[Document Version: 1.02]
[Last Updated: Jan_11_1994]
Author : Keith Lofstrom
E-mail : firstname.lastname@example.org
Date : Jan 11, 1994
WARNING! DANGER! WARNING! DANGER! WARNING! DANGER!
Power supplies contain lethal voltages. That means they can kill YOU. Yes,
YOU! Unless YOU are especially cautious, YOU won't wake up tomorrow. Pretend
you are playing with a nuclear bomb. A bomb can't kill you any deader.
If you don't understand electronics, don't bother with power supplies. This
is analog circuitry, not cookbook logic stuff. You need to have a feel for
currents and voltages and waveforms, or you won't be able to figure out what's
happening. Find some circuit nerd (like me) who needs some software written,
and swap services. Don't risk your life unless you know what you are doing.
Power supply repair is a challenge, like mountain climbing. Probably about as
dangerous. Allow adequate time (3-4 hours, plus trips to the parts store),
don't attempt it while tired or distracted, and have the right tools available.
Make sure somebody is around to haul you to the hospital if you zap yourself.
If you DO hurt yourself, in spite of my warnings, I am NOT responsible. I
never posted this, my signature is forged, the important parts got deleted by
your newsfeed, etc. If you can't take responsibility for your own life, don't
repair power supplies.
WARNING! DANGER! WARNING! DANGER! WARNING! DANGER!
I've fixed about 40 of them. The broken ones were all broken by
bad design or sloppy manufacturing - inadequate component choice, usually. If
you want to fix one properly, you will need to be able to find heftier
components that still meet the other design goals - made somewhat easier by
the advances in components. A proper repair involves a little design.
You will need access to a semiconductor curve tracer, or else be ready to
build a lot of ad hoc kludges for pulse characterization of components. Most
of the power components that fail will have no easily available direct
replacements. Looking at some components with DC sources can cook them.
You will also need an oscilloscope ( 5MHz bandwidth will do ), a voltmeter,
and access to a huge pile of power transistor and diode databooks. National
Semiconductor, International Rectifier, Texas Instruments, Motorola books will
probably do; you may need SGS Thompson, Siemens, and Signetics for some
Without a schematic, be prepared to spend some time tracing out circuits. If
you live in East Armpit, Nebraska, you probably won't be able to find the
components you need, so give up now.
An isolation transformer and a Variac are handy. Also a pile of power
resistors from which you can build a test load. A bench supply to externally
power the secondaries is useful, though I use the curve tracer for this.
[Notes on the diagnosis and repair
of small SwitchMode Power Supplies] (SMPS)
- written by Samuel M. Goldwasser. This is an excellent guide to just about
anything you may want to know about switching power supplies.
A Fil's Must-Read!
- Put at least two switches between you and line. Unplugging counts as
one switch. You are human, and you will forget one switch from time
to time. Forgetting shouldn't have to be lethal. Take off watches
and metal jewelry. Hell, get someone else to do it. Life is too
short anyway. If you are a professor, find a gullible grad student :-)
- Check the fuse. Keep plenty on hand - you may end up blowing a dozen
or so. About a third of the time fuses blow for no reason at all.
- Is everything hooked up and switched properly? I've seen a PhD spend
4 hours debugging a simple little box that "should work but didn't".
When I innocently asked "what does this little switch do?" he thought
for about 10 seconds and turned the prettiest shade of red...
- Good designs rarely fail. If it is broken, it is because of an idiot
engineer, most likely. Don't assume anything is designed right; if a
circuit you've traced out can't possibly work, check it again. If it
simply LOOKS stupid, you have discovered a fundamental truth about
some power supply designers. There are lots of good power supply
designers out there - I've rarely had a chance to explore their work,
because I've never had to fix their designs.
- A toasted resistor means something else failed first. Figure out what
could toast the resistor, or you will toast another one.
- Buy extra replacement parts. You will blow a few things up during
- Before firing the supply up, figure out the primary circuit. Typical
switched 115/230V supplies have a full wave bridge and two filter
caps, resulting in about 340 volts DC on the primary. Make sure there
is a discharge path through a power resistor - if not, temporarily
solder one on while you work on the supply. A 20K ohm, 10W resistor
will discharge a 1000 uF primary in about 1 minute. You can switch
in a smaller value resistor for faster discharge, but you will
probably forget to turn it off, and end up blowing it up when you
turn the supply back on. In any case, get rid of the primary side
voltage somehow before you mess with things.
- When building test loads, load all the supplies. Some switchers get
unhappy if there isn't at least a trickle of current on each supply.
A 20% load is often adequate.
- When you think you have the supply fixed, run it up and down with the
Variac; run it up to at least 1.2X normal line, if you can. Note:
some switchers get very unhappy with 0.5X line - see note on stupid
designers above. If you end up breaking it this way, at least you
know what will happen during brownouts. If the result is overvoltage
on the outputs (I've seen this once) then throw the supply away, and
kludge on a different design if possible. Overvoltage eats expensive
- Anything can be fixed. Sometimes it's too much work. Be willing to
The things I have seen fail, in order, are:
- Electrolytics. Switchers put a lot of AC current into electrolytics, many
of which aren't designed for this treatment. The capacitors cook, and
either short out or unweld themselves open. The nastiest failure is an
"almost open" that opens and closes itself at line rates (I've encountered
a couple of those - murder to diagnose).
One capacitor failed on the input side of a pi filter on a secondary:
--|>|------+----- Inductor ------+--------- 5V out
----- This one -----
----- <--- -----
This was on the 5 volt supply of a 5V/12V switcher, the regulator worked off
the filtered 5V. When the indicated capacitor opened, the inductor
swallowed most of the switching voltage from the transformer and diodes,
resulting in the regulator kicking in much harder to try to make 5V on the
other side. This put 35 volts on the allegedly 12V supply - as a result, all
the 16V tantalums on the 12V supply got blown out, as well as much of the
stuff hooked up to the 12V supply.
The reason for the pi filter in the first place was too much ripple on the
supply - the reason for too much ripple was the high ESR of the capacitors
used. I replaced with better capacitors.
Replacement: Don't just go for the same voltage and capacitance. You need
low ESR, and if the first one failed, the second one will too. One sorta
brain-damaged way to get low ESR is to use more capacitors - and possibly
more capacitance - wired in parallel. You can get some idea of ESR with the
curve tracer - with a small voltage ( approx 200mV ) AC drive, a low ESR cap
looks like a circle, while a high ESR cap looks like an ellipse from the
resistive component. Look for high temperature ratings, too - at least 105C
If there are any power resistors sitting right next to an electrolytic, move
them enough to get airflow between. Electrolytics are too damned temperature
sensitive - most are rated to only 85C.
- The main switch transistor(s). Usually a TO3 power transistor. These
usually fail because something shorted things out (there should be a
current sense shutdown, but this involves buying another 5 cent resistor;
too expensive :-( ). These also often fail from inadequate voltage rating.
These are usually the easiest component to check, assuming a curve tracer is
available, even if they are less likely to fail than one of the 10 or so
electrolytics. I pull the transistor and check it before trying anything
else. If the transistor works, write down some of the curve tracer
measurements. You may accidentally fry it during repair.
You will want a transistor that can stand off at least 1.5X the peak primary
DC voltage, preferably 2X. You will need to find what the current rating
of the transistor you pulled out is, and meet that with the replacement.
When you are looking for a replacement, find a new transistor with:
Advances in components make finding an adequate replacement possible. Curve
trace the replacement.
- Same type - usually NPN, non-darlington
- Same case
- Better voltage rating
- Better current rating
- Similar Beta/Hfe/Current Gain
- Same or Higher Ft
- Same or lower capacitance
- Secondary diodes: I've lost a few of these. The result of a diode failure
is usually an imbalance in secondary voltages.
- Same type - usually Shottkey diode
- Similar case (you may have some lattitude here)
- Better voltage rating
- Better current rating (make sure with curve tracer)
- Lower capacitance
- Regulator ICs - these rarely fail, but a manufacturer can get a bad batch.
The result is a whole bunch of the same supply type failing. There are so
many different ways a regulator can fail that you will just have to figure
it out; though a supply that goes tick ... tick ... tick may have a broken
regulator chip, if you can't find anything else wrong. The ticking noise
means the regulation loop is broken somewhere.
You may have a heck of a time finding a replacement. An exact replacement
is required. Socket the replacement. Use a good socket that grips the chip
tightly - thermal excursions can walk an IC out of its socket.
- Check the fuse.
- Dust the damn thing, with kleenex and Qtips. Of course you've never changed
filters on the fan. Dust kills power supplies. After this, you'll change
- Look at a working version of the supply, if any are available. Get a
schematic if you can. Apple and Sun are lousy for schematics, HP is good.
Clones are impossible.
- Look for all the usual visual stuff, open traces, shorts, burned components.
- Disconnect the supply, writing down somewhere where all the connectors go and
which direction ( if I need to be telling you this stuff, you aren't
qualified). Don't use the regular circuit as a dummy load, though you may
want to find out how much current it draws, to help you build test loads.
Use a bench supply for testing the regular circuit.
- Power up the secondaries - one at a time, with an external voltage source -
first. Look for shorts. Drive capacitors with curve tracer AC. Look for
ESR or opens.
- Load the outputs at about 20% of full load. Power up the primary with curve
tracer in AC mode, slowly, to about 40VAC. You should be able to watch the
primary capacitors charge up. This finds primary shorts and capacitor opens.
- If you bring the supply up to full input and nothing happens, check for
primary voltage - usually 340 volts. If the supply is 115 volt only, and
there is only one capacitor, there is probably only 170 volts across it.
- If there is voltage, AND the primary is separately isolated with an isolation
transformer, locate the primary common, usually the lowest voltage on primary
side (You traced the circuit, remember?). Hook the scope signal input to
this point (at 50V / div, line sweep and sync ), there will probably be a
significant AC signal here. Try connecting this point to oscilloscope
ground with a 100 Kohm resistor - if the signal doesn't diminish, you aren't
properly isolated. If it does diminish, power down, disconnect, and
discharge, ground the scope here, and power back up.
- Check for oscillations around the regulator chip. If none, check for voltage.
Switcher regulators are sometimes powered with their own separate little
supply, or off a big power resistor from primary voltage. Sometimes they are
powered with a little kickstart circuit from the primary, then a separate
winding off the transformer. These are good for the tick...tick...tick
type of failures. You may have to power the regulator separately, with a
bench supply, to get things started.
- From here on out, it's measurement, and debugging, and tinkering, and such.
Let us know what you find wrong. Perhaps we can embarrass some manufacturers
into doing a better job on their supplies.
Again - and I repeat - don't repair power supplies unless you are a pretty good
analog circuit tinkerer, can act safely around high voltage, and have the right
tools. Take the thing to a TV repair shop, instead. It would spoil my day to
find out somebody with more ambition than sense took the above notes and hurt
themselves trying to do something they weren't ready for yet.