[HBR] The long, SLOW HBR project

[Walt Hutchens]

Well, THAT was interesting.

That VHF oscillation on turn-on?  Turned out to be not in the mixer, but in
the LO.   First time I've seen that.

The first couple of hours of work I was getting nowhere with it because it
was only showing up for a second or so at turn on.   I swapped mixer tubes,
no change.  Finally in desperation I put 100 ohm series resistors in the two
mixer grids ... NO CHANGE.

Because tuning the receiver changed the behavior and disabling the RF stage
did not, it had to be either the oscillator or the mixer.   I swapped LO
tubes, and suddenly the problem was solid:   Turn it on and the roaring
(very strong broadband noise covering everything else up) never goes away.

Flipped it upside down and began poking around the 12BZ6 LO with a wooden
lead pencil, and sure enough touching the LO control grid made the roaring
stop.   Touching the screen grid (oscillator plate) changed things and the
results varied slightly along the wire connecting the tube to the coil -- a
distance of maybe three inches.   Clearly in the oscillator section of that
tube AND clearly VHF.

(Why a lead pencil?  Because it is a resistor coupled through a tiny
capacitor to ground, namely, you.  It will eat significant energy from a VHF
circuit but have much less effect on HF signals.  Due caution necessary with
high voltages!)    

A 100 ohm resistor in series with the control grid mounted right at the tube
socket -- resistors for this purpose should have leads not over 1/4" long on
the socket side -- stopped the problem.

All the modern high gain triodes and many late pentodes will oscillate at
VHF when the tube is connected as an amplifier if the grid circuit looks
like a parallel tank circuit at VHF.   (In this case a 3" wire going to the
tuning cap ... 'parallel lines' tuned at the far end for VHF.)   However you
rarely see an HF oscillator pull this stunt because the negative grid bias
from the oscillation won't let the VHF oscillation keep going and there's
generally not enough time during the 'on' pulses of the tube for it to
start.  This circuit was an exception.   In effect the stage was
superregenerating, that is, the VHF oscillation was being quenched by the LO
frequency.   The roaring was the tube noise amplified by this action which
was wide enough bandwidth to cover the 1.7 mcs IF.

WW II radar jammers used amplified tube noise as modulation to transmit
signals that were megacycles wide.   I bet not too many people would have
tried using a broadband noise jammer as a receiver LO!

Ideally grid and plate leads in receiver RF/IF circuits are short enough
that hi-Q spurious resonances don't occur at frequencies at which the tubes
have substantial gain.   However that requirement is almost impossible to
meet in a receiver with plug-in coils when you factor in shielding, access,
the need for space around the oscillator coil to preserve Q and minimize
capacitive effects, and heat transmission considerations with the mechanical
requirements of the tuning cap and dial.   Too much real estate is needed
and there are too many conflicting 'zoning' requirements.

Next I tackled the 20 cps variation of oscillator frequency with AGC action.
While tinkering with circuit values to reduce sensitivity to plate voltage
changes was appealing, it didn't take long to realize this answer would only
be right on a single band.  (Transit time is part of what's being affected
...)  So, I dug a couple of 5651 voltage reference gas tubes out of the
archives and added another socket.   One tube turned out to be end-of-life,
so useful only for EOL testing (striking voltage goes up ...) but the other
was new.   Regulating both the plate and screen voltages on the 12BZ6
oscillator reduced the voltage change with full-range AGC action to ~1 volt
and the frequency shift to around 3 cps.

I had wound a temporary oscillator coil to shake out issues there; the next
job was doing a proper one.   This was a bigger chore than it should have
been because I had forgotten that I had completely rebuilt the receiver (new
chassis with revised tube layout) since calibrating the set with the
temporary coil.   So when I plugged in the new coil (which I thought was
identical L), the calibration was way off at the intermediate points on the
dial.

A couple of hours of head scratching later, I remembered the rebuilding AND
the fact that this setup makes synchronizing of the tuning cap to the dial
mechanism, critical.

(Remember that I generally make errors in filament wiring at least a couple
of times per project ... forgetting something like this is par for the
course!)

The Eddystone 898 and ball bearing-silver plated brass-soldered ceramic
tuning cap came as a pair from a very rough G3RKK receiver acquired several
years ago.  This is a UK 'Radio Communications Handbook' design,
contemporary with the W6TC sets but using a bandswitching front end.   The
front ends were sold as assembled units by the UK 'Electroniques' firm and
the dial was part of that package, already calibrated for (US)
160-80-40-20-10.   Thus I had a first quality capacitor that with the proper
coil and other parts should require only trimming to match a decent
pre-calibrated dial.

That wasn't quite as simple as it seemed, however, because in an effort to
use as much as possible of this excellent dial, the tuning capacitor is
rotated through more than the usual fraction of its range to cover each
band.   As a result, end effects are substantial.

At the low capacitance end, the 'open' end of the rotor starts to approach
the stator, so the rate of decrease of capacitance slows.   And at the high
capacitance end, the reverse is true -- the 'closed' end is approaching the
edge of the stator so the rate of increase slows -- although this effect is
less pronounced.  

The result is that while you can correct the high and low ends of the dial
by tweeking the trimmer capacitor and the coil and get a match somewhere
near the middle with a padder cap (effectively in series with the tuning
cap), the high frequency end can have a severe curvature one way or the
other if the rotor is even slightly off the correct position relative to the
dial mechanism.   Like -- errors at the other 100 kcs points in the ballpark
of a kc or two but 3900 is off by 20 kcs?   And 3800 by 10 kcs?

With great care in positioning the rotor it's possible to bring the errors
at all the 100 kcs points to under 1 kc.   That's as close as you can read
the dial and in fact less than the errors in the calibration of this unit --
if you look closely you can see that a few points are out of place by a
couple of kcs!  The calibrated dial is very pretty and accurate enough for
finding a net or QSO but not suitable for use as a frequency meter!

You'd think that the problem of tracking the various bands would be
simplified by looking at the original parts values.  Unfortunately the Radio
Communications Handbook I have doesn't show those values -- it assumes use
of the manufactured front end.   I do have the front end itself, but the
thing is a mess; even figuring out which coil is for which band isn't
trivial.   

(Looks like someone may have tried to convert a band for CB use or something
like that -- lots of loose wires, replaced wires with 'whatever' was at
hand, bare, DSSC, a couple of broken coils -- plus most caps are TINY and
located down inside a 2" deep chassis under the bandswitch and coils.)

There is some roughness on very strong SSB signals that goes away if you
crank the AGC threshold (RF/IF gain) control way back.  That's most likely
inadequate BFO injection, so I will look at that in the next day or two.
I'll also clean up the most recent oscillator wiring changes, and add a bit
of temperature compensation.

Ideally the chassis and each oscillator coil are independently compensated
so you can change a cold coil into a hot set and have minimum drift but I'm
afraid I'm not serious enough to get there.   Just tweaking the padder cap
inside the coil to get a three point match to the dial is a chore and since
temp compensation requires a final permanent assembly (no coil sitting loose
on the octal plug!), per-coil compensation would be a very tough job.

However in redoing the 80M coil I added one turn and spaced the winding a
bit.  With plastic forms (PVC pipe in this case) that makes the coil
somewhat self-compensating.   And the drift in the second minute -- after
the tube structure is nearly stabilized -- is now upward by a few hundred
cycles, so this lazy man's approach is helpful.   In the fourth minute and
after, the drift turns downward as is usual for uncompensated oscillators.
The total warm up drift (maximum freq. minus minimum) is now about 1 kc.

One point that's often missed in building stable oscillators: Self-heating
of tank circuit parts can be important, even at receiver LO power levels.
Start by operating the oscillator at the lowest possible power level, though
often that conflicts with the requirement to couple output as loosely as
possible.    Then use the physically largest caps you can find, orient them
for good cooling, and use two or more in parallel whenever possible.   Wind
the coil tightly with the heaviest wire practical and space turns slightly
if possible.  

This applies only to parts that are in the tank circuit, not to those
connecting it to other circuitry.  Typically this would be the tuning cap, a
parallel trimmer and any fixed caps in parallel, plus the padder cap that is
in series with the tuning cap.   (The Clapp and Vackar circuits have a
slightly different list of high current caps.)  And of course the coil
itself.   Getting high Q with fine wire and a ferrite core isn't as good as
getting it with heavy wire in solenoid form.

Also the advice to use very heavy wire hooking up the oscillator is not
always right: Heavy wire connected to a tube socket is a pipe for heat; with
a digital thermometer you can watch the temp rise flow out from the socket
to your tank circuit parts.   When fitting out a bomber, use #10 but for a
ham set use the smallest wire that will keep its place under the roughest
expected handling. 

Vibration is also an issue: A set with an integral loudspeaker will need
stiffer oscillator construction.   Loudspeakers should be as far as possible
from HF oscillators, use a heavier than usual panel, put heavy parts near
the speaker, and consider bracing the panel to the chassis in the vicinity
of the loudspeaker to eliminate mechanical resonance problems.

My digital thermometer shows the side of the coil nearest the center of the
set to be a couple of degrees warmer than the other side after half an hour
or so.   This could be either due to air flow -- the hottest part is the
117N7 on the far side of the set so air would tend to be sucked past the
coil in that direction -- or radiation from the 1st IF tube, located about
3-1/2" away on the warmer side.   Further investigation will follow.

I notice that the darn very slight warble is back.  That's gotta be 60 cps
getting into some oscillator -- most likely the LO -- but the effect is so
slight (and it comes and goes) so it's not easy to work on.   I should check
the bypassing of the cathode to chassis ground (cold cathode Colpitts
circuit) and (probably) move that filament to the grounded end of the string
-- It is #4 now and #1 (plate detector) got its problem taken care of in
another way.  I guess the warble is the priority project; an SSB receiver
cannot be satisfactory without rock solid oscillators.

Speaking of rocks, my 3500 kcs band edge marker crystal is a kc or two high;
I will check the shunt capacitance.   Of course it is about 70 years old
(has Signal Corps part number and marked for BC-1306) so this may be the
crystal itself.   

I'll probably do a 20M oscillator coil next, just to see how that goes.  I
won't be able to move the rotor relative to the dial for that one but
hopefully a two point match will be good enough.  The band is of course much
narrower percentagewise which should make it easier.

Anyway, it's a fun project.

Walt 
KJ4KV