In this blog I'm going to describe using the 6424E to look at various aspects of the working and performance of a switching power supply. The power supply has been designed for the Earth Resistance meter project which hangs out in Open Source Hardware using the ARMP tag. (Just search for ARMP).
In order to understand what the scope is up to in this blog, it's necessary to understand the power supply it's testing - so we'll start with that:
It's quite an unusual power supply. It has to provide +/- 25V at up to 35mA, +/-15V at up to 20mA and +/- 5V at up to 40mA. Low noise and reasonable efficiency are absolute requirements. Regulation is not crucial and I'm expecting to use linear regulators on the 15V and 5V sections to get the noise really low. The input will come from a lithium battery and be in the range 5 - 8.5V.
To attempt to get all these supplies each with it's own switcher would be expensive and very complicated, so I've gone for a custom hand wound transformer driven by an LT3439 chip. This chip is designed for the purpose and has slew rate controlled output drivers to reduce noise. It can be synchronised to an external clock, which will be essential in the final application.
I really don't want to put linear regulators on the +/- 25V outputs, they would be hard to protect (this supply is for the ERM output power amplifier) and the high voltage restricts the choice of available parts very significantly.
My way round these problems is to use a linear regulator to control the power supply to the LT3439 and close the feedback loop from the 25V output. For simplicity I'm assuming that the +/- 25V volt rails will be equally loaded and will track each other well enough. Up to 1V of mis-tracking won't hurt.
REF1 is a 2.5V shunt reference. U2 amplifies the difference between the reference and a proportion of the +25V rail, fed back via R13. If teh 25V feedback signal is too high the output of U2 will tend to go low, reducing the voltage on Q2 base and reducing the base current in Q1, and so reducing the voltage on Q1 collector. So we have a feedback loop doing the right thing. R11 controls the gain in the loop to keep it stable, so far C12 has not been required. Q1 is protected against sudden shorts by Q3 which will divert Q1 base current if enough current flows through R7. It shouldn't ever be necessary because the LT3439 has current limiting, but if that chip should fail, I want another level of protection before the Lithium Ion battery gets shorted.
I've already posted some pictures of the power supply board being tested in my Probe Positioning System blog, but for completeness here's another one !
I've loaded the outputs with 5k6, 2k7 and 1k resistors. The croc clips are to attached a DMM and the scope probe is on the +25V rail.
it took a while to get this board to work, I built two and managed to blow the LT3439 on one - due to an undetected (until too late) short on the secondary side of the transformer. I used a prototype board service because the usual Chinese supplier was having a holiday, and this meant that my boards have no solder resist. On this working board there is a short under the LT3439 but I was able to lift the affected pin and connect it by wire. I think I'm unlikely to buy any more boards like this !
Once the thing was working it seemed time to look at the noise:
PSU noise on 25V rail, 9.3mA load, 20MHz bandwidth limit
The 6424E had no trouble showing the noise when AC coupled.
PSU noise on 25V rail, 9.3mA load 1MHz filter
The noise is very low indeed with a 1MHz filter, which takes the nasty spikes away. This isn't actually measuring a signal with the scope, more using it to show what sort of passive filter I might need to get the PSU noise down to where I'd like it.
PSU noise on 25V rail, 9.3mA load, using resolution enhancement to 14 bits
The resolution enhancement has a similar effect to the filter, I've included this screenshot to show that you need to be very careful when using the enhancement - because it can hide signals, I've left the ruler bars showing so you can see what the peak to peak noise really is.
It would be really nice to use DC coupling rather than AC, then I could do noise and load step testing all at once.
But to do that I need to apply an analogue offset of 25V, and the most sensitive range the 6424E can do that is 20V.
PSU noise on 25V rail, 9.3mA load, scope on DC, 25V offset
You can't see the noise at all - the 6424E just can't offset 25V on a sensitive enough range to see the noise.
The Rhode and Schwarz RTA4004 could do no better - it can offset 25V on its +/- 5V range, but the psu noise was still impossible to detect on screen.
At this point you might wonder why I'm mentioning this at all, if no scope can do it - to which the answer is that it can be done:
LeCroy 610zi, PSU noise on 25V rail, 9.3mA load, scope on DC, 25V offset.
The LeCroy can apply 25V offset on it's +/- 1V range (Pico describe the range setting in terms of V for the full screen so I've converted the other scope settings to match. That's why the LeCroy says 200mV, it means 200mV per division.)
It can display the noise, despite being an 8 bit scope, it is using resolution enhancement (by 3 bits) . I've set it to sample a bit too fast here, the 10GS/s results in an 80MHz bandwidth after enhancement.
The secret is being able to offset large voltages on low ranges - a purely analogue feature. Whilst we are on the subject, the LeCroy makes a good job of explaining what is going on - the offset is clearly visible in the C1 setup box. (To be fair, this scope is about 7 years old, the current equivalent costs about 3x as much as the 6424E.)
The 6424E voltage scale does not take the offset into account, and neither do the ruler bars - this should certainly be an option, and in my opinion the default.
Enough of noise, let's try some load step testing. I had already tested the steady state regulation of the 25V supply and found it to be quite good over an input voltage range of 4.6V to 8.5V the +/-25V rails went from 50.71 to 50.925V -that's about 0.4% for a nearly 2:1 change in input - and quite good enough.
I don't have a six channel electronic load , so I used the single channel one, initially to apply load pulses to the 15V rail, while at the same time monitoring the voltage at Q1 collector (to see what the regulator loop is up to) and the +15V rail (to see how the uncontrolled rails behave.) I started with 60mA pulses 25ms long.
Red = 15V rail, Yellow = 25V rail, Blue = Q1 collector, 25ms, 60mA load pulse on 15V rail
The unregulated 15V rail dips but no change is visible on the 25V rail.
I decided to move the load and apply pulses to the regulated rail, reasoning that the filtering effect of the big caps on the supplies was making the transients to slow to really see what the feedback loop was doing.
Red = 15V rail, Yellow = 25V rail, Blue = Q1 collector, 25ms, 40mA load pulse on 25V rail
Red = 15V rail, Yellow = 25V rail, Blue = Q1 collector, 25ms, 40mA load pulse on 25V rail
Here I've set up three views, shortened the time base and applied some resolution enhancement. I'm using DC coupling on all channels and offsets on the yellow and blue - which makes it very hard to tell what the absolute voltages are.
Red = 15V rail, Yellow = 25V rail, Blue = Q1 collector, 25ms, 60mA load pulse on 25V rail
The error has increased dramatically, and we've lost the expected increase in error at the start of the load pulse, followed by a reduction as the feedback loop responds.
It looks as if the regulator is running out of range. The blue trace has been offset by 5V and the bench power supply to 5.1V (I'm not measuring right at the board but at the far end of the wires - so the 4.8V on Q1 collector is very close the maximum it can reach.) I'll guess that I need more volts, and increase the supply voltage to 5.5V.
Red = 15V rail, Yellow = 25V rail, Blue = Q1 collector, 25ms, 60mA load pulse on 25V rail, supply increased to 5.5V
I've switched the yellow channel to AC, so I can get a better look at the noise. Increasing the power supply has done the trick.
Using AC coupling for the 25V rail has a side effect, look at the overshoot when the pulse turns off - is it real or is it an artifact of the AC coupling ?
We can tell that it's probably real, because it's visible on the DC coupled traces too, but if I used a longer pulse it would be distorted by the AC coupling.
I mention this to make the point that being able to offset large DC voltages on low ranges isn't just a "nice to have" - it makes a real difference to measuring perfectly normal stuff.
Although a 60mA pulse + 9.3mA resistive load is about full load for the design, it would be nice to see if it can do better - I tried 80mA and that was too much but it can just about cope with 70mA.
Red = 15V rail, Yellow = 25V rail, Blue = Q1 collector, 25ms, 70mA load pulse on 25V rail, supply increased to 5.5V
As you can see, the PSU is just about coping here - the recovery time has lengthened significantly.
The main purpose of this blog is to talk about the scope - so pushing the PSU a bit more and discussing how it fits into the ERM will have to wait.
I've whinged about offsets and the 6424E has two problems in that area, one is a software issue which I'm hoping will be addressed soon - it really should do the maths for you and apply the offsets to the measurements. Other scopes do that.
The second thing is the amount of offset available, the Pico is no worse than the more expensive R & S scope in the range used for the PSU tests, although the R & S offers a staggering 245V offset on higher ranges. The LeCroys are better, but much more expensive.
For Pico to increase the offset range would be a hardware change so it isn't going to happen tomorrow, but it would be a huge improvement.
The scope has done pretty well. It's ability to show multiple views with their own rulers and zooms puts the R & S RTA4004 and the LeCroy 610zi in the shade.
I was able to measure everything I needed to, and a lot of things more easily than with other instruments.
The noise tests were done using a Windows 7 PC and the load step tests on a Windows 10 machine. I've had the Pico software crash a few times (4) on Windows 7 but had no crashes on Windows 10.