The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Measurements >> Phase Noise and Jitter Measurements >> PNOISE Sources Mode's Phase Noise Result
https://designers-guide.org/forum/YaBB.pl?num=1256925601

Message started by neoflash on Oct 30^th, 2009, 11:00am

Title: PNOISE Sources Mode's Phase Noise Result
Post by neoflash on Oct 30^th, 2009, 11:00am

Hi,

I'm simulating a clock buffer's phase noise.

In pnoise sources mode, there is no way to control the threshold voltage of clock zero crossing. (In jitter mode, there is setting for threshold. ) (This is the case for vco phase noise simulation too.)

This make me curious how simulator knows the correct timing to calculate the phase noise which only matters at zero crossing timing?

Regards,
Neo

Title: Re: PNOISE Sources Mode's Phase Noise Result
Post by Andrew Beckett on Jan 5^th, 2010, 1:36pm

It doesn't. Phase noise is really a misnomer with the result from sources. It's actually plotting the output noise in dBc (i.e. total output noise divided by the magnitude of the carrier). For a switching system, including the amplitude noise in the non-transition regions is of no use to you, so that's why you want to use the jitter mode.

In other cases, seeing the time-averaged output noise, presented in dBc, is what you want, so that's why you have a choice of ways of measuring it.

Although I do feel that we (Cadence) should stop calling it phase noise when computed from the pnoise sources analysis, because it's potentially misleading if you don't know what you're doing...

Unfortunately it's been this way for ever, so we are more likely to upset existing users if we change it!

Regards,

Andrew.

Title: Re: PNOISE Sources Mode's Phase Noise Result
Post by Mayank on Jan 7^th, 2010, 11:51pm

Hi,

I agree completely with Andrew's reply......

@ Andrew : Sir, since you are working with cadence, why don't you suggest them to use the term SSCR [ Single Side-Band to Carrier Ratio ] instead of Phase Noise in Direct Form pop-up from pnoise sources analysis.

--
Mayank.

Title: Re: PNOISE Sources Mode's Phase Noise Result
Post by Andrew Beckett on Jan 16^th, 2010, 8:15am

Not sure that term is that widely known. Probably better to just call it "output noise in dBc" or something like that - that's pretty clear.

Regards,

Andrew.

Title: Re: PNOISE Sources Mode's Phase Noise Result
Post by pancho_hideboo on Jan 16^th, 2010, 11:58am

Mayank wrote on Jan 7^th, 2010, 11:51pm:

@ Andrew : Sir, since you are working with cadence,
why don't you suggest them to use the term SSCR [ Single Side-Band to Carrier Ratio ]
instead of Phase Noise in Direct Form pop-up from pnoise sources analysis.

Again see http://www.designers-guide.org/Forum/YaBB.pl?num=1050465395/26#26

Simple phase noise measurements using Spectrum Analyzer give L_SSB(f) not S_Φ(f)/2.
And a phase noise result of pnoise(sources) is corresponding to this simple phase noise measurements using Spectrum Analyzer.
So it is a "Phase Noise" classically and traditionally.

But in Agilent GoldenGate, L_SSB(f) is called as "Sideband Noise" not "Phase Noise".

Title: Re: PNOISE Sources Mode's Phase Noise Result
Post by Mayank on Jan 17^th, 2010, 9:44am

Hi,
True single side-band phase noise is given by PM option in pnoise modulated analysis....
Isn't SSCR [ Single Side-Band to Carrier Ratio ] a small signal approximation to this PM noise ??
I guess that's why it matches L_SSB plot from spectrum analyzer in F_linewidth << frequency << F_oscillation / 2 ....

--
Mayank.

Title: Re: PNOISE Sources Mode's Phase Noise Result
Post by neoflash on Jun 29^th, 2011, 11:21pm

Andrew:

This is an old post however I revisit recently.

I agree that for a driven circuit we probably need to use jitter mode for phase noise measurement.

However, in oscillator simulation, it seems that source mode works pretty well in plotting the correct phase noise. You think it is just being coincidental or there is some reason behind it.

Thanks,
Neo

Andrew Beckett wrote on Jan 5^th, 2010, 1:36pm:

Title: Re: PNOISE Sources Mode's Phase Noise Result
Post by sacbr on Jul 31^st, 2013, 11:31am

I had the same question (the first question in the post) and it is really misleading. Based on Andrew's reply, it seems like phase noise is a misnomer.

I also wonder where the time averaged output noise (in dBc) for a clock buffer would be useful. Physically, what information does it give and where can we use it.

Also, assuming the voltage noise is sampled at threshold crossing. Can we repeat the the output noise spectrum (obtained from sources) in frequency domain (to account for sampling). Then find the rms of sampled noise and divide by slope of the waveform at threshold crossing to get jitter? Would this result match with one obtained from noise type = jitter mode?

Title: Re: PNOISE Sources Mode's Phase Noise Result
Post by Frank Wiedmann on Aug 1^st, 2013, 1:17am

If you would like to get a response from Andrew, you should post your question at http://www.cadence.com/community/forums/33.aspx. He does not have the time to follow the Designer's Guide Forum anymore.

Historically, pnoise jitter mode was added to SpectreRF many years after pnoise sources mode (which is not a very descriptive name either). For an oscillator, the noise around the oscillation frequency (and its harmonics) is usually almost exclusively phase noise, so that is why this name was chosen for this function a long time ago.

And yes, the method you propose should work, but you would need to use some sampling circuit and also mirror parts of the (input-referred) noise spectrum to correctly account for the sampling (see figure 3 of http://www.designers-guide.org/Analysis/sc-filters.pdf). One result of the pnoise jitter analysis is the sampled noise spectrum in this "folded" form, so that all noise contributions are included between 0 Hz and half the pss fundamental frequency.

Title: Re: PNOISE Sources Mode's Phase Noise Result
Post by sacbr on Aug 14^th, 2013, 10:53am

Thanks Frank.

My question was - Can we take the time averaged spectrum obtained from pnoise=sources option (without the explicit sampler as you mentioned).

This should give the spectrum of average voltage noise (in continuous-time) at the output. Since we are interested only in the rms of this at threshold crossing, we sample this voltage noise waveform in timedomain every FCLK. So we repeat the time averaged spectrum in frequency domain, every FCLK. Then sum all the components that alias into 0-FCLK. All this can be done in matlab with one time averaged spectrum with pnoise=sources option. We would not need an explicit sampler this way right?

Does this method yield the same result as pnoise=jitter option.

Title: Re: PNOISE Sources Mode's Phase Noise Result
Post by Frank Wiedmann on Aug 16^th, 2013, 2:04am

Sorry, I'm afraid that I don't really understand the method you are proposing. I don't think that it's possible in general to calculate jitter from the result of the pnoise=sources analysis.

Title: Re: PNOISE Sources Mode's Phase Noise Result
Post by sacbr on Mar 4^th, 2015, 10:42am

Hi Frank,

I thought I understood it, but I am stuck again. Here is what I am thinking.

Let us say we select "sources" option on pnoise on a driven circuit (say clock buffer). This should give the time averaged voltage spectrum of the clock buffer.

Ideally, using the time averaged spectrum, one could transform back to time-domain and construct the time domain voltage noise at the output. Once we have a time domain voltage noise output, we sample that at the instances of zero-crossing, find the rms and divide the slope of the clock waveform to get the jitter.

The question is - On a driven circuit, does the pnoise=sources option provide all the "information", that one would need if he/she desires to manipulate the spectrum to obtain jitter at zero-crossings?

Please help me understand.

Title: Re: PNOISE Sources Mode's Phase Noise Result
Post by Ken Kundert on Mar 5^th, 2015, 11:55am

Once the noise has been 'time-averaged' information has been lost. Specifically, how the noise varies with time across a cycle. For this reason you can no longer back-calculate the noise only at the threshold crossings.

If you want to use the time-average noise to compute jitter, you need to include an ideal limiter in your circuit that has the effect of eliminating all noise except noise that occurs very near threshold crossings, then you need to rescale the noise that you observe to account for the fact that you are observing the time-average noise but the noise itself only occurs with a small duty cycle. This is the way we did it before Spectre included jitter extraction for you.

-Ken