Intersample over test?

[Dro]

I stumbled across this blog post by Benchmark: https://benchmarkmedia.com/blogs/application_notes/intersample-overs-in-cd-recordings

I think a intersample over test would be interesting to do with future DAC measurements and some of the gear you own. With the high incidence over intersample overs, there could be audible differences between DACs that should be indiscernible according to other measurements.

 

[bennetng]

https://audiosciencereview.com/foru...ation-audible-intersample-clipping-test.2231/

 

[Blumlein 88]

https://audiosciencereview.com/foru...ise-clipping-and-upsampling-dac-effects.1835/

This thread and the one linked in the first post of this thread also have some info, and tests done showing intersample overs which are more common if you do upsampling. Benchmark would be especially careful of this as they upsample all inputs to their own internal sample rate. I think it was 110khz sample rate on the earlier Benchmark gear and is 211 khz on more recent gear. I believe part of the reason they do this is using ASRC to filter jitter an odd resamping rate makes that work better than if you simply double or quadruple rates.

BTW, Don pointed out that with white noise you can have intersample overs with signals as much as 10 db lower than max level. Though such signals are very uncommon. So even 3.5 db headroom may almost eliminate such things, but won't guarantee they cannot happen. It is also one good argument against upsampling. You might create such intersample overs in the upsampled result than aren't there otherwise.

 

[bennetng]

A bit of supplementary information about the test signal I posted in the Reaper forum: The waveform of the test signal is not symmetrical and therefore sensitive to phase change. That means a filter with same steepness but different phase parameter (linear -> intermediate -> minimum) will result in different resampled peaks and give a false impression that some devices have more or less headroom.

In general, end users should not rely on DAC manufacturers or mastering engineers to deal with intersample peaks since this phenomenon is highly dependent on the playback signal chain and music material.

For fun, there are some highly geeky Russian engineers from iZotope tell people how to generate infinitely high intersample peaks to screw up some True Peak meters.
https://techblog.izotope.com/2015/08/24/true-peak-detection/

 

[MusicNBeer]

I just completed a test of two units, 1) Benchmark DAC1 and 2) the new SMSL M500 DAC+HP amp.

Signals:
I created 4 signals in Octave (free version Matlab clone).

Signal 1: No intersample overs --> fullscale 5512.5 Hz (44100/8), or 8 samples per cycle, 0 phase offset
Signal 2: 0.68 dB intersample overs --> fullscale 5512.5 Hz (44100/8), or 8 samples per cycle, 22.5 deg phase offset
Signal 3: 1.25 dB intersample overs --> fullscale 7350 Hz (44100/6), or 6 samples per cycle, 30 deg phase offset
Signal 4: 3.01 dB intersample overs --> fullscale 11025 Hz (44100/4), or 4 samples per cycle, 45 deg phase offset

Results:
As expected, both DACs cleanly reproduced Signal 1.
Benchmark DAC1 clipped on all 3 intersample over files.
SMSL M500 also clipped on all 3 intersample over files with volume set to 40 (max)
For Signal 2, volume reduction to 39 produced clean output.
For Signal 3, volume reduction to 38 produced clean output.
For Signal 4, volume reduction to 35 produced clean output.

ARTA was used on PC through Asus sound card as an ad-hoc audio analyzer.

 

[LTig]

I stumbled across this blog post by Benchmark: https://benchmarkmedia.com/blogs/application_notes/intersample-overs-in-cd-recordings

I think a intersample over test would be interesting to do with future DAC measurements and some of the gear you own. With the high incidence over intersample overs, there could be audible differences between DACs that should be indiscernible according to other measurements.

I did this with a Musical Fidelity V-DAC.

 

[LTig]

I just completed a test of two units, 1) Benchmark DAC1 and 2) the new SMSL M500 DAC+HP amp.

Signals:
I created 4 signals in Octave (free version Matlab clone).

Signal 1: No intersample overs --> fullscale 5512.5 Hz (44100/8), or 8 samples per cycle, 0 phase offset
Signal 2: 0.68 dB intersample overs --> fullscale 5512.5 Hz (44100/8), or 8 samples per cycle, 22.5 deg phase offset
Signal 3: 1.25 dB intersample overs --> fullscale 7350 Hz (44100/6), or 6 samples per cycle, 30 deg phase offset
Signal 4: 3.01 dB intersample overs --> fullscale 11025 Hz (44100/4), or 4 samples per cycle, 45 deg phase offset

I tested a RME ADI-2 PRI fs with a signal similar to your signal 4. It clips with Volume = 0 dB or higher. At -0.5 dB it is clean. So the RME is able to handle intersample overs of up to +2.5 dBFS.

 

[MusicNBeer]

It would be good to make a thread where we could log all the results in a single spot.

 

[MusicNBeer]

It is interesting that these converters do use a multistage attenuation scheme. The ESS chip is doing part of the attenuation on the front end prior to the upsampling. I was very happy to see that.

 

[bennetng]

Try these files:
https://www.audiosciencereview.com/...y-of-software-volume-control.5922/post-172865