Hewlett-Packard 3314A function generator

[synx508]

It's too late, I've replaced the 2N3904 that had a probe clipped to its base earlier. The emitter was cut really short compared to the other two leads, so perhaps there was a soldering defect. All the joints looked great from the top, though. Let's see what happens…

[synx508]

Update, I'm now chatting to the son of the late designer of the 3314A on the fediverse. My 2N3904 repair appears to be holding. Oh god. All that work and it was the cheapest transistor in the whole unit! (maybe)

[tautech]

Update, I'm now chatting to the son of the late designer of the 3314A on the fediverse. My 2N3904 repair appears to be holding. Oh god. All that work and it was the cheapest transistor in the whole unit! (maybe)

Not abnormal.

Once every other passive has be checked one must start asking questions of the normally reliable silicon.
I've had the odd frustrating repair also but not as complex as this HPAK AWG and kept a shitty old DMM with a transistor tester for times as these and it's identified low gain transistors on several occasions ...they look and test okay but gain is outside datasheet spec and this where you need be careful as old devices often had several gain specs, sometimes identified with a letter suffix and other times by a colored dot.
A trap for the less observant !

Fingers crossed you've cracked it.

[synx508]

Don't panic, it's not fixed! It has restarted in the usual time of just over an hour from cold start.

Right, that 1458 is coming out…

[Zenith]

I've known transistors fail but they've gone from good to bad, not complex behaviour, such as working when they are cold but not when they are hot. It's usually been power transistors rather than small signal ones. I've had the occasional op-amp become noisy after being abused, but again, it's been noisy all the time.

At least parts like 2N3904 and 1458 are cheap, and you probably have a few ready in your stock.

[synx508]

Ha ha ha, it wasn't that, either. This is amazing.

[synx508]

This absolutely isn't about hot/cold, but temperature is involved. I guess the ramp down to 11.8V on +15V must be coming from an external overload after all. I suppose the DMM missed it as it was working in chunks of 250ms or thereabouts and the whole event is over with in 22ms. That would explain the slightly higher current measurement that wasn't sufficiently high to cause overcurrent protection.
I need to measure the current with two channels of oscilloscope and inverted summing (I don't have a differential probe), I suppose. I can optimise the capture now I know the time and amplitude of the event and maybe it'll reveal a useful detail.

[tggzzz]

Is it possible to test the 3314's PSU when not connected to the rest of the circuit?

Or to use a bench PSU on the rest of the 3314, with the nominal voltage and a current limit slightly above the normal operating current? Objective would to be to (1) see if normal operating current drifts) and (2) spot any sudden large "breakdown" load current drastically reducing the bench PSU voltage.

[Zenith]

I was going to suggest that. tggzzz beat me to it. It looks as if the glitch comes from the +15V supply, but it's not impossible it's caused by a spike in demand from the AWG circuitry itself.

The AWG circuitry could be powered from a bench PSU. The +/-15V and 5V requirements don't seem too demanding. There may be a problem in the sequence in which the voltages are expected to appear.

The PSU could be used to power dummy loads and the glitch watched for. Powering the AWG from the bench PSU would establish what the current requirements were.

A common fault finding technique is known good sub unit replacement. Given a good unit, swap the PSUs, or other sub units, and see whether the fault follows a particular sub unit. It's not guaranteed to work, because the problem may be subtle incompatibilities between sub units. In this case you only have one 3314A, but the bench PSU and dummy load approach may shed some light.

[synx508]

Substitute power supply seems like a good idea but I'm going to have to carefully check that I don't backfeed anything.

I've just completed a test with probes either side of the +15V current sense shunt, which showed the same ramp shape, unfortunately I'd left the trigger coupling on AC and it didn't trigger where I really needed it to. I don't have the whole event. I'm running that test again now because although I saw the ramp shape it wasn't the way up that I expected it to be, the waveform only shows the part where the computer is rebooting, when overall current draw is lower, so the unregulated voltage rises.

There's one of those yellowish axial tantalum capacitors on the generator board, tucked under a heatsink that's heated by a couple of TO126 transistors. There are also a couple of larger (33µF) yellowish axial tants elsewhere on that board. I have had these fail in my 3325B. These are all on 1995 datecodes because it's perfectly normal to have boards with build years more than a decade apart!

Casually wondering if the mains transformer or maybe the half wave rectifier could have a fun fault where a half cycle goes astray, that would fit the timing of the problem, even if the seemingly fine capacitors would likely prevent such a dip in voltage. The most recent traces are on the unregulated supply but don't show evidence of a drop like the regulated side. Unless it was on the part of the waveform that the scope didn't capture.

[Zenith]

There's one of those yellowish axial tantalum capacitors on the generator board, tucked under a heatsink that's heated by a couple of TO126 transistors. There are also a couple of larger (33µF) yellowish axial tants elsewhere on that board. I have had these fail in my 3325B. These are all on 1995 datecodes because it's perfectly normal to have boards with build years more than a decade apart!

Casually wondering if the mains transformer or maybe the half wave rectifier could have a fun fault where a half cycle goes astray, that would fit the timing of the problem, even if the seemingly fine capacitors would likely prevent such a dip in voltage. The most recent traces are on the unregulated supply but don't show evidence of a drop like the regulated side. Unless it was on the part of the waveform that the scope didn't capture.

Is there any good reason for using tants in those positions? Was it part of the craze for using them everywhere, which I thought was mostly an early 70s thing? Is the voltage rating conservative, or are they right on the edge? Can they be easily replaced with modern electrolytics? I certainly don't replace all tants on sight, but I always feel uneasy about them, especially across power lines.

I suppose the mains xformer could have a fault caused by a dodgy contact causing it to drop the odd cycle when it gets warm, but it seems like an oddly specific fault. Have you remade the soldered joints on the rectifier? They can play up.

[tggzzz]

As zener notes, the PSU rail sequencing would have to be checked before powering it from a bench supply.

The 1740 scopes has known problems with its bridge rectifiers, but I don't think that would match your symptoms.

Missing a half cycle could conceivably be the transformer breaking down internally. (I say "conceivably" because I've conceived it, not because I believe it) If that's the case then I would simply replace the transformer with modern AC/DC converters. I did that in one of my Tek 1520s.

[tggzzz]

Is there any good reason for using tants in those positions? Was it part of the craze for using them everywhere, which I thought was mostly an early 70s thing? Is the voltage rating conservative, or are they right on the edge? Can they be easily replaced with modern electrolytics? I certainly don't replace all tants on sight, but I always feel uneasy about them, especially across power lines.

Until the mid 80s tants were far better than aluminium electrolytics in terms of size, HF performance, reliability especially in hostile environments. There was, of course, a significant difference between beads and axial sealed tants!

Nowadays the size and HF performance favours ordinary electrolytics, leaving sealed axial tants for hostile environments. Where I've had to replace axial tants, I've used decent PTH electrolytics chosen principally so they would fit in the confined space.

[EC8010]

In theory, there was a good reason for using tantalum bead capacitors across the output of a regulator. All regulators have an output impedance that is resistive at low frequencies, then becomes inductive as frequency rises. You can treat that as a Thevenin source of series resistance and inductance. When you put a capacitor across the output, you create a series resonant circuit and if capacitor ESR is zero, the Q is very high and oscillation becomes likely. So, you use a tantalum bead capacitor because it has quite high ESR. However... If you measure tantalum bead capacitor ESR with an ESR meter, you find it's very variable. I prefer to use an aluminium electrolytic and deliberately add a small resistor if necessary.

Axial tantalum capacitors are a completely different beast; and akin to an aluminium electrolytic.

[synx508]

As zener notes, the PSU rail sequencing would have to be checked before powering it from a bench supply.

The 1740 scopes has known problems with its bridge rectifiers, but I don't think that would match your symptoms.

Missing a half cycle could conceivably be the transformer breaking down internally. (I say "conceivably" because I've conceived it, not because I believe it) If that's the case then I would simply replace the transformer with modern AC/DC converters. I did that in one of my Tek 1520s.

One of my 1740s has a wonderfully cooked rectifier board, with the evenly browned Molex connectors. It seems to work fine, though.

My plans to further test this thing have been detected by the fault, which is refusing to show itself. That's two restarts today, both before lunch. But it does this, the restarts aren't evenly distributed. I've even blasted it with the hairdryer today and no response at all.

[synx508]

Until the mid 80s tants were far better than aluminium electrolytics in terms of size, HF performance, reliability especially in hostile environments. There was, of course, a significant difference between beads and axial sealed tants!

There's a mix of axial tants in metal cans, axial tants in the yellow resin and rather conventional electrolytic capacitors. There aren't many of any of these, though, it's a surprisingly big-cap-free design. They're only where they're absolutely needed. Board level guardians of power rails, a time constant or two. Nowhere else. The designer understood that you could get long time constants with big resistors, there's a 22M in the power supply, for instance. And a warning in the manual that the entire thing is susceptible to leakage if not kept absolutely clean…

[synx508]

Axial tantalum capacitors are a completely different beast; and akin to an aluminium electrolytic.

Yup, I think they're used in place of electrolytic capacitors because they can tolerate high temperatures for long periods of time. You wouldn't normally stick capacitors that can dry out under the metal wings of a heatsink, but HP buried a couple of low value sealed axial tantalums there. It's a very dense design that runs quite hot. Oh gaaaaaah, it just restarted and the scope didn't trigger.

[tggzzz]

22Mohm isn't small. Leakage indeed!

Might it have drifted upwards and might that cause the resets?

[synx508]

I checked the 22MΩ, it's within tolerance. Back to waiting again now, as I'm hoping to get more detail of a glitch, not that it'll help me much.

[Zenith]

In theory, there was a good reason for using tantalum bead capacitors across the output of a regulator. All regulators have an output impedance that is resistive at low frequencies, then becomes inductive as frequency rises. You can treat that as a Thevenin source of series resistance and inductance. When you put a capacitor across the output, you create a series resonant circuit and if capacitor ESR is zero, the Q is very high and oscillation becomes likely. So, you use a tantalum bead capacitor because it has quite high ESR. However... If you measure tantalum bead capacitor ESR with an ESR meter, you find it's very variable. I prefer to use an aluminium electrolytic and deliberately add a small resistor if necessary.

Axial tantalum capacitors are a completely different beast; and akin to an aluminium electrolytic.

I'm not a tant hater, although I have come across a few which glowed red and damaged nearby parts and the PSU. For the most part they just sit there and do their capacitor thing. I have come across people who are tant haters, including one who claimed one had exploded, narrowly missed his head, and the remains were embedded in the wall. I have encountered a few which have exploded, although there were no signs that it happened with such violence as to embed itself in a wall. I've also heard horror stories about exploding electrolytics, which I don't believe.

Tants have their strengths and I have replaced them with other tants. I do believe they were overused and under specified back in the day, and the technology was very likely immature in the early 70s. Bead tants at least. I certainly wouldn't touch bargain bags of ancient looking bead tants from a swapmeet. I have little experience of axial tants.

Synx did say he'd had to replace similar tants in other HP gear, which started me thinking along those lines.

As for the temperature requirement close to a heatsink, dunno really. It seems a wing and prayer part of the design. There are 125C electrolytics, uncommon and expensive. And yes, electrolytics are notoriously troublesome, but we all know that.

[tautech]

Axial tantalum capacitors are a completely different beast; and akin to an aluminium electrolytic.

Yup, I think they're used in place of electrolytic capacitors because they can tolerate high temperatures for long periods of time. You wouldn't normally stick capacitors that can dry out under the metal wings of a heatsink, but HP buried a couple of low value sealed axial tantalums there. It's a very dense design that runs quite hot. Oh gaaaaaah, it just restarted and the scope didn't trigger.

I have some nagging thoughts you may have some degraded caps and maybe design shortcomings that don't adequately handle rail current spike loads where some designs utilize remote bulk capacitance some distance from the PSU.....it's just a thought that adding 100+ uF somewhere and maybe close to the IC that does the reset might easily solve your problems.

[synx508]

I don't think adding capacitance will help, there's no IC resetting as such - the reset happens because there's a power good monitor checking the +15V line and that will turn off the 5.1V supply if the voltage drops more than a volt or so. And it is dropping more than a volt or so, around 3.5V, for 30ms, once, maybe an hour or two from cold. Then maybe 3 hours later, maybe not. That extra load might be caused by some sort of unwanted oscillation, but it's not on the +15V rail - during the event there's not a hint of noise on the rail, it just politely ramps down, holds steady, then back up again.
Nothing does that, does it? That's why I must know what the cause is.
I've already replaced the key capacitors near and within the supply, then swapped them back because it made no difference.

One thought i had earlier was that I should replace +15V from the internal supply with +15V at up to 15A and then when the component that is inducing the extra load does so, it'll see what happens when the supply doesn't sag downwards… of course it'd end up being the unobtainium hybrid triangle to sine converter that is currently not very well bonded to the shield can of the A1 Generator board with a rough looking thermal pad and that would be "game over".

My next move is likely to be changing the axial tantalum capacitors on the A1 board, but I don't really want to.

[tggzzz]

Wouldn't it be nice to know whether the ch1 droop was due to a missing input cycle or an extra load on the PSU output.

Key features of the ch1 droop are the voltages, the linear fall and rise tends to imply constant current, and the 10ms falltime and 7ms risetime.

If you used a BJT plus resistor across the PSU line and turned the BJT on/off, the resistor would act as a roughly linear current sink over the 15V-11V range. Just observing the rise/fall time ratio would give a clue about whether such an extra load current was the issue. Tweaking the resistor value would give a clue about the magnitude of the current.

That technique could also be used on the PSU input, to see whether a missing cycle might cause such an effect.

[EC8010]

I certainly wouldn't touch bargain bags of ancient looking bead tants from a swapmeet.

Oh, I would. And have (at the Weston super Mare bash). Tantalum bead capacitors last very well just left on the shelf; it's when you start using them that you can get into trouble. But they were often used inappropriately in the 70s and 80s. We had some Link 110 television cameras (as used for "The Good Life") that had tantalum bead capacitors for decoupling power supply rails. They also had switch mode power supplies and the combination of high frequency noise and essentially zero ripple current rating periodically resulted in bangs and little brown holes in the side of yellow components.

[synx508]

Wouldn't it be nice to know whether the ch1 droop was due to a missing input cycle or an extra load on the PSU output.

Key features of the ch1 droop are the voltages, the linear fall and rise tends to imply constant current, and the 10ms falltime and 7ms risetime.

If you used a BJT plus resistor across the PSU line and turned the BJT on/off, the resistor would act as a roughly linear current sink over the 15V-11V range. Just observing the rise/fall time ratio would give a clue about whether such an extra load current was the issue. Tweaking the resistor value would give a clue about the magnitude of the current.

That technique could also be used on the PSU input, to see whether a missing cycle might cause such an effect.

I've breadboarded one regulator from the PSU and have been poking at it with loads to see if I can make it do the ramp, I've just been picking resistors but thinking about using a current sink. This started as a test of the 1458 that I'd extracted and escalated while I waited for a glitch. I have a programmable load that I made for a project years ago but I've mislaid the software for controlling it (it has a REST API over wifi because that was the least effort with one of those ESP8266 boards), but I might try to get that working.

Last night I ran out of time before I got a (third in one day) glitch. This morning the equally hard to find connector related issue was back but after reseating it the anomalous "E51 output overvoltage" error had gone. This is, as far as I can tell, a separate issue.