HP8640B Signal Generator

[Nymrod121]

Hi Allan - it's covered in HP manual ref 90125 - my email of just now refers
Guy

[allan]

OK Guy, the oscillator came out far enough for me to remove the inspection panel and the output cable is back in place, but not tested just yet.
I couldn't find a suitable new right angle bulkhead socket. The old one had lost it's securing thread. The nut, still threaded to it and the lock washer were lying inside the RF amplifier compartment of the oscillator and RF connection broken off.
A simple fix as it turned out. The remains of the thread had a flat to which I soldered a couple of gold flashed wires cut from a surplus CAT5 socket. The socket was then a force fit into the mounting hole. I superglued it in place so it couldn't free itself, soldered the ground wires to the RF pcb and the RF output back to the connector.
I've noticed from gouges that someone has unscrewed most of the rigid coax fittings and must have over tightened the one on the oscillator output socket.

I also superglued shims into the cracks in the tuning gear hubs. Presumably, once shrunk, the plastic won't expand. The cracks didn't extend into the main part with the teeth.
Allan

[G0HZU_JMR]

To add to my earlier answer about the ALC diode issue, both the 'quick fix' and the 'better solution' have issues. I can try and predict what these will be if it helps in the choice between them?

The quick fix will present the wrong source impedance at low attenuator settings and this will introduce a lot of mismatch uncertainty when trying to set the sig gen level with the rotary vernier at high output levels where the step attenuator is set at 0dB. The mismatch uncertainty would be extremely significant here.

The 'better solution' will overcome this problem but this solution may well cause the ALC to drop out into open loop because the output amplifier has to deliver more voltage at its output node in order to achieve the same target RF voltage at the detector diode in its new location. This is because it is now at the top of a bigger potential divider with 39R + 39R + 50R.

I'm not sure how significant this will be as there will presumably be some excess gain margin in the design of the ALC system and some margin in the output amplifier in terms of how big a signal it can produce before limiting. It may be that the ALC system will only go open loop when trying for full depth AM near the maximum RF output the sig gen can normally deliver with AM. Under these conditions the ALC will be working nearest to its design margins.

It may still be able to stay in closed loop for cw operation at all sig gen levels (and maybe it will also be OK for AM for most sig gen levels) and in this case the sig gen might be OK for the majority of applications. So this solution may be adequate for your needs.

If it goes open loop with full AM modulation I'd expect to see the modulation waveform show some clipping in places as the ALC system drops in and out of closed loop.

[Julesomega]

I was interested to see on the VE7CA blog that he had replaced the failed power amp MMIC with a Minicircuits surface-mount MMIC:
http://www.ve7ca.net/TstM86.htm
and googling brought up another:
http://http://www.dl7maj.de/Repair%20PreAmp%208640B.pdf

K6JCA had used a MAV-11 which it seems will just about squeeze +20dBm with some compression which is not going to matter here.

(Not that this will help you at this time)

[allan]

I'll be studying those two articles later Julian and maybe investigate what else is available in case I need to replace the driver hybrid. I had a similar problem with my Wavetek 2407 sig gen. First the MRF571 driver failed and then the PA. Both were a bit weird with the devices developing serious leaks, perhaps tin whiskers or similar? I replaced the driver with the correct device but the PA (MRF839F the "F" was important) was tricky so I put a reverse bias across the transistor and fused open whatever was leaking and all was well for a spell. Recently the output blips off and on so I bought a vaguely similar device which remains unfitted. When the cheap HP8640B appeared I thought it might serve as a replacement but not so quickly it seems. The major advantage of the HP is its smooth tuning instead of having to push buttons. Digital is not very convenient!
Turning to my quick mod. I used to work at that place in Priors Road donkeys years ago Jeremy. The 35 ohm (39?) resistor seems to have been carefully calculated and I notice the detector diode is fed partly by current through this as its 1K DC bias resistor is on the opposite side. The key thing is the absence of a capacitor which would make the circuit frequency sensitive. The 200 ohm series resistor looks also carefully calculated. Scroll down to the drawing by R.P. in 2007.
http://www.radiomuseum.co.uk/HP8640B.html
If the detector (with 35+50 ohms) results in 0dBm in the AGC loop from the circuit parameters, and I add another 35 ohms for example (or something less?), can I adjust the AGC parameters (using ohms law) to compensate for the extra (=70+50) ohms to produce 0dBm at the attenuator input? Or is this too simplistic?
Allan G3PIY

[Nymrod121]

Originally Posted by Julesomega

I was interested to see on the VE7CA blog that he had replaced the failed power amp MMIC with a Minicircuits surface-mount MMIC:
http://www.ve7ca.net/TstM86.htm
and googling brought up another:

http://www.dl7maj.de/Repair PreAmp 8640B.pdf


(Julian: your second link wouldn't load - hope this helps - Guy)

[G0HZU_JMR]

Originally Posted by allan

Turning to my quick mod. I used to work at that place in Priors Road donkeys years ago Jeremy. The 35 ohm (39?) resistor seems to have been carefully calculated and I notice the detector diode is fed partly by current through this as its 1K DC bias resistor is on the opposite side. The key thing is the absence of a capacitor which would make the circuit frequency sensitive. The 200 ohm series resistor looks also carefully calculated. Scroll down to the drawing by R.P. in 2007.
http://www.radiomuseum.co.uk/HP8640B.html
If the detector (with 35+50 ohms) results in 0dBm in the AGC loop from the circuit parameters, and I add another 35 ohms for example (or something less?), can I adjust the AGC parameters (using ohms law) to compensate for the extra (=70+50) ohms to produce 0dBm at the attenuator input? Or is this too simplistic?

I can't give any firm answers but to get the detector working as the original will require some knowledge about how much frequency compensation there is in the detector across frequency.

For example, if the frequency response through the step attenuator droops by 1.5dB across 0.5-1100MHz then it could be that the diode detector will be designed to have the same amount of 'droop' across this frequency range. I'll give it some deeper thought over the next few days but I can offer the following guesses straight away...

Your 1N5711 detector diode is probably a poor choice because the self capacitance is quite high and it varies with reverse bias. This is not good for this application because the diode has a series 200R resistor ahead of it.

At a couple of volts reverse bias this diode will have maybe 1.3pF capacitance. Work out the RC rolloff of 200R into 1.3pF and this detector will probably droop 3dB by 600MHz. This seems a lot. This assumes the detector is built with low inductance SMD parts. If it is built with old school leaded parts like your version then it could go the other way and become more responsive at UHF because any series inductance will form a step up L match into the diode at UHF. So the choice of diode and the layout will be quite critical here. I would guess the ideal choice would be a low barrier Schottky diode with a fairly consistent 0.3pF self capacitance wrt reverse bias. It would have to be built really tight with SMD parts if the aim was to achieve a graceful 1.5dB downward slope across 0.5-1100MHz. I think one reason the 200R resistor is there is to have some control over the detector droop vs frequency.

But all the above is just a guess, the team at HP that designed the detector will know all the answers to the above and I may have got it all wrong.

[Radio Wrangler]

I think this is the same die but in SMT package

https://www.mouser.co.uk/datasheet/2...,0-1217119.pdf

The problem is that HP begat Agilent begat Avago and a financial consortium bought Avago. With the might of the financial consortium behind them Avago bought Broadcom, liked the name and changed their own name to Broadcom. The new Broadcom saw more growth potential in Broadcom sort of parts, they systematically obsoleted every single useful RF device Avago made, and it was definitely one of the best such ranges. So all HP's nice schottky diodes are no more, not even in SMT.

So you need to go looking for something comparable. MA/COM or Skyworks maybe? There are some NXP possibilities, but sawing your own leg off is more fun that trying to find things by parameter on their website.

David

[G0HZU_JMR]

Thanks David. Those HSMS-282x diodes look to be better suited because the capacitance is lower and it is more consistent wrt changes in reverse bias. They are also good to -15V so they should be quite rugged. At a guess a diode rated to maybe -8V with 0.3pF capacitance might be even better but the HSMS-282x range should be worth trying even if it meant adjusting the 200R resistor for any fine tuning of the response.

By contrast, the 1N5711 capacitance changes from about 1.3pF at -1V to about 0.7pF at -3V. This would mean the bandwidth of the 1N5711 detector would change as the analogue amplitude attenuator/vernier control was adjusted on the HP8640B front panel as this control sets the target ALC detector voltage. The combination of the 200R resistor and the changes in diode capacitance would cause this problem as the rotary control was changed from 0dB through maybe -15dB.

All this analysis may seem a bit critical and OTT but I bet the designer at HP spent a fair bit of time optimising the detector performance across frequency and across all the rotary vernier range. Looking at the design of this sig gen HP tried very hard to keep the amplitude accuracy within tight limits. The detector performance will be a critical part of this. It would be really easy to throw this performance away with a casual replacement for this detector even though the detector just looks like a simple peak detector at first glance.

[Radio Wrangler]

Does Alan have a 432A/478A to check the output against?

David

[allan]

I have a slightly better diode, a 5082-2835 which I could try, otherwise I'll order an SM type.

I have a few bits of test gear David, including an HP 431C and a DSA-815TG, but when I reassembled the oscillator, and the switch on the back of the attenuator, the meter dropped to near zero output although I can still see 180mV at the PA and 2v at the driver.
I think I muddled up the wiring and the modulator circuit must be switching off the output to the PA. I found the vernier pot wiring was wrong and its possible the wipers for one or two functions are not making decent contact. The fault I was trying to fix ie. alternately lighting the 0-3/0-10 scale lamps is now working perfectly though! Pity the meter reading is zero.
A bit more testing is due Thursday.
I'm now clear about the tuning problem so that's on the list.
Allan

[G0HZU_JMR]

I have quite a few of the classic (sadly obsolete) Avago Schottky diodes here plus some low capacitance Infineon alternatives.

I have several HSMS-282x types and I could tack together a basic detector using SMD parts to show you the response if that helps?

I have various sig gens here with excellent amplitude flatness, typically <0.08dB error across LF through 1GHz and it should be possible to prove the response of a 'contender' for your sig gen detector. My guess is that (to offset the droop vs frequency in the step attenuator) you would want maybe 1.5dB to 2dB 'droop' in the detector by 1.1GHz and this would require a diode with about 0.5pF capacitance when driven by a 200R source. The hot end of the 200R resistor will be a voltage source or 'virtual ground' because of the ALC action so I think that only the 200R resistor sets the resistive part of the time constant for the droop. There may be other parasitics that influence the droop/rolloff of a practical design and the layout of the detector components will be critical here. If old school leaded parts are used instead for the detector components then all bets are off with respect to the frequency response. Depending on how sloppy the layout gets the detector could easily have an undesirable peak somewhere up towards UHF. This would give the opposite effect and any positive peaking of the detector at 1GHz (say 2dB) would cause the sig gen output level to appear very droopy by 1GHz. It could easily droop by 4dB by 1GHz in this example.

[Radio Wrangler]

stray-C across that 200 Ohms?

David

[G0HZU_JMR]

Yes, I think that will have some effect too. Over lunch I downloaded an image of the insides of the output amplifier module from the HP memory site.

I think I've spotted the ALC diode and the associated components and I've marked up the image below. It looks like there is a gimmick shunt capacitor shape by the detector diode to deliberately add some controlled capacitance. It looks like the capacitance can be doubled by adding link wires to the left hand capacitor shape within the blue circle. Also the 200R resistor may be adjustable to another value by snipping bond wires, but that is just a guess. This shows that HP were probably experimenting with the RC time constant into the detector and this was presumably to compensate for the droop vs frequency through the step attenuator.

The junction of the 200R resistor and the 39R resistor will be the place where the source impedance will be close to zero due to the ALC feedback action and I've arrowed this in the image below.

I'm not sure the technology needs to be quite this tiny and advanced for a 1GHz diode detector but it looks impressive!

[G0HZU_JMR]

The other reason for selecting the right diode is that the attenuation scaling on the vernier control will presumably be tied in with the IV characteristic of the detector diode. This would probably become more and more critical as the attenuation is increased via the rotary vernier control because changes in detector efficiency will be most apparent at low detector drive levels.

In general terms, the diodes with lowest Vr ratings tend to be the most efficient and have the lowest self capacitance. But they tend to be a bit fragile...

[allan]

I managed to fix the zero RF output fault which I'd self-inflicted.
The rotary switch on the rear of the attenuator was the culprit. I'd dismantled it and fitted a new springy bit, for the meter scale lamp (the "Odd" circuit), which had disappeared. As the adjoining spring contact was wobbly and missing a plastic securing pip I'd moved all 3 to the opposite side where there's a set of spare pips. I superglued the two existing springy bits in place together with my new one made from a bit of relay contact, then reassembled everything only to find the vernier switch via the pot was open circuit. In this state the RF output is reduced to near zero.
I couldn't figure out why the vernier circuit was open-circuit but eventually found the two original spring contacts were coated in superglue. The glue was so thin it wasn't visible. The glue must have an incredibly low surface tension. Cleaning it off brought the vernier back to life with full RF output available.

That picture of the hybrid might come in useful Jeremy if I can't get the external circuit going properly. It looks repairable if I can prise the lid off.

Currently I'm working in the range 4 to 8 MHz (3386-8601KHz apparently) but I plan to overhaul the Peak Deviation/Range assembly next. I haven't tried that yet for fear of breaking the gears which look vaguely rescuable.
Allan

[G0HZU_JMR]

That picture of the hybrid might come in useful Jeremy if I can't get the external circuit going properly. It looks repairable if I can prise the lid off.

It might also be worth a look inside a 'real' example rather than the image from the HP memory site. It could be that the 200R resistance on the schematic represents the maximum resistance if both bond wire links are removed. This may mean that the typical resistance in a real production hybrid could be less than 200R if the bondwire links are fitted as per the image I posted up. It would be easy to measure this resistance with a DMM if the lid was removed.

To get and idea of scale, the 1k resistor trace nearby is a similar width and if you added up all the black resistive segments within the 200R trace they probably add up to a fifth of the length of the 1k resistor trace. But that assumes the resistive material is the same for both and the same depth.

If you can't get the lid off the one you have and your revised detector still gives the wrong response then I don't think the 200R resistance value is set in stone. I think it is there to provide a controlled amount of droop in the detector response.

[Radio Wrangler]

I think the bond wires were selectively applied with the thing running in a test jig. The two capacitive squares near the 200 Ohm resistor and the collector of the transistor (and HP21 die perhaps?) would be bonded into circuit as needed.

Thus the part only goes under the microscope once and only the minimum number of bonds are made - saves time and gold wire.

I used to be able to fix things like this. Had the facilities available. You don't value things until you lose them.

David

[allan]

Here's a progress report folks. There are a few fairly major problems remaining. The A9 assembly is now detached and two of the three rotary switches have only a single spring contact. Only a couple of plastic gears need intensive care and maybe one or two just a spot of glue. The FM slide switch is sticky but seems OK. The AM slide switch is now detached and waiting for inspiration to fit a new lever. The attenuator knob has so far refused all attempts to come off and needs a new printed scale. The multi-turn tuning mechanism needs a tiny adjustment and some scalpel work in-situ on the inner plastic disk. The output amplifier is awaiting further experiments which I'm putting on hold until the range can be moved from 4-8MHz where it's stuck until A9 faults are cleared.
Fortunately, most if not all the problems, seem to be stock faults and have been described with lots of helpful suggestions and pictures.
I see a late mod is adding buffers under the transparent top of the mode switches.
Allan

[allan]

All the remaining HP8640B repairs are complete.. broken AM slide switch, faulty tuning, seized cams and A9. It just remains to refit the A9 assembly, switch on and investigate the AGC loop. I wish it could be so simple...
Now the lawnmower needs attention after emitting a clonking noise, and a rattle before stalling and refusing to start.
Allan