Duet 3.3V supply quality

@dc42: Is it just me, or does everyone else get "Image error, Running getimagesize() on the image data failed" on your pictures? They don't work in Firefox, Safari or Chrome on Mac...

Ian
RepRapPro tech support

Quote
dc42
I agree, it doesn't look easy to add a big capacitor in parallel with C3. However, I've found 470uF 25V and 680uF 16V SMD capacitors that will probably fit in the C3 position, and 330uF 16V caps that will definitely fit. I'll add some to my next Farnell order. Or maybe just get an ATX PSU with better resistance to brownouts.

FWIW It would certainly be possible to bolt a pretty hefty capacitor outside the Duet enclosure upside down by the z-motor, below the bed and tee its terminals into the power supply leads just where they enter the Duet enclosure.

( Watching this thread with interest but without much knowledge... but now a bit wary of firing mine up at all! )

regards
Andy

Quote
droftarts
@dc42: Is it just me, or does everyone else get "Image error, Running getimagesize() on the image data failed" on your pictures? They don't work in Firefox, Safari or Chrome on Mac...

Ian
RepRapPro tech support

Hi Ian, I've complained about it here.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Further to my earlier post regarding the noise I see on the 3.3V rail when the bed heater turns off, I disconnected the scope probe that I was using to trigger from the mosfet drain signal and nearly all the noise went away. So it seems that what I was seeing was coupling between the scope probes via the (short) common ground wire I was using to connect to the scope prove earth clips to pin 2 of the expansion connector. However, I did notice some droop on the 3.3V rail after the bed heater turns off:



Note the change in horizontal scale (50ns/div before, 100us/div now) and verical scale. So the 3.3V rail appears to dip to around 3.0V when the bed heater turns off, then it recovers to about 3.2V.

[EDIT: I now think that this 3.3V supply droop is a measurement artefact, see my later post.]

Edited 3 time(s). Last edit at 12/29/2013 03:24AM by dc42.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Radian

Not likely to get the scope out till tomorrow. Just looked at the regulator and it has 21RA NO5A markings. The no5a seems to indicate an LM1117MP-3.3, 21RA being a date code apparently. Do you have the same? As DC42 has said with high frequency scopes there can be a tendency to see things which are perhaps not there when the scope is not connected. RF pickup in the scope ground lead was mentioned. Either that or we can now see things which in the past we couldn't so we never woried about them. I wonder what current is drawn on the 3.3V rail I did have a quick look but did not get far? The device is 800mA I believe and also has thermal protection. I suppose it could be that one of these is playing silly transients. You do hear of fake components so It could be anything I suppose?

Rory

Edited 3 time(s). Last edit at 12/28/2013 07:46PM by Rory166.

1. I've just checked the 3.3V supply with a DMM across pins 2 & 3. It reads 3.28V whether the bed heater is on or off. So I think the apparent drop in the 3.3V supply on the scope when the bed heater goes off is caused by the fact that ATX supplies have the 0V line connected to mains ground. which unfortunately means you get a ground loop when you connect a scope earth lead as well.

2. Radian's observations on the 3.3V supply here are worrying. I'd put it down to measurement artefacts, but the fact that he doesn't see the same thing on the 5V rail when he uses the same ground pin suggests they are genuine. Maybe his 3.3V regulator is marginally stable, due to a faulty regulator or to faulty or badly-soldered output capacitors.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

@kwikius (and others that have emailed me, worrying about this thread): Please don't be afraid of turning your printer on. The guys here are talking about a specific issue, and many customers HAVEN'T experienced problems printing. While the work here may inform improvements to the Duet board and/or power supply, it doesn't mean that the printer won't work. We will work with every customer to achieve a working printer!

There have been issues with reprap electronics in the past; the electronics are often sensitive to sudden voltage changes, and earth loops. For example, see 'random disconnections' here: [www.reprappro.com]

Ian
RepRapPro tech support

I would add to Ian's comments that I've only had printing stop unexpectedly once, and the mains power where I am has been a little flaky during the high winds over the past few days. Ideally I'd get a UPS, but that is beyond my budget right now. Hence one of my mods is aimed at improving tolerance of brownouts. The 3.3V supply on my board appears to be stable when the bed heater turns on/off, so I think Radian's board has a specific fault.

Edited 2 time(s). Last edit at 12/28/2013 07:41PM by dc42.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Quote
droftarts
@dc42: Is it just me, or does everyone else get "Image error, Running getimagesize() on the image data failed" on your pictures? They don't work in Firefox, Safari or Chrome on Mac...

Ian
RepRapPro tech support

Hi Ian, I've solved the image error problem by only using links to my Dropbox public folder, and I've updated all the relevant posts that had this problem.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Quote
droftarts
@kwikius (and others that have emailed me, worrying about this thread): Please don't be afraid of turning your printer on. The guys here are talking about a specific issue, and many customers HAVEN'T experienced problems printing. While the work here may inform improvements to the Duet board and/or power supply, it doesn't mean that the printer won't work. We will work with every customer to achieve a working printer!

There have been issues with reprap electronics in the past; the electronics are often sensitive to sudden voltage changes, and earth loops. For example, see 'random disconnections' here: [www.reprappro.com]

Ian
RepRapPro tech support

No worries! I always get like this before I apply power to a board for the first time.

regards
Andy

---

Ormerod #318
www.zoomworks.org - Free and Open Source Stuff

Interesting discussions.
I have had similar noise issues and scope probe problems in the past (on a variable speed stepper motor drive project) with my battery/mains powered scope (old Tek). When the scope was mains powered any supply noise was injected into the circuit via the scope probes! When battery powered the probe/leads and scope acted as a large aerial, picking up my electrical noise and injecting that!

My cure for the measuring issues was to use some clamp on ferrite cores on the scope probe leads near the bnc (with a tie wrap to stop it sliding).

For the actual circuit I was amazed at the reduction in electrical noise and spikes using these at switching end of cables, they certainly removed the narrow/fast spikes and with the scope probe mod provided some quality signals.

They also do some ribbon cable ferrite clamps [uk.rs-online.com] is a 40 way example.

Just to add when using an mbed LPC1768 [mbed.org] for a data logging product, USB lead conductive noise was a problem with some laptops (the older ones seemed better!), the project was a data logger and the (ARM Cortex-M3) A/Ds on the chip were very sensitive to any noise, but once again these clamp on ferrites came to the rescue and I just noticed they even do a specific USB lead version now [uk.rs-online.com]

So these may be worth considering as relatively simple add ons with no PCB mods.

Treth

Very intersting. I have never heard of anyone putting ferrites on their scope leads before. I have recently purchased a Hantek battery powered 60 MHz 1062B but yet to use it in anger. The chinese throw in FFT while TEK make you pay through the nose for a dongle. My previous employer was spending so much on Tektroniks they decide it would be cheaper to buy the company instead, so they did. Just my light hearted view on the take over you understand. I used to make the calibrators for scopes and our boffins decided the only way to get a good display of an edge was to generate the edge right on the tip of the BNC socket, the so called active head. They pushed the speed up to 13 GHz last thing I heard. This technology was probably one of the reasons we were taken over by Fluke who were still trying to push edges down coax.

We all know from experience that RF does weird and wonderful things, who hasn't put the coax plus on the TV near the socket and had the picture come on without actually connecting it. I f you look inside a UHF tuner it's a different world looking more like plumbing than electronics. The earth lead on a scope probe looks very much like an antenna at a certain frequency.

Rory

Rory166

Ferrites on scope leads might not be considered good practice, but it worked for me with my low frequency application.

The bigger question with battery powered test instruments is what this 'lump of electronics' and metal looks like to your circuit under test to which it is connected via a scope lead?

Just to add I also placed a sheet of copper clad board connected to mains earth (not a good safety practice) under the scope. So my aim was for the HF noise, block the route via the scope probe to my circuit with the ferrite and provide an alternative capacitive route for noise to earth.

I like your 'take over' story!

Treth

I assume it is normal to use attenuated probes whenever possible to minimise cicuit loading. Normally 10:1 but 100:1 does exist and probably more. At work we were always told to have 10K resistors in our DMM voltage test probes but we never bothered.

I have heard of something called FET probes. Not sure but if you put a Mosfet on the end of the probe with just the gate connecting to the UUT circuitry you would minimise any circuit interference. Does sound very vulnerable to ESD though. Must look this up.

Rory

Edited 1 time(s). Last edit at 12/29/2013 05:39AM by Rory166.

Hi All

Just attached a scope to my duet.

Confirm Radian's results entirely. Infact I saw negative excursions on the +3.3v rail.

Radian does not have a one off faulty board.

My setup had no connection to the mains being powered from my laptop via usb, the scope also running on battery.

It seems that significant RF can be picked up in a usb lead but the fact that the 3.3v regulator appears to amplify them is not at all desirable. I cannot rule out that this is some kind of measurement artifact, but I do not favour this explanation on the balance of probability.

Rory

There is certainly a difficulty in measuting these sorts of glitches. The primary problem is with a single ended coax scope lead which makes it sensitive to EMI... The outer screen of the coax "sees" EM but shields the inner conductor from the same induced voltage resulting in noise imposed (in series) on the signal. A true differential probe would solve this but they are very expensive. I've been using the dirty trick of running two scope leads (twisted together so they both get exposed to the same field) with one probing the node under test and the other probing the common ground point to serve as a reference. Then by using the scope's math function to subtract the "ground" from the signal a better representation can be obtained.

My subsequent measurments have all been taken like this but It's also possible to reduce the pickup by using an overall grounded shield for the test leads, which I had already made from a large folded aluminium sheet I used to test a metal detector design. Depending on ground currents, one or other method can be found to give less in the way of glitches (hence demonstraring that what you see isn't necessarily what's really going on). In the case of the Duet traces I posted earlier I used the metal detector rig. While exagerrating the issue, it is still very much the case that the 3.3V rail is far from being a flat-iron.

@Rory
21RA NO5A is indeed the case code on my LM117. I do wonder if this part could do with an upgrade - it's referred to as an LDO, but isn't really so much. By feeding the 5V from a variable bench supply, the drop-out voltage appears to be around 1.2V which explains why it can be fussy about being powered from the USB lead. And there's a diode (D13) between the USB 5V and the LM117 (ignoring for the moment the SP2525) which my DMM says drops another 0.36V. So with 3.3 + 1.2 + 0.36 we need 4.86V coming in over the USB cable to *just* hit 3.3V

So what about the SP252? It 's a 70mOhm high-side a switch that could bypass D13 if it was enabled (the mosfet is bi-directional). However, it appears to be diabled all the time (why fit it I wonder?). It may be intended for use if the Duet board was configured as a host - maybe to read USB sticks? In which case it would manage 5V out of the USB port - with proper current limiting. Highly commendable! But still, I wonder if there would be any downside to enabling it all the time to give a very handy 0.34V extra headroom when connected as a slave?

---

RS Components Reprap Ormerod No. 481

Yes, the LM1117 doesn't look like a good choice of regulator, because of its high dropout voltage. Even at lower currents, the dropout voltage is still 1V. The best drop-in replacement I found (i.e. same pin configuration and at least 500mA rating) was the Microchip TC2117-3.3VDBTR.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Radian

Have you measured the current drawn by the duet powered from usb? My laptop gave a message saying usb cevice drawing too much power which might indicate exceeding 500mA and that was with nothing connected like IR position board for example.

I agree the regulator might be improved, RS have a 1A device which although slighly bigger might well fit. 36p but you have to buy 25 off.

I am thinking to try the ATX supplying 5V and then 3.3V as well. It may be necessary to solder onto the regulator tab if no better connection point can be found.

Rory

@dc42
I like the look of that Microchip reg. good find. RS don't stock them but Farnell do

@Rory
I've only ever seen the Duet draw around 250mA from the 5V rail. But there's all kind of things that could go on during power-up to briefly add to the load.

Currently my "best 3.3V" is achieved using the supplied ATX with a 2200uF cap across C1/C3 and no USB connection. Without this the Duet typically hangs during more prints than not, but even with this added C it hangs every now and then. Ian has kindly offered to ship me another Duet which will greatly help with the elimination process.

---

RS Components Reprap Ormerod No. 481

Hi All

Thanks for the detailed investigation Radian, Rory, DC and others! I am a bit late to the party due to being away from home - turns out bringing my laptop along was a good idea after all. For info Think3DPrint3D use the Alpine 600W PSUs similar to these ones: [www.amazon.co.uk]
with modifications as shown by Nophead: [hydraraptor.blogspot.co.uk]

Needless to say we did not see the same issues in that configuration.

The on board switching power supply was one of the earlier elements of the design - I have done a quick cut and paste from the lab book for testing of an earlier version of the design (where the power supply circuitry was very similar). That info is live here:
Duet FETs, PSU and Stepper Driver Testing

Unfortunately I wont have the opportunity to do my own investigations until the new year but I really appreciate all the work guys!

Cheers

Tony

---

[blog.think3dprint3d.com]

Hi T3p3,

I don't think there is much wrong with the switching regulator, other than it could do with C3 being a lot larger to better tolerate brownouts (because the 12V supply is going to drop very quickly if the bed heater is on), and there should preferably be a 10nf capacitor in parallel with the 3k92 feedback resistor to improve transient response. The issues seem to be to be mostly around the 3.3V linear regulator: (1) the dropout voltage of the LM1117 leaves very little margin for the 5V supply being low, and (2) some users have reported that small transients on the 5V supply lead to much larger transients on the 3.3V supply.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Your conclusion is probably correct.i.e. ground noise and the relationship between scope ground point and the local ground of an I.C. can result in misleading scope traces.
The only way to deal with this, in order to see what noise an IC is actually experiencing, is to use two scope probes and the scope's maths function.
This is not easy to explain in a few sentences but here we go..

Both scope channels must be set to exactly the same gain settings and AC coupled.
set channel 2 to inverted.
set the scope to 'add' both channels.
If the two scope leads are now connected together the result should be a straight line trace. (i.e Ch1 and Ch2 (Inverted) have the same gain and any common noise will cancel.
If you get the channel gains balanced then you should be able to move the volts/div down into the low mV range and still see a straight line.
Most two channel scopes have the ability to work this way.
Now connect channel one probe directly to the Vdd pin of an ic and channel 2 probe to the same IC's Vss pin. The scope leads' ground croc clips connectors should be linked together but not connected to the PCB
What you should now see on the scope trace is the power noise that the IC is experiencing.
Getting this measurement to work can take a bit of trial and error but this is the way that I check for some EMC issues on PCBS.

and ..Sometimes the safety earth of the scope can introduce problems and be unhelpful but I mustn't tell you to temporarily disconnect this (!)
Ray

Hi Ray

So you are saying if we do the differential technique on the 3.3v o/p and round of the regulator and we still see this problem then it's really there? No worries abut the safety earth on my scope it hasn't got one.

Rory

I got around to adding 100 ohm resistors in series with the bed mosfet and hot end heater mosfet gates. I also removed the flyback diode that I added earlier. Here are some before-and-after pics of the signals on the mosfet drains. First the hot end heater without the gate resistor:



Note that the vertical scale is 10V/div in the first picture and 5V/div in the second. The turn-off has changed from sub-20ns to greater than 200ns. The peak drain voltage has decreased to 30V and the mosfet is no longer avalanching.

Here are the photos for the bed mosfet drain:



Again, the turn-off time has increased to more than 200ns.Peak voltage is about 45V (more than the 40V rating of the mosfet) and the mosfet appears to be avalanching slightly, but in a more controlled way. BTW avalanching is not damaging to the mosfet, provided the avalanche energy is within its rating (as it is here).

I took a look at the 12V supply too. Here are the before and after pics when the bed mosfet turns off:



Note the change in vertical scale from 2V/div in the first pic to 500mV/div in the second. The initial 3V spike and the high frequency ringing have gone.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Just another thought on the power supply arrangement.

Facts:
With 5V feeding the VDO 3.3V supply the operation is marginal.
This is more critical with USB power due to the typically lower "5V" supplied, but for 3D printing we need the ATX supply as this is when the issues are typically observed

Examining the circuit diagram the +5V is supplied via the switching circuit from the 12V input. The +5V is also fed from other sources via diodes.
This +5V line feeds the SP2525A, and the MOSFET drive circuits used to drive the fan and EO, plus the expansion connectors in addition to the 3V3 regulator.

My thoughts are why not re designate the +5V to +5V5 or even +6V by changing R60/R61 for this new voltage. Even a 0.5V increase will significantly increase the margin.

Risks:
SP2525A is spec'd to 7.0V and even has characteristics plotted to +7V, so should not be an issue.
MOSFET drive circuits should be OK with this small change.
The expansion connector is the only unknown. It may be possible to add a diode in series with this later if required to drop closer to +5V.

So a simple change of resistor could provide a significant change in operating margin. With the suggested capacitor in the feedback circuit and additional decoupling we could have a reliable circuit when powered by the ATX supply.

Thoughts?

(Off topic) How do you set the forum to view 'latest post' first, or quickly navigate to the last post?

So dc42, clearly slowing Bed moseft turn off is good!

FWIW Here is my proposed mod to the power supply.

From the schematic its clear there is no decoupling capacitor at Vin. Meanwhile we have supply leads from the ATX PSU that are maybe 600 mm so quite a bit of inductance. Add to this the Inductance of the bed heater leads and heater leads. etc. ect.

So my proposal is to hang some hefty capacitance off the input leads where they enter the enclosure

I ignore the existaing capacitance at the ATX PSU, which I believe is stated by Rory166 to be 2200 uF somewhere, but its 2 ft away down the leads. I assume the ATX PSU can handle the large extra capacitiive load I plan to add! I also ignore the hot end heater and concentrate on the Bed Heter Lets keep it simple.

So some numbers. Lets say we have a current of 20 A with bed heater on.

Assume we want a max voltage drop of 6 V, when the Bed heater switches on (So 5v reg has enough headroom) Lets also arbitrarily assume we want to supply that for 1 ms.
My math gives a required Capacitance of around 3000 uF.

The max ESR of the Cap allowed is 6 V / 20 A == 0.3 R which I think is achieveable. Just to be sure I'll get 3 1000 uF caps and put them in parallel. ( Maplin dont specify ESR)

Anyway off to my local Maplin to see if they have any. Unfortunately I dont have a digital scope so probably wont be able to verify if this is a good mod.

Feel free to jump in and hack my simple maths and plans to pieces !

regards
Andy

Edited 2 time(s). Last edit at 01/01/2014 04:58AM by kwikius.

---

Ormerod #318
www.zoomworks.org - Free and Open Source Stuff

Hi Treth,

I think the 3.3V regulator dropout margin is sufficient when it is running from the 5V switching regulator output. I haven't seen any problems when I power the Duet this way (although I did make a couple of mods). The 3.3V regulator dropout voltage specification is 1.25V max @ 500mA and the output voltage is 3.365V max, giving a worst-case margin of 0.385V when operating from 5V, and a typical margin of 0.55V. The 5V switching regulator output could itself be low due to tolerances of the resistors and the regulator, and it is worth measuring with a DMM.

However, if you provide the 5V power either from USB or from the ATX 5V supply, then there is a SS14M diode in series, with a typical voltage drop of 0.45V @ 500mA. So the margin is now 0.1V typical and negative in the worst case. The LM1117 definitely needs to be replaced by a regulator with lower dropout voltage.

If you do want to increase the output voltage of the switching regulator, then I agree with your risk analysis.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Hi All

DC has sumarised the situation well. Using the 12V rail and internal converter there is an inadequate margin. With 5V and USB there is no margin.

Has anyone tied the following? Put neither link in place,ATX 5v nor enable 5v converter. then feed 3.3V to the tab on the 3.3V regulator from the ATX power board.

Rory

Edit I have clearly got this wrong the 5v from the ATX would need to also be connected and the link in place.

Edited 1 time(s). Last edit at 01/01/2014 05:45AM by Rory166.

You could feed the 5V from the ATX power board into the ATX_5V_EN jumper for the same effect.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Quote
dc42
I think the 3.3V regulator dropout margin is sufficient when it is running from the 5V switching regulator output. I haven't seen any problems when I power the Duet this way (although I did make a couple of mods).

If those mods include adding a few thousand uF across C1/C3 to keep the ATX 12V from browning-out then I mostly agree. I've found this to be absolutely crucial to minimise sudden lock-ups during printing. Unfortunately it hasn't made it 100% reliable but I report that on the basis of only a couple of lock-ups since I did the mod.

I'm still suspecting another vector to PSU related lock-ups because with the additioanl input capacitance in place, the Duet will tick-over all day and night without crashing so long as it's not actually printing. Slowing the high-current switching edges may help with this. I'm now investigating the big flat ribbon cable carrying the 10A to the bed heater to see what kind of EM field it creates. The high flyback voltage, fast edges and large surface area may make it a candidate for capacitive coupling.

---

RS Components Reprap Ormerod No. 481