Author Topic: Is 30mV still acceptable as the lowest voltage for a linear power supply?  (Read 3856 times)

0 Members and 1 Guest are viewing this topic.

Offline bloguetronicaTopic starter

  • Frequent Contributor
  • **
  • !
  • Posts: 354
  • Country: pt
Hi,

This is more a matter of opinion, based on your personal experience. The attached circuit can only achieve an output voltage of 30mV at its lowest (even of the output of the DAC is at absolute 0V). I could solve this by using a different topology (4:1 amplifier and 1:1 error correction, instead of follower and 4:1 correction), or perhaps by driving the output of the first op-amp to negative via a 10K resistor with a charge pump. However, both these solutions will increase noise, with the risk of losing precision as well, and IMHO it is not essential to be able to achieve such low voltage and more precision for the full range is preferable.

What is your opinion?

Kind regards, Samuel Lourenço
« Last Edit: March 12, 2019, 06:24:34 pm by bloguetronica »
 

Online ejeffrey

  • Super Contributor
  • ***
  • Posts: 4032
  • Country: us
For most applications 30 mV is more than sufficient.

It is typical to want lab supplies to have floating outputs.  In this type of regulator that would be accomplished by having the 24 V input from a floating supply.  In that case it is easy enough to slightly raise the "ground" level to allow tuning down to zero volts.  However, I notice that you have SPI pins.  If your supply will be externally programmable you would then need to isolate the communications interface somehow.
 

Offline bloguetronicaTopic starter

  • Frequent Contributor
  • **
  • !
  • Posts: 354
  • Country: pt
Thanks Jeffrey,

The power supply will be isolated, sure. This is just a test module to see if the concept is feasible. Another issue I'm having is a noise of about 2mVpp at the maximum load.

Kind regards, Samuel Lourenço
 

Online David Hess

  • Super Contributor
  • ***
  • Posts: 17427
  • Country: us
  • DavidH
One of the amusing things about the design of the old Tektronix PS501 and PS503 power supplies is that they deliberately included an offset in the voltage error amplifier which is slightly greater than the maximum offset voltage of the 301a or 741 operational amplifier.  Since the output has a pull-down to the negative supply, this allows the output to actually go up to 10 millivolts below 0 volts.  Without the deliberate offset, the minimum output voltage could be higher than 0 volts depending on the input offset voltage of the operational amplifier.

I would consider operating down to 300 millivolts acceptable for general use.  But for a source-meter application, I would want it to support operation down to zero volts or even negative.
 

Offline floobydust

  • Super Contributor
  • ***
  • Posts: 7678
  • Country: ca
30mV is a problem if that is the minimum output voltage when in CC mode, if I remember your other design that has CC.
You short the output and CC activates but only takes the output down to say 30mV and lots of unexpected current flows.

That happened on one of my designs, I'd set it for 100mA and all was fine until I shorted the output and I got 3A  :o
Took me a while to figure that out. The problem was CC could not take the output to zero volts and a small offset remained and so (pass transistor) current-limiting took over.
 

Online David Hess

  • Super Contributor
  • ***
  • Posts: 17427
  • Country: us
  • DavidH
30mV is a problem if that is the minimum output voltage when in CC mode, if I remember your other design that has CC.
You short the output and CC activates but only takes the output down to say 30mV and lots of unexpected current flows.

That is an interesting point but most CV/CC designs have a separate and lower (below zero) minimum output voltages in constant current mode.
 

Offline bloguetronicaTopic starter

  • Frequent Contributor
  • **
  • !
  • Posts: 354
  • Country: pt
30mV is a problem if that is the minimum output voltage when in CC mode, if I remember your other design that has CC.
You short the output and CC activates but only takes the output down to say 30mV and lots of unexpected current flows.

That happened on one of my designs, I'd set it for 100mA and all was fine until I shorted the output and I got 3A  :o
Took me a while to figure that out. The problem was CC could not take the output to zero volts and a small offset remained and so (pass transistor) current-limiting took over.
The only design that has CC, and that I have, was an experimental monster that didn't worked out. The situation you are describing doesn't happen with this one (and would not happen with the CC design that I had). The current is always limited to around 260mA.

Anyways, the problem you describe is unrelated to the matter. Suppose that this PSU could output 0V but suffered from that problem. It would be a problem to set it to 30mV then.

Kind regards, Samuel Lourenço
« Last Edit: March 13, 2019, 10:18:34 am by bloguetronica »
 

Offline SiliconWizard

  • Super Contributor
  • ***
  • Posts: 15795
  • Country: fr
I see very little use of a power supply under 30mV. By that I mean, a real power supply. Meant to power some circuit. So I wouldn't bother.

Now if you think about a voltage source in general, for instance used to provide some voltage reference, I guess that could find some use. I wouldn't call it a power supply though, and wouldn't market it as such. And I'm sure you could then use a different design and settle for much less output current, so you would address a different area. Just my opinion though.

 

Offline bloguetronicaTopic starter

  • Frequent Contributor
  • **
  • !
  • Posts: 354
  • Country: pt
Thanks!

This is for a precision power supply. IMHO, I can get away without the low low range of 30mV downwards. Probably it is better not to sacrifice noise and precision by changing the topology.

Kind regards, Samuel Lourenço
 

Offline bloguetronicaTopic starter

  • Frequent Contributor
  • **
  • !
  • Posts: 354
  • Country: pt
Hi,

I've decided to do some experiments by adding a charge pump to generate the negative drive voltage for the op-amp. This solution should be noisier, but in reality I hope to see no difference.

I've chose the TPS60403 because of its fixed and relatively high switching frequency. The 250KHz noise is definitely easier to clean. Also, at 5mA, the ripple is only 15mVpp according to the datasheet. In this application, be load current will be situated between 0.23mA and 0.45mA, and thus the ripple will be probably less. Plus, the 22K resistor causes the charge pump to have almost no influence on the output of the first op-amp.

Lets wait and see! Once the new board arrives I will be able evaluate if this was a good or bad decision.

P.S.: There is an interesting application note explaining the solution I've used. It works with the OPAx180 family as well. See it here: http://www.ti.com/lit/ug/tidu018/tidu018.pdf.

kind regards, Samuel Lourenço
« Last Edit: March 19, 2019, 01:19:25 pm by bloguetronica »
 

Offline bloguetronicaTopic starter

  • Frequent Contributor
  • **
  • !
  • Posts: 354
  • Country: pt
Hi,

I just wanted to let you know that I've assembled the new module, and now it works beautifully! When the module's output voltage is set to 0V, it now reads a residual voltage of 16uV instead of 30mV. This is much better than before. Pulling the output of the first half of the OPA2180 worked.

I should add that I've measured the output of the charge pump, and observed little noise. Probably due to the 22K resistor, the charge pump does not seem to have an appreciable contribution to noise at that point. The input of the charge pump is filtered, so its switching has little influence over the 5V rail (confirmed that too).

Anyway, I'll post more photos as soon as possible.

Kind regards, Samuel Lourenço
« Last Edit: April 10, 2019, 12:17:55 pm by bloguetronica »
 

Offline duak

  • Super Contributor
  • ***
  • Posts: 1048
  • Country: ca
I recently found a problem with a circuit using an LM358 operated with its inputs and outputs operating near 0 V.  When connected to a PWM servo, it would generate a large rail-to-rail sawtooth output when it should be near zero.  With non zero input voltages, it worked fine.  Turned out that the general common noise noise (100 mVp-p, 16 kHz)  coupled to the inputs, outputs and power supply of the LM358 appeared to cause an internal stage to wind up and produce the output.  Improving the signal input and power supply filtering as well as adding series resistors on the outputs solved the problem.  The LM358 is an old part with some unusual characteristics that newer parts probably don't have. 

I'm only mentioning this because a lab type supply might be used in a noisy environment and be expected to provide a zero output.  This circuit uses a negative supply, so it's probably OK.  For a simple test, try injecting a square wave back into the supply through a capacitor while varying the output voltage to see what happens.

Cheers,
 

Online David Hess

  • Super Contributor
  • ***
  • Posts: 17427
  • Country: us
  • DavidH
Another way this is sometimes done is to place a diode in series with the negative power input (for a positive regulator) to provide about -0.6 volts of bias.  The disadvantage of doing this is that it requires either the input or output to float.

I recently found a problem with a circuit using an LM358 operated with its inputs and outputs operating near 0 V.  When connected to a PWM servo, it would generate a large rail-to-rail sawtooth output when it should be near zero.  With non zero input voltages, it worked fine.  Turned out that the general common noise noise (100 mVp-p, 16 kHz)  coupled to the inputs, outputs and power supply of the LM358 appeared to cause an internal stage to wind up and produce the output.  Improving the signal input and power supply filtering as well as adding series resistors on the outputs solved the problem.  The LM358 is an old part with some unusual characteristics that newer parts probably don't have.

The PNP emitter follower output of the 358/324 can only actively sink current down to about 0.6 volts.  Below that, a 50 microamp NPN current mirror operates down to the Vce(sat) of the NPN transistor.  What I suspect you saw was saturation of the 358/324 when the output could no longer sink enough current to maintain feedback; like most operational amplifiers, the 358/324 lacks protection against saturation which is something which distinguishes operational amplifiers from most comparators. (1)

Improved replacements for the 358/324 like the LT1013/LT1014 and of course rail-to-rail output parts can actively drive their outputs all the way to saturation.

(1) The comparator version of the 358/324 is the 393/339 and also lacks saturation protection which is unusual for a comparator but not so much for a slow one.  Faster comparators, even discrete ones, always include protection against saturation.
 

Offline bloguetronicaTopic starter

  • Frequent Contributor
  • **
  • !
  • Posts: 354
  • Country: pt
I recently found a problem with a circuit using an LM358 operated with its inputs and outputs operating near 0 V.  When connected to a PWM servo, it would generate a large rail-to-rail sawtooth output when it should be near zero.  With non zero input voltages, it worked fine.  Turned out that the general common noise noise (100 mVp-p, 16 kHz)  coupled to the inputs, outputs and power supply of the LM358 appeared to cause an internal stage to wind up and produce the output.  Improving the signal input and power supply filtering as well as adding series resistors on the outputs solved the problem.  The LM358 is an old part with some unusual characteristics that newer parts probably don't have. 

I'm only mentioning this because a lab type supply might be used in a noisy environment and be expected to provide a zero output.  This circuit uses a negative supply, so it's probably OK.  For a simple test, try injecting a square wave back into the supply through a capacitor while varying the output voltage to see what happens.

Cheers,
Hi,

I should mention that this is not a two-quadrant power supply. It is single quadrant, and it is only able to source, and not sink, current. Therefore, if you inject a square wave signal while its output is at 0V, that square wave will appear.

Note that the negative "supply" that this module uses is just for biasing the output of the first half of the op-amp. It does not drive the output whatsoever.

Kind regards, Samuel Lourenço
« Last Edit: April 10, 2019, 07:21:18 pm by bloguetronica »
 

Offline bloguetronicaTopic starter

  • Frequent Contributor
  • **
  • !
  • Posts: 354
  • Country: pt
Hi all,

Did some tests without load. Here are the measurements with the output voltage of the module set to 5, 10, 20, 50, 100, 200 and 500mV. This module is surprisingly precise. Or I was lucky with the components that I've got (resistors, DAC and op-amp). I see no appreciable offset at all!

Kind regards, Samuel Lourenço
 

Offline duak

  • Super Contributor
  • ***
  • Posts: 1048
  • Country: ca
Samuel, you have designed and built a very good power supply with some interesting specifications.  You have used a single supply DAC and a single supply precision autozeroing opamp and expect it to give repeatable results.  I know that your supply is not two quadrant and cannot accept current from the load.  What I propose is a simple experiment to see what happens when pulses are injected into the supply at low voltage settings.  I expect that it will be fine.  However, it may act strangely.

The LM358 circuit I mentioned above is a friend's that misbehaved in an unexpected way when he connected to a purchased PWM servo.  The conducted EMI from the servo was enough to cause a problem.  He is using it now so I can't try different opamps in it.   I wouldn't be surprised if a rail to rail opamp shows anomalous operation with its inputs and output at 0 V and some high frequency noise injected into its various ports.

Another little story: in the 80's I developed a PWM servo using MOSFETs for the switching devices.  I used LEM Hall effect current sensors to measure the motor currents instead of series sense resistors.  The LEM sensors were specified to better than 1% accuracy and stability.  On the bench with a DC or AC test current they were perfect.  In the servo, they had an unpredictable error caused by the high frequency noise on the load current.  Some sensors were better than others.  I found that if I added a 1n0 capacitor inside the LEM sensor, I could eliminate the problem.  LEM analyzed the problem and produced a shielded version that also worked.  Later on, I found that LEM replaced the LM301 opamp with an OP37 opamp in an unshielded case and it worked fine.  I will also say that some 7912 negative three terminal regulators used in the servo would have unstable output voltages when the PWM was on.  The regulators were from one manufacturer but we didn't have a long enough product run to nail it down - they were just replaced.  That part worked fine everywhere else; some just had a problem in a high EMI environment.

Cheers,
 
The following users thanked this post: bloguetronica

Offline bloguetronicaTopic starter

  • Frequent Contributor
  • **
  • !
  • Posts: 354
  • Country: pt
Samuel, you have designed and built a very good power supply with some interesting specifications.  You have used a single supply DAC and a single supply precision autozeroing opamp and expect it to give repeatable results.  I know that your supply is not two quadrant and cannot accept current from the load.  What I propose is a simple experiment to see what happens when pulses are injected into the supply at low voltage settings.  I expect that it will be fine.  However, it may act strangely.

The LM358 circuit I mentioned above is a friend's that misbehaved in an unexpected way when he connected to a purchased PWM servo.  The conducted EMI from the servo was enough to cause a problem.  He is using it now so I can't try different opamps in it.   I wouldn't be surprised if a rail to rail opamp shows anomalous operation with its inputs and output at 0 V and some high frequency noise injected into its various ports.

Another little story: in the 80's I developed a PWM servo using MOSFETs for the switching devices.  I used LEM Hall effect current sensors to measure the motor currents instead of series sense resistors.  The LEM sensors were specified to better than 1% accuracy and stability.  On the bench with a DC or AC test current they were perfect.  In the servo, they had an unpredictable error caused by the high frequency noise on the load current.  Some sensors were better than others.  I found that if I added a 1n0 capacitor inside the LEM sensor, I could eliminate the problem.  LEM analyzed the problem and produced a shielded version that also worked.  Later on, I found that LEM replaced the LM301 opamp with an OP37 opamp in an unshielded case and it worked fine.  I will also say that some 7912 negative three terminal regulators used in the servo would have unstable output voltages when the PWM was on.  The regulators were from one manufacturer but we didn't have a long enough product run to nail it down - they were just replaced.  That part worked fine everywhere else; some just had a problem in a high EMI environment.

Cheers,
Hi duak,

I'm guessing that this power supply might be unstable when dealing with the residual currents that are found in many SMPS. I've had a weird experience with a power supply what used a similar topology as this one, and had some weird fluctuations when it was connected to a constant current load that was being supplied by a SMPS. See the last post of this thread:
https://www.eevblog.com/forum/projects/usb-controlled-precision-power-supply-(or-voltage-reference)/msg2007782/#msg2007782

However, I think this is an issue with the opamp that was used there. This supply doesn't show that susceptibility.

On another note, what voltages do you have in mind for the test (output and square wave voltages)? I don't wish to fry the signal generator, BTW.

Kind regards, Samuel Lourenço
 

Online David Hess

  • Super Contributor
  • ***
  • Posts: 17427
  • Country: us
  • DavidH
I will also say that some 7912 negative three terminal regulators used in the servo would have unstable output voltages when the PWM was on.  The regulators were from one manufacturer but we didn't have a long enough product run to nail it down - they were just replaced.  That part worked fine everywhere else; some just had a problem in a high EMI environment.

Negative regulators built on the common NPN process have a common emitter output so they have special stability requirements which are reflected in the datasheets always showing an output capacitor with a minimum ESR.  They have more design, process, and application variations in performance.
 

Offline duak

  • Super Contributor
  • ***
  • Posts: 1048
  • Country: ca
Samuel, to inject a signal into the power supply output, I would use a 10n capacitor in series with a 47R resistor from a square wave generator set to 1 kHz 1 Vp-p.  The impedance of the generator and the RC network will limit current and prevent any DC current from entering the output the generator.  The power supply output capacitance will attenuate the signal so you will have to observe the output with a 'scope to see the actual signal and if it triggers something unusual.  I recomend 100 mVp-p as a target measured on the power supply because that was the voltage that I observed on the LM358 circuit.

Regarding the other power supply, I don't understand the test conditions and what is connected together and how.

Cheers,
 

Offline bloguetronicaTopic starter

  • Frequent Contributor
  • **
  • !
  • Posts: 354
  • Country: pt
Hi duak,

Setting the supply to 500mV, would be good enough for the test? Anyway, what it is to be expected? I'll test this as soon as I can, and post the results.

Meanwhile, here are some more photos of the tests done so far. Voltage was set to 1, 2, 4, 8 and 16V. Precise as usual. I'm surprised!

Kind regards, Samuel Lourenço
 

Offline babysitter

  • Frequent Contributor
  • **
  • Posts: 899
  • Country: de
  • pushing silicon at work
They just suggest one of the "advanced" tests that separate a good lab power supply from a medium.

(I vaguely remember this being a suggestion of Bob Pease?)

Also test your supply with different loads, as in various R L C combinations. This might catch hints at the regulator getting a oscillator at certain situations. And attach a scope and use a MOSFET and pulse gen as a pulsed load, could also cause some ringing under just the right circumstances. Either there are none, or you get insight to harden against it, or you learn what to prevent.

Its preferrable to know about misbehaviour in advance instead of running into it unprepared.



I'm not a feature, I'm a bug! ARC DG3HDA
 

Offline duak

  • Super Contributor
  • ***
  • Posts: 1048
  • Country: ca
Samuel, I suggest you vary your power supply voltage from 0 V to 100 mV while injecting the AC signal.  If there is a problem it is most likely to occur with the lowest digital codes into the DAC, eg. 0, 1, 2 etc.  I see the reference voltage is 4.096 V and that the DAC is 14 bits.  This gives 16384 steps of about 250 uV each.

The best way to observe a possible problem is with an oscilloscope.  A voltmeter may not show anything unusual because it is designed to filter noise.  However, it may show a different voltage than you expect but if the AC signal is disconnected you get the correct voltage.   The LM358 circuit I mentioned above was built by a friend to connect an Arduino to a servo amplifier that drove a steering wheel for a racing simulator.  He said it worked sort of OK in the corners but it was impossible to drive straight ahead.

Cheers,
 

Offline bloguetronicaTopic starter

  • Frequent Contributor
  • **
  • !
  • Posts: 354
  • Country: pt
Hi,

I haven't made the tests as you suggested yet, as I'm yet to find a suitable function generator. However, I did tests with that problematic CC load at 200mA, and saw no problems at all, even considering that the load was not grounded to earth. This seems to be very promising. I have two hypothesis:
- The op-amp is impervious to the relatively HV currents that are generated by the SPMS feeding the CC load;
- The relatively high resistance resistors are aiding to filter such noise.

I see a 20mVpp noise at the output when feeding the load, though. It is not an oscillation, as it is irregular. This issue doesn't manifest when testing the module with passive resistive load.

kind regards, Samuel Lourenço
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf