Ku波段放大器电路中的频移

您好，本周，安捷伦有一个网络直播+“平面微波电路网络直播的容差分析”+ Lance Lascari，希望很快就可以作为点播网络直播。
Lance展示了基板特性（epsr，厚度）和蚀刻公差的公差如何能够显着地减少滤波器的偏差。
基本上，他参数化了everthing，然后在名义和角落参数上扫描所有组合。
这是开始进行公差分析的简便方法。
> {quote：title = Noman-ul-Haq写道：} {quote}>衬底材料主要是RO4003C或RO5880。
你用过哪个erel值？
罗杰斯指定了两个值，一个标称值（RO4003C为3.38）和“*设计*介电常数”（对于4003C为3.55），这对微带应用有效。
它们以不同的方式测量，“设计”值是您应该用于模拟的值。
> {quote：title = Noman-ul-Haq写道：} {quote}>我怀疑对于某些几何形状，衬底的有效介电常数会发生变化而这会导致偏移，但我也认为ADS2008 Momentum模拟器也应计算出有效的介电常数
在模拟过程中取决于几何形状和频率。
因此，不应该观察到这种转变。
正确。
您始终输入基板介电常数（对于4003C，为3.55）。
衬底和空气中的场分布（以及产生的有效介电常数）将由EM求解器在内部计算。
您只需指定基材本身。
> {quote：title = Noman-ul-Haq写道：} {quote}>我总是在微波模式下模拟具有最大网格密度和最小可能弧分辨率的电路。
听起来不错。
确保启用了* Edge Mesh *，这对于具有窄间隙的耦合线有所不同。
> {quote：title = Noman-ul-Haq写道：} {quote}>这种转变有时是积极的，有时是消极的。
这可能来自制造中的*公差*（见我的帖子开头），这会导致尺寸变化或系统误差。
此外，您的有源器件*（晶体管）的*模型可能是错误的：例如，它可能有一个参数平面，用于与Momentum模型中的参考平面不匹配的S参数，或者它可能只是一个坏的
模型/测量。
像Modelithics这样的公司提供定义良好的良好设备模型。
最好的问候Volker



     以下为原文

  Hello,

this week, Agilent had a webcast +"Tolerance Analysis for Planar Microwave Circuits Webcast"+ by Lance Lascari, which hopefully will be available as an on-demand webcast soon. 

Lance showed how tolerances in substrate properties (epsr, thickness) and etching tolerances can siginificantly de-tune filters. Basically, he parameterized everthing, and then swept all combinations over nominal and corner parameters. That's an easy way to get started with tolerance analysis.

> {quote:title=Noman-ul-Haq wrote:}{quote}
> Substrate material is mostly RO4003C or RO5880. 

Which erel value have you used? Rogers specifies two values, one nominal value (3.38 for RO4003C) and a "*dielectric constant for design*" (3.55 for 4003C) which is valid for microstrip applications. They are measured in different ways, and the "design" value is what you should use for simulation. 

> {quote:title=Noman-ul-Haq wrote:}{quote}
> I suspect that for certain geometries, effective dielectric constant of substrate varies and this causes the shift but I also think that ADS2008 Momentum simulator should also calculate the effective Dielectric constant depending on the geometry and frequency during simulation. Thus, this shift should not be observed. 

Correct. You always enter the substrate dielectric constant (3.55 for 4003C) . The field distributions in substrate and air (and resulting effective dielectric constant) will be calculated internally by the EM solver. You just specify the substrate itself. 

> {quote:title=Noman-ul-Haq wrote:}{quote}
> I always simulate the circuits with maximum possible mesh density and minimum possible arc resolution in Microwave mode. 

Sounds good. Make sure you have *Edge Mesh* enabled, which makes a difference for coupled lines with narrow gaps.

> {quote:title=Noman-ul-Haq wrote:}{quote}
> this shift is sometimes positive and sometimes negative. 

This could be from *tolerances in manufacturing* (see beginning of my post), which lead to variation or systematic error in dimensions. Also, the *model for your active devices* (transistor) could be wrong: for example, it could have a reference plane for the S-parameters that does not match the reference plane in your Momentum model, or it could be just a bad model/measurement. There are some companies like Modelithics who provide well defined, good device models. 

Best regards
Volker

谢谢，霍华德！
实际上，Edge Mesh的模拟速度会慢一些。
但如果结果不好，EM模拟速度是没有价值的。
最初的帖子报告了模拟与测量的问题。
这可能与Momentum完全无关，但我们还不知道。
在我看来，我们最好在优化速度之前建立一个可信赖的参考模拟。
一旦我们知道密集网格的结果是什么，我们就可以优化内存/速度，看看网格仍然“足够好”。
凭借在Sonnet的EM支持方面15年的经验，我发现Momentum默认网格有点过于放松，对速度进行了太多优化。
这就是为什么我建议切换边缘网格：消除网格密度作为误差源，同时排除测量与模拟的差异。
当然，Momentum中的厚金属侧壁可以接收一些推向导体侧面的电流，与薄板（零厚度）金属相比，在这种厚金属模拟中对边缘网格的需求更少。
这很好，通常效果很好，并且在厚金属模型中对边缘网格的需求较少（例如，与Sonnet相比）。
但同样 - 对于这类问题和频率范围，最好检查一下这是否有效。
最好的问候Volker



     以下为原文

  Thanks, Howard!

Indeed, the simulation will be slower with Edge Mesh. But EM simulation speed is worth nothing if results are bad. The initial post reports a problem with simulation vs. measurements. That might be totally unrelated to Momentum, but we don't know yet.

In my opinion, we better establish a trusted reference simulation before optimizing for speed. Once we know what the result with dense mesh is, we can then optimize for memory/speed and see what mesh is still "good enough". With 15 years of experience in EM support for Sonnet, I find the Momentum default meshing a bit too relaxed, too much optimized for speed. That's why I suggested to switch edge mesh on: to eliminate mesh density as an error source, while troubleshooting the measured vs. simulated discrepancy.

It's certainly true that thick metal side walls in Momentum can take over some current that pushes toward the sides of the conductors, and there is less need for edge mesh in such thick metal simulation, compared to thin sheet (zero thickness) metal. That's nice, often works great, and results in less need for edge mesh in thick metal models (compared to Sonnet, for example). But again - it's better to check if that is valid here, for this class of problems and frequency range. 

Best regards
Volker

感谢您的回复Edge mesh已启用。
最大。
可能的网格密度用于模拟Er：3.55，即在该项目中使用Design Dk



     以下为原文

  Thanks for replies

Edge mesh was enabled. 

Max. possible mesh density was used in simulation

Er: 3.55 i.e. Design Dk was used in this project

当在Simulation中，我将Er从3.55更改为4.25时，模拟结果只是吻了测量结果。
即，模拟结果在负方向上移动800MHz。
这表明该基板对该频率的有效介电常数和该几何形状大于3.55，如制造商的数据表所引用。
在开发出这个放大器的第一个版本之后，在得出这个结论之后，我们可以说对这种介电材料和这种设计拓扑结构进行了“电路板验证”。
现在，可以使用Er：4.25设计和开发新版本，以便可以补偿800MHz的负频移。



     以下为原文

  When, in Simulation, I changed the Er from 3.55 to 4.25, simulation results just kissed the measured results. i.e. simulation results shifted in negative direction by 800MHz. 

This shows that effective dielectric constant of this substrate for this frequency and this geometry is greater than 3.55 as quoted by Manufacturer's datasheet. 

After development of first version of this amplifier, and after reaching this conclusion, we can say that "Board verification" is done for this dielectric material and this design topology.

Now, a new version can be designed and developed by using Er:4.25 so that 800 MHz negative frequency shift may be compensated.

> {quote：title = Noman-ul-Haq写道：} {quote}>这表明该频率的这种基板的有效介电常数和制造商的数据表所引用的几何尺寸大于3.55。
从多年的全职EM支持经验来看：没有我看到很多客户都失​​败了这种方法。
您已经证明可以通过调整基板数据来切换到较低频率。
这是预料之中的。
但问题是这种基材材料定义很好，质量控制良好。
这个问题不太可能存在于基质本身中。
当然，您可以通过调整基板数据来部分补偿其他错误（坏组件数据或错误的模拟设置）。
如果真正的问题是某些封装寄生在元件中或者接地过孔中的额外电感（仅举例），您将看到频移。
现在，您还可以调整基板数据，以移动频率，而不是修复封装寄生建模。
这将有效 - 对于这个确切的设计。
当您尝试使用其他设计的“新”基板数据时，您会发现它对其他设计失败了。
因此，我在该领域长期经验的建议是：找到真正的问题，不要试图通过调整远远超出规定公差的基板数据来补偿它。
它只是不能很好地工作。



     以下为原文

  > {quote:title=Noman-ul-Haq wrote:}{quote}
> This shows that effective dielectric constant of this substrate for this frequency and this geometry is greater than 3.55 as quoted by Manufacturer's datasheet. 

From many years of full-time EM support experience: no
I have seen many customers fail with this approach.

You have demonstrated that you can shift to lower frequencies by tweaking the substrate data. That is expected. But the problem is that this substrate material is very well defined, and well quality controlled. It is very unlikely that the problem is in the substrate itself. Of course, you can partially compensate other errors (bad component data or bad simulation setup) by tweaking the substrate data. 

If the real problem is some package parasitic in the components or extra inductance in the ground vias (just an example), you will see a frequency shift. Now, instead of fixing the package parasitic modelling, you could also tweak the substrate data, to shift frequencies. This will work - for this exact design. When you try the "new" substrate data with another design, you will find that it fails for those other designs.

So my advice from long experience in this field is: find the real problem, and don't try to compensate it by tweaking the substrate data far outside the specified tolerances. It just doesn't work well.