Design a better inverter- win a million bucks!

[miguelvp]

They are jealous of Ebay and their Fuel Cell backup generators

But the answer is obvious:
https://www.kickstarter.com/projects/315398504/perpetual-solar-flywheel-energy-storage-by-energie

They didn't get funded and it's dangerous as heck, but flywheels will make a great inverter as far as energy conversion, just make sure the flywheel has no chance to break loose and create havoc in your home or business. One small enough to qualify if it broke loose it would be like a high energy cannon ball that will chew through everything in its path.

[Marco]

In theory you could drive a flywheel with a "magic sine" and then run a generator off it (in essence you use the flywheel as a reactor, which is necessary with the "magic sine" approach to efficiently supply low PF loads). I doubt it's going to be as compact as a solid state device though.

[gxti]

Could it be that Google simply needs better power inverters for their data centers? That this is not about bringing small inverters to the homeowner masses?

The only place you'd find an inverter in a datacenter (not counting VFDs in chillers or whatever) is in battery backup systems, and size definitely doesn't matter for those. But the cool thing these days for server farms is dual-supply servers that take unconditioned AC on one input (often at a higher voltage like 380V) and 48VDC from a battery bank on the other, and thus no inverter is used at all. Google and Facebook use this type of arrangement widely already.

[SirNick]

I thought they were using local backup batteries in their DCs?  It has been a couple of years, so things may have changed, but I remember seeing commodity PC motherboards with a small SLA and a DC-to-DC converter all in the same chassis.  You can lose power all the way to the unit PSU and it keeps running.

It seemed like about as efficient a setup as you could hope for.  Power in -> transformer + rectifier or HVDC-to-LVDC -> battery -> DC-to-DC converter for required rail voltages.  Easy.

Management-wise, it might be preferable to have it all centralized, but there are no shortage of solutions for battery bank to AC inversion, or native -48V power distribution.  Right now, I'm sitting about two floors up from a big room full of lead-acid batteries.  Certainly, the inverters aren't small, but compared to the space consumed by the battery bank, it's probably not on the short list of top problems to solve.

[johansen]

The lead acid battery story is worth mentioning again because what they did was design their own motherboards that ran from 12 volts only, no 5 volt, and no 3.3 volt needed. (which really isn't that hard to do, and would reduce the cost of the power supply significantly)

then you can use a couple cheap mosfets to let the battery float at 13.xx volts while the motherboard runs from 12 volts.
when the power fails, the load pulls the battery voltage down to 12.56 volts fully charged which is well within operational tolerances.

why Goggle needs 450vdc to 240vac inverters i have no idea.

[David Hess]

I thought they were using local backup batteries in their DCs?  It has been a couple of years, so things may have changed, but I remember seeing commodity PC motherboards with a small SLA and a DC-to-DC converter all in the same chassis.  You can lose power all the way to the unit PSU and it keeps running.

I remember seeing the same motherboard design.  There were or still are ATX power supplies available which include battery backup with an externally attached battery but I think they work by stepping the battery voltage up to 340 volts DC to power the existing switching power supply.  A low voltage 12 volt implementation of the same thing is going to be more limited in output power.

Those ATX power supplies need an inverter to supply 120VAC to the monitor or other auxiliary equipment.  One company (Amsdell?) had a tiny inverter which I think was SCR based and they made a big deal about how advanced it was.

Implementing the backup system directly on the motherboard makes sense because the high power point of load regulators already accept a 12 volt input and the individual motherboards and their attachments do not draw that much power anyway.  48 volts would be more efficient as a intermediate bus voltage but then the battery has 4 times as many cells and the point of load buck regulator has to run as 1/4 the duty cycle of the one which accepts a 12 volt input which may be a problem.  Tapped inductors can be used to get around that issue but they would be another complication.

It seemed like about as efficient a setup as you could hope for.  Power in -> transformer + rectifier or HVDC-to-LVDC -> battery -> DC-to-DC converter for required rail voltages.  Easy.

Higher DC voltages are more efficient and needed for heavier loads anyway.

Management-wise, it might be preferable to have it all centralized, but there are no shortage of solutions for battery bank to AC inversion, or native -48V power distribution.  Right now, I'm sitting about two floors up from a big room full of lead-acid batteries.  Certainly, the inverters aren't small, but compared to the space consumed by the battery bank, it's probably not on the short list of top problems to solve.

Maybe it is for reliability?  Centralized solutions tend to fail all at once and it makes interesting news when it happens at data center magnitudes.  Watching the converter or transfer switch self destruct is probably something best viewed from a distance.

[madires]

why Goggle needs 450vdc to 240vac inverters i have no idea.

The HVDC power distribution includes an online UPS, i.e. a bunch of lead acid batteries in series. That's more efficient than using a battery bank with a large UPS (with inverters). But some stuff, like air conditioning, needs AC. Either you install a classic UPS for the AC powered stuff including the AC wiring or you use inverters directly powering the AC stuff.

[David Hess]

why Goggle needs 450vdc to 240vac inverters i have no idea.

The HVDC power distribution includes an online UPS, i.e. a bunch of lead acid batteries in series. That's more efficient than using a battery bank with a large UPS (with inverters). But some stuff, like air conditioning, needs AC. Either you install a classic UPS for the AC powered stuff including the AC wiring or you use inverters directly powering the AC stuff.

The funny part about that is that high efficiency air conditioners use variable speed drives so internally they rectify the AC to DC before chopping it up to drive the motor anyway.  It is turtles all the way down.

[madires]

The funny part about that is that high efficiency air conditioners use variable speed drives so internally they rectify the AC to DC before chopping it up to drive the motor anyway.  It is turtles all the way down.

There are also a lot of pumps involved. But I assume that more and more of the utility systems will be migrated to HVDC too.

[johansen]

no, their custom server boards that i recall reading about ran the board directly from the lead acid battery in the case of a power outage, no reason to waste power converting it back to 340vdc to send it through the power supply.


but you need sealed cells to do that at the server level.
saturated cells running at 450 volts makes a lot more sense.
however, why bother converting the 450vdc back to 240vac, then back to 400 volts in a boost converter, then turn that into 200vac at 100Khz and back to 12vdc.

just go directly from 11kvac transformer to 480 vac to 450vdc through a 12 pulse transformer (no power factor correction required) (insert lead acid storage here)
then take that 450vdc and feed it right into the forward/flyback/LLC "ATX" power supply.

[T3sl4co1l]

That's kind of funny actually, because laptops generally run an internal 12-20V bus from which all the supplies run.  I counted, I think: ~1.2V (core), 1.8V (mobile RAM, PCIe?), 2.5V (peripherals? / PCIe?), 3.3V  and 5V (main logic, peripherals), and a battery controller -- six synchronous converters in all, in the last laptop I took apart.  All of them have the same common bus coming in the top of the converters (easily proven, as it is easy to find the high side drain pins and their meaty ceramic bypass caps).

PCs are kind of silly, always having to maintain backwards compatibility.  It would be very difficult to migrate suddenly, when few ATX+ supplies even tolerate asymmetrical loading of outputs.  So, they haven't done this, gone all to just 12V or whatever.

Tim

[tom66]

To minimise copper losses the next generation PC should have a 28~40V DC bus and a 5V standby only. No need to have -12V,3.3V&5V with associated minimum loads and regulation issues, also much lower copper losses.

[SirNick]

I never understood the 3.3v rail.  Anything less than 5v really should be point-of-load regulated.  This is clearly understood by the inclusion of the 3.3v sense return.  They had the perfect opportunity to leave out another entire rail when they changed from the AT connector years back.

I also wonder how often (if ever) the 12v rail gets used directly.  Some of those mini-ITX (etc.) PSUs take 12v in, and pass it right through with barely a bypass cap.  No assumptions can be made about the quality of power coming in, so I assume (?) the designers of such PSUs either don't care at all, or count on the fact that nothing downstream is going to use that rail for anything but high-current buck converters anyway.

[NiHaoMike]

18V or 24V would probably be the most optimal battery voltage if the load is right next to it. 12V has the disadvantage that the actual battery voltage could be either higher or lower than 12V, meaning a more complex DC/DC converter design.

I have seen a server that used a 20V PSU capable of supplying about 60A.

[Marco]

Has anyone done some math on what kind of input capacitance you need to reach the input ripple?

[David Hess]

I never understood the 3.3v rail.  Anything less than 5v really should be point-of-load regulated.  This is clearly understood by the inclusion of the 3.3v sense return.  They had the perfect opportunity to leave out another entire rail when they changed from the AT connector years back.

There was a fair amount of logic which could operate on 3.3 volts directly and currents were low enough that low dropout linear point of load regulators could be used to generate lower voltages from the 3.3 volt rail at with good efficiency.

I also wonder how often (if ever) the 12v rail gets used directly.  Some of those mini-ITX (etc.) PSUs take 12v in, and pass it right through with barely a bypass cap.  No assumptions can be made about the quality of power coming in, so I assume (?) the designers of such PSUs either don't care at all, or count on the fact that nothing downstream is going to use that rail for anything but high-current buck converters anyway.

12 volts is made available to expansion cards.  Audio outputs and RS-232 can make use of it.

[David Hess]

18V or 24V would probably be the most optimal battery voltage if the load is right next to it. 12V has the disadvantage that the actual battery voltage could be either higher or lower than 12V, meaning a more complex DC/DC converter design.

Higher battery voltages will not help with input voltage variation and DC to DC converters are not going to care.  Multiples of 12 volts has the advantage of cheap lead acid batteries.

[ejeffrey]

I never understood the 3.3v rail.  Anything less than 5v really should be point-of-load regulated.  This is clearly understood by the inclusion of the 3.3v sense return.  They had the perfect opportunity to leave out another entire rail when they changed from the AT connector years back.

I believe the issue is that in the pentium pro era, practical switch mode regulators were not fast enough to supply the load transients demanded by the CPU.  Therefore the point of load regulators were linear regulators and they wanted as low a voltage as practical to limit heat dissipation by the VRM.  Of course, in the early 2000s switchers were getting better and core voltages were dropping rapidly to make linear regulators completely impractical, but that was well after ATX became standard.  IIRC, the move to put everything on 12V started with the P4, which added the extra 12 V connector, but even then that was really only for the CPU.  Everything else still ran off 5 and 3.3, which were the only high power connections available to PCI cards and at least the first generation of AGP.  It wasn't really until PCIe that video cards started pulling a lot of power from 12V.

Now that all the high power stuff has moves to 12V, you could certainly get rid of the extra wires and put 3.3 and 5V DC DC converters on the motherboard, but other than saving a few pins on the power connector I am not sure what that saves you.  It is basically free to hang an extra winding and diode on the main supply, and there are still a lot of things that need those voltages.

[johansen]

Has anyone done some math on what kind of input capacitance you need to reach the input ripple?

oh, you mean the uh, rather insane specification of 20% ripple?

i suspect they want it that low so that they do not have to sychronize the sine wave outputs.
but yes, it is retarded.
three phase ac would demand 16% ripple with no input filter at all.

lol.
I suspect you will need to put a PFC boost or buck converter between the 450volts dc input and the full bridge ac inverter to win this competition.

[SirNick]

There was a fair amount of logic which could operate on 3.3 volts directly and currents were low enough that low dropout linear point of load regulators could be used to generate lower voltages from the 3.3 volt rail at with good efficiency.

12 volts is made available to expansion cards.  Audio outputs and RS-232 can make use of it.

True -- however, is it common practice to use the rails from the PSU directly?  (Honestly don't know.)  E.g., if you were Asus, would you attach your 3.3v logic directly to the 3.3v rail from the PSU, or would you use a 5v-to-3.3v regulator?  If that latter, is there any point in providing key voltages, or does it make more sense to offer something that is within comfortable drop-out range of target voltages?

As for 12v -- agreed, those are the most likely uses.  But, it seems like most of the popular RS-232 interface ICs just use internal charge pumps now.  Admittedly, that might be new since ATX's takeover two decades ago.  (Which is probably why we still have a -12v rail that no one uses.)

For audio, most codecs now run from 5v or less.  It might be useful for op-amps in higher-quality audio circuits, except no one worth their salt is going to use the output from a switchmode PSU without running it through a linear regulator first anyway.

It's kinda looking like you could get away with all kinds of crap on the 12v rail and nothing would care.

[SL4P]

OK, my 2c worth goes into a regenerative approach - a number of simple in-wheel alternators' fixed to and using the rotation of the axles to create a high amplitude, very irregular, alternating, phase incoherent waveform that is effectively free from the forward or reverse vehicle motion. 

Rectify & smooth to charge (large) supercaps (batteries only needed to initiate acceleration from zero) - as the harsh AC is present to sustain the supply when needed.  some smarts could be added to sense the phase relationships between wheels, and route non-synchronous waveforms around the wheels to provide 'free' electromagnetic braking.

All part of existing regenerative charging and braking - but taken to a ridiculous and potentially (no pun) meaningful extreme.

A small block of batteries to provide the first 0-10kmh acceleration, then asynchronous AC-> DC closed loop to sustain drive and braking.  The batteries can also augment instantaneous shortfalls in the bumpy alternator waveform

[ejeffrey]

True -- however, is it common practice to use the rails from the PSU directly?  (Honestly don't know.)  E.g., if you were Asus, would you attach your 3.3v logic directly to the 3.3v rail from the PSU, or would you use a 5v-to-3.3v regulator?  If that latter, is there any point in providing key voltages, or does it make more sense to offer something that is within comfortable drop-out range of target voltages?

Yes, 3.3 V logic on a motherboard would typically just directly off the ATX power, as would any 5V logic although I doubt there is any 5V logic or even IO left.  The 5V USB supply and 3.3 and 5 V supplies to PCIe would also be taken directly from the ATX power (with a polyfuse or active current limit on the USB 5V).  Any chips on expansion cards that needed those voltages would also just use them unless they needed a lot of current which would justify stepping down from the 12 V line.

[David Hess]

There was a fair amount of logic which could operate on 3.3 volts directly and currents were low enough that low dropout linear point of load regulators could be used to generate lower voltages from the 3.3 volt rail at with good efficiency.

12 volts is made available to expansion cards.  Audio outputs and RS-232 can make use of it.

True -- however, is it common practice to use the rails from the PSU directly?  (Honestly don't know.)  E.g., if you were Asus, would you attach your 3.3v logic directly to the 3.3v rail from the PSU, or would you use a 5v-to-3.3v regulator?  If that latter, is there any point in providing key voltages, or does it make more sense to offer something that is within comfortable drop-out range of target voltages?

They would use the 3.3 volt supply directly if possible.  Otherwise there would be no point in having it.  Later processors had lower core supply voltages and initially a low dropout linear regulator was used to provide them powered by the 3.3 volt supply.  Efficiency is good dropping 3.3 volts to 2.5 volts.  The PCI bus supports both 5.0 or 3.3 volt signaling.

As for 12v -- agreed, those are the most likely uses.  But, it seems like most of the popular RS-232 interface ICs just use internal charge pumps now.  Admittedly, that might be new since ATX's takeover two decades ago.  (Which is probably why we still have a -12v rail that no one uses.)

The commonly used 1488 RS-232 line driver uses +/-12 volt supplies.  Later level shifters included charge pumps.

For audio, most codecs now run from 5v or less.  It might be useful for op-amps in higher-quality audio circuits, except no one worth their salt is going to use the output from a switchmode PSU without running it through a linear regulator first anyway.

Now they do but 12 volts is still useful for signal conditioning and if you have a power amplifier.

It's kinda looking like you could get away with all kinds of crap on the 12v rail and nothing would care.

That is the case even now.

[T3sl4co1l]

Has anyone done some math on what kind of input capacitance you need to reach the input ripple?

Don't know why they give different figures for voltage (20%) and current (3%), but clearly the 3% figure is the problem.

3% of 450V is 13.5V, and that's peak to peak over the low frequency content (excluding HF noise).  An average 2kW load goes from 0-4kW during the cycle, and varies according to the sin^2 t = (1+sin(2*t))/2 identity.  Therefore, the AC component of the current draw is sinusoidal, twice the frequency of the output, and peak-to-peak equals the (0-4kW) / (450V) spread, or 8.9A (it's actually a distorted sine, more current at lower voltages due to supply droop).  Let's say it's 10A.  Worst case frequency is 50Hz, or a ripple of 100Hz.  The reactance must be less than 13.5Vpp / 10App = 1.35 ohms, or 1179uF.

Snap-ins would be good, a row of 10 x 100uF 450V parts would be... (consulting the Digikey oracle here) maybe a 22 x 32 mm can, x 10 would fit along an 8.6" length side of a flat rectangular shape enclosure.  So actually, that's not too terrible.

I don't think PFC is going to cut it: you're trying to squeeze a PFC switch, controller, inductor -- and don't forget the outside filtering and EMI components for it too! -- into the space of, say, 5-7 of those electrolytics.

Tim

[Marco]

The 500v ones (lets have just a bit of safety margin) are 40 length, add in the snap ins and 2x1mm sheet metal we get to 46mm ... lets assume non staggered mounting, that's 22^2*46*10 = 222640 mm^3 ~= 13.59 inches. So a third of your volume gone on just the input capacitors (and your board area with a single PCB).