Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[KSTR]

Would it be adequate to simulate sinusodial clock jitter with a FM plugin in a DAW, to study audible effects and to do some FFT measurements? In my understanding a  frequency modulation is effectively a form of "resampling" in the way described in the Hawksford paper using a sinusodual jitter stimulus. Is that correct?

Regards, Klaus

[Tomas Danko]

KSTR wrote on Sun, 20 May 2007 02:05

Would it be adequate to simulate sinusodial clock jitter with a FM plugin in a DAW, to study audible effects and to do some FFT measurements? In my understanding a  frequency modulation is effectively a form of "resampling" in the way described in the Hawksford paper using a sinusodual jitter stimulus. Is that correct? Regards, Klaus

FWIW, I once asked the Apogee people over at the (now closed) Dan Lavry forum here at the RP if they were perhaps doing the noise shape equivalent to their dither in the time domain as far as their Big Ben unit went (since they did say some jitter sounds better than other).

They went completely silent.

[bruno putzeys]

KSTR wrote on Sun, 20 May 2007 03:05

Would it be adequate to simulate sinusodial clock jitter with a FM plugin in a DAW, to study audible effects and to do some FFT measurements? In my understanding a  frequency modulation is effectively a form of "resampling" in the way described in the Hawksford paper using a sinusodual jitter stimulus. Is that correct?

It should be the same. Do the FFT measurement first to see if the plug-in is accurate enough as to produce only FM sidebands, but from there on it's a useable test platform. If you can set the modulation index as low as the kinds of levels we're talking of in the jitter context.

Tomas Danko wrote on Sun, 20 May 2007 03:27

FWIW, I once asked the Apogee people over at the (now closed) Dan Lavry forum here at the RP if they were perhaps doing the noise shape equivalent to their dither in the time domain as far as their Big Ben unit went (since they did say some jitter sounds better than other). They went completely silent.

I wouldn't know if that necessarily meant "yes". It could also mean "umm..."
To me the argument sounds more like "some kinds of jitter are more excusable than others", which basically comes down to retrofitting the requirement spec to match the measured performance.

[Quince]

I've found an online applet that estimates jitter from SSB phase noise specs supplied as dBc at various offsets (Hz) from the fundamental.  What I'm wondering is how such a computation is made; that is, given an SSB phase noise curve, how does one compute the RMS jitter?

In a given datasheet, I see both phase jitter and SSB phase noise specified; the jitter given with condition 1σ (is the sigma to mean standard deviation?) of 0.2 ps, and then says SSB phase noise 100 Hz -95 dB, 1 kHz -125 dB, 10 kHz -140 dB, 100 kHz -145 dB.  So which is more important for audio?  The jitter spec, or the phase noise (I'm assuming they are related, but I don't understand exactly how), and if the latter, at the lower or at the higher offsets?

[bruno putzeys]

http://www.grimmaudio.com/whitepapers/clock%20jitter%20spec. pdf
See section 3.3.

I recall there's a typo somewhere in the formulae (factor of 2), but you get the idea.

[Tomas Danko]

Bruno Putzeys wrote on Mon, 21 May 2007 08:27

KSTR wrote on Sun, 20 May 2007 03:05 Would it be adequate to simulate sinusodial clock jitter with a FM plugin in a DAW, to study audible effects and to do some FFT measurements? In my understanding a  frequency modulation is effectively a form of "resampling" in the way described in the Hawksford paper using a sinusodual jitter stimulus. Is that correct? It should be the same. Do the FFT measurement first to see if the plug-in is accurate enough as to produce only FM sidebands, but from there on it's a useable test platform. If you can set the modulation index as low as the kinds of levels we're talking of in the jitter context. Tomas Danko wrote on Sun, 20 May 2007 03:27 FWIW, I once asked the Apogee people over at the (now closed) Dan Lavry forum here at the RP if they were perhaps doing the noise shape equivalent to their dither in the time domain as far as their Big Ben unit went (since they did say some jitter sounds better than other). They went completely silent. I wouldn't know if that necessarily meant "yes". It could also mean "umm..." To me the argument sounds more like "some kinds of jitter are more excusable than others", which basically comes down to retrofitting the requirement spec to match the measured performance.

Yes, and that sounds kinda like noise shaping to me albeit in the time domain. A simple "no, we don't do that" would have been plenty from them. "Humm..."

Actually, they've also said that sometimes more jitter can sound better, depending on what "type". That's one of the things that had Mr. Lavry go ballistic.

Unless I need to sync lots of units, I ain't going to buy any external clock from that bridge in London or wherever it came from.

[bruno putzeys]

Tomas Danko wrote on Mon, 21 May 2007 14:22

Actually, they've also said that sometimes more jitter can sound better, depending on what "type". That's one of the things that had Mr. Lavry go ballistic.

Formally I might have responded differently, but materially my attitude is exactly the same as DL's. Stating that some kinds of jitter makes equipment sound nicer comes across as a feeble excuse for having exactly this kind of jitter problem, and is definitely a post-hoc hypothesis.

EDIT: Scroll to page 4 where it is clarified that no such claim is made.

Even if it were true that some spectral/statistic distributions of jitter have an euphonic impact, that would be true only for particular kinds of programme. Proposing the existence of *any* signal treatment that always improves the sound is like an ME having an EQ locked in one setting and running all his clients' audio through it. Rubbish of course. Lady Thatcher has recently risen in my esteem when I came across a quote of her saying "there is something logically suspect about a solution that is always right no matter the problem" (but then to realise that she herself backed just such a solution -privatise, privatise- puts her back).

As a result, if anyone believes that certain types of jitter are euphonic, that logically leads to the conclusion that the converter should be made free from it (so its use become elective) and a separate effects box or plugin be made to create and manipulate the effect.

Whether converters sound better when locked to an external clock or not, depends very much on the design of the PLL and the internal clock. Some equipment improves markedly (only to the extent that the external clock is really better), some hardly do, and a few will sound worse. It's unpredictable until you try it or until you analyse the design.

[KSTR]

Bruno Putzeys wrote on Mon, 21 May 2007 02:27

KSTR wrote on Sun, 20 May 2007 03:05 Would it be adequate to simulate sinusodial clock jitter with a FM plugin in a DAW, to study audible effects and to do some FFT measurements? In my understanding a  frequency modulation is effectively a form of "resampling" in the way described in the Hawksford paper using a sinusodual jitter stimulus. Is that correct? It should be the same. Do the FFT measurement first to see if the plug-in is accurate enough as to produce only FM sidebands, but from there on it's a useable test platform. If you can set the modulation index as low as the kinds of levels we're talking of in the jitter context.

Oh well, the latter is the problem, with the plug I use I can only go as as low as 1% "FM depth". Spectrum looks nice though, clean peaks at N*Fmod +- Fsig frequencies. But the first two frequencies are only 22dB down from the signal...
I still think it is usable to find if I and others can positivly abx tracks of audio (24/96) modulated with the plug, at various frequencies. If we can't, no need to dig further into the low sideband levels were jitter artifacts are typically expected...

Regards, Klaus

[Quince]

Thanks for the paper reference.  But that shows jitter as integrated in a range of offset frequencies.  On the other hand, I read articles that claim 1 standard deviation in the fequency domain is the RMS jitter in the time domain.  Uh, I'm not sure when seeing such measures how to compare.  For example, I see the clock with datasheet specified as above, RMS 0.2 ps, and SSB phase noise given at points starting at 100 Hz.  Then I see another clock specified as 2 ps in the range starting from 10 Hz.  So how do I compare, or is it even possible, or I don't have enough data?

[bruno putzeys]

The noise spectrum of an oscillator (in the absence of tones) breaks down into three main regions. At small offsets, noise drops 40dB/dec. This corresponds to amplifier 1/x noise being phase integrated. At medium offsets, noise drops 20dB/dec, corresponding to white noise being integrated. Further still you end up with white noise.

Since you don't know where these corners are, extrapolating a jitter spec is dangerous. At 10Hz you can be quite sure you're in the 40dB/dec region so that gives some certainty, but extrapolating to 100Hz is likely to yield an overly optimistic estimate.

[Quince]

I meant extrapolating inwards, since the datasheet shows 100 Hz and up only.  So from 100 to 10, should I go up by 40 dB?  The datasheet gave 100 Hz -95 dB, 1 kHz -125 dB, 10 kHz -140 dB, 100 kHz -145 dB, so then it would be around -55 dB for 10 Hz right?

[bruno putzeys]

Ah OK I thought you were wondering about the integrated noise figure. Indeed you can extrapolate noise density from 100Hz to 10Hz by adding 40dB.

[Quince]

OK, I had thought you may have meant from near 0 Hz would go down by 40 to 10 Hz.

Using the tool at http://www.raltron.com/cust/tools/osc.asp if I just put in the given datasheet numbers starting at 100 Hz, I get a figure close to the 0.2 ps specified in the datasheet.  But if I extrapolate up by 40 dB for 10 Hz offset and add that datapoint, I get a bit under 5 ps... that's pretty huge.  These are also for the datasheed given value of 155.52 MHz; I wonder if SSB phase noise numbers can be expected to scale to give similar jitter figures for the usual 24.576/22.5792 MHz oscillators.

[bruno putzeys]

5ps @10Hz is huge? I should point out the relativity. If you wait long enough, all oscillators have infinite jitter.

[Quince]

I meant compared to a couple other clocks I was looking at, but this one is much cheaper that's why I was so excited.