Close in phase noise

[jkeny]

There's a spin-off thread started on ASR after Mivera's posts about it on another thread which was closed down & I wished to give my views on it here (as I'm banned from ASR) & possibly advance the topic beyond the blinkered mindset on display on ASR.

First off this is not covered in any AES paper so is anathema to the mindset on ASR

The question is why is close-in phase noise perceptually noticeable? I have tested this myself & can concur with the perceptions of Mivera & others who have heard such low phase noise clocks, in well implemented systems - it results in more clarity to individual sounds & more 3D soundstage. When power supplies are low noise & stable it gives results of the same type. When USB isolation is properly done it gives more of the same results. These changes are additive, not mutually exclusive.

So my logic leads me to consider if there are similar underlying mechanisms at play which might be resulting in this almost exactly similar audible improvements?

What could this be?

Well some clues are given by none other than Amir, although he isn't even aware of it - "Vast majority of close-in jitter is random and hence manifests itself as simple increase of noise floor."
The problem here is that he doesn't think through what he's saying - he's looking at a single tone & close-in phase noise. So what happens with modulating signals (i.e. music) - this would result in modulating noise floor in between the the signal fundamentals & in between the signal harmonics i.e. all the spikes that would be seen on an FFT of real music would have the noise floor modulating as the signals continually changed.

So when someone says it's random, they really don't understand what the term random means & dismissing it based on a lack of understanding.

Another major gaff of his from the same post - "It is also massively masked as I post in the other thread.". Is noise floor "masked" by signal? That's the first time I ever heard this claim. Not to mention modulating noise floor.

Perceptually noise is perceived differently to signal as it is a broadband signal which spans across the frequency bands into which the hearing mechanism divides the signal. When the energy in a waveform crosses these frequency bands (called ERBs or Equivalent Rectangular Bandwidth) it has repercussions in perception. Just as a simple illustration - there is a just noticeable difference graph for noise which is very different to the JNDs seen on the Fletcher-Munson plots for signals. The plot for noise perception comes originally from the BBC research dept & differs by 12dB in places from F-M curves & our max sensitivity is at different frequencies

So rather than trotting out hackneyed words such as "random" & "masking" without any thought, I wanted to open up a real discussion here

First topic - how to correctly interpret the FFTs for close in phase noise. A perfect signal on an FFT would be a razor thing signal line with no width to it i.e the signal is always at the exactly correct frequency. If the FFT of a signal with close-in phase noise is zoomed into, the signal plot has width to the line i.e it is not a single frequency but spreads out both sides of the fundamental frequency. Sometimes the signal is +/- 0.1Hz away from what it should be, sometimes +/- 0.5Hz, sometimes 1Hz & so on. The drop off in the plotted slope as it moves away from the central peak is an indication of how often the signal is found at the wrong frequency - it drops off the further away from the peak is plotted.

So what does this FFT tell us - does it tell us that the phase noise is jittering the frequency at almost the same level as the main signal? What is the perceptual result of such jittering of frequency? Again one has to think about this carefully - if you play a 11KHz tone together with an almost equally loud 11.1Khz tone, is it masked? Probably in a listening test. Now if we combine a number of 11KHz signals paired with various other tones of almost the same loudness but slipped in frequencies - some +1Hz, some -1Hz, some at other frequencies - will this be perceived differently to a single pure 11KHz tone? Now do this with real music - take a copy of the waveform, reduce the amplitude slightly & shift some frequencies by 0.1Hz, some by 0.5Hz, some by 1hz, etc. Play both waveforms back on a device which hasn't got internal jitter which drowns out any possible audible effect & check for audibility

 

[jkeny]

Just to elaborate some more on what I said about the psychoacoustics of what's going on with close-in phase noise.

We have to look beyond the first order effects that Amir & many objectivists use in their analysis.

Two considerations about this - the 'noise floor' seen in an FFT is not the actual noise floor - it is always plotted lower than the actual noise floor because of the method FFT uses (not oversampling, btw). Yet the signal spikes are correctly representative of the actual signal level. Why this is? An FFT splits the waveform into a series of frequency bands or bins (under the control of the user running the FFT) & the it then plots the energy of each frequency bin into which it has analysed the waveform. So for narrow band signals, which cross only a couple of such frequency bins, the energy in those bins will be pretty close to the actual energy of the tone as measured by other means. However, when a broadband signal is being analysed, it's energy is analysed across many frequency bins. So when we look at any FFT plot with a broadband signal on it we need to consider this aspect of FFTs - it's called the process gain & can be calculated & the broadband's actual level adjusted. We can't just read off the plot of such broadband signals & claim that the level is what we read, there is some adjustment for the process gain aspect of distributing the energy in the broadband signal over many bins.

So these 'noise skirts' that we see at the base of the signal spike are not arising from a -160dB noise floor, they need to be adjusted somewhat for the FFT process gain due to their broadband nature but I haven't seen this treatment done anywhere. So we can't just read off the plot & say the noise of these skirts is at -160dB rising to -120dB as we get closer to the signal spike & therefore it is not audible - I don't know what the adjusted figures taking into account process gain would be. If we zoom into the signal spike we will also see this 'noise skirt slope' going up far higher in DB as it nears the signal spike.

This is an example of such a zoomed in FFT - the first image is the usual view we see & the second image is the same FFT but zoomed into the blue boxed section




The second zoomed in FFT clearly shows the skirt in red now reaching about -80dB before it's plot merges with the signal spike plot. This is with an x-axis at 20Hz divisions - at even smaller divisions of say 1Hz we would see how high the 'noise skirt' reaches before meeting the signal spike plot. We don't often see such close grained FFTs.

So when people state that the skirt noise is -120dB down they really are just being lazy in their thinking (sciencey) or just being disingenuous.

I've seen people (Amir et al) claim that masking will cancel such tones. I find this sciencey too & often stated by Ethan Winer. The same type of simplistic, first order viewpoint is seen when auditory thresholds are used to dismiss what is perceptible when complex music signals & auditory perception come into the picture.

The assumption is that full cancellation occurs whereas any bit of knowledge about this will stat ethat roughness & beating of tones will occur when two sounds are withing the same critical band. Critical bands describe how our physical hearing mechanism seems to divide up the audio spectrum into frequency bands - something similar to FFTs but we have overlaps between these bands whereas FFTs do not. Masking only applies to the whether we can perceive two tones as separate entities. How we perceive the roughness of a tone is to do with complex tones, not simplistic pure tones - see here "Roughness is physiologically determined and therefore universal" And roughness is the result of our inability to clearly tell apart two tones which are in the same critical band.

Could this be why close-in phase noise in audio clocks is audibly detrimental to the sound resulting in reports of better clarity & sound stage when such phase noise is reduced?

As I maintained before - the claims that AMir & others make concerning support for their arguments from science & pyschoacoustics are often exposed as pseudo-science, with a little bit of examination, logic & knowledge.

 

[amirm]

We are continuing our discussion on ASR Forum. I won't duplicate that here as the discussion becomes all personal and obnoxious. Anyone interested in my views can read it here: http://www.audiosciencereview.com/forum/index.php?threads/close-in-jitter.1621/. And this specific post I made on listening tests performed and psychoacoustics: http://www.audiosciencereview.com/forum/index.php?threads/close-in-jitter.1621/page-8#post-40804

Did you ever run Arny's 30 Hz jitter tests?

 

[Robh3606]

And roughness is the result of our inability to clearly tell apart two tones which are in the same critical band.

Hello jkeny


Ok but you don't have 2 tones. You have phase noise and a tone. The phase noise is random and changes constantly. There is nothing for the signal to beat against

Rob

 

[jkeny]

Hello jkeny


Ok but you don't have 2 tones. You have phase noise and a tone. The phase noise is random and changes constantly. There is nothing for the signal to beat against

Rob

Maybe you misunderstand what clock phase noise means? It doesn't just produce random noise as Amir maintains

I believe that each sample being processed is randomly mistimed because of the clock's close-in phase noise. All plots of close-in phase show that these timing fluctuations are more numerous as we get to lower offsets from the carrier frequency (closer to the carrier frequency) - in other words it's characteristic is given by 1/f so the clock mistiming is not purely random it has a distribution characteristic.

If we take an example for illustration - the reconstruction filter generates one period of a 20KHZ sine wave from two samples - if both samples are shifted slightly (due to close-in phase noise) the reconstruction actually produces a slightly different period waveform than 20KHz - now look at the next two samples & these will be offset differently this producing another different period waveform than 20KHz & so on for each period in the 20KHz tone producing a waveform with periodicity fluctuating around 20KHz

This fluctuating 20KHz tone is what is perceptually heard as roughness.

Now extrapolate to all the tones & harmonics in complex music & we get the sort of blurring that is reported as being changed into clarity to the sound when a clock with lower close-in phase noise is used.

 

[jkeny]

We are continuing our discussion on ASR Forum. I won't duplicate that here as the discussion becomes all personal and obnoxious. Anyone interested in my views can read it here: http://www.audiosciencereview.com/forum/index.php?threads/close-in-jitter.1621/. And this specific post I made on listening tests performed and psychoacoustics: http://www.audiosciencereview.com/forum/index.php?threads/close-in-jitter.1621/page-8#post-40804

Did you ever run Arny's 30 Hz jitter tests?

How is the jitter in these files generated - is it random jitter, signal correlated jitter, what is being presented here?

 

[amirm]

Hello jkeny


Ok but you don't have 2 tones. You have phase noise and a tone. The phase noise is random and changes constantly. There is nothing for the signal to beat against

Rob

That's not his only crime. The bigger one is that he is assuming near 0 db tone at high frequencies (he has chopped off the top of the scale in his graphs). No music has any such content. Not remotely so. Typical values are 50-60 db lower. Since jitter directly scales with level of music, then you have to lower the jitter levels by proportionally the same level. That smashes those jitter noise levels into oblivion.

 

[amirm]

How is the jitter in these files generated - is it random jitter, signal correlated jitter, what is being presented here?

That was in my message you quoted. So the answer is no, you have not listened. Yet want to give lessons on audibility?

 

[jkeny]

That's not his only crime. The bigger one is that he is assuming near 0 db tone at high frequencies (he has chopped off the top of the scale in his graphs). No music has any such content. Not remotely so. Typical values are 50-60 db lower.

And what do you know about the roughness perceptibility of close-in tones - what amplitude level & frequency delta is JND?

Since jitter directly scales with level of music, then you have to lower the jitter levels by proportionally the same level.

Yes, jitter level is correlated to signal level, so what's your point - oh it's in the next false claim

That smashes those jitter noise levels into oblivion.

There's absolutely no correlation between what you have said before this claim & this claim - it's simply snatched out of the alternate universe of sciencey thinking - the ASR universe, it seems.

 

[amirm]

And what do you know about the roughness perceptibility of close-in tones - what amplitude level & frequency delta is JND?

Covered in the link I provided to ASR Forum post.

Yes, jitter level is correlated to signal level, so what's your point - oh it's in the next false claim There's absolutely no correlation between what you have said before this claim & this claim - it's simply snatched out of the alternate universe of sciencey thinking - the ASR universe, it seems.

This is the obnoxiousness I was talking about. It is like on autopilot. Can't discuss audio for two seconds before resorting to such language.

 

[jkeny]

That was in my message you quoted. So the answer is no, you have not listened. Yet want to give lessons on audibility?

Again what alternate universe are you occupying? I see no answer to my question in your post which I quote (in full unlike you often do)

"We are continuing our discussion on ASR Forum. I won't duplicate that here as the discussion becomes all personal and obnoxious. Anyone interested in my views can read it here: http://www.audiosciencereview.com/fo...n-jitter.1621/. And this specific post I made on listening tests performed and psychoacoustics: http://www.audiosciencereview.com/fo...e-8#post-40804

Did you ever run Arny's 30 Hz jitter tests?"​

So please show me where in your quoted post is the answer to the question I asked "How is the jitter in these files generated - is it random jitter, signal correlated jitter, what is being presented here?"

 

[jkeny]

Covered in the link I provided to ASR Forum post.

This is the obnoxiousness I was talking about. It is like on autopilot. Can't discuss audio for two seconds before resorting to such language.

The obnoxiousness starts with your "That's not his only crime." & your false claim "That smashes those jitter noise levels into oblivion."

Of course I call such outrageous claims as I find it - when you make a claim that bears no relationship to the logic or statements preceding it, I conclude that you are operating according to a different logic that can only arise from a different universe or else you are just being intellectually dishonest & know exactly what you are doing.

The "language" you refer to is this - me calling your outrageous claim from "the alternate universe of sciencey thinking - the ASR universe, it seems."

 

[jkeny]

One other point of conduct, Amir - when you come here with a test & don't give the details of what the test files contains is not appropriate behaviour & i would suggest that you don;t know the answer to my question or are unwilling to provide the answer here.

By linking to your ASR post in answer to my question (which I had already checked before asking you my questions & the answer wasn't there) - it is very obvious you just want to drive traffic to the ASR site (something you demonstrate you are desperate to do at every opportunity) - I don't know what the owners of WBF feel about this behaviour of yours but it would not be tolerated on other forums.

Not only are you being dishonest again (the answers are not on ASR) but your are deflecting & attempting to get people to look into ASR.

Edit: I believe you were permanently banned from AVS for exactly the same behaviour as you are engaging in here - linking to WBF posts on the AVS forum - so why do you continue this behaviour here & think it is acceptable?

 

[jkeny]

So let's just take stock, at this point
I have asked Amir a number of questions - all of which have been deflected & left unanswered
- And what do you know about the roughness perceptibility of close-in tones - what amplitude level & frequency delta is JND?
- How is the jitter in these files generated - is it random jitter, signal correlated jitter, what is being presented here?

Maybe you can show some substance to your claims, made often here & elsewhere, that you have an understanding of psychacoustics?
But your claim that I was wrong about AL's ABX test & soundandmotion correcting you in your misunderstanding (which you yet again avoided answering), show that your understanding is sciencey & intellectually dishonest.

Maybe you might this time, not avoid answering the above questions?

 

[jkeny]

Amir, here's an example of how the preparation of jitter files is described & hopefully you know how the jitter in the files you tried to "test" me with was generated.An answer to my question (rather than a link to a non-existent answer on your seemingly failing forum)

Oh I don't intend to drive traffic to the head-fi forum (they are not desperate for traffic ) - I have no ownership or financial connection to it - I just provie the link fo rfull disclosure & to show that I'm not misquoting or taking out of context what I'm quoting below or engaging in such tactics seen here from you, often

"It is basically a variable delay at 768 kHz sample rate. The signal used to modulate the delay time is a mix of several sine waves (to create sidebands) and lowpass filtered white noise. The noise has uniform distribution before the filtering, and the filter is a -6 dB/octave lowpass with a -3 dB frequency of 4 Hz followed by a -12 dB/octave Butterworth lowpass with a -3 dB frequency of 60000 Hz; the latter one is there mainly for anti-aliasing purposes. The noise generator is seeded from the system time. Also, the noise has some stereo separation (it is a mix of a mono and stereo component with a ~85:15% ratio of power). This might not actually be a good idea or an accurate simulation of real hardware (especially for low frequency noise), and was basically left in the code from an earlier version. Although the resulting ITD should in theory still be well under the threshold of audibility, in a new test, the "stereo noise" should probably be removed, or limited to high frequency noise only.

I doubt the seeding would be of too much significance in practice, except for very low frequency "wow/flutter" components."​

I'm showing this to you so that you know one of the tenets of science - repeatability. The above description gives a reader the ability to repeat the test - one of the fundamental tenets of the scientific approach.

Coming to this thread & suggesting to use files for "testing" which you can't describe how they were generated is typically sciencey & shows a very superficial understanding of what science & scientific tests are about

 

[jkeny]

Back to the normal programs as it seems Elvis has left the building

It has been said that the Jtest is not a suitable test for examining jitter in an environment which doesn't involve SPDIF or AES. The Jtest was developed by Julian Dunn to examine a particular weakness in this protocol which involves Inter Symbol interference in SPDIF (& AES) receiver chips. This weakness has long ago been attended to - mostly because of this test focusing attention on the weakness (& the weakness being identified in the first place by Dunn). Now why has this Jtest remained as the most used test for jitter? Mostly due to laziness. Is it the sensitive enough for jitter testing on modern digital audio devices, particularly USB connected digital audio devices? No, it's not sensitive or specific enough for these use cases & here's an example of why & how it's not suitable.

Let me introduce you to another way of looking at mistiming in digital samples & how the jitter test & FFT analysis is unsuitable. It may also have a bearing on close-in phase noise

I introduced this community to this test here in 2013 - it's a test from Jim Lesurf called IQ-test. Oh & btw, the measurements are taken on the analogue outs of the DAC - the most difficult place to examine such very small timing deviations in the signal.

I'll try to present it briefly here but the full description can be read in my link
Here's a measurement, using this test method, of a Cambridge Audio DACMagic (in red) receiving audio digital samples via USB & the Halide bridge (in blue) USB to SPDIF converter feeding the same DACMagic but this time via SPDIF

What we see in this graph is that DACmagic receiving its digital audio signals via the USB signal is not stable - it does not provide a stable replay on the analogue outs - jumping in replay rate every 25 seconds or so - seen in the red plot. The DACMagic replay in blue when receiving its digital audio signals via SPDIF, is stable in replay rate as measured at the analogue outs.

Stereophile (& users) has this to say about using the USB channel for audio replay "Although its USB input is really of only utility quality and shouldn't be used for serious listening, the Cambridge Azur DacMagic otherwise offers superb measured performance"

Now in that link the Jtest FFT for the DacMagic is shown & it says this "The DacMagic obviously features superb jitter rejection via its conventional data inputs. The USB input, however, performed significantly worse on this test, with both a raised noise floor and significant sidebands apparent (fig.12)."


"Fig.12 Cambridge DacMagic, high-resolution jitter spectrum of analog output signal, 11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 16-bit USB data. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz (left channel blue, right red)."

So what we see in this graph is an increase in 'noise floor', a slight widening of the base (skirt) of the main jitter signal of 11.025KHz & some sideband spurs, none of which are above -104dB or thereabouts
Now in terms of the usual arguments used about audibility - would this FFT (showing spurs n greater than -104dB down) be judged as showing "USB input is really of only utility quality and shouldn't be used for serious listening"? What exactly is audible based on this FFT graph?

Or an alternative viewpoint may exist that if we did have a zoomed in FFT close to the main signal spur would we see a vast difference in the width & height of the skirt at the base of the signal & would this be something worth investigating further as an audible artifact?

The other aspect to this is that FFTs which are regularly cited & claimed as showing that there is nothing of audible consequence to be seen, need to be queried/examined further.

 

[cooljazz]

The "language" you refer to is this - me calling your outrageous claim from "the alternate universe of sciencey thinking - the ASR universe, it seems."

I've taken to just plainly calling this "internet science". It's just a new facet of today's world.

CJ

 

[jkeny]

I've taken to just plainly calling this "internet science". It's just a new facet of today's world.

CJ

The world of alternate facts.
The world of intellectual dishonesty.
The world of deflection & doubling down instead of admitting a mistake.

It's scary how far intellectual honesty & reality have been left behind & I'm not talking about silly issues that exercise our interests on audio forums

 

[marty]

Oh great. Just what we need, another pissing contest.

Gentlemen, with all due respect, few of us here know what the hell you are talking about. This is a very esoteric conversation. At the very least, can you please dumb it down for the rest of us?

 

[BruceD]

Oh great. Just what we need, another pissing contest.

Gentlemen, with all due respect, few of us here know what the hell you are talking about. This is a very esoteric conversation. At the very least, can you please dumb it down for the rest of us?

+1

Bruce