SIW Bandpass Filter Design and Optimization Using SynMatrix

Hello everyone, this is Adel from Ozen Engineering. In this video we will design a 5th order 28 GHz substrate integrated waveguide filter using SynMatrix and optimize it with the AI optimizer, optiSLang.

We will start with the filter synthesis and I'll set the filter order to 5. The filter response is by an Ozen Engineering Ansys software. For the return loss, we want it better than minus 23 dB. Let's set the start to 27.74 and the stop to 28. 26. Let's click here to refresh.

And let's add a second specification for the isolation. Let's set the start to 28.66 and refresh. If I go to Coupling Matrix, here we have the normalized coupling matrix. If we click here, we can change it to coupling bandwidth, where here in green you can see the frequency of each resonator in MHz.

Now we can go to the next step which is the 3D modeling and click on cavity, we will select SIW, we will keep mm and GHz as the units and click confirm and start new design.

For this filter I will be using Hexagon, so here I will start by choosing the substrate and I will choose Rogers Ultralam 1217. I'll change the VIA diameter to 0.171 and click Calculate All. Here you can see the quality factor is 198. Now we can go to the next step which is the single cavity.

I'll set the SIW side length to 3.15 mm and enable Tuning VIAs. Here on the use corner, let's make sure we have the right tuning via diameter and click apply in next step.

Here we will be running a parametric study, the tuning depth, and I'll vary it from 0.65 to 0.7, and I'll change the number of points to 3. Let's set the maximum number of passes to 10, the number of modes to 3, and maximum delta frequency per pass to 1%.

Then click construct model, here we will save it, I'll call it SIW28GHz, click OK. The single cavity model is created in HFSS. We go back to CineMatrix and click Run Parametric Simulation. Here we have the tuning via depth versus frequency.

If I go back to single, in the coupling bandwidth you can see that the frequency of all the resonators is around 27.99 GHz. So this green line corresponds to the required frequency. And here the estimated tuning via depth is around 0.678 mm.

Now we can go to the next step, which is the coupling scheme. Here I'll be using the edge window coupling. Let's enable follow single model tuning via setting. Here we have the right iris via diameter, so let's click set as main coupling and apply a next step.

Then we will run a parametric study with the iris width, I'll vary it from 1.5 to 1. 8. Let's set the maximum number of passes to 10, number of modes to 3, and maximum delta frequency per pass to 1%. Let's click construct model and then run parametric simulation.

Here we have the required coupling range from the coupling matrix highlighted in green, and if I click on this arrow, we can see the target value for each coupling. So for M12 and M45 it's around 0 and for M23 and M34 it's around 0. Let click again on the arrow.

And here the estimated iris width for M23 and M34 is 1 and for M12 and M45, it's around 1.75 mm. So I'll type in those values, 1.75 and 1. 56. Here I click on export data to 3D model design. And then we can go to the next step which is the input and output. Here I'll use the co-planer configuration.

Let's click here to calculate the trace width, enable Trace VIAs. Here we have the right VIA diameter and SIW size length. Let's change the tuning depth to 0.66 mm and change the insertion depth to 0.77 mm.

Here I'll set the branch length to 0.65 mm, then click Set input and output, Apply in next step. Here I'll set the maximum number of passes to 10, minimum coverage passes to 1, and maximum data S to 0. 02. We will run a parametric study.

Let's select the branch length and I'll vary it from 0.5 to 1 mm and change the count to 5. Here I'll click construct model, then run parametric simulation. The target group delay is 0.905ns. Here, if I hide some of the curves, we can estimate the branch length to 0.62mm.

Now we can go to the last step, which is the full 3D modeling. Let's start with the input. So here if I scroll down, I can unhide the branch length and set it to 0. 62. I'll do the same for the output. Now I'll select the first resonator.

And we'll set the tuning depth to 0. 678. We'll use the same value for all the other resonators.

Now let's go to M12 and set it to 1.75, which is the same for M45, and for M23, let's set it to 1.56, which is the same for M 34. We don't have a cross-coupled structure, so we can skip that and go to modeling.

Here I set the maximum number of passes to 10, the minimum coverage passes to 1, and maximum data as 2.02 and click model construction. Now in HFSS, we have our full 3D model of the SIW filter. If I go back to SynMatrix, we can click Run Simulation.

Now that the simulation is complete, we can go back to HFSS and view the results. Here we have the initial response of the filter. We can save this design in a new project and optimize it.

So I will copy this design, go to File, New, I'll name this project SIW28GHz Optimized, and paste the design here. Let's save it somewhere. And now we can go back to SynMatrix to start our optimization. So I go to optimization and click intelligent optimization.

Here we start with the custom optimization. So I'll select it, click confirm. Here I'll click on simulation file and unload the project that we just created. Click open.

In the simulation settings, I'll increase the number of passes to 10, change the minimum coverage passes and delta S, then click save. Now let's go to variable mapping and map the physical variables to the elements of the coupling matrix.

So we'll select the cavity 1 tuning depth, that's M11, cavity 2 tuning depth M22, cavity 3 tuning depth that's M33, cavity 4 tuning depth that's M44, cavity 5 tuning depth that's M 55. The 1-2 iris width is M12, 2-3 iris width that's M23, 3-4 iris width that's M34, and iris 45 that's M 45. Now we have the input branch length, that's ms, and the output branch length which is ml.

So now we have all the variables mapped properly. We can click save, then click save mapping file, and let's click run. Here we have the golden S21 in red, and we have the simulated S21, so let's click extract.

Here you can see there is a difference between the simulated response and the coupling matrix response, which is due to the dispersion effects that are not reflected in the coupling matrix. So if we go to single, we can apply dispersion. So I'll select filter dispersion and let's try four.

Click Apply Dispersion. Let's go back to Intelligent Optimization and click Extract Matrix. Let's try 5. Now the coupling matrix response and the simulated response match. Now let's go and view the Error Levels, which show the difference between the extracted matrix elements and the golden matrix.

Here all resonators are tuned low in frequency. Resonator 1 and 5 by around 1000 MHz, 2 and 4 by around 500 MHz, and resonator 3 by around 400 MHz. Now if I go back to the results from the single cavity simulation, we can see that increasing the VIA depth increases the frequency.

and here if we calculate the slope it's around 1.02 GHz per mm so we need to increase the tuning depth of resonators 1 and 5 by around 0.98 so change it to 1.65 and we need to increase the tuning depth for resonators 2 and 4 by around 0.49 I'll change it to 1.17 and we need to increase the tuning depth for cavity 3 by around 0.39 so I'll change it to 1. 07. Here I click the plus sign and the minus sign to get rid of the error and then click run.

Once the simulation is complete we can go and view the S-parameters. Here let's click on extract matrix. Let's go and view the error levels. Here you can see that the resonant frequency of all resonators is too high, so we need to decrease the tuning depth.

I'll reduce the tuning depth to 1.35 for resonator 1 and 5, 1.12 for resonators 4 and 2, and 1.05 for resonator 3. Then we can click run. Now let check our S-parameters. Here I click Extract Matrix. You can see that the return loss is now better than minus 10 dB.

Now we can switch to the SynMatrix AI Optimizer. So I'll go to Reset, click OK. Here I'll select AI Optimization, AI. Then we load our project.

In AI settings, I'll turn on adaptive error correction and I'll change the global max changes to 0.03 and the global trial modification to 0.01 mm and then click OK. Let's go to the simulation settings. We can keep them as they are for now. Let's click plus and minus to get rid of the error.

Then click get initial performance. Now we can click on extract matrix and start the AI optimization. So here we ran 30 iterations and the minimum overall error occurred at iteration 21. Now we can go to S2P files, select iteration 21, and implement it.

Now let's go and run a second AI optimization. So I click reset. I'll select AI. Let's download our project file. Let's go to AI settings. I'll keep the max changes to 0.0015 and I'll change the trial modification to 0.005 mm. Then click OK.

In the simulation settings, I'll increase the maximum number of passes to 25. Click Save. Let's set the lower bound to 80% and the upper bound to 120%. Then I'll click Get Initial Performance, extract matrix, and now we can start the second AI optimization.

Here the minimum overall error occurred at iteration 21, so we can go to s2p files, select 21, and implement it. Now let's run another AI optimization, so I'll go to Reset. Here I'll select AI. Let's load our project file.

I'll turn on the adaptive error correction and set the trial modification to 0.03 and max changes to 0. 009. I'll click OK. In the simulation settings, I'll increase the maximum number of passes to 25 and I'll click Save. I'll set the lower bound to 90% and the upper bound to 110%.

Now let's click get initial performance. Extract matrix. And now we can run our AI optimization. Here we can see the results of this AI optimization. The simulation stopped at iteration 28 because the filter is now spec compliant. Let's go to S2P files, select iteration 28, and implement it.

Now we can go to HFSS and plot the results. So right click on Results and create a rectangular plot. Here we select the SynMatrix solution setup and plot S11 S 21. As you can see, our SIW filter is now spec compliant.

In this video, we saw how to design a 5th order, 28 GHz, substrate integrated waveguide filter using SynMatrix and optimize it with the optiSLang. Thanks for watching and see you in the next video. Please contact us at https://ozeninc.com/contact for more information.