Designing LC filters for amplifiers?

It's much easier to use a filter synthesis program. I use ELSIE:

Leon

But Leon, that doesn't negate the problem at all: The filter's stopbands will still interact with the amplifier's own limited bandwidth. I.E: Neither stage will see 50 ohms at ALL frequencies, as they were designed to see (in ELSIE), and as they were individually simulated with in a standard linear simulator (with the simulator's own infinitely wide 50 ohm in/out ports)...

-Bill

A passive filter cannot shape response and also provide impedance matching. It provides a frequency response by producing impedance mismatch. This mismatch prevents the passage of power at some frequencies.

A passive filter can be equipped with hybrids and the input and output. So it will look like more or less nominal impedance in the range of frequencies.

Vladimir Vassilevsky DSP and Mixed Signal Consultant

There are passive filters that present a nearly-constant input impedance across a wide, or theoretically infinite, frequency span.

Jeroen Belleman has done some nice work on constant-resistance lowpass filters, and posted about it here some time ago. His CERN papers are probably still available on the web.

John

Do those constant resistance filters include a dummy load where all the rejected signal energy goes?

Wouldn't a unity gain buffer amp solve that issue for the input impedance? just guessing.

Certainly; they have to. The trick is to get the proper filter response and a constant input resistance without a bazillion parts.

Picosecond Pulse Labs and some others make constant-R lowpass filters, so you can stick a filter in a coaxial line and not generate awful reflections.

John

Sure. Usually when an amp drives a filter, the filter's impedance variation isn't an issue. But sometimes you need a passive, non-reflective filter, especially when you get into the GHz range where the buffer amp starts being a problem.

John

This was the problem with the classic image-parameter filter design methods, assuming that each section was uniformly 50 ohms. "Modern" filter design is done by computers, makes no such assumptions, and works in ways I don't understand but appears to involve a lot of fiddling.

A good filter book, Williams maybe, has tables of computer-generated normalized filters and procedures for bandpass transformation and scaling. That seems to work.

John

But if you design a single 50 ohm bandpass LC filter (ideal) in a modern filter synthesis program -- and then duplicate it to make two filters -- and then cascade these two duplicate filters together in series, the graphed frequency characteristic will now be very FUBARD as compared to the single filter. Could someone try this also and see what they get? The filter stopband's high VSWR characteristic really interacts with the next stage, since both filters "expect" to see only

-Bill

Yup. That's why a higher-order filter doesn't look like a string of low-order filters. So don't do that.

I haven't noticed it, because when I want an Nth order filter, I look it up in a book.

John

Can the same thing be done for the filter output, so that it has a constant resistance, also? That would make possible cascadable stages, though I am sure the combinations would not be minimal.

Seems like it should. Just apply the same techniques to both ends.

John

So, simulate it in Spice with some good approximation of the impedances it WILL see, and see how bad things are. Or, perhaps how good. It may surprise you.

For the work I do, op amp amplifiers have pretty well characterized output impedances, and I can design the following stage to have the desired input impedance, at least over a wide enough frequency range that it's not a problem. What I generally find is that even with mismatch from the design impedance, things don't go too far out of whack.

Someone noted that filters have poor return loss outside their passband. That's not necessarily a bad thing, depending on the design goals, but if you don't like it, implement a "diplexing" filter that passes the signal from one port to another in its passband, and outside the passband, it passes the signal to a load resistance. In that way, a filter terminated in 50 ohms (for example) can look like

Cheers, Tom

There are tables for passive LC filters where the source and load resistance are not equal. Thus I don't understand your statement.

Hi, You can try with active filters using Op Amps with Op-Amps separating the stages

They are also called DIPLEX filters, they have complimentary filters where the undersired power is routed to a dummy load instead of being reflected back to the source.

The discussion is about out-of-passband (broadband) impedance matching, not about the filter's characteristic impedance. Standard LC filters *work* when connected to the appropriate resistive sources and load, but themselves don't appear as pure resistors to those sources and loads. With more work, they can be designed to be nearly pure resistances.

John