The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Design >> RF Design >> IM3 sim vs. lab
https://designers-guide.org/forum/YaBB.pl?num=1349805061

Message started by kelly on Oct 9^th, 2012, 10:51am

Title: IM3 sim vs. lab
Post by kelly on Oct 9^th, 2012, 10:51am

Hi,

I have been having problem matching simulated IM3 with lab data for a up-convert passive mixer+ buffer combo. I am consisteantly seeing ~10dB differernce sim vs. lab (with sim being better of course). Here are my questions:
1. for RF signal around 2 - 3G, what is an resonable expectation of matching the sim vs. lab for IM3? 3db? 5db? I assume 10db is just too much.
2. does the matching expectation decrease as the IM3 performance improves? i.e., for target IM3 performance =-40dBc, the matching should around 3dB, for IM3=-60dbc, the matching is relaxed to 5dB? Is there a such rule of thumb?
3. does down convert mixer have a better chance of matching sim vs. lab than up convert mixer?
4. how about active mixer vs. passive mixer? i.e., does one expect to have better sim vs. lab matching for a active mixer than a passive mixer?

Any help/insight/sugestions/comments are greatly appreciated.

Thanks.
Kelly

Title: Re: IM3 sim vs. lab
Post by aaron_do on Oct 9^th, 2012, 5:42pm

Hi,

are you using the BSIM model? If so don't expect good correlation for operating regions around VDS = 0. The results will be way off.

Aaron

Title: Re: IM3 sim vs. lab
Post by kelly on Oct 9^th, 2012, 9:05pm

hi Aaron,

yes I am, but I ran transient with Two tone input not pss and qpss. So won't my sim results still be ok?

thanks.
Kelly

Title: Re: IM3 sim vs. lab
Post by aaron_do on Oct 10^th, 2012, 2:56am

Its not a simulation issue, its a model issue. Simulation results will be wrong. There are threads on this forum regarding that problem. I know cos I asked the same question a while back (don't remember when). If you can get your hands on the PSP model it more or less solves this issue.

regards,
Aaron

Title: Re: IM3 sim vs. lab
Post by rfidea on Oct 10^th, 2012, 11:32am

The model issue Aaron is discussing is that the Id vs Vds has a discontinuity around Vds=0 in the BSIM3 model. The second derivative is not continous. That makes the IM3 simulation of a passive mixer go bananas.

Title: Re: IM3 sim vs. lab
Post by kelly on Oct 10^th, 2012, 12:34pm

Hi Aaron and rfidea,

yes, I remember reading the thread regarding this. I thought by using transient (since it's only using I vs. Vds, unlike the pss needs the diravitives info), is a way to get around the bsim model problem. In another words, if I ran transient and take dft and look at IM3 vs Pin, if it follows the 3 :1 slope, shouldn't I be ok (I am not using the =qpss IM plot to extrapolate the IIP3 and things). I calculate OIP3=IM3/2+Pout by measuring IM3 from the dft plot.

That was what I got from reading thread back then and thought transient sim will not affected by the bsim model deficiency. Maybe I am wrong??

thanks.
Kelly

Title: Re: IM3 sim vs. lab
Post by kelly on Oct 10^th, 2012, 1:09pm

Here is a link from cadence website on how to bypass the bsim model problem for passive mixer, it has been a while, I'll read it again (as soon as I can remember my password)

http://support.cadence.com/wps/mypoc/cos?uri=deeplinkmin:ViewSolution;solutionNumber=11454704

thanks.
Kelly

Title: Re: IM3 sim vs. lab
Post by kelly on Oct 10^th, 2012, 1:31pm

actually, my apology, the article is just stating whar you guys are saying, no mention about the transient method.

My understanding is the pss depends on the smoothness of the derivatives, but not transient, so reansient result should be ok. Is my understanding wrong? (I don't have psp availble to me that's why I am using the go around)

Thanks.
Kelly

Title: Re: IM3 sim vs. lab
Post by Ken Kundert on Oct 10^th, 2012, 1:50pm

You seem to have a fundamental misunderstanding about simulation. Modeling approximates the circuit to create a system of equations that can be simulated. From there the simulator uses numerical methods to approximate the solution to the system of equations. If you make a mistake in the modeling and so create the wrong system of equations, there is nothing you can do in the analysis method to compensate from the fact that your are solving the wrong system of equations.

Perhaps the word 'discontinuity' is confusing you. Sometimes convergence problems are created by discontinuities in the derivatives of the models. That is not the issue here. In this case the discontinuities are in currents, not in the derivatives, and the problem is not a convergence issue, it is the fact that the model does not accurately represent the behavior of the circuit when Vds is close to zero.

-Ken

Title: Re: IM3 sim vs. lab
Post by RFICDUDE on Oct 10^th, 2012, 6:50pm

Hi Kelly,

I agree with the other comments about the BSIM3 and BSIM4 model issue regarding discontinuity when Vds=0 for passive MOSFET switches.

One possible way around the model discontinuity issue is to intentionally force Vds not to be equal to zero when the signal passes through a zero crossing. This means you have to intentionally draw a DC current through each switch when it is ON to force Vds to be big enough that the signal swing never causes Vds to pass through zero.

The better alternative is to use PSP models if they are available.

For future processes, BSIM6 implements a solution to the problem.

Title: Re: IM3 sim vs. lab
Post by kelly on Oct 10^th, 2012, 7:33pm

Hi all,

thanks for all your comments, so I guess I did misunderstand the problem. what you are saying is you can't trust the simulation results even from transient analysis if the BSIM model is used. May I ask, when you used the correct model (such as PSP), do you get pretty close agreement? what's the discrepancy between sim vs. lab typically, within couple dBs ?
Thanks.
Kelly

PS, I'll try to to see whether I can force the small corrent though the swtiches.....

Title: Re: IM3 sim vs. lab
Post by RFICDUDE on Oct 11^th, 2012, 5:55pm

My experience is that the PSP model yields reasonable correlation with lab measurements where as the BSIM3/4 models do not.

Although, while the PSP models yield physical results, I have seen measurements that show measured performance better than what the PSP models predict the IM3 to be. I must admit that I have not closed the loop on the discrepancy (i.e. make sure it is not a measurement issue), but the difference did surprise me.

Title: Re: IM3 sim vs. lab
Post by aaron_do on Oct 11^th, 2012, 7:45pm

My "limited" experience is that your correlation is only going to be as strong as the weakest link. If your RF modeling is off, then IM3 can be off too if the circuit is very "RF sensitive". Also, I see a lot of discrepancy when the transistors are OFF. I guess the model accuracy is not so good (PSP or BSIM) in this region.

cheers,
Aaron

Title: Re: IM3 sim vs. lab
Post by RFICDUDE on Oct 11^th, 2012, 8:19pm

Hi Aaron,

Your post implies the "off" state linearity is the weakest link in modeling linearity?
Can you please elaborate or clarify this issue?
I have not experienced the "off" state linearity of MOSFET switches limiting the linearity performance, so I'm curious to hear more about this issue.

Thanks in advance!

Title: Re: IM3 sim vs. lab
Post by aaron_do on Oct 11^th, 2012, 11:02pm

Hi RFICDUDE,

sure. Obviously its application dependent, and in the design of high throw-count RF switches, the off-transistor nonlinearity is important. The RF switch is the last active circuit (sort of quasi-passive) before the antenna, so the HD from the switches cannot be filtered (that's only partially true). Therefore, the HD requirements for RF switches can be quite stringent.

regards,
Aaron

Title: Re: IM3 sim vs. lab
Post by kelly on Oct 12^th, 2012, 9:12am

Hi RFICDUDE,

Do you mind giving me a ball park number for the situation you are describing?
In the case where PSP yeilds better correlation, is the sim still better the lab? By how much > or < than 3 or 5dB?
In the case where measurements are better than the lab, is the differernce around 5dB or more? Are you surprise by the fact that lab is better than the sim or the differernce ?
Also, when you said the bsim doesn't correlate well, did you just judge that by looking at the discrepancy between sim and lab (I assume the sim is a lot better than the lab)? was the discrepancy mucher larger than 5 or 10dB?

If any of this is sensitive information, just let me know. I just try to get a sense on what kind of the correlation I can expect for a up convert passive mixer design.

(I also have a down convrt passive mixer, the discrepancy was not that bad even though bsim was used. But I think that's probabily because mixer was not the dominant source in that line up since I couldn't measure mixer by itself).

Thanks very much for your help.
Kelly

Title: Re: IM3 sim vs. lab
Post by RFICDUDE on Oct 13^th, 2012, 5:35am

Aaron,
I could see where front-end switches could have issues in what should be an "off" state (thanks).

Kelly,
No I really don't have hard numbers for the discrepancy.
Moreover, my discrepancy may have no correlation to yours (different models, different circuits, etc.).

The only way to really understand if there is a modeling issue is to have test structures in silicon where you can specifically measure the linearity of the device and compare it the models for the exact set of conditions.

Most of the time there is no time to perform such a correlation exercise, so if there are discrepancies then one has to rely on making relative changes to correct the offset (if the offset causes worse than expected performance).

Title: Re: IM3 sim vs. lab
Post by kelly on Oct 15^th, 2012, 9:16am

Hi RFICDUDE,

yes, understood, thanks very much.
Kelly