Power supply for home lab - do I really need a R&S?

[mhsprang]

Thanks to everyone for all the analysis and comparison.  As I mentioned above, I've already communicated all of this to the power supply team in Munich and we are investigating.  It would also be very helpful if the OP could contact our technical support about this and/or contact their R&S sales representative as well.

One (potentially ignorant) question:  has the behavior of the NGE (or other manufacturers' supplies) been outside of its published specifications? 

I've seen lots of test results for lots of supplies (and again, many thanks for those), but I haven't seen any comparisons of experimental results to published specifications.  That would have been one of the first things I checked - is a given supply "in spec"? 

My apologies if that's a dumb question.

Yes, I sent an email to R&S support yesterday about this.

And yes, I feel this is outside the specifications for exactly the reasons ArdWar just stated in his reply #73, just above.

[blackdog]

Hi, :-)

KungFuJosh asked for some explanation of what scope probe I was using, which can be seen below in the picture.
Cost about 1850$ *grin*
The comment above is a hint to the major measuring instrument manufacturers who regularly charge outrageous prices for all kinds of probes.

Probing setup
The tests I did with current limiting around 10mA required a 50 Ohm resistor and about 1-Watt.
I used 2x a 100 Ohm resistor in parallel with a 1M Banana Crock cable from GW Instek and to the scope I used one 1M BNC to Crock cable.
One and all can be seen in the photo below.

I have been thinking about running the measurement setup “HF”.
But on reflection this was not really necessary, this because the LAB power supplies tested have between 100uF and 680uF across the terminals.
This does not allow fast rise times in the measurements.

I use two already older scope probes on my test benches from Micsig for commonmode measurements and current measurements, which are the DP10007 and the CD2100B.
For my work these probes are satisfactory and they are very affordable.
But Micsig also has faster probes available and still affordable prices.
In my measurements, however, these Micsig probes have not been necessary, K.I.S. was a good mindset here, so i used basic cables.

First the pictures


The used probe

.

Scoop setup
In this photo, I left the settings for channel-1 on, so it is easy to see what I worked with.

Coupling
The coupling is DC

Bandwidth
Bandwidth is 20MHz, otherwise too much noise is visible.
These measurements are about waveform and level and not because I want to see everything that happens while measuring.
I do always start with the “window wide open” when making measurements.
Then I have an impression if there are things that also come into play that I need to take into account.

Pressentation
Most measurements I show on forums are “without” noise, so that the attention goes as much as possible to the measured signal of interest.

Scale factor
So the next setting is the scale of the channel used, and this is set by using the “User1” setting of the Siglent scope.
Here the setting is 50V per ampere.

Input impedance
This should be set to 1M Ohm so as not to blow up your scope. 

Unit
This should be set to Ampere unit, this because we are measuring the peak current behavior of a LAB power supply.


.

And now for some comments on LAB Power Supply's.
I've tested a lot of different circuits and I've developed a fondness for the “Harrison Consept.”
That is that the reference section has its own floating power supply and the control current for the power section usually comes from a current source.

Then there are two control loops one for current and the other for voltage.
Which then control the current through two diodes to the Power section.

Nowadays it is a whole lot easier to make a fast responding LAB Power Supply.
The advantage is that it is then possible to keep the capacitor across the output terminals low and still have a fast responding loops with good dynamic behavior.

This has not yet caught on with many manufacturers,
a thick capacitor across the output terminals and slow opamp/power transistors is much cheaper.

Better loop contro en component choices
Furthermore, a current source that always draws current from the output(when the channel is turned on) helps well for excellent dynamic behavior.
There are also opamps available these days that have a large phase and gain margin, look at the ADA4625 models from Analog Devices.

This opamp series has low noise, little bias current, fast and little offset, but above all a phase margin of 88 Degrees at 100pF capacitive load, which is excellent!

Everything I have written here is mine no AI, but translated by deepl.com
And I'm a dyslexic Monkey, which can lead to pretty crooked sentences. 

Kind regards,
Bram

[rf-messkopf]

The inrush duration is also well beyond "Load recovery time" spec of 400µs, so can't use that either to handwave.

That spec exclusively refers to the voltage regulator. A 'normal' bench PSU with a single series pass regulator is unable to sink current. Thus, when in CC mode and the load resistance drops so that the supply has to lower the output voltage in order to keep the current constant, the excess charge in the fixed output capacitor corresponding to the voltage difference has to be dumped into the load. The supply cannot take up this charge. Therefore, we see the exponential decrease of the output voltage shown a number of times in this thread, and this decrease will depend on the load. There is no way around that, unless the PSU is a two-quadrant one like the Agilent 6632B which can down-program the current, or a SMU.

But that is not the point of all of this. If we set a voltage on our PSU, we expect that PSU to output that voltage, no more. Similarly, if we set a current, we expect that PSU to output that current, no more. If a manufacturer specifies that a set current may be +/-5% + 5 mA, I do not expect the output current to be 14 times higher than that.

Well, transients exceeding set values are inevitable when the load resistor suddenly drops in CC mode, as the normal bench PSU cannot sink current, and the transition time will be load-dependent.

The delay time when transitioning between CV and CC mode (I'm not aware of a PSU that actually specifies that delay) is another story. I can see why they seem to be doing that. Otherwise you can get rapid oscillations between CV and CC mode. I've run into that with homebrew PSU designs and complex nonlinear loads long time ago. The question is which delay is appropriate. I think the approx. 150 ms of the Agilent are a bit too long. Would be nice to have control over it, and to have it specified at least.

And if the design shows transients at enabling the output, I want to see that beforehand in the datasheet and not after the purchase on my bench.

That overshoot when the supply is activated with the output enable button and with a load connected which draws more current than set is yet another story. I agree that I'd prefer the supply not actually doing that. But one may also argue that it should behave just the same way at startup as when transitioning to CC mode in order to deal with complicated loads. Again, I'd prefer user control over this behavior.

So no idea whether this is in spec or not. 

[KungFuJosh]

KungFuJosh asked for some explanation of what scope probe I was using, which can be seen below in the picture.
Cost about 1850$ *grin*

Thank you for all the details!

Also... Damn it! I knew it was going to be an expensive probe. 😉

[jayk]

All this is making me feel a lot better about my GPP-4323.  If only it had a decent web interface and were a bit quieter under load it would be perfect.  Still a great value for the money.

[mhsprang]

But that is not the point of all of this. If we set a voltage on our PSU, we expect that PSU to output that voltage, no more. Similarly, if we set a current, we expect that PSU to output that current, no more. If a manufacturer specifies that a set current may be +/-5% + 5 mA, I do not expect the output current to be 14 times higher than that.

Well, transients exceeding set values are inevitable when the load resistor suddenly drops in CC mode, as the normal bench PSU cannot sink current, and the transition time will be load-dependent.

The delay time when transitioning between CV and CC mode (I'm not aware of a PSU that actually specifies that delay) is another story. I can see why they seem to be doing that. Otherwise you can get rapid oscillations between CV and CC mode. I've run into that with homebrew PSU designs and complex nonlinear loads long time ago. The question is which delay is appropriate. I think the approx. 150 ms of the Agilent are a bit too long. Would be nice to have control over it, and to have it specified at least.

And if the design shows transients at enabling the output, I want to see that beforehand in the datasheet and not after the purchase on my bench.

That overshoot when the supply is activated with the output enable button and with a load connected which draws more current than set is yet another story. I agree that I'd prefer the supply not actually doing that. But one may also argue that it should behave just the same way at startup as when transitioning to CC mode in order to deal with complicated loads. Again, I'd prefer user control over this behavior.

So no idea whether this is in spec or not. 

If you go back and take a look at reply #37, second image, you see the current spike first, then drop from 20 mA to 10 mA (setpoint) in several steps over a period of 12 seconds. That has nothing to do with any capacitor discharge but everything with something that goes terrible wrong in the firmware.

[rf-messkopf]

If you go back and take a look at reply #37, second image, you see the current spike first, then drop from 20 mA to 10 mA (setpoint) in several steps over a period of 12 seconds. That has nothing to do with any capacitor discharge but everything with something that goes terrible wrong in the firmware.

Agreed, that should definitely not happen and is out of the setting accuracy spec. R&S should fix that.

I have not seen something similar on the HMP4040 or any Agilent supply, though.

[pdenisowski]

you see the current spike first, then drop from 20 mA to 10 mA (setpoint) in several steps over a period of 12 seconds

Agreed, that should definitely not happen and is out of the setting accuracy spec. R&S should fix that.

Setting accuracy simply refers to how closely the output voltage is to the configured voltage - there's no time component to it.

There is also recovery time, which does have a time component - i.e. how long does it take the voltage to stabilize after a sudden change in load impedance.  This is usually specified as X seconds until the voltage stabilizes, +/- some given range.

These are also the definitions that Tek and Keysight use. (see attached and links below)

Sorry to ask, but could I confirm that we all have the same understanding of the situation?

The current drawn by a power supply is a function of the supply voltage and load impedance, and the current-limiting that occurs when a supply transitions from constant voltage to constant current mode is achieved by (automatically) reducing the output voltage until the current drawn by the load is no greater than the configured current limit.

That is, when the power supply senses (via its internal readback function) that the delivered current exceeds the user-configured current limit, it will begin to reduce output voltage until the current is below the limit.  During the time it takes for this to occur, the current will be above the user-configured current limit.

So is the issue
(a) that the current takes "too long" to fall below the limit?
(b) that the current increases above the "steady-state" level during the transition from CV to CC?
(c) something entirely different?

Sorry to have to ask this way, but I want to be sure I clearly understand what the concern is. There is a lot of discussions and graphs (and many thanks to everyone for those) but I'm not sure I've seen data showing where the NGE specifications are being violated.

Recovery time specifications indicate the time needed to reach a set value +/- some tolerance band.  If the tolerance band is +/- 20 mV, the additional time it takes for the voltage  (and thus the current, assuming a now constant load) to "further recover" from 20 mV to 10 mV is not a part of that specification.

Here is an article on this topic from Keysight:
https://www.electronicdesign.com/technologies/test-measurement/article/21801255/understanding-and-measuring-power-supply-transient-recovery-time

And it's discussed in this Tek whitepaper:
https://www.mouser.com/pdfdocs/Tektronix-Choosing%20the%20Right%20Power%20Supply%20for.pdf

Again, I may be grossly misunderstanding the issue here, so just trying to get some clarification regarding which specifications are not being adhered to.  We're still actively investigating this on our end as well.

[nctnico]

I think you need to define it differently. An ideal current limited power supply has two control loops running at the same time: one loop regulates the output so the set current limit isn't exceeded, the other loop regulates the output so the set output voltage is not exceeded. And this should be an AND function. Set current AND set voltage should not be exceeded.

From the graphs it looks like that in some power supplies the current limit control loop is disabled for a short while causing the output to be regulated to the set voltage but ignoring the requirement for the current limit.

I don't think this is a matter of specification but the very basic behaviour of a lab power supply. The set current limit and set voltage should never be exceeded under any circumstance as this will cause damage to devices connected to the power supply; hence there is no need to specify this. 

[mhsprang]

@pdenisowski: Although I am an electronics engineer, I am also a "user". So, if I purchase a power supply with CV and CC capablities, then to my circuit I design I am the engineer but to the power supply I am "merely" the user. In my opinion, I do not need to get to understand how my power supply works, based on assumptions and experiments. Instead, I rely on its specifications in the datasheet.

If I set it to be used as a current source with a certain max. voltage, I may expect, as a user, that those setpoints are honoured. I don't care about control loops and arguments about how the V and I loops interact, I only care that if I set the max current to be 10 mA, that the current will not exceed 10 mA. I will accept that there is a large capacitor at the output terminals, and that is why I use the "output enable" button to enable the power, instead of plugging in the cables to my delicate circuit, assuming the output cap is after the switch. Even this should be in the specifications, IMO.

To come back to my example in reply #37: I connected the 50 Ohm resistor to the power supply, which was set to 8V and 10 mA. I enabled the power to the resistor by pressing the Output button on the NGE100 while the resitor was connected already. So from that point onwards, it is the power supply's responsibility to deliver what I asked it to: 10 mA of current with a max. voltage of 8V. Then I hit the output button, as a user.

Seeing the current overshoot to 140 mA for 30 ms is one thing, althoug it exceeds the 200 us in the specs, but seeing the current drop from 20 mA to the setpoint of 10 mA over 12 seconds is a blatant violation of the specifications of this power supply. One could also say: I did not change the load. I merely enabled the output to a constant load.

[blackdog]

Hi pdenisowski, :-)

I have a suspicion that you don't understand what test was done....

Preparation
Set a LAB Power Supply to 8V and set the current to max 10mA.

Now connect a 50 Ohm resistor across the output terminals of the channel you set.
Connect a scope across the 50 Ohm resistor.

Set the scope to say 100ms time base time and single shot and the trigger level to say 30% of the expected value.
If you want to do it nicely, set the scope for your used channel to current measurement and then adjust the sensitivity in the probe settings to 50V/A.
Now the peak current is neatly readable on the scope screen.
Connect the scope with a BNC-Crock cable 1:1 low cost probe, as in my previous photo.

If everything is set up correctly and you turn on the LAB Power Supply via the enable switch, you will see the behavior of the Power Supply you are testing.
The maximum value of the current through the 50 Ohm resistor should not exceed 10mA.
This above is a good test for the behavior of the I-Loop control.

This test is slightly different from the test for the dynamic Voltage behavior of a LAB Power Supply.
In this, you vary the load current over a short period of time and see how well the LAB power supply responds in terms of the U-Loop.
Many manufacturers specify from 50 to 90 or 100% load, and there are some that test 10% to 100% as a test in this dynamic U-Loop test for e.g. within 50uSec within a few mV of the set voltage value.

By the way, these are just two of the tests you can do on LAB Power supplies dynamic behavior.
Like this measurement which I also think is important, i have never seen a manufacturer specify how much commonmode interference comes out of a LAB Power supply output.
Especially the ones that have switching behavior, linear post-regulation usually doesn't help with that.

But let me not digress, the current test at issue here shows nicely what a manufacturer thinks is important,
my measurements of a few LAB Power Supply's shows that fast current control at the GW Instek Power Supply is well taken care of.

I hope things are more clear now.

Kind regards,
Bram

[mawyatt]

Think most users work with a Lab Supply the way Bram describes, and this is also how we generally utilize a Lab Supply, especially when sensitive circuits are involved.

The Power Supply is Mains Powered On, Voltage and Current Limits are set, the DUT is connected either before or after the Limits set, then the PS is Enabled turning ON the Output(s).

In this use case the Power Supply should not overshoot the Voltage or Current Limits set by the user IMO.

Some Lab Supplies like the GPP4323 and SPD3303X have an ALL ON/OFF Single Button to Toggle the Outputs ON and OFF which hints at the above described use case. This is a nice feature when one is utilizing multiple outputs which is common with complex DUT circuits/systems.

Best,

[KungFuJosh]

I did blackdog's test with my SPD3303X v6.2 hw.

For values set below 30mA, there's a spike to nearly 30mA before dropping to the set current. For currents 30mA and up, there's no overshoot.

Inserting the probe into the scope was seen as nearly 15mA.

[pdenisowski]

First, thanks to everyone for the detailed replies!

If I set it to be used as a current source with a certain max. voltage, I may expect, as a user, that those setpoints are honoured.

it is the power supply's responsibility to deliver what I asked it to: 10 mA of current with a max. voltage of 8V

At the risk of making an unnecessary observation:  the user normally configures a voltage value and current limit.  I'm not aware of any T&M benchtop power supply where the user explicitly configures the desired output current.

My understanding is that the only way to "force" a supply to output a given current is to set the desired current limit and then set the output voltage so high that the supply will transition to constant current mode and reduce the voltage to the appropriate level when the supply output is enabled. 

However, the power supply has no knowledge of the attached load impedance before it's turned on and current starts flowing.  If you set the output voltage to 10 volts and a current limit of 1 amp, and attach a 1 ohm resistor, then 10 amps (!!!) will flow for some finite amount of time before the supply's readback function is able to detect this and lower the output voltage to 1 volt in order to honor the configured current limit of 1 amp.

So to my (potentially faulty) understanding, the only question is: how long does it take the voltage to be reduced such that the current changes from 10 amps to 1 amp (+/- a tolerance value, specified in volts)?  We want this time to be as short as possible (microseconds), and this is what manufacturers specify.

seeing the current drop from 20 mA to the setpoint of 10 mA over 12 seconds is a blatant violation of the specifications of this power supply.

Sorry for being lazy (because I believe you put this in an earlier post), but what is the voltage drop during these 12 seconds?  As far as I'm aware, all T&M benchtop power supply manufacturers specify recovery times in terms of a voltage band around the target voltage - the timer stops (so to speak) once that band is entered, even though additional adjustments to the output might be made after that. 

It's also worth noting that the NGE (and frankly, most other decent benchtop power supplies) has an overcurrent protection function which will disable output if a user-configured current threshold is crossed.  This is spec'ed at < 10 ms on the NGE.

And again, sorry if I'm being dense:  I completely understand why you would want a "constant current" supply and why you wouldn't want a massive current surge when a power supply is enabled (and I've fried LOTS of things in my career), but the easiest way to avoid this is to set an output voltage only slightly higher than the voltage needed to enter constant current mode, with an electronic fuse configured just in case something goes wrong.  Both of these are easily done on the NGE.

[Martin72]

However, the power supply has no knowledge of the attached load impedance before it's turned on and current starts flowing.

This is the point that should be internalized.
If you set the current limit of a power supply to 30mA, for example, this is no different than setting a controller setpoint.
If you then enable the output and allow current to flow, it takes a certain amount of time for the actual value to match the setpoint.
This is the nature of the controller.
You cannot expect the current to be limited “instantaneously”.

[thm_w]

However, the power supply has no knowledge of the attached load impedance before it's turned on and current starts flowing.

This is the point that should be internalized.
If you set the current limit of a power supply to 30mA, for example, this is no different than setting a controller setpoint.
If you then enable the output and allow current to flow, it takes a certain amount of time for the actual value to match the setpoint.
This is the nature of the controller.
You cannot expect the current to be limited “instantaneously”.

Seems beside the point when we already have demonstrated supplies that do not overshoot, and that is the desired behavior.

[2N3055]

I wanted to stay out of this discussion but am getting an itch..
 
Paul, this is not moment for clever lawyer talk that "technically" specifications are not violated because someone in R&S used clever wording in datasheet...

Current limit is exactly that: a limiter that limits that current does not exceed set current limit. You can accomplish it with some kind of electronic fuse that will disconnect load, you can enable foldback limiting (that will reduce current to minimal until you reconnect load) or you can enable constant current mode that will keep constant current.

In any case, current should not exceed the set limit. It is a limit, not some general recommendation.

If one wants to be able to deal with difficult loads that have large startup current, than that has to be separate, configurable option.

Those are, generally speaking, expectations form average user.
And when they pay 3x the money (or 10x) they expect it behave better that some cheap PSU.
Which here is not the case, because I have Rigol DP831 that behaves very well in this overshoot test, and to add insult to injury, I have two noname analog PSU that are even better in that regard, behaving like decent current source when starting. They also switch between  voltage and current control loops without much saturation in error amplifiers, making transition quite smooth.

I refuse to believe R&S cannot make it better than some noname small manufacture in China copying 50 year old HP design.
But maybe design dept and product leaders should pay more attention to analog design that to "visual identity of the brand".

R&S is brand that is famous for their outstanding devices with state of the art performance.
It is not Versace. Or Apple. Or some luxury brand that sells and designs something that is fancy and status symbol but not mean to be really pushed to the limit.

R&S is not supposed to be Vertu encrusted in diamonds but basically just an ordinary phone.
R&S is supposed to be state of the art military communication device with state of the art performance.

More like LeMans Porsche LMP race car than Porsche SUV with leather seats.

[Martin72]

Seems beside the point when we already have demonstrated supplies that do not overshoot, and that is the desired behavior.

Then they have a very, very fast controller.
You have to reach the point where you have to cut back, in any case, time passes for that.

[2N3055]

I actually looked into user manual.

It explicitly states that PSU functions in Constant Voltage (CV) or Constant Current (CC) mode.

[thm_w]

Seems beside the point when we already have demonstrated supplies that do not overshoot, and that is the desired behavior.

Then they have a very, very fast controller.
You have to reach the point where you have to cut back, in any case, time passes for that.

The other test that could be done is to compare the speed of regulation once the output is on, though at low currents like this its complicated by the size of the output capacitor.
Maybe that would tell us if it is a regulation loop speed issue or simply a software bug (timing or sequencing of the output)?

[rf-messkopf]

However, the power supply has no knowledge of the attached load impedance before it's turned on and current starts flowing.  If you set the output voltage to 10 volts and a current limit of 1 amp, and attach a 1 ohm resistor, then 10 amps (!!!) will flow for some finite amount of time before the supply's readback function is able to detect this and lower the output voltage to 1 volt in order to honor the configured current limit of 1 amp.

As has already been remarked by others, there are supplies which do not behave in this way. They seem to ramp up the voltage slow enough when activated by the output enable button for the current control loop to prevent an overshoot.

However, it is true that when a load is connected to an already activated PSU channel set to a certain voltage, and if the load will draw more current than set, then the current will necessarily overshoot. A number of supplies (among them R&S) seem to wait a couple of milliseconds before switching to CC and the current controller actually starts to ramp down the voltage in order to regulate the current. As I wrote before (see the post above), this behavior could be designed in deliberately.

Also, we speculated that it could be this delay that makes the R&S supplies (among others like the Agilent 6632B) overshoot when activated by the output enable button with a load connected that will draw more current than set. There may be pros and cons for this behavior. I'd prefer no overshoot, or user control over the delay when transitioning between CV and CC mode.

Besides this overshoot (which may be intentional), user mhsprang has seen an issue with the settling time to the set current value in CC mode. To understand this issue one has to realize that a 'normal' bench PSU with a single series pass regulator cannot sink current. Thus, when in CC mode and when the supply has to lower the output voltage to keep the current constant, the excess charge corresponding to the voltage differential in the fixed output capacitor has to be dumped into the load. Therefore, with a resistive load an exponential decrease of the output voltage is expected and has been seen a number of times in this thread. Also, the settling time will depend on the load. There is no way around that, unless the PSU is a two-quadrant one like the Agilent 6632B which can down-program the current, or a SMU. Therefore the settling time spec of normal bench PSUs only refers to the voltage controller. I've never seen a current settling time spec for CC mode.

Now mhsprang has noticed the following: He connects a 50 ohms load to his NGE100 (set to 8V, 10 mA). Then there is the (deliberate?) delay of about 40 ms before the current controller becomes active. Then there is the expected exponential voltage decrease due to the output capacitor (approx. 50 ms). But then it takes another whopping 12 seconds before the output current actually reaches the set 10 mA. Take a look again at the diagrams in post #37. The vertical scale is in volts, but notice that  due to the 50 ohms resistor across the output, it is proportional to output current. After the exponential decrease is over, the output sits at 1 V (i.e. 20 mA) and slowly ramps down to 0.5 V (i.e. 10 mA) within 12 s. There is a clear distinction between the exponential part of the curve and the slow ramp to the set current.

Even though there is only a setting accuracy spec for the current controller (0.1% + 5 mA, so the 20 mA are way out of spec) and no settling time specified (as explained before, there can't be because it depends on the load), I think it is unacceptable for a PSU to take so long before a set value is reached.

I hope this makes things clearer.