ISP Analyzer - When true peak scan cannot represent what happens in a real DAC

[danadam]

I am pretty sure we can come up with some ideal test signal of ISP that can be used in a listening test, which would sound like 'clicks' or 'grainy noise'.

Here's one with tones:

Invitation: Audible intersample clipping test

The test signal is attached at the end of this post: https://forum.cockos.com/showthread.php?t=201868 Basically I don't like the idea of blaming those who make music or DACs for intersample clipping since there is no absolute clipping point. My test signal is a bit more sophisticated than the...

www.audiosciencereview.com

Note, there were some concerns about possibility of equipment damage:

Can you really hear the sound details over 20kHz?

WARNING - PLAY THE ATTACHED FILE ON HEADPHONES ONLY. HIGH LEVEL PLAYBACK MAY DAMAGE TWEETERS. Hello, are you a moderator in Hydorgenaudio? I suppose moderators in HA should have high degree of audio knowledge so my question here: https://hydrogenaud.io/index.php?topic=114816.msg946524#msg946524...

www.audiosciencereview.com

 

[OHtaru]

Note, there were some concerns about possibility of equipment damage

I know that, so I didn't play longer.

Here's one with tones

Wow, this is obvious!!!!!

 

[morillon]

It would be great to get a rough idea, thanks to your approach, of the additional headroom to allow with the most common filters supplied with AKM or ESS DACs.... What about the additional headroom that also seems necessary when using digital compression? Should all these margins be combined as a precaution? ;-)

 

[OHtaru]

It would be great to get a rough idea, thanks to your approach, of the additional headroom to allow with the most common filters supplied with AKM or ESS DACs.... What about the additional headroom that also seems necessary when using digital compression? Should all these margins be combined as a precaution? ;-)

What does "digital compression" mean? Dynamic range compressor or lossy compression like mp3 and AAC？

 

[morillon]

MP3 AAC etc

 

[OHtaru]

MP3 AAC etc

This also requires headroom, and it's more about player software than your DAC. Lossy formats should be decoded to float32. It is recommended that you use Foobar2000, which has no problems at all in this regard. It can not only correctly decode samples greater than 0dBFS, but also output samples greater than 0dBFS, which are handed over to the Windows mixer for final attenuation.

 

[tccalvin]

Thank you so much for the tool (and the idea)!

I've been saying for a while that a sure-fire way to test for ISPs is to use the exact same filters that are present in real DACs. What better way to test a signal than to send it through a simulated DAC LP filter? You can't go wrong with that.

 

[danadam]

Additionally, this function can amplify the differences between linear phase and minimum phase, allowing you to better hear the differences without being confused by vague audiophile descriptions. The figure below shows the spectrum of a single-sample impulse after being filtered 1000 times:

Stacking a linear filter should have rather negligible effects. With a sox sinc that emulates a typical DAC filter:

Here's spectrogram of it applied 1x, 10x, 100x and 1000x, linear phase on top, minimum phase at the bottom:

The null with "1x" for the linear filter is (in dBFS):

• 10x: -30.21 peak, -64.15 rms
• 100x: -26.60 peak, -60.95 rms
• 1000x: -24.74 peak, -59.49 rms

Another interesting result, which you could probably already infer from the spectrograms. Here are peak levels in the files (in dBFS), linear phase:

• 1x: -1.00
• 10x: -1.31
• 100x: -1.47
• 1000x: -1.58

and minimum phase:

• 1x: -3.59
• 10x: -10.41
• 100x: -16.97
• 1000x: -23.59

 

[OHtaru]

Stacking a linear filter should have rather negligible effects. With a sox sinc that emulates a typical DAC filter:

View attachment 503885

Here's spectrogram of it applied 1x, 10x, 100x and 1000x, linear phase on top, minimum phase at the bottom:

View attachment 503884

The null with "1x" for the linear filter is (in dBFS):

• 10x: -30.21 peak, -64.15 rms
• 100x: -26.60 peak, -60.95 rms
• 1000x: -24.74 peak, -59.49 rms

Another interesting result, here are peak levels in the files (in dBFS), linear phase:

• 1x: -1.00
• 10x: -1.31
• 100x: -1.47
• 1000x: -1.58

and minimum phase:

• 1x: -3.59
• 10x: -10.41
• 100x: -16.97
• 1000x: -23.59

I think the 1000x min phase version can't be described as "negligible". It basically loses the spike sounding but is something like sine sweep.
As for why the minimum phase filter continuously decreases the peak, that's because you use an impulse. Try some music and the result will be different.

 

[danadam]

I think the 1000x min phase version can't be described as "negligible".

Sure, and I didn't describe it as such (in case that's what you are implying). I was talking only about the linear one: "Stacking a linear filter ..."

As for why the minimum phase filter continuously decreases the peak, that's because you use an impulse. Try some music and the result will be different.

True. I thought that maybe something with big transients, like "Tricycle", would show some more significant change, but it's only a few dB (though in both directions). The "1.orig.flac" snippet (44k) from the attachment has -2.71 dBFS peak. Upsampling (adding zero samples) and applying the linear phase filter increases it to -2.66 dBFS and then applying the filter again doesn't change the peak anymore. Upsampling and applying the minimum phase filter increases it to -2.63 dBFS and then applying it again, the peak changes to anywhere between -0.41 at 138th iteration and -5.13 at 816th iteration.

 

[OHtaru]

I was talking only about the linear one: "Stacking a linear filter ..."

My bad.

Upsampling and applying the minimum phase filter increases it to -2.63 dBFS and then applying it again, the peak changes to anywhere between -0.41 at 138th iteration and -5.13 at 816th iteration.

This should be the co-effect of phase rotation and peak dispersion (dominant when the number of iterations is sufficient).