Fluctuating USB jitter from CPU activity

[jkeny]

OK, got it - I've edited my original post

And here's JohnW's explanation of the plots

Originally Posted by adamdea View Post
John -thanks this is very interesting.
I'm not 100% sure what I'm looking at, could you help me to understand- is this in effect the plot of many many cycles of usb signal (voltage against time) overlaid ?

The process is called colour Grading - so from Black (no Waveform hits) to Red most Waveform "hit" areas.

If the signal was perfect with no modulation then each waveform would overlay the previous and next waveforms and displayed as purely Red with no fluctuation from one cycle to the next (no other traces or colour) - the background would be Black.

"A digital phosphor oscilloscope (DPO) uses color information to convey information about a signal. It may, for example, display infrequent signal data in blue to make it stand out."

Originally Posted by adamdea View Post
If it is then I can see that the main signal (high crossing to low and back again forming an x) is much sharper in the second case, showing much less fluctuation from one cycle to the next.

Also I see the difference in the background colour which I understand forms some sort of spuria. I also see that there are 3 green horizontal lines on the non-regen version - what are these (are they the rest state of the signal between data?)

The scope timebase was 200pS a Div, so very fast any LF modulation causes the location of the eye pattern to shift in time - so the location of the crossing points fell into other areas which results these hits being colour graded - Lower Frequency events in Blue, more frequent events in light blue etc.

The Green graded hits indicate "medium frequency events" - so these "odd states" happen frequently - Observing the Waveforms on a very fast Analogue scope "WITH IMAGE INTENSIFIER" to allow LF events to be visible on the CRT at such fast sweep speeds (500pS Div) you can see the "runt" events directly related to the processing tasks of the PC - the frequency of their occurrence is visibly modulated by the operating system workload / operation.

Originally Posted by adamdea View Post
I'm trying to understand how we should interpret this information. I suppose it's difficult to be too prescriptive here because it's not clear what the transmission mechanism is for whatever undesirable results we are trying to eliminate

For sure any signal with unwanted modulation is not a good thing to have "flying around" a DAC PCB - these modulation effects will be very visible with an RF spectrum analyser looking at the Audio outputs - how the Audio system copes with this is anyone guess, and this is not the only mechanism this unwanted modulation can cross into the Analogue domain.

Originally Posted by adamdea View Post
1) the signal shape itself on the basis that it somehow directly affects the dac however it is read (as in your example of the jitter on the signal which somehow radiates within the dac even when the data is fed into a buffer via (in my words not yours) the glitch caused each time data is read into a buffer)

With any signal you need a return current path, as there is no such thing as true zero ohms ESPECIALLY at high frequency (Think GHz here with HS USB) the signal currents can be found almost everywhere on the PCB and internal to the silicon. So know you have unknown isolation attenuation between domains with such a signal, then its for sure a good thing to reduce any unwanted modulation of this signal.

Originally Posted by adamdea View Post
2) the signal shape insofar as it causes the usb receiver to have to try hard to read the data (this is what I understand to be JS' hypothesis)

No, I totally don't agree with JS's hypothesis, I think John misses the main point about the unwanted modulation encoded within the USB signal - as the Colour graded plots clearly show.

Originally Posted by adamdea View Post
3 the noise in the signal (not the 5v line) which somehow pollutes the dac through some other mechanism
[have I got this right- are there other mechanisms I have missed]

Yes, unrelated RF noise on the Data lines also needs to be attenuated which requires RF filtering - a separate function to the data repackaging / timing.

Originally Posted by adamdea View Post
4) the noise on the 5v line (not not relevant to this plot I assume)

Not directly related, but any noise on the 5V line (and this Ground) will couple into the Data lines, so should be avoided.

Originally Posted by adamdea View Post
Having regard to those considerations (particularly the first and second). I'm trying to get my mind round the question of what it is about the signal that really matters. Am I right in thinking that because usb is a differential signal the receiver would automatically ignore the common mode noise, as it would be reading the difference signal?

What do you think the Common Mode rejection performance will be at 1GHz? even my US$5K Tek test probe is only around 20dB CMR at 1 GHz, I can assure you that the USB Cable / interface will be less then this!

Originally Posted by adamdea View Post
Apologies for the rambling, I'm just trying to work out what this all means.

Yes, I seem to have awoken some deep seated concerns - but atleast I have been ale to measure quite a significant modulation of the USB Data related to the CPU / System processing which opens the possibility that the "Bodge fix" attempts of optimised software players might have a foundation in "reality" - although they are a very wrong / incomplete band aid to a problem that should not be allowed to cross into or even effect the analogue domain.

[Sligolad]

http://tinyurl.com/ns2rvzg

http://tinyurl.com/qjg5rvl

Looks like John just get there before me!

[Sligolad]

Glad you have an inbuilt immunity to all the dung that gets slung over on PFM John and thanks for posting the relevant bits as I do not have the patience to tolerate the closed minds there.

Interesting that someone as apparently capable as John Westlake comes out with the following:

I'm not in anyway endorsing the idea behind "optimized" media players (as the are trying to resolve an issue that should not effect the analogue domain in the first instance - IMO they are an indirect poor and dubious bodge fix)

Surely he understands that optimized medial players are working towards unloading the hardware and processes that he rightly identifies as contributing to all the mess he is seeing on the USB interface!!

OK so it turns out he is not so against optimized media players after all.....

but atleast I have been ale to measure quite a significant modulation of the USB Data related to the CPU / System processing which opens the possibility that the "Bodge fix" attempts of optimised software players might have a foundation in "reality"

[jkeny]

Glad you have an inbuilt immunity to all the dung that gets slung over on PFM John and thanks for posting the relevant bits as I do not have the patience to tolerate the closed minds there.

Interesting that someone as apparently capable as John Westlake comes out with the following:

I'm not in anyway endorsing the idea behind "optimized" media players (as the are trying to resolve an issue that should not effect the analogue domain in the first instance - IMO they are an indirect poor and dubious bodge fix)

Surely he understands that optimized medial players are working towards unloading the hardware and processes that he rightly identifies as contributing to all the mess he is seeing on the USB interface!!

OK so it turns out he is not so against optimized media players after all.....

but atleast I have been ale to measure quite a significant modulation of the USB Data related to the CPU / System processing which opens the possibility that the "Bodge fix" attempts of optimised software players might have a foundation in "reality"

Sorry, Pearse I missed your first post on the other page - ah, sometimes I like to get down & dirty & the PFM sewer is the place to do it - life can't be all sweetness & light :)

Yes, he is a bit ambivalent but he is only beginning his investigation. When he has looked into all the issues & worked on his own solution I suspect he may have a more rounded viewpoint about what is possible & what isn't (or maybe not, who knows).

I don't really care what attitude he has as long as he remains open to doing the necessary measurements & presenting them - I even subscribed the crowdfund €50 to him to get this investigation rolling.

I'm interested in the whats & wherefores of the issues!

The "runt" signals, as he calls them (the ones that are so far off the line that they cause the bluish background colour) do remind me of some IQ-test measurements I did on MQN & presented on this forum a year or so ago. His "runt" signals are the result of low frequency "stuff" coming from the PC & correlated to its activity.

I vaguely remember presenting some IQ-test plots that showed timing differences in LF areas of the IQ-test plots but I ignored them & focussed on HF parts of the plots as it was difficult to correlate these low frequency differences in parts per billion with audibility. A year later & I have different perspective on this based on my further research into LF noise & audibility & based on these USB eye patterns that are emerging.

Fran, can you do a rebuild of the forum index as I can't find this in a search?

[Sligolad]

Hi John, you certainly got my attention on this and you even coaxed me to go over and look at the PFM thread, and surprisingly I do find it very mature compared with some of the madness that went on with these discussions in the past. Looks it is a slow tide but it does appear to be slowly turning!!

Now we have a trinity of John's leading the way :-))

[jkeny]

I found the original thread & in this post is where I start to talk about the low frequency timing issues that I see on the IQ-test plots viewtopic.php?p=49420#p49420

Just to return to perhaps spur someone along this path, here's something that caught my interest:
I saw this post on DIYAudio about measuring VLF noise http://www.diyaudio.com/forums/lounge/2 ... ost3702082
Quote:
"Hmmm, I have found an interesting way to change the perceived noise, though I am having troubles measuring it.
ARTA running averaging of 2000 or so trials is showing up differences in VLF noise, interestingly.
Subjectively the difference is sort of like reduced jitter/imd, with deeper solider bass and clearer mid/highs.
L/R and depth imaging markedly improved, overall more realistic sound but no FR changes measured. "


VLF (Very Low Frequency) noise is an area I'm interested in & have been following someone else's experiments in this area. His subjective description of the audible changes he hears correlated with the other experimenter & rang some bells with me & with the collective subjective impressions of MQN & also of PS improvement in PCs described in this forum, I believe.

I'm looking at this in light of the IQtest results I have done & Jim LeSurf's published measurements (remember LeSurf called this Wow & flutter but I think it's better termed Wander) - all of which show differences in VLF (<5Hz), particularly below 1Hz. What these graphs show is that there is a drift in timing (of some 10s of picoseconds, yes picoseconds - if it was differences of microseconds, there would be no question of it's audibility) which span timeframes of 1sec. In other words, measuring timing from second to second, there are differences of timing of some 10s of picoseconds. In real music playback there could be a wander (or phase shift?) in the fades of piano reverb tails. What I would like to establish is how audible resulting phase shifts from this level of timing differences (10s of picoseconds) could be? I see 0.1% of flutter cited above as a small amount which makes me doubt the picoseconds - but I don't give up that easily :) Anybody know of a method of applying wander (low frequency) to actual recordings down to the picosecond level in order to establish it's audibility or otherwise? I know this timing is orders of magnitude off what seems to be audible but I was just like to satisfy myself that there aren't some secondary effects resulting from these picosecond fluctuations that are audible i.e the effects of room reflections

Again, this may be grasping as straws :) to find differences but I can see a correlation (the lower VLF measurements map to better SQ) here between the recordings made direct from laptop out Vs via USB to Ciunas DAC. And again I can see the same SQ correlation between measurements made using Foobar Vs MQN.

I also wonder if maybe the differences in these measurements are indirectly signifying an underlying difference in low frequency noise?

It's only with a larger body of such correlating measurements would it signify a worthwhile avenue of investigation to follow!

Come on, guys! Have a go!

[jkeny]

And a following post

More from DIYAudio http://www.diyaudio.com/forums/lounge/1 ... ost3689373
​​

1/f Noise Is The Devil....
Reducing very low frequency system noise (<< 20 Hz) by only a couple or few dB reduces a layer of intermod that intrudes over everything in the band of interest (20-20k).

This effective system intermod behaviour of 1/f noise is also spectrally important..ie the nature of the VLF noise imparts a signature/character to the noise/distortion floor.

IME this is a big arbiter of perceived noise/distortion floor, and further affects/effects all standard descriptions of sound quality...ie overall clarity of sound, L/R separation, depth perception etc.....focus, rhythm, timing etc.
When very low frequency noise reduction is applied to both channels coherently, the perceived noise drops further, thus enabling further improved focus, low level detail, and depth information retrieval.

System intermod behaviour also increases/enables/causes audibilty of low frequency noises, so any reduction/change in VLF system noise would serve to reduce/change 'groove' noise.

I expect the Bybee devices are changing VLF noise level/behaviour/spectrum.....not by a lot, but enough to cause an audible difference.

I've been following this guy's posts on this subject & his latest one is here http://www.diyaudio.com/forums/lounge/1 ... ost4178372:

Ok, I just did a quick and dirty experiment in Cool Edit.
I generated a pink noise signal and then low pass filtered it with 15Hz 4th order cutoff freq.
I then mixed this at relatively low level signal with a known music track and listened on my B mini system out here in the sunroom.

The result is akin to what I get with BQP or my other filter on my A system....ie adding the low freq noise signal to the music signal is analogous to removing BQP or my other filter on my A system.

The noise signal is not really directly audible from my listening position across the sunroom, BUT adding this noise signal subjectively reduces the overall perceived sound level and dynamics, reduces the low bass and messes mids and highs....this effecting/affecting (inversely) correlates with experiences on my A system.

Now, in this case I am dealing with first order effects, ie mixing/adding and I am not really adding modulation/secondary effects such as dynamic THD, IMD or PMD (jitter like) effects.

This all ties in with my experiences to date with BQP and my filter....reduce conductor noise and a whole cloud of secondary effects shite is exponentially reduced.
Subjectively the result is much more realistic dynamics, bass that extends down to the floor, mids clearer/cleaner and highs extended.
By treating (noise reduction) both signal channels together plus power reduces L/R dynamic errors with result of pin sharp lateral imaging and pin sharp depth imaging....as is to be expected.

Excess 1/f noise production is dynamic in nature in a typical circuit using typical componentry...the typical result is the subjective sorry mess as per most listeners experience.

Utilising very low noise componentry goes a long way to providing clean audio reproduction, reducing conductor LF noise takes things to the next level.

Notably Jfets exhibit very low 1/f current noise.....iow, JC ain't stupid.

BQP is short for Bybees, I believe!

[jkeny]

Hi John, you certainly got my attention on this and you even coaxed me to go over and look at the PFM thread, and surprisingly I do find it very mature compared with some of the madness that went on with these discussions in the past. Looks it is a slow tide but it does appear to be slowly turning!!

Now we have a trinity of John's leading the way :-))

Yes, Pearse but that maturity is only skin-deep - it happens because JohnW is a very well respected audio engineer & the usual asshole posters curb their language & their attitude when posting on his threads. On the other hand when answering my posts they ............... :)
So I wouldn't predict that the tide is turning, just that the troops are in a huddle, at the moment, discussing how they will counteract this act of sabotage - presenting measurements which undermine their whole world-view! But you can already see the seeds of what will become their final position - "there is no definitive measurement on the DAC's analogue outs which proves that anything is audible"

[nige2000]

good to see a bit more talk on the subject all the same

funny that one of the first things to do on a pc for audio was to stop cpu throttling and set the cpu frequency
most guys seemed to prefer underclocking but it seems to be a consensus that it needs to be a constant

one thing i always noticed whether hardware, software or power supply improvements their benefit always seemed to be accumulative
or one improvement never seemed to remove the benefit of a previous improvement it was always additional

for that reason id prefer to see devices been designed from the ground up to read files and output a clean usb/i2s signal for its own or external dac rather than try to fix a poor signal from pc etc
the difference between a standard pc and highly modded pc for audio is rather large so given the cost of pc modding surely theres a market for such devices

[jkeny]

As I said - just look at BE718's posts to me on the MDac thread for evidence of how skin deep this maturity actually is.
Admittedly, I did query him on his system when he reported hearing no real difference with the Regen except for a slight reduction in sibilance which he took to be an attack on his listening abilities - it was the equivalent of calling him "delusional" in what he didn't hear :)

Just to go back to something more profitable to discuss - the IQ-tests were used to test a Cambridge Audio DACmagic, a Halide Bridge USB/SPDIF converter & an Arcam R-Dac. In all cases, have a look at the LF side of the plots

First the R-Dac (I think this is the original one which didn't sound great when I tested it - just OK like most run-of-the mill DACs). And here's what Jim LeSurf says about the LF end "Looking at the plot you can see that there is a particular component at about 0·1 Hz. This means that the replay rate is smoothly hunting up and down every ten seconds or so in a regular manner."



"Considered as timing ‘jitter’ these results can be viewed in a different way. The above graph shows the same 44k/16bit results for the rDAC, but this time displayed in terms of the peak Jitter spectrum at low fluctuation frequencies. Frequency variations with a long period have a long half-cycle time over which to build up a large timing error. So, for example, a wow of 100 parts per billion at 0·1 Hz corresponds to a peak timing Jitter component of nearly 200 nanoseconds – i.e. almost 200,000 picoseconds!"

"Compared with the more familiar short-term jitter levels people obtain from the standard ‘J Test’ such a value seems very large. But it should be remembered that the audible impact is different, so we may be comparing apples with oranges. Variations over timescales much longer than a second may be heard as a a tiny periodic variation in pitch, or be inaudible. Whereas relatively small short term variations at higher ‘flutter’ frequencies may alter the timbre or sound quality of what we hear. So it seems fairer to focus on considering the low frequency results of the IQ Test in terms of frequency ‘wow and flutter’ rather than in terms of timing jitter values."

But I don't agree with his evaluation of audibility here as my readings about the psychoacoustic effects of low frequency noise lead me to a different but not necessarily intuitive conclusion.