LM2679 woes

[akis]

Some time ago I experimented with a LM2679 "simple" switcher. Well, "simple" my a**e.

The device works. But when you try to draw some appreciable current from it, it will start to ring and the voltage output drops. So after a lot of head scrathcing and half a dozen PCB layouts, and after eliminating the possibility of daft ICs bought on ebay, I discovered an application note (AN-1135) mentioning that depending on layout (yeah, right) and on the current being drawn, the output will ring in the 10s of MHz and they propose an RC to suppress it. The propose 10R/10nF as a starting point.

Well, I tried that and on pulling some current, the 10R resistor cooks. Literally. So I made it 100R. Still cooks. In the end I dropped the R altogether and went for a C, around 390pF does the trick and nothing cooks.

But why do we have all this ringing that needs to be suppressed in that way?

I thought it may be the catch diode, which they propose to be a Schottky, so it drops a small voltage and does not waste too much energy. On the other hand, the Schottky is very fast and may be partially responsible for the ringing, so what if I substituted it for a silicon diode?

Has anyone tried a LM2679 circuit drawing moderate amps and what diode did they use?

[void_error]

I think it has something to do with the ESR of the output cap(s) and maybe the capacitance of the Schottky diode used, but then again I might also be wrong.

[akis]

Yes I have read about that however I thought, I have no control of the output cap, it is not just what I place on the board, it is also the leads or tracks leading to a load, I cannot really control the load, and as a matter of fact my "load" has 2200uF caps as the first thing. So even if I make it stable by clever choice of output caps, the load , some load, would then make it unstable.

In any case I have tried various values for the output cap which sits close to the LM2679, the one I can control, I have tried 33uF, 100uF, and 470uF. I have settled on the 100uF but that alone doe snot solve the problem.

I tried to calculate the RMS voltages developed over the R (of the R/C ) based on the heat it was generating. At 205R and 2.4W it was at around 60C with ambient 22C therefore an increase of 38C. If we have 38C on 2.4W (rated so at 70C) so say we had 2.0W at 60C, therefore around 100mA RMS of HF, and wasted energy.

I suspect the diode I am using is partially responsible so far I have only tried Schottky diodes: SB5100, MBR10100 and STPS20SM100S. Today I hope to try a silicon like a 6A20G (is what I have in my box).

[mij59]

Never used the thing.
What kind of inductor are you using ?
A photo of the circuit might help.
Waveform of the input, output and schottkey diode voltage might be helpful.

[akis]

I have tried a dozen inductors, some even air wound, all rated for over 4A , some are rated 8A, some are toroidals, some are EPCOS vertical, they all behave in a very similar way, what differs is the current threshold  beyond which the oscillations overtake and suppress the output.

Today or tomorrow I will make a new version and will test it more thoroughly with non Schottky catch diode.

I attach the schematic and the PCB files, the top is shown sans the copper pour it would be too black other wise.

[akis]

And this is the "new" version (the other is two days old...) which I hope to build today (if I can be bothered to go downstairs make the PCB).

As you see I have allowed for two catch diodes, I will try both see which works without ringing.

[JohnnyBerg]

I have tried a dozen inductors, some even air wound, all rated for over 4A , some are rated 8A, some are toroidals, some are EPCOS vertical, they all behave in a very similar way, what differs is the current threshold  beyond which the oscillations overtake and suppress the output.

You have to be very carefull, to chose the inductor. When you draw current from your device, and the inductor saturates, you are playing a complete new ballgame!

The inductor can not be chosen by trial and error, but has to be calculated!

[akis]

I have used TI's own Webench Designer to calculate the inductor values - typically around 33uH and lots of amps, the more the better. But when they talk about trying to eliminate ringing, they talk about "we have found that values of 33uH and 22uF work well" so it appears that is not an exact science.

[mij59]

Could be caused by several things.

The board layout, too much inductance in the high power path, on the evaluation board of the Lm2679 all the high power part are close together with lots of copper.
High esr of the input and output capacitors, evaluation board uses ceramic capacitors.

What the part number  of the inductor your using ?

[MrAl]

Hi,

The ringing usually comes from the capacitance of the catch diode and any parasitic capacitance too and the way it oscillates with the inductance in the circuit.  The small inductance and very small capacitance create a tuned circuit and any energy put into this circuit in the form of a pulse turns into an oscillation that may or may not damp out (usually damps out after some time).  This shows up as a ringing on the scope, and the scope probe capacitance may make it worse.  A small resistance in series with the probe may show that the actual ringing is less.

If the slope of the switch is decreased then the ringing decreases, but that's hard to do because it is all done inside the chip.  So the only other way is to use a snubber circuit as you have been trying or insert a small series resistance into the inductor lead.  Obviously increasing resistance in the inductor lead will reduce efficiency so you'll have to test to see if it is still acceptable.  A small resistance may not hurt your application though, depending on the load current.  Try not to loose more than 1 or 2 percent if you try this, or any other method for that matter.

What else might help is a different kind of diode, like a zero recovery type.  What wont work very well is an ordinary Si diode use for power line rectification as they are far too slow and will kill the efficiency fast if the circuit even survives.  Of course a tight layout is a good idea too.

[void_error]

I wouldn't eliminate ground bounce as a cause for ringing. I'd also try to keep traces that carry high switching currents as short as possible and make them thicker.

[akis]

Hi,

The ringing usually comes from the capacitance of the catch diode and any parasitic capacitance too and the way it oscillates with the inductance in the circuit.  The small inductance and very small capacitance create a tuned circuit and any energy put into this circuit in the form of a pulse turns into an oscillation that may or may not damp out (usually damps out after some time).  This shows up as a ringing on the scope, and the scope probe capacitance may make it worse.  A small resistance in series with the probe may show that the actual ringing is less.

If the slope of the switch is decreased then the ringing decreases, but that's hard to do because it is all done inside the chip.  So the only other way is to use a snubber circuit as you have been trying or insert a small series resistance into the inductor lead.  Obviously increasing resistance in the inductor lead will reduce efficiency so you'll have to test to see if it is still acceptable.  A small resistance may not hurt your application though, depending on the load current.  Try not to loose more than 1 or 2 percent if you try this, or any other method for that matter.

What else might help is a different kind of diode, like a zero recovery type.  What wont work very well is an ordinary Si diode use for power line rectification as they are far too slow and will kill the efficiency fast if the circuit even survives.  Of course a tight layout is a good idea too.

OK so something like 0.1R in series with the choke, it is also in series with the output, so yes there will be a loss there, so at 0.1R and 2A it will be 200mV. But if the resistor is placed at the output of the LM2679 then the feedback is taken at the other end of the inductor, so the regulation should still work?

[akis]

With just a cap across your rectifier your energy in the cap will be dissapated in your switch so check the temperature rise of your controller. If you use an RC snubber the energy in the cap gets dissapated in the resistor, power is just energy x frequency. See this for selecting proper snbubber components.

http://www.ti.com/ww/en/analog/power_management/snubber_circuit_design.html

OK thanks for that. I have established that the power to be dissipated is around 2Watts based on the temperarure rise of the resistors (not very accurate). At the same time the LM2679 is delivering 23.25V into a 15R load thereby 36W of power on load. So we have 2W/36W waste as heat, before we even look at the LM2679 itself but this barely gets over 30C.

This very useful link you sent me basically establishes that the energy has to be dissipated somewhere and that is across the resistor.

In my case the P(Rsnub) seems quite a lot at 2Watts.

[JohnnyBerg]

The diode you are using is hard to read in the schematics?

Have you tried with other diodes? Perhaps try a MUR460?

[akis]

I have redesigned with thicker lines etc, however an older LM2679 design I have already used very thick copper pours and has same issues. So I am not hopeful that I will be able to get rid of the snubber capacitor.

[georges80]

It this a one layer PCB??

cheers,
george.

[akis]

No, I am only showing the bottom where I made the traces much thicker. The top is almost the same as in my previous post.

PCB is made, drilled, and sprayed with Flux. Waiting 12 hours to dry properly before assembly. Will know how it works tomorrow.

[georges80]

No, I am only showing the bottom where I made the traces much thicker. The top is almost the same as in my previous post.

PCB is made, drilled, and sprayed with Flux. Waiting 12 hours to dry properly before assembly. Will know how it works tomorrow.

Where is the top layer copper? I don't see the actually traces/routes. I personally would have a large flooded copper area for ground and then route the rest as needed. I notice though you are using thruhole components which does tend to chop up both layers.

It's not enough to just make traces thicker, you need a good ground plane to all components versus what appears (at least only on the bottom layer) to be a ground trace.

I have designed lots of dc switchers (constant current LED drivers) and layout is critical as power levels increase. I use smt for all my designs since a) most modern components are only available in SMT and b) you reduce a lot of inductance in traces/leads and c) routing is a lot easier and you don't mess up all layers of your copper.

e.g. here's a driver capable of >100W output and is only 1.3" in diameter:



cheers,
george.

[akis]

I will do measurements tomorrow if I build it. I measured (by heat generation) 2W on the R with a 10n C, yes. Then I shorted the R and used 390p in C. The top layer was shown on a previous post. All the traces are at the bottom layer, the top layer is just a ground plane. I have no tools to allow me to work with SMD components unfortunately, and the parts are so small that resemble debris from the soldering on my desk (once I received some parts which were almost not visible by naked eye). My eyesight is not what it used to be, so now I need a strong light and a magnifying glass sometimes to be able to examine faint writing on parts etc. I presume that working with SMD requires a lab set up specially for it.

[georges80]

SMD just needs an ok magnifier (5x sort of thing), solder paste, hot plate, hot air tool etc for prototyping. It's actually very easy to work with as long as you don't go smaller than 0603 for R's and C's.

I'm >50 so it's not as if my eyesight is spectacular either

I assume these boards are not plated thruholes since you are home making them?? If that's the case how are the caps (the big electros etc) going to solder to the top plane (your ground plane)?

cheers,
george.

[akis]

You must belong to a small, and privileged minority of hobbyists that can do SMD work at home to your level. Most people cannot. There is a guide/article at ESP (Rod Elliott) discussing class D audio amplifiers that says, "do not bother, they require to be built with SMD due to the high switching frequencies", implying none at home can do SMD.

Regarding your LED drivers designs, do you take power straight from the mains? If yes I would be interested in a design that can provide 24V (ideal is 23.25V on load) at around 1.25A long term and maybe up to 1.9A for shorter periods. But there must be complete mains isolation, no connection to the mains earth at all, truly floating. It will power a device through 4 metre power leads (talking about lead inductance and capacitances). And there will be two 24V supplies to provide a +/-24V rails for my device, so they must be floating, as I said. Do you have a design that is small, portable, and completely (mains/earth) isolated? Currently I use toroidal transformers at 1Kg each almost... And the toroidals I think have capacitive coupling with mains so they too are not perfect. Still miles better than the LED SMPS I bought on ebay...

My PCBs are old-fashioned through hole made at home. For tracks running at the top layer, I make sure the necessary pins can additionally be soldered on top, some components are bulky and cannot, eg potted relays, most ICs etc. For the capacitors you are asking, it is a pair of 2200uF/35V caps which I am placing horizontally on top of each other, because they do not fit vertically in the box. That means I bend their leads 90 degrees and I have good access to solder them both at the top and the bottom layer. No need to plate holes since I solder the lead on both sides. If I cannot, due to access, I usually insert a via to connect top and bottom tracks.

I attach the latest top layer which I hope to assemble and test this morning. The Silk later shows holes and tracks bnot not the copper pour for clarity.

[MrAl]

Hi,

The ringing usually comes from the capacitance of the catch diode and any parasitic capacitance too and the way it oscillates with the inductance in the circuit.  The small inductance and very small capacitance create a tuned circuit and any energy put into this circuit in the form of a pulse turns into an oscillation that may or may not damp out (usually damps out after some time).  This shows up as a ringing on the scope, and the scope probe capacitance may make it worse.  A small resistance in series with the probe may show that the actual ringing is less.

If the slope of the switch is decreased then the ringing decreases, but that's hard to do because it is all done inside the chip.  So the only other way is to use a snubber circuit as you have been trying or insert a small series resistance into the inductor lead.  Obviously increasing resistance in the inductor lead will reduce efficiency so you'll have to test to see if it is still acceptable.  A small resistance may not hurt your application though, depending on the load current.  Try not to loose more than 1 or 2 percent if you try this, or any other method for that matter.

What else might help is a different kind of diode, like a zero recovery type.  What wont work very well is an ordinary Si diode use for power line rectification as they are far too slow and will kill the efficiency fast if the circuit even survives.  Of course a tight layout is a good idea too.

OK so something like 0.1R in series with the choke, it is also in series with the output, so yes there will be a loss there, so at 0.1R and 2A it will be 200mV. But if the resistor is placed at the output of the LM2679 then the feedback is taken at the other end of the inductor, so the regulation should still work?

Hi,

The efficiency loss due to the series resistance increase when in series with the inductor itself is roughly:
(Iout*Rs)/Vout

where Rs is the extra added resistance.

So for Rs=0.1 ohms and Vout=10 volts and Iout=2 amps, the loss is about 2 percent more.
For Rs=0.1 and Vout=5 and Iout=2, the loss is about 4 percent more.
So obviously there is a limit to how high you can go on the added Rs.

Putting the Rs in series with the chip isnt the same as putting it in series with the inductor.  With Rs in series with the chip it may not help as much when the internal switch turns off because then the resistor is not in series with the diode anymore.  You could still try it though.

[akis]

I have just made another prototype. I did not place a snubber, wanted to see how it behaves without it. It is fed through the bench PSU and not a transformer/bridge. However it seems to be behaving OK, up to around 1.6A that I tried. Of course without a snubber R/C or C, I am afraid it can at any time stop working.

I attach the schematic and the oscilloscope over the catch diode's cathode, at about 1-2 mm distance (cannot touch it).

As you can see there is a bit of overshoot, not sure how to proceed now. Do I add the cap (my typical 390pF) or do I just leave it out?

[mij59]

Why do you think is could stop working?

You could test the circuit with e.g. the 390 pF cap in series with a 10 Ohm resistor, or a 10 nF cap and  see what the effect on the waveform is.

Are you using a 100X probe ?

[rx8pilot]

How are you measuring the switch node that we see in the scope shot? I did not see any discussion on the probing technique that will have a huge impact on what you see on the screen. You MUST be using a very low inductance ground lead on the probe. If you are already doing that, perfect. If not, the majority of ringing you see in the SW node of the circuit is being amplified by the probe and you may not have any problem at all.

On the PCB layout, you have some R's and C's between the SW output of the LM2679 and the inductor - this will lead to ringing. The distance between the output and the inductor/diode must be as short as possible. Taking precedence over all the the other support components. The closer you can get the lower your LC parasitics will be. Some of those parasitics are in the devices, but you cannot control those other than choosing alternate components. In general, getting the PCB layout to prioritize the current paths/loops FIRST can eliminate or at least drastically reduce the snubber dissapation. The RC snubbers should be a last resort since they kill efficiency to do the job that could have been done with a better layout. Also, on the output C7, I would consider putting that very close to the inductor output to catch any ringing before it gets into a trace that has its own resonance that would be added.