9-4-2006

One thing I like about the USMPS ZVS is that it has
appeared much more stable than the original, non-ZVS
version of the USMPS.  I feel that the ZVS supercedes
the original, but I am planning to keep all past information
of my switching power supply design work displayed
at this group.  Eventually, I may have to make some
changes to some picture files to regain some space,
though.

I added the new oscillator circuit to the test ZVS USMPS
unit on my bench.  Now it behaves quite stably even
though it is all point-to-point wired without a circuit
board.

I no longer use breadboards, though, just the
method known as dead-bug.  The 74C14 controller
and IR211x MOSFET driver IC chips lay upside down
on some electrical tape on a flat-surfaced heatsink.
The output pins of the 74C14 chip line up exactly with
the input pins of the IR2110, IR2113, or even the
IR2112.  The lower current outputs of the IR2112 IC
should not matter much since the MOSFET gates
are enhanced more gradually than in hard-switched
power supply circuits.

The heatsink has a hole drilled in it near a corner.
The controller/driver assembly are held in place
by soldering the IR211x to the MOSFETs.  The MOSFETs
are then held down on top of some thermally conductive
insulators by means of a piece of wood or metal which
spans between their plastic TO-220 packages.  The
screw goes through a hole drilled in the spanner
between the two MOSFETs. 

The transformer generally does not need to be
wired to limit leakage inductance.  In fact, for most
windings leakage inductance not only does
nor hurt, it actually boosts efficiency.  For higher
power units, some of this efficiency may need to be
exchanged for lower EMI.  Even though the ZVS
USMPS has reduced EMI in relation to more standard
hard-switched designs, like them as the power level is
increased, the EMI tends to increase as well.

Probably the one reason I ended up needing the ZVS
approach so much is that my surplus-purchased ferrite
toroids probably do not have very high permeability.
I always had problems with turn-off spikes even when
they were wound evenly around the core for lower
leakage inductance.  The lower the coupling coefficient,
the less the transformer drive can damp the spikes.

I still plan to upload a more updated diagram later.
I may wait until the cold weather ends this year's
work in the tinkering room, though.

8-25-2006
Well, since that last really great change to the oscillator,
I think the design of the ZVS half bridge circuit is very
good indeed.

One thing I am not sure that I mentioned is that with the
extra inductance permissible in series with the secondary
side if the transformer, there is really no need for active
current limiting in the circuit.  It limits the power without
stress on the primary side.

This winter season I hope to upload new diagrams of the
ZVS USMPS circuit.  Really, just about the only disadvantage
of the approach is that the frequency does not remain constant.
But in the test circuits I have built, I have found that it
does not vary too much between light load and full load.


8-25-2006
I have redesigned the oscillator circuit.  By adding an extra
timing resistor and capacitor, the ZVS timing can be much more
precise  Also, instead of the circuit starting up at a very
high frequency depending on the value of the formerly single
timing resistor, now the default frequency setting is low
instead of very high.  If the ZVS drops out now, the power
MOSFETs can run cooler at a lower frequency.

Simply pass the output of the oscillator section of the 74C14
through a resistor/capacitor low pass filter before it goes
to the other components of the timing circuit.  Now the original
timing resistor can be decreased is value greatly because there
is no longer a need to use it to depress the frequency.  The 
feedback winding on the transformer can be reduced in turn
number too.

3-28-2006
Some of the latest changes to the circuit have made it test
well so far.  I have changed the secondary from single to
dual secondaries with two single rectifiers instead of a
bridge.  I have wound each secondary in a small window so
that leakage inductance is increased.  Then
a .22uF capacitor connects between the anodes of the
rectifiers.

I have increased the gate turn-on resistor to each IRF840 to
1k ohm.  Efficiency is high and up to a few tens of watts, the
MOSFETs can be run without a heatsink.

10-27-2005
So right now it seems that it is OK to leave out the series resonant
capacitor on the secondary side.  Then, make the parallel resonant
capacitor on the primary side small.  A possible value may range
from no external capacitor to something like .01uf.

Concerning the 9-23-2005 note, I think it is possible to get more
stable circuit operation with a more aggressive feedback circuit as
in the simulated version, though.

9-30-2005
Well, here are some thoughts I have been having concerning ZVS
and power supply design in general.  If the primary side is stiffened
against transients with a capacitor on a half bridge totem pole, and
there is not much series inductance on the secondary side or the
leakage inductance of the transformer is low, meaning there is a
coupling coefficient fairly close to one, then (finally) the transients
will transfer from the primary to the secondary side.  I can see
merit in trying two techniques.  One is to use as low a frequency
as component space and weight will permit to allow a slower circuit
with good efficiency.  The second is to strive for the transformer
not to kick energy back to the primary circuit.  I have observed that
EMI develops when the primary voltage flies by inductive kick from
one power supply rail to the other.  Well, that phenomenon is the
essence of ZVS!

9-23-2005
Try increasing the value of the turn-off resistor at the junction of
the turn-off feedback diodes to even out and smooth the operation
of the feedback.  I found that 330k seems like a good value.  Then
the emitter turn-off resistor can be increased to 10k as well as the
resistor in series with the zener diode on the diode side of the
optocoupler.  Increasing the value of at least the emitter resistor
conserves the already small amount power the control side of the
circuit draws.

If the result of these changes is that the regulation will not hold
down the output voltage under a light a load, lower the value of the
default load resistor a little.  I found that on the simulator that
making its value around 620 ohms when the output voltage is set
up to be 15.5v does the trick there.


6-20-2005

I have been experimenting with a physical ZVS-USMPS circuit over last few months.  I am trying to get as close to the ideal of no EMI without much or any need to include extra filtering or shielding measures.  It involves keeping the frequency of all harmonics and transients as low as possible.  I am tending to observe a direct correspondence between power level and quantity of EMI generated, which is as one would expect.

Lately, I have been designing such a circuit for a battery charger on an electric motor assisted bicycle.   Its output current will be about 7 amps, and I am planning to use a switching frequency below 20khz to allow for lower switching transients and harmonics while maintaining good efficiency.   I aim for light weight and compactness.

The main trick to use for keeping the frequencies of transients as low as possible is to construct the transformer with as many turns as possible for the output power needed.   On the battery charger project, the primary is comprised of about 150 turns of 24 gauge wire applied in two separate windings connected in series.  This allows the resonant capacitance to be made smaller while still maintaining a low switching frequency.  It also keeps the circulating current no larger than necessary.

The ZVS circuit can be considered somewhat the opposite of spread spectrum.  It tends to concentrate any EMI generation in lower frequency bands.  So by making the transformer impedance as high as possible, the resonant current, and therefore the power escaping any lossy areas, are lower.



4-30-2005

Since I like the transformer winding phase feedback method much
more than the capacitive feedback method, I have deleted the
LTspice circuit of the capacitive method but have not ruled out
experimenting further with it.

4-29-2005

I still like the method that uses the transformer winding for phase
feedback much more than the capacitive method.  To the former
circuit, I have added a 430k ohm resistor, R8, to limit the maximum
on-time.  That resistor is needed to set the minimum frequency under full load and to enable the ZVS process under that condition.

4-23-2005

I am considering removing the circuit described directly below
in the 4-19 note because I overlooked an important function of the
variable voltage supplied to the timing circuit capacitor charge.
I think it is required to adjust the duty cycle to maintain the output
state at either high, low, or transition level.  I plan to do
more LTspice experimentation on it and then edit this
note and maybe remove the 4-19 note and the capacitive phase
feedback circuit from the files section. 

4-23-2005

I am experimenting with the two different phase feedback
approaches.  After some contemplation, I suspect, for now at least,
that the transformer way may be superior.  It may not be
detrimental for the resonance of the output filter to delay the turn
on of each MOSFET since the critical requirement is that the
voltage peak be great enough to take the output of the MOSFET
totem pole to each power supply rail.


4-19-2005

In the latest LTspice circuit construct, I have altered the
phase feedback from transformer winding to capacitive.
The capacitor couples the changing voltage at the actual MOSFET
totem pole output to the oscillator.  The resonance associated
with the output inductor no longer interferes with the zero voltage
switching this way.  Since there is no longer any choice of
switching winding polarity to achieve proper phasing,
I've switched the input signals to the IR2110.

4-4-2005

I am uploading the improved LTspice circuit of the ZVS  USMPS
with changes mentioned in the 3-29 note.  I have also increased
the filter capacitor from 1000uF to 2000uF since that restores loop
stability because of the increased maximum ripple due to the higher
over current protection level I have added.  Before I felt it
necessary to limit the power to around 500w, but with the improved
operation, I have been able to raise it to 800w.  I have also
reduced the MOSFET turn-on resistor since the ZVS operation is
now more even, so the quicker turn-on has less risk of dropping the
operation out of ZVS mode, the occurrence of which could cause
the MOSFETs to overheat after a short time.

3-29-2005

I seemed to have found a major improvement in the current
limiting circuit which appears to improve four
simulation aspects of the LTspice construct of the ZVS
USMPS a lot.  By adding a separate current limiting transistor
parallel to the regulator one, the output current gradually ramps
up as in soft start, the feedback response is much smoother,
the transformer drive waveform remains much more balanced
in respect to duty cycle and zero voltage switching,
and now virtual full duty cycle protection is apparent.
After more testing, I hope to post a revised diagram.

3-27-2005

Something to watch out for is power dissipation in the secondary
resonant circuit from losses in the resonant capacitor there as well
as in the inductor and somewhat in the secondary winding.
The greatest losses could come from the skin effect, which is the
tendency of electrons to flow along the surface of a wire at high frequencies.  The normal remedial approach is to compose
windings of multiple strands of thinner gauge wire where needed.
Other times just using a single thicker strand than otherwise
needed may suffice.

3-25-2005

I have uploaded the version of the ZVS USMPS with the largest
resonant capacitance parallel to the primary of the transformer
that I found will work acceptably according to the simulator.
Also, I have  improved the feedback a little.  I have added
half-cycle current limiting as well.  For now, I will leave rev1
available, but may delete it later.

3-16-2005

The feedback circuit cannot cope with a output filter
capacitor smaller than 1000uF due to the slowness of the
optocoupler--my guess.

The maximum size of the resonant capacitor seems to be around
.022uF.  An actual power supply will use one that
 ranges between that value and zero (no parallel capacitor.)
It may be best to use up to .01uF across the lower MOSFET
and the same value across the upper one (drains to sources).
Then they will act as decouplers, I think.  But that way may
also lower the impedance a little too much causing ringing, which
I am not sure will not happen anyway.  Hopefully, the transients
are kept so slow that ringing will not be induced in any
parasitic inductances.


3-14-2005

So far I have gotten good results after reducing the output filter capacitor to 1000uF.  Now I wish to try a lower value too.

3-13-2005

I seem to have great news in that I have been able to get good
performance now with a much larger resonant capacitance
shunting the output MOSFETs.  Something in the ballpark of
.02uF works because I decided to try to get a tank circuit on
the transformer secondary side to add some power back to
the primary side during flyback.  I almost gave up on the idea
because I had been wrongly trying to rely solely on the
inductance of the secondary in the LC tank.  When I placed
the capacitor across the secondary and the output inductor
in series, it seemed to solve the problem.  Now my confidence
that the circuit concept functional is high.

Now the no load frequency is only around 120khz, a great
improvement as far as efficiency is concerned.  Now EMI
can be low too.  It is probably best to divide the resonant
capacitance in half and place half across the drain to source
of each MOSFET.  So, that would mean .01uF across the
upper and .01uF across the lower one.  The secondary
resonant storage capacitor is approximated according to
the turns ratio of the transformer.  For 10 to 1, it needs to
be at least .22uf, but I think right now that making it larger
gives more boost for the primary flyback.  Too large causes too
much circuit losses from high circulating current and can actually
cause the destructive dropping out of the ZVS mode.  I have
simulated it at 1uF well so far.

I would like to do more tests and try things like lowering the
output filter capacitor before replacing the Rev1 circuit with
the new one.  The output loop stability still seems good with
the 10000uF capacitor, but I would try to lower it and observe
the results.

3-11-2005

What I've simulated is the Rev1 circuit with a few changes
to greatly decrease the response time of the ZVS feedback.
I have totally removed the resonant capacitor that was
bypassing the MOSFET totem pole middle point and am
relying solely on the
output capacitance of the MOSFETs.  I have also increased
the transformer coupling capacitors to 1uf each.  The no load
frequency increases to around 700khz.  The frequency at full
power of around 500w is close to 35khz.  The stability of the
DC feedback shows remarkable improvement.  I wonder
how the EMI will be if I put together this circuit.

I have also been able to add half-cycle current limiting.
It was impractical before since it interfered with the ZVS
and would overheat the MOSFETs to their destruction.
If the parallel capacitor to the MOSFETs is relatively large,
half-cycle current limiting cannot be used.  It is safe to use
with no such external capacitor.

Another addition to the Rev1 circuit is differential proportional
integrative control in the feedback loop (edit: but I changed
it to simple phase lead on 3-14-2005). 


3-11-2005

I discovered a problem with the 3-8-2005 noted methodology.
The voltages on the transformer coupling capacitor(s) can
become unrealistically high during resonance.  Those
capacitors will usually only handle 250v, but the simulation
was showing them charging at  over a kilovolt.  So the boost
was really operating, but was unrealistic at those levels.
Maybe the best thing to try will be to greatly lower the value of the resonant capacitor connecting  the middle of the MOSFET
totem pole to ground.  Then, half-cycle current limiting may also work.

3-8-2005

I think that the ZVS USMPS may not require as aggressive of regulation as hard switched topologies.  The inductor in series with the secondary winding of the transformer appears to react with the coupling capacitor in series with the primary to boost the voltage by means of phase shift under a load, partially offsetting the usual voltage drop.  Admittedly, I  have to try to get a lot of physical circuit experimentation done.   

2-07-2005

Theory of Operation of the ZVS USMPS

When power is applied to the circuit, The control IC, the
74C14 receives 10v from the 15v supply through a dropping
resistor.  A 10v zener diode regulates and a .1uF monolithic
ceramic capacitor filters and bypasses noise.

The timing capacitor and charge/discharge resistor initially provide
a starting period of under 5 microseconds.  However, one of  the
MOSFETs is soon turned on, causing the secondary winding, L3,
in series with the resistor to develop a polarity which increases
the current passing through the capacitor very quickly turning that
MOSFET back off.  Now is when the needed action of the phase
control winding begins its very important work.  When that
MOSFET turns off the voltage on the transformer phase winding increasingly opposes the voltage on the series resistor as the
primary voltage flies toward the opposite rail.  Once that waveform
quits traveling--either because it reaches the other supply rail or
because the energy stored in the transformer core can no longer
charge the resonant capacitor parallel to the primary winding, the
charge process on the timing capacitor is no longer being inhibited,
and the period quickly times out, causing the gate to switch states,
turning on the MOSFET on the voltage rail to where the flying
waveform had been going.

Now that one of the MOSFETs is on, the appropriate gate coupling
capacitor is discharged either by the on-time limiting resistor or by
the feedback circuit.  When the MOSFET turns off, the primary
voltage flies toward the other power supply rail all the while inhibiting
the voltage change on the oscillator capacitor.  When it peaks out, it
quits its inhibition of current entering or leaving the capacitor, the
capacitor quickly times out, the gate switches state, and the
appropriate MOSFET on the voltage rail where the waveform either
reached or tried to reach turns on.  The process continues over and
over.

The output of the IR2110/IR2113 is buffered by the relatively large
series resistor, which directly turns each MOSFET slowly through
the diode, but turns it off much faster through the PNP transistor.
Usually, the transformer primary winding voltage is already at the
same power rail supplying the MOSFET turning on.  The slow
turn-on time is extra assurance that if the waveform had not flown
far enough, the inrush current into the resonant capacitor is
gradually increased in order to reduce spiking.

Another important EMI reducing component is the capacitor across
the AC inputs of the output bridge rectifier.  It absorbs transients and
ringing produced by the output voltage snapping away from the
bridge rectifier diodes.

I will plan to offer the basic circuit for cycle-by-cycle overload
protection later.  The ZVS USMPS cannot use only single rail limiting
as the normal USMPS can, or the ZVS process will be severely
compromised.  What it means is that the overload protection needs
to rather symmetrically work with both MOSFETs.

I am considering changing the oscillator charge/discharge resistor
from 220k to 120k to reduce the length of time that the waveform
lingers at each power supply rail before the appropriate MOSFET
turns on.  The smaller value would cause each half cycle of the
oscillator to time out faster once the primary waveform reaches its
peak.

The transformer coupling capacitors play a crucial role in helping the
resonant action occur.  By means of their phase shift, they give the
primary waveform an additional boost in its attempt to reach the
power supply rails during flyback.

Please check back for new editions of this explanation.   



