Finalizing the design of my lab PSU

[erikj]

Thank you guys for the feedback so far!

I have found some time to make a new version of the schematic, more feedback is welcome of course.

Main changes are:

• Got rid of the diff-amp from the current loop to make it faster
• Added a simple peak current limiter
• Paralleled the pass transistors
• Added output cap
• Used a resistor instead of a current sink for minimum load current
• Added some lead-compensation caps

I also found a mistake in the first schematic, the op-amp was supposed to be OPA227 and not OPA277 which is much slower.
The OPA227 was the one I was using for my tests, so the problems I had with the current loop probably weren't due to op-amp speed.

I'm still looking for a nice fast quad op-amp for this project.
I am not quite sure which parameters are the most important for this kind of application, so I made a short list of candidates I have found.

• AD8513 Nice specs, but CMRR seems a bit low. Not sure how important this is.
• AD8674 Is this fast enough?
• TLE2074 Offset drift seems a bit high.

Which one do you think will work best? Or do you have other suggestions ?

[void_error]

You don't really need R1, R3, R5 & R7, you can safely remove them.

If you've done the math right the voltage drop on R2, R4, R6, R8 should be between 0.3 & 0.7V. If it's too low you're going to have the transistor with the lowest V_BE conducting more than the ones with lower V_BE. You could get away with a low voltage drop on those resistors if the transistors have nearly identical gain although I'm not sure that's the case for darlingtons... and their gain and V_BE will also change with temperature.

I don't see any reason for the T9 & R35.

[Andreas]

I don't see any reason for the T9 & R35.

Think twice:
For pulsing loads its the only path that limits 2nd break down fast enough.

with best regards

Andreas

[David Hess]

I also found a mistake in the first schematic, the op-amp was supposed to be OPA227 and not OPA277 which is much slower.
The OPA227 was the one I was using for my tests, so the problems I had with the current loop probably weren't due to op-amp speed.

Usually the output capacitance and output transistor limits bandwidth at low loads so a faster error amplifier does not add much except for the need for more frequency compensation.

I'm still looking for a nice fast quad op-amp for this project.
I am not quite sure which parameters are the most important for this kind of application, so I made a short list of candidates I have found.

• AD8513 Nice specs, but CMRR seems a bit low. Not sure how important this is.
• AD8674 Is this fast enough?
• TLE2074 Offset drift seems a bit high.

Which one do you think will work best? Or do you have other suggestions ?

CMRR will not mean much without precision dividers.  Offset will not matter unless you need absolute accuracy which again, requires precision dividers.  Neither affect load or line regulation.

The AD8674 and OPA4227 are low noise but to take advantage of it, the impedances need to be below 10k.  These amplifiers have input bias current compensation so there is no need to balance the inverting and non-inverting inputs and resistors like R16 and R25 do nothing useful.  The FET input operational amplifiers do not need input balancing because they have such low input bias currents.

Operational amplifiers with high slew rates will recover faster when switching between voltage and current mode since there is no clamping of OP1.1 and OP1.3 depending on the external compensation.  That gives an edge to the JFET operational amplifiers.  Add some clamping and this would not be a consideration.

What are Z1 and Z2 for?  Are they just to protect the LEDs?

I would look into operating at a higher drive current, lowering the impedance of the feedback networks for T7 and T8, and buffering the outputs of OP1.1 and OP1.3 with emitter followers.  This would include replacing the BC series transistors with BC327s and BC337s or something better.  I wish there was a modern equivalent to the old 1+ watt TO-39 transistors.

[erikj]

You don't really need R1, R3, R5 & R7, you can safely remove them.

If you've done the math right the voltage drop on R2, R4, R6, R8 should be between 0.3 & 0.7V. If it's too low you're going to have the transistor with the lowest V_BE conducting more than the ones with lower V_BE. You could get away with a low voltage drop on those resistors if the transistors have nearly identical gain although I'm not sure that's the case for darlingtons... and their gain and V_BE will also change with temperature.

I the voltage drop in this case will be 5/4*0.1 = 0.125V max if the transistors are perfectly balanced (which they probably are not), so maybe that is too low then. It was hard to find any reliable info online so I used something low for the emitter resistors to avoid wasting power.
I added R1,3,5,7 because I it seemed to be a common configuration in other designs, not really sure about their exact purpose there but some kind of balancing I guess. They are useful for measuring the individual base currents though, so maybe I'll keep them until I have the balancing part figured out.

I don't see any reason for the T9 & R35.

Think twice:
For pulsing loads its the only path that limits 2nd break down fast enough.

Yep, they are put there for the really short current spikes that are too fast for the current loop and as an extra fail-safe.

[blackdog]

Hi void_error,

You are wrong, R1, R3, R5 & R7 are Needed!
Have you ever heard of emitterfollower oscillations?

erikj
C1 is on the low side, you need a bigger energy reservoir for dynamic load changes.

And, use about 50uF/ampere current on the output ( rule of thumb ).

Look @ ADA4077-4 as opamp,

Make R16 10K.

Remove T7 and T8, use 1n4148 diodes en place a LED in serie, green for de voltage loop and red for the currend loop, less components and more stability.

I like it more if the peak current limitor is build around the emittor resistors of the TIP142, but your system also works :-)
But test it for stability!

Do stability testing for you start with the circuit board.
Take good care of the wiring around C1 and C20, These capacitors make you Ri @ higher frequenties and your loop stability depents on them.

Kind regarts,
Blackdog

[erikj]

Usually the output capacitance and output transistor limits bandwidth at low loads so a faster error amplifier does not add much except for the need for more frequency compensation.

CMRR will not mean much without precision dividers.  Offset will not matter unless you need absolute accuracy which again, requires precision dividers.  Neither affect load or line regulation.

I am using 0.1% resistors for the differential amplifier. I wanted decent precision as I will be using 22-bit A/Ds and 5-digit displays for the current and voltage meters. I am hoping for <0.1% accuracy for the meters after calibration. If the drift is low I could compensate for offset and CMRR in software, but it would be nicer if I didn't have to do much calibration.

The AD8674 and OPA4227 are low noise but to take advantage of it, the impedances need to be below 10k.  These amplifiers have input bias current compensation so there is no need to balance the inverting and non-inverting inputs and resistors like R16 and R25 do nothing useful.  The FET input operational amplifiers do not need input balancing because they have such low input bias currents.

Operational amplifiers with high slew rates will recover faster when switching between voltage and current mode since there is no clamping of OP1.1 and OP1.3 depending on the external compensation.  That gives an edge to the JFET operational amplifiers.  Add some clamping and this would not be a consideration.

Thanks, I'll consider removing the input balancing resistors.
I am curious to know why impedance < 10k is needed? Is that because of the the current noise density?

I have tried to figure out a way to clamp them but have not been able to find a good solution.
So I think AD8513 might be the best choice then.

What are Z1 and Z2 for?  Are they just to protect the LEDs?

It is more like protecting the supply from future user stupidity (from me probably) . The LEDs will be indicators in the front panel, and usually I would not expect an indicator to be part of any important circuitry. So I was thinking that the supply should work without the LEDs and not lose regulation.

I would look into operating at a higher drive current, lowering the impedance of the feedback networks for T7 and T8, and buffering the outputs of OP1.1 and OP1.3 with emitter followers.  This would include replacing the BC series transistors with BC327s and BC337s or something better.

Some of the transistors might see a voltage of almost 50 volts, so I decided against BC327/337 for that reason and I wasn't able to find any higher voltage equivalents.
How much current do you think might be needed?

I could maybe increase it to 20mA or so. BC546/556 seems to be able to handle about 1W with a heat sink attached.

I wish there was a modern equivalent to the old 1+ watt TO-39 transistors.

Yep, I noticed how hard it was to find good fast transistors in the "a few watts" power range. And those TO-39 parts are easy to keep cool.

[David Hess]

You don't really need R1, R3, R5 & R7, you can safely remove them.

They are often a good idea for snubbing emitter follower oscillations.  The emitter degeneration resistors used to enforce current sharing are usually enough if they are at least Rb/hfe however if the transistors are mounted remotely, the wiring may have enough capacitance to require the base snubber at the transistor also.

I would like to refer to this as snivet protection or moderation but fear enraging the tube enthusiasts.  MOSFETs are more prone to this and maybe it would be more forgivable to use the term for them.

Slow transistors are almost immune to this but it still happens and no Darlingtons are really slow.  Lots of 2N3055 designs failed when they stopped using the 600 kHz single diffused processes.

If you've done the math right the voltage drop on R2, R4, R6, R8 should be between 0.3 & 0.7V. If it's too low you're going to have the transistor with the lowest V_BE conducting more than the ones with lower V_BE. You could get away with a low voltage drop on those resistors if the transistors have nearly identical gain although I'm not sure that's the case for darlingtons... and their gain and V_BE will also change with temperature.

I the voltage drop in this case will be 5/4*0.1 = 0.125V max if the transistors are perfectly balanced (which they probably are not), so maybe that is too low then. It was hard to find any reliable info online so I used something low for the emitter resistors to avoid wasting power.
I added R1,3,5,7 because I it seemed to be a common configuration in other designs, not really sure about their exact purpose there but some kind of balancing I guess. They are useful for measuring the individual base currents though, so maybe I'll keep them until I have the balancing part figured out.

I usually figure on equaling Vbe with the emitter ballast resistor at peak current but the real value needed is less.  If you ignore secondary breakdown effects, the minimum value can be calculated using either the Vbe or hfe change with temperature given a certain level of mismatch but it is seldom worthwhile.

I have found that simple grading of the transistors for Vbe makes a huge difference if you want the lowest possible emitter ballast value.

Darlingtons of course have approximately 2*Vbe so the change of base-emitter voltage over temperature will be doubled but I have never tested it empirically because I have not used them in parallel.  The emitters are ratio matched and track with temperature so it would be fun and not difficult to calculate the actual emitter ballast resistor requirements.  Hfe effects would be more interesting.

I don't see any reason for the T9 & R35.

Think twice:
For pulsing loads its the only path that limits 2nd break down fast enough.

Yep, they are put there for the really short current spikes that are too fast for the current loop and as an extra fail-safe.

I think it is a great idea.  It is simple yet effective and prevents some rather catastrophic failure modes.

[David Hess]

Usually the output capacitance and output transistor limits bandwidth at low loads so a faster error amplifier does not add much except for the need for more frequency compensation.

CMRR will not mean much without precision dividers.  Offset will not matter unless you need absolute accuracy which again, requires precision dividers.  Neither affect load or line regulation.

I am using 0.1% resistors for the differential amplifier. I wanted decent precision as I will be using 22-bit A/Ds and 5-digit displays for the current and voltage meters. I am hoping for <0.1% accuracy for the meters after calibration. If the drift is low I could compensate for offset and CMRR in software, but it would be nicer if I didn't have to do much calibration.

Worst case 0.1% resistor matching with a gain of 0.1 yields a CMRR of 49 dB so the operational amplifier CMRR will not be a limitation.  Worst case 0.01% resistors would still only be 69 dB.

http://www.linear.com/docs/41248

CMRR requirements are relaxed since the change in common mode voltage itself is low.

I have designed and built similar fixed power supplies with remote sense that had load and line regulation measured in microvolts or less.  My favorite configuration is something like this:

http://circuitdiagram-schematic.com/high-stability-regulator-circuit-using-lm108/

The AD8674 and OPA4227 are low noise but to take advantage of it, the impedances need to be below 10k.  These amplifiers have input bias current compensation so there is no need to balance the inverting and non-inverting inputs and resistors like R16 and R25 do nothing useful.  The FET input operational amplifiers do not need input balancing because they have such low input bias currents.

Operational amplifiers with high slew rates will recover faster when switching between voltage and current mode since there is no clamping of OP1.1 and OP1.3 depending on the external compensation.  That gives an edge to the JFET operational amplifiers.  Add some clamping and this would not be a consideration.

Thanks, I'll consider removing the input balancing resistors.
I am curious to know why impedance < 10k is needed? Is that because of the the current noise density?

Yes.  It is only approximate but dividing the voltage noise by the current noise yields the source resistance necessary to double the voltage noise which is a reasonable place to start.  Resistors noise is usually insignificant.

http://www.linear.com/docs/4345

I have tried to figure out a way to clamp them but have not been able to find a good solution.
So I think AD8513 might be the best choice then.

I do not have any good examples to link.  Diode or zener clamps across the feedback capacitor are common but there are other ways using transistors or maybe current switched diodes.  The idea is to prevent the operational amplifier from charging the integration capacitor excessively when operating open loop and the other error amplifier has control.

Some operational amplifiers have special clamp pins or you can do it via an external compensation pin.

Minimizing the integration capacitance shortens the recovery time but clamping is still necessary for the fastest response.

What are Z1 and Z2 for?  Are they just to protect the LEDs?

It is more like protecting the supply from future user stupidity (from me probably) . The LEDs will be indicators in the front panel, and usually I would not expect an indicator to be part of any important circuitry. So I was thinking that the supply should work without the LEDs and not lose regulation.

LEDs tend to fail short but I agree with the concept.  Tektronix just put an LED (and diode) in series with the output of the current loop amplifier to indicate current limiting.

I was thinking it might be better to use a slightly higher voltage zener and a resistor in series with the LED.  Then the bias current from the current source can be raised without affecting the LED.

I am not fond of low voltage zener diodes though and by themselves, zener diodes have an appalling tenancy to fail open.  A slightly more complicated design would use a resistor divider to fix the base voltage of a bipolar transistor so the transistor replaces the zener.  Oddly enough the transistor Vbe would compensate for the LED forward voltage drop for better LED current regulation.

I would look into operating at a higher drive current, lowering the impedance of the feedback networks for T7 and T8, and buffering the outputs of OP1.1 and OP1.3 with emitter followers.  This would include replacing the BC series transistors with BC327s and BC337s or something better.

Some of the transistors might see a voltage of almost 50 volts, so I decided against BC327/337 for that reason and I wasn't able to find any higher voltage equivalents.
How much current do you think might be needed?

I could maybe increase it to 20mA or so. BC546/556 seems to be able to handle about 1W with a heat sink attached.

I was thinking 20 milliamps but as you point out, the voltage is marginal on the BC327/BC337 and I think 1 watt is optimistic for any TO-92 transistor although I have heat sinked them in the past.

I wish there was a modern equivalent to the old 1+ watt TO-39 transistors.

Yep, I noticed how hard it was to find good fast transistors in the "a few watts" power range. And those TO-39 parts are easy to keep cool.

There are lots of surface mount options but I have been slowly compiling a list of suitable TO-126 transistors:

2SA1507/2SC3902   160V 1.5A 1.5W 120MHz
2SB1143/2SD1683   50V 4A 1.5W 150MHz
BD139/BD140   100V 1.5A 1.25W 190MHz(?)

[blackdog]

Hi erikj,

What is de reason that you are using T7 and T8?
With 4x TIP142 you only need about 4ma total base drive, if the transistors getting warm it wil be far below that.
Typical i measured about 0,5mA @ 2Ampere for the TIP142.
A "Pull Up" current source of 7 a 8ma wil be more than enough for this design and the curent its about right for driving your CC or CV LED's.

The best way to be sure that you dont have "switch On" transients is to delay the current source that drive the TIP142.
Also watch out for "Switch Of" transients, a powersupply that kils a connected microcontroler with a 13V puls is a BAD power supply 

Kind regarts,
Blackdog

[erikj]

I usually figure on equaling Vbe with the emitter ballast resistor at peak current but the real value needed is less.  If you ignore secondary breakdown effects, the minimum value can be calculated using either the Vbe or hfe change with temperature given a certain level of mismatch but it is seldom worthwhile.

I have found that simple grading of the transistors for Vbe makes a huge difference if you want the lowest possible emitter ballast value.

Darlingtons of course have approximately 2*Vbe so the change of base-emitter voltage over temperature will be doubled but I have never tested it empirically because I have not used them in parallel.  The emitters are ratio matched and track with temperature so it would be fun and not difficult to calculate the actual emitter ballast resistor requirements.  Hfe effects would be more interesting.

I don't know how to perform those calculations, so I think it would be easier for me to simply get some different values for emitter resistors (maybe 0.1 - 0.4 ohms) and do some experiments later. And also get a few extra transistors for matching.

Worst case 0.1% resistor matching with a gain of 0.1 yields a CMRR of 49 dB so the operational amplifier CMRR will not be a limitation.  Worst case 0.01% resistors would still only be 69 dB.

http://www.linear.com/docs/41248

CMRR requirements are relaxed since the change in common mode voltage itself is low.

I have designed and built similar fixed power supplies with remote sense that had load and line regulation measured in microvolts or less.  My favorite configuration is something like this:

http://circuitdiagram-schematic.com/high-stability-regulator-circuit-using-lm108/

Very useful document, thanks.
I did some measurements of the resistors and found that they are matched to about 0.02%, so op-amp CMRR shouldn't be an issue then.

I do not have any good examples to link.  Diode or zener clamps across the feedback capacitor are common but there are other ways using transistors or maybe current switched diodes.  The idea is to prevent the operational amplifier from charging the integration capacitor excessively when operating open loop and the other error amplifier has control.

Some operational amplifiers have special clamp pins or you can do it via an external compensation pin.

Minimizing the integration capacitance shortens the recovery time but clamping is still necessary for the fastest response.

The self-biasing transistor topology I am using makes this a bit tricky I think. But there should be some way to do it, I'll keep thinking.

LEDs tend to fail short but I agree with the concept.  Tektronix just put an LED (and diode) in series with the output of the current loop amplifier to indicate current limiting.

I was thinking it might be better to use a slightly higher voltage zener and a resistor in series with the LED.  Then the bias current from the current source can be raised without affecting the LED.

I am not fond of low voltage zener diodes though and by themselves, zener diodes have an appalling tenancy to fail open.  A slightly more complicated design would use a resistor divider to fix the base voltage of a bipolar transistor so the transistor replaces the zener.  Oddly enough the transistor Vbe would compensate for the LED forward voltage drop for better LED current regulation.

There are lots of surface mount options but I have been slowly compiling a list of suitable TO-126 transistors:

2SA1507/2SC3902   160V 1.5A 1.5W 120MHz
2SB1143/2SD1683   50V 4A 1.5W 150MHz
BD139/BD140   100V 1.5A 1.25W 190MHz(?)

A shunt resistor is another option I am thinking about to reduce LED-current. Maybe shunting the LED with a 150 ohm resistor or something like that. Also I wouldn't need the zener.

Why is it important to increase the drive current? Is it to make it faster?
I was using BD139 in the first version of the schematic, but it turned out to be too slow. I think that base-collector capacitance might be the problem, as it should be fast enough in terms of bandwidth. The 2SA1507/2SC3902 looks better in terms of capacitance, so that might be an option. Also lowering the impedance as you mentioned might help.

[erikj]

C1 is on the low side, you need a bigger energy reservoir for dynamic load changes.

And, use about 50uF/ampere current on the output ( rule of thumb ).

Look @ ADA4077-4 as opamp,

Make R16 10K.

Remove T7 and T8, use 1n4148 diodes en place a LED in serie, green for de voltage loop and red for the currend loop, less components and more stability.

I like it more if the peak current limitor is build around the emittor resistors of the TIP142, but your system also works :-)
But test it for stability!

Do stability testing for you start with the circuit board.
Take good care of the wiring around C1 and C20, These capacitors make you Ri @ higher frequenties and your loop stability depents on them.

ADA4077-4 is a nice op-amp, I like the low drift and offset. The slew-rate is a bit low, and that might be a problem if I can't get the integrator-windup problem sorted out.

You are right, C1 should probably be higher. How much depends on the voltage margin I will use for the pre-regulator.

What is de reason that you are using T7 and T8?
With 4x TIP142 you only need about 4ma total base drive, if the transistors getting warm it wil be far below that.
Typical i measured about 0,5mA @ 2Ampere for the TIP142.
A "Pull Up" current source of 7 a 8ma wil be more than enough for this design and the curent its about right for driving your CC or CV LED's.

The best way to be sure that you dont have "switch On" transients is to delay the current source that drive the TIP142.
Also watch out for "Switch Of" transients, a powersupply that kils a connected microcontroler with a 13V puls is a BAD power supply 

The reason for T7 and T8 is all about the voltage range of 0-40V. I could maybe use a +-22V op amp and drive directly but I would be running it very close to the maximum voltage rating. I would need an op-amp supply of maybe -1 and 42 volts and extra regulators for that.
There are other options I have considered like bootstrapping a regular op-amp or getting a few of those high-current high-voltage ones that goes up to 60V or more.
Or I could do what you did in your design and reference the control circuitry to the positive output.

[David Hess]

ADA4077-4 is a nice op-amp, I like the low drift and offset. The slew-rate is a bit low, and that might be a problem if I can't get the integrator-windup problem sorted out.

There is probably no good reason to borrow trouble here.  Similar designs using diode switching between constant voltage and constant current get by with slow 741 type amplifiers no clamping.  I would clamp the integrators just for the design challenge although it is pretty straightforward.  I am surprised at the lack of examples online.