Why are my center frequnceys off on my LC filters?

[koffp]

I recently created 2 LC cheb constant K filters, on for 3rd order 115-155 Mhz and one for 5th order 496-560 Mhz. When made the actual physical PCBs I got different results, the first one centered at 125 Mhz instead of 135 Mhz and the second centered at 470 Mhz instead of 530 Mhz, theiur bandwidths are about correct though. Has anyone else come across this problem? I made sure to use silicon RF components with high self resonant frequency and high q around 500.. I dont know what else I could do to negate parasitic impedance, I made the traces as efficient as possible, only thing left would be to change the trace thickness(im using 8 mils) and the trace substance(im using standard copper).

I'm about to resort to attempting calculating micro-strip dimensions by hand and manually drawing them in orcad since my company doesnt have ADS or anything like that..

Could it be something wrong with my parts??


Here are the parts:

530 Mhz 5th order cheb:
C1: https://www.mouser.com/ProductDetail/ATC-Kyocera-AVX/600S120FW250T?qs=ZTdx6reOWK%2FXUH41hO8rGw%3D%3D
L1:https://www.digikey.com/en/products/detail/kyocera-avx/HLC02320HTTR/3528327
C2:https://www.mouser.com/ProductDetail/Kyocera-AVX/0201ZK4R0BBSTR?qs=k2FEcAV%252B92Yoi7sJsjfPBA%3D%3D
C3:https://www.mouser.com/ProductDetail/ATC-Kyocera-AVX/600F240JT250XT?qs=AdTq2PFIamn48Yj%2Fkca%252BNQ%3D%3D
L2:https://www.mouser.com/ProductDetail/Murata-Electronics/LQW18AN34NG80D?qs=hNud%2FORuBR3cm0mPuhqMJw%3D%3D
C4:https://www.mouser.com/ProductDetail/Kyocera-AVX/0201ZJ3R5BBSTR%5C500?qs=%252Br8cYUmWfSeE47%2FWazW4EQ%3D%3D

135 MHz 3rd order cheb:
C1:https://www.mouser.com/ProductDetail/ATC-Kyocera-AVX/600F820FT250XT?qs=AdTq2PFIamkUclHcJDFISw%3D%3D
L1:https://www.mouser.com/ProductDetail/Murata-Electronics/LQW15AN17NG80D?qs=hNud%2FORuBR3E4YmylizoPA%3D%3D
C2:https://www.digikey.com/en/products/detail/kemet/CBR06C609CAGAC/3473936
L2:https://www.digikey.com/en/products/detail/pulse-electronics-power/PA2607-231NLT/2267054


Note:
I realize my sober is sloppy, escecially on the 5th order cheb butthose are old picture, I dont have the pictures but I made other cirucits were my solder was perfect and I managed to move the cneter frequncy from 470-475Mhz but thats not even close to 530 Mhz which I need it to be

[koffp]

Here are sum more relevant pics

[bob91343]

How can you be sure you have the correct element values?  Simulating a circuit is all well and good but since your results are so close to what you want, just trim the values and you should be good to go.  The error you obtain isn't very great, and is probably attributable to component tolerances.

[David Hess]

Isn't the parasitic capacitance from the layout enough to explain the shift in frequency?

[BigBoss]

Because your layout is not appropriate. Why ?
-There is no VIA around GND Connections and you should place VIA stitching all around. This is important
-Components are too close to GND side copper pours. They will create stray capacitances. Especially high impedance nodes are very sensitive to these stray capacitances. ( Midpoint of L-C series circuits )
You should place spaces on the main line.( almost open )
-Component Tolerances play very important role on the Filter's Response therefore you should use trimmer capacitors or variable inductors where possible.
-You have to simulate whole circuit with a EM simulator with realistic and modelled components
-You have also to do the calibration on the same substrate with through line. ( for S21)
-Monte Carlo analysis will give you an insight about shifts on frequency axis.
-You used thermal relief for GND connections of the connectors. This is wrong. Tie them directly.
-Finally you should open the copper pour just underneath of the connectors' pins and main line because it will create parallel stray capacitor between nodes and GND. It will badly shift the center frequency.

[rfclown]

You need to get s-parameter models of the components and simulate with that. If you can't get them, find components for which you can get models for. That and adequate ground (vias). You do have your components close together, which is great. So many people show circuits that have long traces between components and wonder why things are WAY off.

I use QUCS which is free and will do s-parameter simulations. It seems to be modeled after ADS which I used when I worked at a big company. My present day job has one ADS license which we are all expected share (with a hardware dongle) because the CEO and CTO are *heads. I skip the nonsense and use QUCS.

[koffp]

I simulated with worst case senerios, it doesnt account for the shift..

[koffp]

BigBoss

Thanks for the advice but I dont believe that will account for such a large shift(I admit im a greenhorn and ur probably right and im probably wrong). I stitched it on other circuits and it didnt change it significantly. The traces cant be causing more than like 5-10nh induction(when i did arough hand calculation) which wont explaine such a shift.. I havent tried making the gound planes even farther yet though.. I made the same circuit with open copper pour and it didnt change anything significantly..

Can you tell me what you mean by
"You should place spaces on the main line.( almost open )"

I didnt calibrate my network analyzer for this specific run but i will try, thanks for the advice..

[rfclown]

I simulated with worst case senerios, it doesnt account for the shift..

Worst case senerios don't account for the fact that a capacitor (or inductor) isn't just a capacitor (or inductor) with a +/- tolerance. There is parasitic inductance, capacitance, and resistance in the package. You MUST model these. The easiest way to do this is with the manufacturer's s-parameter models.

I simulated your 3rd order with ideal components, and manufacturer's models. I couldn't get to the ATC models (forgot my password), so I used a Johanson part for the 82 pF cap (Hi Q RF component). The Pulse 240nH inductor is completely wrong for this application. In your picture, it is either on the other side of the board, or something is amiss with this one. I picked a small inductor with ferrite (which is wrong here, but yours has ferrite). You can see how the simulation shows a shift in center frequency using component models.

For the 5th order I didn't go to the trouble of getting models for all the parts, just a few. Your schematic didn't show a 12pF shunt on the input, but I added it. You seem concerned with finding parts that are the exact values you want (instead of using standard values). This leads to choosing some inappropriate parts (like the Pulse inductor in the 3rd order design).  With this circuit, I find the 32nH inductor very odd. It has a DCR of 2.2 ohms, which is absurdly high. I don't know what this part is intended for. I kept the ideal component in the simulation for this one because 32 nH isn't a standard value, and if I substituted the 34 nH Murata part (which is an appropriate part) you might think that the frequency shift was due L1 being the wrong inductance.

You MUST model the components correctly. Like I said previously, your pictures show that you have minimized trace lengths between components, which is a very good thing. If you don't do that, then you MUST model the interconnects between components. These "MUST"s related to high frequency design. SMT components might have roughly 1nH series package parasitic. At low frequency, this might be a don't care, but the higher in frequency you go, it can be a deal breaker.

[BigBoss]

BigBoss

Thanks for the advice but I dont believe that will account for such a large shift(I admit im a greenhorn and ur probably right and im probably wrong). I stitched it on other circuits and it didnt change it significantly. The traces cant be causing more than like 5-10nh induction(when i did arough hand calculation) which wont explaine such a shift.. I havent tried making the gound planes even farther yet though.. I made the same circuit with open copper pour and it didnt change anything significantly..

Can you tell me what you mean by
"You should place spaces on the main line.( almost open )"

I didnt calibrate my network analyzer for this specific run but i will try, thanks for the advice..

Yes, it will indeed shift.
Look at your filter's Response with ideal elements.
And then I have attributed 5% deviation for each elements and this deviation is Gaussian Distribution type.
As you can see, filter's response is very pretty shifted to left and right due to tolerances.
You can also consider stray elements on the layout.
So your filter is quite sensitive to components' tolerances. ( plus layout effects )
If this design will be volume production, I recommend you to change the filter topology, modify the layout and use tight tolerance components.
Otherwise use trimmer caps and variable inductors to tune the filter.
If your budget is enough, I recommend you to use Microstrip Based Filter ( Rogers Duroid 6006 )

[Bud]

Component values may change with frequency. You can use a VNA to measure each type of components used across the design frequency range. You can simulate until cows come home but this practical measurement will give you real world insights.

[Joel_Dunsmore]

Late to the game, but in case someone else looks:

Every SMT capacitor needs to replaced with a series LC where L is on the order of 2.5 nH (+-50% I'd guess) and every inductor needs to be replaced with a parallel LC where C is on the order of 0.2 pF; and every resistor (not that there is one in a filter) is replaced with a series RL, shunted by C, of the same value.  These typically cause a filter center freq to shift down 30-50% in freq, in rough numbers.  All the L's appear bigger as do all the C's, so everything goes lower.  Berkeley's EE 142 has this as a principal lab (at least when I co-taught it).

[Hamelec]

maybe you should think about if there is a really need for filters - maybe a lowpass and a highpass would be enough..?

[MathWizard]

Because your layout is not appropriate. Why ?

-Components are too close to GND side copper pours. They will create stray capacitances. Especially high impedance nodes are very sensitive to these stray capacitances. ( Midpoint of L-C series circuits )
You should place spaces on the main line.( almost open )

Would one possible fluid dynamics equivalent be like in a piping system, if in the pipes with the greatest flow restriction, you were adding and taking away indentations in real time, of the pipes, which would make the water pressure to rise and fall accordingly  ? Would that be like stray capacitance in a high impedance section ?

[ejeffrey]

The hydraulic analogy for electronics is... only qualitatively helpful in the best of situations.  Applying it to parasitics in a multi-stage bandpass filter is unlikely to be helpful at all.

[Laurencewilliams]

I recently created 2 LC cheb constant K filters, on for 3rd order 115-155 Mhz and one for 5th order 496-560 Mhz. When made the actual physical PCBs I got different results, the first one centered at 125 Mhz instead of 135 Mhz and the second centered at 470 Mhz instead of 530 Mhz, theiur bandwidths are about correct though. Has anyone else come across this problem? I made sure to use silicon RF components with high self resonant frequency and high q around 500.. I dont know what else I could do to negate parasitic impedance, I made the traces as efficient as possible, only thing left would be to change the trace thickness(im using 8 mils) and the trace substance(im using standard copper).

I'm about to resort to attempting calculating micro-strip dimensions by hand and manually drawing them in orcad since my company doesnt have ADS or anything like that..

Could it be something wrong with my parts??


Here are the parts:

530 Mhz 5th order cheb:
C1: https://www.mouser.com/ProductDetail/ATC-Kyocera-AVX/600S120FW250T?qs=ZTdx6reOWK%2FXUH41hO8rGw%3D%3D
L1:https://www.digikey.com/en/products/detail/kyocera-avx/HLC02320HTTR/3528327
C2:https://www.mouser.com/ProductDetail/Kyocera-AVX/0201ZK4R0BBSTR?qs=k2FEcAV%252B92Yoi7sJsjfPBA%3D%3D
C3:https://www.mouser.com/ProductDetail/ATC-Kyocera-AVX/600F240JT250XT?qs=AdTq2PFIamn48Yj%2Fkca%252BNQ%3D%3D
L2:https://www.mouser.com/ProductDetail/Murata-Electronics/LQW18AN34NG80D?qs=hNud%2FORuBR3cm0mPuhqMJw%3D%3D
C4:https://www.mouser.com/ProductDetail/Kyocera-AVX/0201ZJ3R5BBSTR%5C500?qs=%252Br8cYUmWfSeE47%2FWazW4EQ%3D%3D

135 MHz 3rd order cheb:
C1:https://www.mouser.com/ProductDetail/ATC-Kyocera-AVX/600F820FT250XT?qs=AdTq2PFIamkUclHcJDFISw%3D%3D
L1:https://www.mouser.com/ProductDetail/Murata-Electronics/LQW15AN17NG80D?qs=hNud%2FORuBR3E4YmylizoPA%3D%3D
C2:https://www.digikey.com/en/products/detail/kemet/CBR06C609CAGAC/3473936
L2:https://www.digikey.com/en/products/detail/pulse-electronics-power/PA2607-231NLT/2267054


Note:
I realize my sober is sloppy, escecially on the 5th order cheb butthose are old picture, I dont have the pictures but I made other cirucits were my solder was perfect and I managed to move the cneter frequncy from 470-475Mhz but thats not even close to 530 Mhz which I need it to be

It sounds like you encountered an issue with your LC filter designs not matching the desired results. This could be due to several factors, such as parasitic elements, process variations, and measurement errors.

One possible solution to reducing the impact of parasitic elements is to increase the trace width and thickness. This increases the inductance of the trace, reducing its parasitic capacitance, and reducing its resonant frequency. However, increasing the trace width and thickness also increases the parasitic inductance and resistance, so it is a trade-off.

Another solution is to implement an active filter, such as an op-amp based filter, instead of a passive filter. Active filters have the advantage of having better control over their response and higher stability, although they are typically more complex and consume more power.

It's also worth considering the impact of process variations. The parameters of the components you used, such as the inductors and capacitors, may have variations due to the manufacturing process, and this can affect the overall performance of your filter. To minimize the impact of process variations, you can use higher-quality components or implement design margins to account for the variations.