MQA vs HiRez: an apples-to-apples comparison

[pkane2001]

58 minutes ago, manisandher said:

OK, I can't make it any easier than this. Here's the original 24/88.2 (for a limited time only, because I don't want to get into trouble):

 

https://drive.google.com/open?id=1jAdGtBzHDMnDlX0O3ArdhczdRGDkCMpG

 

It should be really trivial to identify the 'totally molested' MQA now.

 

Knock yourself out...

 

Mani.

 

My analysis:   D = original PCM, C = MQA

 

Here's the delta spectrum of the original minus D:

Volume diff= 0.000dB  Phase offset=701.09ms (61836.000 samples)
Difference (rms) = -132.30dB
Correlated Null Depth=202.93dB
Phase drift before correction: 0.0000 ppm

 

And here's original minus C :

Volume diff= -0.003dB Phase offset=644.02ms (56802.890 samples)
Difference (rms) = -54.84dB
Correlated Null Depth=62.84dB
Phase drift before correction: -0.0007 ppm

 

[pkane2001]

14 minutes ago, manisandher said:

 

Thanks Paul (as always). But you have two 'original minus D' there. Which plot is which?

 

Edit: I can see that the second plot is 'original minus C'.

 

Mani.

 

Yep, sorry, last minute edit messed that up. First one is the original - D.

[pkane2001]

1 hour ago, fas42 said:

So, is something going on in the time domain, in the posted waveforms, that's easy to see ... you betcha!

 

 

Okay, this is left channel only, of C vs D, where C has been aligned with D by removing samples from the beginning of C - the timing is at the top. Both have been upsampled to 352800 to improve aligning accuracy, and to see what's going on - and the key thing is that brickwall equalisation has been applied, you're only seeing the content above 10k in the above, all below 10k is attenuated near 120dB.

 

Note: the amplitudes have not been amplified to make this clearer, the levels at the left are the actual volume, compared to full scale of 1.

 

Here are the same two files (C and D) with a slightly larger scale, taken from about 2:17 mark. No upsampling or filtering applied, but the phase corrected to a fraction of a sample, as well as a minor phase drift corrected out.

 

Red represents D, the original PCM, while blue, C is the MQA encoded one. These are obviously not the same, although the differences are fairly minor, and appear to be in the higher frequencies:

 

There's no additional ringing detectable in the cepstrum plot, at least nothing obvious:

 

 

 

[pkane2001]

7 minutes ago, mansr said:

Here's a clearer plot of that difference. My tools arrive at the same sample offset. There's also some anomaly at both ends that has to be trimmed off before making the comparison. I guess you noticed it too.

 

Yes, I usually trim a few seconds from the front and the end of the files when doing comparisons for this reason. Deltawave can do this trimming automatically when detecting a large deviation between the two files at the beginning and at the end.

[pkane2001]

5 minutes ago, tmtomh said:

You reacted to your own comment, and marked it as Off Topic. That's gotta be a first here. ?

 

That's becoming much more popular on CA lately

 

[pkane2001]

On 10/16/2018 at 8:32 AM, mansr said:

Here's a clearer plot of that difference. My tools arrive at the same sample offset. There's also some anomaly at both ends that has to be trimmed off before making the comparison. I guess you noticed it too.

 

 

This looks exactly like the dither noise Audacity likes to add to anything it touches. That dither has an amplitude of 10 LSB units, meaning these files have not quite 20 bits of actual resolution left. Whether audible or not, this certainly makes any analytic comparisons between the files more difficult.

 

By the way, you are right. It's not dither from Audacity, but rather dither added in DeltaWave to generate a file with the same # of bits as the original. If I turn off dither, the result looks very similar to yours:

 

[pkane2001]

23 minutes ago, manisandher said:

 

If not dither in Audacity, what could be causing this? WASAPI Shared Mode?

 

Could this result be considered 'as good as' a perfect null, for all intents and purposes?

 

Mani.

 

 

 

It was my software that was applying dither. It's something that I can turn off, which results in a lower noise floor. An under -190dB difference and a 200dB correlated null depth are as perfect a match as one can do with the precision of calculations used in DeltaWave. PCM vs MQA captures generated only a 64dB null, and that's good, but not perfect.

[pkane2001]

2 minutes ago, mansr said:

In addition to Audacity.

 

Yes

[pkane2001]

8 hours ago, manisandher said:

 

Not sure if you've read the whole thread, but here's how Roon is configured with the MQA file:

 

 

And I have Audacity set up correctly:

 

 

Mani.

 

Hi Mani, just want to make sure you are quoting the right stats from my posts. The 200dB null was between the two non-MQA captures. The difference between MQA and non-MQA was around 60dB. Still  small, but in the audible range.

[pkane2001]

7 minutes ago, manisandher said:

 

You didn't answer my question. Would you expect a null of around -200dB between the original and the non-MQA capture if dither were being applied somewhere by something?

 

Mani.

 

Based on the shape of the difference spectrum, there appears to be some noise shaping/dither being applied. Where, I couldn’t tell you.