HP 34401a DMM with leaking segments

[qu1ck]

Hi all,

I recently became a happy owner of this used but in good condition DMM (see first pic) which seems to have only one issue that I could find so far: display shows dots where there shouldn't be any and some symbols (4w, buzzer) show up as well.

It is old (cal string says "feb 1996", revision 04-01-01) but perfectly clean on the inside and other than the display everything seems to be in good shape.

So I started searching the forum for similar symptoms and found https://www.eevblog.com/forum/buysellwanted/ebay-many-faulty-hpagilent-34401a-multimers/msg811170/#msg811170

It talks about leaky drivers with NEC part that are used in newer front panels but mine is the old one with unobtanium UPD7527. Is this issue relevant to old style panels too?

Either way I decided to take it apart and look at what the driver is sending to display. Like I imagined the dots and symbols are segment 14 and 15 and they are somehow linked together. See the scope screenshots. Both have the following setup: ch1 - chr11, ch2 - seg14, ch3 - seg15.
Whenever segment 14 is high (+18v) segment 15 goes to around ground, although when 14 is low (-18v), 15 can be driven both high and low successfully.
Btw I did not figure out which segment is 14, it seems to be the lower horizontal one (think underscore).

So here is what I am thinking.
1. Somehow "help" the driver with some outside circuit that will pull pin to -18v when driver outputs anything less than say 5v.
2. Much more involved but also should be great help to others: reverse engineer the protocol between main board and front panel, reimplement it on another MCU and output the display on a nice big OLED screen like this guy did for his 34970a:
https://www.eevblog.com/forum/repair/reverse-ingeeniring-hp-34xxx-display-panel-serial-protocol/

With multi layer board doing 1 may not be possible without splicing pins of the display itself. There may be other issues as well, I have not thought about this too much. Your ideas are welcome.

For option 2 looking at the schematic suggests that likely SPI is used in 34401 instead of usart as in 34970a so it can be completely different in terms of packet format, acks, etc. But should be still doable if it's SPI, would be much harder if it's something else. Protocol is referred to as "4 wire serial" in the service manual but the pins on UPD7527 are marked SCK, SO, SI, what else could it be? I'm not sure what is the purpose of INT(errupt?) pin though. Will have to take a scope to it under different modes and see whats going on there.

If I succeed than the display can be swapped for a 3.2" OLED with a blue pill on a side. Cheaper and more importantly much more accessible than replacement VFD display with UPD7527.

Thoughts?

UPD: segment 14 must be upper-left vertical segment. Deduced from both "V" and "4" having the dot next to it, that's the only segment they have in common.

[DC1MC]

I've never had the opportunity to have such a nice instrument  , but if the display driver exhibits this behavior I think that a quad op amp, put in the most basic comparator mode, should be able to do the trick as long as there is still a way to analogically differentiate between signal levels of the driver output.
That is put an LM614 or similar (it has to accept >= than +/- 20V as supply, and be quad or more)  to the most simple comparator mode, that is the [- Input] to ground or some reference voltage  ( an 10k potentiometer also connected to the voltage supplies) and the [+ Input] to the driver output, the op amp [Output], of course to the display segment. Basically this is it, the not so rich man adjustable signal signal rejuvenator .

I have no idea what is the logical interface to the display board (parallel input, SPI-like sync or actual async serial) but if you can hook a cheap LA to it and post some traces, I promise to have a look and try to decode the protocol, maybe the extra karma and sympathy bonus will allow me to get a similar device sometime.

 Cheers,
 DC1MC
 

[qu1ck]

I will definitely hook it up to a logic analyzer but first I need to find any dip comparator  Everything I got is smt. Not ideal for experiments.

Interestingly the old front panel works with +13 to +18v digital logic levels while new one is on the low side -18v to -13v. The schematic shows that data out from front panel goes straight to an ASIC pin, with a pull up to +5v. It forms a voltage divider with a series resistor that is present in new panel so that logical 0 is around 0v on the ASIC instead of -18v. But on old panel there is no series resistor, are they sending +18v straight into logic board?  I'll have to measure that tomorrow, 3:30am right now, too tired to think

I'll come back with some experiment results soon.

[Samogon]

 
https://www.eevblog.com/forum/buysellwanted/ebay-many-faulty-hpagilent-34401a-multimers/msg811170/#msg811170

[bitseeker]

@Samogon: The OP already referred to that link near the beginning of his post.

@qu1ck: Interpreting the front panel communication with the motherboard in order to completely replace the old display would be interesting and useful for keeping alive the many old 34401A devices out there, if you're inclined to dive into that endeavor.

I had heard that some of the old driver ICs were out in the world, but I had never located any.

[Samogon]

I apologize

[floobydust]

Anyone know the VFD size? Or the front PCB dimensions (width)?

Seeing what would fit....
For glass (no frame) OLED's standard 256x64 size:
3.12" are 88mm W x 28mm H
5.5"  are 146mm W x 45mm H (but expensive ~$100)

Some mentioned if the main CPU firmware is changed, a second-generation front PCB works.
The old uPD7527 seems to use a different com protocol than the 80C51.

[bitseeker]

The tinted lens in front of the VFD is about 153 x 36mm. The top 10mm or so is occupied by the model badging, so a display around 120 x 25mm would probably be OK. That 88 x 28 might work.

[qu1ck]

I've got good news and bad news.

Bad news is that it's not the driver. Since I didn't have suitable comparator at hand I just quickly put together a simple signal repeater using 2 bjt's. It only outputs high when input is 20v or so above negative rail.


I tested it and it worked perfectly and produced expected square wave with no load.
So I cut the pin of the display, hooked my contraption in between the pin and it's pad and what do you know



No difference, zip, nada. Now the input was a clean square but output was the same mess, which means driver was doing it's best to begin with, it's something in the display pulling the pin up. Maybe stronger pull down resistor or push-pull output would have rectified it but at what cost? Increased current, more segments burned? I didn't want to risk it. So I soldered the pin back to it's place and switched to plan B, which brings me to the good news.

Protocol is definitely SPI:


I captured 4 seconds of traffic and with some bash-fu extracted ascii data:

Code: [Select]

$ sigrok-cli -i capture.sr -P "spi:mosi=SI:clk=SCK:miso=SO" -A spi=mosi-data | xxd -r -p
  000027 mVDC       ☺    000027 mVDC       ☺    000026 mVDC       ☺
Now we are getting somewhere

Sigrok capture file is attached for those playing along.

[floobydust]

The VFD drive for the older 34401a's is complicated to troubleshoot.

In the interest of science and keeping these workhorses out of landfill, I did analyze the circuit. It's kind of a bear because of PMOS and six different supplies.

It would be worthwhile to see if the VFD's grid-threshold voltage has just shifted due to aging, or the NEC uPD7527A 4-bit MCU HV output has gone leaky, keeping segments dimly on.

I would confirm:
VFD cathode (filament) -12V bias, from zener CR556/C559 at the Xfmer on the PSU.
VFD MCU IC pre-driver bias -10V, from zener CR600/C607.

Keep in mind you can get grid current flowing, it's not high impedance if the grid is +ve it acts like an anode.

I would be tempted to first vary the -12V cathode bias, carefully (use floating psu with zener removed) and see if the -12V/-18V = -6V cutoff is enough.
My present theory is the tube ages outside the circuit's cutoff voltages and segments stay dimly lit. Other threads said changing the VFD cured the problem, so this is the only explanation I could come up with.

[bitseeker]

@qu1ck: Good to see that it's SPI. Substituting the VFD is still an interesting solution in the long run.

@floobydust: Interesting. I hadn't seen before that replacing the VFD fixed the phantom dots.

[floobydust]

This thread, member says it's the display: https://www.eevblog.com/forum/testgear/hp34401a-display-with-dots-after-every-digit/msg1228322/#msg1228322
This thread, member says it's the driver MCU: https://www.eevblog.com/forum/buysellwanted/ebay-many-faulty-hpagilent-34401a-multimers/msg811170/#msg811170


I was looking at what's involved making an OLED conversion, as this big VFD and driver scheme was used (at one time) across the HP/Agilent product line and is going obsolete for parts.

There are a few flavours of VFD's with special annunciators being the difference.
Universal counters: HP-58503, 53131A, 53132A, 53181A
Power Supplies: E3631A, E3649A, E3642A, E3643A etc.

It looks not bad to convert, but a big OLED display is as expensive as a replacement VFD ~$60 so I didn't think there would be interest.

[qu1ck]

@floobydust
In my case it's 100% the display. When I disconnected that segment in my experiment it was still lighting up, with the pin flapping in the breeze. And driver was sending perfect square wave +-18v as expected, very different from what I was seeing on the scope before (see my first post). Connecting my bjt output stage which is similar to your schematic with fets did absolutely nothing.

You don't have to buy 5" oled, it won't even fit because it is higher. 3.2" oled is $30, add an mcu board and some mounting hardware, that's $35 tops. Half the price of a VFD, much more value in terms of longevity and repairability. I bet you can even make the digits the same size, just make the font more compact, VFD wastes a lot of space.

[DC1MC]

@qu1ck:

IMHO I will still try first an "analogue repair", the segment has to have a cut-off voltage somewhere and if there is a little drive current more, it wont hurt. But if you try to go on the OLED display route, the message structure could be simpler maybe, but is hard to think how. The values are in nice ASCII and the non-printable chars have most likely the bit fields with the status of different aux indicators around. I can't think of a controller that will not be able to decode this message. What is the clock frequency, I'm too lazy to launch Sigrok now ?

 Cheers,
 DC1MC
 

[qu1ck]

@DC1MC

The thing is to shut the segment off you just have to put it at low potential, there is no current. At least that is my understanding of how VFD works. Both grid and segment have to be at positive voltage relative to filament for segment to light up. If one or the other has low potential electrons won't flow to the segment.

Since in my display driving one segment high also induces high potential on another pin it leads me to believe there is some short in the tube itself. I don't want to force that segment low, that might increase the current significantly and those thin threads connecting the dots won't take much.

Maybe someone with old display that had similar issue will try it but I'm reluctant, at least until I have alternative solution in place.

SPI clock is around 93khz. Any controller will be able to decode, but you need something with at least full 8 pin port to quickly push image to the 8080 oled display. Blue pill stm32f103 is super cheap and has plenty of pins, I will likely use that one. Arduino pro mini should work as well and won't require level shifters.

[DC1MC]

No display has such a thin thread that will not take 1mA.

[floobydust]

If a row of segments stays lit with the VFD pin lifted, then there has to be some leakage current.
I don't know of any corrosion or ion migration mechanisms here, occurring inside a VFD. Only a gassy (some air ingress) tube shows leakage currents, the Getter can sometimes show that.

Instead, it might be the VFD stray capacitance to adjacent segment (internal) traces, that you are seeing with the pin lifted, enough to light a segment.

Normally the driver IC has weak current-sink ability to discharge the anode (capacitance and gas leakage current), and a small time delay with the mux driver between selecting digits, to prevent ghosting.

[HighVoltage]

You can test a VFD manually by applying power to the VFD and then apply a voltage to each segment and look what is lighting up.

[bitseeker]

I was looking at what's involved making an OLED conversion, as this big VFD and driver scheme was used (at one time) across the HP/Agilent product line and is going obsolete for parts.

...

It looks not bad to convert, but a big OLED display is as expensive as a replacement VFD ~$60 so I didn't think there would be interest.

In the long run, though, it may still be beneficial. As you said, with parts going obsolete, even if the price is the same for the VFD, it eventually may not be an option. On the other hand, with a custom front panel controller, one could use other available VFD, such as dot matrix, as well as other display technologies. Anyway, interesting stuff to explore.

[qu1ck]

Update:

With a strong pull down (2.7k) I got the ghost segments to finally disappear. But I did it on the disconnected pin, I don't want to try it with pin connected to driver as well, I'm not sure the driver will take it and I don't want to kill the front panel driver because it does input processing too.

On the other front:





It's just a proof of concept at this stage. It doesn't decode lower symbols, it skips bytes sometimes and shows gibberish, but it kinda works.

[floobydust]

Did you measure the filament bias voltage, or try changing it? VFD cathode (filament) -12V bias, from zener CR556/C559 at the Xfmer on the PSU.
I keep thinking this problem is due to the VFD aging and cutoff voltage shifted, making the driver look like the culprit- but it's not.

For substituting the VFD with an LCD or OLED display, I realize radiated EMI is a big concern. It would actually be the hardest part I think.
You have to keep things as quiet as possible. There are no buck converters here, and the metal shield around the VFD is there for good reason.

The display and main cpu do talk back and forth as a self-check, so that might be the occasional garbage you see.

[qu1ck]

Filament bias looked fine on the scope, waveform was completely within -18 and 0v, which is all that is needed, I think. Did not do any more precise measurements.
Problem is that for ghost segments there appears fuzzy 0v to +2v potential on the display pin from some internal short (as I deduced from that potential appearing even on unconnected pin). Can't fix that with bias unless you shift it much higher than it is intended to be which will reduce contrast.

LCD I'm currently using does not have any converters, just a linear reg. Also it has a shield.
For final version with OLED I will have to rip out the 12v boost converter and supply it directly with linear regulator as well. It should be even better EMI-wise because much smaller voltages are involved here instead of +/-18v swings of VFD.

Regarding garbage I'm sure it's my code, I just don't have the interrupts and timings right (I'm doing software SPI to read both input and output at the same time).
I already reverse engineered main portions of the protocol and have a good idea of packet format, there is no meta info exchanged between front panel and main cpu, only "display this text", "display these symbols" and "button/combination pressed".

I will post everything in detail as I do the rest of the work. I don't have as much time as I would like to tinker with this, so updates may be slow

[floobydust]

The 3.12" OLED displays I looked at are about 0.6W-0.7W; 100% pixels lit is max. 50mA at 12-15VDC, and 33mA with 50% pixels on. I don't know the 34401a power budget for how much current is drawn for the VFD and its MCU.

Because the +18V rail goes right to the A/D and analog sections (after 3.3V zener to make +15V) I would not power a display from that rail, in order to keep it super quiet.
It may be the multimeter's limiting factor- noise on the analog +ve rail from the display. Gen2 of the display board also added RFI filtering.
Gen2 display board, if the main MCU does not get a response from the display MCU, the main MCU will reset the display MCU by asserting its reset, as I recall.

I hope the project bears fruit, the multimeters are commanding very high prices now on eBay.

[qu1ck]

We are only displaying some text on the OLED, it will never light up more than 30% of pixels. Power requirement is quite low, but I agree on the noise. I will look for a good place to tap power from.

I fixed the garbage glitch (turns out micros() in stm32duino is just not reliable) and reversed 99% of the protocol. I am not certain about purpose of couple control frames and characters but for all practical purposes I can show exact same information on the display and decode button presses. Unless there are some more special cases in the modes that are not accessible to me at the moment (remote control mainly) I think I am done with this part.
Now on to improving code to display all the annunciators properly.

[coromonadalix]

would be nice to have another type of display adapted  ...