de-embedding of inductor S-par in EM simulation

EDAboard.com Forum Index -> RF, Microwave, Antennas and Optics -> de-embedding of inductor S-par in EM simulation

Author

Message

Alex_IC

Joined: 02 Jul 2007
Posts: 147
Helped: 12
Location: Russia

21 Aug 2008 11:25 a de-embedding algorithm for electromagnetics

I've heard that in EM simulation with HFSS some designers use de-embedding of an inductor, so they simulate open, short and DUT. Why nobody do the same in ADS Momentum? Can I improve ADS Momentum simulation results in this way?

RF-OM

Joined: 03 Aug 2007
Posts: 607
Helped: 113
Location: The Earth

21 Aug 2008 22:26 short open inductor de-embedding

As I remember ADS can do de-embedding, and many other RF simulators as well. I use LINC2 where are two ways to do it. 1st is de-embedding calculator and the 2nd is to use input and output circuit physical lines models. For example, you have DUT and input and output transmission lines on the test board. It is easy to model these lines with physical models using negative lengths and program will automatically de-embed S-parameters for DUT.

Alex_IC

Joined: 02 Jul 2007
Posts: 147
Helped: 12
Location: Russia

22 Aug 2008 7:37 inductor de-embedding momentum

Yes, you are right here. But I mean emulating real de-embedding (with short and open). Because the de-embedding you told possible in HFSS also,actually it is reference plane moving.
Why they use open, short instead of using simple reference moving?

rautio

Joined: 21 Feb 2005
Posts: 297
Helped: 23
Location: North Syracuse, NY

23 Aug 2008 3:03 linc2 standart rf download

Sonnet has used what is equivalent to short open calibration since 1991, and it turns out that doing the calibration this way is perfect (to within numerical precision provided port connecting lines are not overmoded). Perfect calibration can only be done with shielded EM analysis. A recent overview for is in

J. C. Rautio and V. I. Okhmatovski, "Unification of Double-Delay and SOC Electromagnetic Deembedding," IEEE Tran. Microwave Theory Tech., Vol. 53, No. 9, pp 2892–2898, Sep. 2005.

Note that Fig. 6 in the above paper illustrates why unshielded tools can not do perfect calibration (but they can do much better than what they presently do...)

If you want to see what perfect calibration means for design, check out the upcoming December issue of the IEEE Microwave Magazine. Some details are also given in:

J. C. Rautio, “Perfectly calibrated internal ports in EM analysis of planar circuits,” in IEEE MTT-S IMS Digest, (Atlanta, GA), THP1D-01, June 2008.

Basically, perfect calibration, combined with two new design methodologies, is reducing the design cycle by about 10x to 100x, and typically requiring no more than two EM analyses, and often just one, for design closure to be achieved. This is a total revolution, and people are slowly starting to realize it.

The perfect calibration has also been adapted to internal ports. See, for example:

J. C. Rautio, “Deembedding the effect of a local ground plane in electromagnetic analysis,” IEEE Transactions on Microwave Theory and Techniques, Vol. 53, pp. 770–776, Feb. 2005.

The original calibration (which uses an L and a 2L length standard, but is mathematically equivalent to having a perfect short and open circuit standard) is:

J.C. Rautio, “A de-embedding algorithm for electromagnetics,” Int. J. Microwave & Millimeter-Wave Computer-Aided Eng., vol. 1, pp. 282–287, July 1991.

I will email copies of these papers to anyone that wants them. I do not post them here due to copyright. I will be slow in answering any posts/emails for the next week.

RF-OM

Joined: 03 Aug 2007
Posts: 607
Helped: 113
Location: The Earth

24 Aug 2008 19:31 de embedding in ads

I agree with rautio. Also, you need to be very careful with HFSS, often this program give wrong results even for simple cases.

To rautio,

Could you please upload these articles?. Unfortunately my MTT archive stops on year 2004.

Google
AdSense
Google Adsense

24 Aug 2008 19:31 Ads

rautio

Joined: 21 Feb 2005
Posts: 297
Helped: 23
Location: North Syracuse, NY

31 Aug 2008 17:35 open load short deembedding ads

Hello RF-OM -- I just returned from one week in India, my first trip there and it was an amazing experience...but that means I am a little slow in replying, sorry.

Several of the papers are posted on the Sonnet web site,

http://www.sonnetsoftware.com/support/techdocs.asp

Look at item 6 on that page. I will post the rest of the papers listed above at that web site link sometime during the next week.

-- de AJ3K

rautio

Joined: 21 Feb 2005
Posts: 297
Helped: 23
Location: North Syracuse, NY

04 Sep 2008 19:58 sonnet inserire interne calibrate porte

Hi RF-OM -- I have added two papers to our web site that were missing. If there is anything else you want, let me know. 73.

loucy

Joined: 26 Aug 2001
Posts: 432
Helped: 10

04 Sep 2008 20:53 sonnet editor xgeom download

"Why they use open, short instead of using simple reference moving?"

De-embedding is much more than simply moving the reference plane. De-embedding in numerical simulation attempts to remove the effect of higher order modes that are being excited by the "imperfect" excitation scheme. When moving the reference plane, it is assumed that there is only one mode propagating--the focus is to find out what that mode is (its characteristic impedance and propagation constant). The latter job is only part of the difficulties in de-embedding.

"Can I improve ADS Momentum simulation results in this way?"
What is your OPEN/SHORT standard? Is it really OPEN or SHORT? Although ADS Momentum doesn't claim to have a "perfect" de-embedding algorithm, it is unlikely you can improve on the "accuracy" by some sort of a customized "OPEN/SHORT" scheme that you can come up in a short time. (I say short time because if you can spend a huge amount of time implementing your own MOM code with some completely different de-embedding scheme, then you might get a "better" result.) Notice that there is no universal standard for "accuracy" (for the de-embedding scheme). Many people define it as closeness to their measurement result. But unfortunately, the calibration in the real measurement is often much more difficult than the de-embedding.

By the way, the calibration in real mesurement is always based upon some prior error model. There is no "perfect" error model (if it is perfect, people wouldn't call it an error model.) So I don't see how there can be a "perfectly calibrated port"--perfect with respect to what? I will be satisfied if the simulated results has certain correspondence with the measurement that is consistent regardless of the specific geometry parameters. I haven't seen any "perfect" agreement between simulation and measurement.

rautio

Joined: 21 Feb 2005
Posts: 297
Helped: 23
Location: North Syracuse, NY

06 Sep 2008 22:56 Re: de-embedding of inductor S-par in EM simulation

Hi Loucy -- I agree that perfect calibration is not possible for the general case in measurement in typical practice. However, perfect calibration (i.e., to within numerical precision) is possible in EM analysis. I appreciate your sharing your professional skepticism with us. Allow me to offer quantitative proof.

Go to http://www.sonnetsoftware.com/support/techdocs.asp and down load the paper

J. C. Rautio, “Perfectly calibrated internal ports in EM analysis of planar circuits,” in IEEE MTT-S IMS Digest, (Atlanta, GA), THP1D-01, June 2008.

(which I mention above). Look at Fig. 2 (attached to this posting).

We take a 6 resonator filter and split it in two, analyze each half, then connect them back together. That result is compared to the EM analysis of the entire filter. The results are identical. If the port cEalibration is approximate in any way, then this experiment fails. It fails in all other EM tools tried to date. In most tools, it fails dramatically. Only in Sonnet does this experiment work. If you want the Sonnet file for this filter, download SonnetLite, click on Help->Examples->Filters. Click on the hairpin filter, then click on Load into Xgeom (our layout editor). You will need the full version of Sonnet to do this experiment in Sonnet. However, The layout from Sonnet can be easily transferred to ADS if you want (Help->Manuals->Translators).

To do the experiment in the full version of Sonnet, you draw the diving line, divide the circuit, then click analyze (three menu picks total). In all other tools, you must divide the circuit manually and draw a schematic connecting them back together. Takes a few minutes anyway. Try it. Anyone can do this experiment in any tool. The experiment succeeds only if you have perfect port calibration. To date, we have seen the experiment succeed only in Sonnet.

Feel free to share your results with us. Perfect calibration can be done because that is exactly what we have done. And this changes everything.

Sorry, but you need login in to view this attachment

RF-OM

Joined: 03 Aug 2007
Posts: 607
Helped: 113
Location: The Earth

09 Sep 2008 18:20 Re: de-embedding of inductor S-par in EM simulation

To rautio,

Thank you very much Dr.Rautio! I found these papers and downloaded them. Unfortunately I was too busy these days with my last paper and had no ability to check EDA site earlier.

Best regards,
RF-OM

loucy

Joined: 26 Aug 2001
Posts: 432
Helped: 10

16 Sep 2008 16:23 Re: de-embedding of inductor S-par in EM simulation

Hello, Dr. Rautio,

I am afraid the hairpin example you provided above is not sufficient for the readers here to understand what is the "perfect calibration" defined in SONNET. In this particular example, there are several well defined modes at the dividing line, and the coupling between the two sections are through these modes. This is the reason one should obtain almost the same result whether analyzing the whole filter as one circuit or doing it as a cascade of two circuits. If we don't break all the arms at the same time, or not by the same horizontal line (i.e. break the arms at different vertical positions), then this cascading technique will not work.

Now by your description, SONNET's perfectly calibrated internal port seem to be much more general (e.g. it doesn't seem to care if there are highly irregular geometries in the immediate neighbor of the port.) If this is really the case, then one can devide the geometry quite arbitrarily, and indeed you have a very powerful tool for handling large complicated circuit. Unfortunately, it seems too good to be true at this point.

rautio

Joined: 21 Feb 2005
Posts: 297
Helped: 23
Location: North Syracuse, NY

17 Sep 2008 3:38 Re: de-embedding of inductor S-par in EM simulation

Hi Loucy -- Yes indeed it does seem to be too good to be true. And if you violate the underlying assumptions, it will indeed fail.

For the split hairpin filter, it works perfectly with Sonnet ports, for the reasons you correctly specify. It fails with all other EM tools that we have seen it tried on. Actually, you are more advanced than most microwave engineers because most engineers think it can not possibly work with any EM tool.

Splitting a filter in two will fail even in Sonnet if it is not done correctly. This is because EM coupling between the two halves is not included in the overall result. For the hairpin split as shown, this is not a problem. If however, you split the hairpin vertically (instead of horizontally) between two resonators, it fails with S21 = - infinity dB. If you split it vertically between the legs of a hairpin, then the coupling between the two legs is not included. So while the port calibration is perfect, there is still error introduced if the splitting is not done correctly.

For the perfectly internal ports, any single group of ports must all be on the perimeter of a rectangle (this is due to the way the internal port calibration is done, the perfectly conducting sidewalls are used to set perfect short circuit reference planes on the interior of the box, thus a rectangle). If there is important coupling between the sub-circuit that you connect to the internal ports, and the rest of the circuit, then that coupling is not included. Likewise, if there is another group of internal ports too close, stray coupling between the groups of ports is not removed.

So, it is still possible to fail. The big news here is that failure is no longer due to approximate port calibration. The port calibration error (provided the port connecting lines are not over moded!) is zero (i.e., nothing but numerical noise left).

So try the hairpin filter example in your favorite EM tool, any EM tool. Then try it in Sonnet. Look at the results. Sonnet wins.

After you become convinced that Sonnet's ports (properly applied) are indeed perfectly calibrated, there are some amazing things that can happen. After you have run a couple more numerical experiments that convince you that Sonnet's ports are indeed perfect, then let me know and we can discuss this further. If you continue to have doubts, please feel free to speak up, but please do do the suggested numerical experiments first. They are a real eye opener.

P.S. I am traveling in very remote areas for the next two weeks. I will have only occasional access to the internet.