em simulator directional coupler
I design microstrip directional coupler.
I use Ansoft Designer and AWR Microwave Office.
First step, I design and simulate circuit.
Second step, I transfer circuit to layout for 2.5D EM and simulate it.
But I don't have good result. The result is difference about 1 ~ 1.5 dB.
I don't know reason. I increase cell for analysis. But same result.
In case of simple edge coupled, same result.
If you have solution, explain method.
Thank you.
Hi, Hohoh:
For directional coupler, the edge cells are very critial. Normally, if you have an EM simulator, you should try to use edge cells. I know IE3D users can do a single pass design for 20 dB couplers by using the Automatic Edge Cells (AEC) feature on IE3D. They get very good results. They tried to use AEC to design 40 dB coupler on IE3D 10.x. However, there is some difference because the regular AEC is not good enough. They manually create multiple layers of edge cells on IE3D and they could get single pass design on 40 dB coupler. They told me the story on the IEEE IMS 2004. Now, on the IE3D 11, we have implemented the multiple level AEC. You can create multiple edge cells automatically on IE3D. I can see single pass design (simulation and optimization) of 40 dB coupler is possible on IE3D. If you are interested in it, you can give it a try (www.zeland.com).
To check any EM analysis exactly for coupled lines, it is easy to use coupled stripline. For lossless, infinitely thin coupled stripline there is an exact solution. It is calculated in the attached Excel spread sheet. Simply put in your dimensions and the spread sheet will give you the exact (to 8 decimal places) S-parameters that you should expect. Your EM analysis of course gives a slightly different answer. The difference between the EM analysis and the spread sheet is the exact error in the EM analysis.
Jian's suggestion about edge meshing being required for high accuracy is absolutely correct for all types of mesh based EM analyses, both surface and volume meshing. In some analyses edge meshing turned off is default and you must manually turn it on for accuracy. In others (including Sonnet) edge meshing turned on is default and you must manually turn it off if you want increased speed.
To check your EM analysis for correctness, do a convergence analysis. Keep cutting the cell size in half and plotting the result. The result should converge to the exact answer, with error decreasing by half each time you cut the cell size in half. If the analysis does not converge smoothly, then there are additional error sources. When we do this with Sonnet, we converge to the correct answer for coupling values to about 150 - 200 dB down. After that, numerical error can become important.
If the gap between the lines is < 2 times the metal thickness, then you might want to use a thick metal model. There is an automatic multi-sheet thick metal model detailed in the Sonnet documentation. In AWR, such a model can be done manually. If there is metal loss, you should be careful that the skin depth is modeled correctly. For thick metal with narrow gaps, the skin depth allows current to flow inside the metal at some distance from the gap. This increases the effective width of the gap. This is important only with tightly coupled lines. The multi-sheet model allows the current to flow inside the metal provided the cell width is <= skin depth.
To determine if you need a thick metal model, do two analyses. First analyze with no thick metal. Then analyze with thick metal. Note the difference. If the difference is large, you need thick metal. If the difference is small, do not use thick metal.
If your EM analysis converges well for coupled stripline, then it is correct. If the EM analysis and your circuit theory tools give different answers for a similar microstrip coupled line, then your circuit theory answer should be ignored. The EM analysis is correct.
Stripline is just like microstrip except there is a second substrate and ground plane on top of the line. The second substrate and ground plane is just like the microstrip substrate and ground plane, except it is turned up side down, so the ground plane is on top.