Matching a 2-port (typical) SAW filter to 50 ohms

[wb0gaz]

This is about how to get started developing/testing RF matching sections for a 2-port device, when neither port has a working matching network, so the "second" port is not yet properly terminated.

INITIAL SPECIFICS:

The device in this case is a Tai Saw TA0245A SAW filter (139 MHz, PDF datasheet attached.) The ports are not natively 50 ohms (they're basically low impedance dominated by capacitive reactance.)

My questions revolve around this: Can the ports be matched separately? When the SAW filter is in passband (this device has insertion loss about 5 dB at center frequency), the getting-started condition  is that neither port is matched. Does this complicate measurements "looking into" the one port when the other port is open/shorted/(what should I do here?)

ADDITIONAL DETAILS:

The datasheet indicates equivalent circuit for input port is 390 ohms || 300 pF.
The datasheet indicates equivalent circuit for output port is 1164 ohms || 374 pF.

The datasheet shows a representative input  match section with 12 pF series followed by 56 nH shunt at the port.
The datasheet shows a representative input  match section with 9 pF series followed by 56 nH shunt at the port.

I'm testing this on a small bit of double sided/plated thru FR4 (matching the device footprint, so the 6 remaining contact pads have individual vias to ground.) Input/Output paths are sized for 0603 components (shunt, series, shunt "pi" physical arrangement), but are otherwise very short (distance between the device and the match components are a few mm, then a few mm to SMA ports on the test module.)

My initial construction implemented the datasheet's suggested matching section. The filter S21 (insertion) resembles the datasheet, but insertion losses are several dB higher than expected (as I'm at the upper edge of the device passband, the added several dB there make a difference; group delay there won't be an issue.)

Using a VNA (calibrated at the SMA connectors of the small test module) at the device center frequency (139 MHz), with the datasheet matching sections installed, the input port yields 37 ohms || 78 pF; at the output port 16.3 ohms || 58 pF equivalence.

I have made a trial run (just on PC, not building anything yet) using

https://www.analog.com/en/resources/interactive-design-tools/rf-impedance-matching-calculator.html

This has input data options that exactly mirror the physical device requirements (device port is expressed as parallel RC combination, specifying R, C and F).

I'm in the "look before you leap" state right now; the analog.com page suggests match sections, but I want to be sure I'm considering (if necessary) the effect of the second device port properly (that is, within the active passband of the filter.)

Thank you for any comments, questions, or requests for clarification!

Dave

[EggertEnjoyer123]

What power are you putting in with the VNA?
The datasheet says that you can only put in at most 0 dBm (and that's the absolute maximum, so presumably you should test with -10 dBm or less).

[RFDx]

The datasheet indicates equivalent circuit for input port is 390 ohms || 300 pF.
The datasheet indicates equivalent circuit for output port is 1164 ohms || 374 pF.

390/1164 Ohm could be possible, the 300/374pF capacitive part is obviously an error in the datasheet.

[G0HZU]

Golledge have a similar 139MHz SAW filter datasheet here:

https://www.golledge.com/media/1803/ma05254.pdf

Golledge are UK based and I've had good service from Golledge quite a few times over the years at work.

The input impedance is given as 210R || 11pF for both the input and the output at 139MHz. If this is the same SAW filter as yours and you already have the 56nH inductors then maybe try the capacitive tap network shown below on both input and output?

15pF shunt then 22pF series and 56nH shunt as shown in the diagram below.

[wb0gaz]

EggertEnjoyer123 - VNA output (this is a nanovna) is in the -10 dBm range. In the application, -7 dBm (0 dBm then through a mixer with ~7 dBm insertion loss) is the most the device would see. That said, I'm glad you mentioned this as I had glanced past that specification for this device (most other SAW datasheets from Tai Saw I've seen have a higher power tolerance.)

RFDx - I was thinking there might be data sheet error in one respect or another, as the example given did not work as expected.

G0HZU - I've used Gollege from time to time in the past (=when I was employed!), results were excellent and they were altogether helpful, however, none of the small-lot purchases (for prototyping) led to production. In this era I've not got them as a resource (they were not the provider of these parts.) That said, the MA05254 datasheet does indeed cite TA0245A, so I can consider their version of the datasheet likely to supersede (that also corroborates RFDx observation!).

G0HZU - In Section D (MEASUREMENT CIRCUIT) of the Gollege MA05254/TA0245A datasheet, L is not specified. Have I overlooked something? The 160 ohm series in the path it seems would provide the SAW device ports a match from the network analyzer, but would also create non-zero additional insertion loss (which in my case might put me back where I started from, excess insertion loss vs. the datasheet specification.)

G0HZU - I tried 210 ohms || 11 pF in the RF Impedance Matching Calculator page from Analog Devices (set as parallel complex load), one option is shunt C followed by series L (values given); the other option is shunt L followed by series L. It appears these seek to transform 50 ohms to 210 || 11 without attendant loss of 160 ohm series R in the matching sections.

G0HZU - as I can freely change components in the matching sections (my small prototyping board has shunt-series-shunt path into each port), it isn't necessary to retain the shunt L cited in Tai Saw's original (current???, just downloaded!) datasheet.

+++++++++++

One follow-up question - as my last go with SPICE was some decades ago, would anyone reading this have a  suggestion for an on-line passive circuit analysis tool? I'm envisioning something I can set up a few nodes, specify components (L, C or R) amongst the nodes, as sweep (in simulation) the resulting circuit.

+++++++++++

Thanks so much for the various and highly helpful replies!

Dave

[Joel_Dunsmore]

Without studying the problem, I'll remind you that your real matching network, using real components, will have a lot of parasitic response and so you inductors might have 1 or 2 ohms series resistance and maybe 0.05-0.25 pF shunt capacitance (depending on inductor size, maybe much more).  This will move the effective values of these matching networks sometimes as much as 50% in frequency.  ADS has models for SMT components, the smaller ones do have less parasitic than the larger ones. Oh, and an SMT cap might have something like 1-2 nH of series inductance as well.

[G0HZU]

A lot depends on how accurate the datasheet is in terms of the true input impedance of the SAW filter when it is matched.

I think that the matching network I put up in reply #3 should be fairly forgiving of device parasitics at just 139MHz. This assumes SMD parts and a sensible PCB layout. The neat feature of the tapped network is that you can select a fixed inductance (in this case 56nH with a typical Q of 55 at 139MHz) and then adjust the two caps to get a good match. So it should be possible to get a reasonably good match with a simple network even if one or both cap values have to be tweaked slightly.

Assuming the aim is to get a reasonably good match rather than a perfect match, then it should be possible to use just two fixed value capacitors and a fixed inductance. You could also use two fixed caps in parallel for each cap value and this would allow some extra optimisation.

If you want to match it as well as possible then at least one component would have to be tuneable, especially if you wanted to produce lots of copies of the circuit.

[G0HZU]

One follow-up question - as my last go with SPICE was some decades ago, would anyone reading this have a  suggestion for an on-line passive circuit analysis tool? I'm envisioning something I can set up a few nodes, specify components (L, C or R) amongst the nodes, as sweep (in simulation) the resulting circuit.

Assuming you want something that is modern and free, then you could look at QUCS studio.

For many years, the 'free' option was to use RFSIM99.
This is a fairly good linear simulator that is quite intuitive to use. However, this really old program will probably not install and run under Win10 or Win11 without a fair bit of fiddling.

[wb0gaz]

Thank you, Joel -

Mention of ADS reminds me of my attempt to learn ADS given a 30-day eval license from Keysight (some years ago). The no-mentor learning curve was steep enough that I wasn't able get started before the license expired.

G0HZU - Thank you for pointer to QUCS. My current task (seeing if I can simulate what I'm seeing in real life) provides a simple (enough) use case from which to learn QUCS - hopefully there is a user community forum that can help ease the start-up cycle!

(Both) - Inclusion of parasitics in the SMD packages I'm currently using (0603) seems that it would be within the realm of QUCS, particularly if I can define a 2-pin component as an inductor-with-parasitics, capacitor-with-parasitics, or similar.

Thanks again for the very helpful replies and advice!

[wb0gaz]

Hello G0HZU et al.,

Thank you for all of the kind assistance! As of today (6 July 2024) I have a successful match to the SAW filter.

The key to the solution was use of the Golledge version of the TA0245A SAW filter (their part number MA05254) which provided an equivalent representation of the two ports, then use of Analog Devices RF impedance matching calculator which yielded the following matching circuit.



Center of response insertion loss is now about 4 dB (better  than spec); at the upper roll-off position I'll be using it tracks the datasheet response within a dB or so.

I've since downloaded QUCS and will explore that tool, as this task provides a good/simple task to use as a learning project.

Thank you again for the great help!

Dave

[mtwieg]

Do you mean you procured the Golledge version, or you just used the Golledge datasheet to match the Tai-Saw part?

Either way, it's good you seem to have something working, but these datasheets are very strange... especially the Golledge one. Under "Note 2" suggests that 210ohm || 11pF is what is seen at the ports when the other port is terminated with 50 ohms. Once proper matching networks are implemented, the input impedance would change to something else. This means that a matching network designed to match 210ohm||11pF will not give a match on both ports.

Stranger still is the "MEASUREMENT CIRCUIT" section, in which they use 160ohm resistors and some unspecified inductance as some sort of matching network. I'm have no idea why they would characterize the device like this.

Usually to analytically design the matching networks you would start with all the S parameters (including S21/S12). Unfortunately the datasheets don't provide that info.

[wb0gaz]

Thank you, mtweig - good question!

My experiment is from parts procured recently (2024) from asian marketplace seller (aliexpress) that were represented with Tai Saw part number, not Gollege part number. I cannot determine if the parts I received are authentic, but I believe they are as they are working properly once the Golledge datasheet equivalent circuit was used to generate a matching circuit.

By the way, my application frequency is 144 MHz, so I am using upper roll-off, which may create some variation unit-to-unit, but this is not commercial production, so the ~8 dB insertion loss I am seeing is acceptable and my application tolerates slightly more loss there (but not the >13 dB loss I was getting using matching derived from the  Tai Saw datasheet equivalent circuit, which is why I posted this question in the first place.)

The Tai Saw datasheet is dated 2004 (20 years ago). The Golledge MA05254 datasheet has no date information but does have markings citing both V1 and Rev 2, and includes the same Tai Saw part number.

[G0HZU]

The strange looking test fixture isn't meant to be an application circuit, it's just a crude way to get a 210 ohm test fixture.
I think the reason they use 160 ohm series resistors in the test jig is because 160 ohms + 50 ohms = 210 ohms.

You can see they also suggest using two series 160 ohm resistors to 'normalise' the 50 ohm test fixture to a 210 ohm system at a 0dB level. Then you presumably insert the SAW filter and then adjust the shunt inductances to get the best frequency response. The shunt inductance will cancel out the 11pF shunt capacitance at each port of the SAW filter? This lets you measure the frequency response and the insertion loss of the SAW filter.

It may be the case that both the 160 ohm resistors and the shunt inductances were initially adjustable and were optimised for the best response from the SAW filter and because 160 ohms gave the nicest frequency response in terms of passband (and overall match?) the filter was declared as being a 160 + 50 = 210 ohm filter.

They probably didn't bother to quote the shunt inductance because it is implied by the 11pF given elsewhere in the datasheet. It was probably about 120nH. In this case the filter would then be declared as being 210 ohms || 11pF at 139MHz.

[G0HZU]

If it helps, I've used the same type of test fixture (in my youth) to test unknown crystal filters. Use series adjustable resistors and either shunt caps or inductors to try and find out the optimal matching for the filter.

The modern way to do it would be with a full two port VNA. I've done this many times with crystal filters. The VNA produces a two port data file (s11 s21 s12 s22) for the filter and this can then be analysed and matched in a simulator.

[mtwieg]

The strange looking test fixture isn't meant to be an application circuit, it's just a crude way to get a 210 ohm test fixture.
I think the reason they use 160 ohm series resistors in the test jig is because 160 ohms + 50 ohms = 210 ohms.

Hard to not see it as an application circuit if all the datasheet specs are based on it, and they don't provide any other design, or enough data to design an alternative.

You can see they also suggest using two series 160 ohm resistors to 'normalise' the 50 ohm test fixture to a 210 ohm system at a 0dB level. Then you presumably insert the SAW filter and then adjust the shunt inductances to get the best frequency response. The shunt inductance will cancel out the 11pF shunt capacitance at each port of the SAW filter? This lets you measure the frequency response and the insertion loss of the SAW filter.

It may be the case that both the 160 ohm resistors and the shunt inductances were initially adjustable and were optimised for the best response from the SAW filter and because 160 ohms gave the nicest frequency response in terms of passband (and overall match?) the filter was declared as being a 160 + 50 = 210 ohm filter.

Sounds like you're suggesting this is somewhat similar to how datasheets sometimes specify output matching conditions for PAs. That is, the networks don't provide a conjugate match but are still found to be optimal, in this case with regards to the filter transfer function. And thus the impedances shown on the datasheet are not necessarily the actual impedances looking into the ports, but rather the conjugate impedance of what the ports should be terminated with (meaning you will not have a conjugate match). I really dislike such roundabout means of characterizing devices personally... basically lying to the engineer in a way that may lead them to a working design.

I wonder what the transfer function would look like if you made the matching networks from just LC networks which presented the same impedance to the device. Passband insertion loss would certainly be better, but who knows what would happen elsewhere... could be that they suggest lossy matching networks for a reason.

I'm curious what a direct measurement of the device with a VNA would show...

[G0HZU]

If you look at the datasheet this is described in upper case as a measurement circuit and not an application circuit. It looks to me that this was the method used to determine that the filter is 210R || 11pF.   

That's why the other section shows the two series 160R resistors. This is nothing strange. This is there to initially normalise the test fixture to 0dB when using a 50R evaluation system. It's a classic (old) way to evaluate a narrowband filter that doesn't have a 50R design impedance. I've used this same method many times.

With the 160R resistors inline I'd expect the insertion loss to be about 12.5dB plus the 5dB typical insertion loss of the filter (17.5dB) but if the test fixture is first normalised to 0dB (as indicated in the datasheet) the insertion loss might be about 5dB as indicated on the datasheet. The 12.5dB gets normalised out during the through calibration.

Of course, the datasheet could still have errors, and the SAW filter might not have this impedance, but I don't see anything strange about the measurement method they used. Normally this method is used below about 50MHz to look at narrowband filters, but it should still be fine at 139MHz as long as suitable 160R resistors are used.

[mtwieg]

If you look at the datasheet this is described in upper case as a measurement circuit and not an application circuit. It looks to me that this was the method used to determine that the filter is 210R || 11pF.

Why would one add such external components when trying to measure the network parameters of a passive DUT? Especially resistors which will only increase measurement errors after disembedding the DUT. The only reason I can see is if it's actually the recommended implementation of the component (which is what I would call an application circuit). Just to be clear, do you think they're suggesting the user implement the "measurement" circuit? If not, what help is it?

[ftg]

If you look at the datasheet this is described in upper case as a measurement circuit and not an application circuit. It looks to me that this was the method used to determine that the filter is 210R || 11pF.

Why would one add such external components when trying to measure the network parameters of a passive DUT? Especially resistors which will only increase measurement errors after disembedding the DUT. The only reason I can see is if it's actually the recommended implementation of the component (which is what I would call an application circuit). Just to be clear, do you think they're suggesting the user implement the "measurement" circuit? If not, what help is it?

So that the design engineer can see how the numbers presented were found out and if desired, replicate them.
MOSFET datasheets also seem to often have some simplified schematics for the test setups used.

[tszaboo]

If it helps, I've used the same type of test fixture (in my youth) to test unknown crystal filters. Use series adjustable resistors and either shunt caps or inductors to try and find out the optimal matching for the filter.

The modern way to do it would be with a full two port VNA. I've done this many times with crystal filters. The VNA produces a two port data file (s11 s21 s12 s22) for the filter and this can then be analysed and matched in a simulator.

I was going to ask why this method is not used. VNA has a 50 Ohm output/input, the system is presumably 210 Ohm, so the reflections would be quite a lot. Then all the S parameters can be determined, and the system 50 Ohm matched. I don't know how wideband the matching needs to be, but I would guess that with L and C it would be possible to get lower insertion loss (S12) than with a resistor at the filter's frequency. If the filter is designed to handle that much power of course. Am I thinking about this right?

[mtwieg]

So that the design engineer can see how the numbers presented were found out and if desired, replicate them.
MOSFET datasheets also seem to often have some simplified schematics for the test setups used.

I'm familiar with what you're talking about wrt MOSFET datasheets. In those cases the extra components facilitate the measurements. Adding series resistance on the ports of a passive DUT does not facilitate measuring its network parameters. Only makes sense if those same resistors are actually planned to be part of the implementation, in which case they effectively get lumped into the DUT. But it's not clear if that's the intent in the case of these SAW filters.

I was going to ask why this method is not used. VNA has a 50 Ohm output/input, the system is presumably 210 Ohm, so the reflections would be quite a lot.

And with the 160ohm resistors added the reflections will be even greater... definitely not "matched" in any conventional sense.

I don't know how wideband the matching needs to be, but I would guess that with L and C it would be possible to get lower insertion loss (S12) than with a resistor at the filter's frequency. If the filter is designed to handle that much power of course. Am I thinking about this right?

Right, but it's possible that using an LC network for the "matching" might give undesirable behavior further from the passband. The datasheets don't give enough information to tell.

I'm guessing a lot of these questions would be cleared up by the standard referred to in the datasheets (JIS C 6703).

[ftg]

So that the design engineer can see how the numbers presented were found out and if desired, replicate them.
MOSFET datasheets also seem to often have some simplified schematics for the test setups used.

I'm familiar with what you're talking about wrt MOSFET datasheets. In those cases the extra components facilitate the measurements. Adding series resistance on the ports of a passive DUT does not facilitate measuring its network parameters. Only makes sense if those same resistors are actually planned to be part of the implementation, in which case they effectively get lumped into the DUT. But it's not clear if that's the intent in the case of these SAW filters.

That setup allows one to measure the filter passband width and overal shape.
The test circuit with the resistors is not for measuring the S-parameters, it is for the filter passband.
It's an easy way to determine the 3dB and 20dB bandwidths, before involving any other matching that will have a bandwidth of it's own.
This way the presented passband is independent of the exact models of caps and inductors later used to implement the matching circuitry.

[mtwieg]

That setup allows one to measure the filter passband width and overal shape.
The test circuit with the resistors is not for measuring the S-parameters, it is for the filter passband.
It's an easy way to determine the 3dB and 20dB bandwidths, before involving any other matching that will have a bandwidth of it's own.

You seem to be claiming that the DUT has some inherent passband characteristic which is independent from its port terminations. The only way that's true is if you know the DUT's full S parameters (or Z parameters, etc), which could then be used to solve for its transfer function with other port terminations.

This way the presented passband is independent of the exact models of caps and inductors later used to implement the matching circuitry.

If the port terminations can affect the filter characteristics significantly, then why do you think they chose to characterize it with that circuit, as opposed to connecting it directly to a VNA?

[ftg]

That setup allows one to measure the filter passband width and overal shape.
The test circuit with the resistors is not for measuring the S-parameters, it is for the filter passband.
It's an easy way to determine the 3dB and 20dB bandwidths, before involving any other matching that will have a bandwidth of it's own.

You seem to be claiming that the DUT has some inherent passband characteristic which is independent from its port terminations. The only way that's true is if you know the DUT's full S parameters (or Z parameters, etc), which could then be used to solve for its transfer function with other port terminations.

This way the presented passband is independent of the exact models of caps and inductors later used to implement the matching circuitry.

If the port terminations can affect the filter characteristics significantly, then why do you think they chose to characterize it with that circuit, as opposed to connecting it directly to a VNA?

I'm not claiming that the passband is somehow independent of the port terminations.
What I do claim is that the resistive matching is an easy way to get that termination.
It is a much more easily repeatable measurement than some LC matching circuit.
It is also something that can be done to verify that you are getting the right passband shape with your matching circuit.
So, for sanity checking and verification, not production use.
And it does not require posessing a VNA, a spectrum analyzer with a tracking generator would have been enough.


I do agree that proper two port S-parameters would be the way to go, but those seem infuriatingly rare for SAW filters.
Especially for the SAW filters I would have required them for.

[mtwieg]

I'm not claiming that the passband is somehow independent of the port terminations.

Ok, thanks.

What I do claim is that the resistive matching is an easy way to get that termination.

Sorry, I just don't understand what you mean here. "That termination" refers to...? An optimal termination other than what's shown in the measurement circuit?

It is a much more easily repeatable measurement than some LC matching circuit.

Not sure what you mean by "repeatable" here. Making the resistors 10K would probably make the raw measurement more repeatable, but the disembedded measurement of the DUT would not be repeatable, and therefore not useful.

If a device has some inherent issue with "repeatability" then that's a fundamental issue which one shouldn't willfully mask, IMO.

It is also something that can be done to verify that you are getting the right passband shape with your matching circuit.

Again sorry, I don't understand... what do you refer to as "your matching circuit"? Wouldn't the "measurement circuit" simply supplant it?

[G0HZU]

I suspect that the original datasheet was released in paper form in a databook about 20 years ago. At some point it was transferred to (text searchable) PDF and if you look closely at the attached datasheet in post #1 it has inconsistent fonts and some obvious errors. This often happens when the transfer from scanned paper to (text searchable) PDF isn't very successful and all kinds of strange errors can creep in.

It may be the case that someone used the resistive jig technique to reverse engineer the correct Rp and Cp for this filter and this was then added to the amended datasheet supplied by Golledge.

Also, if you look closely at the plots in both datasheets, the noise floor response is exactly the same for both. So clearly the same plot can't be derived from both measurement techniques as the exact same plots from the original datasheet are used in the Golledge datasheet.

If you reverse engineer the original L matches shown in the original datasheet they look to be about right for about 250R || 13pF at 130MHz.

The resistive test fixture in the Golledge datasheet obviously isn't meant to be an application circuit, but it will be fine to use it if you want to. It won't be matched to 50R and the insertion loss is going to be about 17.5dB though...