplers, DACs and ADC's are the same.

 

[ElNino]

I'm not worried either. Just was wondering if at the recording side something was 'properly' filtered then the higher reconstruction filter would not be as 'problematic' as some make it out to be. I have looked at quite a few recordings in the past and noticed the vast majority of recordings was clipped at 20kHz. These, would not give IM products not even with the filters that are tuned a bit higher.

If a super high order brickwall filter was applied before saving the 44.1kHz digital file, e.g., passband to 19kHz, stopband at 20kHz, then you still get aliases from 24.1kHz onwards. Knowing this would make digital reconstruction filtering easier, but it actually doesn't solve any problem, because by applying that super high order brickwall filter, the cost in the time domain for recording thru reconstruction will end up being worse. There is an inherent trade-off that's actually a mathematical result of performing the original bandlimited sampling.

 

[JohnYang1997]

If a super high order brickwall filter was applied before saving the 44.1kHz digital file, e.g., passband to 19kHz, stopband at 20kHz, then you still get aliases from 24.1kHz onwards. Knowing this would make digital reconstruction filtering easier, but it actually doesn't solve any problem, because by applying that super high order brickwall filter, the cost in the time domain for recording thru reconstruction will end up being worse. There is an inherent trade-off that's actually a mathematical result of performing the original bandlimited sampling.

nonsense

 

[ElNino]

nonsense

Sorry, are you being facetious? I did a decent amount of signal processing theory in grad school, I know what I’m talking about.

 

[mansr]

Sorry, are you being facetious? I did a decent amount of signal processing theory in grad school, I know what I’m talking about.

Maybe you do. What you said above still doesn't make sense.

 

[ElNino]

Maybe you do. What you said above still doesn't make sense.

Can you identify what the issue is? Maybe I wasn’t being clear. John’s “nonsense” comment doesn’t help me understand in this regard.

 

[mansr]

Can you identify what the issue is? Maybe I wasn’t being clear. John’s “nonsense” comment doesn’t help me understand in this regard.

OK, I'll try to break it down.

If a super high order brickwall filter was applied before saving the 44.1kHz digital file, e.g., passband to 19kHz, stopband at 20kHz

Well, that's not a particularly sharp (i.e. high order) filter.

then you still get aliases from 24.1kHz onwards.

Uh-huh?

Knowing this would make digital reconstruction filtering easier

Uh-huh?

but it actually doesn't solve any problem

Uh-huh?

by applying that super high order brickwall filter, the cost in the time domain for recording thru reconstruction will end up being worse

Uh-huh?

There is an inherent trade-off that's actually a mathematical result of performing the original bandlimited sampling.

Uh-huh?