Can we talk about Charge Pumps? Design.

[italianguy63]

I posted another thread the other day, and the comment was "Why are you using a MAX1044 (they suck)?"  Well, the short answer was because "it worked."

There were some other questions about some other board traces, and the rest of the story there is I built the circuit around a TC962, and it "didn't work."  It whined.  I tried several things that did not help.

There are proponents of the LT1054 here, they say it works.  But, in looking at the post archives, Govmnt Lacky had the same whining problems, it it does not appear he was able fix it either.

Hence, the purpose of this thread, please.  Looking for a real solution that "works."  Looking for a voltage inverter to go from +9V to -9V in a quiet fashion.  Yes, the MAX1044 works (I'm using the MAX1044CPA).  But, they are deemed unreliable and fragile because they are working on the edge of the designed voltage rating.  I would like to use a more robust LT1054.

There has to be several of you out here that use charge pumps sucessfully.  I am hoping someone will come off the "iggy" and give some real design values of working circuits so we can all stop chasing this problem.

Here is my stab at it and notes:



D1 is just reverse polarity protection.  D2 (D3) is reverse voltage protection.  I have already witnessed it, and reverse voltage is reported to kill charge pumps.  So, for a few cents, and a small voltage drop, so what?

C1 and C5 are filter caps.  Trying to keep overall capacitance down.  In my current MAX1044 circuit, I have used 100uf (worked), 47uf (worked), and have yet to try 22uf, but expect it will work.

L1 and L2, for filtering, and I assume the small resistance they create make C1 and C5 more effective.

C2 and C4 are the charge pump "buckets."  I usually see these as 10uf on most charge pump data sheets.  But, the LT1054 spec sheet specifically calls for 100uf.  Should be fine, they should just both be the same size.

C3 is also on the LT1054 spec sheet.  Its purpose is to boost the efficiency and the frequency of operation.  The key to charge pumps it seems is to get the frequency high, so you don't hear the dreaded whine.

The BOOST solder jumper is for the MAX1044 option.  Allows the frequency boost to be used with that chip.

That pretty much sums up my "suggestion."  If anyone has a working (quiet) model with the LT1054, I would appreciate if you would share your actual component values or schematic.  I know it looks messy, and a lot of components—but we are only talking about 50 cents in parts here excluding the charge pump.  I have ordered some LT1054's to try, but they have not arrived yet.

Thanks again!  Mark

[italianguy63]

(deleted - error fixed on schematic) MC

[duck_arse]

has D2 gone invisible, or changed its name? what type inductors are you planning? 10uH seems v. small a value.

[italianguy63]

Sorry it is labeled D3.  (changed name).  

Yes, small value.  They are the cheap resistor looking inductors.  I believe the purpose is to introduce a small amount of resistance too.. They are like 10 ohm resistance as well.  See Tayda #A-448 or Mouser #434-MICC/N-100J-01.

MC

[Govmnt_Lacky]

Since I have been named in this thread, I guess I better speak up....

I believe you are referring to my Gilmour build where I tried to power the flanger with an LT1054 used as a voltage doubler. You are correct in the fact that I abandoned it and went with a transformer to power the circuit. I have SUCCESSFULLY used the LT1054 as a doubler in several circuits before using the Road Rage circuit from Madbean 

I believe there is a LOT of information from PRR, RG, etc. regarding using the charge pump circuits and the "sewer ground" effect and so much more.

It looks like you have done a bit of research so...... good on you and I hope you find what you are looking for! 

[italianguy63]

Thanks!  I am the persistant type, and will eventually figure it out.  I look at forums like this to help you check your logic with people that have "been there, done that!"  I don't WANT to have to invent the wheel.  Again, hoping somebody will share successful builds.  I don't have a LT1054 in my hands yet.  I could just plug it in and have it work; hence; all this could be moot.  I am thinking aloud mostly. 

[R.G.]

There has to be several of you out here that use charge pumps sucessfully.  I am hoping someone will come off the "iggy" and give some real design values of working circuits so we can all stop chasing this problem.

There is not a magic bullet, at least not a single one. The real answer is to understand the electronics underneath it all and be able to apply it.  So we'll all stop chasing this problem when there is one problem, and that one problem amenable to a single solution.

The MAX1044 is a great chip, works fine for audio if you boost the frequency above audio. It is too near its working voltage limit at 9+ volts in a pedal. However, I designed a MAX1044 into a 9V-powered pedal that was produced for some years without ever running into the voltage limit problem. I spent a bit of overkill on making sure it was quiet, and those steps also prevented overvoltage issues as a side effect, proving once again that it's better to be lucky than smart.

The problem I was solving was noise - what would be whine at audio, but I was trying to solve it for above-audio too. So I took great care with trace routing on both the supply and ground sides, great care with local decoupling, and also worried about damping parasitic resonances in the supply/ground traces. That fixed the issues with power supply voltage, accidentally. It was only a couple of years later that I found out about the power supply sensitivity when debugging field failures of another company's designs. They had a quiet result too, without my elaborate precautions, but were getting sudden-death syndrome from voltage.

I designed switching power supplies for a living once upon a time, and the issues with switchers in general are that they pull sudden, high-magnitude current spikes through the wires leading to the power parts. The edges of those current spikes are as sharp as the designer can make them, because that's where the losses come from. Sharp edges mean high frequency content, and this is what causes whine, as well as electromagnetic interference (EMI) for the higher harmonics. The physical size, shape, and relationships of the traces make differences in what EMI comes out.

The issue is often one of layout, because wires are not short circuits. All conductors are really low-value resistors. If you pull sudden pulses of current through a wire or circuit trace, it causes corresponding voltages across the length of that conductor by Ohm's Law. If that voltage causes a perturbation in your power or ground in the circuit, you get whine, full stop. If you do your layout/wiring with this knowledge in hand (and are lucky: Mother Nature is not only pig-headed, She's perverse), you can minimize the ill effects in to inconsequence.

Here is my stab at it and notes:
D1 is just reverse polarity protection.  D2 (D3) is reverse voltage protection.  I have already witnessed it, and reverse voltage is reported to kill charge pumps.  So, for a few cents, and a small voltage drop, so what?

Again, not one solution. Some designs need the diode's voltage drop not to be there. Some don't care. The devil is in the details. I have not run into the reverse voltage issue in my designs (yet).

C1 and C5 are filter caps.  Trying to keep overall capacitance down.  In my current MAX1044 circuit, I have used 100uf (worked), 47uf (worked), and have yet to try 22uf, but expect it will work.

Here some understanding needs applied.  Why are you trying to keep overall capacitance down? Cost? Space/volume? In a switching power supply, which this is, albeit a small one, the decoupling before and after the power supply are CRITICAL. These caps are best viewed as local buckets of charge from which the actual switching takes local bites. They are there to keep these bites from being sucked through the distributed L-C of the power supply lines to/from the switcher. In bigger switchers, even the ESR and ESL inside the caps makes a difference in noise.

What really matters is the impedance right at the pins of the chip (i.e. power switching parts) versus the impedance out at the remote power supply and circuits supplied. The trick in suppressing noise is to keep the impedance right at the chip as low as possible while deliberately mis-matching this to the remote power supply/load so that noise power is transferred very poorly. A quick rule of thumb is to put a big electrolytic, perhaps a smaller electrolytic, a large MLCC ceramic and a small MLCC in parallel at the input side, leaving out some of these to match the circumstances. These form a low impedance local "bucket" of charge so the switcher can get its current needs met locally without having to suck them through the distributed L-C of the power supply lines.

The LT1054 datasheet says:

An input supply bypass capacitor will supply part of the peak input current drawn by the LT1054 and average out the current drawn from the supply. A minimum input supply bypass capacitor of 2μF, preferably tantalum or some other low ESR type is recommended. A larger capacitor may be desirable in some cases, for example, when the actual input supply is  connected to the LT1054 through long leads, or when the pulse current drawn by the LT1054 might affect other circuitry through supply coupling

... which is in accord with the general principles. You're not limited to 2uF - that's a minimum. If you have space and money, making this bigger and lower ESR further cuts noise.

L1 and L2, for filtering, and I assume the small resistance they create make C1 and C5 more effective.

It gets tricky here. Yes, L1 and L2 are for filtering, to make a Pi filter with the caps both before and after. It's part of that deliberate mismatching of the local power needs to the impedance of the lines back to power and out to load. But you need to watch the resistance carefully. It both damps any ringing of the L and Cs and causes unexpected voltage sag if it's too high. L1 and L2 are also expensive, and sometimes if you do good layouts, you can dispense with them. I included positions for them in the "Neutron" auto filter, but suggested they be replaced with a jumper if possible. Small L's are high resistance. In nearly all cases, the resistance of an inductor is as or more important than its actual inductance.

C3 is also on the LT1054 spec sheet.  Its purpose is to boost the efficiency and the frequency of operation.  The key to charge pumps it seems is to get the frequency high, so you don't hear the dreaded whine.

I don't thing the value of C3 affects the frequency much. But it does have an effect on efficiency. It needs to be as big and low-impedance as needed.

The key to charge pumps is double-sided: get the frequency above audio to prevent the basic switching frequency residuals in the output from being heard as audio, and to get the frequency as high as possible to make the filtering components small. Well, actually, it's three-sided: the layout has to prevent the switched current pulses from flowing in conductor/resistors that cause changes in the audio output. Switching frequency residuals even above audio can be cross-modulated down into audio by the audio signal itself.

Did I mention that Mother Nature is both pig-headed and perverse?   

[italianguy63]

Thank you Mr. Keen for the reply.  I read it a few times and got about 80% of it.  BTW, your Germ testing article has been very helpful to me.  I am having a great time wallowing around in FuzzLand.

I am not an EE, just a layman.  Your brain is bigger than mine on this obviously.  But, that is why I was posting the question in the first place anyway!  I will keep absorbing knowlege as I go, but this is an advocation, and not my job... if you know what I mean.  I promise to try to keep up though.

Maybe if we make some assumptions it can help.  I will fire off some thoughts again... I don't mind that you quote me (even my misinformation at times)-- as long as you understand my simplified understanding of this.  I am in "Sportsman" class, and not "Formula One."  At least I'm not walking or crawling.

Some more random thoughts that might help:

Assume this circuit is being used to power some flavor of Fuzz.

I put the diode in after the circuit because I did witness voltage backflowing into the charge pump.  I also read that could be damaging to allow pin 5 to go positive.  So, just call it insurance-- I think it is a good idea to leave it there.  The slight voltage drop does not really affect the FF.

L1 and L2 are not expensive.  Only $0.08 each.  From the data sheet, 1.7 ohms each.

No particular reason to limit capacitence other than believing "more isn't necessarily better."  No real constraints on size or cost.  Really no difference in component costs-- pennies.

With the MAX1044 my circuits are silent.  With the LC962 I was getting background whine at high gain.  It worked pretty well, but not well enough.  The MAX is nice, but I am concerned with durability, expecially with the use of wall-worts.  So, I am hoping to develop this LT1054 solution.  I wish I had some in my hand I could test.  I have done my best to keep my leads short, shielding good, etc.  Again, whine with the 932, and not with the 1044.  Tells me the problem is (probably) chip related, and not the surrounding circuitry.  But, I do understand your warning.

I am seeing right at a 0.3V drop across the pump using the inductors and the diode.  Not a problem.

Can you determine what the optimum cap and inductor values might be?  Again, I was hoping others might post schematics of solutions they have used that do work, and are quiet.

Thanks again!

MC

[R.G.]

I am not an EE, just a layman.  Your brain is bigger than mine on this obviously.  But, that is why I was posting the question in the first place anyway!  I will keep absorbing knowlege as I go, but this is an advocation, and not my job... if you know what I mean.  I promise to try to keep up though.

That's the right attitude. The only way to get knowledge firmly attached to you is to keep rubbing it on you. 

Maybe if we make some assumptions it can help.  I will fire off some thoughts again... I don't mind that you quote me (even my misinformation at times)-- as long as you understand my simplified understanding of this.  I am in "Sportsman" class, and not "Formula One."  At least I'm not walking or crawling.

I'm with you.

Assume this circuit is being used to power some flavor of Fuzz.
I put the diode in after the circuit because I did witness voltage backflowing into the charge pump.  I also read that could be damaging to allow pin 5 to go positive.  So, just call it insurance-- I think it is a good idea to leave it there.  The slight voltage drop does not really affect the FF.

For this one, I'd probably go with using the Schottky diode as a shunt protector, not series. That should clamp pin 5 to no more than half a silicon diode drop above ground and should be enough to protect the chip without the series diode loss. The series connection is more certain, but does lose you some voltage. Whether this matters or not depends on the circuit that's attached, so there are at least two desirable solutions.

No particular reason to limit capacitence other than believing "more isn't necessarily better."  No real constraints on size or cost.  Really no difference in component costs-- pennies.

Sometimes less is more, sometimes only more is more. Depends. At low loads, 2uF is probably enough. At higher loads, the current pulses get bigger, so more is more.

Here's a way of thinking about that circuit. When the circuit switches the bucket cap to be across the output cap, the output cap is placed in parallel with the bucket cap with only the impedance of the switch devices between them. So the capacitors do I=Cdv/dt where dv is the difference in voltage that the output cap has run down since the last charge cycle, and dt is the period of the oscillator. Well, really probably the half-period, as with both caps in parallel, for half the oscillator cycle, both caps supply the output current. So both caps are run down a bit. When the oscillator flips the bucket cap back to being in parallel to the input cap across the power supply, it is now a lower voltage and starts sucking current from the input cap.

The series L to the remote power supply keeps C3 from charging instantly from the remote supply, so it's on its own for a few microseconds. C2 is 50 times as big as C3, so the charge it sucks out of C3 to charge itself reduces the voltage on C3 by 50 times as much as C2's voltage rises, at least until the inductor can charge from C1 and the remote power supply. So C3 sags.

This may be desirable for some issues of noise out at the remote power supply, but it may not let C2 charge fast enough. I would not make C3 any smaller than C2 at all. I'd be tempted to make it bigger. This will reduce the noise going back to the remote power supply - if that matters. And critically damping C1/L1/C3 may or may not be desirable or possible, but if things happen to get wrong, it can bite certain portions of your anatomy, causing an irremovable tone if this resonance happens to hit the main oscillator frequency.

With the MAX1044 my circuits are silent.  With the LC962 I was getting background whine at high gain.

I looked at the TC962 datasheet. That's where you got grounding the LV pin - it doubles the oscillator frequency to 24kHz, which should have helped with whine if you did it. The MAX1044 wants grounding the LV pin for supply voltages below 3.5V - so the TC962 wants the opposite condition for the LV pin from the TC962.

Can you determine what the optimum cap and inductor values might be?  Again, I was hoping others might post schematics of solutions they have used that do work, and are quiet.

If it were me, I would try to ditch the inductors, replacing them with wires. However, notice that I left positions for them in the Neutron. I'm not sure there is an optimum - many values will work fine. I would tend to make the local decoupling better - making c3 at least 100uF, and I'd definitely put a 0.1uF MLCC cap across it as well. I'd put the same 0.1uF MLCC across the output cap. If you have to use inductors, so be it, but I got away without them in a relatively critical noise application.

Here's a Big Deal. You're using ground symbols on the theory that all grounds are the same. They're not. The (-) end of C3 should go directly to pin 3, as should the (-) end of C4. This makes the resistance between these points as small as possible, and prevents as much as possible the generation of switching current spikes. A wire should go from C1(-) to the pin-3 clot, and another wire from C5(-) to the pin-3 clot. I would try putting D3 with cathode to ground, anode to V-.

[italianguy63]

OK you got me.  What's a CLOT?  The DC in jack?  And when you say (-) side of C4 and C5, I assume you mean the ground side and NOT the -9V out (pin 5) of the IC?

The MLCC would be in parallel? with C3 and C4?  I assume they are faster and would clean up ripple?

Yes, I did have LV pin grounded with the 962.  It whined JUST ENOUGH to be an annoyance.  I guess you could live with it.  But, I wanted better.

Agreed on the Diode to ground.  I can certainly see that.  Thanks for staying with me here.  Again, thank you for taking the time and being so helpful.  I know your time is valuable... I know I feel mine is.

MC

[R.G.]

OK you got me.  What's a CLOT? 

Gelatizing lump of stuff that used to be unconnected. Blood clots. The result is one example of a clot. There are others. 

The DC in jack?

Yes.
 

And when you say (-) side of C4 and C5, I assume you mean the ground side and NOT the -9V out (pin 5) of the IC?

Sorry - had an OFBF. Yes, the ground side.

The MLCC would be in parallel? with C3 and C4?  I assume they are faster and would clean up ripple?

Yes. MLCC = Multilayer Ceramic Capacitor. Good high frequency characteristics when the electrolytics' internal construction makes them stop being caps and begin being inductors at some high frequency. Much better RF performance.

You're welcome.

[italianguy63]

Thanks again.. I'll report back when I have some chippy's to try.. They are on the slow boat from China now.

MC

[Morocotopo]

Just a little info: I have used the ICL7660S many times in high gain circuits and no noise problems at all, always tried to follow correct ground practice as per R.G.´s periodic explanations on the subject since the dawn of times. Think of it as a more robust 1044. Also used the 1054 for higher current circuits. No noise either.

[Govmnt_Lacky]

Just out of curiosity...

Is there a "definitive" posting here or elsewhere on the proper usage of charge pump circuits?

Im off to Geofx.. 

[italianguy63]

I think our conversation has us here now:     

Testing to follow in the future!  MC

[alanp]

Just to clarify -- I was told once that using inductors when any alternative is available is not best design, most of the time?

[R.G.]

Just to clarify -- I was told once that using inductors when any alternative is available is not best design, most of the time?

As a statement of practicalities, that's not far wrong. Inductors are less "perfect" than capacitors because the conductors used to make them are less notionally perfect than the insulators used to make capacitors, and because the magnetic field container (the cores, whether air, iron, or ferrite) are less perfect containers of M-field than the insulators of capacitors, not to mention being heavier. And cored inductors have limits on the field they can maintain internally due to saturation effects.

This means that inductors are heavier and less perfect for containing the same amount of energy in a field than a capacitor. They are almost always bigger, heavier, less standardized and more expensive.

So there is a big premium to finding ways to design the same function without them. As with any question containing the word "best" you have to say how you're measuring "better" and "best", and one should always be suspicious of anything saying "best design", but things are cheaper, smaller, and lighter if you can design away the inductors.

But, as Albert Einstein was reputed to say, everything should be as simple as possible - but no simpler. There are times when inductors are simply needed.

[jubal81]

I noticed you listed using a 1044CPA. I never had good results with those, but the 1044SCPA has always been quiet for me.

[ggedamed]

I redrew the last schematic to help my brain cope with the whole situation.



I'm thinking that C1 and C3 can be merged. C4 and C5 too.
Q: Plain ceramic caps can be used with good results, right?

Adding R.G.'s PNP transistor power switching trick to these, what I could understand led me to this (TO_R goes to the R pin from a TRS input jack):



What do you think?
Then I said, why not try my hand on the PCB:



What do you think about the layout?
Q: Is there a significant advantage to soldering the 100n directly on the IC pins as opposed to the above layout?


OFF-TOPIC: R.G., I would love to let me help you redesign your site for free. I'm simply not able to find anything on it without resorting to Google.

[italianguy63]

Nice work!  But it does bring up some questions...

Good question about combining C1/C3 an C4/C5.  In the original submission, that is why I used the inductors to seperate them (so they would act indepentantly as filters).  Have we lost that?

R.G. indicated two seperate grounds is desirable.  So, having 2 different grounding pads would be better?

I've seen his transistor switching trick before, and considered it.  Since I was able to clean up my current circuit without it-- I determined it wasn't needed.  Maybe another point to discuss?  Easy enough to add it.

I didn't post it, but add another solder pad from pin 6 to ground.  That gives the LV option for other chips like the TC962.  Loose the naming of the specific chip, and just call it IC1.  Can make it generic/flexible.  Also a mounting hole would be helpful.



EDIT:  (Both solder pads could live under the IC, for space).

Looks like we are cooking with bacon.  MC