Blue or red pill?

[pkane2001]

24 minutes ago, Summit said:

 

Phase noise is measurements of variations in signal timing presented in the frequency domain.

 

Jitter is measurements of the variations in the time domain. We have two different types of jitter: deterministic and random.

 

Phase noise or jitter are not measurements, they are the actual variation in timing of digital samples, which can be measured in frequency domain or in time domain. Regardless of how you measure them, they represent the same physical effect.

[pkane2001]

34 minutes ago, Summit said:

 

Why should we?

 

Present tense third-person singular of be

Dialectal present tense first-person and third-person singular of be

Dialectal present tense plural of be

 

Where were you when the then president of USA needed help with this?

[pkane2001]

Just now, Summit said:

 

Yes phase noise and jitter are different ways of measuring (quantifying) the same phenomenon.

 

But remember that you there the one that asked “Whats the difference between phase noise and jitter? And I simply answered the question, which is that phase noise is measurements of variations in signal timing presented in the frequency domain and jitter is measurements of the variations in the time domain.

 

The very difference IS which parameters of the signal that has been measured no way around it!

 

You missed the context of my question, which was an answer to another statement.

[pkane2001]

1 minute ago, Summit said:

 

No matter the context the very difference IS still which parameters of the signal that has been measured no way around it!

 

Again, no. The context was not measurement but the cause of audible differences. In that context, phase noise and jitter are the same physical effect and it makes no sense to distinguish them.

[pkane2001]

Just now, Summit said:

 

OMG! You asked “Whats the difference between phase noise and jitter? I answer that. You then try to lecture me about thing you know nothing about.  

 

Maybe you should go back to defining the word ‘is’. That was much more helpful.

[pkane2001]

1 hour ago, semente said:

 

It wasn't an answer but a question. It didn't clarify, correct, elucidate, educate...

Wow, tough crowd today  it was a leading question. I’m now sorry I asked.

[pkane2001]

1 hour ago, mansr said:

Those spikes in the test file are caused by rounding errors in the LSB of the 24-bit signal. If the DAC simply truncates the input to 16 bits, this will create bigger rounding errors resulting in the measured spikes.

 

That makes sense. Did the Altmann DAC have dither and was that enabled for the captures?

 

[pkane2001]

50 minutes ago, adamdea said:

There 's not a whole lot in it but the spectral difference components look to be maybe slightly higher for A B (both 65k points)

 

track 3 (A) and 4(B)

 

tracks 6 and 8 (both Bs)

 

 

Make sure you remove the first 0.5 second or so from both captures. Mans indicated there were some differences at the beginning of the captures.

 

[pkane2001]

Just now, adamdea said:

I'll happily leave someone else to do this, although i can't see anything special about the first 0.5s. [at a glance fft with and without it look the same] More important to miss off the last 2 seconds.

 

Here are two digital captures, overlaid, that include the entire length of both captures. Once I lopped off the beginning and the end, the captures show no differences at all:

 

[pkane2001]

Just now, adamdea said:

Are you referring to the s/pdif captures? I am referring to the analogue outs. You will never get two analogue captures to null perfectly.

 

As I said, "Here are two digital captures"

 

[pkane2001]

2 minutes ago, adamdea said:

More to the point you started this hare running 

 

"Make sure you remove the first 0.5 second or so from both captures. Mans indicated there were some differences at the beginning of the captures."

 

I have never been talking about the recording of the orginal bitstream but the analogue outs. [btw they are of course all digital captures....]

 

Sorry if this wasn't clear. If digital SPDIF captures (as opposed to analog) show differences, then it's reasonable to assume that the same differences might exist in the analog captures. Hence my suggestion to lop off those parts.

 

[pkane2001]

17 minutes ago, testikoff said:

A real PITA trying to align the analogue captures 1 & 2, but here are the spectral graphs:

 

- Spectra/delta (frequency range 20...88200Hz; logarithmic scale):

 

- Delta (frequency range 45000...50000Hz; linear scale):

 

Some strange spiking near ~48.5kHz frequency...

 

So I'm clear: you phase aligned the two analog captures in the time domain, and then subtracted one from the other and then did an FFT plot of the difference?

[pkane2001]

9 minutes ago, testikoff said:

No, just lined up samples the best I could on time axis (almost impossible to do with all the HF noise present), plotted their spectra & then calculated spectral delta.

 

Ah, thank you, I understand. Not sure if you really need to carefully align samples in the time domain to get the spectrum plot differences, but it shouldn't hurt anything.

 

[pkane2001]

47 minutes ago, esldude said:

For what its worth, without sample rate drift correction Audio Diffmaker shows a null depth of 37.3 db left and 37.5 right.  With sample rate drift correction (when it doesn't blow up) it shows a null depth of 94.5 db left and 92.3 db right.  

 

I complain about Diffmaker quite a lot.  When it works and shows deep null depth it isn't wrong in my experience.  More often it blows up or shows something ridiculous like a null depth of less than 10 db when its obvious by other means the null is deeper than that. 

 

So curious and interesting results.  I am not sure what could be left with nulls into the 90s with a 16 bit DAC with a noise in that same vicinity.  

 

So here's what I tried (similar to what was proposed earlier): align the A and B analog captures in Audacity, using sample # display.

 

Invert one B track, combine A and B. Plot the spectrum of the resulting track. There's definitely noise as the result, but also a few frequencies. Here's the resulting spectrum plot:

 

 

 

Here are the original tracks (top one is A, middle is B inverted, and bottom is the combined two):

 

[pkane2001]

3 minutes ago, psjug said:

Audio DiffMaker is able to get better alignment than doing it the way you did.  See Section 5.2 in their paper:

http://www.libinst.com/AES Audio Differencing Paper.pdf

 

I’m sure it can, but I didn’t have it or know how to use it.  I did see perfect matching between peaks and valleys in a bunch of  areas where I spot checked it. I zoomed in so that individual samples were visible.

[pkane2001]

4 hours ago, PeterSt said:

 

Huh ?

 

If an FFT shows differences but you can't perceive that, then you are deaf or your system is not on par.

Aparenty you think the other way around, somehow.

 

If an FFT does  not show differences, the sound can still be vastly different.

(I'd say that everybody knows this and it is a sad thing it is so)

 

The FFT shows some differences. I posted this yesterday. The differences rise close to an audible dB level, so I'm not sure why we keep saying there are no differences in the two captures.

 

Today, I used diffmaker to match up the A and B analog tracks and compute the difference. A similar set of frequencies appear in the delta (with differences), although with diffmaker the level is a bit higher than in my manual differencing process. Not sure why that is, I have to assume diffmaker is better at matching up using correlation technique.

 

Here's my manual spectrum of differences:

15 hours ago, pkane2001 said:

 

And here's one produced by diffmaker:

 

 

[pkane2001]

2 minutes ago, manisandher said:

 

Because:

 

 

And:

 

 

Mani.

 

And this explains the -74dB signal at around 400-500Hz left over after subtracting A and B?

 

[pkane2001]

5 minutes ago, manisandher said:

 

There isn't a -74dB signal at around 400-500Hz, according to Testikoff's analysis:

 

Mani.

 

I see. I found these frequencies doing the calculation two different ways, but who knows, maybe I did something wrong. I'll go back and recheck all the settings in diffmaker, although I got a similar result without it.

 

[pkane2001]

1 minute ago, manisandher said:

 

Testikoff was pretty thorough:

 

 

Mani.

 

That's exactly what I did manually. Then, I let diffmaker do the alignment. Both showed some left over frequencies in the delta. Diffmaker is a bit of a pain, some settings cause it to crash. I'm trying to find some settings that produce the best result without crashing. Not easy

 

[pkane2001]

5 minutes ago, adamdea said:

I don't think it is the same as what you did. 

 

Maybe I misunderstood what he did, then. Here's what I did (manually):

 

In Audacity, I tried to match the phase of the signals as close as possible. Zoomed in down to individual sample view, I made sure the peaks/valleys aligned on the same sample number in both captures. This was mostly already well aligned in the captures, but a bit of deleting of samples was needed to make the match better.

 

Then, I inverted one and added them together to produce the difference. Then, plotted the spectrum.

[pkane2001]

1 minute ago, adamdea said:

ie he took the [magnitude only?] fourier transform  of each and then subtracted to yeild a [magnitude only] fourier transform.

 

Ah, that makes sense. I did something similar early on and found no visible differences in the spectra, which is why I went with matching up samples in the time domain and then subtracting the two signals and plotting the spectrum of the difference.

 

[pkane2001]

2 hours ago, adamdea said:

No surprises. No two analogue files null perfectly. Say they null to -90dB. Where does that take you until such time as some evidence emerges about there being an audible difference between the files? I have lost the thread as to whether there is any evidence as to whether the As and Bs are more like other As and Bs respectively than they are to each other.(either with test tones or music)

 

Here's the spectrum of the difference of two A captures (3 and 7):

 

Here's the spectrum of the difference of two B captures (4 and 6):

 

Note the absence of the extra frequencies that were there when examining A3-B4 residuals (posted earlier):

 

[pkane2001]

1 minute ago, adamdea said:

That looks much better. This is after time aligning? When I compare 3 and 7 without any realignment I get the usual spectrum but peaking at about -101dB. for 5 and 7 much worse (-83 or so).

 

I've done it both ways, manual alignment, and align using diffmaker. I'm not sure about drift compensation (that was not turned on in diffmaker) as that seems to attempt to adjust phase of the second file to match that of the reference. If there are phase differences, I'd like to know about them rather than to try to adjust them out

 

 

[pkane2001]

1 hour ago, manisandher said:

 

Hey, nice!

 

It looks like the differences are measurable in the analogue captures after all!

 

Well done!

 

Mani.

 

By the way... your capture 8-B appears to be an A, instead, if my method is working correctly.

 

When compared to the other two B captures, it shows the same residue as I've posted in the A-B plot. But when compared to other A's (including 8A) it shows no residue at all. 

 

[pkane2001]

1 minute ago, manisandher said:

 

Unfortunately, '8. analogue capture _ B' is definitely a 'B', and not an 'A'. Mans will confirm that we recorded all the 'B's after the 'A's, and here are the details:

 

 

Here, '200' denotes that it's a 'B'.

 

And here are the details for file 8:

 

 

It's definitely a 'B'.

 

How does this now affect your analysis?

 

Mani.

 

If correct, this invalidates it, although it's very strange how perfectly 8B nulls out all other A captures, but is different from all the other B captures.