Forums

Removing voltage-multiplier-related flyback SMPS switching glitches

Started by Unknown January 17, 2017
A problem encountered with medium-to-fast current-mode flyback switchers,
when combining high-ratio multi-winding flyback inductors with multipliers:

When Q1 switches on, a large current surge flows as Cpump charges from Cmult1
via D2. At Q1, the surge magnitude is multiplied by the transformer ratio.
On simulation, U1 senses this via Rs and cuts Q1 off instantly, crippling the
switcher -- it balks/stalls, and refuses to produce any output.

                              Cpump
               +12V     .------||-------. DA2FJ8100L (x 3)
               -+-   1  |  D1      D2   |   D3
                |    .--+-->|---+-->|---+--->|----+----> +HV
               o )||( PA0367A   |                 |
            L1   )||(          --- Cmult1        --- Cout
         n=1:12  )||( o        ---               ---
                |    |          |                 |
  U1            |   ===        ===               ===
LTC3803-3   ||--'
------.     ||<-. Q1
  Gate|->---'|--+
      |         | | i1
      |         | V
 Sense|<-[Rslp]-+
      |         |
------'        [Rs]
                |
               ===


Several blanking schemes to clamp, filter, or otherwise ignore the
current spike were considered and rejected before settling on a
solution.

Recognizing that i(Cmulti) is -i(Q1)/n, a cancellation scheme
removes the forward-mode i(Cpump) artifact from U1's worldview ...

                            Cpump
             +12V     .------||-------. DA2FJ8100L (x 3)
             -+-   1  |  D1      D2   |   D3
              |    .--+-->|---+-->|---+--->|----+----> +HV
             o )||( PA0367A   |                 |
          L1   )||(          --- Cmult1        --- Cout
       n=1:12  )||( o        ---               ---
              |    |          |                 |
  U1          |   ===         |                ===
LTC3803-3 ||--'               |
------.   ||<-. Q1            |
  Gate|->-'|--+               | ^
      |       | | i1          | | i2 = i1/n
      |       | V             |
      |       +-----[R1]------+ R1 ~= n*Rs
      |       |               |
      |      [Rs]             |
      |       |               |
      |      ===              |
      |                       |
 Sense|<--------[Rslp]--------'
      |
------'


After modification, U1.sense only sees L1's charging current (the
current storing energy for flyback operation); the forward-mode
Cpump-charging spike is eliminated.

Probing with the oscilloscope confirms that cancellation works
essentially ideally in practice, a pleasant result.

Cheers,
James Arthur
On Tue, 17 Jan 2017 18:00:43 -0800 (PST), dagmargoodboat@yahoo.com
wrote:

>A problem encountered with medium-to-fast current-mode flyback switchers, >when combining high-ratio multi-winding flyback inductors with multipliers: > >When Q1 switches on, a large current surge flows as Cpump charges from Cmult1 >via D2. At Q1, the surge magnitude is multiplied by the transformer ratio. >On simulation, U1 senses this via Rs and cuts Q1 off instantly, crippling the >switcher -- it balks/stalls, and refuses to produce any output. > > Cpump > +12V .------||-------. DA2FJ8100L (x 3) > -+- 1 | D1 D2 | D3 > | .--+-->|---+-->|---+--->|----+----> +HV > o )||( PA0367A | | > L1 )||( --- Cmult1 --- Cout > n=1:12 )||( o --- --- > | | | | > U1 | === === === >LTC3803-3 ||--' >------. ||<-. Q1 > Gate|->---'|--+ > | | | i1 > | | V > Sense|<-[Rslp]-+ > | | >------' [Rs] > | > === > > >Several blanking schemes to clamp, filter, or otherwise ignore the >current spike were considered and rejected before settling on a >solution. > >Recognizing that i(Cmulti) is -i(Q1)/n, a cancellation scheme >removes the forward-mode i(Cpump) artifact from U1's worldview ... > > Cpump > +12V .------||-------. DA2FJ8100L (x 3) > -+- 1 | D1 D2 | D3 > | .--+-->|---+-->|---+--->|----+----> +HV > o )||( PA0367A | | > L1 )||( --- Cmult1 --- Cout > n=1:12 )||( o --- --- > | | | | > U1 | === | === >LTC3803-3 ||--' | >------. ||<-. Q1 | > Gate|->-'|--+ | ^ > | | | i1 | | i2 = i1/n > | | V | > | +-----[R1]------+ R1 ~= n*Rs > | | | > | [Rs] | > | | | > | === | > | | > Sense|<--------[Rslp]--------' > | >------' > > >After modification, U1.sense only sees L1's charging current (the >current storing energy for flyback operation); the forward-mode >Cpump-charging spike is eliminated. > >Probing with the oscilloscope confirms that cancellation works >essentially ideally in practice, a pleasant result. > >Cheers, >James Arthur
You could also use smaller caps in the multiplier or lowpass filter the current sense signal a bit or use a less aggressive transformer and more multiplier stages. This worked without tricks, somehow: https://dl.dropboxusercontent.com/u/53724080/Circuits/Power/HV/28S840A.pdf 1400 volts out at the top, with a 1:1 autotransformer. This may current-limit for a few cycles at startup, but it works its way up. -- John Larkin Highland Technology, Inc picosecond timing precision measurement jlarkin att highlandtechnology dott com http://www.highlandtechnology.com
On Tuesday, January 17, 2017 at 9:26:46 PM UTC-5, John Larkin wrote:
> On Tue, 17 Jan 2017 18:00:43 -0800 (PST), dagmargoodboat@yahoo.com > wrote: > > >A problem encountered with medium-to-fast current-mode flyback switchers, > >when combining high-ratio multi-winding flyback inductors with multipliers: > > > >When Q1 switches on, a large current surge flows as Cpump charges from Cmult1 > >via D2. At Q1, the surge magnitude is multiplied by the transformer ratio. > >On simulation, U1 senses this via Rs and cuts Q1 off instantly, crippling the > >switcher -- it balks/stalls, and refuses to produce any output. > > > > Cpump > > +12V .------||-------. DA2FJ8100L (x 3) > > -+- 1 | D1 D2 | D3 > > | .--+-->|---+-->|---+--->|----+----> +HV > > o )||( PA0367A | | > > L1 )||( --- Cmult1 --- Cout > > n=1:12 )||( o --- --- > > | | | | > > U1 | === === === > >LTC3803-3 ||--' > >------. ||<-. Q1 > > Gate|->---'|--+ > > | | | i1 > > | | V > > Sense|<-[Rslp]-+ > > | | > >------' [Rs] > > | > > === > > > > > >Several blanking schemes to clamp, filter, or otherwise ignore the > >current spike were considered and rejected before settling on a > >solution. > > > >Recognizing that i(Cmulti) is -i(Q1)/n, a cancellation scheme > >removes the forward-mode i(Cpump) artifact from U1's worldview ... > > > > Cpump > > +12V .------||-------. DA2FJ8100L (x 3) > > -+- 1 | D1 D2 | D3 > > | .--+-->|---+-->|---+--->|----+----> +HV > > o )||( PA0367A | | > > L1 )||( --- Cmult1 --- Cout > > n=1:12 )||( o --- --- > > | | | | > > U1 | === | === > >LTC3803-3 ||--' | > >------. ||<-. Q1 | > > Gate|->-'|--+ | ^ > > | | | i1 | | i2 = i1/n > > | | V | > > | +-----[R1]------+ R1 ~= n*Rs > > | | | > > | [Rs] | > > | | | > > | === | > > | | > > Sense|<--------[Rslp]--------' > > | > >------' > > > > > >After modification, U1.sense only sees L1's charging current (the > >current storing energy for flyback operation); the forward-mode > >Cpump-charging spike is eliminated. > > > >Probing with the oscilloscope confirms that cancellation works > >essentially ideally in practice, a pleasant result. > > > >Cheers, > >James Arthur > > You could also > > use smaller caps in the multiplier
No. Cpump is already quite small, 1nF.
> or > > lowpass filter the current sense signal a bit
Tried that first, but no. A filter large enough to gobble the glitch destabilizes the supply. The main difficulty is that a fast current-mode switcher's built-in blanking interval is too short to blank out the forward-mode surge reflected back from charging Cpump.
> or > > use a less aggressive transformer and more multiplier stages.
Yes, the problem is diminished or absent with lower turns-ratios. A custom-wound transformer and no multiplier would have avoided the problem completely. But I needed the multiplier to stick within the limits of COTS magnetics. The one-time pain for me is recouped 1000x over, though, every time the product is built.
> This worked without tricks, somehow: > > https://dl.dropboxusercontent.com/u/53724080/Circuits/Power/HV/28S840A.pdf > > 1400 volts out at the top, with a 1:1 autotransformer. This may > current-limit for a few cycles at startup, but it works its way up.
Nice. A bonus of my topology is that there is no d.c. path from input to output, allowing indefinite short-circuit operation without excess draw, or damage. Cheers, James Arthur
On Tuesday, January 17, 2017 at 9:26:46 PM UTC-5, John Larkin wrote:

> This worked without tricks, somehow: > > https://dl.dropboxusercontent.com/u/53724080/Circuits/Power/HV/28S840A.pdf > > 1400 volts out at the top, with a 1:1 autotransformer. This may > current-limit for a few cycles at startup, but it works its way up.
Are those low-duty 350V spikes any problem for your inductor's 200Vac-rated insulation? I eliminated several magnetics options because I was uncomfortable with their inter-winding insulation specs. My supply takes feedback from the bleeder chain to regulate the HV output. Being able to make regulated HV from +12v in a couple square inches is cool. Cheers, James Arthur
On Wed, 18 Jan 2017 08:57:52 -0800 (PST), dagmargoodboat@yahoo.com
wrote:

>On Tuesday, January 17, 2017 at 9:26:46 PM UTC-5, John Larkin wrote: > >> This worked without tricks, somehow: >> >> https://dl.dropboxusercontent.com/u/53724080/Circuits/Power/HV/28S840A.pdf >> >> 1400 volts out at the top, with a 1:1 autotransformer. This may >> current-limit for a few cycles at startup, but it works its way up. > >Are those low-duty 350V spikes any problem for your inductor's >200Vac-rated insulation? I eliminated several magnetics options because >I was uncomfortable with their inter-winding insulation specs.
Well, it seems to work. I once took a twisted pair of #30 magnet wire and applied voltage; it arced at about 1500 volts. I don't think I've ever observed voltage breakdown in an inductor. It would be interesting to test a DRQ127 to destruction.
> >My supply takes feedback from the bleeder chain to regulate the HV output. >Being able to make regulated HV from +12v in a couple square inches is cool.
Yeah, I seldom do HV, so it's always interesting. PCB surface creepage limits are confusing. Here's my board: https://dl.dropboxusercontent.com/u/53724080/PCBs/T840_E1.jpg The layout is uglified by the clearance complications. I did that layout myself, which doesn't help either. Note the heat sinking on T1! -- John Larkin Highland Technology, Inc lunatic fringe electronics
dagmargoodboat@yahoo.com wrote...
> > A problem encountered with medium-to-fast current-mode > flyback switchers, when combining high-ratio multi- > winding flyback inductors with multipliers: [ snip ]
OK, I've thought about it and I like your scheme. And also, thanks for the awesome ASCII drawings. We don't see them much anymore and they really do facilitate understanding and discussion. -- Thanks, - Win
Not wise -- you might as well go with a 555 at that point.

Not hyperbole -- you're defeating the instantaneous peak current detection 
the chip offers!

The better solution is simply not to run a multiplier: it's absolutely the 
wrong second half to pair with the first half of the circuit.  You're 
putting a lion head on fish...

So how to deal with it?  A few options:

1. Use a resonant converter, so the switcher doesn't care about peak 
currents (it will probably be a combination of average current mode control, 
and phase control).

2. Use a current limited switch, so that you don't get "transistor 
hard-switches into brick-wall capacitance" action.  This is done here for 
example,
https://www.seventransistorlabs.com/Images/Deadbug_Sch.png
the 2N4401 is switching, in the sense that it's going from "conducting" to 
"off".  But it's not saturating, so it does dissipate a lot of power 
(relatively speaking).  How much power, depends on the ratio of input to 
output voltages.  (If the simplified discrete design is throwing you for a 
loop, plug this into a simulator.)

3. Use a fully current-sourcing topology.  Set up a constant current supply, 
then drive a full wave converter (push-pull or full bridge).  The supply 
should be a buck converter (running at an even harmonic frequency, or a 
non-harmonic frequency) for efficiency purposes.

Example here:
https://www.seventransistorlabs.com/Images/HVPower1.png
Linear current source, so it's not made for efficiency.  Also, the 
oscillator doesn't work very well at low currents/voltages, and can't be 
adjusted to zero because of gate bias (which flows into the drain circuit). 
It's also resonant, though not very; the transformer ended up with a much 
higher coupling factor than I was expecting, so it's really quite 
transformery in nature, rather than acting as coupled inductors.  Either way 
will work, though.

4. Use more turns.  Usually the least desirable option for circuits like 
these, but at least you don't have to endure massive winding capacitance: by 
dividing one high voltage winding into a stack of smaller DC supplies (each 
with a manageable winding length and diode Vrrm rating), the equivalent 
capacitance goes as O(N), not O(N^2).

It can even go as O(1) with the right winding design -- namely, if the high 
voltage windings are single layers each, so there's "DC" at one end and "AC" 
at the other end, for each winding.  This puts the same AC voltages on each 
point of each winding, so there's zero AC voltage between windings -- their 
capacitances don't stack at all!  This is how CRTs achieved >100kHz 
bandwidth at 30kV in their flyback transformers.  (The bandwidth is given by 
whatever harmonic of 15.75kHz is needed to finish retrace.)

(FYI, multisync monitors used a separate FBT for HV power, so they didn't 
need to drive stupendous sweep rates (>100kHz!) through the high voltage 
circuit.  They used an independent, dedicated deflection system.)

Tim

-- 
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Contract Design
Website: http://seventransistorlabs.com


<dagmargoodboat@yahoo.com> wrote in message 
news:b6488e07-e4a0-4859-a5df-17e61e5fd927@googlegroups.com...
>A problem encountered with medium-to-fast current-mode flyback switchers, > when combining high-ratio multi-winding flyback inductors with > multipliers: > > When Q1 switches on, a large current surge flows as Cpump charges from > Cmult1 > via D2. At Q1, the surge magnitude is multiplied by the transformer ratio. > On simulation, U1 senses this via Rs and cuts Q1 off instantly, crippling > the > switcher -- it balks/stalls, and refuses to produce any output. > > Cpump > +12V .------||-------. DA2FJ8100L (x 3) > -+- 1 | D1 D2 | D3 > | .--+-->|---+-->|---+--->|----+----> +HV > o )||( PA0367A | | > L1 )||( --- Cmult1 --- Cout > n=1:12 )||( o --- --- > | | | | > U1 | === === === > LTC3803-3 ||--' > ------. ||<-. Q1 > Gate|->---'|--+ > | | | i1 > | | V > Sense|<-[Rslp]-+ > | | > ------' [Rs] > | > === > > > Several blanking schemes to clamp, filter, or otherwise ignore the > current spike were considered and rejected before settling on a > solution. > > Recognizing that i(Cmulti) is -i(Q1)/n, a cancellation scheme > removes the forward-mode i(Cpump) artifact from U1's worldview ... > > Cpump > +12V .------||-------. DA2FJ8100L (x 3) > -+- 1 | D1 D2 | D3 > | .--+-->|---+-->|---+--->|----+----> +HV > o )||( PA0367A | | > L1 )||( --- Cmult1 --- Cout > n=1:12 )||( o --- --- > | | | | > U1 | === | === > LTC3803-3 ||--' | > ------. ||<-. Q1 | > Gate|->-'|--+ | ^ > | | | i1 | | i2 = i1/n > | | V | > | +-----[R1]------+ R1 ~= n*Rs > | | | > | [Rs] | > | | | > | === | > | | > Sense|<--------[Rslp]--------' > | > ------' > > > After modification, U1.sense only sees L1's charging current (the > current storing energy for flyback operation); the forward-mode > Cpump-charging spike is eliminated. > > Probing with the oscilloscope confirms that cancellation works > essentially ideally in practice, a pleasant result. > > Cheers, > James Arthur
On Wednesday, January 18, 2017 at 7:02:29 PM UTC-5, Tim Williams wrote:
> Not wise -- you might as well go with a 555 at that point. > > Not hyperbole -- you're defeating the instantaneous peak current detection > the chip offers!
Not true. The chip already has to blank the FETs Cgs glitch, and has to ignore L1's parallel capacitance-charging surge. I've extended that facility to cover the multiplier, too. Peak current detection is not degraded, it's enhanced--now the SMPS can detect actual current much earlier, during times when the Vsense ramp was previously overwhelmed by the undesired signal. By adding one resistor I permit the SMPS to ignore a confounding signal and operate normally over a much wider dynamic range. (That's important in my application.) A (trivial) circuit senses actual output current and initiates a 'chirp' mode when the supply is overloaded. Icc drops from 100's of mA to 20mA in that condition, and dissipation is essentially nil.
> The better solution is simply not to run a multiplier: it's absolutely the > wrong second half to pair with the first half of the circuit. You're > putting a lion head on fish... > > So how to deal with it? A few options: > > 1. Use a resonant converter, so the switcher doesn't care about peak > currents (it will probably be a combination of average current mode control, > and phase control). > > 2. Use a current limited switch, so that you don't get "transistor > hard-switches into brick-wall capacitance" action. This is done here for > example, > https://www.seventransistorlabs.com/Images/Deadbug_Sch.png > the 2N4401 is switching, in the sense that it's going from "conducting" to > "off". But it's not saturating, so it does dissipate a lot of power > (relatively speaking). How much power, depends on the ratio of input to > output voltages. (If the simplified discrete design is throwing you for a > loop, plug this into a simulator.)
Have you considered positive feedback to turn your switch transistor off faster, rather than passively? That would save you some dissipation. A small capacitor from the switch's collector to the base of your current-sensor transistor would make the whole thing snap off.
> 3. Use a fully current-sourcing topology. Set up a constant current supply, > then drive a full wave converter (push-pull or full bridge). The supply > should be a buck converter (running at an even harmonic frequency, or a > non-harmonic frequency) for efficiency purposes. > > Example here: > https://www.seventransistorlabs.com/Images/HVPower1.png > Linear current source, so it's not made for efficiency. Also, the > oscillator doesn't work very well at low currents/voltages, and can't be > adjusted to zero because of gate bias (which flows into the drain circuit). > It's also resonant, though not very; the transformer ended up with a much > higher coupling factor than I was expecting, so it's really quite > transformery in nature, rather than acting as coupled inductors. Either way > will work, though. > > 4. Use more turns. Usually the least desirable option for circuits like > these, but at least you don't have to endure massive winding capacitance: by > dividing one high voltage winding into a stack of smaller DC supplies (each > with a manageable winding length and diode Vrrm rating), the equivalent > capacitance goes as O(N), not O(N^2). > > It can even go as O(1) with the right winding design -- namely, if the high > voltage windings are single layers each, so there's "DC" at one end and "AC" > at the other end, for each winding. This puts the same AC voltages on each > point of each winding, so there's zero AC voltage between windings -- their > capacitances don't stack at all! This is how CRTs achieved >100kHz > bandwidth at 30kV in their flyback transformers. (The bandwidth is given by > whatever harmonic of 15.75kHz is needed to finish retrace.) > > (FYI, multisync monitors used a separate FBT for HV power, so they didn't > need to drive stupendous sweep rates (>100kHz!) through the high voltage > circuit. They used an independent, dedicated deflection system.) > > Tim
Cheers, James Arthur
On Wednesday, January 18, 2017 at 12:42:37 PM UTC-5, John Larkin wrote:
> On Wed, 18 Jan 2017 08:57:52 -0800 (PST), dagmargoodboat@yahoo.com > wrote: > > >On Tuesday, January 17, 2017 at 9:26:46 PM UTC-5, John Larkin wrote: > > > >> This worked without tricks, somehow: > >> > >> https://dl.dropboxusercontent.com/u/53724080/Circuits/Power/HV/28S840A.pdf > >> > >> 1400 volts out at the top, with a 1:1 autotransformer. This may > >> current-limit for a few cycles at startup, but it works its way up. > > > >Are those low-duty 350V spikes any problem for your inductor's > >200Vac-rated insulation? I eliminated several magnetics options because > >I was uncomfortable with their inter-winding insulation specs. > > Well, it seems to work. I once took a twisted pair of #30 magnet wire > and applied voltage; it arced at about 1500 volts. > > I don't think I've ever observed voltage breakdown in an inductor. It > would be interesting to test a DRQ127 to destruction. > > > > >My supply takes feedback from the bleeder chain to regulate the HV output. > >Being able to make regulated HV from +12v in a couple square inches is cool. > > Yeah, I seldom do HV, so it's always interesting. PCB surface creepage > limits are confusing. > > Here's my board: > > https://dl.dropboxusercontent.com/u/53724080/PCBs/T840_E1.jpg > > > The layout is uglified by the clearance complications. I did that > layout myself, which doesn't help either.
Looks decent to me. I fretted the clearances until I noted the creepage on a 1kV D-PAK switching FET elsewhere--20V/mil. I think I'm under 8 volts/mil on the board, worst-case. Then we're coating, which will cover all the bare metal.
> Note the heat sinking on T1!
Neat. Curiously, my inductor doesn't get warm. I expected it would, but it doesn't. Cheers, James Arthur
On Wednesday, January 18, 2017 at 5:58:29 PM UTC-5, Winfield Hill wrote:
> dagmargoodboat@yahoo.com wrote... > > > > A problem encountered with medium-to-fast current-mode > > flyback switchers, when combining high-ratio multi- > > winding flyback inductors with multipliers: [ snip ] > > OK, I've thought about it and I like your scheme. > And also, thanks for the awesome ASCII drawings. > We don't see them much anymore and they really > do facilitate understanding and discussion. > > > -- > Thanks, > - Win
I'm more and more pleased with the cancellation. One always wonders what new can of worms one opens with a new technique, but I haven't seen it yet on this one. I'm also an ASCII fan. It's an easy way to spread ideas and preserve them. Cheers, James Arthur