Recipe for creating microstrip filters?

I'm trying to create a microstrip filter from an elliptic filter
schematic made out of inductors and capacitors. The problem is that I
can't really find a description of a method on how to do this. What
I've found so far is using the Kuroda identities but those lead to
unfeasable thin traces. Another way I've seen is using thin traces
where the inductance is dominant or wide traces where the capacitance
is dominant to form the inductors and capacitors.

I have been trying to get a 3rd order filter to simulate properly
using Sonnet lite but so far no luck. I think I'm still missing a
step. Does anyone know a book or a paper which has a clear recipe? Do
it this way and it will be right (after a few tries)?

 

Does it really make any sense to try such transformation ? After all,
there are different loss mechanisms in lumped LCR circuits and at
least with low cost PCB materials when used for microstrip filters.

Why not start from scratch and design a low pass filter followed by a
band stop filter, possibly with an isolation stage (amplifier) between
the sections ?

 

What I want to build is a 1.5GHz low pass filter with a reasonable
sharp cut-off.

 

Nico Coesel Inscribed thus:

Try "Puff" ! I haven't used mine for a few years but I seem to recall
that it was ideal for this sort of thing. The manual had an example
design for a filter.
HTH.

 

Baron Inscribed thus:

 

Preface: I haven't designed a microstrip (or whatever) filter yet, myself.

My impression of such filters is like this:

Suppose you want a, say, 6 pole bandpass filter, very narrow. You need 3
L's and 3 C's. The general design of such a filter is a parallel resonator,
coupled to another parallel resonator, using a series resonator between (for
a Pi design). A very sharp bandpass means the impedance of each resonator
must be very different from the transmission line impedance, while the poles
are kind of on top of each other (give or take pulling interactions). So
the parallel resonators need a very low impedance to successfully shunt the
line, while the series resonator needs a very high impedance to keep
coupling to a minimum, except in the narrow frequency band where it's
desired.

But with microstrip or what have you, it's very difficult to get such a
large impedance ratio, so your filter Q (sharpness) is way down and you need
more stages instead. This is not done with discrete components, because you
can wind an arbitrarily good inductor, and one expensive inductor is better
than matching three, smaller, custom inductors.

As I'm sure you're already familiar with, the basic idea of microstrip (or
whatever) is to alternate between high and low impedance segments, where the
low impedance segments look like low-Z parallel resonators and the high
impedance segments look like high-Z series resonators. Or vice versa.
Using the impedance of a resonator as the corresponding quantity, it should
be very easy to calculate a simple bandpass by trace widths, of course you'd
need to model it to verify dimensions are correct and the poles are in the
right place.

A lowpass filter doesn't need large impedance ratios (high Q resonators), at
least until the higher order poles. Getting a sharp corner could be
challenging in that case, but using more stages always works.

You can save on trace width by giving it some height over the ground
plane -- you can cut out a hole to give the field some room, but I don't
know how to calculate the cutout required. Would also kill EMC.

Tim

 

I figured out myself! I found another paper which had an example of
using wide traces for capacitors and thin traces for inductors:

http://www.sciencepubco.com/index.p...ng+microstrip+low+pass+filter+rectenna+system

I decided to follow the same path as the authors and see where that
would get me.

The next big problem was finding a piece of software which could
calculate the inductance and capacitance of a copper strip to the same
results as in the paper. That turned out to be harder than one would
expect. Even the tool you can download from Rogers doesn't work
properly! The problem is that there are two formulas and which one is
right depends on the ratio between the width of the track and the
height of the substrate. Most tools only work when the trace width is
less than the height of the substrate.

I was just about to give up when I found this web page. The microstrip
calculator gives the same results as in the paper:
http://wcalc.sourceforge.net/cgi-wcalc.html

Now I could start to translate the lumped element diagram I got from a
program called SVCfilter ( http://tonnesoftware.com/svcfilter.html )
into a distributed one. I first calculated the track widths for a 3rd
order elliptic low pass filter and simulated that with Sonnet Lite.
The results where quite good so I decided to go ahead with the 7th
order elliptic filter I actually want to build. I had to tweak the
sizes a bit to get the resonating frequencies the same as given by
SVCfilter. After that I used the sizes from the simulated layout for a
PCB layout and tested it for real and it actually seems to work
reasonably close to the simulation. I used 0.2mm lines for the
inductors. That is about the limit of what I can etch myself so there
is quite some tolerance on the final width of the tracks which
probably contributes to the error.

It looks kinda cool though:
http://img40.imageshack.us/img40/3585/img0942ui.jpg

 

Thanks

I gave up on running Windows software in Wine. Too much doesn't work
and the Wine crowd seems to focus on games and office. Instead I
installed virtualbox and installed Windows in a virtual machine. There
is a Windows license sticker on the Linux box so why not

 

Google for KK7B

One of his design is at
http://www.arrl.org/files/file/Technology/microwave/23cmxv.pdf
A bandpass filter section seems to have a Q of about 5.

 

I recall doing some simulations on those hairpin filters as an
experiment a few years ago. It turned out that they won't work at all
at 300MHz. Since then I acquired some equipment to be able to measure
in the GHz region.

 

I am a bit confused, since I have always assumed that loaded Q,
unloaded Q and insertion loss are all related.

Googling for various kinds of resonators, there appears to be a huge
number of papers how to make microwave resonators with Q _less_ than 5
(apparently for some wide band services). Traditionally, the unloaded
Q has been quite low due to the PCB material losses.

 

Right -- there are [at least] two "Q"s one could define in a filter
circuit. The one that causes power loss* is the Q of the components
alone -- simple resistive (or equivalent) losses. The other is the
impedance of any given component in relation to the circuit impedance at
that point -- typically, the line impedance. An inductor with a Q of 100
and a reactance of 50 ohms at some frequency, connected to a transmission
line of 50 ohms, has an overall Q of about 0.99 (i.e., about 1/100th less
than 1.0).

*It's total power loss, not insertion loss necessarily. When the
insertion loss is high, reflected power is usually also high, so that
power is (mostly) conserved. Consider a coupled resonator type bandpass
filter: if the coupling is very low, bandwidth will be minimal, and
insertion loss will be high. But power needn't be lost; the first
resonator reflects the excess back. The total power reflected and
transmitted is always less than incident, and this loss is due to
component Q.

Tim

 

It was a long time ago but I recall the resonators started resonating
at a multitude of 300MHz. Something like 1.2GHz (which makes sense). I
used Sonnet for simulation. If the simulation results wouldn't be
close to real life they probably would be out of business real quick.

I'm not specifically working in a range At that time I choose
300MHz because that was in the range of what my equipment could
handle.