Transmille 8081 ADC

[Kleinstein]

The TL431 should operate with some 1-2 mA. More is only produching extra heat. The AD8628 take up to 1 mA ( a little less typical, and not significant load), so something like 1 K from the 5 V supply or some 5.6 K maybe 4.7 K from the 15 V would be suitable.

The dividers for the lm311 trigger level don't need to be very stable - normal 100 ppm/K resistors are good enough. This also applies to the slope amplifier. The OPA209 is slight over-kill (but also low power) there, a NE5534 could be sufficient too (but needs a compensation cap).  Only the drift over the time of a single conversion would matter.  I would still connect pin3 of IC3 directly to ground and do the divider at pin 2. So the 680 ohms (maybe change to 1 K) and than 100 K to +15 V. With 1 K and 100 K on each side the trigger levels would be at some +-150 mV or some +-20 mV before the amplification. With a slope of some 3 mV (mainly from the forcing signal) per µs, this would be length of some 6 µs for the phase with no reference active. This should be long enough for settling at the integrator. Up to about twice the separation would be possible.
There is no need for a fine trim - if at all it would be more like exchanging the resistors in large steps (e.g. 1 K or 2.2 K). I would not expect much of a difference. A very large gap with IC3 essentially at 15 V looks more like a problem to me and the 100 K add quite some noise. I consider this more like an unnoticed mistake.

Which ground to use for the LM311 trigger levels and the slope amplifier is a good question. Chances are it should be the same ground.

[rigrunner]

The TL431 should operate with some 1-2 mA. More is only produching extra heat. The AD8628 take up to 1 mA ( a little less typical, and not significant load), so something like 1 K from the 5 V supply or some 5.6 K maybe 4.7 K from the 15 V would be suitable.

I've changed the TL431 resistors to 1k.

The dividers for the lm311 trigger level don't need to be very stable - normal 100 ppm/K resistors are good enough. This also applies to the slope amplifier. The OPA209 is slight over-kill (but also low power) there, a NE5534 could be sufficient too (but needs a compensation cap).  Only the drift over the time of a single conversion would matter.  I would still connect pin3 of IC3 directly to ground and do the divider at pin 2. So the 680 ohms (maybe change to 1 K) and than 100 K to +15 V. With 1 K and 100 K on each side the trigger levels would be at some +-150 mV or some +-20 mV before the amplification. With a slope of some 3 mV (mainly from the forcing signal) per µs, this would be length of some 6 µs for the phase with no reference active. This should be long enough for settling at the integrator. Up to about twice the separation would be possible.

Thanks for that reasoning. I've amended the divider accordingly.

I noticed that in the Solartron 7081 the negative rail for the high speed parts of the comparator and both LM311s are lifted by a 68 Ohm resistor.
Does that add anything useful to the circuit?

There is no need for a fine trim - if at all it would be more like exchanging the resistors in large steps (e.g. 1 K or 2.2 K). I would not expect much of a difference. A very large gap with IC3 essentially at 15 V looks more like a problem to me and the 100 K add quite some noise. I consider this more like an unnoticed mistake.

Which ground to use for the LM311 trigger levels and the slope amplifier is a good question. Chances are it should be the same ground.

Slope amplifier GND amended.

[Kleinstein]

The resistors ins the negative supply isolate the supply, like already done with the max312 switches and some OPs. I don't think the 68 Ohms in the Solatron are about reducing the supply voltage, more about reducing possible interference as a possible source of INL errors. It may be a bit overkill to have most parts separated, but with SMD parts this may not be such a bad idea, even if the resistors and capacitors are smaller. The nasty parts are the high frequencies (e.g. > 100 MHz) from fast swtiching that are hard to control. Here just a MLCC at every chip may not be enough, as there can be CLC resonances with the trace-inductance. Spooky RF is nothing I like in a precision circuit.

I would consider a comon pair of resistors for both LM311 could be enough, as they don't switch together. I would more consider them as source than targets of interference.

[rigrunner]

The resistors ins the negative supply isolate the supply, like already done with the max312 switches and some OPs. I don't think the 68 Ohms in the Solatron are about reducing the supply voltage, more about reducing possible interference as a possible source of INL errors. It may be a bit overkill to have most parts separated, but with SMD parts this may not be such a bad idea, even if the resistors and capacitors are smaller. The nasty parts are the high frequencies (e.g. > 100 MHz) from fast swtiching that are hard to control. Here just a MLCC at every chip may not be enough, as there can be CLC resonances with the trace-inductance. Spooky RF is nothing I like in a precision circuit.

With everything except the LT5400, AD8628 and OPA209 I can go with through hole parts. The only thing I have seen producing any kind of high frequency noise so far is the PLL LM311 ringing a little at the 50Hz zero crossing. The noise seemed localised and I have that isolated from the supply too.

I would consider a comon pair of resistors for both LM311 could be enough, as they don't switch together. I would more consider them as source than targets of interference.

I think that's the new comparator mk1 circuit ready  Time to measure the existing mounts and connectors and see how layout goes.

The original PCB has the two metal cans over most of the components, along with that annoying potting compound 
Was there really a need to have two cans to separate the micros from the analogue side? I haven't noticed any ill effect running with the shield removed from the micro side of the board. Solartron didn't even bother with full shielding, just a copper clad cover over the ADC area.

[Kleinstein]

With the LM311 for the 50 Hz signal there should be no ringing at the output. The supply should have isolating resistors if needed. At least the extra signal would be mains synchronous and thus suppressed quite well.
The PLL should definitely have a clean supply.

The can for the µC is a little odd - there are plenty other µCs and possibly EMI sources in the DMM. Also with no realy filtering in the supply the cans are not very tight. As far as I can see the PLL was outside the cans.  So chances are the can around the µCs is not really needed. Improved decoupling / supply filtering is likely more effective.

The can around the integrator has a dual effect. One is a little EMI reduction - again without porper filtering the supply not very effective. The other is to improve thermal stability for the integrator and switches. I still don't think the can is really needed.  With an AZ OP the integrator would not be that sensitive. The more sensitive part would be the reference amplification.

For many parts I see no real advantages with THT parts.  SMD resistors are much easier to exchange if needed. 0805 size are still reasonable readable and easy to solder. The film capacitors kind of need to be THT. Combined THT and SMD can be a bit tricky: the THT parts can block the access to soldering the SMD parts. So it helps if the SMD parts are on the bottom.

[rigrunner]

With the LM311 for the 50 Hz signal there should be no ringing at the output. The supply should have isolating resistors if needed. At least the extra signal would be mains synchronous and thus suppressed quite well.
The PLL should definitely have a clean supply.

I saw ringing at the output on both the rise and fall of the square wave, and some noise on the ZC at the input.
Just checked again and it's quiet. I think I screen captured the earlier ringing. If I can locate the captured images I'll post them.

I experimented with isolating the PLL supply and lowering the voltage. With the I2C lines from the 18F252 now being higher they introduce significant noise to PLL as pulses on the supply. Is it OK inserted resistors into the I2C lines to mitigate this?  Not sure I want to lower the supply to the timing PIC.

The can for the µC is a little odd - there are plenty other µCs and possibly EMI sources in the DMM. Also with no realy filtering in the supply the cans are not very tight. As far as I can see the PLL was outside the cans.  So chances are the can around the µCs is not really needed. Improved decoupling / supply filtering is likely more effective.

The can around the integrator has a dual effect. One is a little EMI reduction - again without porper filtering the supply not very effective. The other is to improve thermal stability for the integrator and switches. I still don't think the can is really needed.  With an AZ OP the integrator would not be that sensitive. The more sensitive part would be the reference amplification.

I have my suspicions that the cans on the ADC has a lot to do with obfuscation.
I'll experiment and see what effect running without cans has.
The most significant EMI output is directly under the ADC PCB. The 10MHz clock that is used to output a PPS trigger for the main micro.
The oscillator itself isn't that bad, but on my revision the supply resistor (R119 10R) is cut and a pair of wires tun to a internal/external 10MHz switch. These wires hoot quite a signal. I'm thinking of switching the oscillator supply with a FET rather than have the supply running to the switch and back.

It's a same the 10MHz is on the unguarded size of DMM. It could be used to clock everything on the ADC otherwise.

For many parts I see no real advantages with THT parts.  SMD resistors are much easier to exchange if needed. 0805 size are still reasonable readable and easy to solder. The film capacitors kind of need to be THT. Combined THT and SMD can be a bit tricky: the THT parts can block the access to soldering the SMD parts. So it helps if the SMD parts are on the bottom.

Mounting the SMD parts on the underside of this PCB isn't something I'd considered. I'll bear it in mind though whilst I'm laying it out.

[Kleinstein]

I2C uses open drain/collector outputs and pull-up resistors. It would be OK to lower the voltage with an additional resistor (likely 20-50 K range, depends on the pull ups installed) from the signals to ground. Something like 3.5 - 4 V should be OK for the PIC and other 5 V chips to be safely recognized as high.

A 10 MHz signal running all over with lose wires sounds like an EMI nightmare. I would still be more worried about the supply / ground, not so much simple capacitive coupling. The supply still should be clean - so local decouling and a resistor in the wire to the switch.

If the 50 Hz signal is now clean, all is fine. It could have been a poor ground / supply.

A agree, obfuscation could be a reason for the can and potting. The 100 K at pin3 of IC3 need a kind of cover up.

[rigrunner]

I2C uses open drain/collector outputs and pull-up resistors. It would be OK to lower the voltage with an additional resistor (likely 20-50 K range, depends on the pull ups installed) from the signals to ground. Something like 3.5 - 4 V should be OK for the PIC and other 5 V chips to be safely recognized as high.

I'll test with a few values to see how it runs.

A 10 MHz signal running all over with lose wires sounds like an EMI nightmare. I would still be more worried about the supply / ground, not so much simple capacitive coupling. The supply still should be clean - so local decouling and a resistor in the wire to the switch.

It's quite a sloppy implementation. The wires to the switch aren't shielded, nor are they twisted.
I forgot to add the image to my previous post.

Red and Blue wires are supply to the oscillator via the switch.
Coax is 10MHz in from rear.
Black and Red are the reference in to the ADC - they don't usually dress in this place, but they do run close by.

A agree, obfuscation could be a reason for the can and potting. The 100 K at pin3 of IC3 need a kind of cover up.

  very true.

[rigrunner]

The 100 K at pin3 of IC3 need a kind of cover up.

It would seem the joke is on me 

Pulling R5 to test with a different value revealed a joining trace to IC4_divider_GND. The 100K resistor doesn't look silly now.  It's used to offset the negative voltage on IC4_divider_GND from R5+R11.

[rigrunner]

I was asked if I did re-spin the ADC board in the 8081?

It is still a work in progress but I did indeed create a new PCB and it does work, mostly...

With the main clock at 10MHz all the measurements are x1.6. I'll need to do something with the firmware in the 4620 pic to fix that.
The CS2000 is now outputting 12.5MHz rather than 6.25MHz, so although I have jumpers in place to swap between clocks that's broken.
The pics are using the original firmware so I don't see why the CS2000 is being programmed differently. Something else to sort out.

Only one bodge wire needed to get the new board running. I wired the negative supply on the 50Hz LM311 to 0V when it should have gone to -5V.

To try and resolve the oscillations I had I also made a daughter board for the +/- 10V refs. Board works perfectly wired up outside the 8081, but oscillates when installed. 

Picture of the new ADC board attached (jumper wire over slope amp for testing).