Flyback power supply instability issue

Question

I have created a switching power supply:
Input: 230 V AC
Output: 12 V DC, 1 A
Controller: UC3844
Topology: flyback
Switching frequency: ~ 100 kHz

Schematic:

When I start this power supply, output voltage oscillates between 11 V and 14 V.

Switching is turning on and off (blue line = signal at switching transistor)

When I put some load on it (like 100 mA) then oscillations stop. And oscillations won't return after I remove this load, output voltage remains stable after that.

I assume this instability has something to do with feedback compensation (R120, C118 close to TL431), but I am not sure.
What could be causing this?

UPDATE 1:
Current sense waveform (when oscillating after power up):

Voltage at drain of switching transistor (when oscillating):

Current sense waveform (loaded, 400 mA, not oscillating):

Voltage at drain of switching transistor (when loaded, not oscillating):

Could you please put the primary current (i.e. the voltage across the sense resistor) and drain voltage oscillograms for both unloaded and loaded states? — Rohat Kılıç, Aug 12 '21 at 18:20
The first thing I'd say is that the converter operates in CCM even in unloaded state (look at the primary current). Remember that stabilizing a converter running in CCM is hard compared to that in DCM. So, you may want to try decreasing the primary inductance to make the converter run in DCM for the full range. Additionally, the response of the converter seems slow. So you may try decreasing C118 to a few nF. — Rohat Kılıç, Aug 12 '21 at 19:03
Could be a inductive path from IC output to the gate that is too high - long trace,..etc. You could increase the gate resistor to see if it helps. — Marko Buršič, Aug 12 '21 at 19:43
@RohatKılıç I have changed C118 to 1nF. Oscillations are gone. But now the power supply is squealing, there is heavy pulse bunching. So I changed R120 to 22k. Now it is quiet at no load, squealing between 100mA and 500mA, and silent between 600mA-1A — Chupacabras, Aug 12 '21 at 19:46
Classic CCM -> DCM feedback issue. How did you calculate your feedback network? — winny, Aug 12 '21 at 21:43
@winny Unfortunately I did not calculate it. I could not find any practical example how to do that. All I saw were just general explanations about zeroes and poles. But I could not find anything how to take this circuit and to define zeroes and poles actually calculate values from that. So I googled some schematics for UC384x controller and TL431 with type 2 compensation. — Chupacabras, Aug 12 '21 at 21:54
Forget the zeroes and poles for now. They are the last things to care about. The power level you are working on is quite low, so you can run the converter in DCM for the full operating range (i.e. from unloaded to fully loaded state). There are tons of online articles and tutorials for DCM flybacks. Redesign the transformer and recalculate the primary current sense resistor accordingly, then test. Regarding the buzzing, there could be several reasons: Magnetostriction (xfmr), ceramic cap vibration (snubber), or any other inductive element. You can use diode-Zener snubber instead, by the way. — Rohat Kılıç, Aug 13 '21 at 05:21
Try adding a 1k resistor at the output so that your PSU always runs with some minimum load. You said that the oscillations go away once you connect a load, right? — Prathik Prashanth, Aug 13 '21 at 05:54
There’s your problem! You need to calculate it for your specific situation, manly C and L at the output — winny, Aug 13 '21 at 06:06
@PrathikPrashanth actually I have 10k resistor soldered at the output, so I can easily connect DC load. Oscillations were solved by changing C118 from 100nF to 1nF. Now squealing is the new problem — Chupacabras, Aug 13 '21 at 11:21
@winny I really would like to learn how to calculate it. But I was not able to find any "tutorial" on the internet. I learned about flyback and forward topologies, about transformers, I was able to build my power supply and wind my own transformer for it, but I was not able to find and learn how to calculate that feedback network. I really would like to learn it. I hate to not know it. — Chupacabras, Aug 13 '21 at 11:26
@RohatKılıç Duty cycle is limited to 50%, so transformer should demagnetize every cycle. So IMHO it already runs in DCM. — Chupacabras, Aug 13 '21 at 12:07
@Chupacabras 50 % duty cycle does not guarantee DCM. Simulate it! — winny, Aug 13 '21 at 12:17
@Chupacabras nope. DC limit is not an indication for CCM. DC is limited to 0.5 by [UC3844](https://www.onsemi.com/pdf/datasheet/uc3844-d.pdf) itself. Also, look at your primary current and drain voltage waveforms again and compare to [these](https://media.monolithicpower.com/wysiwyg/3_45.png). For DCM, the primary current should look like triangular pulses instead of pulses with ramp-tops (like yours) and the drain voltage should look like pulses with damping oscillation like [these](https://media.monolithicpower.com/wysiwyg/4_39.png). — Rohat Kılıç, Aug 13 '21 at 13:34
@Chupacabras also check DN 2013-01 from Infineon and AN-4137 from Fairchild/ON Semi for DCM and CCM flyback design. — Rohat Kılıç, Aug 13 '21 at 13:38
If you want to learn about the flyback converter and a way to stabilize it, have a look at my [seminar](https://cbasso.pagesperso-orange.fr/Downloads/PPTs/Chris%20Basso%20APEC%20seminar%202011.pdf) taught in 2011 at an APEC conference. In your case, the issue could be located in the FB path. You should turn the internal op-amp off by a) grounding the FB pin and b) load the CMP to \$V_{ref}\$ via a 10-k resistance. Then ground the opto emitter and connect the collector to the CMP point with a 1-nF cap. closely located between CMP and the IC Gnd. The opto emitter must go straight to the UC GND. — Verbal Kint, Aug 14 '21 at 20:30

score 3 · Accepted Answer · answered Aug 14 '21 at 21:00

The UC384x series hosts an internal operational amplifier (op-amp) whose output is capable of delivering 1 mA per the data-sheet specs. This has been done on purpose to disable the op-amp and let an external signal drive the FB-to-CS path directly when the regulation is kept on the secondary side as in your case. And this is what you need to do in your application: disable the op-amp and have the opto drive the peak current directly. This is what I have shown in the below picture. It is an industry standard I have seen in many high-volume designs:

The 1-nF capacitor introduces a high-frequency pole and rolls-off the gain. This cap. together with the TL431 and the single 0.1-µF capacitor, form a type 2 compensator. Please note the direct connection of \$R_3\$ to \$V_{out}\$ which is wanted in this implementation. If you disable the fast lane via a transistor as in your circuit, then you need another \$RC\$ across the 0.1-µF cap. Nothing wrong but you need more components and it is not needed here. If you stick to the circuit I drawn - which is an industry standard - a single cap. is ok and the given value should roughly do the expected job for you. See my seminar The Dark Side of the Flyback Converter for more details or my recent APEC 2018 seminar available from my webpage.

The 1-nF capacitor should be placed very close to the IC, between the CMP and GND pins of the UC384x. Same for the GND copper trace to which the opto emitter connects: it must go straight to the UC as an individual string. No connection to a noisy path or instability is likely to occur. See the below recommended layout for another controller in which the control pin is also pulled down for regulation purposes:

Excellent! The suggested changes work pretty well. Output voltage is stable and the audible noise is gone. I went through your presentations, half of the stuff I understand, the other half will take some more time :) It's very motivational to get help from guru like Chris Basso. — Chupacabras, Aug 18 '21 at 17:48
Hey, glad if my suggestions worked and happy to help when I can! : ) — Verbal Kint, Aug 18 '21 at 20:25

Flyback power supply instability issue

1 Answers1