Broadband Matching Between Cascaded Filters

Huh? What in heaven's name does the microstrip have to do with the mismatch?

If the microstrip is lossless, then the SWR at the input side will be the SWR of the load hanging of the output side. It won't affect the severity of the mismatch, just its relative phase.

The shortness of the microstrip won't do anything to get you out of a broadband match. Sorry.

If you want an accurate answer, you could find out what happens in the cascaded connection of the three two-ports thay you have: #1 and #3 the bandpass filters and #2 a short length of line, with S parameters

S2= [exp(-j*theta) 0 0 exp(-j*theta)]

Then, see what happens with a resistive pad with

S3= [0 alfa alfa 0 ] with your choice of alfa

....though changing the length of the microstrip between the filters might allow the response to be screwed up differently. (Maybe there is one value of length that is less bad than others.)

A resistive attenuator between the filters sounds like a good idea if the attenuation can be tolerated, otherwise an isolator, though it will have to isolate out of band too since that is where the filters will have the worst matching. Also a better SAW filter might avoid the whole mess.

Chris

Obviously, the impedance mismatch exists regardless of the length of interc onnect, but the length of interconnect does relate to the effects of reflec tions due to the mismatch. Generally, if the length is small compared to th e wavelength of the highest frequency of interest, then transmission line e ffects become less of a concern.

The constructive and destructive effects of the out-of-band reflections cha nge with the interconnect length. The interconnect length can influence the out-of-band response of the cascaded filters and can produce a stop-band r esponse that is less desirable than using only one of the filters.

Yes, I've been running some simulations that indicate a broadband match is not required for an interconnect propagation delay that is 5% of the shorte st out-of-band wavelength of interest. But it's so common to add a broadban d match between filters I'm wondering if there is something I'm missing.

ency of 1.35 GHz and bandwidth of 20 MHz. The filters have input/output imp edances of ~50 ohms in the passpand. The impedances can vary widely outside the passband. Generally, out-of-band impedance variances require using a b roadband impedance network (e.g., resistive pi) between the cascaded RF fil ters to obtain a true cascaded response. If a broadband match is not used, out-of-band reflections can result in degraded stop band performance. The h ighest frequency presented to the filters is 4.5 GHz and there is a 0.1 inc h long microstrip (Zo=50 ohms) that connects the output of the first filt er to input of the second filter. My question: Is there any reason to add a broadband match since the length of my microstrip is much shorter than the wavelength of my highest frequency, which means that I don't really need t o treat the connection as a transmission line? Thanks, Darol Klawetter

------------------------- To the extent you can tolerate the loss of the high frequency resistive att enuator, yes, you should add it. The reason to use a pad is somewhat unrela ted to the line length. "Shorter lines" may be better, in any case, for pu re isolation concerns.

The rejection of passive filters works on the basis of reflectivity in the stop bands. However, their transmission function (H(s), Out(s)/In(s) if yo u like) is based upon broadband terminations at both ports. The return los s of the resistive attenuator is double (dB) the attenuation; they can inde ed improve the rejection performance for this sort of cascade situation. S trictly speaking, you're doing something you are not supposed to do: direct ly cascade two SAWs, or *any* "reflective" filter, and the loss of "theoret ical rejection" is expected. A "pad" would recover some of the supposed re jection performance, since you are violating the test conditions by cascadi ng SAWs directly, or through some near lossless TX line of a given length.

In principle, a 50 ohm line will simply rotate the "reflection impedance" a round the smith chart. Theoretically, I'll guess it would not matter. To the extent the situation is non-ideal, and you happened to end up transform ing to a less reflective impedance, then some rejection could be lost. How ever, this is highly speculative. Put the s-params in AWR and mess with th e line length. The "broadband match" is not so relevant to the line length in an obvious way. But I would be tempted to test it, and don't be surpris ed at hazards. Network analyzers are most accurate near 50 ohms. Precisio n at very high or low immitances will be less, and could hold some hazards if you make idealistic presumptions regarding vendor supplied s-parameters.

Also, you may run into simple isolation problems when it comes to the casca ded rejection (for *any* high rejection filter). Anytime the requested rej ection is up into the 60-70-80 dB range, basic isolation becomes a problem aside from the performance of the filter under somewhat ideal conditions. Since the filters reflect, there will be standing waves on the lines at the out-of-band frequencies. Some will be radiated. An antenna works on the basis of standing waves. For testing a filter on an "open pcb," you'll ne ver see the tiny amount radiated. Enclose these standing waves lines in a low loss enclosure (a shield) and you may see rejection plumment, as coupli ng between them will increase, since it no longer radiated into your office /lab. Keep those lines short, and possibly bury them, or keep the in and o ut "antennas" in separate "chambers." This *may* have implications regardi ng line length between the filters too.

As others have said, an attenuator is a pretty good impedance match device (at least, you control the output impedance well); I'd consider using an amplifier, too, unless there's some reason to think the SAW devices' losses aren't going to imperil your signal/noise headroom. It's pretty easy to bias a common-base transistor for 50 ohms input impedance.