Matching a monolithic xtal filter

Crystal manufacturers are still rather "old school" and don't provide as much assistance (at least on data sheets/app notes) as you get these days from, e.g., the IC manufacturers.

No, it's 3dB in an "ideally matched" system. That is, if you terminate the crystal in the termination impedances specified on the data sheet (keep reading...).

The data sheet there tells you the termination impedance that the crystal wants to see. If you compute the impedance at the specified frequency (e.g., for the ECS-96SMF45A30, termination impedance is listed as 1200 ohms in parallel with 1.8pF = 1200 ohms in parallel with 1.8pF @ 45MHz = -j1965 ohms =

874 - j533.8 ohms), the complex conjugate (874 + j533.8) will be the input impedance of the crystal (give or take -- often the "termination impedances" specified have been rounded to "nice numbers"). You now have the standard problem of, "I have a 50 ohm source to match to an 874 + j533.8 ohm load -- how do I do that?" and this can be solved in many different ways. See, e.g., "RF Circuit Design" by Chris Bowick.

---Joel

Point of clarification: do you really mean that the filter, terminated in 874-j534, will look like 874+j534 at its input (in the passband...), or that you want the driving source to look like

874+j534 to get the maximum power transfer?

I'd have assumed from the data sheet that I want to terminate both the input and output with 1200 ohms in parallel with 1.8pF to get the correct frequency response from the filter, but I would not have assumed that the input to the filter, when terminated at the output in

1200||1.8pF, would necessarily look anything like 874+j534. Of course it certainly could not outside the passband, if it's a non-dissipative filter.

Cheers, Tom

Thanks Joel, we were working on coming to that same conclusion ourselves, but it's nice to have it clarified. It should be relatively easy to match with a Smith Chart program and a network analyzer.

I'll say old school!! I don't see why engineers have to go through the interpretation and then the matching process each and every time when the crystal manufactures could just put a 'typical matching circuit' right on the datasheet. They're all the same too. Sheesh!!

Thomas

Usually like this: Series resistor from 50ohm source into filter input, matching the spec'd input impedance minus 50ohms. Resistor of spec'd output impedance value to ground on output, then to base of an ordinary emitter follower which feeds the next 50ohm stage. Plus a wee capacitive load if they spec it.

--
Regards, Joerg

http://www.analogconsultants.com

Wait until you need a detailed spec for a piezo. That can really drive you up the wall.

--
Regards, Joerg

http://www.analogconsultants.com

I spoke to soon. When trying to model the matching network from 50 ohms to (874 + j533.8) I get unrealistic values of components. For instance (from 50 ohm to xtal), a series capacitor of 15 pF and a shunt inductor of 770 nH. That's one huge inductor. So how do I do this then? What are the thousands of other engineers doing to get decent insertion loss?

Thomas

Hi Tom,

Now that I think about it, the later ("for maximum power transfer terminate in

874-j534 ohms") makes sense whereas the former is not necessarily true. I suspect you have more experience with this than I do anyway. :-)

---Joel

What's wrong with 770 nH. Looks good to me. Or did you mean uH? Motorola ap note AN267 is full of good stuff on matching. I think you can still download that from the web. Also, why are you matching the filter to 50 Ohms, instead of the input impedance of the IF amp?

Tam

I'd be surprised if the mixer output impedance, or the gain block input, were actually 50 ohms. So the filter response may not match what the mfr designed it to be, since they no doubt assumed perfect 50 ohm resistive source and load.

What's the gain block? I've measured a bunch of MMICS and they're usually low, as low as 30-ish.

John

Huh??

If that's "huge," just what are you working on?? It's more of a problem that you need to use fairly high Qu parts in a matching network that goes between such different impedances. Instead, use a transformer such as MiniCircuits TC16-1T. That and a 6.8pF series cap on the filter side gets you from

50 ohms to about 800-j520, pretty close to the recommended load for the filter. It's pretty unlikely you'd need to match better than that, for either frequency response or signal loss reasons. (I have a feeling that you are trying to match to the wrong polarity of reactance....or maybe I got it wrong...).

Often these filters are used in circuits that aren't implementing 50 ohm signal paths at that point...

Cheers, Tom

You use either coupled inductor or capacitor transformers. Your step-up is only a factor of 18 or so.

Why not a common-source input amplifier? An rf jfet and a drain resistor, with maybe a source resistor for matching.

I'm not an rf type (for me, 45MHz is rf), so please be gentle :-).

john perry

...duh, I meant, of course, a common-gate amp.

jp

Tom. Interested question..............

That 1:4 stepup transformer and 6.8pF raises the 50 Ohm source to nearly the 874-j534 series-equivalent (at 45MHz only) of 1200//1.8pF.

What is the disadvantage of using a 1:5 stepup to get the 50 Ohm up to 1250, plus a parallel 1.8pF, in order to get a direct (wideband) output Z of 1200//1.8pF?

--
Tony Williams.

Except that method loses you 6 dB right off the crack of the bat in a place where loss adds directly to noise figure.

Most monolithic crystal manufacturers presume that people buying a crystal filter have done a little thoughty work up front and know how to match from a mixer's collector load into the front end of a gain block that has been designed for a conjugate match. It's like buying a Ford and expecting the owner's manual will teach you how to drive. They do expect a bit of a priori experience, right or wrong.

Me? I'd see if an L-network gives reasonable answers for real components, both in and out, with a tunable inductor that can give you a little variable X to take care of the imaginary component. Mouser has some real cheap variable inductors these days.

I'm just presuming that you are using a 10.7 or 21.5 MHz. monolithic?

Jim

--
"If you think you can, or think you can\'t, you\'re right."
        --Henry Ford



"Joerg"  wrote in message 
news:eVjEi.24330$eY.9105@newssvr13.news.prodigy.net...
>>
>
> Usually like this: Series resistor from 50ohm source into filter input, 
> matching the spec\'d input impedance minus 50ohms. Resistor of spec\'d 
> output impedance value to ground on output, then to base of an ordinary 
> emitter follower which feeds the next 50ohm stage. Plus a wee capacitive 
> load if they spec it.
>
> -- 
> Regards, Joerg
>
> http://www.analogconsultants.com

That would be 3dB in power and who says it has any effect on total NF at this stage of IF processing anyway. The mixer is always a bad actor when it comes to NF degradation and harmonic distortion. If it's a narrowband communication application, it's always better to lose a little signal strength and buy spec'd harmonic distortion performance by isolating all the mixer ports with attenuation. >

If he's way down there at that frequency it's almost not RF anymore. He should use a stock step-up transformer from MCL. Joerg's suggestion gets him some ridiculous refection coefficient like 10LOG(1-rho^2) loss at the filter, and that really would degrade NF.

I'm not using one, it's Thomas. Anyhow, if losses are a concern you can also rig up a common gate stage up front between mixer and filter. I think someone had already suggested that. These have tons of dynamic range, you can taylor the output impedance, their input impedance typically suits a mixer nicely, they give you the needed gain and cost next to nothing. In commercial designs I often employed switcher FETs like the BSS123 for that because they are just a few cents. When I had enough dynamic range I even skimped on the output inductor because it would have added 3-4 cents.

Oh, did I mention that a professor at my alma mater taught us that common-gate stages are stupid? Well, that is until I brought him a stack of schematics. Not just mine but also some from reputable companies such as Collins and Atlas.

--
Regards, Joerg

http://www.analogconsultants.com

Hi Tony,

That would be just fine, too, 'cept that in my quick look, it looked like the 4:1 turns ratio from MiniCircuits was much cheaper than any suitable 5:1 in their catalog. M/A-com has some similar small transformers and may have something in a 5:1 turns ratio that would be appropriate.

I'm not sure how sensitive these particular filters are to load impedance. With many filters, if you're not trying for the absolute best conformance to the specified filter shape, a modest mismatch from the recommended load and source impedances doesn't really matter that much. It's one of those "YMMV" things--test to be sure you get what you want.

Cheers, Tom

Well, actually, I have a fair amount of experience with LC filters of various types, but not so much with crystal filters and especially not that much with monolithic crystal filters. However, what I know of filters suggests to me that the recommended load is probably more to insure proper frequency response than it is for the absolute maximum power transfer. The monolithic filters I do know about have nothing besides plated quartz blanks inside, AFAIK, so capacitance in the load at each end would be important for response shaping.

Cheers, Tom