EEVblog #861 - Rigol DP832 PSU FAIL & REPAIR

[exe]

I think you're missing the point of a LINEAR regulation power supply.

Why do you think so? I was talking about mosfets in linear region, that is still valid even for switching applications: http://www.infineon.com/dgdl/Infineon+-+Application+Note+-+PowerMOSFETs+-+OptiMOS%E2%84%A2+-+Linear+Mode+Operation+and+SOA+Power+MOSFETs.pdf?fileId=db3a30433e30e4bf013e3646e9381200

[Kleinstein]

A full SOA curve has curves for different pulse times and also DC operation. The pulse times assume a low duty cycle and fully open MOSFET in between. However for a linear power supply the DC SOA curve is the relevant - so the MOSFET is for a long time at the specific conditions.

With SOA curves for MOSFETs there are a few problems:
- many MOSFETs don't specify a DC curve, though there will be curve somewhere way down.
- the SOA curve is not production tested for normal types - so they are more like expected by design
   SOA testing is expensive and even if done only checking a small part, e.g. a few points for pulsed operation
- there is chance that not all samples meet there SOA specs, reliable part would need individual SOA testing.
  One parameter that enters is the degree of local inhomogeneity - a parameter prone to scattering
- the SOA measurements may be limited to a few points. So even if there is a curve it may not be reliable. Especially if the thermal instability range is not shown in the curves, chances are this effect is not checked for and curves are just extrapolated.
- The SOA curves shown are usually for something like 20 C case, so derating for higher temperature is needed.
- for TO220 and similar chips non homogeneous contact to the heat sink might degrade the SOA curve.
- many mosfets are made by several sources, the SOA curve may no be considered an important parameter and may vary.

[fish]

I have followed Dave's trials and tribulations with this supply with some interest and concern as I only bought mine on his recommendation!

That said I am extremely pleased with it. It offers me SO much more than the cheaper power supplys I have had in the past and the current limit has save some of my components form untimely deaths. And all the bells and whistles are great for understanding what's going on, even just noting the effect different bits of software have on the current draw from a pi etc.

However like Dave I have had some issues with mine, ( https://www.eevblog.com/forum/repair/dp832-lose-screws/msg786388/#msg786388 ) thank fully not to the extent of his, as I would not have managed to work out what was up. But all the same there is clearly some QC issues on the DP832 line.

As for, is it cheap? Well it is very reasonably price for the performance but if you are going for hobisist, self employed, start ups and small companies then they really cant afford for things to just go pop when they cost that much! If Rigol get a bad name for this then people might just make do with less equipment which would be a big shame for them and for Rigol.

Once again a great video, Thanks Dave.

[exe]

SO much more than the cheaper power supplys I have had in the past and the current limit has save some of my components form untimely deaths.

Actually, has anyone measured its load response? A few years ago I was quite surprised to see 1000uF output caps (and even with them it has spikes on turn on, sometimes negative spikes, are they good for caps? ). Now I realised out they used a MOSFET pass element with gate capacitance of almost 10nF... Never seen anything like that before. So I'm curious how good it actually is (I would expect it to be rather slow in response). And what about stability with capacitive load...

[fish]

So I'm curious how good it actually is (I would expect it to be rather slow in response). And what about stability with capacitive load...

I think the point I was trying to make is that for the price I don't expect supper duper performance BUT I do worry about the failure rate.

Also I do find the feature set very use full.

I was thinking of writing this up as a more formal question but I will note it hear first.

NewFile1 shows the start up (pressing the channel button not the main switch) with no load, I have tryed it with a tiny bit of load and it was similar, (the transience of my circuit is similar so its annoying.)

NewFile2 shows flicking a in line switch. (I have plots of some nasty bouncing but they are over well before the few milliseconds transience of what I was interested in)

Sooo, is this pretty standard for a power supply? Is this standard for this power supply (is mine broken) ?

I had a bit of a look when I got it and really struggled to get anything nearly this good anywhere else, even some  poor one channel thing was well over a third of the price but maybe I didn't look hard enough. So to be honest just using a switch every now and again is really not a issue for me when I get so much more than I would other wise. That said would like to take Daves advice and build one my self, but it still will not come close to the DP832.

Is there a better strategy than a switch? I was thinking about using a Micro controller and a transistor or something as that wouldn't bounce but if it did anything funky I wouldn't have the experience to tell so I think I will stick with my switch and a cap.


*** I also tested the supply with the probes on the terminals and got the same answer.

[DuPe]

SO much more than the cheaper power supplys I have had in the past and the current limit has save some of my components form untimely deaths.

Actually, has anyone measured its load response? ...

If you let me know a little in detail what exactly you are interested in, I may post some measurement results with
this PSU DP832 using electronic load (Maynuo M9811)
Cheers
Peter

[exe]

If you let me know a little in detail what exactly you are interested in, I may post some measurement results with
this PSU DP832 using electronic load (Maynuo M9811)

Hi Peter,

I'm no specialist in benchmarking PSUs, but I think it would be great to see:

1) set 12V and 3mA current limit and connect LED. We'll see if it survives . If so, set to 30V and repeat the experiment

2) 12V 200mA -- 1A change in load

3) 12V 200mA -- 3A change in load

4)  3.3V 10mA-1A change in load

5) 12V No load => 1A load.

For every measure it would be great to see it in both directions. E.g., in case 4, first step load from 10mA to 1A and then back to 10mA again. I.e., 10mA => 1A, then 1A => 10mA. I think your load is very smart and can do this with ease .

If your ever decide to make such test I'd like to thank you with a small paypal donation.

[DuPe]

No problem. I am getting so much information from this blog, so I am happy to contribute a little.
Only condition: An answer will take some days till I am back in may lab. I am currently travelling abroad.
Cheers
Peter

[DuPe]

If you let me know a little in detail what exactly you are interested in, I may post some measurement results with
this PSU DP832 using electronic load (Maynuo M9811)

Hi Peter,

I'm no specialist in benchmarking PSUs, but I think it would be great to see:

....

Hi,
here are the first results. Edit: Values are correct

Power supply settings were 29V/3A, channel 1. I see nothing to be worried about.
Current change was 100mA to 2A and vice versa (see file name)
More to follow as soon as there is more time.
Cheers
Peter

PS: I hope the attachements do work

[exe]

here are the first results. Measurements taken with 1/10 probe, so all Y-values have to be divided by 10.

Thank you very much! Just one question, should the values be divided by 10, or multiplied by 10? If divided then it's a very-very good result. If multiplied that would be... uhm, not great at all . So, 1/10 probe means attenuation by the factor of ten, or 10x amplification?

PS I wrote you a personal message about the stuff I promised.

[DuPe]

here are the first results. Measurements taken with 1/10 probe, so all Y-values have to be divided by 10.

Thank you very much! Just one question, should the values be divided by 10, or multiplied by 10? If divided then it's a very-very good result. If multiplied that would be... uhm, not great at all . So, 1/10 probe means attenuation by the factor of ten, or 10x amplification?

PS I wrote you a personal message about the stuff I promised.

Edit: Voltage readings are correct (probe ratio was set on scope)
What I am interested also is what will happen if you set to 30V @ some mA and then short the output.
In other words: Ho much energy is stored in the output capacitor and would a sensible device like a LED survive that.
I'll post the result as soon as available. I am also planning to compare to my Keithley PSU.

[exe]

I am also planning to compare to my Keithley PSU.

Ho-ho-ho, Keithley should "destroy" that rigol hands down . Actually, the results you obtained is already enough to conclude DP832 is a "snail" PSU.
Here is, just for comparision, step response of TI's lm1085.pdf 0-3A with 100mA preload: (see first attachment).

And, one of the fastest  MOSFET drivers on the market, lt1575, step response 0.2-5A: (see another attachment). It was designed for Pentium 2 CPUs, that's why it is fast like hell. I can't imagine what modern CPU power supplies can do... They should respond to change of tens of A in some hundreds of nanosecond. But they all are very low-voltage, while lt1575 works up to 18V or so.

PS all pictures from datasheets, so they should be taken with some grain of salt.

[DuPe]

here are the first results. Measurements taken with 1/10 probe, so all Y-values have to be divided by 10.

......... I am also planning to compare to my Keithley PSU.

Here we go: Keithley 2304A, Setting 19V/3A, Load 100mA to 2A and vice versa.
Consider the Y scale! Does not look bad for the Rigol at all

[DuPe]

I am also planning to compare to my Keithley PSU.

Ho-ho-ho, Keithley should "destroy" that rigol hands down . Actually, the results you obtained is already enough to conclude DP832 is a "snail" PSU.
Here is, just for comparision, step response of TI's lm1085.pdf 0-3A with 100mA preload: (see first attachment).

And, one of the fastest  MOSFET drivers on the market, lt1575, step response 0.2-5A: (see another attachment). It was designed for Pentium 2 CPUs, that's why it is fast like hell. I can't imagine what modern CPU power supplies can do... They should respond to change of tens of A in some hundreds of nanosecond. But they all are very low-voltage, while lt1575 works up to 18V or so.

PS all pictures from datasheets, so they should be taken with some grain of salt.

Hi exe
The load response especially of a universal/Lab PSU depends by far more on the design of the voltage control loop than on particular chips or transistor.
The design has to take care on regulator loop stability and static load regulation as well (the latter value is well suited for outnumbering in data sheets).
I assume the Rigol performs bette due to a bigger output capacitor. To verfiy takes a little bit more effort.

I think the weaknesses of the Rigol are somewhere else than load regulation: E.g. they do not recommend to use active loads. I.e: You should not try to
charge an accumulator with the DP832 (which is a mess)

PS: Just took a quick and dirty (I know, I know) test mesuring the capacitance between outputs: Rigol 500..800µF, Keithley 130..150µF (... depending on polarity of meter)

[exe]

The load response especially of a universal/Lab PSU depends by far more on the design of the voltage control loop than on particular chips or transistor.
The design has to take care on regulator loop stability and static load regulation as well (the latter value is well suited for outnumbering in data sheets).

Yeah, but those chips are complete regulators... Only that lt1575 requires an external pass element and a simple compensation, other than that they are ready to go. So I think it's a fair comparison (though they do not provide CC mode out of the box).

May be I'm missing something, just I expected much better load regulation.... I'll try to benchmark my DIY PSU based on lt3080 with 4.7u output capacitor as soon as I get access to a decent scope... May be bench PSUs are intentionally slowed down to be more robust for variety of loads. But there are models on the market with ~1MHz loop bandwidth.

[DuPe]

I am also planning to compare to my Keithley PSU.

.....
PS: Just took a quick and dirty (I know, I know) test mesuring the capacitance between outputs: Rigol 500..800µF, Keithley 130..150µF (... depending on polarity of meter)

Epilogue: Your question wheter a LED would survive (set 12V and 3mA current limit and connect LED):  I belive not:
To finally measure the output capacitor I had a very old fashioned idea: C=I*t/U.
I did set the current regulation (which is quite precise) to 5mA, voltage to 30V, did short the output and measured the time until 30V are reached after the short (3.4 sec)
C=(5e-3A * 3.4s)/30V = 566.66 µF (my fluke was almost right . ( I know, that a look into a schematic would be easer, but it was more fun this way)
I.e. it is unlikely that a LED survives a discharge of such a capacitor. If you want to light a led, I'd recommend to use lower voltage setting.

It does not speak for the DP832 that such a big ouput capacitor is needed to hold the data sheet values. 

[DuPe]

I made a bad beginner mistake : I have not noticed, that on the scope ratio was correctly set and values are corrected for probe divider: All voltage vallues in the oszillogamms are 1:1 correct.
I am going to edit my post for correctness.
So sorry

PS: Done. Information now should be ok

[Kleinstein]

The step response is not that bad. It's about what you expect with such a large capacitor at the output. Even other wise fast supplies will get slow with 500-1000µF at the output. Many supplies are rather slow, as to make the layout / cables less critical. For a fast regulator you have a hard time having the power devices at the back and output at the front.

Fast regulators like the LT1575 also need a suitable layout to get fast. Also the compensation for highest speed will not be that tolerant to higher capacitive or other problematic loads. So it's ok for a CPU, but not a lab supply. Its also a question if you really need a supply with lower output impedance than a foot of 20mm² cable.

An active load circuit could also be a a very difficult load and might make the supply / load combination oscillate. Ideally it should not happen, but both sides could be guilty - there can be unstable active loads as well. Especially a constant power more realized in software can very easy be unstable - it's more like what you expect. So if a supply with a active load in constant power mode is oscillating, I would blame the load first.

Supply circuits that use tap switching or preregulation with SCR might not like very fast changing or oscillating loads.  That a way to get magic smoke out, as not properly working preregulation might overload the main output stage.

[exe]

Fast regulators like the LT1575 also need a suitable layout to get fast.

I just conducted a few simple experiments with a step load of my LT3080 PSU. Basically, the observation is that the distance from output/smoothing capacitor dominates the response. So, placing a good cap (large electrolytic with short leads or a 10uF SMD mlcc cap) just right at the load changes results dramatically. Placing any capacitance on long leads does not make any sense. I can post some pics from my oscilloscope if anyone cares.

The load itself is just two metal-oxide resistors in parallel (of 33 and ~3.9Ohm) and a mosfet (PSMN9R5-100XS) switching the smaller resistor. It was driven from a signal generator. At 3.3V this gives approximately 0.1A => 1.0 step response.

Interestingly, putting polyester/MLCC caps at the point of load creates dumping/decaying/fading (which word is better?) oscillation. But this is explainable, mixing different caps and long wires leads to resonance problems (the output cap in PSU is 10uF MLCC).

Conclusion: measuring step response makes only sense if wires are as short as possible.