Fluke 5220A bursts into song and even flames.

[bdunham7]

Well, smoke anyway.  My Fluke 5220A current calibrator (transimpedance conductance amplifier) has some issues and I haven't managed to figure them out.  I acquired the unit damaged, it had been dropped on the right front top corner and is a bit bent with a broken handle.  It didn't look too bad and when I tested it, the unit functioned and I set it aside until now.  The problems seem to be evolving--the first thing I noticed was that DC worked fine but AC didn't.  Then I realized that only positive DC was working properly, negative DC didn't work well at all  After taking it apart, pulling the boards in and out, soldering on some test leads to the output board and maybe some other things I've forgotten, it now basically functions on both postive and negative DC but with some issues.  It also works somewhat on AC, again with issues.

The main issue is that it oscillates under certain conditions and it has taken me some time figure out what those are.  When you short the output at the terminals, the unit appears to function pretty much normally.  If I short it with a 24" banana lead, it may oscillate and I can change it by moving the lead around.  If I connect it to a bench meter with 36" leads, it goes into oscillation all the time.  It oscillates right around 700kHz and this is reasonably consistent.  Clearly it is reacting to either the additional resistance, inductance or both.  I've run into this with audio amplifiers where the driver transistors have been replaced.  The 5220A is mainly just a big Class AB amplifier with a current shunt feedback. 

I'm out of ideas at this point.  I've tested all the relevant capacitors, I've tried adding capacitance in a few places as I might with an audio amp, with no real success.  Once the oscillation starts, it sometimes continues even if you remove the input signal and I can measure the oscillations all the way back to the input terminals.  The specified BW of the unit is only 5kHz, so 700kHz is quite an achievement.  It seems to be putting out a frightening amount of current at 700kHz and at one point the driver transistor resistors were smoking.

I've attached relevant schematics and some scope pictures if anyone wants to play along.  The complete manual is at xDevs.com.

https://xdevs.com/doc/Fluke/FLUKE%205220A%20Instruction.pdf

[Kleinstein]

The amplifier looks like a class B, as there as it looks like the standing current would not be stable.  The circuit spread out over multiple pages makes the plan a bit confusing.
With an old instrument a point to look at could the supply decoupling. Some amplifiers don't like it when the ESR of the supply capacitors goes too hight.


When driving a current, the critical load is an inductive load.  One could get away with some RC element (e.g. a few ohms with low inductance and maybe 100 nF)  in parallel to the "load" / terminals.

[bdunham7]

The amplifier looks like a class B, as there as it looks like the standing current would not be stable.  The circuit spread out over multiple pages makes the plan a bit confusing.
With an old instrument a point to look at could the supply decoupling. Some amplifiers don't like it when the ESR of the supply capacitors goes too hight.

The amplifier layout is a bit confusing and the circuit theory section of the manual is not as helpful as I'd hope.  They claim that the ouput A7 is driven differentially by A6, yet it has a fairly typical arrangement where the center (grounded to common #2, see flag 6 between the two boards) is connected to the drivers by the 2.4R resistors and then each driver is driven by the same signal with an offset determined by the bias circuit.  It has to be Class AB or something like it--they have a spec of 0.04% distortion which would be impossible otherwise.  I confirmed that it seems to operate in an AB fashion by looking at V_BE for Q102.  I just used the scope, so no real precision, but it was about 0.7V at zero output, then went up very slightly as the current was increased in its direction (negative) and then went down to 0.3V (turned off) when the current went the other direction (positive).  So AFAIK, both drivers are on at rest, which is what I'd expect.  The bias and DC offset adjustments do exactly what you'd expect as well. 

Adding to the difficulty in understanding the operation of it, the preamp and analog parts are driven by a +/-15V regulated supply with one common (COM1) and the output is driven by the large +/-12V unregulated supply with a different common (COM2) and the shunt is connected across both commons (I'm sure there's a reason, I just don't see it yet) so that the COM2 of the output section is also the shunt feedback signal.  And then that feedback signal goes to both the A6 driver board and the A5 preamp board, both of which seem to use it as negative feedback.  That makes two separate feedback loops unless I'm misunderstanding something. 

I've looked at the output of the +/- 12V supply and it has the typical sawtooth plus image of the output that you'd expect.  It has a total of 8 50mF blue Sprague capacitors, so a total of 0.2F per side.  The problem occurs at low currents as well as high, so they'd have to be pretty degraded to cause an issue at 1A of output.  The 12 volt rails do droop quite a bit under heavy load.

When driving a current, the critical load is an inductive load.  One could get away with some RC element (e.g. a few ohms with low inductance and maybe 100 nF)  in parallel to the "load" / terminals.

I actually "fixed" this last night, although it isn't a permanent fix.  I attached a 3.3µF film motor run capacitor directly across the outputs and now it is able to drive my 48" leads and bench meter cleanly and accurately. 

I measured the impedance of my test leads and DMM with an LCR meter, just unplugging the leads from the calibrator and testing them right in the same position.  At 1kHz, I got Z = .097R with Θ = 5.7^o, or 1.57µH + 0.88R.  With the leads and the 3.3µF capacitor installed and the scope across the output, at 1A 1kHz I get 280mV_p-p on the scope.  I was able to go through the calibration routine on the calibrator and once adjusted it is stable and accurate.  So that all works out and makes sense, but I haven't gotten to the root of the problem. 

The manual says the calibrator is rated to drive up to a 200µH load but mentions that this might become an issue at higher frequencies and step transitions.  I'm so far below that at 1.57µH that there still must be an issue somewhere in there.  Even if the calibrator were challenged by an inductive load, I would think it would just trip off due to the compliance voltage being over the limit or else put out a distorted or inaccurate output.  It shouldn't react by oscillating at 700kHz.

[Fried Chicken]

Certainly if there was smoke there should be something obvious?

Carbon composition resistors have a tendency to go bad afaik... and I think their failure is in a destructive direction?  Their resistance goes down?

[bdunham7]

Certainly if there was smoke there should be something obvious?

Carbon composition resistors have a tendency to go bad afaik... and I think their failure is in a destructive direction?  Their resistance goes down?

Yes, the smoke was from two carbon comp resistors in the driver circuit.  They're 2.4R, 2W and despite being charred, they still measure about 2.5R and the unit works.  I'll replace them when I order parts after I figure out the root of the problem.  The good news is that it works well enough for now to complete the job on your meter.

[Fried Chicken]

Certainly if there was smoke there should be something obvious?

Carbon composition resistors have a tendency to go bad afaik... and I think their failure is in a destructive direction?  Their resistance goes down?

Yes, the smoke was from two carbon comp resistors in the driver circuit.  They're 2.4R, 2W and despite being charred, they still measure about 2.5R and the unit works.  I'll replace them when I order parts after I figure out the root of the problem.  The good news is that it works well enough for now to complete the job on your meter.

Woah.... don't push it!  I wouldn't power on a meter if I were suspect of it like that.... of course I want my meter calibrated, but don't fry your equipment in the process!  Do it right!

[bdunham7]

Woah.... don't push it!  I wouldn't power on a meter if I were suspect of it like that.... of course I want my meter calibrated, but don't fry your equipment in the process!  Do it right!

No worries--those big-box Fluke calibrators are pretty much burning something inside whenever they are on.  Toasted parts, shorted tantalums and burned PCB traces are common.  Those carbon comp resistors can sure take some punishment and I'll probably burn more stuff up trying to figure this issue out.  My 30A fuses sure lit up when whatever happened happened. 

[bdunham7]

Today I tore apart the output assembly, took out the driver and output transistors and tested them on my small curve tracer.  The big curve tracer awaits restoration, so 50mA is all we have.  Results are as shown.  Q103/104 are 2N5883, Q105/106 are 2N5885.  The drivers are Q101-D45C5 and Q102-D44C5.  Comments, suggestions, ideas or magic fixes all welcome.

[Kleinstein]

Q105 and Q106 are quite a bit different. It may be worth checking (at TP5 to TP6) if the current sharing is still working OK.
The resistors for current sharing are quite small and I would expect that this may need well matched transistors.  The use of 4 wire resistors is a bit odd: the parallel connection partially defeats the 4 wire connection.

With the transistors out, one could also check if the 2 are also similar in VBE and maybe also the speed.

Are the transistor on the PCB or mounted separately on a heat sink. If extra wires are used, the position / spacing of the wires can make a difference from parasitic inductance.

With the transistors out, one could also check R108, R109 in circuit. Drifting resistor values could change the gain / speed of the the stages.

[bdunham7]

Q105 and Q106 are quite a bit different. It may be worth checking (at TP5 to TP6) if the current sharing is still working OK.
The resistors for current sharing are quite small and I would expect that this may need well matched transistors.  The use of 4 wire resistors is a bit odd: the parallel connection partially defeats the 4 wire connection.

With the transistors out, one could also check if the 2 are also similar in VBE and maybe also the speed.

Are the transistor on the PCB or mounted separately on a heat sink. If extra wires are used, the position / spacing of the wires can make a difference from parasitic inductance.

With the transistors out, one could also check R108, R109 in circuit. Drifting resistor values could change the gain / speed of the the stages.

The calibrator is fixed, unfortunately mostly due to serendipity not hard work or brilliance on my part.  But the wild goose chase was educational at least.  Here's what I did:  (TLDR:the outputs and drivers aren't the problem)

I still had the transistors out, so I checked V_BE with a 1mA test current and got 0.48V for Q103/104 and 0.53V for Q105/106.  I tested R108/109, they were a bit high at 5.6 and 5.9R, but I let that go.  I tested most of the resistors on the driver board and they were all OK.  I very carefully tested R111/112/113/114 (these are loops of wire that extend out into the heat sink tunnel) and they were 87.4mR +/-1mR the long way and about 1.1mR between the Kelvin terminals.  Regarding that, the circuit isn't using Kelvin sensing but I think it is set up so that the voltage supply to the driver board stays close to the voltage at the driver and output transistors, probably to improve PSRR.  The outputs and drivers are both on a large heat sink and also directly on a PCB via sockets, a picture is attached.  It's really a neat, compact arrangement. 

Thinking that I was going to quickly prove that the transistors were not behaving and already shopping for replacements, I set out to measure the currents using the accurate measurements of the emitter resistors and some extra test leads that I soldered in.  I tested at standby, active with zero output, 1A, 2A, 5A and 10A, positive for R110/111 - Q103/104 and negative for R112/113-Q105/106.  Here's a table, hopefully the formatting works:

CURRENT    R110/Q103       R111/Q104        R112/Q105     R113/Q106

STBY           -0.5mV               -0.5mV            +1.0mV           +1.0mV

0A                50mV*               50mV*             -80mV*           -40mV*

1A                90mV*               90mV*             -110mV*         -80mV*

2A               0.125V               0.121V              -0.136V           -0.117V

5A               0.219V               0.225V              -0.230V           -0.218V

10A             0.435V               0.448V              -0.455V           -0.435V

I_C@10A (both)        10.096A                                      10.187A
I_B@10A                   0.096A                                        0.187A

GAIN (combined)   104                                            53

So the first lesson is that curve tracing at voltages and currents far from the actual use don't indicate much!  This might seem obvious in hindsight if you look at the datasheets.  The second is that emitter resistors really do work for load sharing because very small changes in V_BE result in big current changes.  The curve tracing is probably more appropriate for the drivers since the 50mA max of the curve tracer is reasonably close to the currents in the actual circuit.  Those curve traces indicated that the gain of Q101 is about 40 and Q102 about 80, and that ratio almost exactly compensates for the difference in gain of the outputs.  I wonder if that is a coincidence or if this is someone's idea of driver matching. 

So, on to the anticlimactic solution.  I decided to follow the signal through and see where the phase shifts were occuring.  When the 700kHz+ oscillation was occurring, I found that signal had a greater than 90 degrees shift between TP4 (input) and TP2 on the driver board.  I decided to short the output to quell the oscillation and then give it higher frequency inputs and observe the phase shifts.  Unfortunately I didn't take any photos at this time, planning on getting some later.  I ran the frequency up to 50kHz and thought I could see a small shift but I couldn't go any higher because 50kHz is the max for the 5101B calibrator I was using.  I switched to using an AWG for the input and as soon as I connected it the DUT stopped working entirely, the LEDs went off and there was no logic control.  Well, the 5100B has an isolated output, but the AWG has the negative grounded.  It turns out that there was some physical damage to the control panel that essentially shorted a 5V supply to the common of the analog preamp (COM1) and this must have provided some sort path for feedback from the output.  I fixed that and the problem went away--it now can drive a load through 20 feet of leads and three meters if I like.  Now as I raise the frequency, there is a phase shift as the frequency goes up but the strength of the signal drops into the noise long before the phase difference nears 90 degrees.  I can give it 5V @ 700kHz input and it essentially does nothing.

I then reassembled it, calibrated it and finished up with Fried Chicken's meter.   A happy ending all around.