of variable frequency
Can ferrite beads do the trick,
Are the ICL7660 or MAX232 really out of your budget?
Also, please share a bit more of your circuit (schematic and real life photos if possible) so we can help you better.
I designed an audio circuit containing a charge pump generating up to -10V for driving an optional circuit which hardly draws over 1µA when properly used.
I've had success using RLC filters on both the input and output of a 7660, but they're designed to invert an input voltage so that the output is roughly -1*VinWhat's the point of the 7660 if you place an inductor to filter it? Might as well use a regular inductor based DC-DC at that point.
Maybe you'd have better luck with a different topology rather than a switched capacitor. Maybe some LLC switchmode coverters, like Cuk converters?
The square wave frequency determines the output voltage
Charge pumps are usually more efficient compared to buck boost topologies, which is useful when creating any battery powered audio device.You have it kind of backwards. Inductor based converters are very efficient across a wider frequency range, and wider current range. Charge pumps are only efficient when they are operated in the audio range, and that's not very good. Not only it's harder to filter them, with larger value components required, they are also more noisy in general.
As mentioned in this case, the charge pump won't create a regulated output without additional circuitry (there's examples in the app notes for the 7660), and it can't create a voltage greater in magnitude than -Vin. At least not without cascading multiple of them to make voltage doublers etc, which I think is a bit of a waste of space/parts at that point.
From your first post it was not clear that your frequency varies intentionally. It looked that you just can't get it being stable.
You have it kind of backwards. Inductor based converters are very efficient across a wider frequency range, and wider current range. Charge pumps are only efficient when they are operated in the audio range, and that's not very good. Not only it's harder to filter them, with larger value components required, they are also more noisy in general.
I would second PCB.Wiz's suggestion of using a photovoltaic isolator to generate the negative voltage, as it requires no switching circuits at all.If you have a noisy supply for your PV, you get a noisy supply for your isolated supply. So I'm somewhat perplexed what it would actually solve. Plus the efficiency would be like 1% or something close to that. Negative supply is a solved engineering problem. You use a switched capacitor for low effort. Negative output DC-DC or a buck converter with grounded output for higher currents. If you need low noise, then you post regulate, with a regulator that has meaningful PSRR in the switching frequency and the required harmonics.
As I understand your current circuit doesn't fulfill the design requirements. I don't advocate throwing out entire circuits, but if it doesn't work well, there is only so many things you can do. I would try to hack in a negative LDO if you really don't want to change too much.You have it kind of backwards. Inductor based converters are very efficient across a wider frequency range, and wider current range. Charge pumps are only efficient when they are operated in the audio range, and that's not very good. Not only it's harder to filter them, with larger value components required, they are also more noisy in general.
I'd be happy to put a better circuit on the existing pcb as long as it doesn't cost significantly more, but space is quite limited. Unless something can fit in the layout I posted above.
If you have a noisy supply for your PV, you get a noisy supply for your isolated supply. So I'm somewhat perplexed what it would actually solve.I'm operating under the assumption that the charge pump itself is the source of the audible noise (not sure this has been verified though). In that case, removing the switching circuitry eliminates the OP's problem.
Plus the efficiency would be like 1% or something close to that.AFAIK efficiency is not a concern of the OP.
Negative supply is a solved engineering problem.Agreed, and there are many different solutions. It's likely that the charge pump could be modified to resolve the issue, but it's not clear how the noise is contaminating the audio circuit (if at all). There's no guarantee that adding an LDO would help either. One way around this is to not use a switching DCDC at all.
The charge is a simple chain of BAT54 shottky diodes driven by a +3V square wave of variable frequency using the existing MPU of the initial design.
The charge pump generates a low audible background noise, which I would like to suppress.
IMHO the first thing to try would be to drive the charge pump from a well decoupled single gate buffer, with a resistor (try 10 ohms) feeding it from logic Vcc. It probably should also have a 1K resistor between it and the GPIO driving it, as I expect a fair bit of ground bounce.
Fine suggestion. I would suggest doing some simpler tests first:
1. Disable the charge pump entirely and provide the negative voltage from a low noise DC bench supply. If the noise issue persists, then the charge pump isn't the issue. Otherwise goto 2...
2. Enable the charge pump, but with its output still disconnected from the rest of the circuitry (but give the charge pump a dummy load). If the noise does not come back, then it was likely coupled through conduction on the output, and you could resolve it with a simple RC or LC filter on the output. If the noise does come back, then goto 3...
3. It must be coupling either through conduction in a supply rail (or gnd), or via EM fields (not likely at audio frequencies). You should experiment with other methods of driving the charge pump like what Ian suggested above.
IMHO the first thing to try would be to drive the charge pump from a well decoupled single gate buffer, with a resistor (try 10 ohms) feeding it from logic Vcc. It probably should also have a 1K resistor between it and the GPIO driving it, as I expect a fair bit of ground bounce.
However, the noise is definitely the charge pump frequency.What frequency does the MCU generate ?
Due to size restrictions, I wonder if an equivalent negative charge pump can be made out of inverters rather than all those diodes. I don't have time to experiment this at the moment, but will try to find online examples to save some time.You need some switching elements in a charge pump.
However, the noise is definitely the charge pump frequency.What frequency does the MCU generate ?
You could also soften the edges before they enter the charge pump.
eg A 74LVC1GU04 with input R and feedback C will let you define a slew rate.
Due to size restrictions, I wonder if an equivalent negative charge pump can be made out of inverters rather than all those diodes. I don't have time to experiment this at the moment, but will try to find online examples to save some time.You need some switching elements in a charge pump.
What current do you actually need ?
Then the simplest solution is a Photovoltaic generator as already suggested.QuoteWhat current do you actually need ?Current needed is very low, typically 1µA if the target circuit is used correctly.
I'm unclear what that last comment means ?QuoteWhat frequency does the MCU generate ?Frequency: 200Hz to 20,000Hz roughly.
You could also soften the edges before they enter the charge pump.
eg A 74LVC1GU04 with input R and feedback C will let you define a slew rate.
....
The problem is, I can hardly increase the frequency without making the output voltage increase drastically (it must be within -10V and 0V).
Current needed is very low, typically 1µA if the target circuit is used correctly.Then the simplest solution is a Photovoltaic generator as already suggested.
I'd love to use this option but such devices are rather big and would require significant pcb size increase unless miniature versions exist.Frequency: 200Hz to 20,000Hz roughly.I'm unclear what that last comment means ?
....
The problem is, I can hardly increase the frequency without making the output voltage increase drastically (it must be within -10V and 0V).
You can add a zener to lightly clamp the charge pump, and push the MCU drive source ultrasonic, so you cannot hear the crosstalk.
If you have excessive clamping, you might be able to reduce the number of stages at higher frequency.
If you need to vary the output voltage, you can vary the supply of the 74LVC1GU04 buffer.
The LVC1GU04 has
• Overvoltage tolerant inputs to 5.5 V
• Wide supply voltage range from 1.65 V to 5.5 V
So you could vary the supply up to 5.5V if you have that available, to reduce the pump chain stages.
If you have more voltage available, you could use a SOT23 gate driver like x27517, that has 3V in and up to 20V Vcc, so a single stage can get you -10V, with ~ 11V in.
You can use a linear regulator after the charge pump, this will decrease the noise (ripple rejection) but it'll waste more energy while being used.
Can you divide the supply in half with 2 capacitors to create a negative voltage?
Over what variable range, to what precision and what is the exact load ?
The output voltage is variable, so no regulator at the output..
I use a dual LDO delivering 3.3V and 5V.From what main supply voltage ?
You have it kind of backwards. Inductor based converters are very efficient across a wider frequency range, and wider current range. Charge pumps are only efficient when they are operated in the audio range, and that's not very good. Not only it's harder to filter them, with larger value components required, they are also more noisy in general.
I tried using 1nF capacitors for testing at higher frequencies, but the maximum output voltage was significantly reducedOdd, you should be able to get the same output by just increasing the drive frequency by a factor of ten (BAT54 should be fast enough for ~1MHz). Maybe there's parasitic leakage and capacitance on various nodes of the charge pump bogging it down.
and the frequency range didn't change much.....huh? Isn't switching frequency something you control directly?
Then you've disregarded one of the quickest and cheapest ways to solve the problem of "I want a low noise 0 to -10V variable output and dont have a negative supply"You can use a linear regulator after the charge pump, this will decrease the noise (ripple rejection) but it'll waste more energy while being used.The output voltage is variable, so no regulator at the output.
Can you divide the supply in half with 2 capacitors to create a negative voltage?
This is not a voltage regulator. This is a converter, that halves the input voltage, and probably generates so much ripple on the input and output, that maxim didn't dare to place it into any documentation. Let's just try to compare apples to apples.You have it kind of backwards. Inductor based converters are very efficient across a wider frequency range, and wider current range. Charge pumps are only efficient when they are operated in the audio range, and that's not very good. Not only it's harder to filter them, with larger value components required, they are also more noisy in general.
ADI has some DC-DC converters based on charge pumps that operate in the MHz+ range with 98%+ efficiency (example (https://www.analog.com/media/en/technical-documentation/data-sheets/MAX77932C.pdf)). Charge pumps are universally more efficient than inductor based converters for similar load currents. Their major limitation is relatively low output currents because your switch impedance is the only thing limiting the current flow in/out of the fly capacitor. But there are applications I've designed where a charge pump converter is absolutely the most appropriate solution and achieves excellent efficiency. Inductor based converters always have magnetic losses to contend with, ideal efficiency will always be better with a charge pump but it's hard to get without extremely careful layout due to the very high peak currents involved.
This is not a voltage regulator. This is a converter, that halves the input voltage, and probably generates so much ripple on the input and output, that maxim didn't dare to place it into any documentation. Let's just try to compare apples to apples.
Possible, but getting off into things the OP isn't interested in such as the LTC7820.This is not a voltage regulator. This is a converter, that halves the input voltage, and probably generates so much ripple on the input and output, that maxim didn't dare to place it into any documentation. Let's just try to compare apples to apples.Ok, but the point is it's possible to make an efficient charge pump converter. Though, I suppose unlike an inductor-based converter, you cannot make an infinitely variable converter with excellent efficiency. You can only make some pre-defined fraction efficient. The most efficient converters switch between e.g. 1.33x, 1.5x and 2x output and then pulse modulate between those modes to obtain high efficiency. But switching at audio frequency is definitely not needed.
Quoteand the frequency range didn't change much.....huh? Isn't switching frequency something you control directly?
Did you try and of the tests I proposed (where the charge pump is disabled and/or its output disconnected)?
Could you explain what the negative bias is used for? Does it directly interface to the audio circuitry in some way?
You can also place a series R at input and a cap feedback.
I was a bit busy yesterday but got time to test the 74LVC1GU04 buffer with a 10R at its Vcc and 1K at its output on the breadboard prototype.
That means you probably do not care too much about ripple on the supply.
It's to drive a 6FG6 magic eye vaccuum tube as a percentage level indicator (so not a VU). The charge pump goes to the grid, hence the very low current draw.
Wait, now I'm even more confused. You said that with 1nF the voltage range decreased greatly, but now you say you were still able to get -10V without changing the drive frequency significantly... those two things can't both be true...Quoteand the frequency range didn't change much.....huh? Isn't switching frequency something you control directly?
Sorry I didn't express myself correctly. I meant that the frequency range required to obtain -10V to 0V using 1nF capacitors didn't really change and still covered a good part of the audio spectrum.
I was a bit busy yesterday but got time to test the 74LVC1GU04 buffer with a 10R at its Vcc and 1K at its output on the breadboard prototype. This board doesn't have the background noise issue because the charge pump isn't as close to the audio circuit as on the target pcb, and doen't use tiny smd components. As it can only improve things, I added the buffer to the pcb, which is really tight now.Hard to give feedback on this without some pictures of the breadboard/pcb setups...
I also thought about some possible crosstalk with the Vcc layer underneath and added a ground area under the charge pump. Can only help as well.
My biggest problem is that I can't modify the target pcb while testing.
I'm not a tube expert, but I'm assuming ripple on the grid will cause ripple on the plate/cathode current, which might be the path for this to contaminate the rest of the system. If you disable the tube by disconnecting the plate bias, does the audio noise go away?Could you explain what the negative bias is used for? Does it directly interface to the audio circuitry in some way?
It's to drive a 6FG6 magic eye vaccuum tube as a percentage level indicator (so not a VU). The charge pump goes to the grid, hence the very low current draw.
Fine suggestion. I would suggest doing some simpler tests first:
(...)
2. Enable the charge pump, but with its output still disconnected from the rest of the circuitry (but give the charge pump a dummy load). If the noise does not come back, then it was likely coupled through conduction on the output, and you could resolve it with a simple RC or LC filter on the output. If the noise does come back, then goto 3...
3. It must be coupling either through conduction in a supply rail (or gnd), or via EM fields (not likely at audio frequencies). You should experiment with other methods of driving the charge pump like what Ian suggested above.
I found some info, these are cute devices !
It's to drive a 6FG6 magic eye vaccuum tube as a percentage level indicator (so not a VU). The charge pump goes to the grid, hence the very low current draw.
What MCU are you using?
Does that model have a 8MHz or 32MHz PWM clock ?What MCU are you using?
PIC16F (up to 32MHz).
That means you probably do not care too much about ripple on the supply.
It's to drive a 6FG6 magic eye vaccuum tube as a percentage level indicator (so not a VU). The charge pump goes to the grid, hence the very low current draw.
How critical is it to get to 0 V ? ie what is the control range of the 6fg6 ?
I'd still suggest you get some of these - they are low cost (39c 10+) and claim 12V 8uA
https://www.lcsc.com/product-detail/Transistor-Photovoltaic-Output-Optoisolators_SUPSiC-GAQV1123S_C22385005.html (https://www.lcsc.com/product-detail/Transistor-Photovoltaic-Output-Optoisolators_SUPSiC-GAQV1123S_C22385005.html)
and experiment with linear or PWM drive into RC filter load for gate.
Addit: I find the Vishay Photovoltaic models include nice curves for Io/Vo and applications circuits
https://www.vishay.com/en/product/84639/ (https://www.vishay.com/en/product/84639/)
connect the grid to ground, and place a positive control voltage to the cathode, and the noisy charge pump is non longer needed
or, skip the triode entirely and apply amplified control voltage directly to the control rod of the indicator system
You do not have to feed to the data sheet max, if you look at the 8uA curves, it is a more modest drive.Addit: I find the Vishay Photovoltaic models include nice curves for Io/Vo and applications circuits
https://www.vishay.com/en/product/84639/ (https://www.vishay.com/en/product/84639/)
Too expensive unfortunately: $3-$4 as opposed to the 3 BAT54 arrays which cost $0.05 each. Not that I can't afford the part itself, but production cost must stay reasonable.
Other problem: its 'If' is 50mA. Even if I added a buffer, this would increase the current draw too much as to what the dual 3.3V+5V LDO can provide (80mA for both outputs and 50mA max per output when both are used).
You do not have to feed to the data sheet max, if you look at the 8uA curves, it is a more modest drive.
I linked the (new) Vishay parts because they are the only ones I find that have full curves, a great help to understand how these niche photovoltaic parts work.
You mean the slope difference between the VO1263 (p. 3, Fig. 2, orange curve) and APV1121S (p. 10, first graphic) ?Not really, the Vishay parts have Io/Vo load lines, which few others give. The APV is Vo/If
Indeed the APV seems to raise much faster, which could be a problem.
I ordered a few pieces for the sake of testing. I plan to use two in series, which allows up to -16.4V and a direct LED connection.Toshiba have Spice models, if you want to experiment.
Toshiba have Spice models, if you want to experiment.
The Photovoltaic parts with a box called [control circuit] have an additional active turn off (a modest current sink) to discharge FET gates.
The Toshiba parts with [control circuit] simulate slightly better with a series diode, when used as PWM to DC generators with output RC loads.
If you read back the gate voltage on the MCU via a resistive divider, a second clamp diode to stop positive valve gate would be needed.
If you can rewire the valve as suggested in #58, you then only need to sink cathode current with ~ 20V compliance.
You could even current drive the cathode.
A quick update after testing a pair of APV1121S in series: driving the magic eye works, but the output rapidly raises from 0 to 3V or so when powering the LED between 1.03V and 1.07V (just above the 1V required to light it), causing the magic eye to deflect too quickly. Same behavior with a pulsed output, where the frequency range giving the same effect seems too narrow (I need to experiment a bit further on this).You should current drive the led, not voltage, and experiment with load resistors.
I'm happy to see that such components could be used, the only problem being to find one with an appropriate curve. I will try the Vishay you recommended.
Thanks a lot for taking the time to share these detailed examples. The op-amp circuit looks really interesting even though I won't have enough pcb space to fit everything on.You can get opamps in SOT353 packages, which may help (eg LMV321) ?
I found that a rather correct deflection could be obtained with a pulse ranging between 8KHz and 10KHz. I didn't put a resistor before the LED and will test further tomorrow.Do you vary the duty cycle ?