Fixing insufficient DAC filter roll off frequency with digital EQ?

[anphex]

Greetings everyone,

since even in high end DACs we usually see the stock rolloff filter hitting it's full attenuation at only 24Khz @ 44,1Khz, which is by definition of Nyquist a little too much.
Now, could I theoretically fix this by just doing this:
(Tiefpassfilter = low pass filter)

Resulting in this:

?

I'd probably get this filter steeper and more precise, but it works for now to achieve no resonance while still maintaining a steep curve.

 

[solderdude]

The reason why many reconstruction filters attenuate max at around 24kHz (with 44.1kHz) seems illogical.
Some reconstruction filters do attenuate max. at 22kHz which makes sense.

The reason most filters do have max attenuation 24kHz is done because the bandpass is 'flat' well up to 21kHz or so.

The good news is that at the recording stage there is an anti-alias filter and before 'pressing' the CD the audio bandwidth is limited to around 20kHz.
So 2kHz short of the Nyquist.
This also means that 2kHz past Nyquist there should be no mirror images that need to be filtered. Those start at 24kHz and are the highest frequencies in the audio signal so very small in amplitude anyway.
Thus filtering at 24kHz (max. attenuation) makes kind of sense. There is no need to do additional filtering... unless you are using very slow filters or filterless NOS.
But in those cases adding an extra steep filter makes no sense if the goal is to get the 'slow filter' or 'NOS' sound.

So when steep filtering is done with enough attenuation at 24kHz there is no need to 'fix' this.
Another option is to buy one of those DACs with a proper filter that does adhere to the sampling theorem and does not count on the BW being filtered at 20kHz in the 'creation' stage.

This is where Amir's reviews come in handy.
Me... I am not bothered by it (anymore).

 

[DVDdoug]

And.... It's really not that important! Not as important as the anti-aliasing filter on an ADC.

Several years ago I was doing some experiments on a soundcard and oscilloscope at work (after hours). I don't remember what the experiment was, but I was SHOCKED to see a completely unfiltered stair-stepped output!!!

I didn't have great speakers at work and it was a fairly-average computer-and-speakers setup. Nor do I have "golden ears". But I had never noticed anything wrong with the sound. Then I realized that any harmonics were above the audio range. Plus, the amplifier probably had some filtering, and the speakers certainly had "mechanical filtering" (I doubt the tweeter went to 20kHz).

 

[anphex]

The reason why many reconstruction filters attenuate max at around 24kHz (with 44.1kHz) seems illogical.
Some reconstruction filters do attenuate max. at 22kHz which makes sense.

The reason most filters do have max attenuation 24kHz is done because the bandpass is 'flat' well up to 21kHz or so.

The good news is that at the recording stage there is an anti-alias filter and before 'pressing' the CD the audio bandwidth is limited to around 20kHz.
So 2kHz short of the Nyquist.
This also means that 2kHz past Nyquist there should be no mirror images that need to be filtered. Those start at 24kHz and are the highest frequencies in the audio signal so very small in amplitude anyway.
Thus filtering at 24kHz (max. attenuation) makes kind of sense. There is no need to do additional filtering... unless you are using very slow filters or filterless NOS.
But in those cases adding an extra steep filter makes no sense if the goal is to get the 'slow filter' or 'NOS' sound.

So when steep filtering is done with enough attenuation at 24kHz there is no need to 'fix' this.
Another option is to buy one of those DACs with a proper filter that does adhere to the sampling theorem and does not count on the BW being filtered at 20kHz in the 'creation' stage.

This is where Amir's reviews come in handy.
Me... I am not bothered by it (anymore).

This all makes sense, but I doubt that in this day and age of self releasing artists or small labels through digital services there isn't really that much strict "mastering" left that actually takes those details into account. Unless there is some security-feature in most DAWs that takes care of that I bet there are plenty of songs around that go up to 22 Khz.

 

[solderdude]

Well you can just check out the frequency response of the files. You will see not many 44.1 recordings ever have much signal over 19kHz-20kHz.
Youtube is 48kHz so the filter is higher up too.

Most self respecting home recording artists record at 96kHz and some even at 192kHz.
There won't be many recordings made in native 44.1kHz.
When they downsample to 44.1kHz there is always a sharp filter in place which ensures the FR is limited to around 20kHz.

Do you have any 44.1kHz recordings with a significant signal all the way up to 22kHz ?

 

[voodooless]

Also mind the phase rotation all those minimum phase filters will give you.

 

[AnalogSteph]

If what little aliasing is generated above fs/2 from signal components above 20 kHz is bothering you, just use some decent-quality upsampling (SoX resampler for Foobar @VHQ etc.). I don't think a parametric really is the right tool for the job, even if it would basically work.