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

[G0HZU]

Here's a quick youtube demo I put together showing the resistor method in action. You can see that there is also a normaliser network used, just as in the datasheet for the SAW filter. In this case the normaliser is between ports 3 and 4.
The software is able to auto normalise by computing s21 - s43 and plotting this with the blue trace.


With the normaliser, you can predict the true insertion loss of the filter when it is matched. In this case the blue trace shows 2.04dB.

Once the R and C is optimised for best match (960R and 23pF) the video then shows the response when tuned with an adjustable complex source. The passband response and insertion loss are virtually identical.

So the series resistor method is a valid way to estimate the Rp and Cp (or Lp) of a narrowband filter.

[G0HZU]

I believe that the above KVG 9MHz LSB filters were designed to be 1000 ohm devices and a small amount of external capacitance is required to match them properly.

960 + 50 = 1010 ohms which is very close!

The filter I have here is the same as the one tested by someone else in the link below:

https://awsh.org/80m_superhet_receiver/

By using a couple of trimmer resistors and a couple of trimmer caps it should be possible to accurately reverse engineer the optimal Rp and Cp required for this filter using this classic method. There is no need to use a VNA, just something with a 50R generator and some sort of tracking receiver with enable the operator to quickly establish that this is a 1000R filter and there also needs to be about 23pF added in shunt to it.

If this is done correctly, you can see how flat the passband can be. There's hardly any ripple at all. This is a great way to find out how good the passband response can be if you reverse engineer the correct combination of Rp and Cp for the filter. Sometimes Cp is negative and therefore a shunt inductance is needed for some filter types.

[tszaboo]

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.

No doubt, resistive matching is indeed the easiest to achieve up to a few GHz. It's also lossy, which might or might not be a problem.
I'm guessing if the matching is a Pi or similar made from LC, then there really only ever going to be one frequency where the filter is matched*. Which again, might or might not be an issue.
As I understand, SAW filters are always bandpass, and narrow.
I mean if the filter isn't matched and the signal bounces back from it's input as opposed to it disappearing in the filter, if it doesn't blow up your source...

*Maybe more above the self resonant points for example, that's beside my point.

[G0HZU]

It doesn't really matter that the resistive method is so very lossy because the aim of the resistive jig is to tell you the correct insertion loss and passband response for the filter once you normalise the test jig using the two 160R resistors in series between the two 50R ports.

It also indicates the correct Rp and Cp for the filter if you then want to design a suitable low loss matching network for the filter.

What you end up with isn't meant to be a practical application circuit. It is just a means to an end. Or, in other words, a stepping stone towards designing a low loss matching network if you don't know what the correct Rp and Cp of the filter is. Using the resistors and trimmers you can work out Rp and Cp and then progress to designing a suitable matching network.

This technique is many decades old, and predates vector network analysers. Back in those days, some people would have used this method using a 50R signal generator and a 50R receiver and a pair of series trimmer resistors and a pair of shunt trimmer inductors (or trimmer capacitors). It would have taken a long time, but they would have little choice back then. If done correctly, they would be able to estimate the ideal Rp and Cp for the filter quite well.

[tszaboo]

It doesn't really matter that the resistive method is so very lossy because the aim of the resistive jig is to tell you the correct insertion loss and passband response for the filter once you normalise the test jig using the two 160R resistors in series between the two 50R ports.

It also indicates the correct Rp and Cp for the filter if you then want to design a suitable low loss matching network for the filter.

What you end up with isn't meant to be a practical application circuit. It is just a means to an end. Or, in other words, a stepping stone towards designing a low loss matching network if you don't know what the correct Rp and Cp of the filter is. Using the resistors and trimmers you can work out Rp and Cp and then progress to designing a suitable matching network.

This technique is many decades old, and predates vector network analysers. Back in those days, some people would have used this method using a 50R signal generator and a 50R receiver and a pair of series trimmer resistors and a pair of shunt trimmer inductors (or trimmer capacitors). It would have taken a long time, but they would have little choice back then. If done correctly, they would be able to estimate the ideal Rp and Cp for the filter quite well.

Right, so you match it to 50 Ohm to measure it. You don't match it this way in the application circuit.
I'm familiar with the SWR meter based matching method, even though I never did it myself.

[G0HZU]

Here's a link to someone who has used the resistor method the old school way using a sig gen and a Hi Z (low capacitance) probe.

https://www.qsl.net/g3oou/filtertestjig.html

The aim here is to estimate the ideal termination impedance for the filter.

A shunt resistor and shunt cap is used at the output because a High Z probe is used instead of a series trimmer followed by a 50R receiver.

The important conclusion at the end is quoted below:

The termination impedance may now be estimated from the trimmer settings and measuring the resistor values with a test meter. If you have a lot of local RF interference then the jig may need to be screened.

[ftg]

Now that I really think about this and properly look at both Tai-Saw TA0245A and Golledge MA05254 datasheets, it feels like they are completely separate parts and that the Golledge part is offered as a functional replacement for the Tai-Saw TA0245A, down to the 6.5dB insertion loss.
That would explain why the lossy resistive matching is presented instead of some lumped element circuit.

But G0HZU's point about it potentially just being a copy paste job or a bad scan also sounds plausible.
When one compares the charts and the package drawing the jpeg artefacts look the same on both datasheets. 
At least in these two:

Golledge MA05254
https://www.golledge.com/media/1803/ma05254.pdf

Tai-Saw TA0245A:
https://www.cdiweb.com/datasheets/tai-saw/ta0245a%20_rev.2.0_.pdf

[mtwieg]

Here's a quick youtube demo I put together showing the resistor method in action. You can see that there is also a normaliser network used, just as in the datasheet for the SAW filter. In this case the normaliser is between ports 3 and 4.
The software is able to auto normalise by computing s21 - s43 and plotting this with the blue trace.


With the normaliser, you can predict the true insertion loss of the filter when it is matched. In this case the blue trace shows 2.04dB.

Once the R and C is optimised for best match (960R and 23pF) the video then shows the response when tuned with an adjustable complex source. The passband response and insertion loss are virtually identical.

So the series resistor method is a valid way to estimate the Rp and Cp (or Lp) of a narrowband filter.

First of all I really appreciate the effort you put into this.

Just want to check my understanding of your video.
1. You're adjusting Rmatch and Cmatch with the objective of getting a good S21 "shape", while ignoring S11, S22, and overall insertion loss.
2. This does not tell you anything about the "input impedance" of the DUT. But it tells you the "optimal" terminating impedances of the DUT (call it Zopt, in your case around 361 - j487 ohms).
3. You would then devise impedance transformers to transform whatever impedance(s) are facing the DUT (50ohms or whatever) into Zopt. If those those impedance transformers are lossless you will get an overall insertion loss (shown in your video as S56) much better than what was seen with Rmatch/Cmatch (S21 in your video). But the return loss could also be very poor on both sides.
4. On the datasheet, you would then say that the "input impedance" of the DUT is the conjugate of Zopt, assuming that this will lead designers to make matching networks presenting Zopt to the DUT.

Is that correct?

[G0HZU]

You're adjusting Rmatch and Cmatch with the objective of getting a good S21 "shape", while ignoring S11, S22,

Yes, but I am also looking at insertion loss via the normalised trace in the simulation. However, in a typical setup, this wouldn't be possible so the aim is to get the best S21 shape across the passband with minimum passband ripple.

If this is done correctly, it should also coincide with the best overall s11 shape across the passband even though it can't (yet) be measured.

This does not tell you anything about the "input impedance" of the DUT. But it tells you the "optimal" terminating impedances of the DUT (call it Zopt, in your case around 361 - j487 ohms).

Yes, it lets me know Zopt. If you compare 361 - j487  and 1010R || 23pF at 9MHz they are essentially equivalent.

You would then devise impedance transformers to transform whatever impedance(s) are facing the DUT (50ohms or whatever) into Zopt. If those those impedance transformers are lossless you will get an overall insertion loss (shown in your video as S56) much better than what was seen with Rmatch/Cmatch (S21 in your video). But the return loss could also be very poor on both sides.

I'm not sure what you mean here. I quickly added an L match to the input and output as in the image below and you can see that the passband is still essentially identical with 2dB loss. The traces overlay almost perfectly. I've also added s55 which is the input of the L match. This is as good as it gets with this crystal filter. It isn't possible to get better than 11dB return loss across the whole passband and this is because this filter was designed for this return loss. In other words, no matter what I do with the L match values, I can't get a nicer return loss plot that what you see in the plot below.

On the datasheet, you would then say that the "input impedance" of the DUT is the conjugate of Zopt, assuming that this will lead designers to make matching networks presenting Zopt to the DUT.

Yes, that's it. Obviously, this technique isn't as elegant or as powerful as using a VNA, but it does usually give accurate results for Zopt

[mtwieg]

You would then devise impedance transformers to transform whatever impedance(s) are facing the DUT (50ohms or whatever) into Zopt. If those those impedance transformers are lossless you will get an overall insertion loss (shown in your video as S56) much better than what was seen with Rmatch/Cmatch (S21 in your video). But the return loss could also be very poor on both sides.

I'm not sure what you mean here. I quickly added an L match to the input and output as in the image below and you can see that the passband is still essentially identical with 2dB loss.

Right, this is what I meant by "impedance transformers to transform whatever impedance(s) are facing the DUT (50ohms or whatever) into Zopt". Produces the same response as S56 in your video (around the passband, anyways).

I've also added s55 which is the input of the L match. This is as good as it gets with this crystal filter. It isn't possible to get better than 11dB return loss across the whole passband and this is because this filter was designed for this return loss. In other words, no matter what I do with the L match values, I can't get a nicer return loss plot that what you see in the plot below.

I don't see the S55 trace in your screenshot, but I think I get what you're saying. 11dB isn't that bad, actually. One could probably get it much lower at specific frequencies, but not across the entire passband. And trying to do so would make the S21 ripple terrible.

Yes, that's it. Obviously, this technique isn't as elegant or as powerful as using a VNA, but it does usually give accurate results for Zopt

To be clear, I'm not saying the method isn't useful (especially after having been walked through it). My issue the way the results are communicated in the datasheet. They could just say "present the ports with this Zopt". But instead they say "the input impedance is Zopt*" which is simply untrue and will lead to confusion for engineers who notice that.

[G0HZU]

I don't see the S55 trace in your screenshot, but I think I get what you're saying. 11dB isn't that bad, actually. One could probably get it much lower at specific frequencies, but not across the entire passband. And trying to do so would make the S21 ripple terrible.

Whoops, sorry, the S55 ident is missing off the legends at the bottom. This sometimes happens if I set the fonts too large and it pops off the graph area. The S55 trace is the pale blue trace (trace d) in the plot and it has the 1d marker pip on it at -11.06dB.

If I try and retune anything, this S55 trace loses symmetry and so it is tuned to the best result I can get.

Note that the scale for S55 is the light blue scale on the right side of the graph at 5dB/div.