Linear lab power supply

[JuanGg]

Ideally the the digital to analog ground path should not carry any currents but it's difficult to avoid.
The return current path current for the 8V is low and doesn't fluctuate much at all.
Even a possible HF loop isn't really a big problem because the 100nF capacitor would filter a lot of it out at the CV op-amp.

Ok, I'll do the grounding as suggested. What values should I use for R1 and C1?

I have prototyped the CV loop. After making a couple mistakes I got it working. It regulates from 0 to around 30 V and with no load there is barely any ripple on the output. There is no overshoot at power on. I have only done the analog side, using an LM317 as the 8 V reg and a pot for setting the voltage. As I haven't got the SOIC to DIP adapters yet, I made my own for the op-amp. I also had to parallel and series some resistors to accound for values I haven't got around (some on the way), so resistor values aren't spot on, but within tolerance. I'll attach the power transistors to a heatsink and do further testing.
    Juan

[xavier60]

Ideally the the digital to analog ground path should not carry any currents but it's difficult to avoid.
The return current path current for the 8V is low and doesn't fluctuate much at all.
Even a possible HF loop isn't really a big problem because the 100nF capacitor would filter a lot of it out at the CV op-amp.

Ok, I'll do the grounding as suggested. What values should I use for R1 and C1?

I have prototyped the CV loop. After making a couple mistakes I got it working. It regulates from 0 to around 30 V and with no load there is barely any ripple on the output. There is no overshoot at power on. I have only done the analog side, using an LM317 as the 8 V reg and a pot for setting the voltage. As I haven't got the SOIC to DIP adapters yet, I made my own for the op-amp. I also had to parallel and series some resistors to accound for values I haven't got around (some on the way), so resistor values aren't spot on, but within tolerance. I'll attach the power transistors to a heatsink and do further testing.
    Juan

47Ω and 100nF will be fine.
Good to see it working there. Some tests should be done to make certain that it isn't just working in a marginal way.

The Base of Q2 in the schematic from post #177 should be close to 5V.
A load transient recovery test should give a recovery time of about 15µs with very little overshoot.
What I'm mainly curious about is the ratio of voltage change at the Base of Q2 and output current.
This needs to be done with the load current stepped between 0.1A and 0.5A or higher, at about 50Hz.

[JuanGg]

       47Ω and 100nF will be fine.
Good to see it working there. Some tests should be done to make certain that it isn't just working in a marginal way.

The Base of Q2 in the schematic from post #177 should be close to 5V.
A load transient recovery test should give a recovery time of about 15µs with very little overshoot.
What I'm mainly curious about is the ratio of voltage change at the Base of Q2 and output current.
This needs to be done with the load current stepped between 0.1A and 0.5A or higher, at about 50Hz.   

I'll make those measurements. I have a CC load that I can hook up to my function generator. Testing is to be done at 0.1 A, 50 Hz square wave, then 0.2 A etc right?
What about output voltage, do I test at, say, at 5, 15, 25 V?
    Juan

[xavier60]

       47Ω and 100nF will be fine.
Good to see it working there. Some tests should be done to make certain that it isn't just working in a marginal way.

The Base of Q2 in the schematic from post #177 should be close to 5V.
A load transient recovery test should give a recovery time of about 15µs with very little overshoot.
What I'm mainly curious about is the ratio of voltage change at the Base of Q2 and output current.
This needs to be done with the load current stepped between 0.1A and 0.5A or higher, at about 50Hz.   

I'll make those measurements. I have a CC load that I can hook up to my function generator. Testing is to be done at 0.1 A, 50 Hz square wave, then 0.2 A etc right?
    Juan

The load transient test can be stepped from 0A to 0.5A or up to anything the supply can handle. The duty cycle can be reduced to minimize dissipation.
The other test needs to step from 0.1A because the trans-conductance is non-linear at lower currents. You can even use a ramp current to observe the linearity.  The reason for the 50Hz is so that the capacitor in your DSO doesn't alter the waveforms much if AC coupled.
The voltage doesn't really matter so long as the regulator doesn't drop out of regulation.

[xavier60]

The transfer characteristics starting at the Base of Q2 to the Emitter of the TIP35C is mainly a voltage controlled current source, or transconductance amplifier. A change in voltage at the Base of Q2 causes a change of current at the power supply's output.
The amplification factor or gm is the ratio of voltage change at the Base of Q2 to the current change at the output.
The one main remaining thing that I have not been able to test is how much the gm changes for individual TIP35C transistors.

[JuanGg]

47Ω and 100nF will be fine.
Good to see it working there. Some tests should be done to make certain that it isn't just working in a marginal way.

The Base of Q2 in the schematic from post #177 should be close to 5V.
A load transient recovery test should give a recovery time of about 15µs with very little overshoot.
What I'm mainly curious about is the ratio of voltage change at the Base of Q2 and output current.
This needs to be done with the load current stepped between 0.1A and 0.5A or higher, at about 50Hz.

Attached is a photo of my setup: Function generator is connected to the electronic load, which I can configure to give me a set CC or in relation to the waveform from the function gen. I don't trust this load too much (as I made it). I used the High-res adquisition as the load was injecting a fair ammount of noise. I am monitoring current Q2 base or and output voltage on the scope.
I attached a small heatsink to the TIP35C, and only ran it for a couple seconds each. It got a bit warm (far from too hot to touch) when running it for a minute or so.

All measurements have been done with the power supply set to 20 V.

-Drawing 0.2 A CC or 0.5 A CC, Q2's base remains at 4.54 V.
-Other measurements as attacched.
-Load transients give about 300 uS recovery time (as I see it, see screenshots)
    Juan

[xavier60]

The result of the transconductance test is rather low, looks like a gm of about 7.
The cause of that is actually here. I have just realized that I have been using a Darlington power transistor after all of my fake TIP35C's failed.
Reducing the emitter resistor on Q1 will raise the gm to a more reasonable figure.
This doesn't explain the poor load transient response. The step in the response indicates that the voltage measurement is being taken at some distance from the regulator's output and/or something wrong with the compensating components.
This is my result with a 2N3055 now fitted.


EDIT: change Q1's Emitter resistor to 47Ω

[JuanGg]

The result of the transconductance test is rather low, looks like a gm of about 7.
The cause of that is actually here. I have just realized that I have been using a Darlington power transistor after all of my fake TIP35C's failed.
Reducing the emitter resistor on Q1 will raise the gm to a more reasonable figure.
This doesn't explain the poor load transient response. The step in the response indicates that the voltage measurement is being taken at some distance from the regulator's output and/or something wrong with the compensating components.
This is my result with a 2N3055 now fitted.


EDIT: change Q1's Emitter resistor to 47Ω

I'll change that resistor. The oscilloscope probe was right on the output of the power supply, hooked to the output capacitor. I'll check in case I have forgotten something/used different value capacitors/whatnot.
    Juan

[JuanGg]

Ideally the the digital to analog ground path should not carry any currents but it's difficult to avoid.
The return current path current for the 8V is low and doesn't fluctuate much at all.
Even a possible HF loop isn't really a big problem because the 100nF capacitor would filter a lot of it out at the CV op-amp.

Would a groundplane shaped as attached fulfill the outline in post #250? It just seems cleaner and simpler to me, as I can't possibly take every single ground to the star point, and have to join them before. That would be, of course, keeping the groundplane out of the shunt resistors so main current paths are on the top only.
    Juan

[xavier60]

Ideally the the digital to analog ground path should not carry any currents but it's difficult to avoid.
The return current path current for the 8V is low and doesn't fluctuate much at all.
Even a possible HF loop isn't really a big problem because the 100nF capacitor would filter a lot of it out at the CV op-amp.

Would a groundplane shaped as attached fulfill the outline in post #250? It just seems cleaner and simpler to me, as I can't possibly take every single ground to the star point, and have to join them before. That would be, of course, keeping the groundplane out of the shunt resistors so main current paths are on the top only.
    Juan

Although ground planes allow current paths to share, the cross talk will be minimal because of the ground plane's low impedance.
Just keep a close eye on the fan current path.

[JuanGg]

Fan current path is as short as it can be. I think it'll be fine.
    Juan

[JuanGg]

Here is where I'm at with the PCB. I have included the groundplane, and rearranged things a bit so I don't cut it too much. Also consolidated hole sizes and made higher voltage traces have bigger clearances. The only thing missing is a way to measure current. Maximum drop on the shunt is 0.5 A * 0.025 Ohm = 0.0125 V Would that be too low for an acurate measurement?
EDIT: I bought some INA 196 IC which are high/low side shunt amplifiers. I could use that, and it could maybe added to the CC loop. It has a gain of 20 V/V. They are fairly expensive at $2.45 each in one-off quantity.
I also have one ADC channel left that could be used to detect CV or CC operation.
    Juan

[xavier60]

Here is where I'm at with the PCB. I have included the groundplane, and rearranged things a bit so I don't cut it too much. Also consolidated hole sizes and made higher voltage traces have bigger clearances. The only thing missing is a way to measure current. Maximum drop on the shunt is 0.5 A * 0.025 Ohm = 0.0125 V Would that be too low for an acurate measurement? I also have one ADC channel left that could be used to detect CV or CC operation.
    Juan

You could monitor the CC op-amp's output via a 2:1 resistor divider.
I could be missing something, I have not found a parametric search that helps me find op-amps with wanted combinations of specs.
The only suitable op-amp that I'm aware of is the OPA2192. Actually the same high precision op-amp should be used for the CC op-amp also.
The data PDF for the OPA2192 says "Common-mode (V–) – 0.5 (V+) + 0.5" witch means that it suits the CC job.

 

[xavier60]

Rail to rail auto zeroing op-amps do exist. ISL28134

[xavier60]

The  INA196  has  ±0.5mv typical offset error. That's a ±20ma measurement error. The worst case offset will cause a ±80ma error.
Even with using an op-amp, because we are using single rail, an offset error that wants to cause the output signal to go negative, can't. It is lost signal that can't easily be compensated for.
On the other hand, the OPA2192 has ±5µv offset which gives ±0.2ma error. I think that this might be over kill depending on what resolution is required.
Some time ago I did test the design with a 50mΩ shunt.

ADC inputs usually ignore negative voltage anyways unless small positive offset is intentionally added which still won't help in this case.

[JuanGg]

Ok. What I was aiming for is 0.05 V resolution (can't go any lower with 10 bits ADCs without switching ranges) and 1 mA resolution. I do have 12 big and even 16 bit PWM.
The OPA2192 fortunately has the usual footprint. The thing is that it is fairly expensive (4.5 € each) and it is not available at LCSC, so if I go and buy it at digikey or mouser, shipping would cost a fortune. It isn't that cheap on eBay and who knows if they are fake. I don't mind getting it if absolutely necessary though, although I am in kind of a budget.

Would it be possible to increase the shunt resistor value, so offsets are negligible, maybe switch to high side? Is there any reason to have such small values on shunt resistors for this sort of currents?
I just want to keep the thing simple. Thanks again.
    Juan

[xavier60]

Ok. What I was aiming for is 0.05 V resolution (can't go any lower with 10 bits ADCs without switching ranges) and 1 mA resolution. I do have 12 big and even 16 bit PWM.
The OPA2192 fortunately has the usual footprint. The thing is that it is fairly expensive (4.5 € each) and it is not available at LCSC, so if I go and buy it at digikey or mouser, shipping would cost a fortune. It isn't that cheap on eBay and who knows if they are fake. I don't mind getting it if absolutely necessary though, although I am in kind of a budget.

Would it be possible to increase the shunt resistor value, so offsets are negligible, maybe switch to high side? Is there any reason to have such small values on shunt resistors for this sort of currents?
I just want to keep the thing simple. Thanks again.
    Juan

I could try testing it with 100mΩ but there could be complications. The way the CC op-amp is configured,  it is not a true Miller Integrator. Its minimum gain is one. Increasing the shunt resistor is the same as increasing the  gain.
And the offset error gets improved by  factor of 4.
I bought the OPA2192 from UTsource.
For now, I'm keen on seeing the CV loop working properly.

[xavier60]

It's mostly fine with a 100mΩ shunt,just getting some current undershoot when tested into a dead short, not a practical problem really.
Removing C5, 33pF fixes it.

[JuanGg]

.I could try testing it with 100mΩ but there could be complications. The way the CC op-amp is configured,  it is not a true Miller Integrator. Its minimum gain is one. Increasing the shunt resistor is the same as increasing the  gain.
And the offset error gets improved by  factor of 4.
I bought the OPA2192 from UTsource.
For now, I'm keen on seeing the CV loop working properly. 

.  It's mostly fine with a 100mΩ shunt,just getting some current undershoot when tested into a dead short, not a practical problem really.
Removing C5, 33pF fixes it.

I'll get the op-amp from there, shipping seems cheaper. So, one would be needed for the CC loop and another for current measurement. I could get two dual ones, and use the remaining two for the CV loop and as a buffer for filtered PWM. Or even use the TLC072 for this last two things.

I'll continue testing the CV today. I wasn't able to do so yesterday.

If nothing, by increasing the shunt resistance we are decreasing parts count. However, with the new op-amp it wouldn't be needed right?
Thanks a lot for the effort.
    Juan

[xavier60]

.I could try testing it with 100mΩ but there could be complications. The way the CC op-amp is configured,  it is not a true Miller Integrator. Its minimum gain is one. Increasing the shunt resistor is the same as increasing the  gain.
And the offset error gets improved by  factor of 4.
I bought the OPA2192 from UTsource.
For now, I'm keen on seeing the CV loop working properly. 

.  It's mostly fine with a 100mΩ shunt,just getting some current undershoot when tested into a dead short, not a practical problem really.
Removing C5, 33pF fixes it.

I'll get the op-amp from there, shipping seems cheaper. So, one would be needed for the CC loop and another for current measurement. I could get two dual ones, and use the remaining two for the CV loop and as a buffer for filtered PWM. Or even use the TLC072 for this last two things.

I'll continue testing the CV today. I wasn't able to do so yesterday.

If nothing, by increasing the shunt resistance we are decreasing parts count. However, with the new op-amp it wouldn't be needed right?
Thanks a lot for the effort.
    Juan

It's happy with 250mΩ also. I'm aiming for 4 amps for mine, I keep overlooking  that yours is lower.
So, a lot of options there.  Is there really a need to buffer the references depending on how you are doing the filtering?
You could use the TLC072's and high resistance shunt for now and upgrade to the OPA2192 if it's necessary to go back to the low resistance shunt.

[xavier60]

My earlier 20 amp bench supply has a 10mΩ shunt. Before I replaced the TLC072 CV/CC op-amp with an OP2192, I resorted to trying all of my spare TLC072's in the socket to find the one with the lowest offset.

[Kleinstein]

One usually could compensate for an offset of the OP for the current reading / control in software or if needed in hardware. So I don't think one would really need the rather expensive OPA2192.  If single OPs are used, there would be the option to do offset adjustment with a pot.

Using a 250 mOhms shunt for 4 A would cause quite a lot of heat at the shunt and thus potential drift of shunt value. So it would need a rather high power one.  So the more normal way is to aim for something like 100-200 mV burden voltage, which would be more like 25-50 mOhm for a 4 A range.

[xavier60]

One usually could compensate for an offset of the OP for the current reading / control in software or if needed in hardware. So I don't think one would really need the rather expensive OPA2192.  If single OPs are used, there would be the option to do offset adjustment with a pot.

Using a 250 mOhms shunt for 4 A would cause quite a lot of heat at the shunt and thus potential drift of shunt value. So it would need a rather high power one.  So the more normal way is to aim for something like 100-200 mV burden voltage, which would be more like 25-50 mOhm for a 4 A range.

I think JuanGg is targeting 0.5 amps, so 250mΩ and TLC072's should do fine.
Ill stay with 25mΩ and 4 amps for mine. And because I'm using analog meters, Ill likely use the TLC072 also.

[JuanGg]

Yes, I am after about 0.5 A. I could go a bit higher but not much, given the transformers I have. This PSU would be used for low power stuff. I have a switching PSU that can go up to 5 A  if I need it.

I suppose changing shunt and op-amp would be a matter of resistor values, so the PCB would remain the same. I can go with the TLC072 for now and upgrade later. As I am planning on building two channels, I could even do one of each kind.

I was thinking of using a simple RC low-pass filter.
There is no need to buffer the CV ref, as the op amp has "infinite" input impedance. The CC ref has a rather high (470 k) input impedance, so if using a stiff RC filter shouldn't be a problem. The thing is, if I have one op amp left of the four (CV, CC, Cmeas, remaining one), it could buffer the CC ref instead of being left unused.
How would the current measuring op-amp be connected?
Please correct me if I'm wrong.
    Juan

[xavier60]

Yes, I am after about 0.5 A. I could go a bit higher but not much, given the transformers I have. This PSU would be used for low power stuff. I have a switching PSU that can go up to 5 A  if I need it.

I suppose changing shunt and op-amp would be a matter of resistor values, so the PCB would remain the same. I can go with the TLC072 for now and upgrade later. As I am planning on building two channels, I could even do one of each kind.

I was thinking of using a simple RC low-pass filter.
There is no need to buffer the CV ref, as the op amp has "infinite" input impedance. The CC ref has a rather high (470 k) input impedance, so if using a stiff RC filter shouldn't be a problem. The thing is, if I have one op amp left of the four (CV, CC, Cmeas, remaining one), it could buffer the CC ref instead of being left unused.
How would the current measuring op-amp be connected?
Please correct me if I'm wrong.
    Juan

The left over op-amp is inconvenient. If it's used for the CC buffer, there will be more offset uncertainty. It could be used for the voltage reference to separate 2  RC filters so that the same value R and C can be used for both stages.
The current measurement op-amp would have to be connected as a inverting amplifier because of the negative going shunt voltage.
The CC op-amp already has a 10MΩ resistor connected to the inverting input to make sure it regulates to zero. You will need to add a settable amount of offset to the CC reference in software.
You can do the same sort of thing with the current measurement op-amp so that it has a small positive offset when the shunt current is zero.
Then hope they don't drift much.

Edit: Just realized that the spare op-amp would be buffering the high level CC reference signal and the offset would have little affect. I think.