EM模拟中的细胞数选项问题

你好RinC，> {quote：title = RinC写道：} {quote}>这里我有一个关于EM模拟中“细胞数/波长数”选项的问题。
>我已经阅读了有关EM模拟中Mesh选项的文档。
从文档中，随着细胞数量/波长的增加，您可以获得更准确的结果。
在这种情况下，我假设忽略模拟的时间，主要目的是模拟的准确性。
正确。
>我用这个选项进行了测试，这对我来说是一个问题。
>在我的工作中，50欧姆TL与RO4003c，导体厚度= 17um，sub。
厚度= 0.504 mm，2.45 GHz为1.12mm（这是LAB中的实际尺寸）。
根据这些信息，我将RO4003C的er调整为er = 3.62。
然后，我通过EM模拟模拟这个物理宽度。
> 1.case：80个细胞/波长：S11：-40dB> 2.case：180个细胞/波长：S11：-55dB>在两种情况下，它都是“匹配的”，180个细胞/波长给出更好的结果。
在这一步，您可以得出结论：80cells /波长就足够了，因为实际上对于非常好的校准，VNA只能得到-65dB的最佳结果（这是我们在LAB中得到的最好结果），所以我们无法区分这些差异。
正确。
无论我们的回波损耗是-40dB还是-55dB，都非常出色。
在实验室中很难*准确地*测量这么小的S11值。
>正如您所看到的，现在差异非常明显。
我们关心耦合分支的匹配条件（S33），耦合（S31）和隔离（S41）。
S31没有变化，但S33和S41变化很大。
对于180个细胞，S33是匹配的，但在我看来，80个细胞不匹配。
>所以，我的问题是，我们应该依赖哪些价值观？
>如果你需要一些检查，我附上了我的工作区有趣的耦合器结构，因为耦合是如此薄弱。
正如您已经发现的那样，如果您希望获得非常准确的回波损失结果，则默认网格密度为20cells /波长+边缘网格是不够的。
前向耦合更加稳健，并且随着网格化而略微变化。
现在真正困难的部分是隔离：因为前向耦合太弱，我们无论如何都有小电平，测得的S41是耦合器隔离和在负载反射回来的信号的混合（不匹配）。
这使得您的结果对设置/网格和端口校准中的最小变化非常敏感。
我查看了你的结果，并在Sonnet中验证了它们与另一个非常准确，非常密集的网格EM仿真。
* 180 *细胞/波长的动量结果与S11的Sonnet结果一致，S11优于-50dB。
前向耦合也同意（-34dB @ 2.2GHz）。
我从Sonnet获得的隔离度在整个频段上优于26dB，因此我们在整个频段内S41小于-34dB-26dB = -60dB。
所以我认为你可以信任180cells /波长的结果。
我不会尝试进一步提高精度，因为网格密度不是唯一的误差来源：无论如何都不包括与制造相关的小变化（边缘轮廓，粗糙度）。
显然，具有更紧密耦合的耦合器将更容易准确地模拟，因为来自S41的不匹配的相对影响将小得多。
最好的问候Volker
sonnetmesh.PNG20.5 KB



     以下为原文

  Hello RinC,

> {quote:title=RinC wrote:}{quote}
> here i have a question about "Number of cells/wavelength" option in EM simulation.
> I already read documents about the Mesh options in EM simulation. From documents, as you increase the number of cells/wavelength more, you get more accurate results. In this case, i assume to neglect the time of simulation, the main aim is the accuracy of simulation.

Correct.

> I made a test with this options, here it arises a question for me.
> In my work, a 50 Ohm TL with RO4003c, conductor thickness = 17um, sub. thickness = 0.504 mm, at 2.45 GHz is 1.12mm ( this is a real dimension in LAB). From this information, i adjust the er of RO4003C as er=3.62. Then, i simulate this physical width by EM simulation.
>  1.case: 80 cells/wavelength: S11: -40dB
>  2.case: 180 celss/wavelength: S11: -55dB
> in both cases, it is "matched", and 180 cells/wavelength give better results. At this step, you can conclude that: 80cells/wavelength is enough, because in reality for really good calibration, VNA only gives best result of -65dB (this is the best result we got in LAB), so we cannot distinguish the differences.

Correct. No matter if we have -40dB or -55dB return loss, both is excellent. It is difficult to *accurately* measure such small S11 values in the lab.

> As you can see, now the difference is dramatically visible. We care about the matching condition (S33), coupling (S31), and isolation(S41) of Coupling branch. There is no change in S31, but S33 and S41 change alot. For 180 cells, S33 is matched, but with 80 cells, in my opinion, it is not matched.
> So, my question is, which values should we rely on?
> I attached my workspace for you if you need some checks

Interesting coupler structure, because the coupling is so weak. As you have already found out, the default mesh densities of 20cells/wavelength + edge mesh are not sufficient if you want very accurate results for return loss. Forward coupling is more robust, and changes only slightly with meshing. Now the really difficult part is isolation: because the forward coupling is so weak, we have small levels anyway, and the measured S41 is a mixture of coupler isolation and signal reflected back at the load (mismatch). This makes your results so very sensitive to even the smallest changes in the setup/mesh and port calibration.

I looked at your results, also verified them against another very accurate, very dense mesh EM simulation in Sonnet. The Momentum results for *180* cells/wavelength agrees with the Sonnet results for S11, with S11 better than -50dB. The forward coupling agrees as well (-34dB @ 2.2GHz). The isolation that I get from Sonnet is better than 26dB across the band, so that we have S41 less than -34dB-26dB= -60dB across the band. 

So I think you can trust your 180cells/wavelength results. I would not try to push it even further for accuracy because the mesh density is not the only error source: manufacturing-related small changes aren't included anyway (edge profile, roughness).

Obviously, a coupler with tighter coupling would be much easier to simulate accurately, because the relative effects from mismatch on S41 would be much smaller.

Best regards
Volker

附件

• compare.PNG27.4 KB
  
• sonnetmesh.PNG20.5 KB
  

> {quote：title = volker_muehlhaus写道：} {quote}>>有趣的耦合器结构，因为耦合是如此弱。
正如您已经发现的那样，如果您希望获得非常准确的回波损失结果，则默认网格密度为20cells /波长+边缘网格是不够的。
前向耦合更加稳健，并且随着网格化而略微变化。
现在真正困难的部分是隔离：*因为前向耦合太弱，我们无论如何都有小电平，测得的S41是耦合器隔离和在负载反射回来的信号的混合（失配）。
这使得您的结果对设置/网格和端口校准中的最小变化非常敏感。*>所以我认为您可以信任180cells /波长的结果。
*我不会试图进一步提高准确度，因为网格密度不是唯一的误差来源：无论如何都不包括与制造相关的小变化（边缘轮廓，粗糙度）。*>最好的问候> Volker早安Volker，
我同意你的解释。
我真正的旧耦合器对耦合分支的匹配条件非常敏感。
参数S11，S31与模拟一致。
然而，S33不再匹配（约-25dB），而隔离S41为-60dB（在我的模拟中为-75dB）。
我将使用增加的网格密度进行新设计，并将再次检查结果。
从你的回复中，我意识到我没有包括边缘轮廓，粗糙度......这些误差源应该有助于改变性能。
顺便说一句，我认为在我从新设计中获得测量后，我们可能会有更详细的讨论:)。
感谢您的帮助，致以诚挚的问候，RinC



     以下为原文

  > {quote:title=volker_muehlhaus wrote:}{quote}
> 
> Interesting coupler structure, because the coupling is so weak. As you have already found out, the default mesh densities of 20cells/wavelength + edge mesh are not sufficient if you want very accurate results for return loss. Forward coupling is more robust, and changes only slightly with meshing. Now the really difficult part is isolation: *because the forward coupling is so weak, we have small levels anyway, and the measured S41 is a mixture of coupler isolation and signal reflected back at the load (mismatch). This makes your results so very sensitive to even the smallest changes in the setup/mesh and port calibration.*

> So I think you can trust your 180cells/wavelength results. *I would not try to push it even further for accuracy because the mesh density is not the only error source: manufacturing-related small changes aren't included anyway (edge profile, roughness).*

> Best regards
> Volker

Good morning Volker,
I agree with your explanations. My real old coupler was very sensitivity to matching conditions at coupling branch. Parameters S11, S31 agreed with the simulation. However, S33 wasn't matched anymore (around -25dB), and Isolation S41 was -60dB (in my simulation was -75dB). I gonna make new design with increased mesh density and will check the results again. 
From your reply, i realized that i did not include edge profile, roughness... These error sources should contribute into the changes of the performance. Btw, i think we might have some more detailed discussion after i get measurement from new design :).
Thank you for your helps,
Best regards,
RinC

早上好！
> {quote：title = RinC写道：}>从你的回复中，我意识到我没有包括边缘轮廓，粗糙度......这些误差源应该有助于改变性能。
是的，预计会对这些极端准确度产生一些影响......但是为这些影响建立可信数据非常困难，因此我们可以将其包含在模拟中。
>顺便说一句，我想在我从新设计中获得测量后，我们可能会有更详细的讨论:)。
太好了，我很期待。
祝你的设计好运！



     以下为原文

  Good morning!

> {quote:title=RinC wrote:}> From your reply, i realized that i did not include edge profile, roughness... These error sources should contribute into the changes of the performance. 

Yes, some influence at these extreme levels of accuracy is expected ... but it's so very difficult to establish trusted data for these effects, so that we can include it in simulations. 

> Btw, i think we might have some more detailed discussion after i get measurement from new design :).

Great, I look forward to it. Good luck for your design!