Linear Power Supply based on HP/Agilent E3610A

[timsu]

Another update, now the digital circuit is done as well, with isolated serial port.
I decided to use an Arduino instead of using just an Atmega, because it does not really need that much more space.
The component selection is also mostly done.
Is the protection for all the op amps done right?

[Kleinstein]

The protection at the OPs looks Ok. The current regulator might not need the diodes, as there usually will not be so much current to make the diodes work or even damage the OP.

If you want really low noise, RC filtering after the DAC might be helpful, but this slows down changes in the set voltage.

The TL072 can have quite some offset, this might limit the minimum current setting.

[StillTrying]

You're asking too much of the LM393 to pull down 100R from 5V. (10mA max)
I'd try to invert the sense there and make the opto part of the pull up to 5V because I hate wasted milliamps. 
R3 seems to have changed from 10K to 1K, needing the opto to sink ~28mA.
R22 doesn't do anything.
Didn't notice anything else.

[Dave]

The serial interface isn't actually isolated.

[timsu]

The TL072 can have quite some offset, this might limit the minimum current setting.

Maybe the TLC 272?

You're asking too much of the LM393 to pull down 100R from 5V. (10mA max)
I'd try to invert the sense there and make the opto part of the pull up to 5V because I hate wasted milliamps. 

Can you explain this a little bit more?
I thought about using something like this: http://www.mikrocontroller.net/attachment/45300/active-pullup.png

R22 doesn't do anything.

You're right, I forgot to remove it after I changed the protection diodes.

R3 seems to have changed from 10K to 1K, needing the opto to sink ~28mA.

Does 10K ensure fast enough switching of the MOSFET?
LTSpice shows about 2u seconds switching time with 10k.

The serial interface isn't actually isolated.

TX has to use an external 5V source, right?

Why did HP added resistors between GND and op amp input? (Example: R26 at the inverting input of the voltage error amp)

[Dave]

The serial interface isn't actually isolated.

TX has to use an external 5V source, right?

Yup.

Why did HP added resistors between GND and op amp input? (Example: R26 at the inverting input of the voltage error amp)

To cancel out the offset voltage caused by the input bias currents.

[sync]

Why did HP added resistors between GND and op amp input? (Example: R26 at the inverting input of the voltage error amp)

To cancel out the offset voltage caused by the input bias currents.

I don't think it's about the offset voltage. It's a LF411 JFET op amp in the HP design. I think it's for the op amp compensation. Without R26 the inverting input would be grounded and R21, C22 from the op amp output to the inverting input have no effect.

[Dave]

I don't think it's about the offset voltage. It's a LF411 JFET op amp in the HP design. I think it's for the op amp compensation. Without R26 the inverting input would be grounded and R21, C22 from the op amp output to the inverting input have no effect.

You're absolutely right. This is what happens when you just glance at schematics instead of taking the time to give it a proper look.

The bias current compensation resistor like I had in mind would be R33, for example (though it should be a higher value - around 50k).

[StillTrying]

You're asking too much of the LM393 to pull down 100R from 5V. (10mA max)
I'd try to invert the sense there and make the opto part of the pull up to 5V because I hate wasted milliamps. 

Can you explain this a little bit more?

On the data sheet graphs the open collector output of the LM393 it can only sink about 10ma before its saturation voltage is 1V and rising. With 100R pull up it just won't be able to pull down enough. I think you'll have to invert it and use the opto as the pullup so that all the current goes through the opto.

I started to draw it and then decided to make it 2 working versions. Just to be awkward the simulations version of the LM393's output saturation seems to be about 4 times better than what's shown on the data sheets graph.

Does 10K ensure fast enough switching of the MOSFET?
LTSpice shows about 2u seconds switching time with 10k.

I think a few uS is fine for this occasional switching of 12V at up to 2A, but there's much better expurts than me here!

[StillTrying]

To cancel out the offset voltage caused by the input bias currents.

And especially their drift with temperature.
I'm surprised at how every everybody just sticks 10Ks, on op amp inputs, it seems like almost 50% of the time they're actually unbalancing the inputs!

[Kleinstein]

With FET OPs like the like LF412 (HP design), TL072 or TLC272 there is no real need to balance the input impedance. This also applies to some other OPs with compensated input bias (like OP27) or those with a relatively low bias and not so high input impedance (e.g. LT1013). So there are a lot of exceptions where you don't need balanced impedance. The 10 K value is about right to prevent excessive currents that might damage some OPs and for the typical 1 MHz GBW OPs it's low enough you don't have to look at parasitic capacitance so much.  Also bias currents are often not that critical at this resistance and thus no real need for balanced impedance with most OPs. So it a natural choice if not critical.

The LM393 has a limited drive capacity, so one should not use the 100 Ohms to make it drive some 40 mA, which is more than it can.

[kridri]

When I was looking to the schematic I don't see the point of adding resistor R15 (parallel to the Current sink). When you add a resistor parallel to it. It won't be constant anymore and at a higher voltage, the current increases. With more losses! Please can you explain me why you did it?

[timsu]

To slowly discharge the output caps when the device is turned off or at very low voltage.
It does not have to be exactly constant current.

[kridri]

To slowly discharge the output caps when the device is turned off or at very low voltage.
It does not have to be exactly constant current.

Thanks for the reply and the explanation. If forgot that a bench supply must be turned down. Silly me!

[Robaroni]

I have an E3610A that the current adjust seems to drift. That is, I can't set the current to zero. I tried to recalibrate the supply but after awhile it drifts above zero and can no longer be set to zero.

I swapped out the current opamp U4 as I didn't see anything that stuck out as being bad but it didn't fix the trouble. It might be something as simple as the wire wound current pot breaking down but I was wondering if any of have seen this problem before I dig it apart.
Thanks for the help,
Rob

[timsu]

Should the ADC/DAC should be level shifted up to get a higher accuracy at lower output voltages and allow better calibration?

[Kleinstein]

It depends on the quality of the ADC / DAC whether it might be worth using different ranges.

As the modern sigma delta ADCs offer rather good resolution, there is usually not need for a range switching. Also many ADCs measure negative values as well, so no problem with offsets around zero. It might be a good idea to make use of the differential inputs.

For the DAC it might be worth using two ranges, as really good DACs are expensive. A little shifting near zero might be needed if digital offset calibration is wanted.

[timsu]

How would be the differential inputs used?
What would be connected to the inverting inputs other than GND?

[Kleinstein]

The GND for current sensing (e.g. pin of shunt -> amplifier for current reading) can be a little different from the Ground for voltage sensing (e.g. Sensing line to output connector).  For a low ohms shunt the wires/lines going to the resistor can add to the resistance. So voltage should be read close to the resistor, possibly even using a shunt with 4 wires. The difference will not be large, maybe a few mOhms times the output current, but it's possible with little effort to avoid the error.

[mcckevin]

Hey Roberoni --
I just stumbled across an E3610 with the same problem. I also replaced U4 (great minds ...) with little success. Did you have any luck with this? Thanks.
Kevin

I have an E3610A that the current adjust seems to drift. That is, I can't set the current to zero. I tried to recalibrate the supply but after awhile it drifts above zero and can no longer be set to zero.

I swapped out the current opamp U4 as I didn't see anything that stuck out as being bad but it didn't fix the trouble. It might be something as simple as the wire wound current pot breaking down but I was wondering if any of have seen this problem before I dig it apart.
Thanks for the help,
Rob

[Robaroni]

Mcckevin-

I have several supplies so I just use this in situations where I don't need to control low currents. I think the trouble is the 10 turn current pot may be breaking down and not holding resistance. I haven't pulled it apart to absolutely verify it though.

Rob