New for June 2013: We now have a great diplexer example using singly-terminated Butterworth filters, derived from Matthaei Young and Jones.
What's a diplexer? It's a three-port
network that splits incoming signals from a common port into two
paths (sometimes called "channels"), dependent on frequency.
A diplexer is the simplest form of a
multiplexer, which can split signals from one common port into
many different paths. The incoming signals must be offset in frequency
by an appreciable percentage so that filters
can do their job sorting them out. Diplexers can use low-pass, high-pass or band-pass filters to achieve the desired result.
could be used to route signals to two different receivers, based
on frequency. Or it could be used to create a "matched"
filter that is non-reflective outside of the intended passband.
It could also be used as a bias tee, to feed your favorite active
device with DC power.
Don't confuse the word diplexer
which is the three-port network that permits a transmitter and receiver
to use the same antenna, at essentially the same frequency. Duplexers
are often used in radar, because the transmitter
and receiver commonly share an antenna, and the returned signal
doesn't vary much from the transmitted signal in frequency (just
shifted slightly by the Doppler effect).
Diplexers are employed more
often in communications, seldom in radar. But never say never,
as Chris points out! Wideband multifunction radars can use diplexers
to split received signals to different receive chains based on frequency.
Examples of this appear in Modern Radar Systems by Hamish
Meikle which is available on Google
A diplexer is yet another example
of a microwave concept with an audio analogy. In audio, a "crossover
network" is used to route bass signals to your sub-woofer and
woofer, and treble signals to your tweeter.
Below is a very simple example
of a diplexer that requires that a signal at 2.0 GHz be split from
a signal at 2.2 GHz, modeled on Agilent's ADS. It was a college
homework problem, not an example of how someone might actually attempt
to do this. One requirement was that only a pair of open stubs be
used on each arm.
Try to think of the diplexer
in terms of what each arm must block. In order to block 2.0 GHz,
quarterwave open stubs (E=90) are used in the right arm. In order
to block 2.2 GHz, quarterwave stubs at 2.2 GHz are used in the left
Refer to our section on quarterwave
tricks. Here you will learn that two mismatches of equal magnitude
can be made to cancel each other if placed approximately one-quarter-wave
apart by rule of thumb. This
diplexer application stretches our rule of thumb, you will see that
the "quarter-wave spacing" between the left arm is 16
degrees, and on the right arm it is 152 degrees... hey, we said
approximately, right? You always have to "tune" the length
of line between the mismatches, and when the mismatches are extreme
like they are here, you get results that seem far from 90 degrees.
In the response below you will
see that that port 3 blocks the 2.2 GHz signal and passes the 2
GHz signal, and port 2 does the opposite. The problem with this
circuit is that because the two frequencies are so close together,
the bandwidth is very poor. In real life you would do this with
a pair of higher-order bandpass filters, maybe lumped element, but
more likely coupled line, if you have the room.
coupler bias tee
You can use a pair
of Lange couplers, cascaded back-to-back, to make a diplexer for
use as a bias tee. Referring to the figure below, port 1 is the
input port, port 2 is the DC port and port three is the RF port.
From 6 to 18 GHz you have less than 1 dB RF loss from port 1 to
3. Note that the impedance match at port 1 is excellent, from DC
to 18 GHz.
An interesting point
was raised on our message board recently. If you follow the ADS
symbol for a Lange (shown below), you don't see that both sets
of diagonal ports are DC-connected. What's missing in the picture
are the wirebonds. Look on our Lange
coupler page and you'll get a better look at a real Lange at
the bottom of the page (the wires are hard to see, but they're there!)
Example using singly-terminated Butterworth filters
This example provides a diplexer with some amazing properties, and takes a shallow dive into some of the math. This is Microwaves101, not Microwaves201, OK? We used Matthaei Young and Jones as the reference for this work, in three different chapters. Look for this book and more on our book page. It is important for younger engineers to appreciate that many of the problems that are thrown their way have already been solved for all time, you just need to know where to look for them. You can download a free and legal electronic copy of this book at our download area, put that on your laptop for in-flight discussions with attractive seat-mates of the opposite sex!
Two important properties of this particular diplexer are:
It provides perfect -3.01 dB split at the crossover frequency. Thus it is said to provide a "contiguous passband".
If the two diplexed arms are terminated in Z0, the imaginary parts of the reflection coefficients looking into the filters are equal and opposite, so they cancel out and you are left with exact match to Z0 at the common port!
This diplexer topic is discussed starting on page 991 of MY&J.
Singly-terminated filters assume that there is a matched impedance at the load side but not in the generator. This topic (at least for LPF) is discussed starting on page 104 in MY&J. The Butterworth filter has property of being maximally flat in passband. It also is the only filter that will work perfectly in the diplexer we are describing, and it has to be singly terminated.
Before we get to far, "singly" rhymes with "tingly", a word that was put to good used by the Travelocity Roaming Gnome back around 2011:
The LPF circuit model in Microwave Office (below) is for singly-terminated Butterworth filter of N=3. The G-values are given in MY&J, we did not try to derive them. On page 107 there is a table of G-values up to N=10 if you want to try this on a higher-order filter. G values represent the inductance or capacitance of any element in the LPF, at ω=1 and Z0=1. We did some simple algebra to convert the G values to values that are derived from a specified frequency in GHz.
Note that the singly-terminated Butterworth filter is NOT symmetric. Also, the Nth element has N times the value of the first element, interesting but not useful to know. Component values for 10 GHz are L1=0.3979 nH, C2=0.4244 pF and L3=0.1.194 nH.
Below is the response of the ideal N=3, 10 GHz singly-terminated Butterworth LPF. It does not provide equal ripple reflection coefficient in the passband like the Tchebycheff, if it did, it would be called "Tchebycheff".
Now we need a singly-terminated Butterworth filter that is the direct transformation of the LPF. On page 412 of MY&J you can learn how to transform a LPF to a HPF: basically CHPF=1/(LLPF*omega^2) and LHPF=1/(CLPF*omega^2) does the job and the same G values are used.
Component values for 10 GHz are C1=0.6366 pF, L2=0.5968 nH and C3=0.2122 pF.
Below is the response of the N=3 10 GHz Butterworth HPF.
Note that for either the LPF or the HPF, an equivalent pi network could be substituted; the G-values are correct for either topology but the algebraic solution to cough up the inductor and capacitor values as functions of frequency in GHz would be slightly different.
Now let's put together our diplexer. It is a three-port network made from the two filters. You have to be careful which way the filters are facing as they are not symmetric.... the common port is connected to the sides of the filters with the higher component values in the tee network case.
Now we look at the response: we have achieved perfect -3.01 dB crossover at the design frequency of 10 GHz. In many designs much higher roll off will be required in order to maintain maximum use of the overall frequency band, so you will need to look at solutions higher than N=3 that we used here.
What do the return losses look like? For one thing you don't need to plot S11, in this perfect example we hit the bulls eye of the Smith chart at all frequencies and there is no need to even plot it! Of course in a real design nothing is perfect. At the center frequency both arms (ideally) exhibit -6.02 dB return loss, which is perfect 2:1 VSWR corresponding to perfect -3.01 dB reactive power split.
Now let's take a look at the group delays of the individual filters and then compare to the group delay of the diplexer. Here are the filters by themselves, they are very different.
And here are the group delays of the diplexer: what's this, and other magic property of the singly-terminated Butterworth filter diplexer? The group delays of the two arms are equal to each other! Remember where you learned this useless property of the singly-terminated Butterworth filter diplexer. Too bad the group delays are not flat over frequency, that would be a usable result.
In conclusion, MY&J is a great resource, if you want to get involved in filter design you need to blow the dust off this book and start using it!