uSupply - LM350T instead of LT3080

[IrejectYourReality]

Hi, I am building a lab power supply based on the uSupply. Instead of the LT3080, I want to use LM350T, because it is cheaper and can take 3 amps. But it has 1.25V interntal reference, so I cannot bring it down to 0V using the original uSupply circuit. So my idea was to bring the adjust pin on LM350T down to -1.25V. To do that, I devised this circuit:



It is basically just a summing amplifier, which will "subtract" the -1.25V. Output of the opamp should be -1.25V to 23.75V, to give 0-25V at the output of the regulator. To get the negative voltage needed for the opamp and -1.25 voltage reference, I plan to use this -

Will this circuit work?
Also, is the LM358A a good opamp for this application, or should I go with another one?

[Kleinstein]

The circuit should work as an adjustable voltage regulator. However it is not such a good idea to use the OP to implement a kind of current regulation this way.

The LM358 is ok for this application, unless you need very fast changes. The offset could be an issue - so one might have to include this in the control voltage. Due to limited accuracy, there is no need to use really accurate resistors - so 1% (or even 5%) types should be good enough.

[IrejectYourReality]

My plan is to use the same current regulation circuit as Dave used:



Current will be measured by the MAX4080. Is this the correct way to do it, or not?

[Kleinstein]

The current regulation like Dave used in his µSupply is not that good. It is more of a bad example, or how not to build a lab supply.

The voltage regulator just has very high gain (current controlled by voltage), when using it to drive current through a low impedance load. So the circuit would be either prone to oscillation or ridiculous slow.
To use a regulator chip like the LM350 (or the LT3080) one would have to have a resistor behind the regulator and add your own voltage regulation loop to compensate for the drop at the resistor. So the regulator chip is more or less used as a kind of protected transistor.

[IrejectYourReality]

Thanks for the advice!
I was also thinking about making the constant current regulation software-only. Since I can measure voltage and current, I can reduce the voltage to get a set current, but I fear that the regulation might not be fast enough. I plan to use  ATXMEGA32E5-AU running at 32MHz to control the whole thing. Would that be a good idea or a bad one?

[Kleinstein]

Software control is not a good idea - it is slow and resolution is limited. For a lab supply one would want something like a 100 kHz control loop bandwidth and maybe 16-24 Bit of amplitude resolution. SO nothing easy to do in software.

[IrejectYourReality]

So, what would be a good solution for my power supply? Some modification of Dave's circuit, or something completely different?

[wraper]

LM350T has 10 times higher dropout voltage than LT3080 has. So besides other issues, you would need to ensure that voltage on it's input is at least 2.5V higher. Then as a consequence comes huge increase of heat dissipation.

[Kleinstein]

I would prefer something completely different. More like the typical old HP power supplies work: Have a MOSFET or darlington transistor as a power stage, and a floating regulator with 2 OPs to control. One for the voltage and one for the current.

[IrejectYourReality]

I know about the higher dropout and I can cope with that. My heatsink can easily take 50W and will add a fan, just to be sure.

[David Hess]

This will work and I have done it myself but there are some issues with your implementation:

1. Your LM358 is configured with positive feedback.
2. The LM350 (and LM317) expects a minimum output load current whether you actively drive the adjustment pin or not.  If your output is going to go to 0 volts, then this current must be drawn down to the negative bias supply.  This is easy enough to do with a single resistor between the output and adjustment pin which will allow the operational amplifier to sink the constant 1.25V/R current into the negative bias supply but this would defeat your additions for current control.  Fixed regulators like the 7805 do not have this problem because their quiescent current flows out their common (adjust) pin instead of their output pin (their voltage divider is internal) but I would still use the LM350.
3. The reference and regulation errors of the LM350 are being added to other errors.  If feedback is taken from the output of the LM350 instead of its adjustment pin, then the errors and noise from the LM350 will be removed by the external error amplifier.  This adds some additional frequency compensation requirements but nothing complex.

[IrejectYourReality]

Thanks for the feedback. Did you build your supply with current limiting? Or just the voltage regulation?

1) Which one? The current limiting one is straight from uSupply schematics and I tried the other one in circuit simulator and it works as it should...

2) I was going to solve the constant load the same way Dave did, with a LM334 or something similar (I know that LM350 needs about 5-10 mA, not 1mA like LT3080).

3) So, more like this?

[JoeN]

Thanks for the advice!
I was also thinking about making the constant current regulation software-only. Since I can measure voltage and current, I can reduce the voltage to get a set current, but I fear that the regulation might not be fast enough. I plan to use  ATXMEGA32E5-AU running at 32MHz to control the whole thing. Would that be a good idea or a bad one?

Software control is not a good idea - it is slow and resolution is limited. For a lab supply one would want something like a 100 kHz control loop bandwidth and maybe 16-24 Bit of amplitude resolution. SO nothing easy to do in software.

I don't know anything about power supplies, but if you need raw speed and the ability to read a wide (16-24 bit) ADC quickly and make adjustments, doesn't this scream out for a CPLD or something like that?  Small CPLDs are very inexpensive these days and some of them have quite a few pins so you can interface a wide ADC easily and look at its output every single clock cycle.

http://www.mouser.com/ProductDetail/Altera/5M160ZE64C5N

[KerryW]

Note that T1 can only pull the ADJ pin to 0V, so you would have 1.25V out of the supply minimum during current limit.  You need it to pull down to -1.25V.

[IrejectYourReality]

What if I connect the current limiting straight to the input of the other opamp?



(I have a feeling that this is not a very good idea)

[Kleinstein]

Moving the current limit to before the other OP does not change very much. Still the same high gain, but the problem gets only worse because of extra delay of the extra OP involved.

[David Hess]

Thanks for the feedback. Did you build your supply with current limiting? Or just the voltage regulation?

1) Which one? The current limiting one is straight from uSupply schematics and I tried the other one in circuit simulator and it works as it should...

I have designed them with and without adjustable current limiting.  Without adjustable current limiting and maybe with it, the current limiting of the voltage regulator provides protection or I add a fixed current limit protection circuit.

As you have implemented it, the added gain of the transistor will make the current control loop unstable.

2) I was going to solve the constant load the same way Dave did, with a LM334 or something similar (I know that LM350 needs about 5-10 mA, not 1mA like LT3080).

A 100 ohm resistor between the output and adjust pin will draw a constant 12.5 milliamps which then must be handled by the operational amplifier which will be no problem.  Add a PNP emitter follower to the output of the amplifier if necessary and it will not even see the load and the constant current will provide an ideal operating point for the transistor.

3) So, more like this?

IC1A still has its inputs reversed.

What if I connect the current limiting straight to the input of the other opamp?

(I have a feeling that this is not a very good idea)

The first bench power supply I designed worked like that with the current control loop driving the voltage control loop.  It worked but having two operational amplifier stages in series made frequency compensation very difficult and performance of the current control loop was poor.

It is better to combine the outputs of the error amplifiers with a pair of diodes (or elegantly and cleverly two PNP emitter followers that have a high Vbe breakdown voltage) so each can pull the output down independently through the adjust pin.  This works well with the single resistor current sink suggestion above.

[Zero999]

It is basically just a summing amplifier, which will "subtract" the -1.25V. Output of the opamp should be -1.25V to 23.75V, to give 0-25V at the output of the regulator. To get the negative voltage needed for the opamp and -1.25 voltage reference

The 1.25V reference won't exactly match the LM350's reference, which can vary from 1.2V to 1.3V.
http://www.ti.com/lit/ds/symlink/lm350a.pdf

You could use an op-amp to look at the LM350's voltage reference and subtract it from the output.

Note, that you should use a better op-amp than the crappy old µA741 and make sure the R2 to R5 are selected from the same batch, so they're well matched.

[David Hess]

Note, that you should use a better op-amp than the crappy old µA741 ...

There is little reason to use an operational amplifier better than a 741 or 358/324.  Precision will not be improved unless remote sensing is used, low frequency noise will not be improved without a low noise reference, and transient response will not be improved without careful frequency compensation.  The dedicated supplies I have build using a precision OP-27 or LT1007 type of operational amplifier to control an integrated regulator had all three of these features but they are overkill for general purpose bench supply.

[Zero999]

Note, that you should use a better op-amp than the crappy old µA741 ...

There is little reason to use an operational amplifier better than a 741 or 358/324.  Precision will not be improved unless remote sensing is used, low frequency noise will not be improved without a low noise reference, and transient response will not be improved without careful frequency compensation.  The dedicated supplies I have build using a precision OP-27 or LT1007 type of operational amplifier to control an integrated regulator had all three of these features but they are overkill for general purpose bench supply.

The main improvement will be the offset, especially in the schematic I posted that uses 470k resistors will drop a fair voltage due to the relatively high bias currents of the old 741. The values could be reduced to 10k so an old op-amp will give respectable performance.

[IrejectYourReality]

As you have implemented it, the added gain of the transistor will make the current control loop unstable.

What should I change to make it stable?

A 100 ohm resistor between the output and adjust pin will draw a constant 12.5 milliamps which then must be handled by the operational amplifier which will be no problem.  Add a PNP emitter follower to the output of the amplifier if necessary and it will not even see the load and the constant current will provide an ideal operating point for the transistor.

I planned to use the negative voltage inverter to supply the opamp with negative voltage. Wouldn't this mess up the negative voltage, if the opamp had to sink 12.5 mA?

[David Hess]

The main improvement will be the offset, especially in the schematic I posted that uses 470k resistors will drop a fair voltage due to the relatively high bias currents of the old 741. The values could be reduced to 10k so an old op-amp will give respectable performance.

The offset current is what matters in this case because of the deliberately balanced source resistance.  The errors from the 741's high bias current cancel and the offset current only contributes an error of about 5 millivolts typical into 235k which is not so bad compared to 2 millivolts of typical voltage offset error.  The error from the common mode rejection is almost as great and in practice the resistor matching will create the largest error term.

I agree that the 741 is the wrong operational amplifier for such high source resistances but for a different reason;  Tektronix used the 741 in such high impedance circuits with balanced inputs to minimize voltage offset error from the bias currents but at very low bandwidths.  The problem is that the current noise from a 741 is pretty high making it noisy with high source impedances.  The typical choice in the past would be the 308 but nobody makes them anymore.

I wonder what the least expensive modern alternative to the 308 would be.  The OP07 is not really a low input bias current operational amplifier but it is close and inexpensive.  The LT1012, LT1097, and OP97 certainly count but cost twice as much as the multiple sourced OP07.  I used to use the LM11 for this sort of circuit if I needed something better than a 308 but they are also discontinued.  JFET operational amplifiers lack precision.

That is a clever circuit.  Did you come up with that yourself or find it somewhere?

[David Hess]

As you have implemented it, the added gain of the transistor will make the current control loop unstable.

What should I change to make it stable?

The problem with the transistor as shown is that it adds voltage gain to the error amplifier for the current control loop making it much more difficult to frequency compensate.

You have two separate error amplifiers so "or" the outputs together so whichever one is active is the one pulling the LM317 adjustment pin low.  Usually this is done with a pair of diodes but in the past when they were more available like the 2N404A, PNP emitter followers with high Vbe breakdown could be used.  Or use diodes and PNP emitter followers.  I have also seen p-channel JFETs used to do this.  I have not tried it but low Vgs p-channel MOSFETs or p-channel depletion mode MOSFETs could be used also.

Most of the power supply designs I would copy use diodes or diodes plus PNP transistors; the later unloads the output of the operational amplifiers for better precision but this would not matter in a general purpose bench power supply.

A 100 ohm resistor between the output and adjust pin will draw a constant 12.5 milliamps which then must be handled by the operational amplifier which will be no problem.  Add a PNP emitter follower to the output of the amplifier if necessary and it will not even see the load and the constant current will provide an ideal operating point for the transistor.

I planned to use the negative voltage inverter to supply the opamp with negative voltage. Wouldn't this mess up the negative voltage, if the opamp had to sink 12.5 mA?

If you sink the current to the negative supply voltage, then it does not matter whether it goes through the operational amplifier or not except that the design is simpler if the operational amplifier does it.  The problem with sinking it to ground is that your current sink has to operate with very low compliance when the output voltage is low; this might not matter except it seems like you want an output which can actually go all the way to ground.  This is almost possible to do using a bipolar transistor or MOSFET as the current sink but again, it is more complicated than sinking the current to the negative supply.

[IrejectYourReality]

So, if I understand it correctly, it should look like this?



I am still concerned about sinking the current to the negative supply. As I said, I want to use a PWM voltage inverter, which is a very weak supply and can generate only couple of mA. If I sink the 12.5 mA to this supply, isn't it basicaly the same process as if the supply itself sourced equivalent opposite current? I fear, that by sinking too much current into this supply, the negative voltage will drop and the supply will no longer be able to go close to 0V at the output. Is this a valid concern?



Another thing, more for the sake of curiosity, about the full digital control of the constant current, as Kleinstein wrote:

Software control is not a good idea - it is slow and resolution is limited. For a lab supply one would want something like a 100 kHz control loop bandwidth and maybe 16-24 Bit of amplitude resolution. SO nothing easy to do in software.

I was thinking about this. I plan to use an ATXMEGA mcu running at 32 MHz. It has 12 bit DAC capable of 1 Msps and a 12 bit ADC capable of 2 Msps. I do not need more than 12 bits of precision. What if I dedicated this mcu to just running the control loop and nothing else, and handled the rest like reading rotary encoders and writing to an LCD somewhere else? I know this is probably a bad idea, but I am just curious.

[Zero999]

The main improvement will be the offset, especially in the schematic I posted that uses 470k resistors will drop a fair voltage due to the relatively high bias currents of the old 741. The values could be reduced to 10k so an old op-amp will give respectable performance.

The offset current is what matters in this case because of the deliberately balanced source resistance.  The errors from the 741's high bias current cancel and the offset current only contributes an error of about 5 millivolts typical into 235k which is not so bad compared to 2 millivolts of typical voltage offset error.  The error from the common mode rejection is almost as great and in practice the resistor matching will create the largest error term.

I agree that the 741 is the wrong operational amplifier for such high source resistances but for a different reason;  Tektronix used the 741 in such high impedance circuits with balanced inputs to minimize voltage offset error from the bias currents but at very low bandwidths.  The problem is that the current noise from a 741 is pretty high making it noisy with high source impedances.  The typical choice in the past would be the 308 but nobody makes them anymore.

I wonder what the least expensive modern alternative to the 308 would be.  The OP07 is not really a low input bias current operational amplifier but it is close and inexpensive.  The LT1012, LT1097, and OP97 certainly count but cost twice as much as the multiple sourced OP07.  I used to use the LM11 for this sort of circuit if I needed something better than a 308 but they are also discontinued.  JFET operational amplifiers lack precision.

That is a clever circuit.  Did you come up with that yourself or find it somewhere?

I designed it myself, when I noticed many people were using separate voltage references and trimmer resistors to get 0V from the LM317. I realised there's a more effective way to zero the output.

So, if I understand it correctly, it should look like this?

It will still be unstable, I'm afraid. As long as you have another gain stage inside the op-amp's feedback loop (T1 in this case) it will be prone to oscillation.