The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> RF Simulators >> Estimate pk-pJitter using eye diagram in SpectreRF
https://designers-guide.org/forum/YaBB.pl?num=1107968758

Message started by ic_engr on Feb 9^th, 2005, 9:05am

Title: Estimate pk-pJitter using eye diagram in SpectreRF
Post by ic_engr on Feb 9^th, 2005, 9:05am

Hello All,

I am trying to estimate peak-peak jitter with Time Interval Error of 30us for my Oscilltor. The fundamental frequency for my oscillator is only 2MHz.

I ran pss-pnoise simulations in SpectreRF. Therafter, in the simulation directory I plotted trans-pss plot of my oscillator output and used the Eye Diagram function in the calculator. I enetered the interval of 30usec and used an estimate for the period and plotted the eye diagram. I am measuring the spread on the eye diagram as the pk-pk jitter. Will this method give me proper pk-pk jitter for TIE of 30usec.
Does the tran-pss output generated in spectre is same as the one generated in eldo using .noisetran, i.e includes all the noise affects since my models have flicker noise parameters.

Thx

ic_engr

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by Andrew Beckett on Feb 9^th, 2005, 1:07pm

No, this will not give you any kind of jitter information.

If you're looking at the pss-td results, then this will only be a single period of the fundamental frequency. If you're looking at the pss-tran (or whatever it is called), then this is just the initial transient used for partial settling before the shooting interval starts. In either case, these do not include any noise sources.

In pnoise, the noise sources are added as small signal noise sources around each of the devices in the circuit. It will then compute the transfer functions of these to the output (including any noise folding effects) giving you the total noise at the output of the circuit in a desired band. You can also strobe the noise using tdnoise.

All of this is discussed in Ken's paper on jitter and phase noise in the Analysis section (see links at the top of the site's web page). Also, there is a "jitter" mode for pnoise now (introduced in IC5141) which allows you to measure jitter more conveniently/directly for both driven and autonomous circuits.

Spectre does now have a transient noise analysis (as from MMSIM60 onwards), but in general you would usually be better off using pnoise/tdnoise if you can (it would normally be quicker and more accurate). There have been discussions to this effect in this forum before, if my memory serves me correctly. Transient noise analyses are best for looking in a qualitative way where noise results in some large signal effect, particularly in non-periodic circuits (or at least circuits where the period is not constant).

Regards,

Andrew.

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by ic_engr on Feb 9^th, 2005, 1:53pm

Andrew,

Thanks for the reply. If I do use the pnoise then based on Ken's paper I can compute accumulating jitter. I did compute the accumulating jitter using the pnoise. However, what I am interested in is the pk-pk jitter in Time Interval Error of 30usec, how would I estimate that.

Your help will be appreciated.

The output of the oscillator I am designing is a square pulse and would pnoise analysis be still accurate?.

Tx

ic_engr

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by Ken Kundert on Feb 10^th, 2005, 8:43am

The concept of peak-to-peak jitter is not well defined in this context. The noise sources are Gaussian, which are unbounded. And as such, the resulting jitter is unbounded. The best you can do is give the likelihood that a particular bound is exceeded.

-Ken

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by ic_engr on Feb 10^th, 2005, 8:55am

Thanks Ken,

I did take note of that in your paper. However, the specification is that in a Time Interval error of 30usec I need not to exceed a jitter of 1.0nsec for my oscillator.

I am trying to use the tdnoise feature for this purpose.
Would I be able to predict the jitter based plotting the voltage vs Spectrum and then perform integration across the BW. Should I square before integration using the calculator or not.

Thanks for your time Ken.

ic_engr

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by Ken Kundert on Feb 10^th, 2005, 1:14pm

You should always integrate the noise power (V²/Hz) and not the noise voltage. So you can either square the noise voltage or use the noise power directly, Spectre computes both.

-Ken

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by ic_engr on Feb 10^th, 2005, 1:55pm

Thanks Ken,

I did integrate the sqaure of the voltage over 0 to fo/2. fo=4.04MHz.

I got a number like 800. Now I divided this with the Slew Rate of my PSS fundamental and I got something very large like 40nsec.

Did I do something wrong.

Tx

ic_engr

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by Ken Kundert on Feb 10^th, 2005, 3:16pm

At a minimum, you must take the square root after integrating to convert V² to V. Then dividing through by the slew rate will give you units of seconds.

-Ken

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by ic_engr on Feb 10^th, 2005, 4:52pm

Thanks Ken,

I will do that, In the PNOISE form when I select the peak, and compute the jitter the way you mentioned would it still be rms jitter or pk jitter.

Also calculating jitter this manner will include the 1/f noise or still assume only white noise using the strobe method (tdnoise)

I appreciate your time on this matetr.

ic_engr

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by Ken Kundert on Feb 10^th, 2005, 5:16pm

I am not familiar with the peak option on the PNoise form, but as I said before, it cannot be the peak jitter because Gaussian jitter is not bounded. Peak jitter really only makes sense when the jitter is deterministic.

You should avoid the flicker noise (remove it before integrating). Jitter due to flicker noise is not defined.

-Ken

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by ic_engr on Feb 10^th, 2005, 7:29pm

Ken Sorry for the mis-information. Its actually the
signal selection in the PSS form (peak or rms).

Should I plot the noise from pnoise or from pss after I run the tdnoise simulation.

Can I run PNOIS simulation without tdnoise and just plot the output noise A^2/Hz and integrate over the BW.

Thanks for your input.

ic_engr

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by Andrew Beckett on Feb 10^th, 2005, 9:55pm

I wondered whether you might be using the pss results instead of the pnoise/tdnoise results (from another post you made in another thread).

It sounds as if you're doing Direct Plot->Main Form, and then you'd get a choice of pss and pnoise. If you're interested in the noise, looking at the pss results will not give you the information you're after, since that is the large signal results (as I explained earlier), before the noise sources are enabled. It's the pnoise tab which will give you access to the pnoise or tdnoise results.

I think you need to read the SpectreRF documentation (user guide, theory guide) in cdsdoc, as I think your understanding of what each analysis does is a little misplaced.

Regards,

Andrew.

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by ic_engr on Feb 11^th, 2005, 7:04am

Andrew,

Thanks for pointing that out and I did read the SpectreRF guide on this issue. Things are much clearer.

I did run tdnoise and in the tdnoise form (not PSS form anymore) I plotted the integrated noise and I got 10.5uV. If I divide this number by the Slew Rate of 20e9V/sec I get a rms jitter of 0.52fSec. I feel this is very low. Is it due to the fact that in this method we are assuming only white noise and neglecting 1/f noise.

I was wondering if now I go back to the standard PNOISE simulation and plot output noise as A^2/Hz and integrate that lets say from 100Hz (to include some appreciable 1/f noise contributions) to fo/2 and then divide this by Slew Rate, would'nt this be a more realistic estimate of Jitter for my oscillator. As I have to make estimate for long term jitter and we know in log term jitter the 1/f noise contributes.

I do agree that if all the 1/f noise (1Hz since 1/f is at lower freq) is included, as Ken pointed out ,we may have a problem that jitter will be unbounded, but if we include some portion of low frequency 1/f noise e.g 100Hz then will this provide a better estimate for jitter?.

I really appreciate your feedback.

ic_engr

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by Ken Kundert on Feb 11^th, 2005, 9:06am

If you use standard PNoise rather than the strobed PNoise, then you are including noise that occurs at times other than when the signal is crossing its threshold. Noise produced when the output is all the way high or all the way low does not contribute to jitter, but often dominates the noise that standard PNoise measures.

Actually, I went back and took a closer look at my notes, and I think you should be able to include the flicker noise. Before you do though, make sure you have carefully resolved the flicker noise. I often see people use equal frequency point spacing during their PNoise analysis with an insufficient number of points to accurately resolve the flicker noise corner, and when they do the integration they get a highly inflated number for the flicker noise contribution.

-Ken

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by ic_engr on Feb 11^th, 2005, 9:12am

Thanks Ken,

I already have Flicker Noise included in the models. So when I run strobed Pnoise and on pnoise form after simulation select Integrate output noise and use varying BW I get the same number, 10.5u.

Woudl you know what may be happening.

Or should I go to pss page after the simulation of strobed pnoise and plot output SQR(V) and integrate across the BW.

Please let me know

ic_engr

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by Ken Kundert on Feb 11^th, 2005, 9:35am

Initially you indicated that your oscillator frequency was 2MHz, then you said it was 4MHz. Either way I would not expect you to have a slew rate of 20GV/s. Is this really right? If so, why do you believe 0.5 fs is too small?

Also, did you multiply by sqrt(f₀*30us) to convert period jitter into k-cycle jitter?

-Ken

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by ic_engr on Feb 11^th, 2005, 9:41am

Ken,

The oscillator frequency is fo=4MHz. what I said was fo/2 = 2MHz.
The oscillator output is a square wave with a rise time from 100mV to 1.1V of 40pSec, therefore, the Slew Rate is indeed 1/40psec=20GV/sec.

No I did not multiply the period jitter by sqrt(fo*30us) yet. The reason I suspect 0.5fSec is too low because the current in each stage of my ring oscillator is very low. Its a low power design and therefore I am not optimisitic that my jitter will that good.

Thx

ic_engr

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by Ken Kundert on Feb 11^th, 2005, 5:36pm

It does seem a little too good, but I don't know what else to suggest.

-Ken

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by ic_engr on Feb 12^th, 2005, 6:08am

Ken,

I made some modifications to the ring oscillator design. I increased the current and maintained same frequency. When I run standard PSS and Pnoise simulation what I note is that the slope of the Phase noise plot is 30dB/decade upto near the fundamental freq. The fundamental is 4MHz.

The usual fc (1/f noise corner I used to see is now gone.

Can you help me understand what's happened.

Tx

ic_engr

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by ic_engr on Feb 12^th, 2005, 9:12am

Does this mean that fc doesnot exist in the new design. The slope is all the way -30dB/dec.

Pleas let me know.

ic_engr.

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by Andrew Beckett on Feb 13^th, 2005, 10:02am

This usually occurs because you set maxsidebands in the pnoise analysis to "0" - whereas it should be a number large enough to include all significant contributions. Setting it to 0 in particular will omit any up-converted flicker noise from DC, so that's often a cause.

Regards,

Andrew.

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by ic_engr on Feb 14^th, 2005, 3:23am

I did check the Maxsidebands and it is set to 12. So I am not sure what's happening. The change I did is that I made the bias circuit with a resistor. The slope is all the way -30dB/dec.

Any suggestions will be greatly appreciated.

Thx

ic_engr

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by Andrew Beckett on Feb 14^th, 2005, 4:52am

You might want to look at the noise contributions. I suspect that your noise may be dominated by the noise from the bias circuitry (I don't think that's necessarily unusual, if there's high gain from the bias circuitry to the output), and since that doesn't have any flicker noise, you don't see the flicker knee.

Can't remember if you're running in ADE - but if you are, use the noise contribution UI (under the results menu) to take a look at which devices contribute the most noise.

Regards,

Andrew.

Title: Re: Estimate pk-pJitter using eye diagram in Spect
Post by ic_engr on Feb 14^th, 2005, 1:32pm

You are right I did check and its due to the resistor. Okay So can I then estimate jitter with the phase noise plot I have or not.

Since the expressions in Ken's paper do mention that the slope should be -20dB/dec and I have -30dB/dec will these be valid to use.

ic_engr