ADS中的allegro PCB提取

> {quote：title = y19085写道：} {quote}>我的EM设置为0~20GHz，带20pts，这对于那些条件是否足够？
对于这样简单的线路，最多20点的自适应扫描应该没问题。
如果点数不足以构建聚合模型，您将在Momentum日志中获得*警告*。
您现在可以检查*模拟日志*：打开ADS布局并转到菜单：EM>显示最新> S-Param模拟摘要。
> {quote：title = y19085写道：} {quote}>我认为这可能是我的一些选项设置错误。
我真的很感谢你的帮助。
谢谢。
高级别，听起来不错，所以我使用的细节/设置一定有问题。
你能为这个简单的2行案例上传工作区吗？



     以下为原文

  > {quote:title=y19085 wrote:}{quote}
> My EM setup is 0~20GHz with 20pts, is that enough for those condition? 

Adaptive sweep with maximum of 20 points should be fine for such simple lines.
If the number of points isn't enough to build a converged model, you will get a *warning* in the Momentum log.
You can check the *simulation log* now: open the ADS layout and go to the menu: 
EM > Show Most Recent > S-Param Simulation Summary.


> {quote:title=y19085 wrote:}{quote}
> I think it may be some of my option setting errors.   I really appreciate for your help. Thank you.

High level, this sounds ok, so I there must be something wrong with the details/settings that you used. 
Can you upload the workspace for that simple 2-line case?

> {quote：title = y19085写道：} {quote}>我认为这可能是我的一些选项设置错误。
是。
。
。
。
。
。
。
。
。
。
。
。
。
。
。
但是，如果我们不知道你做了什么，我们怎么能帮忙呢？
没有屏幕截图，没有工作空间，没有。
如果您希望论坛提供帮助，您应该上传您的工作区，并更详细地了解您评估的结果以及评估方式。
从您的初始帖子开始，我不清楚您使用SPICE模型进行的瞬态测试是在ADS还是HSpice中完成的，以及您看到的实际结果。



     以下为原文

  > {quote:title=y19085 wrote:}{quote}
> I think it may be some of my option setting errors.   

Yes.

.
.
.
.
.
.
.
.
.
.
.
.
.
.


But how can we help, if we have no idea what you did? No screenshots, no workspace, nothing. 
If you want the forum to help, you should upload your workspace, and some more details what results you evaluated, and how you evaluated. From your initial post, it is not clear to me if your transient test with the SPICE model was done in ADS or HSpice, and what actual results you saw.

谢谢你的回复，这对我很有帮助。
但是我很抱歉我无法上传我的工作区，因为我在安全实验室中运行这些工作人员，我无法轻松地显示我的数据。
但是，在阅读了你的回复并开始研究ADS后，我想我找到了原因，我可以解决一些问题。
正如我之前提到的，我正在尝试提取PCB以进行高速ADC / DAC的协同仿真。
时域中的DAC输出包含瞬态部分和静态/稳定部分，其行为类似于方波函数。
为了处理瞬态部分（Tr / Tf = 20ps~50ps）和静态部分，我的频率设置范围是0~20GHz，但只有20个点。
因此，近DC波段EM模拟点太粗糙，S21拟合曲线远低于实际曲线。
这导致了我的第一和第二个问题。
我可以通过将EM设置重新排列到更窄的频率范围并使用线性扫描的更多点（0~4GHz，30pt）来解决它们，这可以计算更多的DC部分。
这确实有帮助，DC附近的S21结果是正常的，并且从HSPICE仿真计算的DC电阻类似于allegro的结果。
因此，在DC附近运行更多点可以解决我的问题1和2.但是如何为方波输入信号运行EM仿真？
将我的频率范围扩大到50或60GHz？
如果我使用与P1＆amp; P2解决方案相同的模拟点密度。
EM模拟时间可能非常长。
更不用说更大的PCB板了。
或许这是我能做的唯一方法？
谢谢你的帮助，希望不要浪费你的大量时间。
我会尝试用相机拍摄一些有用的模拟照片。



     以下为原文

  Thanks for your reply, it's helpful to me. But I'm sorry I cannot upload my workspace because I'm running these staff in a Security Lab where I cannot easily bring out my data. 

However,after reading your reply and working on ADS, I think I found up the reason and I can solve some of my problems.
As I mention before, I'm trying to extract PCB for co-simulation of high-speed ADC/DAC. The DAC output in time domain contains transient part and static/stable part, which behave like a square wave function. In order to deal with the transient part (Tr/Tf = 20ps~50ps) and static part, my frequency setup span is 0~20GHz but only with 20points. So the near-DC band EM simulation point is too coarse and the S21 fitting curve is far lower than the real curve. That causes my 1st and 2nd problem. And I can solve them by rearrange the EM setup to narrower frequency range and with more point (0~4GHz, 30pt) with linear sweep,which can calculate more of the DC part. It really helps, The S21 result near DC is normal and the DC resistance calculated from HSPICE simulation resembles the result in allegro. 

So, to run more point near DC can solve my problem 1 and 2. But how can I run EM simulation for a square wave input signal? Enlarge my frequency span to maybe 50 or 60GHz? If I use the same simulation point density as in the solution for P1&P2. The EM simulation time can be incredibly long. Not to mention a larger PCB board. Or maybe this is the only way I can do?

Thank you for your help, hope is didn't waste lots of your time.
And I will try to take some useful simulation picture by my camera.

> {quote：title = y19085写道：} {quote}>但对不起，我无法上传我的工作区当然，没有问题，谢谢你的回复。
我不确定你是否见过这个答案。
> {quote：title = y19085写道：} {quote}>我的频率设置范围是0~20GHz，但只有20个点。
所以近DC波段EM模拟点太粗糙我很困惑。
您是否使用20个离散点运行，或使用自适应扫描？
您是否在模拟日志文件的末尾看到类似“自适应：覆盖频率范围的100.0％”的内容？
使用自适应扫描进行模拟时，插值结果将具有低频数据点。
附加的是用于耦合线的0-20GHz自适应扫描的输出，其获得29（！）点。
该模拟器足够智能，可以输入9Hz，90Hz，900Hz，9kHz，90kHz等，因此DC和第一个常规GHz频率点之间没有那个间隙。
你的输出中还有这样的额外数据吗？
> {quote：title = y19085写道：} {quote}>因此，在DC附近运行更多点可以解决我的问题1和2.但是如何为方波输入信号运行EM模拟？
将我的频率范围扩大到50或60GHz？
如果我使用与P1＆amp; P2解决方案相同的模拟点密度。
EM模拟时间可能非常长。
更不用说更大的PCB板了。
或许这是我能做的唯一方法？
随着0-20Ghz的自适应扫描，Momentum已经投入输出中的超低频点（！），而实际上没有花费太多额外的时间来模拟EM所有*这些低频点。
我认为这是一种明智的默认行为。
如果你想要一些额外的EM模拟点，你可以混合对数扫描和自适应扫描。
在频率规划中，只需添加一个额外的输入行，然后为额外扫描选择日志扫描。



     以下为原文

  > {quote:title=y19085 wrote:}{quote}
> But I'm sorry I cannot upload my workspace 

Sure, no problems, and thanks for your reply. I wasn't sure if you had seen the answer at all. 

> {quote:title=y19085 wrote:}{quote}
>my frequency setup span is 0~20GHz but only with 20points. So the near-DC band EM simulation point is too coarse 

I'm confused. Did you run with 20 discrete points, or used the adaptive sweep? 
Did you see something like "Adaptive: 100.0 % of frequency range covered" at the end of the simulation log file?

When simulating with adaptive sweep, the interpolated results will have low frequency data points. Attached is the output from a 0-20GHz adaptive sweep for coupled lines, which took 29 (!) points. The simulator was smart enough to put in 9Hz, 90Hz, 900Hz, 9kHz, 90kHz etc. so that there is not that hughe gap between DC and the first regular GHz frequency point. Do you have such extra data in your output as well?

> {quote:title=y19085 wrote:}{quote}
> So, to run more point near DC can solve my problem 1 and 2. But how can I run EM simulation for a square wave input signal? Enlarge my frequency span to maybe 50 or 60GHz? If I use the same simulation point density as in the solution for P1&P2. The EM simulation time can be incredibly long. Not to mention a larger PCB board. Or maybe this is the only way I can do?

With the adaptive sweep from 0-20Ghz, Momentum had already thrown in the extra low frequency points in the output (!), without actually spending too much extra time on EM simulating *all* these low frequency points. I think that's a smart default behaviour. If you like to enfore some extra EM-simulated points, you can mix log sweep and adaptive sweep. In the frequency plan, just add an additional entry line and choose log sweep for that extra sweep.

附件

• sweep.PNG8.0 KB
  

谢谢你的帮助。
您混合不同频率设置的建议很有帮助。
但我仍然有一些问题。
Q1。
网格密度设置是否有任何原理？
较低的密度导致网格密度过于粗糙的警告，但较高会导致内存不足问题。
Q2。
我用“无”校准将每个端口设置为50 + 0i欧姆，我很好奇如何在端口设置中设置校准。
“无”是对的吗？
Q3。
在宽带Spice Generator中，SPICE输出和EM仿真之间仍存在一些S参数差异。
我试图安排配件公差，但似乎没有明显改善。
还有其他提高准确度的方法吗？
感谢您的热情和善意。编辑：y19085于2013年11月20日下午7:07



     以下为原文

  Thanks for your help. Your suggestion of mixing different frequency setting is helpful.

But I still have few question. 
Q1. Are there any principal of mesh density setting? Lower density result in warning of mesh density too coarse, but higher will lead to out of memory problem.

Q2. I set each of my port to 50+0i Ohm with "None" Calibration, I'm curious about how to set the calibration in port setting. Is "None" right?

Q3. In Broad Band Spice Generator, there is still some S parameter differences between SPICE output and EM simulation. I have tried to arrange the fitting tolerance, but it seems to be no significant improvement. Are there any other method to raise the accuracy?

Thank you for your enthusiastic and kindness.

Edited by: y19085 on Nov 20, 2013 7:07 PM

> {quote：title = y19085写道：} {quote}> Q1。
网格密度设置是否有任何原理？
较低的密度导致网格密度过于粗糙的警告，但较高会导致内存不足问题。
决定任何EM结果的准确性有两个主要主题：良好的网格划分和正确的端口设置。
通常，更高的网格密度更好，因为求解器可以以更逼真的方式对电流建模。
但正如您所提到的，我们不希望在不改变结果的地方浪费网格密度。
在提供对网格，全局和每层，甚至每个多边形（如果需要）的控制时，动量非常好。
我没有具体的建议可以给你，因为你没有详细说明你的模型。
在这里问的问题：是否包括地面飞机？
如果有地平面，可以将它们建模为无限平面（不需要网格划分 - >内存效率）或者我们是否需要绘制必须进行网格划分的有限地平面？
如果地面是绘制的有限金属平面，这也会影响正确的端口定义。
您是否查看过在线帮助或知识库中的网格文档？
本文档提供了一个很好的概述，虽然它比应用程序相关更加以功能为中心：http：//edocs.soco.agilent.com/display/ads201306/Defining+Mesh+Settings+for+Momentum> {quote：title = y19085写道
：} {quote}> Q2。
我用“无”校准将每个端口设置为50 + 0i欧姆，我很好奇如何在端口设置中设置校准。
“无”是对的吗？
EM中的端口引入了一些不连续性，就像硬件中的连接器一样。
校准是为了消除这些错误，并为DUT本身获得更准确的结果。
端口校准有一些要求，在某些情况下，端口校准可能会引入副作用和错误。
在不知道布局配置的情况下，很难评论端口校准。
对于低精度要求或电气小的布局，端口校准并不重要。
知识库文档：​​http：//edocs.soco.agilent.com/display/ads201306/Ports+for+Momentum> {quote：title = y19085写道：} {quote}> Q3。
在宽带Spice Generator中，SPICE输出和EM仿真之间仍存在一些S参数差异。
我试图安排配件公差，但似乎没有明显改善。
还有其他提高准确度的方法吗？
我还没有使用过ADS Broad Band Spice Generator。
在我之前与Sonnet的合作中，当输入数据良好时，Sonnet宽带香料生成器与低阶多项式非常吻合，并且在输入数据出现问题时难以适应（如端口校准问题导致错误
/非物理S参数）。
所以你首先要确保你的输入数据是好的。
对不起，我对这个话题没有更好的答案。
以下是您可以在知识库中找到的内容：http：//edocs.soco.agilent.com/display/ads201306/Creating+Broadband+SPICE+Models



     以下为原文

  > {quote:title=y19085 wrote:}{quote}
> Q1. Are there any principal of mesh density setting? Lower density result in warning of mesh density too coarse, but higher will lead to out of memory problem.

There are two main topics that decide about the accuracy of any EM result: good meshing and proper port settings. 
In general, higher mesh density is better because the solver can model the currents in a more realistic way. But as you have mentioned, we don't want to waste mesh density in places where it doesn't change the result.

Momentum is pretty good in offering control over the mesh, global and per layer, and even per polygon if needed.

There is no specific advice that I can give to you, because you haven't given much detail about your model. Question to ask here: Are ground planes included? If there are ground planes, can these be modelled as an infinite plane (needs no meshing -> memory efficient) or do we need to draw finite ground planes that have to be meshed? If the ground is a drawn, finite metal plane, this also affects the proper port definition.

Have you looked at the mesh documentation in the online help, or in the Knowledge base?
This document gives a good overview, although it's more feature-centric than application related:
http://edocs.soco.agilent.com/display/ads201306/Defining+Mesh+Settings+for+Momentum


 
> {quote:title=y19085 wrote:}{quote}
> Q2. I set each of my port to 50+0i Ohm with "None" Calibration, I'm curious about how to set the calibration in port setting. Is "None" right?

Ports in EM introduce some discontinuity, just like connectors in the hardware. Calibration is about removing those errors, and obtain more accurate results for the DUT itself. 

There are some requirements for port calibration, and in some cases port calibration might introduce side effects and errors. It's difficult to comment on port calibration without knowing the layout configuration. For low accuracy requirements, or layouts that are electrically small, port calibration is not that important.

Knowledge base document:
http://edocs.soco.agilent.com/display/ads201306/Ports+for+Momentum 

> {quote:title=y19085 wrote:}{quote}
> Q3. In Broad Band Spice Generator, there is still some S parameter differences between SPICE output and EM simulation. I have tried to arrange the fitting tolerance, but it seems to be no significant improvement. Are there any other method to raise the accuracy?

I haven't used the ADS Broad Band Spice Generator yet. In my previous work with Sonnet, the Sonnet Broad Band Spice Generator was fitting nicely with low order polynomials when the input data was good, and was having trouble to fit when there was something wrong with the input data (like port calibration issues leading to bad/unphysical S-parameters). So you first want to make sure that your input data is good. Sorry I have no better answer on this topic.

Here's what you can find in the knowledge base:
http://edocs.soco.agilent.com/display/ads201306/Creating+Broadband+SPICE+Models