controlling linear regulators from DACs: any potential issues?

[exe]

Hi :)

I'm trying to design my own bench power supply. I decided to use a discrete regulator (LT1575) because it has very good characteristics. And I want to "program" it digitally, e.g., from a DAC. But it does not have an input for voltage reference, it has only feedback pin (FB). So I adopted this scheme to control it via FB: http://www.microchip.com/forums/m688260.aspx (I also attached the actual scheme of what I'm going to implement just in case). What I'm worried about is that with DAC and its output buffer (opamp) we now have two loops:

1) one that stabilizes PSU output: FB connected to the output via resistor divider.
2) DAC output buffer that also connected to the output via the divider.

My question is: do I sacrifice any performance by controlling the regulator not via a pot, but via DAC? Any potential stability issues?

PS I'm also filtering DAC output with an RC-filter. This is just in case (like, bad routing, bad grounds separation, etc), I don't really know if I need this or not.
PPS I'm not using digipots because they have quite limited number of steps. And my stm32 MCU has built-in DAC.

[macboy]

You are already using a pass transistor/MOSFET and an opamp, so why bother with the regulator chip? Try something like attached. This goes from 0 - 30 V for a DAC output of 0 - 5 V. The accuracy of the output depends entirely on the accuracy of the DAC, so if the DAC output drifts a lot with changes in temperature, etc. then so will the output of this circuit (or any other controlled by that DAC). R5 is the load. You will need suitable voltage rails for the opamp.
(Ignore the randomly chosen MOSFET part number, use a suitable one. same for opamp).

[exe]

Thanks!

Unfortunately, I spent 3 months in TI-TINA and LTSpice trying to create "an ideal bench PSU" and like 8 moths studying the theory of regulators (I don't have electronics education, it's a hobby). And failed to achieve the performance of LT1575 (load regulation, PSRR, minimum input capacity, capacitive load tolerance, stability, etc). That's why it is here.

May be I'm just too obsessed with an idea of designing a PSU that is "in par with serious bench supplies"... But LT1575 cost like ... four pounds?

Concerning your scheme. Thanks for sending it. I'll try it to play with it in LTSpice. At first glance there is no bias for driving mosfet? Why we need Q1? What if I use opamp directly? There are "capacitor-friendly" opamps (though they still do a bit of overshooting).

[bktemp]

Why we need Q1? What if I use opamp directly?

It acts as a level shifter. Most opamps only work upto 30V, so the maximum possible output voltage is limited to 25V when the gate is driven directly by the opamp.

[macboy]

Concerning your scheme. Thanks for sending it. I'll try it to play with it in LTSpice. At first glance there is no bias for driving mosfet? Why we need Q1? What if I use opamp directly? There are "capacitor-friendly" opamps (though they still do a bit of overshooting).

The gate is driven from the input voltage through R1, which is not ideal (it limits the output voltage due to Vgs threshold). You could do better by having a bias supply higher than the input and driving it from that. When Q1 conducts, it draws current through R1 which reduces the gate voltage, reducing the output voltage.

If you are willing to have your "ground" (which is also the ground for the micro with the DAC) at the postive terminal instead of the negative one, then the following is better, in my opinion. The MOSFET can easily be replaced by a darlington BJT (in either case).

[macboy]

Note that the performance is very highly dependent on the op-amp used, and the LT1001 I chose is not ideal. The capacitor C1 is there for stability, and its value is dependent on the op-amp and the pass transistor stage, and should be tweaked for good dynamic response and stability.

[exe]

The capacitor C1 is there for stability, and its value is dependent on the op-amp and the pass transistor stage, and should be tweaked for good dynamic response and stability.

Thanks you for your time and effort. I tried to build a MOSFET driver from scratch and the best I achieved is this: (see attachments). As we can see, LT1575 does the job considerably better. It is also MUCH friendlier to capacitive load, my scheme will oscillate if the output capacitor is big-enough and has low ESR. I didn't figure out how to stabilize it without sacrificing performance.

Now, the problem is: designing a compensation network for an opamp is extremely difficult for me. I read tons of literature just to conclude that I'm not good-enogh to do so for a wide-range of input capacitance. I played month in TINA-TI meditating on bode plots and trying to achieve "good response and safe phase and gain margins". But one single chip (LT1575) just works and all component values are in data sheets .

That's why I'm trying to figure out how to build a PSU around LT1575.

[exe]

(bigger scheme screen shot  just in case)

[Kleinstein]

The regulator chips are usally made to drive resonable well behaved and defined loads. In this case they can get very good and fast regulation.  The difficult part with a bench supply is, that one usually wants it to be stable with essentially any resonable load. This is usually only possible with a slower regulator.
The commercial lab supplies are also muchslower than the LT1575, for this reason.
There may be a chance to use the LT1575, as you can set the compensation externaly and thus make it slow enough. But than no real need for such a fast chip.

I think daves supply design using a LT3080 or similar failed because he also wanted to much and used a regulator IC.

The second point is that a bend supply usually should have a varaiable current limiting / current regulation. Most regulator chips don't have this. Adding current limiting is not that simple and can cause quite some problems if the regulator does not provide for it.

Using a Mosfet as a power device seem attrective, but MOSFETs are relatively nonlinear and slow at low currents. So you may need quite some standing current on the MOSFET. BJT's might be easier.

[exe]

to be stable with essentially any reasonable load.

Yes, these all are true, but, according to the simulator, LT1575 "just works". I don't know if the simulation is accurate though...
It seems they really put an effort to make it stable driving capacitive load. Probably because it supposed to drive CPUs where load may change dramatically in tens of ms.
From what I understand, they did the following. They use opamps with  GBW 70MHz and 1MHz crossover frequency. The compensation network was designed experimentally (http://cds.linear.com/docs/en/application-note/an69f.pdf) and it was checked to provide performance goals. In that application note they also cover stability issues (like board layout considerations, recommended MOSFETs, how to slow down the loop, what to look for when checking, design margins, etc). So, it looks like an interesting chip to play with, doesn't it?

The second point is that a bend supply usually should have a varaiable current limiting / current regulation. Most regulator chips don't have this. Adding current limiting is not that simple and can cause quite some problems if the regulator does not provide for it.

Yeah, I got issues with this too. So I added another pass element, MOSFET, that does just this (please see attachment). So voltage drop shouldn't be a problem, though it is not very fast to respond when saturated...

Using a Mosfet as a power device seem attrective, but MOSFETs are relatively nonlinear and slow at low currents. So you may need quite some standing current on the MOSFET. BJT's might be easier.

Yeah, absolutely. I'm using MOSFETs because I'm obsessed with the idea of low dropout regulators. So my goal is to achieve a dropout less than 1V (with a buck pre-regulator of course). This is why MOSFET. But I understand I may not get what I want. I don't expect the design to work well in the first revision (but I have a hope )

[Kleinstein]

The LT1575 allready has provisions for current regulation / limiting inside. So I would use this, and not a second independent regulator. At crossover the two seperate regulators have a hard time to work together. The internal curretn regulator should work resonably good.

Within the limited voltage range (max. about 16 V at output) the LT1575 might be good and one can allways slow the regulator down via the compensation circuit. At high speed layout gets really difficult - the application note allready mentions inductance of the MOSFET case. So you may wan't to include some of these parasitics in the simulations. Also don't assume ideal caps.

The usuall lab power supplies you can buy are much slower, than the CPU supplies. To get the supply stable with varaiable load a lower speed is likely needed and possibly an additional speedup capacitor in the feedback path. I usually also helps to have the output capacitor not just as a single cap, but a combination of a very low ESR one and one with some ESR (e.g. 10-50 mOhms range with a fast regulator).

With a mosfet output it helps to have some kind of minimum load in the form of a constant current sink - this makes compensation less dependent on the current level. Idealy one would have a kind of push pull outout with quite some bias, but this will interfere with the current regulation of the chip.

[exe]

The internal current regulator should work reasonably good.

Thanks for suggestion! Unfortunately, I didn't figure out how to implement CC mode or mixed CC/CV mode (when it is CV and if the current is too high it does CC). The built-in current limit is more like to cut off the output in case of a short circuit. With separate CC and CV regulators this supposed to be easy to implement, though I haven't tested this configuration hard in LTSpice.

PS You are right, non-zero ESR can dramatically change the picture. I'm also trying to take ESL into account, though my approach is just to a small cap (0.1uF) in parallel to other caps. I'm going to use murata caps and I use their online simulator to see |Z| vs freq.

> constant current sink

Thanks, will do! Hope to use one big heatsink for all power mosfets in the scheme, if layout allows...

[Kleinstein]

The LT1575 has provisions for a CV/CC mode. It just has the additional timer for a shutdown after some time. One can disable the timer by pulling the shutdown pin low, instead of having the delay cap there. Current from that pin could also be used to signal CV mode is active: e.g. have a NPN transistor with the base to the shutdown pin and emitter to GND drive a LED.

CC mode might not be very low noise, because of the rather low voltage drop. So if a large current range is needed switchung the shunt might help.

For the output capacitor one could aim for something like 1 µF of very low ESR and maybe 5 µF with a slight ESR value (e.g. 50 mOhm).  How low one can go depends on the layout - so measurements will tell. Simulations are more like a first guess and to identify which is the most critical load/current. Its also import to have a shunt with low inductance and have the fiter caps for the preregulated voltage close by.

For a MOSFET like BUZ10, I would suggest somthing like 10-50 mA as a minimum current - you can't speed up very much thrugh the current, because the might always be nasty loads that can absorb the current after transients. In this cases a fast compensation setting could cause large signal oscillations, despite the regulator being stable in small signal case.

[exe]

The LT1575 has provisions for a CV/CC mode.

I didn't figure out how to make current limit settable. The built-in circuit just kicks in when INEG pin is 50mV or above... There should a programmable amplifier or something... Do you have any ideas?

if a large current range is needed switching the shunt might help.

I'm going this way. But how can I switch it? My initial intention was just to use a panel switch, but I'm afraid any long wires would work as antennas and could potetially cause problems... Should I use a relay?

For a MOSFET like BUZ10, I would suggest somthing like 10-50 mA as a minimum current

I think I will go for 100 or even 200mA I think I should have stated the desired parameters of the PSU. 1-18V range, 0-4A current limit. So, 0.2A is not a big deal.

[Kleinstein]

18 V is already quite high, as the LT1575 must deliver the output voltage plus gate source voltage.

There are several ways to adjust the current limit: one can add a series resistor to one (lookup in DS which one) of the sense lines this increases current. One can also add a voltage between shunt and chip. This way the drop can be even set to less than 50 mV for currents at the low end.

Switching the shunt might in deed be tricky if long cables are used. On might be able to use a MOSFET parallel to the larger shunt, and switch the sense voltage to. Very small currents (e.g. less than 10 mA) may be tricky with the shunt on the drain side of the MOSFET.

[exe]

18 V is already quite high, as the LT1575 must deliver the output voltage plus gate source voltage.

One can also add a voltage between shunt and chip. This way the drop can be even set to less than 50 mV for currents at the low end.

Switching the shunt might in deed be tricky if long cables are used. On might be able to use a MOSFET parallel to the larger shunt, and switch the sense voltage to. Very small currents (e.g. less than 10 mA) may be tricky with the shunt on the drain side of the MOSFET.

Thanks for your help, adding a voltage between shunt and chip sounds interesting, but I cannot figure out how to do this (other than separate isolated voltage source, see attachment). Could you please give a hint?

On might be able to use a MOSFET parallel to the larger shunt, and switch the sense voltage to. Very small currents (e.g. less than 10 mA) may be tricky with the shunt on the drain side of the MOSFET.

Mmm, I'm a bit afraid Rds(on) is close to the shunt value... But I'm going to use lt6105 which has programmable (via an external resistor) amplification. Since the resistors are of the order of kilo-ohms this way it should work more predictable.. What do you think?

[Kleinstein]

The LT6105, like most high side current sense amplfiers is quite slow. So it won't work with the LT1575. Also interfacing can get tricky due to variable DC levels. So I think this chip will not really help for regulation - it might be OK for dispaly purpose.

Adding a voltage can be something like a resistor (e.g. 100 Ohms) in the positive side line and a adjstable current source.
The tricky thing is just that the chip itself allready has the 50 mV inside, so stability at low levels will not be very good, as you substract from the internal value.

MOSFET R_on can be comparable to shunt resistance - so you need more than one switching FET. The R_on will than only appear in the voltage lost, but not in the measured voltage.