a basic doubt abut deembedding

[abhi_kolkata]

shift in reference planes lossy lines

hi all,

i really want to know why the simulators (like IE3D or some others also) use deembedding schemes for the ports.what they r meant for is not clear to me.

could anybody plz explain this to me!!

waiting for replies.

with regards

abhijit

 

[kleinesmurf69]

microwave deembedding

hello,

if I am correct, deembedding is shifting the reference plane of the phase of the applied signals.
this is used for instance in antennas when using a wavequide kind of port. with deembedding you can shift the reference of the signals, and thus the calculated S-parameters, to let's say a more useful position than where the sigals are applied in the (sub-)model.

best regards

 

[plasma]

It useful for calculating the input impedance of the antenna. Cause you can't put
port. at input of the antenna. you can get worng impedance.

 

[jian]

Hi, Plasma: De-embedding is related to how people treat microwave netoworks. It should be a long story. Traditionally, a microwave network is the cascading of transmission lines (TLNs) and discontinuities (Ds):

p1=TLN1==(D1)==TLN2==(D2)==TLN3==(D3)==TLN4==p2

Microwave designers want to model it as a set of components TLN1, D1, TLN2, D2, .... Each D is studied separately. For example, we try to study D1. We may add some additonal lengths at the ports lead to TLN1 and TLN2. Adding additional lengths are required in order to remove the higher order modes. If TLN1 and TLN2 are closed waveguides, people even assume they are semi-infinite long in theoretical and numerical analysis. In this way, they can model D1 with its 2 ports very close to the reality except the lengths of the additional TLNs can be different. For closed waveguides, the lengths of the additional TLNs are not critical as long as they are not significantly smaller than a wavelength because higher order modes get decayed very fast. In order to make the analysis consistent, people try to define the reference planes at the junctions of the D's and the TLN's.

For modern planar circuit applications, it really is not necessary to define the reference planes at the junctions of the D's and the TLN's. In fact, I do suggest users to try model the sub-circuits as:

p1=TLN1=(D1)=half_TLN2=

=half_TLN2=(D2)=half_TLN3=

=half_TLN3=(D3)==TLN4==p2

In this way, we really don't need to do the shifting of reference planes. For traditional microwave circuits, both ways should yield same results ideally. For RFIC circuit, the 2nd way should yield more accurate results because it does not need to shift reference plane. As I have mentioned before in early posting, any TLN with loss in the transverse direction does not follow waveguide theory precisely. Applying waveguide theory to it is an approximation. Therefore, shifting reference plane in RFIC is also an approximation. If the shifting is short, you may not see significant error. However, of the TLN is very lossy and the shifting is long, you may introduce numerical error into it. This is also related to the topic of complex Zc. Please try to find my comment on complex Zc. Using complex Zc for long shifting of reference plane may introduce significant numerical error. Best regards.

 

[rautio]

An exact (to within numerical accuracy) deembedding is completely described in the following paper:

Unification of double-delay and SOC electromagnetic deembedding
Rautio, J.C.; Okhmatovski, V.I.;
Microwave Theory and Techniques, IEEE Transactions on
Volume 53, Issue 9, Sept. 2005 Page(s):2892 - 2898

If you do not have access to IEEE Xplore, I will email you a copy of the paper.

One type of excitation for EM analysis is a voltage source impressed across an infinitesimal gap. This is the only type of excitation used in Sonnet (I work for Sonnet). The gap has fringing fields around it. These fringing fields result in extra capacitance. The SOC deembedding exactly characterizes and removes that capacitance. It also optionally shifts the reference plane to where you want "virtual" ports for your resulting data. Using the SOC, again, this reference plane shift is exact, even for highly lossy lines and very complex Zo, see the example (Fig. 5) in the above paper.

EM deembedding is the same thing as deembedding (i.e., calibrating) a physical measurement. You do not want the S-parameters of your feed structure included in the measurement. The ANA calibration characterizes and removes the feed structure. In EM, deembedding does exactly the same thing.

For shielded analysis, shifting the reference plane using SOC is exact (again, to within numerical precision) for all single mode transmission lines, both lossless and lossy. Transmission line theory is completely valid for lossy lines. We use it extensively directly because it is exact. In fact, transmission line theory for lossy transmission lines is how we form our Green's function for shieleded analysis when we have lossy substrates. In a typical EM analysis, we use lossy transmission line theory literally millions of times. It is exact. Suggestions that it is approaximate are simply wrong.

While the deembedding is exact for all single mode transmission line lengths longer than the fringing field extent, the calculation of Zo is not. If the through line (used in the deembedding) is very very close (<1 degree) to within an integral mulitple of a half wavelength, then the Zo calculation has increased error. This increased error is much larger for unshielded EM analysis, see Fig. 6 in the above paper. One reason for this increased error is explained in the paper.

I will not challenge statements that unshielded EM analysis deembedding is approaximate. But lossy transmission line theory and shielded EM analysis deembedding are both exact to within numerical precision. We have performed extensive numerical tests that show this is the case. Many of these tests have been published. I can describe some and post referneces if they are of interest. Most of the tests are very easy to duplicate on any EM analysis.

If anyone wants to say they can't figure out the exact answer, that is no problem. It is not OK to say that the exact answer does not exist, because it does (and it is described in the above paper).

 

[mwmmboy]

I think there are two main reason for deembedding in EM simulators:
1)Eliminate the discontinuity at the port when a port is close to the bondary.
2)to behave like a network analizer..

bye

Added after 2 minutes:

I think there are two main reason for deembedding in EM simulators:
1)Eliminate the discontinuity at the port when a port is close to the bondary.
2)to behave like a network analizer..

bye

Added after 35 seconds:

I think there are two main reason for deembedding in EM simulators:
1)Eliminate the discontinuity at the port when a port is close to the bondary.
2)to behave like a network analizer..

bye

 

[rautio]

Just got some new results in. Remember how Zo calculation fails (but the deembedding is still valid) when the through line used for deembedding calibration is within +/- 1 degree of a multiple of a half wavelength? Turns out our development guys just happened to be investigating this in detail in the last couple days. They found it is more like +/- 0.75 degrees, except at low frequency.

As we approach zero frequency, for reasons not yet understood, it fails when the through line is less than about 0.01 degree (yes, that decimal point is correct!), not 0.75 degree. Sonnet stops displaying Zo at about 1 degree with the informational message that the line is too short. We will be changing that in the next release, so you can get good Zo data all the way down to 0.01 degree long. Neat, huh?

Can anyone else do that?