Designer’s Guide Community

tromeros

Junior Member

Offline

Posts: 11
Greece

Strobed noise analysis
Jan 18^th, 2006, 4:26am

Hi guys,
I want to specify the edge-to-edge jitter of a frequency divider.
I have read the paper of Ken Kundert "Predicting the Phase noise and jitter of PLL - based synthesizers" which suggests the use of pnoise analysis in SpectreRF and specifically the timedomain noise feature (also known as strobed noise analysis)

However I cannot understand how to use this feature. I have to fill one of the fields "Noise Skip Count" or "Number of Points" but I cannot understand their meaning nor the manual is clear on this terms. Huh

I would be grateful for any help. Thanks a lot!

Back to top

IP Logged

Andrew Beckett

Senior Fellow

Offline

Life, don't talk to
me about Life...

Posts: 1742
Bracknell, UK

Re: Strobed noise analysis
Reply #1 - Jan 20^th, 2006, 3:22am

Essentially the tdnoise feature (the strobed noise analysis) of pnoise allows you to strobe the noise at each timestep that was performed in the PSS analysis. In practice you don't want all of them, so Noise Skip Count allows you to strobe every nth timepoint, and Number of Noise Points allows you to distribute n points evenly over the timepoints in the PSS.

For measuring jitter however, you're mainly interested in the noise at the transitions - so you want to specify the timepoint for the transition point.

Better still, if using a recent version (there were some problems in IC5033 when this was introduced, but should be OK in IC5141), you can use the "jitter" mode of pnoise, which allows you to specify the threshold value - then it will automatically do the strobing at the threshold, and also capture the slope of the PSS time domain results at that point.

Best Regards,

Andrew.

Back to top

IP Logged

bhargav

New Member

Offline

Posts: 2

Re: Strobed noise analysis
Reply #2 - Feb 24^th, 2006, 1:04pm

Hello ,

I am new to this concept of verilog-a modeing and noise analysis. Am pretty impressed with a lot of relevent articles and model code avaiable in this site . I need to know a little more from the experts on the following areas.

1) I have come to know that Spectre RF analysis suffers from hidden state . But many of this verilog-a code for the PLL (esp the divider !) have this hidden state . So is there any way I can use this code directly to run PSS and Strobed Noise analysis? ( I have a urgent requirement to use them)

2) I presume verilog-ams is a superset of verilog-a . So does Spectre(RF) understand verilog-ams language ? Can i write a "PFD" model containing verilog-D constructs like "reg" , "wire" , "always" etc and itegrate with other models.

3) To perform PSS and Psrtobe noise analysis . Do we need to use phase-domain models or voltare domain models ?

4) Is PSS a requisite for performing Pnoise analysis and is transient is prerequisire for performing PSS ?

Sorry for barging with so many queries but an early response is very much apppreciated

Thanking You,
Bhargav

Back to top

IP Logged

Ken Kundert

Global Moderator

Offline

Posts: 2386
Silicon Valley

Re: Strobed noise analysis
Reply #3 - Feb 24^th, 2006, 3:26pm

1. You can use the frequency divider model given in the RF Models section of verilog-ams.com. Though generally you do not want to do PNoise simulations with the divider at the behavioral level. See http://www.designers-guide.org/Analysis/PLLnoise+jitter.pdf for an alternate approach.
2. Verilog-AMS is a superset of Verilog-A, but SpectreRF only supports Verilog-A.
3. Its really the other way around. If you have phase-domain models you can use AC and Noise analysis. If you have voltage-domain models, you need to use PAC and PNOISE.
4. PSS is a prerequisite for PNoise. A transient analysis is not a prequesite for PSS, but a PSS analysis always simulates the transient behavior of the circuit for at least a small period of time (under control of the tstab parameter) before starting its process for computing the steady-state solution.

-Ken

Back to top

IP Logged

bhargav

New Member

Offline

Posts: 2

Re: Strobed noise analysis
Reply #4 - Feb 25^th, 2006, 1:36am

Ken ,

Your answers were much useful in understanding the features in verilog-A and SpectreRF .

From your reply I can understand that the following things need to be done .

First develop noisy -phase domain models , use AC and noise analysis and get PSD .

Then , develop jittery voltage-domain models and run transinet analysis , got to mat-lab with these resullts and do noise analysis there .

This was you can by-pass PSS and PNoise with out having to change simple models to avoid hidden states !.

Please confirm the above . Any comments on the consistency and accurary of both these reuslts ?

Also in the furute as I have schematics developed I can keep replacing the models with the spectre netlist and run siumations and go to matlab for noise analysis !

Also the nest phase of my project is jiter optimization .

Any pointers regarding this very much welocme

Thanks Once again ,
Bhargav

Back to top

IP Logged

Ken Kundert

Global Moderator

Offline

Posts: 2386
Silicon Valley

Re: Strobed noise analysis
Reply #5 - Feb 25^th, 2006, 10:59pm

You need the PSS/PNoise analyses to characterize the noise of your blocks, which are at the circuit level, and so are not plagued with hidden state. Once you have the noise of the blocks, you can either build phase domain models, which you simulate in AC or noise analysis to get the composite noise of the PLL, or you can build a voltage-domain model and simulate the resulting PLL model in the time domain with transient analysis. These models may have hidden state, but it causes no trouble because you are using transient rather that PSS analysis.

-Ken

Back to top

IP Logged

savithru

Junior Member

Offline

Posts: 11

Re: Strobed noise analysis
Reply #6 - Jul 5^th, 2007, 10:13pm

hi

Still I could not understand the "Noise Skip count" in the strobed Noise analysis ( time domain analysis) part of the Phase Noise analysis in spectre.

Suppose my refence freq. in my PFD is 4MHz, then Can I use the Noise Skip count as 250n... (=1/4M)

Kindly clarify.

Also I want to compare the Phase noises(nemerical values, not the graph) of (PFD+CPLPF+Divider) with trhat of VCO alone.

What kind of Phase noise analysis I can do, to compare them?

Kindly reply..

Thanks & regards
SavithRu

Back to top

Thanks & Regards
SavithRu

IP Logged