[Design Review/PCB] 137 MHz Front-End optimized for Strong FM Interference

[daarkess11]

Hi everyone!

I’m a 2nd-year electronics engineering student working on my first RF project for a future ground station. Since this is my first time designing an RF Front-End, I’d really appreciate some feedback from the community before I send it to fab, especially with the PCB layout since I’m sure there will be a lot of room for improvement.
Just ~4.3 km from a 240 kW EIRP FM tower (raw FM power at my antenna is around -1.3 dBm in worst-case).

Design:
Input Filter: 5th-order Cauer notch [88–108 MHz] (1.09 dB IL @ 137 MHz).
LNA: PGA-103+ E-PHEMT Mini-Circuits
Stability (10 MHz – 4 GHz): R+L shunt.
Output Filter: 3rd-order Butterworth bandpass (2.5 dB IL @ 137 MHz, compensated with Friis)
Used High Q High SRF components and 2% tolerances for the Cauer.

Simulated performance (with .s2p files): ~1.84 dB NF,~22 dB cascaded gain, -13.4dB for S11.

Two things I'm not sure about:
I know Cauer filters are pretty sensitive to component tolerances. I used 2% across the board for the notch but i'm not sure if that's enough in practice. Anyone run into this before?

I can't orient the inductors at 90º in the input filter, so I just alternated their orientation (LC/CL/LC) trying to reduce magnetic coupling but i don't know if that's enough either.

All files (schematic, KiCad & sims) are open-source on GitHub if you want a deeper look.
Thanks for reading, feedback is really appreciated!

[tchicago]

I think it is not a good idea to run ground underneath capacitors and inductors that are in series with your signal line. Unless you want to add an undetermined (1-3) pF's of capacitance to the ground there. But since you mention that the component tolerances are tight, you probably don't want that.

[BigBoss]

I don't know the filter response, but you can use trimmer capacitors in this band.
Because you used a pretty strict filer topology. It's generally hard to maintain to get an acceptable response from this kind of filter due to components' tolerances, as you said.
If I were you, I would use adjustable inductors and capacitors to tune the filter.
If you are able to access ADS or a similar RF simulator, I would do a "sensitivity analysis" to determine which components are critical in this filter.
Then you replace those components with adjustable ones. Finally, a Monte Carlo simulation will give you an insight into the filter characteristic.
OR
Loosen your constraint and decrease the filter order that will be less sensitive to the tolerances.

Edit: Your filter does not seem appropriate. Is the schematic correct??

[G0HZU]

Yes, the band reject filter values look a bit strange in places.

There's a DC blocking capacitor missing at L8. Without it the PGA103 will be biased off.

There don't appear to be enough ground vias around the PGA103+.

It might be worth it to relieve some solder resist in places around the notch filter to allow screens or shields to be attached and played with if necessary.

The Q of the coils isn't going to be that great at 100 MHz but maybe this is still OK in your case?

Will the regulator get hot at 100 mA current? I don't know what the input PSU voltage is.

Adjustable coils would be good IMO but a lot depends on what is acceptable in terms of rejection if the circuit turns out to be a bit detuned with the fixed coils..

[G0HZU]

If you want to stick with fixed component values then maybe consider using two caps in parallel for some or all of the resonator caps in the band reject filter. This will allow some 'select on test' fine tuning if this is a one off design.

[rf-fil]

As others said, if this is a one-off or a small batch, I'd add trimmer capacitors to all the resonators in both filters. It's then easy to tune them up on the bench.

One other thing is the inductor Q. It looks you're using small chip inductors, which have very low Q. Having so many low-Q inductors in your input filter may result in a much worse noise figure than you think. Probably worth including in your simulation, if you haven't yet.

The observation that coupling between inductors in the input filter will impact its performance is also probably right. Specifically, the band stop depth won't be anywhere near the ideal value, unless you add shielding around each filter section. You can get small pre-made shielding cans at Digikey etc.


Oh.. and for your output filter.. you could just use a SAW filter. There are some SAW filters available for that frequency. It would completely eliminate any trouble with LC output filter design. That's the "professional" solution to this.

[dobsonr741]

I would put the Cauer onto two separated chambers, and the LNA to a third. Did you simulate with real world component parameters? The SMA inductors will be far from ideal. And an input protection, with antiparallel diodes would make sense. Possibly a switchable input attenuator, too.

[daarkess11]

Thanks everyone for taking the time to go through the design, really appreciate all the input!

1. Filter Simulation & S-Parameters (@BigBoss, @G0HZU)
I should have clarified this upfront: the notch filter was designed and tuned in Qucs-S using vendor Touchstone (.s2p) measured models for every component (Murata GRM series capacitors and Coilcraft CS/LS series wirewound inductors) rather than ideal parts. The nominal values and resonant tanks already compensate for component ESR, parasitics, and Self-Resonant Frequencies (SRF), which is why simulating with ideal lumped elements at these exact values shows a distorted response. I'm running sensitivity and Monte Carlo sweeps next to check notch depth stability against standard component tolerances.

2. Tuning Strategy (@G0HZU, @BigBoss, @rf-fil)
Thanks for the trimmer suggestions (@BigBoss, @rf-fil). Given the Q degradation compared to C0G/NP0 ceramics at this frequency, I'll go with @G0HZU's idea instead: adding secondary parallel pads across the critical tank capacitors for bench "select-on-test" trimming with high-Q discrete caps.

3. Inductor Coupling & Spacing Question (@G0HZU, @rf-fil, @tchicago)
Ground under series line (@tchicago): Noted on the parasitic capacitance risk, I'll review the ground plane relief under the series CPWG pads/traces.
Coil coupling question (@G0HZU,@rf-fil: Since I'd prefer to avoid adding individual shielding cans if possible: would increasing the spacing between filter tanks (e.g. 4–5 mm separation with continuous ground between them) provide enough isolation or would shielding cans still be strictly necessary in practice?

4. Quick question on the R19+L8 branch (@G0HZU)
Regarding the DC block on the R19+L9 branch: I tied it straight to GND, my assumption was that since pin 1 is just RF-IN and the DC supply comes in through pin 3, that branch didn't need isolating. Could you help me understand why grounding it there without a DC block cap would bias off the PGA-103+? Trying to understand the internal bias behavior better.

5. PGA-103+ Ground Vias & LDO Thermals (@G0HZU)
Ground vias: Good point, the PGA ground paddle already has 6 vias (0.2mm drill), trying to minimize solder wicking, but I don't have many around the footprint for a low-inductance RF ground return. I'll be adding more around it.
LDO dissipation: The LDO tab (5V net) has a dedicated top copper polygon for heat spreading, I'll size it once I lock down the final input voltage for the ~100 mA.
Solder mask relief: Good call on leaving that flexibility for shields later, I'll keep it in mind depending on how testing goes.

6. Output Stage (@rf-fil)
Swapping the discrete output bandpass filter for a dedicated 137 MHz SAW filter is a great suggestion. I`ll look into available parts and footprints now.

7. Chambering & Input Protection (@dobsonr741)
Good point on chambering, I'll look into splitting the Cauer notch, LNA and output BPF into three separate shielded compartments instead of just spacing them out. On real component sims, yes, same as mentioned above, all discretes were simulated using vendor .s2p files rather than ideal models, so SRF/ESR/Q are already accounted for. Input protection is something I'd missed completely though, antiparallel diodes at the input plus a switchable attenuator both make sense. I'll look into low-capacitance diode options that won't mess with the notch filter's input impedance.

Thanks again to all of you, this has been super helpful for a first RF design.

[fourfathom]

I've built similar preamps and filters using 5% SMT inductors and capacitors, and the Monte Carlo simulations as well as the actual performance of several builds of about 50 units each has been quite good.  I use a four-stage elliptical (Cauer) filter with some of the notch frequencies adjusted to null out the 88-108 MHz FM band.  The inductors are ceramic core (solenoid-wound) in the "1008" size.  I wouldn't call these particularly high-Q, but they're better than a lot of other SMT inductors.

You will probably find that the self-resonant frequency of the notch capacitors will result is significant spikes in the stopband response (in the 200 - 500 MHz region).  I've added a simple C/L/C Pi network post-filter to minimize the SRF feedthrough.

I try to place adjacent inductors with 90-degree orientation to minimize inductive coupling, but haven't really looked into this -- perhaps this makes no difference with practical inductor separation.  I also have groundplane relief under the parallel L/C sections, but I've also tried it with solid groundplane and not seen a big effect (at my frequencies of interest).

Here's a document for a 60 MHz filter / preamp that I sell, with schematic and response plots: https://turnislandsystems.com/wp-content/uploads/2025/06/FP6m-Manual-1.pdf

Your extremely strong FM environment may require more effort that I've done, but I think that you can make usable high-performance filters using 5% parts, without needing tuning options.

[daarkess11]

@BigBoss Thanks a lot for taking the time to put this schematic together in ADS, really appreciate the help and the effort!

I recreated your design in Qucs-S using the vendor .s2p models for the Coilcraft and Murata parts. The rejection response looks great, and the slight IL increase (from ~1.09 dB to ~1.4–1.5 dB) is totally fine for my VHF link margin.  I actually prefer your approach over my initial design, which definitely suffered from an excessive component count.

When simulating the filter cascaded with the LNA stage, S11 sits at around -9.2 dB at 137 MHz. I’ll manually tweak the end resonator values and coupling to see if I can pull return loss below -15 dB without distorting the notch shape or hurting IL.

Really looking forward to seeing your MC run in ADS. Thanks again! I've attached my schematic and plot below too.

@fourfathom:
Really useful to hear this from actual production runs, thanks. Good to know 5% parts have worked out fine in practice, gives me more confidence in loosening the 2% constraint if my sensitivity analysis allows it, and the SRF spike warning in the 200-500 MHz stopband is something I honestly hadn't considered at all, I'll check my simulated response there and look into that C/L/C Pi network if I see the same issue.
Also interesting that you've tried both relieved and solid groundplane under the L/C sections without seeing much difference, same with the 90-degree orientation, curious if that'll hold at 137 MHz too. Thanks for the FP6m manual! I'll go through the schematic and plots.

[G0HZU]

Quick question on the R19+L8 branch (@G0HZU)
Regarding the DC block on the R19+L9 branch: I tied it straight to GND, my assumption was that since pin 1 is just RF-IN and the DC supply comes in through pin 3, that branch didn't need isolating. Could you help me understand why grounding it there without a DC block cap would bias off the PGA-103+? Trying to understand the internal bias behavior better.

The PGA103+ is an enhancement mode PHEMT device so you can expect to see about 0.6V DC bias at the gate pin when running normally. If there is a 50 ohm path to ground at DC via L8 then this will starve the PGA103+ of bias and it might end up with a drain current of maybe less than 1 mA (instead of the expected 100 mA).

Also, 137 MHz is a long way away from the VHF FM band so you could fit a tunable 137 MHz BPF here using tuneable (and screened) coils if you have the height and budget for the increased cost. I'm not sure you need a huge rejection at the FM broadcast frequency so a low loss 137 MHz BPF could be fitted here and it would provide selectivity against other VHF and UHF traffic locally.

[daarkess11]

G0HZU Thanks, that makes a lot of sense, I hadn't considered the internal bias network also references pin 1.

Also, 137 MHz is a long way away from the VHF FM band so you could fit a tunable 137 MHz BPF here using tuneable (and screened) coils if you have the height and budget for the increased cost. I'm not sure you need a huge rejection at the FM broadcast frequency so a low loss 137 MHz BPF could be fitted here and it would provide selectivity against other VHF and UHF traffic locally.

On the BPF suggestion, I get your reasoning and it's a fair point given the distance between 137 MHz and the FM band, but one thing I should've flagged earlier: my -1.3 dBm number was just a quick calculation using the main 240 kW transmitter, and I'd already added some extra margin to the notch, figuring there'd probably be other stations on that same tower.
I checked the official data for that site and there are 17 more stations (8 to 50 kW ERP each). Summing their power gives around +4.2 dBm at my antenna (worst-case), and that's likely still a lower bound since the dataset I used may not cover every station at that site.

With that much FM power hitting the input, I think sticking with the Cauer notch is probably my safest bet to avoid saturating the LNA.

BigBoss Thanks for running the MC simulation, really appreciate it!

It's great to see that the insertion loss at 137 MHz stays relatively clean and stable across all trials. I'll tweak the nominal values to give the 108 MHz edge more margin and post an update soon with the new IL, S11, and rejection results.

Thanks again!

[G0HZU]

It will be interesting to see how much in the way of jelly wrestling you will end up doing when trying to set up the 5 SMD coils in that notch filter. The isolation between them could cause you some issues when adjusting the notch frequencies as they can interact with each other. Normally it's best to isolate the sections in a notch filter using shielding or screening.

OK, I know you don't want to use a BPF but for a bit of fun I modelled a two resonator BPF using shielded coils. It only has three different cap values and it uses just two resonator coils each with the same part number but these are adjustable 10mm coils.

I've modelled the coupling between the coils (because it won't be infinite despite the shielding) and here is the result. This filter only uses two inductors and not five. I think the stopband performance is OK. A three resonator version would obviously be better.

No jelly wrestling would be required when adjusting this filter

[daarkess11]

Thanks @G0HZU, that 2-resonator response looks really good!
Would you mind sharing the schematic and the 10mm coil part number? I'd like to test that topology against the rest of the chain.
I'll look into both options soon to see whether 10mm coils are practical for my build or if sticking to SMD makes more sense.

I'll also work on an SMD layout considering all the input received here so far, using individual shielding and parallel tuning pads to see if I can reduce that "jelly wrestling" and what kind of isolation I can really achieve.

Thanks again for taking the time to model this!

[David Hess]

I have done this sort of thing in the past by using a helical resonator notch filter.

[G0HZU]

In terms of gaining experience, it's probably best to continue with the multi-resonator notch filter. It would be interesting to see how well it behaves in terms of tuning due to the finite isolation between the SMD resonators. You may experience what I refer to as jelly wrestling as each notch squishes the others sideways as you try and tune them. However, you might be OK and it might be well behaved in this respect.

I just had a go at making the dual resonator BPF. I didn't etch or mill a proper PCB and just built it in a couple of minutes on a bare sheet of copper PCB. I actually got slightly better isolation than I was expecting. However, the result is quite close to the simulation. See the image below. The only tuning required was to the hex tuning slugs in the 10mm coils using an insulated tuning tool.

I used less than ideal 10mm coils from Coilcraft in that I used the 142 series (shielded can fitted) and these have close-wound turns which is not ideal for Q.

If I had space-wound coils here I'd expect to see the insertion loss go below 1dB. However, it as about 1.2 dB as you can see below.

It really is worth it to try and get as much isolation as possible between the resonators. This puts you in control of where the notches appear rather than having the squishy jelly wrestling effect where they over couple and push each other about as you tune one of the resonators.

Try building the notch version first. The filter I've built below is very crude and far from ideal but it was meant as a demo to show what can be achieved with just two (tunable) inductors and a few fixed capacitors. There are no other forms of inductor in the filter, just the two Coilcraft 10mm types and they are both the same part number.

[Uunoctium]

I'm going to make myself unpopular here by suggesting a simple SAW filter for 137.5 MHz (NOAA weather satellites).
e.g.
https://elekitsorparts.com/product/137mhz-saw-filter-137mhz-saw-filter-with-lna-137mhz-saw-filter-for-noaa-satellites/

[G0HZU]

Yes, someone has already suggested using SAW filters. I think a fairly basic HPF ahead of the SAW might be a good thing. I'm still not convinced why so much rejection is being searched for (at 88-108 MHz) so a SAW filter might be a quick and easy solution here. I don't know what system noise figure is acceptable either.

[G0HZU]

FWIW here's the simulated response of a triple inductor BPF. It has loads of stopband rejection across 88-108 MHz but do you really need this much rejection? I'd have thought that 50 dB would be easily enough. The simulation below assumes space-wound shielded 10mm inductors.

[daarkess11]

@Uunoctium @G0HZU Totally fair point regarding the SAW filters. I'm aware there are commercial modules available for NOAA reception, but as a 2nd-year student, the primary goal here is getting hands-on experience designing, simulating and laying out discrete RF stages from scratch, while also targeting a lower NF than what an input SAW would give.

@G0HZU Regarding the triple BPF and required rejection: you're right, I definitely don't need that much rejection well beyond 50dB. With ~4.2dBm of FM power at the antenna (being a lower bound), a rejection of around 30dB to 35dB across 88–108 MHz is my comfortable target.

Thanks again for running that simulation, and I really appreciate you taking the time to actually solder and bench-test that dual-resonator build on copper sheet.

[Uunoctium]

tl;dr
I suggest an helical filter in separate enclosure. Cheap, easy to buildt. On p. 31 one can see, that it will match the 137.5MHz without altering.
https://www.f4htb.fr/wp-content/uploads/2023/06/VHF-Low-Loss-Band-Pass-Helical-Filter-for-145-MHz-English-new.pdf

And don't expect, an commercial filter like these Daden-Anthony Assoc. from last century having an better internal appearance.
The upper ist a narrow tuned to 76.9MHz, the lower has a wider BW from 89 to 90.5MHz

[David Hess]

I suggest an helical filter in separate enclosure. Cheap, easy to buildt. On p. 31 one can see, that it will match the 137.5MHz without altering.

That was my suggestion also because they provide incredible performance with easy construction, but a helical filter is not conducive to the objective of printed circuit board construction, at least at VHF.

[iMo]

.. another question is whether an "LNA" fits in such a design actually. I would rather focus on the first mixer (if any) and its parameters.
The input filter will never be ideal as off a simulation.
So a decent filter and a very good mixer (like with a high linearity and a high IP3)..

[David Hess]

.. another question is whether an "LNA" fits in such a design actually. I would rather focus on the first mixer (if any) and its parameters.
The input filter will never be ideal as off a simulation.
So a decent filter and a very good mixer (like with a high linearity and a high IP3)..

Unlike at HF below 30 MHz, at VHF the noise level is low enough that an RF amplifier will improve sensitivity, however superheterodyne receivers always include an RF amplifier to block local oscillator leakage from the first mixer stage.

If maximum dynamic range is required and some loss of sensitivity is acceptable, and you do not care about local oscillator leakage, then the RF amplifier can be left out.

[G0HZU]

I kind of assumed this was for NOAA weather satellite stuff so the noise figure of the LNA should be quite low. If I had to comment on the choice of the PGA103+ device then I'd say that it is quite hungry in that it requires 100 mA at 5V and it is a broadband device. It also produces about 25 dB gain at 137 MHz which is quite high.

Also, filter simulation can be quite representative of the real hardware, especially at VHF and a lot depends on the experience of the engineer and the tools and models used. I've been designing RF filters for over 35 years so I'm usually quite good at simulating filter designs