You just covered my masters thesis from back in '91 while I was at BT Labs. I was modeling thin-film circuits of gold on alumina substrates at 2 - 60GHz. I developed software from scratch that split the space into variable sized cubes to run the finite element analysis. I'd run the simulation and then based on the magnitude of any change in a cube would then split into 8 smaller cubes and re-run. The simulation took 2 weeks on a DEC Alpha of the time. I found it was really important to include a sealed metal box, which I had custom manufactured in the lab's machine shop, and to include that as part of the simulation. I've not run the Rogers materials but have heard great things about them though. We took the measurements on an HP 8510C network analyzer which was about as good as you could get back in the day.
The difference in the simulated and measured center frequency is probably just due to the difference in the dielectric constant you used in your simulation and the design value recommended by Rogers. The dielectric constant of RO4003C is 3.38 but the recommended value for design is 3.55. In your simulation increase your dielectric constant and you will see that the center frequency of your filter shifts to lower frequencies.
As the response is slightly too low in frequency you might want to try cutting or grinding off the ends of the hairpin filters - 1mm at a time . I had a similar hairpin filter for 1152MHz on FR4 board which was too low and had to do this to get the filter to play nicely. A great video as always.
It's the mass of solder to the sma, I did these designs in cad and verified in prototypes. Had to have not even a fillet in the solder. None of the other team could match the minimum amount of solder joint.
"The blackest of the black arts in analog electronics", as CuriousMarc called microwave engineering. And now it's all over the place - in our routers, in our cells... PCBWay has been reaching out to me offering sponsorship - as if my teeny tiny channel mattered to anyone, haha! I'll certainly try them for a bunch of projects.
Hi IMSAI Guy. I think the Er to use in MOM is more like 3.48 and it would shift down slightly. The response proves the model is great. Just the lengths are a little long because the Er is too low. Or PCB Way shipped you 4350 boards instead of 4003. 4350 is the most common today, but a little more ceramic content. You can get a precise calipers thickness of just the dielectric, then with the area of your traces just calculate the parallel plate capacitance you should have versus measuring the same with your LCR meter and you can see if they used the right material. Or with the area and measured capacitance value, solve for Er and see what you get. Whatever that Er is, enter that in GENESYS and you will probably see the frequency shift down a bit.
The copper thickness and roughness changes the effective dielectric constant which is most likely the reason for the frequency shift. Rogers actually have a good paper on the topic. Also, the conductivity of the gold plating, which I assume is ENIG, can have its own effect on higher frequencies.
My takeaway? If you think about something like a military radar system and then the large expenditure in time and money that just Rogers spent in developing the materials that are used in order to make a small part of it work you start getting an understanding of why they cost so much.
With so many filters it might be interesting to see how a couple of them look in series. In parallel one will have to add splitters and combiners. 9:50 sometimes it might work to understand the frequency dependency while looking at temperature, - never mind.
Thanks! I always learn something new on your channel. N1BUG's channel has excellent microwave projects on it and has designed some filters using copper pipe caps.
PCB with Rogers dielectric? So fancy that I need a second monocle :D But in all seriousness, measuring S11 (hopefully it is better than -20dB) really helps to determine what is the cause of the frequency response shift (interesting as a follow-up video). Also check out the losses in the passband , it seems that you measured 4dB of in-band attenuation which is a lot compared to your simulations. Also, sometimes simulators assume infinitely large ground plane, which is not the case in your design. This surely changes the freq response.
Not being an RF person, it would be nice to know how those hairpins act as a filter. Also knowing what the application for such a filter might be, would again be useful.
They are coupled resonators - each strip is a half wavelength long. Did Keysight give you an evaluation license? I've only used ADS when I worked at Qualcomm, Anritsu, and Z-Comm. By the way, I bought a license for Sonnet Software - I wouldn't mind optimizing your filters.
Excellent first effort. This method would be a realistic entryway for a designer making custom circuits except for the exorbitant cost of software. You could use this series for a new job application if you gave a sh...
@@jms019 consistency is a measurement of variation from any point in time. I would need to monitor over several months. The total thickness is two layers of prepreg and one layer of FR4. the manufacture of the boards already know these numbers and publish +/-10%
I needed a high tolerance on thickness and dielectric. I wanted to show modeling accuracy and needed to rule out other variables. FR4 is fine but I'm not sure what the variability would be
You just covered my masters thesis from back in '91 while I was at BT Labs. I was modeling thin-film circuits of gold on alumina substrates at 2 - 60GHz. I developed software from scratch that split the space into variable sized cubes to run the finite element analysis. I'd run the simulation and then based on the magnitude of any change in a cube would then split into 8 smaller cubes and re-run. The simulation took 2 weeks on a DEC Alpha of the time. I found it was really important to include a sealed metal box, which I had custom manufactured in the lab's machine shop, and to include that as part of the simulation. I've not run the Rogers materials but have heard great things about them though. We took the measurements on an HP 8510C network analyzer which was about as good as you could get back in the day.
The difference in the simulated and measured center frequency is probably just due to the difference in the dielectric constant you used in your simulation and the design value recommended by Rogers. The dielectric constant of RO4003C is 3.38 but the recommended value for design is 3.55. In your simulation increase your dielectric constant and you will see that the center frequency of your filter shifts to lower frequencies.
As the response is slightly too low in frequency you might want to try cutting or grinding off the ends of the hairpin filters - 1mm at a time . I had a similar hairpin filter for 1152MHz on FR4 board which was too low and had to do this to get the filter to play nicely. A great video as always.
I could move the center frequency but the top of the filter was no longer flat.
It's the mass of solder to the sma, I did these designs in cad and verified in prototypes. Had to have not even a fillet in the solder. None of the other team could match the minimum amount of solder joint.
"The blackest of the black arts in analog electronics", as CuriousMarc called microwave engineering. And now it's all over the place - in our routers, in our cells...
PCBWay has been reaching out to me offering sponsorship - as if my teeny tiny channel mattered to anyone, haha! I'll certainly try them for a bunch of projects.
Hi IMSAI Guy. I think the Er to use in MOM is more like 3.48 and it would shift down slightly. The response proves the model is great. Just the lengths are a little long because the Er is too low. Or PCB Way shipped you 4350 boards instead of 4003. 4350 is the most common today, but a little more ceramic content. You can get a precise calipers thickness of just the dielectric, then with the area of your traces just calculate the parallel plate capacitance you should have versus measuring the same with your LCR meter and you can see if they used the right material. Or with the area and measured capacitance value, solve for Er and see what you get. Whatever that Er is, enter that in GENESYS and you will probably see the frequency shift down a bit.
The copper thickness and roughness changes the effective dielectric constant which is most likely the reason for the frequency shift. Rogers actually have a good paper on the topic.
Also, the conductivity of the gold plating, which I assume is ENIG, can have its own effect on higher frequencies.
My takeaway? If you think about something like a military radar system and then the large expenditure in time and money that just Rogers spent in developing the materials that are used in order to make a small part of it work you start getting an understanding of why they cost so much.
Why not go back and run the numbers with 0.0021, see if that corrects the shift? Cliffhangers galore! 😭😅
yea at 11:00 is see the error
I did, it didn't mater
thanks for the feedback @@IMSAIGuy
With so many filters it might be interesting to see how a couple of them look in series. In parallel one will have to add splitters and combiners.
9:50 sometimes it might work to understand the frequency dependency while looking at temperature, - never mind.
Thanks! I always learn something new on your channel. N1BUG's channel has excellent microwave projects on it and has designed some filters using copper pipe caps.
Do more of this !
PCB with Rogers dielectric? So fancy that I need a second monocle :D But in all seriousness, measuring S11 (hopefully it is better than -20dB) really helps to determine what is the cause of the frequency response shift (interesting as a follow-up video). Also check out the losses in the passband , it seems that you measured 4dB of in-band attenuation which is a lot compared to your simulations.
Also, sometimes simulators assume infinitely large ground plane, which is not the case in your design. This surely changes the freq response.
I bet the launches from the sma connectors need to run into a 45 degree cut corner first rather than perpendicular
Some "hacks" that can be done are adding copper tape pieces or layers of teflon tape to change the dieletric constant on the top...
Great series, thanks. Now I wonder what you will use them for, or if you will give them away?
Not being an RF person, it would be nice to know how those hairpins act as a filter. Also knowing what the application for such a filter might be, would again be useful.
Off by 50 Mhz for first run at a first run is pretty good. Zarquan! The Chinese would just ship it if it were off by 150mhz.
Hi, Great video. How are you measuring 2.4Ghz on a 2.1Ghz max analyzer.
it has a firmware upgrade to 3.2GHz
@@IMSAIGuy cool, how do you get the firmware. Was it a simple upgrade.
Would you like another shadow? I'm hungry, eager, and not afraid to fail on the way to figuring something out.
Well now I know !....cheers !
The design looks like a SAW filter. And may be you can test again by adding a metal shield with grounding on the top side.
I want to look at the SA graph while you go at it with a scalpel!
I could move the center frequency but the top of the filter was no longer flat.
They are coupled resonators - each strip is a half wavelength long. Did Keysight give you an evaluation license? I've only used ADS when I worked at Qualcomm, Anritsu, and Z-Comm. By the way, I bought a license for Sonnet Software - I wouldn't mind optimizing your filters.
Little baby J poles on the ends, what is the input impedance.
the filter was designed for 50ohm input/output
Excellent first effort. This method would be a realistic entryway for a designer making custom circuits except for the exorbitant cost of software.
You could use this series for a new job application if you gave a sh...
You can cut it to increase frequency
I could move the center frequency but the top of the filter was no longer flat.
Go back to the modelling software, try that correct figure and see if it matches reality please. Also is an FR4 design possible ?
FR4 has +/-10% thickness tolerance. so hard to predict.
Measure your actual PCBWAY FR4 and I’ll bet the thickness is way more consistent than that.
@@jms019 consistency is a measurement of variation from any point in time. I would need to monitor over several months. The total thickness is two layers of prepreg and one layer of FR4. the manufacture of the boards already know these numbers and publish +/-10%
Waste of money at those kinds of frequencies. Build the same filter for FR4 and test them side by side?
I needed a high tolerance on thickness and dielectric. I wanted to show modeling accuracy and needed to rule out other variables. FR4 is fine but I'm not sure what the variability would be
@@IMSAIGuy Most people watching probably understood that, the comment wasn't derisive. As for FR4 it would be interested to see the comparison ...
Definitely would be nice to see the result on FR4. What would that cost? $20?
@@chiraldude I believe that PCBWay sponsors the boards?
@@chiraldude the problem with FR4 is the thickness tolerance is +/-10% so you might get lucky or not