MAX1044 / ICL7660S High Voltage SMPS Experiment - don't try this at home!

[frequencycentral]

There's quite a bit of talk about HV boost SMPS around here, for use in our tube projects. I've built a few different types, NE555 and MAX1771 based. It occurred to me that it might be possible to make one using a MAX1044 / ICL7660S, of which I have many lying around. ICL7660S are really cheap...... so just out of curiosity I decided to try it out. This is really just a record of my experiment, not sure if it will go anywhere, or if this is a viable SMPS when under load, how many ma it's good for, or how long it will last before burnin' up! But the prototype appears stable at the moment.........

My first experiment had the source of the IRF840 connected directly to ground, the HV was really high, like over 500v, and the MOSFET got really hot real quick, even though it was heatsinked. So I tried reducing the oscillator frequency of the ICL7660S I'm using by adding a cap to ground from pin 7. Voltage dropped, but MOSFET is still very hot. Next I tried adding a resistor between the IRF840's source and ground, I happen to have a lot of 18R/2watt lying around, so that's what I used. Voltage is now stable and the heatsinked MOSFET as well as Rsense are 'acceptably' hot IMO. The boost feature of the MAX1044 / ICL7660S has been ulitised, but it's also possible to fine tune the chip's frequency by adding a cap to ground from pin 7. Therefore it's possible to tune the circuit by combination of the value of the Rsense resistor and by adding a cap to ground from pin 7 of the MAX1044 / ICL7660S. A combination of 9R for Rsense and a 1n cap to ground from pin 7 yeilds 215v for example.

[Hides-His-Eyes]

So does wiring in a "load" of say 1k (hundreds of mA depending on the output) give you some stable consistency or is there something I'm missing about that approach?

[iccaros]

There's quite a bit of talk about HV boost SMPS around here, for use in our tube projects. I've built a few different types, NE555 and MAX1771 based. It occurred to me that it might be possible to make one using a MAX1044 / ICL7660S, of which I have many lying around. ICL7660S are really cheap...... so just out of curiosity I decided to try it out. This is really just a record of my experiment, not sure if it will go anywhere, or if this is a viable SMPS when under load, how many ma it's good for, or how long it will last before burnin' up! But the prototype appears stable at the moment.........

My first experiment had the source of the IRF840 connected directly to ground, the HV was really high, like over 500v, and the MOSFET got really hot real quick, even though it was heatsinked. So I tried reducing the oscillator frequency of the ICL7660S I'm using by adding a cap to ground from pin 7. Voltage dropped, but MOSFET is still very hot. Next I tried adding a resistor between the IRF840's source and ground, I happen to have a lot of 18R/2watt lying around, so that's what I used. Voltage is now stable and the heatsinked MOSFET as well as Rsense are 'acceptably' hot IMO. The boost feature of the MAX1044 / ICL7660S has been ulitised, but it's also possible to fine tune the chip's frequency by adding a cap to ground from pin 7. Therefore it's possible to tune the circuit by combination of the value of the Rsense resistor and by adding a cap to ground from pin 7 of the MAX1044 / ICL7660S. A combination of 9R for Rsense and a 1n cap to ground from pin 7 yeilds 215v for example.

Rick- are you using the same MOSFET you were using before ( I can not see the drawing from work)? this one http://www.farnell.com/datasheets/39198.pdf has a good impedance which should be more efficient. based on this site. http://www.desmith.net/NMdS/Electronics/NixiePSU.html, which should allow for faster switching.

At the bottom he list changes to get 129MA @ 187v from the changes to MOSFET and inductor.  I was looking at building one, but the ground plane makes me wonder if I can do it with out a board. 

[iccaros]

Danger...-- This has not been tested

Also this circuity puts out possible dangerous voltages, I take no responsibility if you do not understand how to handle high voltages and get hurt and or DIE...

Here is the PCB layout I am going to test with this --> http://dl.dropbox.com/u/14312589/Max1044Power%20Supply%20-%20mirror.pdf
component layout -->


U1 = max1044 or ICL7660S
D1 = UF4004
L1 = 100uh
T1=IFR840
C1= 100uf
C2= 10uf
C3= .1 uf

C5 see Ricks notes on voltage out
C6 = 4.7uf 350v

R sense = see Ricks Notes on Voltage out

This has not Been Tested

[R.G.]

The 1044 has no feedback from the output. What voltage you get out is dependent on the inductor, the switching frequency and duty cycle, and any loading on the inductor.

The 1044 has a 50% duty cycle on its bucket pins. The inductor is then turned on for a time 1/F, where F is half the 1044 oscillator frequency. Call this "t".

For an inductor, V = L di/dt, or said another way di = (V/L)*dt, where di = the change in current and dt is the elapsed time. If the inductor is initially at zero current, it ramps up at a rate of V/L amps per second. If dt is short enough that L does not saturate, then it charges up to an energy of E = LI²/2 each cycle.

When the MOSFET turns off the energy has to go somewhere. The inductor flips its voltage to keep the current flowing, forward biases the diode, and dumps the energy into the output cap. The energy in a cap is E = CV²/2, and the cap eats all the energy in the inductor if the MOSFET is off long enough for the inductor to discharge it. The cap voltage changes by an amount to make CV²/2 equal  E = LI²/2 each cycle. As the cap charges up to higher voltages, and the inductor discharges faster into the higher voltages. This is important because if the cap can't eat all the inductor's energy before the MOSFET turns on again, the inductor current is not reset to zero, and the next cycle it charges to a higher current, and so on each cycle. This is called "inductor walking"  and eventually saturates even an inductor which does not fully saturate on single cycles.

With no regulation, simply a 50% duty cycle at some repetition rate 2*t, whether the MOSFET, inductor, diode and cap live or not depends on whether the inductor saturates, and how hard. If the time is long enough (that is, frequency is slow enough) the inductor will saturate and no longer limit current. The MOSFET is then doing its best to drag down the supply voltage by acting as a short circuit. It gets hot. Sometimes very. Speeding up the frequency shortens t, and keeps the inductor from ramping up as high (di = V/(F*L) and keeps it out of saturation. This is easier on the MOSFET.

Inserting Rsense lowers the MOSFET dissipation by raising the source voltage, cutting down Vgs, and making the MOSFET stop conducting at some current which cuts the Vgs back. It's not really feedback, sensing, or limiting, but the MOSFET conducts less current when the inductor saturates, so it is cooler than if it were trying to be a short circuit.

How high does the output voltage go? The capacitor is being charged with a pulse of energy of E = LI²/2 at a rate of F times per second, so the cap voltage increases ... without limit   until something breaks, or until some output load exists and eats that much power (power = F*E). So the output is the voltage which makes the output load eat F*LI²/2. The voltage increases until that happens. What "loads" it may be the capacitor dielectric breaking down, contaminants on the PCB, diode leakage, your finger, etc.  If you put a resistor on the output, the voltage drops until the resistor power (V²/R) equals F*LI²/2. Whatever that is.

With no means of changing either the frequency or the pulse width to the inductor, the voltage varies inversely with the square root of the load resistance.

Choosing an inductor and an operating frequency is a Big Deal in a circuit like this. The choice of the inductor needs to include it's saturation current as well as its inductance.

[iccaros]

The 1044 has no feedback from the output. What voltage you get out is dependent on the inductor, the switching frequency and duty cycle, and any loading on the inductor.

The 1044 has a 50% duty cycle on its bucket pins. The inductor is then turned on for a time 1/F, where F is half the 1044 oscillator frequency. Call this "t".

For an inductor, V = L di/dt, or said another way di = (V/L)*dt, where di = the change in current and dt is the elapsed time. If the inductor is initially at zero current, it ramps up at a rate of V/L amps per second. If dt is short enough that L does not saturate, then it charges up to an energy of E = LI²/2 each cycle.

When the MOSFET turns off the energy has to go somewhere. The inductor flips its voltage to keep the current flowing, forward biases the diode, and dumps the energy into the output cap. The energy in a cap is E = CV²/2, and the cap eats all the energy in the inductor if the MOSFET is off long enough for the inductor to discharge it. The cap voltage changes by an amount to make CV²/2 equal  E = LI²/2 each cycle. As the cap charges up to higher voltages, and the inductor discharges faster into the higher voltages. This is important because if the cap can't eat all the inductor's energy before the MOSFET turns on again, the inductor current is not reset to zero, and the next cycle it charges to a higher current, and so on each cycle. This is called "inductor walking"  and eventually saturates even an inductor which does not fully saturate on single cycles.

With no regulation, simply a 50% duty cycle at some repetition rate 2*t, whether the MOSFET, inductor, diode and cap live or not depends on whether the inductor saturates, and how hard. If the time is long enough (that is, frequency is slow enough) the inductor will saturate and no longer limit current. The MOSFET is then doing its best to drag down the supply voltage by acting as a short circuit. It gets hot. Sometimes very. Speeding up the frequency shortens t, and keeps the inductor from ramping up as high (di = V/(F*L) and keeps it out of saturation. This is easier on the MOSFET.

Inserting Rsense lowers the MOSFET dissipation by raising the source voltage, cutting down Vgs, and making the MOSFET stop conducting at some current which cuts the Vgs back. It's not really feedback, sensing, or limiting, but the MOSFET conducts less current when the inductor saturates, so it is cooler than if it were trying to be a short circuit.

How high does the output voltage go? The capacitor is being charged with a pulse of energy of E = LI²/2 at a rate of F times per second, so the cap voltage increases ... without limit   until something breaks, or until some output load exists and eats that much power (power = F*E). So the output is the voltage which makes the output load eat F*LI²/2. The voltage increases until that happens. What "loads" it may be the capacitor dielectric breaking down, contaminants on the PCB, diode leakage, your finger, etc.  If you put a resistor on the output, the voltage drops until the resistor power (V²/R) equals F*LI²/2. Whatever that is.

With no means of changing either the frequency or the pulse width to the inductor, the voltage varies inversely with the square root of the load resistance.

Choosing an inductor and an operating frequency is a Big Deal in a circuit like this. The choice of the inductor needs to include it's saturation current as well as its inductance.

RG I thought the frequency or was changed in combination of C5 (mine ping 7 on the MAX1044)  and R sense.
I was planning on using a 1.5 amp 100uh for L1

[R.G.]

The Max1044 datasheet says that frequency is nominally 10kHz without "boost" and 25kHz with boost. Capacitance on pin 7 decreases that.

So the time for the inductor to charge is half of 1/10KHz and half of 1/25kHz, or 50uS and 20uS respectively, and that increases as you put capacitance on pin 7. As you increase the capacitance from 1pF up to 0.1uF, frequency decreases down to the 1Hz and 10Hz range.

A 100uH inductor lets current rise at a rate (i.e. di/dt) of V/L, or 9V/100uH = 90,000 amps/second. It reaches 1.5A in 1.5/90,000 = 16.7uS. So as fast as you can run the MAX1044 with boost on will saturate a 1.5A/100uH inductor if 1.5A is a cliff the inductor falls off of. In reality, that is probably the beginning of a soft saturation, so it's not a binary disaster.

But it does account for the MOSFET getting hot.

Increasing Rsense does not a thing for the timing on the 1044. All it does is put some feedback into the MOSFET. The MOSFET gate is being fed a 9V square wave (roughly) and so it tries to conduct a current equal to the input voltage times its transconductance in amps/volt minus its turn-on voltage. For the IRF840, turn on voltage is between 2 and 4V, and transconductance is a minimum of 4.9S (i.e., amps per volt), so the IRF840 tries to conduct

(9V-Vgth)*(4.9A/V)-I*Rsense. 

If the available current is less than the Vgs tells it to conduct, it will saturate to 0.85 volt/ampere. As Rsense is raised, the conduction current times Rsense lowers the available Vgs and pulls the MOSFET out of saturation and into linear/current limited conduction, which limits the current through the inductor and dissipates the current times Vds voltage in the MOSFET. This may reduce the dissipation in the MOSFET by preventing the MOSFET from conducting much higher currents from a saturating inductor.

At saturation, the ferromagnetic core of the inductor changes from a permeability of several thousand down to a permeability of one, so the incremental inductance drops from the low-current case down to the inductance the inductor would have if it had no ferromagnetic core. How big this change is depends on how gapped the core is. For rod cores, it may be a minor change, since the core is a minor component of the M-field path length, most if it being the return path through the air around the core. For gapped pot cores, E-cores, C cores, etc., this can be a change of thousands to one.

The drop in incremental inductance means the rate of change of current with time goes from 90,000 A/sec up to between maybe ten and several thousand times more than that, or from about an amp per microsecond up to tens to hundreds of amps per microsecond, which means that either the power source limits or something breaks, limits, or burns up.

Or all of the above.  I have an assortment of psychic scars from all of these mechanisms.

[frequencycentral]

Having read RG's posts twice or more, how much I don't know is Keenly brought into focus. I think he's saying 'don't try this at home'.....

[R.G.]

Having read RG's posts twice or more, how much I don't know is Keenly brought into focus. I think he's saying 'don't try this at home'.....

Most absolutely not. Trying it at home is how Georg Ohm figured out his Law. What I was trying to say or show was that some simple algebra can really bring this stuff into perspective. There are very few principles involved here, and figuring out what affects what is not all that complicated. You need to know:

1) Ohm's law
2) the time response of an inductor - V = Ldi/dt
3) the time response of a capacitor - I = Cdv/dt
4) the energy in an inductor -  E = 1/2 *L* I²
5) the energy in a capacitor -  E = 1/2 * C * V²
6) the fact that power = energy * frequency
7) the fact that inductors do saturate at some current
and something about the 1044's switching on and off.

For instance: if you want to keep the inductor out of saturation, don't slow down the switching by adding a capacitor to pin 7, and increase the inductance to 100uH, possibly by putting another 100uH in series. Since di/dt = V/L and V is fixed, doubling L means you ramp up current half as fast.

Also: slowing down the frequency of oscillation does lower the output voltage, but it does it by reducing the number of buckets of energy being dumped over into the cap every second. With fewer buckets coming over, the same cap charges less quickly because fewer energy buckets are coming in.

Also: changing the load resistor changes the voltage because P = V²/R, and with a fixed P (same inductor, same peak current, same frequency), the power remains the same, so the voltage must change inversely with the square root of P/R.

Quite the opposite of "don't do this", it's a great opportunity to learn some stuff.

The stuff I just listed up top is the fundamental basis of electronic power conversion. Things get fancier, faster, and more complex, but moving power around and storing/dumping inductors and caps is where it starts. It's easy to learn just a few things, and start understanding power conversion.

[iccaros]

So I am not going to Lie and say this is simple math, I have to look up half of the variables, but I think I get the drift. And please forgive this post, as I am trying to put this together, using what I think I know and matching to what I think I am reading.

question, since the input power supply is a limiting factor, but your saying that the Fet is going to try and draw max current so it could damage the power supply if it can not supply max current? Should we not see this as lower voltage as power in = power out. So I would expect that if we supply 1.5 amps at say 12 volts and try to put out 120 volts we would be limited to .15 amps? no ?

In Ricks 555 based smps or this one http://www.ledsales.com.au/kits/nixie_supply.pdf
q2 is a safety part and voltage control, by shutting down the oscillator when the voltage hits a specific point determined by the voltage divider.  By shutting down the timer until voltage reaches a specified level by the adjustment of VR1, creates a duty cycle of the transistor state. While this could exceed saturation time, we see the effects of more heat on the FET, which is protected by a limited power supply and the fact that the voltage divider restricts max voltage out. no?

While some things could be changed, like say put a 300uh inductor, so that it takes about 37us (300/12)*1.5 = 37.5 to reach saturation at 1.5amps, I am not smart enough yet to figure out how to set duty cycle out of a max1044, except by lowering frequency with a cap on pin 7, unless more parts are added, like an and gate with a cap and resistor calculated at 37us, God knows I do not know how to do this, this would make the turn FET on cycle at frequency with a given duty cycle.


then finding max power out
Power=(((rise time)*(Volts in)2)/(2*Inductor uH))
37.5*12 squared / 2* 300 = 8.7 watts

convert to mA out
mA=((Power Watts)/(output volts))*1000

8.7/165 *1000 = 52.72 ma out






this example uses a PIC and PWM, but you have to calculate source voltage and inducter fill rate to program the frequency
http://www.instructables.com/id/High-Voltage-Switch-Mode-Power-Supply-SMPSBoost/

[iccaros]

for anyone intrested, here is a spreadsheet from the link I gave
http://dl.dropbox.com/u/14312589/InductCoilSMPS.xls

It lets you calculate inductor charge time and max current out..  RG please let me know if this look right to you.. I have applied the little math I know,, but wow