Racal RA17C tips

[Radio Wrangler]

I thought I had one in the attic... somewhere! But only yesterday did I have a spare hour or so to have a hunt. I must have had it for 30 years. I knew it would be useful!

It's yours, free.

Now, how do I get it to you?

Keep personal addresses to private messages. The forum is open and gets trawled for personal info, so anything in the open risks an avalanche of spam. The moderators are pretty good at spotting and removing risky info, but they can't be sure to be quick enough.

David

[dtvmcdonald]

After doing lots of checking I put all tube shields on and the "fingered"
bottom covers. I do say that the latter caused very explosive cussing.

It now seems to work almost right. But there is one very serious flaw:
the AGC does not work. As far as I can tell, it does not work at all.
Changing signal strength with the attenuator does nothing. It overloads
at about the same setting of the RF/IF gain control in the MAN or AGC positions.

I don't understand the circuit. I tried disconnecting the 50K resistor to B+
and that did not make it work, though there was a small effect.
I checked the circuit again, as this was the part affected most
by the made capacitor remover and wire clipper.

Suggestions? Its hard to fix a circuit I don't understand.

[dtvmcdonald]

The AGC is fixed. The mad snipper strikes again! The ground for
the filament was cut, underneath another part. This was discovered
after I pulled the tube and it was less hot than the
other 6AL5. I first thought it was a bad tube, then a bad socket pin.

The set is now behaving more normally.

[dtvmcdonald]

I fixed the bent front panel with a big hydraulic press.

In the process I found out the origin of the mysterious plastic piece
I mentioned earlier. It is the plastic cover for the main kHz panel window.
Its hopelessly scuffed. I will make a new one of Plexiglass (polymethylmethylmethacrylate). I think you call this Perspex, is that correct?

Question: does anybody know what the correct glue is to hold this on, or
should I use double sided sticky tape?

[Radio Wrangler]

That switch is sitting here on my bench, ready to go into a jiffy bag. You're welcome to it for free. Click on 'Radio Wrangler' in blue to the left of this post, and select 'send private message' from the drop-down menu. This way you can send me the address to mail it to, without your address appearing on the open internet.

Does it need anything writing on the outside besides the address?

David

[dtvmcdonald]

Yesterday I gut the calibrator working. This took HOURS! It wanted
to oscillate at 90.909 kHz (1MHz/11) . The alignment procedure did not work ... it got a
nice stable 90.909. It required tweeking just so, both coils off resonance.

I found it a great help to use a scope viewing a Lissajous pattern of its output and
a 100 kHz synthesizer output). It would snap in and out of all sorts of strange stable
patterns, with unstable ones in between. A big problem was that I could tweek
it to be correct, then turn it off and when turned back on, it would
be at 90.909 again! Getting it to start at 100 required setting both coils off
resonance in just the right way.

The 1Mhz oscillator is within 1 Hz of correct!

[Colinaps]

Told you the calibrator could be a pain!

Good job well done though - it'll be a grand radio now. If the kc/s scale is tracking properly you'll hardly need it. Drift should be insignificant for voice purposes after half an hour or so.

Colin.

[dtvmcdonald]

The drift, after 10 minute warmup, is much less than the width of the
sharpest crystal filter position.

This morning I tested the alignment of the 100 kHz system. (I had checked the
three 30-40 MHz multistage filters before cecapping and they were OK.)
The 100 kHZ LC filter system was slightly off but responded to tiny tweeks
of all adjustments except one (the very first capacitor., C147) which was way off.
The 6 and 66 dB bandwidths are reasonably good, but the actual shapes are
far off what is in the manual. What is striking, however, is that they are off
in a very similar way to the way the 50kHz IF in my SX-88 is off. I suspect that the
published curves in both cases are not actual measured ones from a real set.
The bottom frequency slope is softer than the top one, though the 66 dB points
are reasonably symmetrical.

I normally check such things with a circuit simulator, but without coil specs including
the "coupling coefficient" that's impossible.]

The crystal filter alignment was perfect.

[dtvmcdonald]

I decided to try once more on the 100kHz L-C box.

Opening it up again, the problem was obvious: the second tuning cap
was wide open. I disconnected the two fixed caps, a mica and a ceramic,
and both measured near perfect. What could be wrong? I never
figured it out. So I changed the red mica one from 290 pF to 250pF,
then it tuned nice and sharp near the center of the variable cap.

The response curves were much nearer spec.

[dtvmcdonald]

Yesterday I tried checking the RF alignment. I did not do
the full alignment in the manual, with resistors, just little tweeks.

The coils tweeked sharply, quite close to where they were.

The trimmer caps had almost no effect at all. Is this normal?
The main antenna tuner works well, but sensitivity is not
uniform across the bands. Its not later stages because its
quite different for different RF ranges. I this normal?

[Colinaps]

Is the 37.5MHz drive reasonably constant through the range? You should be able to monitor it on one of the TPs above deck if I remember correctly.

73,

Colin.

[dtvmcdonald]

Yes, that's not the problem. That would not effect this problem, is uneven
sensitivity across one 1MHz spread, and the ratio of sensitivity
at X MHz+100kHz to that at X mHz + 900hHz is the same at all MHz except 0.

Tests show that the problem is (was?) uneven output of the 40MHz filter

I tried aligning the filter all day. No matter where I tried aligning the center frequency
there was a huge falloff below 39.8 MHz. I eventually had an
idea and shorted out the last coil, connected my scope
across the lat one with a (noncritical, I used 30K) resistor. Doing that it was
easy to get a nice perfectly OK response.

I tried disconnecting the output (which has a 2.2k resistor across it). That too was
passable. I eventually discovered that C108 was set to null that frequency!

There are no instructions for aligning C108. I had already aligned L50 per instructions
(just peak it) L50 tunes the 38.5 MHz output while C108 adjusts some sort of balance.
Adjusting it got a reasonable 40 MHz filter response. The null is between
38.5 and 39.5 MHz.


A side effect is an enormous increase in stage gain.

Does anybody know how this is supposed to be done.

How many microamps are the MHz "chuffs" supposed to be
with no antenna at all, using a tuned, not broadband, RF?

[Radio Wrangler]

There have been instances reported of failed mica capacitors in those filters. I've not come across one myself but it does shoot down the frequent saying that 'micas never need changing'. There is the possibility that one is affecting your filter and cannot be adjusted out.

The manufacturer and military organisations had special setting jigs to aid filter adjustment. I've never seen one.

With a coupled resonator filter of this sort, one adjustment method (I'm avoiding the use of the word 'alignment') is to sweep one resonator at a time with its neighbours shorted and to adjust it for resonance on the geometric centre frequency, repeat for the the resonators. When the shorts are removed, the filter should be very close to its correct shape.

Resistors at the ends of filters aren't attenuators, they are there to define impedances terminating the ends of the filters. They waste signal, but are needed to define the in-circuit Q of the end resonators. Without them you get more signal but less bandwidth.

High order rectangular shaped filters are not easy to adjust. The adjusters interact quite strongly and the whole process tends to diverge, not converge on the wanted response. This is where the shorting links and peak individual resonators approach wins. Also, if you do have a duff capacitor, you find one stage can't peak in the right place and you only have to check parts in its vicinity.

Those filters were designed in the fifties. I'm not sure how they did the maths. There were the old m-derived and constant-k techniques, but the much better k and Q techniques really came in after Zverev's book, which came out in 1967.

The idea is that each resonator in the chain creates a pole exactly on the geometric centre frequency. This would look like it was heading to produce a peaky, synchronously tuned filter, but it doesn't. When the coupling factors come into play, the poles split up into the wanted pattern (around a half-ellipse on the s-plane to create a Chebyshev response) With the coupling working, it is counter intuitive, but each resonator stage does NOT create an individual pole. The number of poles is equal to the number of resonators (well, double if you include the poles in the mirror-image response at negative frequency) but one adjuster does not move one individual pole. This is where the fiendish interaction of the adjusters comes from.

My apologies if that last bit seemed meaningless, but filter design is done by looking at the equation of the transfer function of a circuit (gain and phase responses) and finding the roots of that equation. Bizarrely these roots involve positive and negative and imaginary numbers of Hertz of frequency. Obviously these are impossible, but the patterns of where those roots sit on a chart of real and imaginary frequency turns the design task into a bit of simple geometry. Much nicer than having to solve high order differential equations.

Shape is important. These filters must not be adjusted by peaking them up at a spot frequency. That gives you varying sensitivity over the range of tuning of the kHz knob. from the peaked response in the passband. RA17s tuned this way are rather disappointing, and you find that as you tune across one of the 1MHz bands, you have to keep peaking up with the MHz knob.

Back in the day, adjustment of these filters required a wobbulator (great word!) and an oscilloscope. A modern spectrum analyser with tracking generator or a network analyser is a better bet. The last time I ventured into this area in an RA17 was 20+ years ago and I didn't keep the details in my head. I have a frequency synthesised spectrum analyser with a tracker so I can just type in centre frequency and span and expect them to be accurate.
I used resistive probes (a length of 50 Ohm cox with a 5.6k series resistor and a blocking capacitor on the end) to inject and sample my swept signal.

David

[dtvmcdonald]

I used the adjustment method you describe ... it worked perfectly .. so long as
the output termination was resistive. The input termination in operation is the plate
resistance of the tube V7, the output termination is that 2.2k resistor in parallel
with the adjustable network L50 - C113 - C108 - tube capacitances of V9 and V10.
This latter means to do a passable test I needed to pull V9 and use a pure capacitive
scope probe at about 8-10 pF at TP3 to match it when turned on to test this
mechanism (using antenna input of the signal and inserting or removing V8
as needed for oscillator). This got it close enough ... I think. This morning
I managed to simulate the whole circuit and think I now understand it.
It clearly was seriously misadjusted; the simulator matched reality well enough.

The set seems to match specs well now.

Oh yes ... no mica caps were actually bad, somebody replaced them in perhaps
the 80's by their look. But two were barely in tuning range so I replaced
those with slightly (4 or 5 pF) different values.

As I now understand it I can devise a truly "proper" way to adjust it.

[Richard - F4VPR]

Originally Posted by Radio Wrangler

. . . . . . . .
Back in the day, adjustment of these filters required a wobbulator (great word!) and an oscilloscope. A modern spectrum analyser with tracking generator or a network analyser is a better bet. The last time I ventured into this area in an RA17 was 20+ years ago and I didn't keep the details in my head. . . . . . .

David is of course quite right, last RA17 I did at GCHQ was 52 years ago and also the memory is (long) gone.

Wobbulator yes! Many a long year now to re-discover that terminology. . . .

A hulking great lump of R&S Polyskop and quite some time to get that set-up appropriately before starting in on the filter which as has been noted tended to move away from the ideal gentle "W" shape rather than towards it.
Hours of fun and frustration for us trainee radio techs.
I hope you succeed!
R.

[dtvmcdonald]

Originally Posted by Richard - F4VPR

Back in the day, adjustment of these filters required a wobbulator (great word!) and an oscilloscope. ...

I hope you succeed!
R.

Oh I already succeeded ... the "do one coil at a time" method is very easy and
worked great.

I got a new switch thanks to a guy in the UK and have it finished and
all buttoned up even with rubber feet.

The only serious design problem is that the AGC do not work on SSB.
Really all it needs is a 10 uF cap in place of the 1uF one. I've had
good luck using top grade electrolytics rather than film caps
(but a 10 uF 100V one will fit easily). I may try that when a night comes
that will really test it (meaning that one of our local hams is on the air on HF.)

[dtvmcdonald]

Originally Posted by dtvmcdonald

The only serious design problem is that the AGC do not work on SSB.
Really all it needs is a 10 uF cap in place of the 1uF one.

Not true!

The design is amazing ... I simulated it (I finally designed a Spice simulation
for the actual 6AL5.) With the RF gain all the way up it
has a hopeless "AGC delay" due to a +25 volt bias at the bottom of R116
and R122. Also, 20 Meg is too low for R81B+R81C. I propose
changing R81B+R81C to 100Meg and the R121/R122 at 50K to 2 Meg .

This is very easy and I will see how it sounds (assuming I get a
sufficiently strong SSB or slow CW signal tonight, propagation
has been so hthemorrible it has to be local.)

WHY would they design it like that?

[dtvmcdonald]

Well, those changes did not work. I had not simulated the full AGC feedback loop,
only the generation of the AFC voltage. The thing is marginally unstable
(highly underdamped) and those changes make it worse.
Do the only way to make it work better on SSB is increase capacitor
values. Increasing C159A to 0.47uF made it stable, and
increasing C188A to 3.3 uF made SSB sound better, though I suspect
5.6 or 6.8 would be better. I will try a 5.6 or 6.8 after my next capacitor order.

[Radio Wrangler]

The RA17 was never designed for SSB. That came later.

Its intended use was RTTY and CW with occasional AM. It just shoves a BFO signal into the AM detector. It isn't a proper product detector and the AGC suffers. Far less sophisticated than the rest of the set and you find you have to drive it for SSB in much the same way as an HRO/AR88/CR100 era set.

Use for SSB was supported via an add-on box, RA63. This was effectively replacements for the 100kHz IF right through to the speaker. There was an ISB version as well but it was really designed for a pilot tone system. Even with one of these, AGC is the Achilles' heel.

By the time of the RA1217 Racal had started fitting proper SSB filters, product detectors and effective AGC. However they are early transistor sets and the front ends are very 'soft'.

David

[dtvmcdonald]

Proper product detector and AGC are completely different things. Truly proper
AGC is really impossible for SSB or any signal that is not "on" for a substantial fraction of the fading time, and/or when on never falls into the noise. If
the latter is true then AGC falls into the same category as automatic gain riding for broadcast radio or TV, and Mr. Orban finally got that really right.

The complete separation of signal and AGC detectors as in the RA-17 or R390A are sufficient to supply all the info needed for the AGC. High fidelity SSB is
a very very different matter.