The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> RF Simulators >> Augmented Pnoise of Cadence Spectre
https://designers-guide.org/forum/YaBB.pl?num=1241894034

Message started by pancho_hideboo on May 9^th, 2009, 11:33am

Title: Augmented Pnoise of Cadence Spectre
Post by pancho_hideboo on May 9^th, 2009, 11:33am

Cadence claims about "Augmented Pnoise" of Cadence Spectre like following.

Quote:

We have improved Pnoise analysis for autonomous circuits with a new "augmented" mode.
Augmented Pnoise will give more accurate results for oscillators with large time constants.
In augmented Pnoise, the frequency-aware PPV method (rather than Floquet deflation) is used.
The new method gives a Lorentzian plot.
However, the price you pay is slower simulation speed.

http://www.cadence.com/Community/blogs/rf/archive/2008/07/29/why-do-shooting-and-harmonic-balance-phase-noise-results-differ.aspx?postID=10396
http://www.cadence.com/Community/blogs/rf/archive/2008/08/26/guidelines-for-simulating-oscillators-phase-noise-simulations.aspx?postID=10863
http://www.cadence.com/Community/blogs/rf/archive/2009/05/01/enhanced-pnoise-algorithm-to-compute-phase-noise-for-vcos-with-bandgap-voltage-reference.aspx?postID=17305

To my regret, I can't access Cadence Spectre which supports "Augmented Pnoise".

Is there anyone who compared results of "Augmented Pnoise" of Cadence Spectre with Agilent GoldenGate or ADSsim, especially regarding phase noise in very small offset frequency region ?

"Augmented Pnoise" of Cadence Spectre truely can give good results ?

Title: Re: Augmented Pnoise of Cadence Spectre
Post by ACWWong on May 12^th, 2009, 5:48am

We use "Augmented Pnoise" in our last design tape-out, it gave us more conservative (i.e. higher) estimation of phase noise than vanilla pnoise. This was especially true in our VCO (& VCO regulator) design at low kHz frequency offsets.

We did not benchmark with ADS or other tools, but hope to benchmark with silicon in the near future!

cheers
aw

Title: Re: Augmented Pnoise of Cadence Spectre
Post by pancho_hideboo on May 12^th, 2009, 6:41am

Thanks for comments

ACWWong wrote on May 12^th, 2009, 5:48am:

Vanilla Pnoise tends to give phase noise of more than 0dBc for very small offset frequency and not to show f^-3 region in phase noise spectrum plot clearly regardless of phase noise of more than 0dBc.

However according to your comments, "Augmented Pnoise" for your design does give higher phase noise for very small offset frequency than Vanilla Pnoise.

"Augmented Pnoise" doesn't give phase noise of more than 0dBc for very small offset frequency, although it gives higher phase noise than Vanilla Pnoise ?
It does result in a Lorentzian plot truely ?
If so, "Augmented Pnoise" has to suppress f^-3 region in phase noise.

ACWWong wrote on May 12^th, 2009, 5:48am:

We did not benchmark with ADS or other tools, but hope to benchmark with silicon in the near future!

I'm looking forward to your report.

Title: Re: Augmented Pnoise of Cadence Spectre
Post by rfmems on May 12^th, 2009, 8:47am

I thought augmented pnoise should give less noise compared to the old pnoise simulation for large rc. And it should give a lorentzian plot.

BTW, why it is called vanilla pnoise?

Title: Re: Augmented Pnoise of Cadence Spectre
Post by pancho_hideboo on May 12^th, 2009, 9:08am

rfmems wrote on May 12^th, 2009, 8:47am:

I thought augmented pnoise should give less noise compared to the old pnoise simulation for large rc.
And it should give a lorentzian plot.

What do you mean by "rc" ?
Expression of "old pnoise" is not proper, it should be "conventional" or "normal" Pnoise.

rfmems wrote on May 12^th, 2009, 8:47am:

BTW, why it is called vanilla pnoise?

It means "conventional" or "normal" Pnoise.

Title: Re: Augmented Pnoise of Cadence Spectre
Post by rfmems on May 12^th, 2009, 9:10am

rc means RC time constant.

Title: Re: Augmented Pnoise of Cadence Spectre
Post by rfmems on May 12^th, 2009, 9:12am

pancho_hideboo wrote on May 12^th, 2009, 9:08am:

rfmems wrote on May 12^th, 2009, 8:47am:

BTW, why it is called vanilla pnoise?

It means "conventional" or "normal" Pnoise.

Well, it seems that this forum improved my English as well.

Title: Re: Augmented Pnoise of Cadence Spectre
Post by rfmems on May 12^th, 2009, 9:22am

pancho_hideboo wrote on May 12^th, 2009, 9:08am:

Expression of "old pnoise" is not proper, it should be "conventional" or "normal" Pnoise.

Hi pancho_hideboo,

I think expression of "normal" is also not proper, because I won't call augumented pnoise abnormal.

Anyway, it seems to me that we both used improper English more or less. But we understood each other perfectly. Surprise!

have fun
rfmems

Title: Re: Augmented Pnoise of Cadence Spectre
Post by oermens on May 12^th, 2009, 11:48am

rfmems wrote on May 12^th, 2009, 9:22am:

pancho_hideboo wrote on May 12^th, 2009, 9:08am:

Expression of "old pnoise" is not proper, it should be "conventional" or "normal" Pnoise.

I think saying just vanilla is short for plain vanilla, like the ice cream flavour - something which is as simple and basic as you can get, no bells and whistles.

Quote:

A major use of vanilla is in flavoring ice cream. The most common flavor of ice cream is vanilla, and thus most people consider it to be the "default" flavor. By analogy, the term "vanilla" is sometimes used as a synonym for "plain".

http://en.wikipedia.org/wiki/Vanilla#Usage 8-)

Title: Re: Augmented Pnoise of Cadence Spectre
Post by pancho_hideboo on May 12^th, 2009, 8:26pm

oermens wrote on May 12^th, 2009, 11:48am:

I think saying just vanilla is short for plain vanilla, like the ice cream flavour - something which is as simple and basic as you can get, no bells and whistles.

Anyone can infer such a meaning of "Vanilla" easily.

rfmems wrote on May 12^th, 2009, 9:22am:

I think expression of "normal" is also not proper, because I won't call augumented pnoise abnormal.

If I see a dictionary, an antonym of "normal" is surely "abnormal".

However an antonym of "normal" is "special" or "enhanced" or "extented" for me.
It is not "abnormal" for me.
As another expression, I choose "Ordinary Pnoise".

"Vanilla Pnoise" is a expression which was introduced by ACWWong not me.
Should I use expessions such as "Chocolate Pnoise" or "Strawberry Pnoise" for expressing "Augmented Pnoise" ?

Title: Re: Augmented Pnoise of Cadence Spectre
Post by pancho_hideboo on May 12^th, 2009, 8:28pm

"Augmented Pnoise" of Cadence Spectre reminds me of "FM Noise analysis ability in phase noise evaluation of Agilent ADSsim which is based on DC operation points and not implemented in current ADSsim".
See pp.143-146 of http://cp.literature.agilent.com/litweb/pdf/ads2003c/pdf/cktsimhb.pdf
This "phase noise from frequency sensitivity analysis" was removed since ADS2004A.

If there is no 1/f noise parameter in model parameters of MOS or BJT,
there is no f^-3 region(Flicker FM Noise, -9dB/oct=-30dB/dec) in phase noise.
In this case, phase noise is naturally a Lorentzian plot.

But if there is 1/f noise parameter in model parameters, f^-3 region appears in phase noise.
In this case, phase noise is not a Lorentzian plot.

"FM Noise analysis ability in phase noise evaluation of Agilent ADSsim" gave this f^-3 region clearly.

Does "Augmented Pnoise of Cadence Spectre" underestimate or overestimate this f^-3 region ?
If "Augmented Pnoise of Cadence Spectre" claims to give a Lorentzian phase noise plot, it must give underestimated phase noise for f^-3 region.

Title: Re: Augmented Pnoise of Cadence Spectre
Post by rfmems on May 13^th, 2009, 2:58am

pancho_hideboo wrote on May 12^th, 2009, 6:41am:

Conventional pnoise/hbnoise shows clearly f^-3 region in phase noise, which gets non-physical when the offset are close to and below the line width frequency. Phase noise above 0dBc/Hz can only be observed at offset frequency below line width.

pancho_hideboo wrote on May 12^th, 2009, 6:41am:

However according to your comments, "Augmented Pnoise" for your design does give higher phase noise for very small offset frequency than Vanilla Pnoise.

According to my simulation, using hb/hbnoise for LC oscillators, the phase noise gets smaller for small offset (close and below line width) when augmented hbnoise is chosen. And phase noise clearly shows a Lorentzian plot.

However if shooting pss/pnoise is used, I still see above 0dBc/Hz phase noise and a non-Lorentzian plot. So I believe the augmented pnoise simulation is not working properly.

I heard that augmented pnoise is not a completed work and cadence is still releasing enhancemented versions. So the accuracy of simulation heavily depends on the MMSIM version you are using.

pancho_hideboo wrote on May 12^th, 2009, 6:41am:

If so, "Augmented Pnoise" has to suppress f^-3 region in phase noise.

I believe that is wrong, augmented hb/hbnoise simulation shows both a lorentzian plot and f^(-3) region.

Title: Re: Augmented Pnoise of Cadence Spectre
Post by pancho_hideboo on May 13^th, 2009, 4:00am

rfmems wrote on May 13^th, 2009, 2:58am:

Conventional pnoise/hbnoise shows clearly f^-3 region in phase noise

No.
If we compare phase noise results of f^-3 region between Agilent ADSsim and Cadence Spectre Conventional Pnoise, Agilent ADSsim often gives more matched results with actual measurement data.

rfmems wrote on May 13^th, 2009, 2:58am:

pancho_hideboo wrote on May 12^th, 2009, 6:41am:

However according to your comments, "Augmented Pnoise" for your design does give higher phase noise for very small offset frequency than Vanilla Pnoise.

Your tendency about "Augmented Pnoise" is different from one of ACWWong.

rfmems wrote on May 13^th, 2009, 2:58am:

However if shooting pss/pnoise is used, I still see above 0dBc/Hz phase noise and a non-Lorentzian plot.

"Augmented Pnoise" has been already implemented in Shooting-PSS/Pnoise ?
I think it is still only for HB-PSS/Pnoise.

rfmems wrote on May 13^th, 2009, 2:58am:

I heard that augmented pnoise is not a completed work and cadence is still releasing enhancemented versions.

Before completing or improving, this "Augmented" option in Pnoise might be forgotten as same as many imcomplete options of Cadence SpectreRF.

rfmems wrote on May 13^th, 2009, 2:58am:

pancho_hideboo wrote on May 12^th, 2009, 6:41am:

If so, "Augmented Pnoise" has to suppress f^-3 region in phase noise.

I believe that is wrong, augmented hb/hbnoise simulation shows both a lorentzian plot and f^(-3) region.

Based on "Lorentzian Spectrum", the spectrum is flat up to the 3dB-linewidth point, and roll-off at 20dB/decade thereafter.

I don't know definition of "Lorentzian Spectrum" which include f^-3 region.

Title: Re: Augmented Pnoise of Cadence Spectre
Post by rfmems on May 13^th, 2009, 6:26am

pancho_hideboo wrote on May 13^th, 2009, 4:00am:

Based on "Lorentzian Spectrum", the spectrum is flat up to the 3dB-linewidth point, and roll-off at 20dB/decade thereafter.

The full spectrum of the oscillator has the shape of a Lorentzian around the carrier, and
away from the carrier, the white-noise sources contribute a term
that has a f^-2 frequency dependence, and the flicker noise
sources contribute terms that have a f^-3 frequency dependence.

The complete phase noise spectrum is not Lorentzian, but the part around the carrier is.

That is what I meant by

rfmems wrote on May 13^th, 2009, 2:58am:

augmented hb/hbnoise simulation shows both a lorentzian plot and f^(-3) region.

Title: Re: Augmented Pnoise of Cadence Spectre
Post by rfmems on May 13^th, 2009, 6:41am

pancho_hideboo wrote on May 13^th, 2009, 4:00am:

Your tendency about "Augmented Pnoise" is different from one of ACWWong.

Whether we were talking about the same thing depends on if ACWong compared his hb/hbnoise "augmented" to conventional hb/hbnoise or shooting pss/pnoise.

BTW, I got the feeling that you have strong faith in ADSsim. People with different experiences have different opnions. I am happy with Spectre so far.

Title: Re: Augmented Pnoise of Cadence Spectre
Post by pancho_hideboo on May 13^th, 2009, 6:59am

rfmems wrote on May 13^th, 2009, 6:41am:

BTW, I got the feeling that you have strong faith in ADSsim.
People with different experiences have different opnions.
I am happy with Spectre so far.

I'm now planning to move SmartSpiceRF due to cost reduction purpose.

I'm very enhancing in actual measurements maybe than you.
I've always said to people in this forum, You have to learn measurements using actual instruments. Not "EDA Tool Play".

Generally Cadence Tool Players have no experience of actual measurements.

I have Agilent ADSsim, GoldenGate and Cadence SpectreRF.
I've intensively compared simulation data with actual measurement data over long years since Agilent MDS and EEsof Series-IV Libra days when SpectreRF didn't exist.

In my country, Cadence Spectre is not used for RF circuit design mainly.
However only for VCO design, there are many people who are using Cadence Spectre.

Title: Re: Augmented Pnoise of Cadence Spectre
Post by pancho_hideboo on May 13^th, 2009, 7:03am

rfmems wrote on May 13^th, 2009, 6:26am:

The complete phase noise spectrum is not Lorentzian, but the part around the carrier is.

Again, I don't know definition of "Lorentzian Power Spectrum" which include f^-3 region.

(1) {1/(a²+f_offset²)} * {1+k₁/(b+f_offset)}

(2) {k₁/(a²+f_offset²)} + {k₂/(c³+f_offset³)}

Which is "Lorentzian Power Spectrum" including f^-3 region ?

Title: Re: Augmented Pnoise of Cadence Spectre
Post by rfmems on May 13^th, 2009, 7:49am

pancho_hideboo wrote on May 13^th, 2009, 6:59am:

I'm enhancing in actual measurements maybe than you.
I've always said to people in this forum, You have to learn measurements using actual instruments. Not "EDA Tool Play".

I am really curious to know what are you based on to make these claims. It may surprise you that I measure my chip myself as well.

pancho_hideboo wrote on May 13^th, 2009, 6:59am:

Generally Cadence Tool Players have no experience of actual measurements.

I guess lots of people in this forum disagree with you. Actually, I think nowadays most of the industries and academies using spectreRF for analog/rf designs. And most of the designers measure their critical block (if not the whole chip) themselves!

pancho_hideboo wrote on May 13^th, 2009, 6:59am:

I have Agilent ADSsim, GoldenGate and Cadence SpectreRF.
I've intensively compared simulation data with actual measurement data over long years since Agilent MDS and EEsof Series-IV Libra days when SpectreRF didn't exist.

It may surprise you again, I, who is not "enhancing in actual measurements" as you think, also use ADSsim. I said I am happy with Spectre based on my experience of simulation and measurement. Maybe there is some flaw of spectre which I have not discovered yet. But I said "so far"!

I thought you started this topic for discussion, not for confrontation. But I was wrong! You first pulled up my English, then outbursted to "command" me to learn measurement. Come on, this is a forum, not a colosseum!

Title: Re: Augmented Pnoise of Cadence Spectre
Post by rfmems on May 13^th, 2009, 7:54am

pancho_hideboo wrote on May 13^th, 2009, 7:03am:

Which is "Lorentzian Power Spectrum" including f^-3 region ?

I did not say "Lorentzian Power Spectrum" including f^-3 region, I said you can see both Lorentzian and f^-3 region. You understood me wrong.

I would suggest you to read Demir's paper "Phase noise and timing jitter in oscillators with colored-noise sources", if you still find it not crystal clear. (I won't say you don't understand phase noise, that is your style though).

Title: Re: Augmented Pnoise of Cadence Spectre
Post by pancho_hideboo on May 13^th, 2009, 8:06am

rfmems wrote on May 13^th, 2009, 7:49am:

I guess lots of people in this forum disagree with you.
Actually, I think nowadays most of the industries and academies using spectreRF for analog/rf designs.

Are you student or any relation with Cadence ?
I think almost all people except for a few people in this forum are students.

I think you just started to use SpectreRF since two or three years ago.
Did you resolve the following ?
http://www.designers-guide.org/Forum/YaBB.pl?num=1207830622

rfmems wrote on May 13^th, 2009, 7:49am:

And most of the designers measure their critical block (if not the whole chip) themselves!

No.
In old days, designers, especially RF designers were engaged in all tasks.

But nowadays, in most companies, works required for making IC are divided into small tasks.

architectual design, circuit simulation, physical layout design, verification, building block measurements, system evaluation, etc.

Nowadays, designers are engaged in limited tasks in most companies.

Indeed there is no stage for measurement/evaluation and even no feedback path from measurement in design flow of Cadence Top Down Methodology.

rfmems wrote on May 13^th, 2009, 7:54am:

I did not say "Lorentzian Power Spectrum" including f^-3 region, I said you can see both Lorentzian and f^-3 region.

Maybe you just mean some spectrum which have flat top in very small offset region.
Such flat top spectrum is natural result of actual non-linear effects.

Title: Re: Augmented Pnoise of Cadence Spectre
Post by rfmems on May 13^th, 2009, 9:15am

The conversation is getting harder and harder. Well, I didn't start it, but I will finish it!

If you are doing a PhD, you have to measure your chip all by yourself.
For small startup comapnies, more or less the same situation, you have to verify your own block.
In big companies, I worked in one (top 10 in IC industry), if I designed a critical RF block, I need to measure it for the first fun, see if it works properly, get the right settings and write a measurement proposal for the verification guy. Then he will do the rest, corners, temperatures, and so on.

So you see, almost every RF designer has some measurement experience.

Back to phase noise, I don't think you understand oscillator phase noise thoroughly. Even if flicker noise is presented, you will still have a Lorentzian near the carrier in the form of
c/(pi^2c^2+f^2). You definitely should read that paper!

Title: Re: Augmented Pnoise of Cadence Spectre
Post by pancho_hideboo on May 13^th, 2009, 9:20am

rfmems wrote on May 13^th, 2009, 9:15am:

For small startup comapnies, more or less the same situation, you have to verify your own block.

I also think so.

rfmems wrote on May 13^th, 2009, 9:15am:

So you see, almost every RF designer has some measurement experience.

I don't think so. There are many who have only simulation experiences.

rfmems wrote on May 13^th, 2009, 9:15am:

Even if flicker noise is presented, you will still have a Lorentzian near the carrier in the form of
c/(pi^2c^2+f^2).

Where is f^-3 ?
Do you mean (2) in the following ?

pancho_hideboo wrote on May 13^th, 2009, 7:03am:

(1) {1/(a²+f_offset²)} * {1+k₁/(b+f_offset)}
(2) {k₁/(a²+f_offset²)} + {k₂/(c³+f_offset³)}
Which is "Lorentzian Power Spectrum" including f^-3 region ?

Title: Re: Augmented Pnoise of Cadence Spectre
Post by rfmems on May 13^th, 2009, 9:35am

c/(pi^2c^2+f^2) f close to 0
a/f^2*(b+c/f) f>>0

Title: Re: Augmented Pnoise of Cadence Spectre
Post by pancho_hideboo on May 13^th, 2009, 9:42am

rfmems wrote on May 13^th, 2009, 9:35am:

c/(pi^2c^2+f^2), f close to 0

I think this is not proper if flicker noise is included.
See pp.12-14 of http://www.designers-guide.org/Analysis/PLLnoise.pdf

rfmems wrote on May 13^th, 2009, 9:35am:

a/f^2*(b+c/f), f>>0

Correct. This is modified Leeson's Equation.

{1/(a²+f_offset²)} * {1+k₁/(b+f_offset)}
or rather
{k₁/(a²+f_offset²)} + {k₂/(c³+f_offset³)}

Title: Re: Augmented Pnoise of Cadence Spectre
Post by rfmems on May 14^th, 2009, 1:19pm

The modeling of flicker noise differs with foundries. I have two PDKs at hand, one models the flicker noise with 10dB/dec down to 600uHz, then flats out. while the other keeps the 10dB slop even at 1a Hz.

Since the augmented hbnoise has no above 0dBc/Hz values, Spectre must have defined a bandwidth below which the flicker noise flats out. That is why simulation shows both Lorentzian (close to carrier) and f^-3 region.

Theoretically, if flicker noise is presented, oscillator noise will be unbounded and not a Lorentzian. But with the flicker noise bandwidth defined, "c/(pi^2c^2+f^2) f close to 0" is valid. If this bandwidth is chosen according to typical measurement (accumulation) time, then the simulation should have a good accordance with measurement. If measure with very long time scale, flicker noise will anyway become indistiguishable from temperature drift.

What is your opinion? pancho_hideboo.