Charge pump NE555 vs LM386!

[R.G.]

OK, I can see that if the not-well-performing part is that the LM386 in question has crossover distortion and you can get another if you need it.

On the need-another-half-a-volt item: I believe you're on the right track, Rob. I dimly remember from the stuff I read in the older charge pump datasheets from when dinosaurs ruled that you can model the output impedance of a charge pump as a switched-capacitor resistor. The idea being that even with perfect switches charging and discharging the bucket capacitor, the current is limited to the charge that can be transferred by that capacitance filled and dumped at that frequency. Drat. Now I have to go look up my stuff on switched-capacitor resistors!
Anyway, the datasheet (which I remember as the 7660) said that the output impedance approximated an 80 ohm resistor. That was for a fixed-frequency self-self oscillator with (IIRC) 10uF caps. In newer charge pumps, the frequency is up-able, and the switches may or may not saturate to a lower resistance (for MOS devices). 
The issue of switch performance depends on frequency too. When you turn on the switches to charge the bucket cap, the top and bottom switch transistors have some voltage loss depending on their saturated resistance. The bucket cap charges at a rate determined by 2x the switch resistance and the cap value, so its a declining exponential approach to the charge voltage. At slow frequencies, the switch resistance matters less because there is longer to charge, so the bucket cap charges more fully. But at slow frequencies, the switched-capacitor equivalent resistance delivers less charge to the output cap, so the switched-capcitance resistance goes up.
From all this mess, you want to switch the bucket cap with the lowest possible switch devices, as fast as you possibly can, with the biggest cap. Give the competing effects of frequency, capacitance, and resistance of the switches, there's likely to be varying results.

Dag nabbit!! Once again, details matter! 

[Rob Strand]

Anyway, the datasheet (which I remember as the 7660) said that the output impedance approximated an 80 ohm resistor. That was for a fixed-frequency self-self oscillator with (IIRC) 10uF caps. In newer charge pumps, the frequency is up-able, and the switches may or may not saturate to a lower resistance (for MOS devices).

That's what I remember as well.   If you dig through the datasheets and app notes for *all* the chip variants somewhere in there is some plots with different caps, and in some cases different frequencies.   (I've got a big pdf stash of those.)

The issue of switch performance depends on frequency too. When you turn on the switches to charge the bucket cap, the top and bottom switch transistors have some voltage loss depending on their saturated resistance. The bucket cap charges at a rate determined by 2x the switch resistance and the cap value, so its a declining exponential approach to the charge voltage. At slow frequencies, the switch resistance matters less because there is longer to charge, so the bucket cap charges more fully. But at slow frequencies, the switched-capacitor equivalent resistance delivers less charge to the output cap, so the switched-capcitance resistance goes up.
From all this mess, you want to switch the bucket cap with the lowest possible switch devices, as fast as you possibly can, with the biggest cap. Give the competing effects of frequency, capacitance, and resistance of the switches, there's likely to be varying results.

Dag nabbit!! Once again, details matter!

I remember a document from Analog Devices which gave some quite simple formulas for the charge pumps with MOSFET switches.   The output resistance of the converter was the sum a set of different factors.  One was 4 or 8 times the on-resistance of the first switch, then another term for the second switch and one for the cap-size+frequency.

Back in the days before low ESR caps the ESR of normal Al-Electro's had a big impact on performance over about 10kHz.   For some of the converters I used Tantalums to run at higher frequencies.  These days the low ESR caps kind of make that mute.

Another quirk of the roll-your-own stuff is the use of bipolar output stages and diodes.  These all introduce largely fixed voltage drops  whereas the MOSFET switches are have more load dependent drops.

But yes, many details are required to get the estimates to match the built unit.   As far a tinkering goes bigger caps, bigger diodes and low ESR caps all help.  The circuit might grow in size scraping that extra performance  .

[11-90-an]

I'm still trying to understand charge pumps a bit so please bear with my questions...

So i am quite interested making a inverter for my +9v without a charge pump ic...
After reading the ne555 vs lm386 charge pump article, i am quite confused as to the use of A and B in the lm386 schem like in here...



When you scroll up, it says that they are for "stopping the oscillators with elements having open collectors or open drains". Does that mean that there would be a transistor somewhere?

Also, is this circuit enough to compensate for a charge pump IC? I want to build jonny.reckless' ET redux, but the charge pump used is about the same price as the 13700 in tayda...

Also, do the diodes have to be schottky? Would there be a different way of positioning the diodes that i can replace rhem with 1n4148s?

[Rob Strand]

After reading the ne555 vs lm386 charge pump article, i am quite confused as to the use of A and B in the lm386 schem like in here...

When you scroll up, it says that they are for "stopping the oscillators with elements having open collectors or open drains". Does that mean that there would be a transistor somewhere?

Just ignore all of that stuff.  You don't need to do anything more than the circuit that are there.

The oscillator on the NE555 can be stopped, so all they are saying is the LM386 can also be stopped.    It's not a very useful function.  A lot of modern regulators can be turned off and take tiny amount of power in the idle state.  You do that stuff to save power.   Both the NE555 and LM386 don't take tiny amounts of current when stopped.  Also, the output isn't switched off because of the diodes.   It all seem a bit pointless.

Also, is this circuit enough to compensate for a charge pump IC? I want to build jonny.reckless' ET redux, but the charge pump used is about the same price as the 13700 in tayda...

These things don't compete with the good charge pumps but the it's highly likely the ET redux doesn't need to max-out the power requirements of the charge pump.  So as far as sub-ing an NE555 ckt or LM386 ckt  you shouldn't have any problems.  All you need to do is set the frequency.

Also, do the diodes have to be schottky? Would there be a different way of positioning the diodes that i can replace rhem with 1n4148s?

The Schottky's  will give you a higher output voltage.    If you are willing to chuck away a volt or two at the output you should be able to get the 1N4148's working.      Technically you shouldn't use 1N4004s especially at high clock frequencies, however, I think some people tried them and there wasn't a big difference.

[11-90-an]

Forgive me for asking but, how can I set the frequency? Does it have to do with some resistors and capacitors in the given schem?

[Rob Strand]

Forgive me for asking but, how can I set the frequency? Does it have to do with some resistors and capacitors in the given schem?

Yes.   

25kHz would be a good start.   You probably don't want to go any lower than that because you could get into noise problems.  The best frequency depends on the caps.

For the LM386 the formulas were discussed earlier on in the thread.


For the NE555,

https://ohmslawcalculator.com/555-astable-calculator



I don't recommend using R1 less than 2.2k and 10k would be better.  You see a lot of schematics will low values like 1k and 900ohms.  There's no need for such low R1 values, it just burns up power and it can make cause a small amount of lost performance.

[11-90-an]

So if I incorporate the lm386 charge pump circuit to the ET redux, it should end up like this... right?



I used 2.2nF since i don't have any 18nF caps... I followed this from the AMZ design... is there a differnce in why the voltage DOUBLER cap value is 18nF and why the voltage INVERTER is 2.2nF, yet have supposedly the same frequency? (The ones in the article)

[Rob Strand]

I used 2.2nF since i don't have any 18nF caps... I followed this from the AMZ design... is there a differnce in why the voltage DOUBLER cap value is 18nF and why the voltage INVERTER is 2.2nF, yet have supposedly the same feequency? (The ones in the article)

It should be 18nF.   The 2n2 value is a copy and paste error in the article; see reply#7 in this thread.   If you have 10n or 12n or 15n it's probably OK.  The closer to 18n the better, well at least to start with.   For a one-off build you can use parallel caps.

[11-90-an]

But what about your measurements calculation thing? It said there that the AMZ version said 2.2n....




Unless that too, was an error...

With 2.2n, your calculations said that the ckt had a frequency of around 126kHz...

[Rob Strand]

But what about your measurements calculation thing? It said there that the AMZ version said 2.2n....

With 2.2n, your calculations said that the ckt had a frequency of around 126kHz.

It's a bit messy to explain.

Jack built the unit according to the article with 2n2 but he found the frequency was really high, 126kHz (later more accurately given as 103kHz or something).    The article said 25kHz.   So that brought up the question why the big difference.   Then I think Jack noticed there were two schematics in the article one with 2n2 and one with 18n.   So that's the point where it was clear the article had a typo in that they copied the 2n2 cap value from the NE555 circuit.     So the bottom line is the cap should be 18n.

The second issue that came up was how to calculate the frequency for the LM386.   That's when I did the spread sheet.   The aim of my calculations was to show that the 18n cap got near the 25kHz in the article and also matched the high frequency Jack saw when he used 2n2.    That confirmed the 18n caps is "more" likely.

The third issue was: there are oscillator circuits using the LM386 which give frequencies and formulas but there  seemed inconsistencies between those.   One reason for that is some circuits have 30k resistors and others have 10k.

So after all that, the 2n2 definitely looks wrong.   The 18n still doesn't seem to match the article I calculated 15.5kHz and the article implies 25kHz.   I guess the only way to find out would be to build it and check the frequency.  In all honesty I'm expecting the 18n to produce 15.5kHz.    So it would be best to start out with a smaller cap than 18n to raise the frequency to 25kHz.     That would be  C  = (15.5kHz/25kHz) * 18n  = 11.2n,  so a 10n would be a good start.

So to check that with the formula,

f  = 1 / (0.36 * Ct * Rt )  = 1 / (0.36 * 11.2n * 10k)  =  24.8kHz

and if you use the closest but lower value of 10n (so the frequency is higher not lower than 25kHz),

f = 1 / (0.36 * 10n * 10k)  = 27.8kHz

[For reference that formula came from the PDF attached to Reply #7.  Rt is the timing *resistor* value.   The formula automatically takes into account the 50k's inside the LM386; no need to account for Thevenin equivalent impedance driving the cap etc.    It also matches the formula in the extract Jack posted in Reply #11.]

[11-90-an]

So is there any "preffered" frequency in charge pumps? Does higher frequency give more ripple? I know that the frequency should, in practice, not go below 20khz to prevent some buzzing in psu... (But there are ways if doing this, but requires bigger brains... )

[Rob Strand]

So is there any "preffered" frequency in charge pumps? Does higher frequency give more ripple? I know that the frequency should, in practice, not go below 20khz to prevent some buzzing in psu... (But there are ways if doing this, but requires bigger brains... )

I'd say 25kHz to 50kHz would be a good range.  25kHz gives a reasonably safety margin to be sure no whistling or buzzing occurs.   The "best" frequency depends on the caps.   Generic non Low-ESR caps probably should be run the low side of that frequency range.    Low ESR caps, tantalums can be pushed up higher.   You generally get lower ripple at higher frequencies but if the cap ESR is high you can get higher ripple.