YAPS Part Eleven - Further Testing

18 Jul 2019

EDITS: 21/07/19 - added a PDF of the schematic

11/08/19 - updated navigation links

Introduction

My intention was to document, in some detail, testing of the power supply with both a 10 Ohm and 5 Ohm load (to drive the current from 0A to 3A.) To summarise, I’m seeing a number of problems - see later - and I thought I’d document where I’ve got to as I need to do some thinking about this. See my conclusion at the end. TL;DR; It isn't working as well as I hoped!

I’ve broken the testing down as measured testing and Noise and Ripple testing. If you can see problems with how I’ve done this testing, or ways of improving it, or indeed other tests you’d like to see, please let me know so I can improve or report additional results.

For the measured testing, I’ve captured data for the following tests:

Voltage and current comparison test
Steady state operation/voltage stability
Rise Time
Overshoot/Undershoot when turning on the mains switch; turning off/on load switch; lowering voltage level down to 0V; raising voltage level up to 15V
Soak test over 1 hour to measure temperatures and voltage stability

My scope was warmed up for 30 minutes before taking any measurements.

Environment and Setup

By necessity, the following equipment is turned on and plugged in to the same power strip as the scope and power supply: FTTP Modem, Router. Also operating, but plugged in to a different power strip: iMac, AOC monitor, USB hard drive.

The probe was used with a short ground connector

Images on other blog posts show the setup and layout of the prototype board.

10 Ohm Load

The load is a 10.08 Ohm, 100W resistor, connected through its solder lugs (although not soldered.) For all these tests, using the scope, the probe was set at 10x, full bandwidth, and a timebase that allowed for 1GSa/s where possible, measuring DC, and connected across the control stage output terminals.

Voltage and Current measurement

Essentially, this is a comparison test of DMM measurement vs INA260 vs 4Duino. I took readings at various voltage settings, reported in the table below.

DMM voltage was measured at the Control Stage output terminal
DMM current was measured in series between Control Stage output and load
INA260 voltage was measured at Vin to the chip
INA260 current was calculated using ohms law from the measured voltage at INA260 Vin.
4Duino voltage and current as reported on the screen
10.08 Ohm load

The Extech, measuring current, has a resolution to 2DP (i.e. 10mV). These results seem pretty good to me, particularly current.

The table following shows measurements with NO load attached (voltage only of course:

Again, very good.

Steady State Measurement

These images were captured off the scope

15V - 3.02V / division

10V - 2V / division

5V - 1V / division

0.5V - 100mV / division

As you can see voltage is steady but the noise is more apparent with the scope set in the 100mV division range. More on noise in later tests.

Rise Time

3.02V/div. Normal trigger set at 181mV. Timebase at 5mS.

LTSpice simulation rise time:

The actual rise time to the set voltage is 12.7ms actual vs 2.1ms simulated. There are similarities in the waveform but it's not clear why the rise time is so slow - my suspicion falls on the MOSFET, see Conclusions.

Overshoot

The scope timebase was set at 100uS and sample rate at 1GSa/s.

Voltage control on the supply was set so 4duino was reporting 15v; Trigger set to 15.46V, rising edge, normal mode.

Mains Switch test: mains turned off then back on. No triggering occurred. Trigger set to 15.4V and test repeated. Waveform displayed - noise is triggering the waveform.

Load Switch test: mains is on, load switch turned off then back on. No triggering occurred. Trigger set to 15.4V and test repeated. Waveform displayed - noise is triggering the waveform.

Rising Voltage test: voltage control turned to 0V and waveform confirmed on the scope. Trigger set back to 15.46V, normal. Voltage control rapidly turned to increase volts. No waveform triggered before measured volts reached 15V. No overshoot observed. Reset voltage control back to 0V and trigger to 15.4V and repeated. Waveform was displayed at a reported 15V (4Duino). Again, noise is triggering the waveform.

There doesn't appear to be any actual overshoot occurring - no high spiking - but noise is making it difficult to accurately assess if there’s a low-mV overshoot.

Undershoot

The scope timebase was set at 100uS and sample rate at 1GSa/s.

Voltage control set so scope was seeing 15V. Trigger set to -0.6V, falling edge, normal mode.

Mains Switch test: with voltage control set at 15V the mains switch was turned off. Waveform displayed.

Load Switch test: with voltage control set at 15V and mains switch turned on, the load switch was turned off. Waveform displayed

Lowering Voltage test: Voltage control rapidly turned from 15v to decrease volts. No waveform triggered until around 0.24V - again, this seems to be the result of noise (and is in line with the triggering on the overshoot.)

I’m not convinced I’m measuring these undershoot tests correctly, when operating the switches, as whatever position I set the trigger to (e.g. -60mV, -340mV) a waveform is displayed at that value when the switch is activated, but there is no spiking, just a steady voltage line. Lack of experience with scopes I think, but thoughts would be appreciated?

Soak Test

This didn’t involve the scope, just readings from the 4Duino - I'm particularly interested in reported temperatures. Power supply was left off to cool to ambient and then turned on for 1 hour at 15.5V and measurements recorded. At the end of the hour, I used a thermocouple on a multimeter to cross-check recorded temperatures and sample other components on the Control Stage PCB. Other components’ are the Schottky Rectifiers, 5V regulator and LT3092 current control.

Other components measuring between 40c and 45c.

These results seem pretty reasonable to me - no components seemed to overheat anywhere near their limits. Voltage had drifted down by 50mV from its initial peak setting - I’m wondering if that is due to the heating of the load resistor which is quite warm given it’s dissipating over 23W of power.

5 Ohm Load

The load is actually 5.08 Ohm, 100W resistor connected through solder lugs (although not soldered). I had intended to undertake the same testing as for the 10 Ohm load but the supply will only provide for 8.5V / 1.6A (approx) rising slowly over 6 minutes to 8.87V / 1.7A, and still rising. Something clearly isn’t right so there doesn't seem much point until I investigate further. I tested with a 1000 Ohm resistor (I only have 0.25W unfortunately) and that was ok. I don’t have another power resistor to try with. Not only that, I have noticed that touching the -ve line from the current control potentiometer drops about 2V on the output - see the video

No other leads do this and I have tested continuity (power off) and it’s solid. I’ve even replaced the connector. If I connect the 10 Ohm load and try this experiment, it has no affect; ditto no affect if no load is attached.

As a thought, I’m wondering if the bad output is the result of current limiting from the circuit? I can drop it down from whatever the current reading is to 0A with the potentiometer. Could the 5 Ohm resistor be faulty? It’s a 1% resistor measuring 5.08 Ohms so is slightly out of spec but I put this down to the DMM?

Noise and Ripple Testing

Noise and ripple are poor which I need to sort out. For the sake of documenting the testing, I’ve included images below to show what I mean.

Baseline

I took some measurements with the power supply off to get a baseline level. Same timebase and voltage division; bandwidth limited; 1x probe, AC coupled.

Background noise, no probe connected to the scope, power supply off:

Background noise, probe held, power supply off:

Background noise, probe on ESD mat (connected to ground), power supply off:

Background noise, probe held away from power supply, power supply on:

Background noise, probe held by power supply, power supply on (good heavens!):

Supply Measurements

Supply on at 15v. Timebase and/or voltage division selected to show noise; bandwidth limited; 1x probe, AC coupled.

10 Ohm Load attached, powered through RCD (1st image), no RCD (2nd Image):

No load attached, powered through RCD (1st image), no RCD (2nd Image):

And here's a couple with the timebase changed, both without RCD. With load (1st image) and no load (2nd image):

I think it's fair to say these are NOT good results and I need to do something about it. Again, I suspect the MOSFET.

Current Position

As I noted at the start, I decided to document the testing progress so far as there are clearly issues and I guess part of this whole project was to document how it came together, good and bad! The 5 Ohm load results I’m not sure what to make of or how to proceed - it clearly doesn’t like such a low resistive load but I don’t have any other way I can think of of testing it. I can’t understand what is up with the current control connection either - why would touching it with such a low load affect the voltage?

I’m wondering if the Mosfet is causing some of the issues. In this thread, Jon Clift points out it may well cause problems and it is definitely out of conformance to the requirement of the LTC1624 which requires a maximum 2000pF of input capacitance on the connected Mosfet (the one I have is 3247pF so way over.) I’ve hunted down another one, IRL8721PBF, but it’s not easy! There are 10s of thousands of the things and the parameteric search on Farnell/CPC is hopeless for purpose. I’m also not sure what the important specs to look for are but I have read around and here’s a comparison of what I think the important ones are:

The rise/fall/delay times seem to be heavily dependent upon the specific circuit they were tested with but I quote them for comparison purposes. The Total Gate Charge seems to be a better indicator of switching time and the IRL is significantly better. The input capacitance is well within requirements as well.

My approach now is to:

replace the MOSFET with the IRLB8721.
replace the power switch with an filtered module as per Michael Kellett’s suggestion in this thread.
bodge in a 0.4mH Common Mode Choke between the Power stage output and Control Stage input.
test voltages at various points of the circuit to compare with what the LTSpice simulation is reporting. It may give me a clue as to where issues lie - perhaps I’ve mis-soldered a resistor or capacitor (shame these can’t be tested in-circuit.)

Why a filtered power module and Choke? The frequency I was seeing the noise at seems to range from 60kHz to 130kHz and the datasheet indicates this would attenuate; I’m wondering if some noise is coming in through the power lead; I’m also wondering if this thing switching is potentially introducing noise on other equipment which I need to deal with. I may have grasped the wrong end of the stick here completely though

If anyone has any other suggestions then I’m really open to hearing about them. I’ve come a long way and I feel so close - It’s a bit disheartening if I’m honest: it all seemed to be going so well!

Follow on results I’ll post as a comment.

Next: Part Twelve - Design revisited: reworking the layout and PCB

Back: Part Ten - The Control Stage and Initial Functional Testing

Attachments:

Schematic.pdf

Top Comments

Parents

Andrew J over 6 years ago
So, keeping things up to date...I hope nobody thought I'd given up. I've made a few changes following input from you guys, particularly Jon. I now have a supply which will go to the full 15V and 3A!

Note the iMon current and U3/U4 temperatures: I did indeed blow those analog pins when I was probing but as far as I can tell the Linear Regulators are fine. Fortunately, the zener diodes (I think) protection on the 4Duino saved the chip.

The issue that I was having was caused by an LT and data sheet application note error in the schematic. They have tied the iLim pins of the linear regulators together with a 3k92 Ohm in parallel with a 5K pot, both then in series with a 100Ohm. The calculation they conveniently provide buried in the data sheet is 360mA / 1kOhm + 450Ohms = output current and even that isn't clear! That gives current = (2197 + 100 + 450) / 1000 * 0.36 = 0.989A per linear regulator. I'd used a 3K3 (1%) and a 5K pot (actually 4973Ohms) so I could achieve: (1984 + 100 + 450) / 1000 *.36 = 0.912A per Linear Regulator = 1.82A which, not coincidentally, was what I was achieving with the 5Ohm load. Clearly that's enough for 10Ohms which only needs 1.5A.

The annoying thing is that using LT Spice with LT components and an LT schematic (and my updated schematic) shows it working perfectly fine with the parallelised resistors! The lesson here is that you should still do the calculations if they are given, you can find them and understand them!

Doing the calculation with just the 5K pot would give a total of 3.996A. I desoldered the 3K3 and bingo. Sort of....rise time to nearly 3A (actually 2.9 eventually) was around a minute. In fact it never got there and was stuck at <15V and <3A (actual values following ohms law based on 5Ohm load) Not good - I have access to a very simple dummy load that was able to draw 3A out of it fairly easily - it wasn't stuck with 5Ohm resistance - so I knew it could do it.

I'd also looked at the feedback circuit with some more simulation - I mention this in a comment above in response to Jon. I can't yet change the Inductor or the Output cap but I have bodged in a 22nF cap across one of the 100K feedback resistors and re-routed the 1nF capacitor with a bit of wire (and cut the trace to its original destination). Firing it up and it shoots immediately to 15V, 3A! Yay. It still drops to 0V and 0A as well through the voltage and current controls.

Ok, so I think I've resolved the output functionality, leaving the noise. Here's some piccies. Firstly, ripple at the output terminal with the 5Ohm load. Pretty poor BUT I don't see the switching spikes.
Next, ripple at the output terminal with the 10Ohm load:
If you look carefully it is possible to see a bit of the switching noise just before the rise to a peak - I'd assume it is there on the 5Ohm waveform as well but the timebase is too large to notice it. Still poor ripple but the switching noise looks a lot more under control in my opinion (I'd be interested in other's opinion though.)

I also checked the voltage on the feedback pin of the Switching Regulator, Vfb. The whole point is to try and bring that under a bit more control. Prior to the bodges:
This is ranging from 460mV to 2.16V (10Ohm load). After the bodges with the 10Ohm load:
It's now ranging between 0.97V and 1.37V - firmly in the range for a 200kHz operation - I should say, I think it is still switching at 100kHz, I need to do more measurement. LTSpice (which I'm not sure whether I trust or not anymore) has this at 1.15 to 1.22V but I haven't modelled the output cap ESR or parasitics as per the actual build. Waveform is similar though. It gives me a bit more confidence that changing the Inductor and output cap for low ESR along with, say, a 20nF on the output, the Vfb would improve.

There's still the output ripple though. I need to find a way to squash that so this is what I intend to do:
- Redesign the PCB layout, particularly parts position and the grounding.
- Add in low ESR Input and Output caps for the Switching Regulator - I've found some that are 28mOhm.
- Add in a 20nF cap to sit alongside the Output Cap
- Provision optional pads for 2 more Output caps of 100uF (but could be other values)
- Change the inductor to 50uH to help reduce output ripple
- Test points for various pins so I don't blow it up anymore.
- Use small pots for the feedback circuit so I can make adjustments. I think I'd ultimately want to swap these for resistors at some point so I shall design around that.
I'm hoping to make use of LT's samples provision to get replacement parts to save cost! I want to keep the existing board as-is for comparison purposes and I can't be sure the parts I have are not damaged in some way. Remember, the first thing I did was solder the Schottky in the wrong way and blew up the Mosfet and sense resistor - it could just as easily have damaged the switching regulator.

I have been reading about proper grounding, including a paper from Analog Devices (LT.) It's difficult: there are definitive recommendations for the use of a ground plane but just as definitive recommendations for star grounding (Gene, for example, is a definite fan of star grounding as he's commented above.) Analog Devices are very much in the ground plane camp. One thing I'm fairly sure of, and open to correction, is that copper fills on the top layer that are via'd to the ground plane and used as a grounding plane by the top layer components is not a good thing - I think that's a capacitor. So I intend to tie ground pins with a short track and a via. Is it worth, then, 'filling the gaps' with copper on the top layer?

Finally, I will re-layout the PCB but I have previously gone with 2-layer to keep the cost down. That inevitably means I have to place tracks on the bottom layer which, if I go with a ground plane, will cause breaks - you can see in the images earlier in the comments. Should I bite the bullet and go for a 4-layer board - I could keep the current carrying tracks on one layer and signal tracks on a different layer?
- Cancel
- Vote Up +2 Vote Down
- Sign in to reply
- More
- Cancel

Comment

Andrew J over 6 years ago
So, keeping things up to date...I hope nobody thought I'd given up. I've made a few changes following input from you guys, particularly Jon. I now have a supply which will go to the full 15V and 3A!

Note the iMon current and U3/U4 temperatures: I did indeed blow those analog pins when I was probing but as far as I can tell the Linear Regulators are fine. Fortunately, the zener diodes (I think) protection on the 4Duino saved the chip.

The issue that I was having was caused by an LT and data sheet application note error in the schematic. They have tied the iLim pins of the linear regulators together with a 3k92 Ohm in parallel with a 5K pot, both then in series with a 100Ohm. The calculation they conveniently provide buried in the data sheet is 360mA / 1kOhm + 450Ohms = output current and even that isn't clear! That gives current = (2197 + 100 + 450) / 1000 * 0.36 = 0.989A per linear regulator. I'd used a 3K3 (1%) and a 5K pot (actually 4973Ohms) so I could achieve: (1984 + 100 + 450) / 1000 *.36 = 0.912A per Linear Regulator = 1.82A which, not coincidentally, was what I was achieving with the 5Ohm load. Clearly that's enough for 10Ohms which only needs 1.5A.

The annoying thing is that using LT Spice with LT components and an LT schematic (and my updated schematic) shows it working perfectly fine with the parallelised resistors! The lesson here is that you should still do the calculations if they are given, you can find them and understand them!

Doing the calculation with just the 5K pot would give a total of 3.996A. I desoldered the 3K3 and bingo. Sort of....rise time to nearly 3A (actually 2.9 eventually) was around a minute. In fact it never got there and was stuck at <15V and <3A (actual values following ohms law based on 5Ohm load) Not good - I have access to a very simple dummy load that was able to draw 3A out of it fairly easily - it wasn't stuck with 5Ohm resistance - so I knew it could do it.

I'd also looked at the feedback circuit with some more simulation - I mention this in a comment above in response to Jon. I can't yet change the Inductor or the Output cap but I have bodged in a 22nF cap across one of the 100K feedback resistors and re-routed the 1nF capacitor with a bit of wire (and cut the trace to its original destination). Firing it up and it shoots immediately to 15V, 3A! Yay. It still drops to 0V and 0A as well through the voltage and current controls.

Ok, so I think I've resolved the output functionality, leaving the noise. Here's some piccies. Firstly, ripple at the output terminal with the 5Ohm load. Pretty poor BUT I don't see the switching spikes.
Next, ripple at the output terminal with the 10Ohm load:
If you look carefully it is possible to see a bit of the switching noise just before the rise to a peak - I'd assume it is there on the 5Ohm waveform as well but the timebase is too large to notice it. Still poor ripple but the switching noise looks a lot more under control in my opinion (I'd be interested in other's opinion though.)

I also checked the voltage on the feedback pin of the Switching Regulator, Vfb. The whole point is to try and bring that under a bit more control. Prior to the bodges:
This is ranging from 460mV to 2.16V (10Ohm load). After the bodges with the 10Ohm load:
It's now ranging between 0.97V and 1.37V - firmly in the range for a 200kHz operation - I should say, I think it is still switching at 100kHz, I need to do more measurement. LTSpice (which I'm not sure whether I trust or not anymore) has this at 1.15 to 1.22V but I haven't modelled the output cap ESR or parasitics as per the actual build. Waveform is similar though. It gives me a bit more confidence that changing the Inductor and output cap for low ESR along with, say, a 20nF on the output, the Vfb would improve.

There's still the output ripple though. I need to find a way to squash that so this is what I intend to do:
- Redesign the PCB layout, particularly parts position and the grounding.
- Add in low ESR Input and Output caps for the Switching Regulator - I've found some that are 28mOhm.
- Add in a 20nF cap to sit alongside the Output Cap
- Provision optional pads for 2 more Output caps of 100uF (but could be other values)
- Change the inductor to 50uH to help reduce output ripple
- Test points for various pins so I don't blow it up anymore.
- Use small pots for the feedback circuit so I can make adjustments. I think I'd ultimately want to swap these for resistors at some point so I shall design around that.
I'm hoping to make use of LT's samples provision to get replacement parts to save cost! I want to keep the existing board as-is for comparison purposes and I can't be sure the parts I have are not damaged in some way. Remember, the first thing I did was solder the Schottky in the wrong way and blew up the Mosfet and sense resistor - it could just as easily have damaged the switching regulator.

I have been reading about proper grounding, including a paper from Analog Devices (LT.) It's difficult: there are definitive recommendations for the use of a ground plane but just as definitive recommendations for star grounding (Gene, for example, is a definite fan of star grounding as he's commented above.) Analog Devices are very much in the ground plane camp. One thing I'm fairly sure of, and open to correction, is that copper fills on the top layer that are via'd to the ground plane and used as a grounding plane by the top layer components is not a good thing - I think that's a capacitor. So I intend to tie ground pins with a short track and a via. Is it worth, then, 'filling the gaps' with copper on the top layer?

Finally, I will re-layout the PCB but I have previously gone with 2-layer to keep the cost down. That inevitably means I have to place tracks on the bottom layer which, if I go with a ground plane, will cause breaks - you can see in the images earlier in the comments. Should I bite the bullet and go for a 4-layer board - I could keep the current carrying tracks on one layer and signal tracks on a different layer?
- Cancel
- Vote Up +2 Vote Down
- Sign in to reply
- More
- Cancel

Children

No Data