The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> RF Simulators >> phase noise of ring oscillator using spectre
https://designers-guide.org/forum/YaBB.pl?num=1050465395

Message started by kuolance on Apr 15^th, 2003, 8:56pm

Title: phase noise of ring oscillator using spectre
Post by kuolance on Apr 15^th, 2003, 8:56pm

Hi All,

I'm using spectre to estimate the phase noise of my ring oscillator. The results I got was +90 dBc/Hz at 1 Hz offset and -70 dBc/Hz at 1MHz offset. I'm not sure how to interpret the result. Is it possible to get a positive dBc/Hz for phase noise? Shouldn't phase noise be 0 dBc/Hz maximum? Any insight you can provide is greatly appreciated.

Thanks!

kuolance

Title: Re: phase noise of ring oscillator using spectre
Post by Frank Wiedmann on Apr 16^th, 2003, 12:38am

While the phase deviation of a noisy free-running oscillator will indeed grow without bound over time, the voltage deviation of course will not. This, however, is a nonlinear effect that is not taken into account when SpectreRF presents its results.

You can find the solution to your problem in equations (24) and (26) of Ken's paper "Predicting the phase noise and jitter of PLL-based frequency synthesizers" (http://www.designers-guide.com/Analysis/PLLnoise+jitter.pdf). SpectreRF presents the approximation according to equation (26), the real result, however, is given by equation (24).

By the way, the integral over equation (24) from -infinity to +infinity is always equal to 1. This means that the total power of the oscillator signal remains constant, it is only spread over a wider frequency range when there is more phase noise.

Title: Re: phase noise of ring oscillator using spectre
Post by Tommy on Feb 14^th, 2005, 8:57pm

Thanks Frank,
So If I have followed you correctly, By plotting eqn 26 from Spectre, I will end-up with a quantity which has the dimiensions of (rad*V)**2/Hz.
Or am I making a mistake here

Thanks
Tommy

Title: Re: phase noise of ring oscillator using spectre
Post by Frank Wiedmann on Feb 14^th, 2005, 11:51pm

You can see the dimension of L in equation (22) where it is defined. S_v is in V²/Hz (see footnote 5), a₁ and b₁ are in V (see equation (23)). So you have (V²/Hz)/V² or 1/Hz.

Title: Re: phase noise of ring oscillator using spectre
Post by tommy on Feb 14^th, 2005, 11:59pm

Guess Iam little confused now.

The dimensions of Sphi(delta-f) is rad**2/Hz.

So does eqn27 mean only a numerical equality & not dimensional?

Thanks
Tommy

Title: Re: phase noise of ring oscillator using spectre
Post by tommy on Feb 15^th, 2005, 12:38am

I got the answer to my above question from "phase noise or L" thread.
http://www.designers-guide.com/Forum/?board=jitter;action=display;num=1036525104

Spectre plots oscillator-phase-noise in dBc/Hz. To plot this curve in rad**2/Hz, I need to multiply the
10**(phasenoise(in_dbc/Hz) / 10.0) with the carrier power.(assuming that phase-nosie button in analog artist considers the Carrier power while plotting it in dBc/Hz)

PSS analysis gives the carrier power in V**2. If I multiply with this value with 10**(phasenoise(in_dbc/Hz) / 10.0), I will be left with the dimensions of Sphi as (rad*V)**2/Hz.

Or am I wrong somewhere?

Thanks
Tommy

Title: Re: phase noise of ring oscillator using spectre
Post by Tommy on Feb 15^th, 2005, 1:01am

AAh,
I think I can answer my own question ;D

The phase-noise button plots with the dimensions dBc/Hz (or 1/Hz in a non-log scale). Multiply this with the carrier power leaves us with the dimensions of V**2/Hz. Using eqn 27, this is equated to the phase noise with the dimensions of rad**2/Hz.

Thanks
Tommy

Title: Re: phase noise of ring oscillator using spectre
Post by Frank Wiedmann on Feb 15^th, 2005, 11:29pm

Equation (27) equates L in (V²/Hz)/V² (or 1/Hz) with the phase noise in rad²/Hz (or 1/Hz), so the dimensions are ok. Remember that rad is a dimensionless quantity (arc length divided by radius).

Title: Re: phase noise of ring oscillator using spectre
Post by Tommy on Feb 16^th, 2005, 12:30am

Thanks a lot Frank for clearing up.

Title: Re: phase noise of ring oscillator using spectre
Post by Mayank on Nov 23^rd, 2009, 1:04am

Hello all,
I am sorry for bringing up this thread again...I dug up into some of the old posts including http://www.designers-guide.org/Forum/YaBB.pl?num=1102124964/0

.....Now point is -- i understand that cadence uses the approximation of cfo²/Δf² as indicated in eq. 26 in Ken's paper -- http://www.designers-guide.com/Analysis/PLLnoise+jitter.pdf

Also i do understand that total power is constant, it's just that it's distributed over the spectrum....

BUT, Frank pointed out a way of obtaining correct plots by using eq. 24 instead of cadence's default eq 26....Can someone elaborate how exactly can i obtain a plot of eq. 24 from cadence ???

--expecting some help from big simulator gurus.....

-thanx in advance,
mayank

Title: Re: phase noise of ring oscillator using spectre
Post by pancho_hideboo on Nov 23^rd, 2009, 1:13am

Mayank wrote on Nov 23^rd, 2009, 1:04am:

BUT, Frank pointed out a way of obtaining correct plots by using eq. 24
instead of cadence's default eq 26....
Can someone elaborate how exactly can i obtain a plot of eq. 24 from cadence ???

I can't understand meaning of your questions.
Please explain in detail with correct terminologies.

There is no tool which name is Cadence.

See
http://www.designers-guide.org/Forum/YaBB.pl?num=1241894034
http://www.cadence.com/community/forums/T/12466.aspx

EDA Tool Play is no more than Tool Play.
There is nothing superior to the actual measurement using actual instruments.

Title: Re: phase noise of ring oscillator using spectre
Post by Mayank on Nov 23^rd, 2009, 3:01am

Hello Pancho & everyone,
My Problem :-- I am getting +ve values of Phase Noise plot ( i.e. > 0 dBc ) that i obtain from pss/pnoise analysis using spectre(MMSIM7.0) in cadence Analog Design Environment. The ckt is a 10-stage single-ended Ring Oscilllator.

My Question :-- How to get correct Phase Noise plot ??

My Analysis :--
I went through the following threads :--
1. http://www.designers-guide.org/Forum/YaBB.pl?num=1102124964/0
2. http://www.designers-guide.org/Forum/YaBB.pl?num=1241894034
3. This thread itself.

In thread 1, Ken explains that > 0dBc plot is because spectre,like almost every other simulator, uses small-signal approximation.

In thread 3/this thread, Frank, in relation to Ken's paper -- http://www.designers-guide.com/Analysis/PLLnoise+jitter.pdf explains that --
Quote:

SpectreRF presents the approximation according to equation (26), the real result, however, is given by equation (24).

I want to know how to backcalculate the data required in eq. 24 from the standard outputs in cadence's spectreRF, and then apply equation 24 on that data, & obtain actual and correct plot ??

& btw, i read the 2nd thread that you posted -- http://www.cadence.com/community/forums/T/12466.aspx ---says MMSIM7.1 has it fixed in HB/HBpnoise analysis of spectreRF.....dnt have the new updated version.....But it would be helpful if i could obtain it in this version of spectreRF through backcalculations...

seeking help,
mayank.

Title: Re: phase noise of ring oscillator using spectre
Post by pancho_hideboo on Nov 23^rd, 2009, 3:09am

Mayank wrote on Nov 23^rd, 2009, 3:01am:

My Problem :-- I am getting +ve values of Phase Noise plot ( i.e. > 0 dBc )

What is "+ve" ?

Mayank wrote on Nov 23^rd, 2009, 3:01am:

Quote:

SpectreRF presents the approximation according to equation (26), the real result, however, is given by equation (24).

I want to know how to backcalculate the data required in eq. 24
from the standard outputs in cadence's spectreRF,
and then apply equation 24 on that data,
& obtain actual and correct plot ??

Again seen http://www.designers-guide.org/Forum/YaBB.pl?num=1244213153/6#6

If Cadence Spectre's results could be reasonable,
Lssb(f) has to follow eq.(24), while S_Φ(f) follow eq.(26).

Mayank wrote on Nov 23^rd, 2009, 3:01am:

http://www.cadence.com/community/forums/T/12466.aspx ---says MMSIM7.1 has it fixed in HB/HBpnoise analysis of spectreRF.....dnt have the new updated version.....
But it would be helpful if i could obtain it in this version of spectreRF through backcalculations...

What do you mean by backcalculations ?
Do you mean "Post Processing" ?

Have "Augmented Pnoise of Cadence Spectre" come to be reliable ?

Title: Re: phase noise of ring oscillator using spectre
Post by Mayank on Nov 23^rd, 2009, 4:03am

hello Pancho,
i was always a big fan of the very particular post u just pointed out to on the other thread....It was very clear & exhaustive....Coming back to this,

By +ve, as i already indicated in the statement, i meant value of pnoise plot > 0 dBc at some offsets below linewidth frequency.

By backcalculations, i meant how to calculate Sv, fourier coeffs a1 & b1, or how to calculate c so that i can use eqns. 22 & 24

As from your post which you just pointed, you said
Quote:

L_SSB(f) is small signal mixing noise, that is, sideband noise which include both AM and PM noises.
L_SSB(f) is composed of L_LSB(f) and L_USB(f).
They are not equal. But they are almost same.

What exactly did you mean by composed of ? Is it the equal to (L_USB + L_LSB) / 2 or just the summation ??

And you said,

Quote:

In Cadence Spectre Pnoise Analysis,

For noisetype="sources",
you can get Phase Noise in "pnoise of Direct Plot Form" as L_USB(f) if you use postive offset frequency value.

For noisetype="jitter" or "modulated",
you can get L_LSB(f), L_USB(f), AM noise, PM noise in "pnoise modulated of Direct Plot Form",
and also you can get Phase Noise in "pnoise jitter of Direct Plot Form" as true S_Φ(f).

How do i obtain L_SSB plot from cadence which will follow eq. 24 not eq 26 ???

& Augmented HB-PSS/PNOISE with Oversampling factor=10 is giving me phase noise plot with a FLAT SPECTRUM uptil Linewidth frequency.... Is is CORRECT ??

--mayank

Title: Re: phase noise of ring oscillator using spectre
Post by pancho_hideboo on Nov 23^rd, 2009, 4:15am

Cadence Spectre has not been able to calculate phase noise correctly for long time.

Papers you see were written first in such days when Cadence Spectre didn't have HB Analysis
and didn't have ability to calculate PM and AM noise separately.
Of course Cadence Spectre didn't have Augmented option in Pnoise.

Also it seems still an augmented option can not show effect for Shooting-PSS.

Mayank wrote on Nov 23^rd, 2009, 4:03am:

What exactly did you mean by composed of ?
Is it the equal to (L_USB + L_LSB) / 2 or just the summation ??

Neither.

Consider actual Spectrum around Carrier. L_SSB(f) is completely different from S_Φ(f).
I mean
L_SSB(f)=L_LSB(f) for lower frequency than carrier.
L_SSB(f)=L_USB(f) for upper frequency than carrier.

Mayank wrote on Nov 23^rd, 2009, 4:03am:

& Augmented HB-PSS/PNOISE with Oversampling factor=10 is giving me phase noise plot with a FLAT SPECTRUM uptil Linewidth frequency.... Is is CORRECT ??

If Cadence Spectre is reasonable, it is L_SSB(f) which follows eq.24.

Mayank wrote on Nov 23^rd, 2009, 4:03am:

By +ve, as i already indicated in the statement, i meant value of pnoise plot > 0 dBc at some offsets below linewidth frequency.

I don't think "+ve" is common expression which everyone can understand.

Title: Re: phase noise of ring oscillator using spectre
Post by Mayank on Nov 23^rd, 2009, 5:04am

Hello pancho,
Gotcha, So summing up our discussions so far,
1. For +ve Offset Frequencies, Lssb = Lusb,
2. For -ve Offset Frequencies, Lssb = Llsb...
3. Cadence plots Lssb, which is Lusb or Llsb depending on whether the offset specified as +1 or -1. defualt being +1, it is Lusb.
4. pnoise modulated gives all components separately
5. pnoise jitter plots true Sphi which should be equal to PM noise from pnoise modulated .

Am i correct so far ??

I simulated with HB-PSS/augmented-PNOISE and i am attaching the waveforms obtained for LSB, USB, AM & PM noise plots....

In HB-PSS/AUGMENTED-PNOISE, To my surprise, AM plot is coming too close to the PM plot....Also L_LSB is totally wierd and nothin similar to L_USB...Totally confused by the results.....

Title: Re: phase noise of ring oscillator using spectre
Post by pancho_hideboo on Nov 23^rd, 2009, 5:19am

Mayank wrote on Nov 23^rd, 2009, 5:04am:

1. For +ve Offset Frequencies, Lssb = Lusb,
2. For -ve Offset Frequencies, Lssb = Llsb...
3. Cadence plots Lssb, which is Lusb or Llsb depending on whether the offset specified as +1 or -1. defualt being +1, it is Lusb.
Am i correct so far ??

I can't understand your expressions.

Mayank wrote on Nov 23^rd, 2009, 5:04am:

In HB-PSS/AUGMENTED-PNOISE, To my surprise, AM plot is coming too close to the PM plot....Also L_LSB is totally wierd and nothin similar to L_USB...Totally confused by the results.....

I think you treat L_LSB and AM noises wrongly.
Maybe your AM noise is L_LSB and your L_LSB is AM noise.
Confirm your AM noise and L_LSB.

Do you have Agilent GoldenGate(4.3.6 or 4.3.7) or Agilent RFDE(ADSsim 2008U2 or 2009) ?

Title: Re: phase noise of ring oscillator using spectre
Post by Mayank on Nov 23^rd, 2009, 6:33am

Hello Pancho,

At first i also thought so.....But i rechecked it twice....The image i attached previously is correct...I plotted USB,LSB, AM & PM from pnoise_modulated of Direct plot Form.....It is showing L_LSB wierdly.....& it shows PM Noise = AM Noise ....Totally wierd !!!

However, the plots without augmented pnoise analysis are the ones i am attaching now....

Is this a bug in spectreRF that it exchanges AM & L_LSB plots OR is it a bug in augmented pnoise analysis OR is it really possible ? i have no clue....

And i dont have access to any tools from Agilent... :(

mayank

Title: Re: phase noise of ring oscillator using spectre
Post by pancho_hideboo on Nov 23^rd, 2009, 6:39am

Mayank wrote on Nov 23^rd, 2009, 6:33am:

Is this a bug in spectreRF that it exchanges AM & L_LSB plots OR is it a bug in augmented pnoise analysis OR is it really possible ? i have no clue....

Post your questions in http://www.cadence.com/Community/forums/33.aspx

I think these are bugs and augmented pnoise is still unreliable and suspicious..

Title: Re: phase noise of ring oscillator using spectre
Post by Mayank on Nov 23^rd, 2009, 9:02pm

Hello Pancho,
Thanx for all the help and your quick replies....They helped a lot....One of my best discussions on the forum....

mayank

Title: Re: phase noise of ring oscillator using spectre
Post by pancho_hideboo on Nov 24^th, 2009, 3:21am

Mayank, again see the followings.
http://www.designers-guide.org/Forum/YaBB.pl?num=1241894034/12#12
http://www.designers-guide.org/Forum/YaBB.pl?num=1241894034/23#23

Originally Lorentzian Spectrum Models were derived in the absence of flicker noise.

In the absence of 1/f noise, it is well known that phase noise should show a Lorentzian spectrum. We can easily apply this to evaluation of the phase noise characteristics.

But modeling as Lorentzian Spectrum does not yield meaningful results if 1/f noise is present.
There is an issue for the use of the Lorentzian spectrum in the presence of 1/f noise.

PM noise can become very large at low offset because of continued accumulation of phase close to the carrier, so it could go above 0dBc/Hz. Properly, units of PM noise should be "dBrad/Hz".

SideBand Noise, L_SSB(f) is an approximation combining AM and PM noise, so L_SSB(f) could also go above 0dBc/Hz at low offset especially in the presence of 1/f noise.

Your following HB-PSS/AUGMENTED-PNOISE results are all completely flat for small offset frequency region.
http://www.designers-guide.org/Forum/YaBB.pl?num=1050465395/15#15

Even PM noise is also completely flat. These results are very suspicious.

Such completely flat phase noise reminds me phase noise of Agilent MDS or Agilent earlyday's ADS which underestimated phase noise for small offset frequency region.
See "Phase Noise Overview" of "Chapter 10: Oscillator Noise Simulation" in the following.
http://cp.literature.agilent.com/litweb/pdf/ads15/cktsim/index.html

I think Cadence Spectre's HB-PSS/AUGMENTED-PNOISE is still very suspicious.

Apart from Spectre's HB-PSS/AUGMENTED-PNOISE, some analytical models have been proposed for phase noise with flicker noise.
These decompose the model into a Lorentzian for the white noise sources and Gaussian for the flicker noise source.

So these share the property of the Lorentzian spectrum so that it has an integrated area of one ; it conserves the overall power of the noiseless oscillator,
and has a finite value as the offset frequency approaches zero.

At small offsets the spectrum looks like a Gaussian when dominated by flicker noise ;
at larger offsets 1/f³ behavior is seen, followed by 1/f² behavior.

[1] F. Herzel, "An Analytical Model for the Power Spectral Density of a Voltage-Controlled Oscillator and Its Analogy to the Laser Linewidth Theory", IEEE Transactions on Circuits and Systems . I: Fundamental Theory and Applications, vol. 45, pp. 904.908, Sept. 1998.

[2] G. V. Klimovitch, "Near-Carrier Oscillator Spectrum Due to Flicker and White Noise", Proc. of ISCAS 2000, IEEE International Symposium on Circuits and Systems, Geneva, pp. I-703.706, 2000.

[3] G. V. Klimovitch, "A Nonlinear Theory of Near-Carrier Phase Noise in Free-Running Oscillators", Proc. of Third IEEE International Conference on Circuits and Systems, Caracas, pp T80/1.6, 2000.

[4] A. Demir, "Phase Noise in Oscillators: DAEs and Colored Noise Sources", Proc. of ICCAD-98, pp. 170.177, 1998.

Title: Re: phase noise of ring oscillator using spectre
Post by Mayank on Nov 25^th, 2009, 12:14am

Hello Pancho,
I want to reserarch more on this so that i can resolve the issue....Could you provide me the 3rd paper you mentioned --
Quote:

[3] G. V. Klimovitch, "A Nonlinear Theory of Near-Carrier Phase Noise in Free-Running Oscillators", Proc. of Third IEEE International Conference on Circuits and Systems, Caracas, pp T80/1.6, 2000.

It would be very helpful....
Also i wanted to ask Can AM noise become equal to PM noise at such low offsets as 10kHz ??
Have you verified any silicon vs. CAD results so that i can know what matches silicon results better ?? -- the flattened Pnoise plot OR the actual > 0dB plot ??

mayank

Title: Re: phase noise of ring oscillator using spectre
Post by pancho_hideboo on Nov 25^th, 2009, 1:45am

Mayank wrote on Nov 25^th, 2009, 12:14am:

Also i wanted to ask Can AM noise become equal to PM noise at such low offsets as 10kHz ??

I don't think so.

Mayank wrote on Nov 25^th, 2009, 12:14am:

Have you verified any silicon vs. CAD results so that i can know what matches silicon results better ?? -- the flattened Pnoise plot OR the actual > 0dB plot ??

In actual Measurement using Actual Instruments, L_SSB(f) never go above 0dBc,
while PM(f) can be go above 0dBrad/Hz.
Using Vector Signal Analyzer, we can evaluate AM and PM noises.

https://forums.tm.agilent.com/community/viewtopic.php?f=33&t=2504
https://forums.tm.agilent.com/community/download/file.php?id=570

I have Agilent 4352B VCO/PLL Signal Analyzer and 89400 & 89600 Vector Signal Analyzer.
I don't have E5052B which is new replacement of 4352B.

Agilent E5052B Signal Source Analyzer can measure AM noise directly.
http://cp.literature.agilent.com/litweb/pdf/5989-7273EN.pdf

Title: Re: phase noise of ring oscillator using spectre
Post by Mayank on Nov 25^th, 2009, 9:08pm

Heloo Pancho,
The Document on Choosing Phase Noise Techniques by Agilent was quite helpful.....

But i didnt exactly understand this concept....PM Noise refers to Phase Modulation due to noise.....& Phase Noise is a cumulative effect of both AM & PM.....Then shouldn't Phase Noise also go higher than 0 dBrad/Hz if one of it's components are going above 0dBrad/Hz .....

If L_SSB doesnt go above 0dBc , then is it the correct representation of Phase Noise ??

--mayank.

Title: Re: phase noise of ring oscillator using spectre
Post by pancho_hideboo on Nov 25^th, 2009, 10:36pm

Mayank wrote on Nov 25^th, 2009, 9:08pm:

Then shouldn't Phase Noise also go higher than 0 ~~dBrad/Hz~~dBc/Hz
if one of it's components are going above 0 dBrad/Hz .....

SideBand Noise, L_SSB(f) is an approximation combining AM and PM noise,
so in simulation using current EDA Tool, L_SSB(f) could also go above 0 dBc/Hz at low offset especially in the presence of 1/f noise.

Mayank wrote on Nov 25^th, 2009, 9:08pm:

If L_SSB doesnt go above 0dBc, then is it the correct representation of Phase Noise ??

What I can say surely is
"Cadence Spectre's HB-PSS/AUGMENTED-PNOISE is very suspicious".
That's all.

Modeling as Lorentzian Spectrum does not yield meaningful results if 1/f noise is present.
Again see the followings.
http://www.designers-guide.org/Forum/YaBB.pl?num=1241894034/12#12
http://www.designers-guide.org/Forum/YaBB.pl?num=1241894034/23#23

Also see last append by "RFMEMS" in http://www.cadence.com/community/forums/T/12466.aspx
Even RFMEMS has a doubt now.

I think results of HB-PSS/Pnoise without Augmented is rather proper.

But even using Agilent GoldenGate Simulator, L_SSB for small offset frequency can not be matched to actual measurement well.

Title: Re: phase noise of ring oscillator using spectre
Post by Mayank on Dec 10^th, 2009, 10:17pm

Hi Pancho,
I am too confised at this stage :(
Just confirming Once Again :----

1. According to the doc you poited out on Choosing PhaseNoise Techniques by Agilent, I should trust S_Φ/2 values instead of L_USB.

So does the true S_Φ plot (given by pnoise jitter option in Direct Plot Main Form of cadence as pointed out by you in your post http://www.designers-guide.org/Forum/YaBB.pl?num=1244213153/6#6) never go above 0 dBc/Hz ??

2. If you have compared Silicon vs CAD results, which is a better approximation to actual measurements out of these ----->
1. Augmented Pnoise ??
2. S_Φ/2 below linewidth & L_Φ after Linewidth ??
3. Only L_USB plot of vanilla Pnoise ??

-- mayank.

Title: Re: phase noise of ring oscillator using spectre
Post by pancho_hideboo on Dec 11^th, 2009, 12:55am

Mayank wrote on Dec 10^th, 2009, 10:17pm:

1. According to the doc you poited out on Choosing PhaseNoise Techniques by Agilent, I should trust S_Φ/2 values instead of L_USB.

Strictly speaking, S_Φ/2 is diffefrent from L_USB physically and conceptually.

Mayank wrote on Dec 10^th, 2009, 10:17pm:

So does the true S_Φ plot (given by pnoise jitter option in Direct Plot Main Form of cadence as pointed out by you in your post http://www.designers-guide.org/Forum/YaBB.pl?num=1244213153/6#6) never go above 0 dBc/Hz ??

S_Φ(f) could go above 0 dBrad/Hz.

Mayank wrote on Dec 10^th, 2009, 10:17pm:

2. If you have compared Silicon vs CAD results, which is a better approximation to actual measurements out of these ----->

It depends on whether you measure L_SSB(f) or S_Φ(f)/2.
Simple measurements using Spectrum Analyzer give L_SSB(f) not S_Φ(f)/2.

Mayank wrote on Dec 10^th, 2009, 10:17pm:

1. Augmented Pnoise ??

I don't use Cadence Spectre as RF Reference Simulator.
Due to lack of Agilent licenses, I often have to use Cadence Spectre reluctantly
http://www.designers-guide.org/Forum/YaBB.pl?num=1227668750/8#8

Of course, I don't trust Augmented Pnoise at all.

Mayank wrote on Dec 10^th, 2009, 10:17pm:

2. S_Φ/2 below linewidth & L_Φ after Linewidth ??

How do you define "linewidth" in your results of http://www.designers-guide.org/Forum/YaBB.pl?num=1050465395/17#17 ?
From very suspicious results of Augmented Pnoise ?

Modeling as Lorentzian Spectrum is not proper if 1/f noise is present.
http://www.designers-guide.org/Forum/YaBB.pl?num=1050465395/20#20

Mayank wrote on Dec 10^th, 2009, 10:17pm:

3. Only L_USB plot of vanilla Pnoise ??

In Cadence Spectre, L_SSB(f) is completely same as S_Φ(f)/2 although they are physically and conceptually different.

You may trust "3" less than -10dBc/Hz region where assumption of small signal noise seems to be valid.

Title: Re: phase noise of ring oscillator using spectre
Post by Mayank on Dec 11^th, 2009, 1:22am

Pancho,
Also pls have a look & reply to my recent post...It's a related one :---
http://www.designers-guide.org/Forum/YaBB.pl?num=1260521995

There i had a query regarding obtaining true Sphi plots in cadence....

regards & thanx,
mayank.

Title: Re: phase noise of ring oscillator using spectre
Post by Andrew Beckett on Jan 4^th, 2010, 3:58am

Finally had the time to read through this thread properly.

First of all, some comments (in no particular order):

I believe in most simulators, the phase noise at low offsets is problematic, because it breaks the small-signal limitations of the noise analysis. In spectre, we tell you this frequency (the "line width") where the results become non-physical.
The "augmented" mode in pnoise was originally introduced to add an improved algorithm for cases where there is a long time constant in the circuit - in such cases, the previous algorithm could sometimes incorrectly include the noise contributions from the long-time-constant parts of the circuit (things like the bias circuitry in an oscillator). This is not a universal problem, and most customers have seen simulation results that match with measure data.
The "augmented" mode in shooting pnoise in particular had some problems initially, but as far as I'm aware, there aren't outstanding problems.
The "augmented" mode in hb pnoise is in fact now used all the time (for the algorithm).
Augmented is not enabled by default for shooting, because it's actually more costly to compute, and isn't always needed.
The traditional "sources" mode in pnoise doesn't really compute phase noise (although the the direct plot form calls it that) but instead plots the output noise in dBc (so it's just the output noise divided by the magnitude of the carrier) - which is often close to the phase noise for many oscillators. The "modulated" mode does the separation of AM and PM noise by simulating both sidebands around the carrier, and computing correlations between the two sidebands.
Augmented (with HB) flattens the phase noise curve below the line width frequency. Strictly this is not really anything to do with the augmented algorithm, but because we had a request that the phase noise over 0dBc should not be displayed. You can argue too that the flattened phase noise is also not physical - so perhaps this is where Pancho Hideboo's "suspicion" comes from? I've certainly discussed with R&D that we need to be very careful about displaying results which are not physical however we choose to do it.
Augmented (with shooting) doesn't do this flattening - but I don't think that means that the low-frequency results are really better or worse - they're just displayed differently.
I believe other simulators use some sort of heuristics to shape the phase noise at low offsets. I'd like to hear if others are really "accurate" at such low offsets.

Please note that none of the above is any kind of official "Cadence" statement. This is just my personal take, and I am stating these things in the interest of trying to clarify what is going on, as well as trying to understand what is missing or wrong with SpectreRF's pnoise (bearing in mind that it is seen as accurate for the majority of our customer base, despite what some posters would say ;-> )

Kind Regards,

Andrew.

Title: Re: phase noise of ring oscillator using spectre
Post by pancho_hideboo on Jan 4^th, 2010, 6:40am

Andrew Beckett wrote on Jan 4^th, 2010, 3:58am:

Augmented (with HB) flattens the phase noise curve below the line width frequency.
Strictly this is not really anything to do with the augmented algorithm,
but because we had a request that the phase noise over 0dBc should not be displayed.
You can argue too that the flattened phase noise is also not physical
- so perhaps this is where Pancho Hideboo's "suspicion" comes from?

No.
Again compare the followings.
http://www.designers-guide.org/Forum/YaBB.pl?num=1050465395/15#15
http://www.designers-guide.org/Forum/YaBB.pl?num=1050465395/17#17

Title: Re: phase noise of ring oscillator using spectre
Post by Andrew Beckett on Jan 4^th, 2010, 8:33am

I agree, something looks strange in those waveforms, but I'd like to know more about how those simulations were set up. Maybe something is wrong is in the setup, maybe it's a bug. In general I don't see behaviour like those plots are showing.

Mayank, is it possible to send me a testcase showing the problem? I'll send you a personal message with my Cadence email address.

Regards,

Andrew.

Title: Re: phase noise of ring oscillator using spectre
Post by Mayank on Jan 20^th, 2010, 4:03am

Hello,

@ Andrew :
Quote:

On one of your points, Augmented (with HB) flattens the phase noise curve below the line width frequency. Strictly this is not really anything to do with the augmented algorithm, but because we had a request that the phase noise over 0dBc should not be displayed. You can argue too that the flattened phase noise is also not physical - so perhaps this is where Pancho Hideboo's "suspicion" comes from? I've certainly discussed with R&D that we need to be very careful about displaying results which are not physical however we choose to do it.

I did some MATLAB post processing on the phase noise plot obtained from pnoise sources without augmented.
---> I clipped the plot above linewidth to Fosc / 2 and imported in matlab.
---> I extrapolated the curve using linear extrap in interp1 command of MATLAB....
---> I got a plot looking quite similar to the one shown with an augmented pnoise !!! i am referring to the flattening below line-width...

So, do augmented pnoise option just extapolates the curve below line-width freq. ?

Of course, these flattened values do not have any correlation, whatsoever, with actual pnoise values in the region below line-width. I think we can include a disclaimer for this in the cadence documentation. :D

@ everyone : The phase noise near the carrier frequency is almost Gaussian, instead of Lorentzian. I dont know, but i honestly wanna ask, is it that tough to model this gaussian noise in simulators that it has yet not been implemented successfully in any simulator till date ??

--
Mayank.

Title: Re: phase noise of ring oscillator using spectre
Post by pancho_hideboo on Jan 20^th, 2010, 5:11am

Mayank wrote on Jan 20^th, 2010, 4:03am:

So, do augmented pnoise option just extapolates the curve below line-width freq. ?

How does Cadence Spectre determine this "linewidth" ?