Need help with a UC3843 based SMPS repair

[SCR]

I received an UC3843 based SPMS (driving an audio amplifier) for repair, and it turned out slightly more complicated than anticipated, so would be happy about some help and inspiration...

According to the user, the amp kept blowing fuses as soon as it was switched on. The PCB looked fine without any visual damage of the components. I started checking components and found out that R521 (gate resistor between UC3843 and MOSFET) was open, also U506 and Q513 were gone, replaced all of these. I also noticed that R509 had a tiny hole, so i also replaced R508 and R509 (current sensing).

There is one inconsistency in the (manufacturer supplied) circuit diagram, R508 in situ was actually 1K2 (diagram 1K) and R509 was 10? (diagram .1?). The soldering of both looked quite original, not sure if the 10? was a design change to handle current spikes?

For a first test i put a  1K load on the +22VDC output and (bulb in line) powered up the unit. The SPMS did not fully power up as it apparently was stuck in the UC3843 startup. I then removed the 1K load and tried again and interestingly now it seemed to power up properly.  I then applied the 1K load while it was powered up and it still seems to run fine (and it supplied all external voltages), so i assumed it is just a startup problem.

I checked the UC3843 VCC supply and replaced C510 (support cap) as well as C512 (VR support cap for Q515). That did not change anything, unfortunately. R518/R519 for the initial (boot up) voltage support to UC3843 were also ok, and i could measure 165V before R519 but only 8V at VCC with the external load. Without an external load, the startup seems to work as the initial voltage support came up to 11V.

The primary voltage suppy measured after C517/C518 delivers a clean and ripple free 165V DC, so this seems to be fine.

When i lower R519 (limiting resistor between 165V supply and UC3843/11V Zener) from 47K to 22K, the startup seems to work as expected with the unit powering on and supplying secondary voltages as expected. I can also see that the UC3843 now takes up 1-2mA as required for the initial power up sequence.

Then there is the secondary voltage supply from the transformer winding 1-8 to take over after the startup has happened. With a 1K load in place, i just measure 8V after D532 (rectifier between transformers 22V AC and UC3843) and before R521 (limiting resistor before 11V Zener), without the load it's 16V should be enough to supply 11V after the Zener and run properly.

I then removed the transformer and measured the windings with my LCR set at 1 KHz. The values all seemed to be healthy (4-6: 2 mH/660 Ohm, 1-7: 10uH/0.2 Ohm, 2-7: 1.97mH/418 Ohm, 10-9 and 10-11: 40uH/0.8 Ohm, 15-14 and 15-16:  9,9uH/0.2 Ohm).


So... what i don't really understand is how the 1K load at any of the secondary outputs can destabilize the startup process, any idea? It seems to be both a problem with the startup voltage as well as with the transformer supply voltage, but D507 seems to be ok and C510 was replaced.

Anything else i'm missing?

Thanks!

[bktemp]

There is one inconsistency in the (manufacturer supplied) circuit diagram, R508 in situ was actually 1K2 (diagram 1K) and R509 was 10? (diagram .1?). The soldering of both looked quite original, not sure if the 10? was a design change to handle current spikes?

10 ohms for the current shunt is too high. 0.1 ohms is most likely the correct value (Ipeak=1V/Rsense=10A, that's ok for two paralleled BUK456 mosfets). Did you replace R509 with 10ohms? If yes, that's probably why the smps is unable to produce any output power.

[SCR]

Ok, thanks, i'll replace it against .1 and test again.

[tautech]

Ok, thanks, i'll replace it against .1 and test again.

Yes that would seem to be a plausible explanation, that R509 had been damaged and just as bktemp has suggested ten Ohms would be an unusually high value for current sense.

[T3sl4co1l]

Ew...

Just ew...

I'd say kill it with fire, but it's already dead, good riddance...

Replace it with something actually good, if possible.  I'd imagine pairs of 48V and 24V COTS supplies would do the job... might be able to find duals?

Tim

[SCR]

I've checked the photos again i took before starting the repair, the solder spots look indeed a tad different than the factory ones... looks like someone already tampered with the SPMS before i got it.

[SCR]

Ok, have replaced R508 against 1K and R509 against 0.1 ohm, as in the circuit diagram. Put a 820 ohm ballast load on the 22V output and did a first power up to check UC3843 VCC, which went up to 11 volt as expected, directly after this the MOSFETS overheated and the main fuse blew after a second.

Hmm... have to get another couple of BUK456 and check what happens on the UC3843 output pin to the MOSFET gates, no idea why it would do that.

[bktemp]

Did you check all secondary side diodes?
Maybe one is shorted.

The power supply is a odd design:
There is no regulation at all. After a short delay after powerup the current sense signale gets shorted, so the converter is running at maximum duty cycle, acting as a simple transformer. The current sensing is only used for soft start.
It looks like a  forward converter, but is missing the freeweeling diodes after the rectifier. Strange design.

[TerraHertz]

How about a fault (possibly thermal) with D523, or D525? And check C524.
Something is making the main FETs bear a lot of current faster than the current limit circuit can switch them back into off-state.

[bktemp]

Something is making the main FETs bear a lot of current faster than the current limit circuit can switch them back into off-state.

There is no current limit after a short initial delay, because Q515 disables the current sense signal.
So any excessive overcurrent will destroy the mosfets.

[SCR]

There is no current limit after a short initial delay, because Q515 disables the current sense signal.
So any excessive overcurrent will destroy the mosfets.

That's why i want to take a closer look at the output pin (PWM signal) to see what happens at startup. I did not properly check out the position of R516 so first thought it would be kind of a bias for the IB input (via Q515 ground), but actually the VR 5V supply directly drives the base of Q515, so fully disables the current sense feature. Strange!

I'll check D523 and D525 as well as C524, thanks!

The secondary diodes should be ok as the output voltage was correct with the 10 Ohm resistor and without load, but i'll check them anyway.

[bktemp]

You can apply an external +12-14V supply voltage after D532 to power the UC3843.
Then you can apply a low voltage to the input and look at the switching waveforms. If there is a shorted diode somewhere or the duty cycle is too high (it should be <50%) it will draw a significant amount of current.

[Ton]

Hi

This is a really odd ball of a converter, so here are my 2 cent ideas.

Since there is a suspicion that somebody else have worked on this converter, then the more I look at it the more I think it should run with a UC3845 controller instead.

Basically the voltage loop is disabled, pushing the controller towards 100% dutycycle , the current loop is disabled shortly after the startup has finished, pushing the controller further towards 100% duty-cycle.

So now there is no control loops to turn off the Mosfets => we get close to 100% dutycycle, if it ran with a UC3845, this would be 50% duty-cycle due to the build in t-flip flop in this model. and a 50% duty-cycle fits quite well with a forward converter style power transfer.

A forward style power transformer do not like duty-cycles going towards 100%, the core will probably saturate due to accumulated magnetic flux due to no time for core reset.

and a saturated transformer means hot mosfets, and then blown mosfets, just as OP has experienced.

I just cant get my head around how this dogs breakfast of a converter is supposed to work with a UC3843 with disabeled loops.

Or I might have misunderstood everything, it is after all around 15 years since I worked with a UC384x family controller. and it was in a flyback configuration.

Ton

[rob77]

Since there is a suspicion that somebody else have worked on this converter, then the more I look at it the more I think it should run with a UC3845 controller instead.

Basically the voltage loop is disabled, pushing the controller towards 100% dutycycle , the current loop is disabled shortly after the startup has finished, pushing the controller further towards 100% duty-cycle.

+1

i think it should be UC3845/3844 as well - those have limited duty cycle to 50%

btw.. that SMPS design is kind of stupid.. but i have an idea how to make it even more stupid...  replace the UC3845 with a 555   

[Ton]

Got my head a little more around it, so the converters off time is controlled by the value of CT, made up of C501,C502,C503,C504 and JU505 and the switching frequency is then adjusted with R516.

Based on the values in the schematics(C501 12nF), then they try to have a dead time of 3us-4us and a frequency of 100kHz => a duty-cycle of approximately 65%

the bad thing about this is that the Rt value is below 1Kohm, and the data sheet has no specifications for RT lower than 3K. so it might work ---- or not.

Why do they disable the current loop ??  cant see any good reason for that, just making the converter more fragile. Hummm

running this converter from a lab psu, and then checking some waveforms sound like a really good idea.

I had the same thought about the 555, but then you would loose the softstart 

Ton

[T3sl4co1l]

I had the same thought about the 555, but then you would loose the softstart 

Don't worry, just use a bigger supply filter cap -- comes up more slowly, and the low gate voltage limits switching current!

The MOSFET dissipation will be higher though, so you might replace it with a vacuum tube, which can dissipate the extra power.

Tim

[rob77]

I had the same thought about the 555, but then you would loose the softstart 

Don't worry, just use a bigger supply filter cap -- comes up more slowly, and the low gate voltage limits switching current!

The MOSFET dissipation will be higher though, so you might replace it with a vacuum tube, which can dissipate the extra power.

Tim

hmmm... actually.... that's not a bad idea at all   Vacuum tube SMPS , that would be a great technology for audiophools  warmer supply rails resulting in richer sound  but we would have to replace the 555 with an astable multivibrator made of tubes to be more "pure" 

[SCR]

With external power supplying the UC3843 the PWM output is at 54% duty cycle and 142 Khz (factory set, R516 allows to vary the frequency between 80-145 KHz).

When applying 0-20V to the primary input after BR327 the PWM duty cycle stays at 54%, regardless of the input voltage.
The scope shot shows the PWM input (yellow) and voltage output (green) at the MOSFETS (ignore the duty values of channel 1, they are an artefact of the switching spikes in the MOSFET input).

At 20V the input current is at 210 mA, which seems to be a tad high?

C524 and D525 are ok, still have to check D523. The only part damaged by the last full test was one of the MOSFETS (replaced both though) and the fuse, everything else seems to be still fine. I wonder if i should add a reverse diode to the UC3843 output, as well as flybacks to the MOSFETS as recommended above - if everything else is up and running again.

[bktemp]

210mA is way too high for no load condition, even at full mains voltage. At 20V that's already 4.2W. At 300V it will be 63W if the current stays constant (most likely it will rise with the voltage, so the power dissipation will be around 1kW!).
At 4.2W some component should get warm. Maybe that helps to find the faulty part.

The switching waveform looks a bit strange: The voltage typically swings to 2x supply voltage during off phase while discharging the inductor current. In your waveform the peak is lower and much shorter, it looks like there is almost no energy stored in the inductor. But it is hard to tell because of the odd topology used in this SMPS.
My guess would be one winding of the transformer is shorted either internally or there is a shorted diode.

[T3sl4co1l]

hmmm... actually.... that's not a bad idea at all   Vacuum tube SMPS , that would be a great technology for audiophools  warmer supply rails resulting in richer sound  but we would have to replace the 555 with an astable multivibrator made of tubes to be more "pure" 

It works pretty well, really...  http://seventransistorlabs.com/tmoranwms/Elec_Compound2.html (from 2007)

Tim

[SCR]

210mA is way too high for no load condition, even at full mains voltage. At 20V that's already 4.2W. At 300V it will be 63W if the current stays constant (most likely it will rise with the voltage, so the power dissipation will be around 1kW!).
At 4.2W some component should get warm. Maybe that helps to find the faulty part.

Had a look with a thermal cam, the two MOSFETS went to 70ºC after 5-10 seconds, but they were not mounted to the heatsink while testing. Still a bit too hot for a MOSFET driving just 200 mA.
D525 went up to 60ºC (but not sure why, winding 2-7 looks like some kind of push-pull /inverted phase circuit?)

The switching waveform looks a bit strange: The voltage typically swings to 2x supply voltage during off phase while discharging the inductor current. In your waveform the peak is lower and much shorter, it looks like there is almost no energy stored in the inductor. But it is hard to tell because of the odd topology used in this SMPS.
My guess would be one winding of the transformer is shorted either internally or there is a shorted diode.

I'll check the diodes again, as next step i would remove the transformer and only connect the primary windings to check if the input/output square wave looks similar. Good chance that the insulation of the primary winding got damaged, since it is a simple e-coil type transformer it should be possible to rewind it though (BTW, Heiser Transformatorenbau is a good address for such custom work in DE).

[bktemp]

Had a look with a thermal cam, the two MOSFETS went to 70ºC after 5-10 seconds, but they were not mounted to the heatsink while testing. Still a bit too hot for a MOSFET driving just 200 mA.

That's probably because there are much higher peak currents flowing due to a fault somewhere.

D525 went up to 60ºC (but not sure why, winding 2-7 looks like some kind of push-pull /inverted phase circuit?)

D525 should stay fairly cool. Look at the waveform on pin 2 of the transformer. My guess would be D525 is conducting in reverse direction, therefore shorting or at least loading the winding.

D525 and the winding at pins 8-2 are used to remove the magnetic flux from the transformer, generated during the on phase.
Here is a short description of how a forward converter works:
http://schmidt-walter-schaltnetzteile.de/smps_e/edw_hilfe_e.html#funprin

[SCR]

Thanks for the forward converter link!
I tested D525  before and it looked fairly good. Don't have a spare BYR 29-800 at hand, could get a STTH812DI as replacement though. Waveform/swing on pin 2 looks similar to the voltage output, except for the overshoot.

[TerraHertz]

There is no current limit after a short initial delay, because Q515 disables the current sense signal.
So any excessive overcurrent will destroy the mosfets.

Oh god. I'd ignored that bit since why would anyone do that?
Ha ha, actually strictly according to the schematic, there is NEVER any current feedback. What's that dot doing there?

[b_force]

Not having a feedback loop in a SMPS made for audio isn't so weird at all.
In audio the load varies a lot, which is a big problem for SMPS with a feedback from the secondary winding.
Second, in audio circuits it's also not very important to have an exact output voltage.
In fact, most guitar amplifiers use the voltage sagging for extra compression effect.

Only working on 99% duty cycle wouldn't be my choice, that leaves not a whole lot of room to play.

Another problem is that now the output voltage dependent on the input voltage.
Although, that's also similar compared to a linear power supply.

[SCR]

B_force, excellent guess, it is indeed a guitar amp.

I removed all conponents from the transformer output and checked again but the symptoms did not change. Good chance the transformer is gone, will do some more tests.

Fallback plan, are there small commercial power supplies available with symmetric outputs? (+-48V @2-4A, +-16V @1A)  There is not enough space to just install four small Meanwells.

[SCR]

P.s. What confuses me somewhat right now is that the output voltage (the setup with the 10 ohm shunt) is exactly like in the specs, but that should not be the case if some of the primary windings got shorted out since this would change the winding ratio, hmm...

[bktemp]

Did you try to replace D525?
To make the diode get hot, it needs to dissipate a lot of power. This can happen in 2 ways for a diode:
- due to a high forward current: Because of the low voltage drop you need a couple of amps to make the diode get hot)
- due to a higher than normal reverse current: Because the reverse voltage can be high, a rather small current can generate a significant amount of power.

Can a high forward current flow in this circuit? Probably not, because all of the current flowing thru D525 must have been stored in the transformer, and forward SMPS transformers have a high inductance to avoid any high magnetization currents. If the core gets satured because of a too low switching frequency or a too high duty cycle the current can get quite high, but the stored energy is still rather low. Maybe you can try to increase the switching frequency and look if the diode gets less warm.

My guess is still a faulty D525: I had diodes that tested fine at 5V but had a leakage current of several mA at 100V and even more at their rated 400V. From the switching waveform it looks like something is shorting the winding where D525 is connected to for the first part of the on-cycle. Maybe D525 has a too high turn-off time therefore generating a lot of heat (high recovery lossed because of some fault).

[bktemp]

There is no current limit after a short initial delay, because Q515 disables the current sense signal.
So any excessive overcurrent will destroy the mosfets.

Oh god. I'd ignored that bit since why would anyone do that?
Ha ha, actually strictly according to the schematic, there is NEVER any current feedback. What's that dot doing there?

I noticed that, but it makes no sense, therefore it is most likely just an error in the drawing. Also some parts of the cicruit are connected to GND, other ones to -165V but there is no connection between GND and -165V.

But it is still a bad design: They could have added an overload protection easily by putting a thyristor on the current sense resistor, shorting UC3843's power supply when the peak current gets too high.

[SCR]

Did you try to replace D525?

Not yet, i need to order some STTH812DI and MUR8100EG first, the ones i have at hand are not fast enough. Should have some results on tuesday.
I only did a low voltage test on D525, so a defect sounds quite likely given that the winding ratio seems to be still ok - thanks!

[SCR]

Tried both diodes but the symptoms did not change unfortunately. D525 and the MOSFETS still get hot (about 58C). I've fed 20V DC after the rectifier and the consumption while idle is around 190mA. Maybe the recovery time of STTH812DI and MUR8100EG is too short (currently at 50-75ns) as you mentioned before, i'll see if i can find something slower.

Could the overheating take place because the field/input at 4-6 is faulty?
E.g. the origin of the problem is at 4-6 and we just see some symptoms at 2-7?

[bktemp]

You could try to increase the frequency by adjusting R516: When the resistance gets lower, both the frequency increases and the duty cylce decreases, reducing the peak current. If it does not reduce the current, you could also try removing all secondary side diodes.

If one of the windings is shorted partially, the peak current will be much higher. If you have a LCR meter you can try to measure the inductance, but you probably have to desolder the transformer to get correct measurements. The inductance is probably in the range of 500uH to a couple of mH. It can be higher, but it should not be lower.

[SCR]

You could try to increase the frequency by adjusting R516: When the resistance gets lower, both the frequency increases and the duty cylce decreases, reducing the peak current. If it does not reduce the current, you could also try removing all secondary side diodes.

The power supply came set at 142 KHz, the maximum i could set up with R516 is 145 KHz, unfortunately.

If one of the windings is shorted partially, the peak current will be much higher. If you have a LCR meter you can try to measure the inductance, but you probably have to desolder the transformer to get correct measurements. The inductance is probably in the range of 500uH to a couple of mH. It can be higher, but it should not be lower.

Transformer was removed and tested after i checked the MOSFETS in the beginning, i only tested the transformer at 1KHz though:

4-6: 2 mH/660 Ohm
1-7: 10uH/0.2 Ohm
2-7: 1.97mH/418 Ohm
10-9 and 10-11: 40uH/0.8 Ohm
15-14 and 15-16:  9,9uH/0.2 Ohm

2-7 and 4-6 are slightly different but probably should be identical (1:1 windings), that might imply a small winding short in the beginning or end of 2-7. Still not a good explanation about the power uptake in idle when applying 20V. Will have a look at the MOSFET output and winding outputs with the scope tomorrow.

BTW, i've removed all secondary diodes before the D525 tests to make sure there is no influence from the secondary side.

[b_force]

Did you actually replaced the UC3843 already?

All looks pretty fine to me.

[bktemp]

The transformer looks fine. If you have removed all secondary side diodes, there is not much else left. You could disconnect R532 and D523 and C524 (should be ok while testing using low voltages).
If it still draws a high current, you could look at the current waveform (R509).

[SCR]

UC3843 was replaced in the beginning together with R521/Q513/Q522/R509/R508 and the PWM output looks fine. I'll focus on the waveforms around the transformer next and take a look at D523 and R532, as well as R509 if my current clamp manages measuring mA's.

Have replaced C524 against a WIMA MKS yesterday after D525 still got hot, but did no see any change afterwards.
I'm sure it's something very simple, i just haven't been able to discover it yet

[SCR]

A quick update, i put the scope on the MOSFET's gate and the drain (first image), as bktemp commented the wave does not look correct for a healthy forward transformer.

Next was the gate again and this time the transformers pin 2 which is the output of the second 330V winding for demagnetizing the core. I would have expected some kind of square wave but it looks like a DC voltage with spikes, maybe due to the phase cancellation with the main winding.

For this test, D524 and D525 have been exchanged against new components. As soon as i have found my differential probe i'll check transformer output 4-6 compared to 2-7 (the one in the second scope shot).

[bktemp]

No signal at pin 2 is strange, because your inductance measurement showed 1.97mH. So the winding can't be shorted or open.
It looks more like a shorted D525. That would explain all measurements (DC level at pin 2 and reduced inductance of the primary winding causing high current in the mosfets).
Can you verify D525 is not shorted?

[megajocke]

With that largeish 10 nF capacitor C524 connecting the demagnetization winding in that way to the drain node, we know that the windings must have 1:1 ratio.

This means the duty cycle can not be allowed to ever exceed 50 %. If it does, the transformer will saturate and a large current will flow through the FETs and D525. This matches the symptoms you observe and I think you also said the duty cycle was above 50 %.

Are you sure it's not supposed to be a UC3845 in there? Could it be that the schematic is wrong? Has someone been at it before you? (does soldering look non-factory made on the IC?)

[megajocke]

After a second look, it looks like they might be using an unusally large timing capacitor instead to provide the duty cycle limit as Ton earlier wrote...

Adding jumpers to JU505 would decrease the duty cycle (and frequency). But why would one design something this way? It's probably going to be unusually sensitive to temperature variations and differences between individual ICs, requiring adjustment at production or if the IC is ever replaced.

[rob77]

After a second look, it looks like they might be using an unusally large timing capacitor instead to provide the duty cycle limit as Ton earlier wrote...

Adding jumpers to JU505 would decrease the duty cycle (and frequency). But why would one design something this way? It's probably going to be unusually sensitive to temperature variations and differences between individual ICs, requiring adjustment at production or if the IC is ever replaced.

yes that whole design is literally stupid. 
a self-oscillating half-bridge driver driving 2 mosfets in push-pull would be a better solution for such a feedback-less SMPS ...

[SCR]

So, after doing some real work for a change back at trying to fix the SMPS

I played around a bit with the frequency/PWM setting of the UC3843 by adjusting R516 and discovered that i can get rid of the overshoots shown in my last posting by adjusting the frequency below 140 kHz.

At that frequency the overshoots vanished and the signal turned into a nice square wave (see scope image: yellow is the PWM output to gate, green is the MOSFET output, blue the secondary transformer output without load applied) the output current of the supply voltage also went down to 3 mA @20V (previously 190mA). Turning up the frequency creates the overshoot again, i am not sure why this happens though, looks like the transformer is resonating?



 

[bktemp]

It seems like the transformer was saturating, because the duty cycle was too high. Now it is close to 50%. Try going a bit lower to have more margin, because the slightest drift and it may go back to >50% and fail again.
That's a really bad design.
Maybe it could be a good idea to use a UC3845 instead of UC3843, because it is limited to 50% duty cycle (it blanks every other pulse, therefore the oscillator frequency needs to be doubled).

[SCR]

I think that the duty cycle was always at 50% or below, but the frequency seemed to matter. My previous tests at 43% duty and 147KHz yielded 190 mA @20V input (which would blow out the fuse when fed line voltage), whereas 49% at >140KHz results in 3 mA based on my testing today.

EDIT: The trimmer also changes the duty cycle above 140KHz, you were right. At around 50% and upwards the saturation starts and you can see the overshots appearing, also the primary input current rises up sharply.

[SCR]

The SMPS is up and running again, thanks for the support!

[MagicSmoker]

The SMPS is up and running again, thanks for the support!

Yeah, but for how long? This design is fatally flawed. I get the rationale for skipping voltage regulation in a guitar amp, but every other design choice made here either needlessly complicated the designer's life (e.g. - like choosing a controller IC with a 100% duty cycle limit in a transformer-isolated power supply and relying on grossly distorted timing component values to limit max duty cycle) or was done to intentionally defeat an intrinsic feature or function of the controller IC (e.g. - shorting out the current sense signal... to a current-mode controller IC).

I'm very curious about the model/brand of the guitar amp, and I suspect other participants in this thread are as well.

[SCR]

I've set the duty cycle to ~30-35% to be on the safe side which seems to work well even when applying some good load. These amps are known to be quite reliable, i think they have been sold since the early nineties. Good chance that the SMPS design has been changed in the meantime though.

http://www.gallien-krueger.com/mb-150s112-specs/

P.s. just noticed that the website shows a photo of the newer power supply, which features a toroidal transformer