Isolated DC-DC Transformer with 6 Secondaries??.......

[Smokey]

From the paper dobsonr741 linked:



I'm thinking down on the right side in "Converter based" land, but I like the idea of generating the supply for the converter with a cap charge pump (if that's possible).

[Someone]

At least ESR isn't hard to come by; a 10uF X7R will still have plenty of value even in 0805 size, at 4.2V.  10uF might have to run pretty fast still (say, a MHz for <20mOhm), so probably a somewhat larger size would be chosen (100uF?) or enough used in parallel, or electrolytic (even bigger to keep ESR down), or maybe polymer (if economical).  MOSFETs aren't a problem, indeed it's almost a problem if anything that they're hardly available at low enough voltages to keep Qg nice and low.  (IIRC, selection is narrow -- but indeed available.)

At lower frequencies there are some discrete MOSFETs demonstrated for low loss in charge transfer duty with the LTC7820, and the rather elegant inductor integrating LTC7821
can't recall the other integrated switched capacitor parts I had come across recently but the Renesas DA9313 gives some examples of the practical capacitances.

[T3sl4co1l]

So I need separate cell referenced rails in the 4.5V-6V range.  But I think Weston was right, that they don't all actually need to be fully floating.  If I have an external 24V and I used a flyback to make one floating 5V rail and reference it to the bottom cell, I should be able to use that rail to generate the other for the remaining 5 cells in series with some sort of charge pump.  That was the idea.  I can see through how to do that with a multi-tap transformer (thread title...), and conceptually it makes sense that should be doable with a capacitor charge pump but it's the details of that which I'm not cool enough to see all the way through (and how to do it inexpensively and at the required current).  The closest circuit my brain can see is a high side gate driver supply generator that ends up floating on the high side FET, but I'm not sure that sort of thing can supply a continuous 150mA.

But anyway yeah, that's easy enough with multiple windings, and, since there's no AC common mode, you could cap-couple channels together ala SEPIC.  Whether you want to do some outputs, or all but one main channel, with coupled inductors, or independent ones, or just with caps and diodes (charge pump style, in which case you need Vpp correct, not just peak), I suppose is whatever.  You probably don't want to stack capacitors (Cockroft-Walton style) as regulation declines along the stack, but with a star topology (one cap from main to each rectifier) you need more voltage rating per cap as you go up.  Which shouldn't be a big deal, like, 20V would be a big battery (say the main channel is in the middle so +/-20V is a 40V pack) and 20V is quite feasible in 1206s or 1210s in useful values (couple 10s uF?).

Tim

[Someone]

So I need separate cell referenced rails in the 4.5V-6V range.  But I think Weston was right, that they don't all actually need to be fully floating.  If I have an external 24V and I used a flyback to make one floating 5V rail and reference it to the bottom cell, I should be able to use that rail to generate the other for the remaining 5 cells in series with some sort of charge pump.  That was the idea.  I can see through how to do that with a multi-tap transformer (thread title...), and conceptually it makes sense that should be doable with a capacitor charge pump but it's the details of that which I'm not cool enough to see all the way through (and how to do it inexpensively and at the required current).  The closest circuit my brain can see is a high side gate driver supply generator that ends up floating on the high side FET, but I'm not sure that sort of thing can supply a continuous 150mA.

But anyway yeah, that's easy enough with multiple windings, and, since there's no AC common mode, you could cap-couple channels together ala SEPIC.  Whether you want to do some outputs, or all but one main channel, with coupled inductors, or independent ones, or just with caps and diodes (charge pump style, in which case you need Vpp correct, not just peak), I suppose is whatever.  You probably don't want to stack capacitors (Cockroft-Walton style) as regulation declines along the stack, but with a star topology (one cap from main to each rectifier) you need more voltage rating per cap as you go up.  Which shouldn't be a big deal, like, 20V would be a big battery (say the main channel is in the middle so +/-20V is a 40V pack) and 20V is quite feasible in 1206s or 1210s in useful values (couple 10s uF?).

I think the problem with the capacitor + diode coupled outputs is they will return current through to the common, which sums in the battery string. By the time you make the converter bidirectional to sink and source current at each output you've re-invented the whole charge balancer.

(and how to do it inexpensively and at the required current).

Since you are working along the plan of having the charge control per cell, the cheapest option to match that may well be a separate isolated supply per controller/cell: 30c synchronous buck + 50c coupled inductor + 2c secondary diode?

[dobsonr741]

Why to charge all cells individually and not as one single chain? You have plenty of BMS for the later. Capable to properly terminate charge when the highest reaches target, then rebalance to all be equal a bit lower. Do I feel you are opposing rebalancing for some reason?

[T3sl4co1l]

I think the problem with the capacitor + diode coupled outputs is they will return current through to the common, which sums in the battery string. By the time you make the converter bidirectional to sink and source current at each output you've re-invented the whole charge balancer.

You sure about that?

Tim

[Someone]

I think the problem with the capacitor + diode coupled outputs is they will return current through to the common, which sums in the battery string. By the time you make the converter bidirectional to sink and source current at each output you've re-invented the whole charge balancer.

You sure about that?

Thanks for the clear schematic, it confirms we are talking about the same thing. The DC current through the lower cells on that string is 0 but the AC current isn't. At high frequencies that could be bypassed by significant capacitors in parallel with the batteries but not sure if that will affect the operation of the charge controllers.

It certainly is an interesting cost optimisation problem for the OP.

[T3sl4co1l]

I wouldn't think you'd do more than say three above and below a given cell, an amount easily bypassed away.  The AC ground return can be tied to any tap you like, since it's completely AC coupled.

Or, if cumulative AC is a problem, pump it bipolar, using twice as many coupling caps and rectifiers.

Tim

[Weston]

T3sl4co1l posted a schematic before I could get around to it, but thats more or less what I was talking about.

You can also replace the capacitor with an LC circuit, which gives you better switching waveforms with a smaller capacitance value. But inductors probably cost more than a bigger capacitor.

One possible issue is the inrush current, but with suitably placed resistor dividers you can pre-charge the capacitors.

[Smokey]

Sweet.  You guys are the best.  LTSpice confirms.  That looks like it works (at least with ideal components). 

I found this paper, which shows what looks like a similar thing:
https://minds.wisconsin.edu/bitstream/handle/1793/11132/file_1.pdf



And this app note has descriptions of cap and diode selection for a circuit using a 555 as the pulse frequency and output:
https://www.ti.com/lit/an/slva444/slva444.pdf


Another TI app note: https://www.ti.com/lit/an/slva398a/slva398a.pdf
And another forum thread: https://www.eevblog.com/forum/projects/capacitor-type-for-charge-pump/

[Smokey]

Working on turning this multi-channel charge pump from ideal simulation to actual circuit... check me if I'm off track on any of this.... I'm targeting about 125mA for the output load on each of the 6 supplies.

Because of the high peak pulse current, I'm looking at using independent push-pull drivers for each of the 6 charge pump supplies.  That sucks because it increases parts count/cost, but the equivalent single solution would have to be able to handle the sum of the peak currents which would make it bigger and more expensive itself.  There is also an output voltage imbalance from different output rail loading that I think having independent drivers mostly does away with.  I'm not sure how much I need to worry about layout parasitics, but I would imagine that would be worse with a single driver.

Because I have the adjacent cell voltages available, I could in theory mostly pre-charge the output cap on all cells but the top one before turning on the charge pumps (the next cell isn't available on the top one).  But that again adds parts cost and sequencing complexity.

I haven't got a full BOM yet, but I'm wondering if the flyback + transformer solution will actually be less expensive than the charge pump after a real world circuit is together.  The flyback is certainly less involved, and less finicky. 

[Weston]

Are you referring to the current spikes at the start of each transition caused by the capacitor charging up?

A few comments on that:

You are fundamentally charging a capacitor with a resistor (on resistance of FET, diode resistance, capacitor ESR). Loss is proportional to delta V, which depends on the capacitor size and switching frequency. Independent drivers / lower ESR fets will not necessarily reduce loss.

A higher capacitor value will reduce the loss and make the current less peak-ey. If the RC time constant is much larger than the switching frequency you get flat current waveforms and good efficiency (assuming ESR is low). I did not look too closely, but this app note seems to talk about the relationship between capacitor size, switching frequency, and output impedance / capacitor ripple: https://www.analog.com/en/technical-articles/guide-to-integrated-charge-pump-dcdc-conversion.html

You can also use a resonant charge pump by adding an inductor in series with the flying capacitor and driving at the series resonant frequency. You can get a lower impedance and sinusoidal current waveform with a smaller capacitor value