I'm looking into doing something I would call a guarded resistance measurement as it can measure a resistance without disconnecting it from the rest of the circuit.

Would anyone have an idea of a general name for this technique? I'm sure I am not the first to do this and wish to see some other information.

This technique isolates R1 from the rest of the circuit since the amp forces the current in R2 to be zero. Thus the ohmmeter only sees R1. Do note the amp does supply current to the circuit, that is returned thru the "ground" connection not thru R1.

 

I'm looking into doing something I would call a guarded resistance measurement as it can measure a resistance without disconnecting it from the rest of the circuit.

Would anyone have an idea of a general name for this technique? I'm sure I am not the first to do this and wish to see some other information.

This technique isolates R1 from the rest of the circuit since the amp forces the current in R2 to be zero. Thus the ohmmeter only sees R1. Do note the amp does supply current to the circuit, that is returned thru the "ground" connection not thru R1.

Quick, delete this image and file a patent!

Oops, too late, I already copied it.

Anyhow, very clever. Have you done this already or just a paper exercise so far?

 

Since I worked this out from something I remember 20 or so years ago I doubt there is anything patentable here.

I have several VME backplanes to either rework or pronounce dead. The common symptom is the termination voltage is off, and the resistance is 102 instead of 98 ohms. The termination resistors look like R1 and R2 above, but there are over 200 pairs in the system.

I'm just looking for a bit more information, mostly something practicable from someone who already did this.

For example, instead of building an amp with very low offset perhaps a voltmeter and a power supply would do: adjust the voltage A-C until the voltage A-B is zero, then read resistance B-C.

 

By rule, VME terminating should be the thevenin equivalent of 330 ohms to Vcc and 470 ohms to GND. It sounds like you have a passive-terminated backplane, as opposed to active termination where the network is a single 194 ohm resistor to a 2.94 V source. So I'm very curious about your resistor values. Can you share the backplane manufacturer, slot count, etc. Photo?

Also, is this for work, or something much more fun like a massive mad-scientist home computer project?

ak

 

There's a 5 terminal technique too. See https://www.google.com/url?sa=t&rct...qJ5eD3EJR2N9_RVA&sig2=ZgLciY8znj3WI8gmrRKCTQV for a White paper.

I'd have to dig for the 5 terminal version. it's in one of Keithley's manuals.

 

Nice find. The device in figure 1 is a typical VME (and others) passive terminating resistor network.

ak

 

I'm looking into doing something I would call a guarded resistance measurement as it can measure a resistance without disconnecting it from the rest of the circuit.

Would anyone have an idea of a general name for this technique? I'm sure I am not the first to do this and wish to see some other information.

This technique isolates R1 from the rest of the circuit since the amp forces the current in R2 to be zero. Thus the ohmmeter only sees R1. Do note the amp does supply current to the circuit, that is returned thru the "ground" connection not thru R1.

This looks like it only works in a very special case sincefor one things, it requires that the resistance you want to measure be in series with another accessible resistor in the circuit.

 

Excellent find KISS, nice to have a backup. With they had an app circuit. <g>

This is a VME chassis of origin I cannot state as It is work related and I try to keep that private. Yes the resistors are 330 and 470 (not 440) between ground and 5V. That is a thev resistance of 189 but there are two networks, one on each end of the bus, so a nominal reading at each pin should be 94.3 ohms and 2.94 volts.

A good line reads some 99.1 ohms and 3.06 volts. The issue is many lines read just 103.8 ohms and 2.17 volts. All networks get the same voltage, and there does not appear to be any bus line shorts to any other pin on the backplane busses on bad lines.

I might believe a resistor or two may go bad over time, but here there are so many out (over 50 lines on one unit) bad in the same way (same bad voltage and resistance readings) it screams for a root cause determination that so far escapes me.

Note: edited to correct 470 ohm resistor.

 

1. Are these backplanes VME, VME64, or VME64X?

2. I assume you meant 470, not 440 ohms. The Thevenin equivalent is 193.875 ohms. This is exactly half-way between two standard 1% resistor values, but some backplane manufacturers have custom resistor networks built with 194 ohm resistors for active termination circuits. Much lower static current, less complex routing, etc.

3. A perfect line with no VME cards installed should read 96.9375 ohms and 2.9375 V. 3.06 V is a 4.2% error, which is within the 5% tolerance of discrete terminators but outside the tolerance of 2% resistor networks. BTW, are your terminating resistors discrete or networks?

4. Since the Thevenin equivalent source is well known, try this: +5 V power to the backplane, no VME or other cards installed. At one of the VME J1 connectors, install a 97.6 or 100 ohm resistor from a known-good line pin to a GND pin. 1/4W resistor leads should push into the IEC connector socket easily. Voltage on the line should drop by 50%. Then install the resistor between the same signal line and a Vcc pin. Line voltage should increase to ((5.0 - Vopen) / 2) + Vopen. Now that you have benchmarks for a good line on that particular backplane, do the same tests on a suspect line and compare.

If you really have 50 bad lines on one backplane, that has to be either a soldering problem across multiple resistor network Vcc or GND pins, or a problem where Vcc enters the plane that powers the networks. How sure are you that Vcc is 5.0 V at the terminator network pins, at the bottom row of the J1 connectors, etc?

ak

 

A good line reads some 99.1 ohms and 3.06 volts. The issue is many lines read just 103.8 ohms and 2.17 volts. All networks get the same voltage, and there does not appear to be any bus line shorts to any other pin on the backplane busses on bad lines.

Any possibility it's contamination of some sort. Corrosion around the terminators? What's the calculated wattage for the resistors? Is the "package" wattage exceeded? Has cooling been compromised? Power supply ripple?

Just some ideas.

 

Let's see if I can answer from my device.

VME64X backplane. SIP resistor networks. Solder good,no corrosion on a conformal coated multilayer board. PS stable but hot at 5.2 volts, which explains the hi reading on good lines.

Tried the 100 ohm load which dropped the bad lines from 2.16 to 1.07 so the thev resistance on bad lines seems to really be what is measured.

Thanks for the help but still at a stone wall till I guess I build up something to take a 6 wire measurement.

 

Is this a new backplane or one that failed? If it is a new one then maybe some of the SIP networks are the wrong part or installed backwards.

 

I think they all are failed units that once worked OK. And the SIP resistor networks date them.

ak

 

Got a guarded measurement (or 6-wire) set up, and it became instantly clear several SIP resistor networks were installed backwards. That explains how 50 of 110 lines fail with identical symptoms.

Units are all returns but I can see how they "escaped."

I now have a root cause and a corrective action, just lack a complete rework procedure. These networks are near impossible to remove the way they are wedged in between connectors, so boards may be scrap, but I know why and can keep it from repeating.

Here's my measurement setup:

The ohmmeter is there to make the resistance measurement, of course.

The voltmeter indicates the current flowing into the rest of the network. One will adjust the power supply to obtain zero volts here before recording the reading of the ohmmeter.

The power supply (a bench lab supply with course and fine voltage adjustments) runs the rest of the networks to keep R2 biased off. Rx is there to decrease the sensitivity to the power supply: the ohmmeter only puts out perhaps 50 mV. By inserting the resistor the P/S needs drive this too, increasing the voltage it need emit. It is much easier to use both course and fine voltage adjustments to set the zero than just the fine knob.

In practice I found it best to leave the P/S disconnected (unplug the banana) till the ohmmeter stabilized on one reading, then turn the P/S on. Most times in this case once the supply was set it worked for other similar reading too.

 

Got a guarded measurement (or 6-wire) set up, and it became instantly clear several SIP resistor networks were installed backwards. That explains how 50 of 110 lines fail with identical symptoms.

Units are all returns but I can see how they "escaped."

I now have a root cause and a corrective action, just lack a complete rework procedure. These networks are near impossible to remove the way they are wedged in between connectors, so boards may be scrap, but I know why and can keep it from repeating.

Here's my measurement setup:
View attachment 102969
The ohmmeter is there to make the resistance measurement, of course.

The voltmeter indicates the current flowing into the rest of the network. One will adjust the power supply to obtain zero volts here before recording the reading of the ohmmeter.

The power supply (a bench lab supply with course and fine voltage adjustments) runs the rest of the networks to keep R2 biased off. Rx is there to decrease the sensitivity to the power supply: the ohmmeter only puts out perhaps 50 mV. By inserting the resistor the P/S needs drive this too, increasing the voltage it need emit. It is much easier to use both course and fine voltage adjustments to set the zero than just the fine knob.

In practice I found it best to leave the P/S disconnected (unplug the banana) till the ohmmeter stabilized on one reading, then turn the P/S on. Most times in this case once the supply was set it worked for other similar reading too.

It always amazed me how some brands of resistor networks lack markings to indicate Pin 1 (except part number printing which implies pin 1 is on the left). Even those brands that do mark Pin 1 with a "1" or black printed bar, they do not make it obvious once installed on a full board. Some color Change or dimple on one end would help a lot.

 

I agree 100%. In this case we have a black body with pin 1 marked on the side. We have the case where three parts are installed tight against each other with one tight against a backplane connector. At best you may be able to see one part marking, but it's 50/50 depending on which side it is marked.

Totally our fault in first installing incorrectly then not doing the test that would have raised the issue. (I can safely say that because that work was not done at my site <g> )

I am actually considering marking a plain white dot on the top of the part so it is damn obvious if they get put in backwards: as easy mistake to make and one so hard to correct for later.

Thanks to everyone who supported me on this. I'll be off to a meeting with most of the top level staff to make corrective actions.

 

Now I'm more curious. Are you working with new production backplanes that failed a system-level test, or returns? If returns, then the systems worked for a while? And then failed? curious...

And it's amazing to me that things I lived with every day can fade away. The low termination voltage should have been a giant red flag.

Three rows of SIPs, not socketed....Hybricon?

ak

 

Return units, unsure how long they have been out, doubt they ever worked "out there." These backplanes are our own custom design for a "ruggedized chassis." More I cannot say.

Since "rugged" no sockets allowed anywhere.

Our internal test uses a single board computer to run some perifial board, that combo works just fine despite the termination error.

I agree in hindsight the root cause should have been more obvious. I even missed the error was confined to certain SIPs.

Still, my intuition was screening for a simple common error, and I am happy I finally tracked it down 100%.