Programmable Electronic Load - Analyse the Summing Node Zero Point

Does it still affect the other ADC channels (going negative by only 5mV) or is it going to be usable in normal operating conditions?

You've still got the problem with abnormal conditions - ie if someone connects the load output the wrong way round. Perhaps the way to deal with that would be to make it a 'perfect diode' circuit, though that wouldn't give you any warning that the output was connected wrongly.

5mV error is a bit better than 200mV, so it's a definite improvement (worth having for two extra resistors). But it's only 40 times better and substituting an op-amp with bias currents maybe a thousand times less would give much more benefit.

If you take a look at the datasheet above, you'll see that the voltage offset error is typ 0.5mV and may be more. Amplified by the 6.8 gain of the differential amplifier, that gives the potential for several millivolts of error at the ADC input. Saying it another way, the reduced bias current error is now around the same level as the voltage offset errors, which means, if you want to go further and try and make it even more precise, you're trying to work with (and disentangle) two error sources rather than just the one.

Robert Peter Oakes  wrote:

 

I just noticed that you have a 680K going from the UC3 + to ground, this should not be populated and will throw things off in the real circuit when the common mode volts start being applied etc.

I've just put it in for the exercise in this thread. I can easily remove it again (happy that I didn't go for crazy 0402 parts on this PCB).

jc2048  wrote:

 

...

 

After Jan had done that (and assuming that it worked), he could then move on to the differential amplifier (U3C). To achieve the same effect there he merely needs to install the 680k to ground. Then we can see how close it brings things overall to the kind of accuracy you want. I suspect you'll want to move to a more up-to-date precision part (particularly if Jan is hoping to get anything close to the accuracy implied by working 16 bits), but let's see (and it's an interesting experiment to do, anyway).

With R32, 680K in place, the voltage at the entry of ADC1 is -0.005V as measured by my DMM,

The ADC measures -0.00581250 V (65506)

jc2048  wrote:

...

 

A third option, that a textbook would give you, is to balance the input resistances. It's something that Jan could try quickly with what he's already built, if he wanted to.

 

For the inverting amplifer U3B, lift the non-inverting pin and connect a 50k resistor from the pin to ground. (If you don't have a 50k 1% resistor, use two of the 100k ones in parallel.)  That gets you this

 

the output of U3B should now be much closer to the gain of -1 and with 0.108V going in should read within a millivolt or two of -0.108V. At least that's what theory says. Try it and see.

 

To understand why it works, look at the line above the one I highlighted on the datsheet (offset current). That's saying that, although the bias current is fairly high, the difference in the bias current of the two inputs is quite small (no more than one part in a hundred - less than 1%). That comes about because, although the bias current is the base drive to a transistor and is a bit variable  (it will vary from batch to batch), the transistors on a particular chip match very well in their characteristics (because they're all fabbed together). That matching can be made use of.

 

For this part, the bias current is out of the input pin (it's the base current of a PNP transistor). When it flows through whatever resistance is connected to the pin it generates a voltage. If the resistance seen by both inputs is the same, then those voltages will be the same (we don't know precisely what the voltage will be, because the bias current varies quite a lot, but we do know that both inputs will see a very similar additional voltage because the currents match well). Since the two inputs are difference inputs, the two voltages will then pretty much cancel out as part of the normal action of the op-amp.

 

...

I have added the 50K from non-inverting input to ground of U3B.

The voltage at that input is 0.051 V.

On the output, it's - 0.106 V.

Bodging is done, I can now test the first part:

For the inverting amplifer U3B, lift the non-inverting pin and connect a 50k resistor from the pin to ground. (If you don't have a 50k 1% resistor, use two of the 100k ones in parallel.)  That gets you this

 

the output of U3B should now be much closer to the gain of -1 and with 0.108V going in should read within a millivolt or two of -0.108V. At least that's what theory says. Try it and see.

Years of hoarding passives finally pays off:

started ... can you believe that I just have one 100K left.

Maybe it's time to buy one of those SMD resistor kits that you can backfill when stock runs out.

Tomorrow I'll check in my stack of discarded PCBs if I can recycle something close to 50K ...

note to self: lift pin 5

I'll try the 3rd option next weekend. I don't specifically need the precision but it's an interesting exercise. I think I'll pass on using a higher precision opamp for now. Time to move on to the high piwer part, then get operational firmware.

As you say, changing the op-amp is one possibility. There are bipolar op-amps with much lower bias currents, but an alternative nowadays is a precision CMOS op-amp (you'd have to check whether other characteristics were good enough).

Lowering the resistor values would be good anyway, but there's a limit to how far you can go and it still leaves quite an error. The measured value is way out and though you could try and compensate in software it's messy because it will vary from unit to unit.

A third option, that a textbook would give you, is to balance the input resistances. It's something that Jan could try quickly with what he's already built, if he wanted to.

For the inverting amplifer U3B, lift the non-inverting pin and connect a 50k resistor from the pin to ground. (If you don't have a 50k 1% resistor, use two of the 100k ones in parallel.)  That gets you this

the output of U3B should now be much closer to the gain of -1 and with 0.108V going in should read within a millivolt or two of -0.108V. At least that's what theory says. Try it and see.

To understand why it works, look at the line above the one I highlighted on the datsheet (offset current). That's saying that, although the bias current is fairly high, the difference in the bias current of the two inputs is quite small (no more than one part in a hundred - less than 1%). That comes about because, although the bias current is the base drive to a transistor and is a bit variable  (it will vary from batch to batch), the transistors on a particular chip match very well in their characteristics (because they're all fabbed together). That matching can be made use of.

For this part, the bias current is out of the input pin (it's the base current of a PNP transistor). When it flows through whatever resistance is connected to the pin it generates a voltage. If the resistance seen by both inputs is the same, then those voltages will be the same (we don't know precisely what the voltage will be, because the bias current varies quite a lot, but we do know that both inputs will see a very similar additional voltage because the currents match well). Since the two inputs are difference inputs, the two voltages will then pretty much cancel out as part of the normal action of the op-amp.

After Jan had done that (and assuming that it worked), he could then move on to the differential amplifier (U3C). To achieve the same effect there he merely needs to install the 680k to ground. Then we can see how close it brings things overall to the kind of accuracy you want. I suspect you'll want to move to a more up-to-date precision part (particularly if Jan is hoping to get anything close to the accuracy implied by working 16 bits), but let's see (and it's an interesting experiment to do, anyway).

So a couple of possibilities

Change the op-amp or reduce the 100K s to reduce the effect of the op-amp node, this would increase the risk of a fault condition blowing the op-amp but would fix the error, relatively of course, by going to say a pair of 20K we would improve things by 5*

Thoughts?