Need a DC +12v to -15v converter good for at least 50mA. Any ideas?

[italianguy63]

I'll make another stupid suggestion...

What about a cheap DC to DC Buck Converter?

MC

Edit-- $1.26 shipped on FleaBay

https://www.ebay.com/itm/2A-DC-DC-Boost-Buck-Converter-Step-up-Voltage-Module-Adjustable-3V-5V-9V-12V-24V/323013308636?hash=item4b3517f0dc:g:-OUAAOSwEIJbNKc5

Nevermind!  You still have to invert the voltage.  Back to a charge pump!

MC

[Rob Strand]

As far as board space is concerned, absolutely. Even if SMT packages are used for all of the drive and pump components, it'll still be bigger than a single 8-PDIP IC. In terms of economical gain, it's still "feasible", soft of. A decent charge-pump IC that works above 12V, LTC1144 (which still drops 2 volts under 40mA load...), costs about $4.95 in a single quantity. NE555 with a complementary pair of MOSFETs, Schottky's and caps cost slightly above $2. And then there is an educational value. Throwing parts until it "just works" can be satisfying too. 

Very true.   At home I often try to build the best circuit I can with what I have on-hand; and spend a lot of hours thinking and tweaking things.    I guess that's a challenge in itself.   At work I'd never do such a thing, although I do factor in longevity of special parts.

FWIW,  those circuits I posted aren't new.   Circuit 1 and Circuit 2 have been around for some time.   I used a circuit similar to Circuit 2 back in 1990.  I believe it might have originated in the Motorola TMOS applications notes or data sheets - the early days of power MOSFETs.  Circuit 1 is used in a zillion consumer goods which drive synchronous motors, like washing machines.

A completely different approach would be to use something like a CMOS 4017 driving some MOSFETs.  Basically leave out one clock cycle on each edge to prevent shoot-through.

[Rob Strand]

I had a poke around in the old Motorola TMOS data and found this.




A trick for using this on high voltages was to connect the anode of the diode to a lower voltage rail like +12V.

[anotherjim]

Next minor improvement was to slew the 555 output with a small value RC network. This has reduced the shoot-thru and improved the pump action. The R in the RC can only be small otherwise it lets shoot-thru increase due to the path from PNP to NPN via base resistors - as you might guess.

[anotherjim]

Update, RC to slew it doesn't need an R, just a C.
A few tweaks and it's much better. Transistors cool. All of it takes 120mA for the 40mA load.

D1 did a bit of magic. I suppose it helps the 555 output not quite reaching +V and makes sure the PNP switches off faster. Output is stronger, even though D1 robs 0.6v or so from the pump supply.

[Rob Strand]

D1 did a bit of magic. I suppose it helps the 555 output not quite reaching +V and makes sure the PNP switches off faster. Output is stronger, even though D1 robs 0.6v or so from the pump supply.

You have done pretty well to get it going.  D1 allows the top transistor to turn off fully.

The output circuit will have quite a bit of shoot-though.  If you imaging the DC output of the 555 slowly changing from 0 to Vcc-1.4V   there is a wide zone on the middle where both transistors are on an they conduct a lot of current (+V ->  Q1.e -> Q1.c -> Q2.c -> Q2.e -> 0V)

If you change the base-emitter resistors you can reduce the degree of overlap where both transistors are conducting.  For a 12V rail the 555 will swing roughly 0 to 10.6V.  So ideally you don't want the top transistor conducting until the 55 output voltage is below say 5.3V and you don't want the bottom transistor conducting until the 555 input voltage goes over 5.3V.

I've done these changes and you can compare the two schematics.   I've also moved the level shifting to the base to remove the diode drop on the + rail.     The addition of the Base-emitter resistors should also speed-up the switching.  (There's two types of shoot through: one from the DC transition and one from the transistor switch time.)   In order for apples to apples comparisons I've set the base drive current in the modified circuit to be the same as the original.  I don't know if you need more or less base drive.  With the 2N3904/2N3906 the Vce_sat levels off so you don't gain by pumping more base current.

So here's the modified schematic,


Here you can see the base currents as the 555 output voltage goes from 0 to 10.6V.   You can see on the original design the base current start increasing very early on but with the modifications the base current doesn't start to increase/decrease until the 555 output gets near mid rail.  Note also the final base drive when the 555 output is sitting at 0V or 10.6V is the same for both circuits.


Here's the improvements in the shoot-through current for the same DC transition.   The original circuit is dumping a heap of current through the whole DC transition:


I'm not sure what the slow-down cap is doing on your circuit because slowing down the transition will make it stay in the shoot-through region longer.    If the transisitons are fast it stays in the DC shoot-through region for less time and has less losses.  After that you just have to worry about shoot-through due the transistors not turning-off fast enough (a dynamic effect).


Note R555 and R556 are not real resistors, they are not part of the circuit.  They are just a crude way to model the internal voltage drop of the 555.

[I forgot to add that the penalty of removing the shoot-through current is smaller input resistors on the bases.  There is a static loss due the power dissipation in these resistors.  The "optimum" would be more like when the static loss is about the same as the dynamic loss through the transistion.  I haven't calculated it (as it is a function of the switch frequency) but that *could* mean *more* over-lap is better for efficiency as a whole.]

[anotherjim]

Wow! Thank you Mr Strand. I'll try that later.
I did try the diode in the PNP base, but duh, forgot to lower the base resistor to compensate & got reduced output!

[Rob Strand]

I did try the diode in the PNP base, but duh, forgot to lower the base resistor to compensate & got reduced output!

The diode mod is definitely a step forward.

As for the overlap: I had a quick look at the power dissipated assuming a 100nS rise/fall time from the 555.  Since the 100nS transition time is so small, the wasted power due to shoot-through doesn't seem much at 50kHz.   The power "wasted" in the series base resistors is much more than the shoot-through losses.   I'd probably go with Rbe resistors of around 1k and use the highest series base resistors as possible (maybe 1k's?).   While I'm not happy with overall shoot-though current the overall efficiency says not to worry about it!

[diffeq]

D1 did a bit of magic. I suppose it helps the 555 output not quite reaching +V and makes sure the PNP switches off faster. Output is stronger, even though D1 robs 0.6v or so from the pump supply.

You have done pretty well to get it going.  D1 allows the top transistor to turn off fully.

The output circuit will have quite a bit of shoot-though.  If you imaging the DC output of the 555 slowly changing from 0 to Vcc-1.4V   there is a wide zone on the middle where both transistors are on an they conduct a lot of current (+V ->  Q1.e -> Q1.c -> Q2.c -> Q2.e -> 0V)

If you change the base-emitter resistors you can reduce the degree of overlap where both transistors are conducting.  For a 12V rail the 555 will swing roughly 0 to 10.6V.  So ideally you don't want the top transistor conducting until the 55 output voltage is below say 5.3V and you don't want the bottom transistor conducting until the 555 input voltage goes over 5.3V.

I've done these changes and you can compare the two schematics.   I've also moved the level shifting to the base to remove the diode drop on the + rail.     The addition of the Base-emitter resistors should also speed-up the switching.  (There's two types of shoot through: one from the DC transition and one from the transistor switch time.)   In order for apples to apples comparisons I've set the base drive current in the modified circuit to be the same as the original.  I don't know if you need more or less base drive.  With the 2N3904/2N3906 the Vce_sat levels off so you don't gain by pumping more base current.

So here's the modified schematic,


Here you can see the base currents as the 555 output voltage goes from 0 to 10.6V.   You can see on the original design the base current start increasing very early on but with the modifications the base current doesn't start to increase/decrease until the 555 output gets near mid rail.  Note also the final base drive when the 555 output is sitting at 0V or 10.6V is the same for both circuits.


Here's the improvements in the shoot-through current for the same DC transition.   The original circuit is dumping a heap of current through the whole DC transition:


I'm not sure what the slow-down cap is doing on your circuit because slowing down the transition will make it stay in the shoot-through region longer.    If the transisitons are fast it stays in the DC shoot-through region for less time and has less losses.  After that you just have to worry about shoot-through due the transistors not turning-off fast enough (a dynamic effect).


Note R555 and R556 are not real resistors, they are not part of the circuit.  They are just a crude way to model the internal voltage drop of the 555.

[I forgot to add that the penalty of removing the shoot-through current is smaller input resistors on the bases.  There is a static loss due the power dissipation in these resistors.  The "optimum" would be more like when the static loss is about the same as the dynamic loss through the transistion.  I haven't calculated it (as it is a function of the switch frequency) but that *could* mean *more* over-lap is better for efficiency as a whole.]

Thanks for this circuitry and thorough analysis. Never thought there's more nuances to BJT switching.

Can control pin 5 of the 555 theoretically be used to reduce shoot-through by varying the duty cycle?

[amptramp]

I am not sure a charge pump is the answer for currents of more than 10 mA and the OP wants 50 mA.  The 555 has an internal shoot-through even if you manage to avoid it in the external circuitry.  A negative flyback converter topology would easily handle the current but it would require an inductor.  The flyback could be implemented with comparators so you would not need specialized converter IC's.

[Rob Strand]

Can control pin 5 of the 555 theoretically be used to reduce shoot-through by varying the duty cycle?

Can't be done.  For "proper" avoidance of feed-through you actually need gaps in the switching between when the bottom transistor turns off and the top turns on (and the same idea on the other transition).  In power electronics the gap is call the "dead time".   Simple circuits use diodes + resistors + caps and CMOS Schmitt-trigger inverters.  Modern stuff uses a microprocessor and the timing is done in software.

The 555 has an internal shoot-through even if you manage to avoid it in the external circuitry.

Yes.  (IIRC the datasheet recommends a 100nF across the rails to help filter the glitch.)

At this point I'd probably go for 1k's for the series base resistors and say 330ohm for the base-emitter resistors.   That should be able to drive 100mA at the output of the transistors. The shoot through current is in the order of 100mA.   Overall efficiency not bad. 

The exact values depend largely on the maximum output current.  It's easy to over tweak a circuit on the bread board.  The breadboard values might not produce maximum output current if the transistor gains were low on another build.

[anotherjim]

With the changes in the booster, it was putting out -17v on a 330R test load (which is pretty much 50mA). So there's room for a Zener & series pass transistor output.
Nothing is getting hot, although input current is now 130mA. I'm not sure if it's worth taking the time to shave that down, although I'm sure some reduction is possible. In the final power supply, it can run off a 7812 supplying the units +12v and still come under 1A total from that.
I found the trimmer on the 555 timing can be twiddled for maximum voltage at D6 anode. It's close to 50kHz on the breadboard. Duty is under 50%, it doesn't seem too fussy about that.

[Rob Strand]

Nothing is getting hot, although input current is now 130mA. I'm not sure if it's worth taking the time to shave that down, although I'm sure some reduction is possible. In the final power supply, it can run off a 7812 supplying the units +12v and still come under 1A total from that.

The base resistors probably kick the current up a bit.

It's preferable to move R4 to the base-emitter of Q1 as is speeds up Q1's turn-off time by removing stored base charge.

I found the trimmer on the 555 timing can be twiddled for maximum voltage at D6 anode. It's close to 50kHz on the breadboard. Duty is under 50%, it doesn't seem too fussy about that.

I think that's because you multiplier input caps are small.  Most designs use larger input caps.  The difference in your case is you don't what a to get double the supply you only want 1.5 times or so.  The small input caps help do that.

[anotherjim]

Drawing error, R4 is on the base of Q1 before the diode. Drawing Corrected...



FYI,
Output -14.4v @ 40mA. Series pass transistor costs a junction drop. Could be corrected with an extra diode added to the 15v Zener, but only 5% error I can live with.
Input 12.1V @ 130mA
555 alone takes 13mA with output disconnected.
555 and Transistor booster take 36mA with charge pump disconnected.

[Rob Strand]

Output -14.4v @ 40mA. Series pass transistor costs a junction drop. Could be corrected with an extra diode added to the 15v Zener, but only 5% error I can live with.
Input 12.1V @ 130mA
555 alone takes 13mA with output disconnected.
555 and Transistor booster take 36mA with charge pump disconnected.

Thanks a heap for posting your results.
That's actually pretty good.

With no load the output stage burns up 36-13=23mA.
For no load, my simulation produced about 23.4mA, 
which could be decomposed into:
    Average shoot-through = 5.3mA (no load)
    Average 470R = 10.6mA
    Average 680R = 7.5mA

I thought of a few schemes to knock those down but they make the circuit more complicated.  The thing is even if you halved the losses in the output stage it's only a small percent overall.

When running, the input power is about 1.6W and the output power is 0.58W.
I haven't tried pin-point the main losses but I suspect it's from the diodes.

[Rob Strand]

I had a poke around in the old Motorola TMOS data and found this.

FWIW,  the other day I looked at the performance of this circuit.    The biggest
downfall is the turn-on time of the top MOSFET.  It is quite slow when the gate
resistor is something reasonable like 1k to 10k.   That's certainly going to limit
the upper switching frequency.   Lower gate resistor values are better but then
you are wasting quite a bit of power when the top MOSFET is turned off, since
the gate resistor then sits across the supply rail.   It dissipates power without
contributing to much.  The whole idea of using a MOSFET is avoid those type
wasted drive currents.

[diffeq]

anotherjim, Nice to see the circuit working up to spec! Will we see a drum machine demo when it's up and running? 

Can control pin 5 of the 555 theoretically be used to reduce shoot-through by varying the duty cycle?

Can't be done.  For "proper" avoidance of feed-through you actually need gaps in the switching between when the bottom transistor turns off and the top turns on (and the same idea on the other transition).  In power electronics the gap is call the "dead time".   Simple circuits use diodes + resistors + caps and CMOS Schmitt-trigger inverters.  Modern stuff uses a microprocessor and the timing is done in software.

Thanks for explanation. So if we consider these transistors as switches, they must have "break before make" action. Reminds me of a flying capacitor topology:

Makes one wonder if this bottom inverter propagation delay is useful.

[Rob Strand]

So if we consider these transistors as switches, they must have "break before make" action. Reminds me of a flying capacitor topology:

There's two mechanisms behind shoot through.  One is the overlap under DC conditions.  If you reduce the overlap you reduce the shoot through.  That's the idea in my early post showing the overlap and shoot-through current.  The other mechanism is a dynamic one.  When the input signal to a switch changes state the switch device does not change state immediately.   From outside both MOSFETS and BJTs show this behaviour but the cause is different.

A lot of small stuff only bothers with with DC overlap.  They just put up with the shoot-through based on the idea that the devices will limit the shoot-through current and the fact the shoot-though doesn't occur for very long.   They might try to reduce the shoot-through by making sure the timing for the on and off parts are symmetrical.  A lot of logic gates (eg 74LSxx and CMOS 4000) use this idea.  As someone mentioned before the 555 actually has quite a bit of shoot through.  It's there but it's under control and not going to blow something up.   

When you get to motors and larger stuff the turn-off time for the devices gets much longer and the devices have low impedance.  In this case you really can't let it go  you have to actively avoid the problem with delays and non-overlapping switching.

Here's a paper on how these guys dealt with shoot-through on a voltage doubler device,
https://www.utdallas.edu/~hxl054000/publications/Journals/SC-powerstage.pdf

The example you gave probably doesn't specifically have non-overlapping edges. It probably just relies on controlled overlap.  Note also while the block diagram shows an inverter,  real circuits have a whole heap of junk in there.  Like there may be a buffer which is not shown that has a delay which matches the inverter.  You just don't know.

[anotherjim]

Drum machine demo? Sure...
"Bum-tickaticka-tock, Bumtickaticka-tock...."

Probably will post it up when done, but over in the Lounge.

One scary element of this design is that it ought never be powered up with the 555 absent. Without its grip on the BJT base drive the pair will both turn on and sit there getting very hot. I usually socket IC's, but it might be wise in this case to solder it in.

The ICL7660 looked interesting, but can't quite produce the goods in this case. They can be parallel and/or series ganged, but that's still more parts.
One thing that I do find interesting about the 7660 is that it could replace the 555 as the oscillator. See the applications in AN-051 (fig.23). An antiphase clock is also available but only from an N-channel drain, so pull-up is weak. I think maybe that could drive the PNP independently giving an opportunity to separate the base currents and reduce shoot through.

[duck_arse]

it's probably too late for me to have thought of it by now, or it's outside the scope of "parts in the bin" [although I have one in mine], but what about something like the IR2155? hi and low side drivers, and with dead-time inbuilt, might be a thing.