KSTR
Major Contributor
There is a lot of discussion on how different DAC reconstruction filters sound like, besides the pure technical aspects which are easily explained and visualized with frequency response plots where the imaging -- mirrored frequencies above half the sample rate -- can be seen (if any), as well as impulse responses shown. I won't repeat that in all detail here as it's been covered already multiple times (for example, here).
Direct listening tests are not easy as we normally don't use any material with sample rates less than 44.1kHz and hence artifacts (both in frequency and time domains) start at close to fs/2 (22.05kHz) which is outside the range most adult individuals can hear.
Therefore, I've made some test files where I use a sample rate of 11.025kHz, 1/4th of 44.1kHz. Frequency response thus is restricted to 5.5125kHz and then artifacts above that range will be easily heard. Further, any effect of ringing etc also is much better exposed as that will happen right at this 5.5125kHz.
The test snippet is a music sample that doesn't have too much high frequency energy and fast transients as that would probably overexpose the effects since I've effectively scaled the audio two octaves up, from the sample rate point of view. Some of you may have heard that snippet already, in @Blumlein 88 's 8th generation DAC-ADC loopback test.
What I did is the following:
1) Downsample the 44.1kHz source file with Audition to 1/4x, using the most precise settings. This a standard (linear phase) close to ideal "sinc" filter with pre and post filtering and is representative of the situation when an analog stream is AD-converted with a textbook proper anti-aliasing filter.
2) Upsampling 4x back to 44.1kHz with NOS "filter" (which simply repeats each sample four times), using Foobar2000's "multiresampler" plugin (with Zero Order Hold, which is the technical name for NOS).
The somewhat irregular shape comes from the actual samples being quadrupled which are just the sample values of the previous impulse which are not fully symmetric (would show up at higher zoom levels).
3) Another upsampling 4x of the downsampled track with a true maximised sinc filter (passband to 99%, linear phase, filtered), using Foobar2000's "SoX resampler" plugin . This gives an impulse reponse with the usual pre-ringing and post-ringing, quite long in this case, from the maximally extended passband.
4) A third upsampling 4x with with the mimimum phase version of the above (otherwise same settings) This gives an impulse reponse with only post-ringing and even longer duration.
Basically, this covers the extremes. Actual filters in DACs usually are in between, notably the filters have only little rejection at fs/2 (where a true sinc filter would have already full attenuation) and reach their stop-band up higher -- not as bad as NOS (which does not attenuate at all) but still images will be created.
Listening to the NOS clearly shows that nasty mirrored content.
Telling the two steep filters apart is not so easy (unless one would listen to the impulse shown, directly, where the long tail of the minimum phase filter clearly can be identified). You may post ABX logs of attempts to do so (I failed in quick run).
Have fun!
https://www.dropbox.com/s/xddak5sjsh8eyzs/RecontructionFilterComparison.zip
Direct listening tests are not easy as we normally don't use any material with sample rates less than 44.1kHz and hence artifacts (both in frequency and time domains) start at close to fs/2 (22.05kHz) which is outside the range most adult individuals can hear.
Therefore, I've made some test files where I use a sample rate of 11.025kHz, 1/4th of 44.1kHz. Frequency response thus is restricted to 5.5125kHz and then artifacts above that range will be easily heard. Further, any effect of ringing etc also is much better exposed as that will happen right at this 5.5125kHz.
The test snippet is a music sample that doesn't have too much high frequency energy and fast transients as that would probably overexpose the effects since I've effectively scaled the audio two octaves up, from the sample rate point of view. Some of you may have heard that snippet already, in @Blumlein 88 's 8th generation DAC-ADC loopback test.
What I did is the following:
1) Downsample the 44.1kHz source file with Audition to 1/4x, using the most precise settings. This a standard (linear phase) close to ideal "sinc" filter with pre and post filtering and is representative of the situation when an analog stream is AD-converted with a textbook proper anti-aliasing filter.
2) Upsampling 4x back to 44.1kHz with NOS "filter" (which simply repeats each sample four times), using Foobar2000's "multiresampler" plugin (with Zero Order Hold, which is the technical name for NOS).
The somewhat irregular shape comes from the actual samples being quadrupled which are just the sample values of the previous impulse which are not fully symmetric (would show up at higher zoom levels).
3) Another upsampling 4x of the downsampled track with a true maximised sinc filter (passband to 99%, linear phase, filtered), using Foobar2000's "SoX resampler" plugin . This gives an impulse reponse with the usual pre-ringing and post-ringing, quite long in this case, from the maximally extended passband.
4) A third upsampling 4x with with the mimimum phase version of the above (otherwise same settings) This gives an impulse reponse with only post-ringing and even longer duration.
Basically, this covers the extremes. Actual filters in DACs usually are in between, notably the filters have only little rejection at fs/2 (where a true sinc filter would have already full attenuation) and reach their stop-band up higher -- not as bad as NOS (which does not attenuate at all) but still images will be created.
Listening to the NOS clearly shows that nasty mirrored content.
Telling the two steep filters apart is not so easy (unless one would listen to the impulse shown, directly, where the long tail of the minimum phase filter clearly can be identified). You may post ABX logs of attempts to do so (I failed in quick run).
Have fun!
https://www.dropbox.com/s/xddak5sjsh8eyzs/RecontructionFilterComparison.zip