Andy,

I realise when I re-read your post that I have not included the loss of
the tuned circuit each side of the filter due to Rp of the inductors.

Taking this into account, the max. loss should be around 1.1dB per tuned
circuit + 3dB max. for filter = 5.2dB.

I re-measured the response using the digital power meter as this is more
accurate than the spectrum analyser relative measurements.

The insertion loss of filter AND tuned circuits was measured at 3.7dB

I then swept the filter and measured the -3dB bandwidth and found this
to be -18kHz, + 15.5kHz (very close to the expected +/- 15 kHz).
Next I measured stop band attenuation at +/- 60kHz and measured -47.3 dB
on low side and -50.7 dB on high side (manufacture data was for at least
40dB).

I then looked at ripple in the pass band. There was 1 ripple that
produces a dip at +9kHz of 1.6dB and a slight peak at +13 kHz of 0.86dB.

This is very close to the 1dB figure mentioned in the datasheet. There
was no ripple below Fo.

If I can now get the impedance transformation in my circuit working the
same as this then I'm on my way.

I'll leave the output match as-is and if there are issues then it
eliminates the output load and just leaves work to do at the mixer end.

This is the first discrete mixer and crystal filter I have constructed.
I'm not sure if I am overlooking anything here but to me those figures
now look good.

Does this all sound reasonable ?

Thanks heaps

Regards

David



AndyS wrote:
David wrote:
I noted in We Hayward's book "Experimental Methods in RF Design" that he
suggests the gain of a Dual Gate Mosfet in a mixer circuit is about 1/4
of the gain of the same device in an RF amp circuit. If this is true
then the 6dB gain I now see in the mixer stage would be about right.

I cranked up the LO level and now have +/- 4V on G2. I tried biasing the
Gate up to 3 but it made no difference than when just 47K to ground
bias was used on G2.

The main issue I am having now is matching to the 4-pole filter.
I have tried several approaches and the performance is disgusting.

The current mixer to filter circuit is....

This "should" have matched down to 800R

100nH (Q=100) inductor to Vcc from Drain
6-30p trimmer to ground from drain
Split capacitor tap from drain to ground (220pF in series with 200pF)

1st filter, 4p7 to ground between 1st and second filter.

Output match to 50R from 800R for testing...
100nH to ground from filter output
82pF to ground
6-30p trimmer to ground
15pF in series to 50 Ohm load.

The loss through the filter is around 10dB instead of 3dB, the ripple is
around 6-8db instead of 1dB. The filter response shows double peaks with
dip between, either side of the peaks falls off extremely quickly at
around 2 kHz off (should be +/- 15kHz bandwidth).

I would appreciate any help I can get to determine what is happening and
to correctly match into this filter that requires 800R//3pF terminations
at 45 MHz.


Andy writes:

Ok. Well, using +/- 4 volts ( i am guessing rms) will certainly
drive
the mosfet from full on to cutoff so DC biasing wouldn't be required.
Less LO could be used with a DC bias, but if you have the LO power
available, there is no reason to change.

I am assuming that you have gotten rid of the 800 drain load.....
The split cap approach is fine and should give you about a 4/1
step down, so if the drain output imp is 3200 ohms or thereabout, the
match should be close... ( I am doing this in my head, so forgive me if

I am off by a thousandfold :))) )

Now, the filter loss is something else. You did not tell me how you

measured it, and I am assuming that you just measured the voltages
and used that. This is a common mistake as there can be substantial
impedance change. The only accurate way to characterize filter loss
is with a special test jig which can measure the power into the load
without the filter and then the power into the load WITH the filter,
without
changing any of the tuning.... A purely resistive jig with a highZ
probe
can be fairly accurate, measuring voltage loss..... But you have to
allow
for any impedance changes in source and load.......

The inter match between the two filters should have both and L and
a C
making a parallel tank to make sure both the filter reactance, and the
stray reactances are tuned out. Usually filters are slightly
capacitive
, a couple pf, on their parallel terminal impedance......a tuned
circuit
will tune all this out..... giving you 800 to 800... You tune the tank
for minimum ripple in the passband... not for max gain at one of
the ripples.... tho it will be close...

The output transformation from 50 ohms up to 800 ohms represents
a 4 to 1 voltage transformation, or 12 db voltage loss..... In other
words,
if you used an 800 ohm resistive load (with a parallel tank to get rid
of the reactance and strays) you would, with a voltage probe, measure
12 db higher than you would when you transformed the load down to
50 ohms.... If you already knew this, and accounted for it, I
apologize for
assuming you didn't,, , but it is a common mistake some people have
made....

I don't disagree with Wes's book or explanation at all, but I am
surprised
that 12 db was the best the mosfet would do.....but, as I said, I have
never
used that particular one...

Remember, you have to match the RF generator UP to the input
Z of G1 in order to calculate the conversion gain. If you have just
connected G1 to a 50 ohm RF source, you are losing a lot of voltage
since the input Z of G! is probably a couple K..... That is quite a
voltage gain.... and that is before the conversion process even
starts....

It looks to me like you are on the right track. Again , I apologize
if it seemed
I was "talking down" to you, but I am just doing an all-purpose memory
dump
of all I remember about when I did this..... And I know for sure that

I got a hell of a lot more conversion gain out of a 3n141..... after
matching
both the input to G1 AND the output to a matched load....

Goodluck,,

Andy W4OAH