On 8/30/2011 6:01 AM, Rob wrote:
Geoffrey S. wrote:
Rob wrote:

No. There are two sides, one of them for the high frequency and the
other for the low frequency. The filter response is such that it passes
on one frequency and notches on the other.

Ok thanks. Note that they are tunable anywhere in the 136-17? range.

Assuming the pass side is set to pass 145.600 (the input) and the notch
side is set to notch out 145.000 (the output) and they are connected in
series, would that work?

It is not a pass side and a notch side. Both sides are pass/notch.

But on one side the pass freq is above the notch freq, and on the
other side the pass freq is below the notch freq.

On the Chinese duplexers you have found, the Q is probably not high
enough to tune the pass and notch to frequencies 600 kHz apart.

one might be able to cascade two of them, but Rob's point about Q is
well taken.

It depends on whether the problem is the "depth of rejection" or the
"steepness of skirt". A single resonant unit has a narrow spike on top
of gentle shoulders. Say one section has Q of 1000 (i.e. the 3dB
bandwidth is 100kHz for 100MHz center frequency), but the rejection at 1
MHz away is only 20dB. I could cascade 3 sections, tuned exactly the
same, and I'd have 9dB loss at 100kHz, and 60dB rejection at 1 MHz away.
The 3dB bandwidth might be around 30kHz, or, you could stagger the
tuning slightly to get a little broader flat top, but keep the 60dB
(i.e. moving the center frequency of one cavity by 10kHz won't change
the attenuation at 101 MHz)

The design challenge is that you need to suppress the transmit signal
(at, say, 50W, +47dBm) low enough so that the front end of the receiver
can handle it without blocking (say, you want it down around -50dBm).
With our not so hot cavities described above, you'd need to stack up 5
sections at 20dB per section to get the 100dB suppression. Well, that's
not so great, because you probably now have a bunch of additional loss
in the receive path, AND you've got a real tuning chore on your hands to
make sure that all of them are tuned appropriately to get the required
bandwidth.

Enter the idea of a notch. Let's say our example is 101 MHz Tx and 100
MHz Rx. Rather than rely on the "far away" response of a resonator to
suppress the transmit, I can put a 101 MHz notch filter on the input to
my receiver. If I can get 30dB/section rejection, then 3 or 4 sections
will knock the Tx power down low enough to not block the receiver's
front end.

But wait, there's more... that Tx isn't a narrow spike. It has phase
noise sidebands that go out fairly far. Unfortunately, the run of the
mill transmitter might only suppress "off channel spurious and noise"
by, say, -50dBc (and a rig designed to operate half duplex is probably
worse). That means that your 50W (+47dBm) transmitter is putting out
-3dBm *at the receive frequency*, so no amount of filtering on the
receive path will help. Again, enter the notch filter.. you put a 100
MHz notch on the output of the transmitter.

Now.. as Rob points out, the Q might be kind of low. If the Q is, say,
100, then that 100MHz filter is 1MHz wide, and by the time you cascade
enough to get the BW down, the loss will be huge.


As you can see, this whole duplexer/diplexer/multicoupler design thing
can get pretty complex, and there's not usually a simple cookbook answer.