EMRFD Message Archive 2029
Message Date From Subject 2029 2008-09-08 03:53:09 timshoppa Crystal filter success! Now matching to a homebrew RX.
I got 50 4.443 MHz microprocessor crystals from Mouser, circa $20
total.
Built the tunable RF source from EMRFD, restricted its tuning range
to 4.443 Mc +/- a few kc. Used it and a test jig to parameterize all
50 crystals including series resistance (as low as 17 ohms for the
best, 22 ohms for typical, and some lousy ones at 28 or so). Took a
little while, but not any slower than the G3UUR scheme. All 50
crystals out of the package match to +/- 200kHz, but there were a
couple obvious piles in the middle of the frequency range which had
highish Q (circa 110,000 for average and 140,000 for the best) and
each pile had 8 crystals matched to +/- 8 Hz. Outside those two piles
there were a lot of randomish numbers that often had poor Q (as low
as 70,000).
Finding the piles was pretty easy if you take a piece of graph paper
and plot each crystal for F and delta-F. The highest delta-F's were
the worst Q's and highest equivalent series resistance.
Found my Xerox of Zevrev, looked up the parameters of a Gaussian-to-
12dB filter, fed them into XLAD along with my crystal parameters for
a 500Hz wide filter. It gave me a bunch of capacitors to use ranging
from 20-some to 300-some pF. If I made the filter much narrower the
capacitor values went up a lot, into low 1000 ohm range in some
cases. Hmm, maybe for my next filter I'll use that choice, but other
than a contest I don't really like narrow filters.
Anyway, I had a bunch of NPO ceramics and I could match the capacitor
values from XLAD to within a percent or two by paralleing at most two
units.
Built a little compartmentalized ladder strip like Carver's article
out of PCB, built the filter with one pile of 8 good crystals.
Took the RF generator, whipped up some matching transformers, and put
the filter through it's paces. WOW. The measured bandwidth was
exactly 500Hz. Filter response shape was indeed a Gaussian to 12 dB
and then falling off just like GPLA predicted. I had no idea that
with circa a few $10 in parts, a little test RF source whipped up in
the basement on a Saturday morning while the kids were watching
cartoons, some thinking and a pecking at a keyboard, and a handful of
capacitors, could I produce in my basement a filter that so exactly
matched the theoretical curve for response. WOW!
Rigged up a quick-and-dirty ditter (literally my keyer with a 1 pound
roll of solder on the dit key switchin power on and off to the RF
generator) and it looks like a group delay of a few ms at the peak.
If I go way out into the skirts I can see that around the 12dB down
point, group delay goes up and ringing sets in, each leading and
trailing end of the ditter makes a smallish ring. Not complaining,
just saying that I can notice it out on the skirts with test
equipment.
One thing I did note (and again it's reality matching GPLA
predictions), having the impedance match at the filter wrong really
really messes up filter response. From the bench experiments it's
perfectly clear to me that if I don't match this filter to the rest
of the rig impedance-wise, I'm not going to get the beautiful theory
filter response.
So now it's time to put the filter into a receiver. I have a hybrid
cascode IF strip that I built a little bit before, and have a
question: what's my choices for impedance matching this 500-ohm
filter to the post-mixer amp and the IF amp? Poking at my calculator
leads me to believe a transformer with a 19:6 turn ratio will do the
match. W7ZOI seems to often use L-matches to put 500-ohm filters into
his rigs (all the way back to the Progressive and as recently as the
hybrid cascode article I see it in the schematics) featuring a
variable cap for fine tuning, but is the variable cap adjusted "by
ear" or is there an official test procedure e.g. a 500 ohm resistor
and some clever adjustment procedure?
Tim N3QE2042 2008-09-09 16:56:17 kerrypwr Re: Crystal filter success! Now matching to a homebrew RX.
Gaussian-6 filter but my attempts to build a wide one have been
disastrous!!
The question of impedance is one that has caused me to ponder.
Suppose we have an amplifier; the type that abounds in EMRFD and
elsewhere. This amplifier "likes to see" 200 ohms in the collector
circuit so we match it to 50 ohms with a transformer.
In a QRP transmitter we follow the amplifier with a 50-ohm antenna
(let's forget the LPF for now); the 50-ohms is reflected through the
transformer to "look like" 200 ohms to the collector circuit.
Power is transferred.
But perhaps we are building a receiver (for example) and the amplifier
is a "strong" one after the mixer. We then follow the amplifier with
a filter that "likes to see" 50 ohms at its input.
But there is no "50 ohms" anywhere in circuit.
The wise designer interposes, at least, a 50-ohm pad; each end of the
pad looks, at least, something like 50 ohms to the relevant device.
The "something" depends on the attenuation; gain is traded for match.
In some instances a simple resistor is used; that "forces" 50 ohms (or
another desired impedance) on the system.
Examples of both approaches can be found in EMRFD.
The "pad" approach may be seen in Figs 7.28 & 7.60 and the "resistor"
approach in Fig 6.57 where I think FL1 must require a 300-ohm termination.
(These are just quick & random examples; there should be many more in
EMRFD).
So I think it's important to distinguish a real impedance from a "like
to see" impedance.2043 2008-09-10 02:09:36 timshoppa Re: Crystal filter success! Now matching to a homebrew RX. 2044 2008-09-10 03:25:42 kerrypwr Re: Crystal filter success! Now matching to a homebrew RX.
but, I hope, more sophisticated. One application will be noise figure
measurements.
I built a 200 Hz "narrow" Gaussian-to-6 dB filter that came-out very
well; nice rounded top and good steep skirts. Actual bandwidth was
195 Hz; noise bandwidth was 215 Hz iirc.
The actual width of the "wide" filter isn't important in my
application; knowing the width (and the noise bandwidth) is the
important thing.
I aimed at about 3 kHz bandwidth but anything up to, say, 10 kHz would
do. This receiver is not intended for "listening"; it will have no
audio output, just metering.
Results were very poor; I have put it aside out of frustration.
I measured crystal parameters very carefully, using both the
series-resonance method and the G3UUR method; there was good
correlation. I even measured using the phase-shift method (I have a
vector voltmeter) as a "sanity check".
I measured all the capacitors carefully & accurately.
When I tested the result it was terrible; several iterations did not
improve it.
Wide filters are more difficult than narrow filters.
Bill Carver put it neatly in an e-mail to me; "...... you cannot make
a filter narrower than the difference between series resonance of the
crystal and parallel resonance ......".
So, if the bandwidth of your filter is a small fraction of that
difference, it's not too difficult. As the fraction grows, so do the
difficulties.
Termination is one; the natural termination of wide filters is high
(the various programs refuse to accept low termination impedances) and
I wound-up with figures like 10 k-ohms.
I transformed to 50 with L-networks.
I am considering using an L-C design for my "wide" filter.
For the present I have "retired hurt" and gone-on to design &
construction of the IF section; that is looking good so far.2045 2008-09-10 10:59:27 John Kolb Re: Crystal filter success! Now matching to a homebrew RX.
John
At 03:25 AM 9/10/2008, "kerrypwr" wrote:
>My application is a Measurement Receiver; similar to the EMRFD one--
>but, I hope, more sophisticated. One application will be noise figure
>measurements.
>
>I built a 200 Hz "narrow" Gaussian-to-6 dB filter that came-out very
>well; nice rounded top and good steep skirts. Actual bandwidth was
>195 Hz; noise bandwidth was 215 Hz iirc.
>
>The actual width of the "wide" filter isn't important in my
>application; knowing the width (and the noise bandwidth) is the
>important thing.
>
>I aimed at about 3 kHz bandwidth but anything up to, say, 10 kHz would
>do. This receiver is not intended for "listening"; it will have no
>audio output, just metering.
>
>Results were very poor; I have put it aside out of frustration.
>
>I measured crystal parameters very carefully, using both the
>series-resonance method and the G3UUR method; there was good
>correlation. I even measured using the phase-shift method (I have a
>vector voltmeter) as a "sanity check".
>
>I measured all the capacitors carefully & accurately.
>
>When I tested the result it was terrible; several iterations did not
>improve it.
>
>Wide filters are more difficult than narrow filters.
>
>Bill Carver put it neatly in an e-mail to me; "...... you cannot make
>a filter narrower than the difference between series resonance of the
>crystal and parallel resonance ......".
>
>So, if the bandwidth of your filter is a small fraction of that
>difference, it's not too difficult. As the fraction grows, so do the
>difficulties.
>
>Termination is one; the natural termination of wide filters is high
>(the various programs refuse to accept low termination impedances) and
>I wound-up with figures like 10 k-ohms.
>
>I transformed to 50 with L-networks.
>
>I am considering using an L-C design for my "wide" filter.
>
>For the present I have "retired hurt" and gone-on to design &
>construction of the IF section; that is looking good so far.
>
>
>
>------------------------------------
>
>Yahoo! Groups Links
>
>
>
>
>
>--
>No virus found in this incoming message.
>Checked by AVG.
>Version: 7.5.524 / Virus Database: 270.6.19/1659 - Release Date:
>9/8/2008 7:01 AM
>
>
>
>
>--
>No virus found in this incoming message.
>Checked by AVG.
>Version: 7.5.524 / Virus Database: 270.6.19/1659 - Release Date:
>9/8/2008 7:01 AM
No virus found in this outgoing message.
Checked by AVG.
Version: 7.5.524 / Virus Database: 270.6.19/1659 - Release Date: 9/8/2008 7:01 AM2046 2008-09-10 10:59:27 timshoppa Re: Crystal filter success! Now matching to a homebrew RX. 2047 2008-09-10 16:47:30 kerrypwr Re: Crystal filter success! Now matching to a homebrew RX.
Yes; the two resonances are more widely spaced at higher frequencies;
perhaps that's why 9 MHz, 10.7 MHz etc are used in SSB-bandwidth filters.
I'm using a 5 MHz IF.
"Did your constructed filter's response look lopsided like Figure
3.39 in EMRFD?"
No; worse!!! It looked rather like the solid line in Fig 3.44(A) only
much worse; wavy across the top and with pronounced "ears" before the
passband drops-away.2049 2008-09-11 14:23:58 victorkoren Re: Crystal filter success! Now matching to a homebrew RX.
parallel, maybe it also helps implement a higher bandwidth filter?
Victor - 4Z4ME
2064 2008-09-14 14:45:34 Wes Hayward Re: Crystal filter success! Now matching to a homebrew RX.
Wow, great work. Glad to get the report on your excellent
results. Yea, I agree -- it's a real hoot to get a pile of cheap
crystals, do some measurements with simple homebrew gear, and then
whip up a filter that one would have paid over $100 for in 1970
dollars.
I would have commented earlier, but I've been off in the mountains.
It appears that you are working through the problems and having a lot
of fun with the discovery of the solutions, or at least the
problems. Keep up the good work.
One comment: I tend to use L-networks for impedance transformations
rather than wideband transformers, for they seem to fall into a
better controlled situation for me. That is, when I grab a powdered
iron core with a typical Qu>100 and put it in a very low Q circuit,
losses never get in the way and I end up with the impedance I want.
If I grab a ferrite core, especially one with quite a bit of loss
like the favorite for many of us, the FT-37-43, the transformation
can be off because of the loss. One can certainly design around
this, so it is not a bid deal. In a commercial situation where I
had to feed a high impedance (perhaps 3K) crystal filter, I ended up
with a wideband transformer wound on a core that was the same
material as the Micrometals -61, which is a pretty high Q mix.
Bottom line: It doesn't matter if you design it.
One detail that you discovered was quite interesting. If I
understood it, you found a correlation between crystal Q and the
frequency spread from the center of your group of 50. That is, you
found some of the lowest Q parts to be2072 2008-09-15 07:44:54 timshoppa Re: Crystal filter success! Now matching to a homebrew RX. 2077 2008-09-15 20:49:59 Wes Hayward Re: Crystal filter success! Now matching to a homebrew RX.
Many of us really like simple device models, and I am certainly in
that group. My view of a JFET amplifier for small signals is just a
current generator with gm that depends upon the DC standing bias
current with an open circuit for RF for the gate. But that simple
model is not enough. There will certainly be some input C, and it
is more with the J310 than with smaller area parts such as a 2N4416.
There is going to also be a resistive part to the input impedance.
Frankly, I don't know where these things sit at HF with these
parts. My new toy, a N2PK-VNA, should let me look at such things.
I've also noticed with the hycas IF amp that the thing tunes with
less C in the trimmer than seems appropriate. You have to really do
a C measurement to tell what is happening, for the usual 65 pF
plastic film trimmers have a max C that is more like 70 or a bit
more. Still, adjustment suggests that there is quite a bit of input
C with the FET. My ultra simple model is not good enough.
My version of the G3UUR oscillator runs all the time during a
measurement session. The crystal socket is one that will accept
about any leaded part. The socket was a piece of junk from the
semiconductor industry. A zero insertion force socket for ICs is
also a good part for this application. I then use gloves to grab
the crystals and to cram them into the socket. This keeps the
thermal transfer down and they seem stable enough.
One of the goals on my list is to examine the results of various
measurements and then to compare them. The gold standard here will
probably be the VNA using the N2PK program "Xtal.exe" to extract
data.
Have fun. It appears that you are doing some great work.
73, Wes
w7zoi