Linear Power Supply based on HP/Agilent E3610A

[timsu]

Hello,
I'm trying to build a power supply based on the HP E3610A (15V/2A).
You can find the Schematics as attachement to this post or http://sites.fas.harvard.edu/~phys191r/Bench_Notes/A1/agilent_e3610a.pdf on page 16.
On the pages before there are also the component values.

I think I understood most of the circuit, except Q2/CR8( (HP schematic) Q3/D7 (my schematic).
The simulation seems to work fine even without it.
Can somebody explain to me in what situation the transistor is passing current and when not? What is the point of it?

In a service note HP is recommending to remove CR8/D7 to prevent oscillations.
But here https://www.eevblog.com/forum/testgear/hp-agilent-e3610a/ it is suspected, that this modification can destroy the pass transistors because of a too high base-emitter voltage.
Also, should I include the modification, the added cap in the voltage error amplifier, from the linked thread?

Did HP make more "bad" choices which I can prevent?

[Kleinstein]

The transistor Q3 is working as an emitter follower - so it's always in a linear range, essentially never off. The diode limits the lower output voltage to something close to -100 mV. It also provides higher current sinking capability. This should not be important with the shown BJT power stage, but is can be important with HPs MOSFET stage. The high sinking current allows to turn off the FETs fast and the 1.2 K resistor limits the speed when turning them on. 

The values for C1, R2 look odd - looks way to slow. I would more expect something in the 500 pF and 5 K range. It depends on the output caps and power stage too.

One poor choice HP made was adjusting the voltage by changing the feedback divider. It's better to keep the divider fixed and adjust the set voltage like indicated. However with a variable ref voltage for the divider one might need an extra minimum load (e.g. constant current) to bring the output back to zero.   

Also the supply voltage for the OPs could have be chosen lower, not to allow the OPs to go way to high in output voltage. Limiting the voltage excursion for the regulation part that is not active can speed up the switch over from CC to CV mode and back. Especially with a BJT output stage the voltage range is well predictable, so it's easy to chose more reasonable limits. So going for a Rail-Rail OP might be one option - the LM1001 is rather slow anyway.

The output caps should be a combination of a low ESR (e.g. ceramic / foil in the 100 nF to 1 µF range) and low ESR electrolytic. HP might not have modern low ESR electrolytic as an option back then. So one might get away with lower capacitance - the large caps chosen by HP might be needed to get the ESR down in the 1 Ohm or lower range, not the capacitance itself. The old style high ESR caps are usually not helping - so I would prefer lower capacitance low ESR (e.g. 0.1-0.5 Ohms range).

[timsu]

So why would HP decide to remove the diode because of oscillations?
That could cause harm to the transistors, right?

The values for C1, R2 look odd - looks way to slow. I would more expect something in the 500 pF and 5 K range. It depends on the output caps and power stage too.

I thought this was also confusing, because in the high voltage model E3612 they used 22 pF and no resistor in the voltage feedback loop.

However with a variable ref voltage for the divider one might need an extra minimum load (e.g. constant current) to bring the output back to zero.   

The 330 Ohm R9 should work, right? This was also added by HP in a later revision.

Also the supply voltage for the OPs could have be chosen lower, not to allow the OPs to go way to high in output voltage.

I would prefer going for -5V/+5V because for other parts of the circuit (DAC/ADC) I will need 5V.

Also why use a voltage divider with 1R and 500K before the current shunt?
Are they compensating some error with this?

[Kleinstein]

Removing the diode (CR8) to limit negative control voltage should not change much with oscillations - it could be an issue if the output stage (gate capacitance) is setting much of the compensation. The diode might limit the ability to go all the way to 0 V output voltage, especially if the transistors are hot. However the more logical step would be using two diodes in series instead of removing it all together. Allowing the voltage to go to negative (much below -7 V) without the diode might cause trouble for the output transistors (to negative a base-emitter voltage), so it is not a problem with only a +-5 V supply there.

R9 could to some degree act as a minimum load, though it is not very efficient at low voltages and gets quite hot at high voltage. A constant current sink might be more appropriate.

The divider at the shunt looks odd - though it could be just to use the same meter with different supply models.

[timsu]

I think HP was limited in their ability to change something aftwerwards because they had no space on the board for two diodes, or a constant current sink, so they added just a big resistor as load.

I started to draw up the schematic (attachement). The OP-Amps are just placeholders, maybe someone has a recommendation.

I will have to do some more simulations to determine the right component values for the voltage error loop and the output cap.
Also the digital stage is still missing, it will be 12 Bit ADC/DAC and an ATmega.

I would appreciate feedback.

[Kleinstein]

The constant current sink could use the +5 V auxiliary supply to work all the way down to 0 V (or even a little below): could be just a PNP transistor with base to GND, collector to the neg. output and emitter through a resistor to the +5 V. The down side is that the current will flow through the shunt too, but there is base current flowing anyway. Also output voltage could go slightly below zero (a resistor/trimmer at the base could be used to adjust this to a few mV).

ESR of the output capacitor might be important for stability - keep that in mind for simulation. Low ESR electrolytics have about the right ESR. The cap might be a little smaller with a low ESR version.
Stability with very large caps might need a capacitor (or RC) in parallel to R29.

A 10000 µF filter cap may not work very well with a relay for transformer tap switching, especially at higher voltage (e.g. 30 V). I would consider using a transistor to do the transformer tap switching. For a transformer with a split coil, like 2 x 15 V this works rather good and today a transistor might be cheaper than a relay. As a side effect no more problems with highly dynamic loads that might cause the relay so switch to often. The circuit is even simpler than relay control.

For OPs something like the LF412, TL072 or TLC272 would be OK for the regulator. The readback of the current and possibly voltage might want more precision, like an OP07.
The Output disable could use the second half of the LM393.

[timsu]

Your solution for the constant current sink works fine (mine stopped working at around 0.4V). I can remove the error current through software.

Also output voltage could go slightly below zero (a resistor/trimmer at the base could be used to adjust this to a few mV).

What would be the advantage of this?

The maximum voltage will be only at 15V. With 2 Amps this means 30W of heat in the pass transistor. If I use an large transistor (TO-247/264) maybe tap switching is not even needed.

As a side effect no more problems with highly dynamic loads that might cause the relay so switch to often. The circuit is even simpler than relay control.

Something like a half bridge with two mosfets? Should I use an IC for that?

Can I use the OP07 in the current feedback loop or is it too slow for that?
The second half of the LM393 was planned as comparator with hysteresis for tap switching.

[Kleinstein]

The OP07 is relatively slow. Depending on the speed planed it might work too. However current regulation is usually not needed to be as stable as the current reading.

Using a second transistor for the second tap would use two filter caps get a raw voltage of something like a 0, 10 V and 20 V. Transistor stage is more like a class H audio amplifier: the 10 V connect via a diode to the collector. The 20 V via an NPN transistor with the basis connected through a resistor to the 5 V auxiliary supply. The transistor needs to be cooled as well, as it will take over much of the loss at some output voltages.
For just 30 W maximum loss (even if more like 40 W since for a 15 V output the rectifier should be more like 20 V) there is no real need for tap switching, but it could permit a smaller heat sink (e.g. 60% the size) at relatively low extra costs (2nd capacitor, 1 diode, 1 power transistor - though smaller TO220 transistors could be used than, as the maximum power is about half).

Having the constant current sink go down to below 0 might be an issue with the output disabled. In this case the output would go slightly negative, if the current in not turned off too. So adjusting the minimum voltage to something like -100mV instead of -500 mV might be a good idea.

[timsu]

However current regulation is usually not needed to be as stable as the current reading.

I agreee that this is not as important, but the set current limit should not be too far off from reality. With an 0.1 shunt resistor, each mA is 100uV. If the used op amp in the feedback loop has 10mV offset the current limit would be quite inaccurate.

I don't know if the tap switcher is correct, I attached an simplified version. Why would the switching transistor need cooling? It is either saturated or off, the power dissipation should not be over half a watt.
This solution seems nice, except it wastes quite a bit pcb space with the huge caps doubled.

The negative voltage when the output is switchted off is about -200mV, limited through the output schottky diode.
I could not limit this further, it would require an base resistor of several 100k , which is too large too let current flow through the transistor.

[Kleinstein]

The transistor to use the higher raw voltage is not switching but trying to give the "normal" transistor something like 3-4 V to work with. So at about half the voltage it will handle much of the power loss. Not as much as the single transistor, solution but still half the voltage times all the current in worst case. However its only one of the two transistors at a time that has the high power. With a low voltage a MOSFET is even better for the switchover. Attached is a crude plan to show how the second tap could be used.

For the caps: you need two caps, about twice the capacitance but only half the voltage rating. So about twice the mechanical size (and price). On the other side you save on the heat sink. So tap switching might be more expensive, but could be smaller if you want it without a fan. Usually a passive heat sink is larger than the caps.

Just 15 V and 2 A output could still work with a single transformer tap, even with a passive heat sink.

[timsu]

Thank you for your help so far.
I would prefer to have only one transistor which needs active cooling (even if the total heat stays the same), it makes mounting and cabling easier.
The solution from my previous post should still work, right? I just need a comparator to do the switching.
For the constant current sink I think will settle for the solution which makes the output to go to 0V if it is switched off, even if it justs works down to 0.5V.

The LF412 looks good so far for the error loops. For the readout maybe OP07, does something comparable exists in a dual package?

[Kleinstein]

There are a few precision OPs in a dual package, e.g. LT1013, OPA2177, MCP6V12 and similar.
The really high precision ones can be quite expensive though. There are many that are better than LF412 / LM358, but still relatively cheap.

When doing range switching with a relay, one should not make the filter cap so large as this adds quite some load to the contacts. It might also be worth using two extra diodes and keep the lower voltage always on and only add the path for the higher voltage. With this low power there still is the option of not switching at all.

[timsu]

I meant switching with one transistor as shown in post #8, not a relay.
To summarize my tap switching options:

• no switching: (+) simple, small pcb (-) large heatsink
• relay switching: (+) filter caps only one required once (-) relay wear, dynamic loads, comparator required
• switching as in post #8: (+) no wear, only one power transistor (-) comparator and two filter caps required
• switching as in post #9: (+) no wear, few components (-) two filter caps and two power transistors

I still have not deciced yet, but maybe I will use two channels. In this case tap switching would be nice.
Most heatsinks I looked at go down to 0,8K/W. For one channel this would not be a problem, with two it could get a bit hot.

[Kleinstein]

For the circuit from post #8, there is still the question on how to power the comparator / driver. If powered from the auxiliary (5V) supply the drive current will flow through the shunt and an extra diode again to negative a base voltage is needed.
So for switching the taps I would prefer a MOSFET over a NPN. At the rather low voltage there should be no big problem in protection against too much gate voltage. However using an N-MOSFET at the high side like drawn would need an extra supply, as the +5 V auxiliary supply for the regulator is no enough. So a slightly different circuit is needed.

There is also one more difference between relay and transistor use for tap switching: the transistor solution can be fast enough to even include the ripple on the raw voltage. So the transition could be set lower and with less (no) hysteresis.

For a dual channel version, the two channels will normally be fully separate, except the digital control signal going from one part to the other. It might get difficult to find a transformer with two spilit windings. So this  might make tap switching more difficult.

[timsu]

Is it even possible to do the switching without another floating supply circuit?
It seems kind of difficult because GND has no static potential to the tap switching.
I tried using a P-MOSFET with a pullup.
But how to pull the gate down with a ground-referenced circuit?

I know for two channels the transformer becomes a bit of a problem. I have to see what a custom made one costs.

[StillTrying]

In case it's of any use.

[Kleinstein]

You could do the switching with a N_MOSFET, if you choose to switch the other half of the two voltages. It looks a little like a bridge circuit. The MOSFET goes from the most negative voltage to the negative output, with the diode from the middle voltage as an alternative. The controlling comparator is powered from the main supply and testing the voltage over the output transistor. So should be reasonably easy at low voltages (e.g. < 30 V), so the comparator can handle.

Though slightly larger and less efficient, one could still use two separate power transformers for the main power and one small one for the two auxiliary supplies. For the auxiliary supply one can get a way with just a single winding, as there is no real need for a large negative supply something like a -1 V and +5 V or maybe +8 V can be enough. The -1 V could use just two diodes as shunt stabilization. The down side might be limit in the voltage between the outputs.

[timsu]

I kind of got it working.
At around 7V the taps should be switching, the comparator U5 does this.

But I still have some trouble controlling the MOSFET. With some opamps/comparators there is some weird oscillations. The LT1077A worked, but it is not rail-to-rail, so it does not fully turn the MOSFET M1 off.

[Kleinstein]

You don't want an rail to rail OP for the MOSFET control, as the output voltage would be to high. A comparator like LM393 with a little hysteresis should do it. The gate voltage could be from the lower 10 V supply.
I would compare the collector side of the power transistor to something like 3 V higher than GND (at emitter resistor).  One  might need to check how the current from the hysteresis flows, but I won't expect trouble there.

[timsu]

I can't quite get it to work.
I used the LM393 as comparator.
The inverting input is collector-3V (with a diode drop).
The noninverting input the emitter voltage.
Output of comparator with a pullup to the 10v low tap.

Because the comparator has such a high imput impedance, a parallel resistor is required, to have a diode drop.
If the comparator switches to on, the voltage goes on only 3V higher than in the off state (100R pull up), which turns the fet not complete on.
Lower pullups make the circuit oscillate.

Maybe it is obvious but all the different voltages which different reference to each other confuse me

Edit:
I tried an different circuit with an optocoupler, but even here I could not get over 3V Gate-Source Voltage.

If i disconnect the gate it goes up to 10V

Edit2:
I got it working

[ZeTeX]

You might want to have a look here:
https://www.eevblog.com/forum/projects/lab-psu-design-ideaquestions/msg205205/#msg205205

[timsu]

This circuit has the disadvantage of high power dissipation in the MOSFETs and an additional higher voltage to drive the gates is required.
I managed to save a few components compared to my last post with inverting the comparator input.
I will post an updatete schematic soon.

I looked for pass transistors and found these:
MJL 3281A
MJH 11022G
TIP142

What are your opinions on these?

[sarepairman2]

keep in mind this was designed for mass production. you can do better  O0

[Kleinstein]

The tip142 is good enough for 15 V 2 A, even without tap switching. So no need for the larger and more expensive ones.
With tap switching even a smaller version might be OK, but this won't save that much and a larger transistor could alow for a higher temperature or slightly smaller heat-sink.
The MJH 11022 looks a little slower - so I would not prefer that one.
The MJL 3281A is faster, but you need a good layout and likely quite some real life tests to really make use of it.

The FET tap switching from the old linked task is very similar to the one I suggested before, just for more taps and a regulator with emitter follower type output stage.  The main disadvantage is that the FETs at some times also have quite some power - so usually the FETs and the main output transistors should be mounted on the same heat-sink. Especially the 2 tap version I showed is really simple. For the relatively low voltage (e.g. < 20 V) the MOSFET sees it is not that difficult to find MOSFETs that can work in linear mode.

There is some extra effort to make the FETs switching and there can be small glitches when switching taps - so hysteresis is needed to avoid this (and possible instability) to happen just from ripple or load changes.

[timsu]

Here another update:
- fixed some smaller bugs (wrong values)
- added tap switch
- opamp model selection
- beginning digital part
- simpler output disable

Todo:
- passive and transistor selection
- more protection
- digital circuit
- ...

Edit:
Should I add an RC Filter at the Output of the DAC?