[amluto]

I have to admit that I really dislike using ~12V batteries for high power applications like this. I say this having built a ~400A ~14V system. It’s miserable.

1 kW at 100V or 250V or similar uses a nice, small, flexible wire. It can be quite safe because it can be fused or otherwise protected at low currents, which mitigates the risk of welding things, starting fires, or arcing. Ground fault protection, arc fault protection, and general loss-of-isolation protection are available. It’s easy to rework (lever nuts! screw terminals!).

400A (or even 80A or so like in this article) is a whole different ball game. Sure, you have to work hard to electrocute yourself. But you can easily set things on fire or weld things together without coming close to blowing a fuse. And you need to protect both ends of wires in a parallel arrangement. And the wires are enormous, expensive, and hard to terminate.

I would much prefer one of three alternative designs to become popular:

1A: a series arrangement of batteries at a civilized 48V or so. You can do this with an aftermarket BMS, but they tend to be janky.

1B: same but actually high voltage (a few hundred V, like an EV)

2: batteries with microinverters and a civilized way to share current. A manufacturer could make a single package with a 1kWh battery, a BMS, a low voltage, low current DC auxiliary output, and a ground-fault and overcurrent-protected 110-250V AC input/output. And an RS485 or 10BASE-T1S or CAN connection so that they can coordinate their I-V characteristics to appropriate distribute charge or discharge current.

Now you can connect as many microinverter-batteries as you like in parallel, using #14 wire, to one ordinary circuit breaker per battery plus (depending on the overall arrangement) one big breaker to protect the common bus.

edit: Also, with this design, no one, not even the manufacturer, needs to touch a heavy-gauge wire. Everything in the battery would use cheap, painless busbars or small wires, depending on the internal voltage, and the manufacturer could set the voltage however they like. Although 12V internally might be entirely reasonable if the end user also wants to consume 12V at very low currents through the aux output.

[samtho]

I use 48v for all my high power DC applications and I usually spin PCBs. Part of the reason I use 48vdc is that there are bunch of ICs and reference designs for PoE applications that run at this voltage level.

12v is a miserable voltage to work with in general: too high for most logic circuits and for LDOs to work effectively but it’s too low for any large load with high-power applications become cost prohibitive quickly due to the cost of the conductors. Fun fact: Auto makers have gotten away with under sizing starter conductors, despite it drawing 80-100A, because it is only energized briefly and the length of the run is very short.

[gaudat]

And I think 48VDC is the highest convenient voltage you get before exceeding ELV. There are a lot of surplus telecom and server power supplies that gives 48V at decent current. Above that there is a real risk of electrical injury, at least according to regulations.

[samtho]

You can actually go up to 120VDC (with no ripple) before you’re out of ELV - which I always find surprising. You’re probably thinking of the A/C limit which is 50V RMS.

[gaudat]

Interesting, I thought the limit is at 60VDC.

[omegabravo]

IEC limits are 120VDC, 50VAC which quite a few countries will use (eu*/au/nz).

US NEMA/NEC is a very different standard.

[auxym]

60 VDC / 30 VAC in Canada per CSA Z462 (workplace electrical safety).

[Animats]

Much industrial automation runs on 24VDC. High enough that the losses within a cabinet are tolerable, low enough that there's no electrocution hazard.

[amelius]

If you keep your hands dry or if you're a male, then 48VDC can be quite safe too.

https://www.allaboutcircuits.com/textbook/direct-current/chp...

We're running out of copper. So better keep your voltages on the high end (but still safe of course).

https://www.cnbc.com/2023/02/07/there-isnt-enough-copper-in-...

[mbreese]

To save someone else looking for the answer to why being male is a factor here… (which I didn’t know about!)

> Oh, and in case you’re wondering, I have no idea why women tend to be more susceptible to electric currents than men!

The pain thresholds are given in a table that is split by men and women. I believe the assumption is different conductivity between “typical” male/female bodies, but it’s really not spelled out.

[amelius]

Yes, on one hand it's strange because body fat decreases conductivity, and women have typically larger amount of body fat (25% versus 15%).

https://www.ncbi.nlm.nih.gov/books/NBK218162/

However, men are typically larger, so perhaps that's the reason.

[ilyt]

Soo fat people are more shock-resistant too ?

[mjevans]

There's a push for '12vo' or 12 volt only PSUs for PCs... but with the major power consumers mostly the GPU, CPU, RAM (?), and possibly spinning drives, would a similar to telecom gear 48V make more sense?

For GPUs and any USB PD connections ('standard' is up to 20V 5A while 'extended' is 48V 5A) a system power level of ~48v might be very useful. It would also give a chance to replace the recent higher density connectors that aren't designed with a sufficiently robust user experience with improved versions that more clearly lock into solid connection.

[crote]

Not really.

Pretty much the only PC part which wants native 12V would be the fans. All the other parts drop it down to 5V / 3V3 for auxiliary components, or 1-2V for the CPU and GPU cores - which use the vast majority of power.

Dropping 12V DC to 1.5V is reasonably doable, but dropping 48V DC to 1.5V is a bit of a pain. In general you do not want to go beyond a 1:10 ratio for efficiency reasons, so 48V doesn't really gain you anything, while at the same time resulting in a massive compatibility break.

The push to 12VO is driven by a desire to get rid of technical debt. The 3V3 / 5V wires can't handle the current you need on those rails, so those are converted from 12V anyways. And literally nobody is using the -12V and -5V wires, so keeping those around is pointless.

[namibj]

Open compute platform switched to 48V years ago, to afford having multiple PSUs feeding multiple servers, with (those days) a fixed 4:1 switched capacitor down converter right next to the existing 12V-to-0.5~1.5V DrMOS VRM power stages.

There are iirc GaN devices from epc-co these days that make it feasible to go directly from 48V with the VRMs to the 0.5~1.5V for the core.

[mkj]

Why does the ratio affect efficiency? Naively it seems like it should just be able to use differently sized inductors.

[myself248]

Yes, I thought 48vo was already on the drawing board.

[londons_explore]

> And an RS485 or 10BASE-T1S or CAN connection so that they can coordinate their I-V characteristics

Note that simply voltage/frequency alone can communicate everything necessary to evenly distribute load. For example:

Output frequency = 60 + X Hz

Where X varies between -0.1 and 0.1, indicating the percentage of max load the inverter is under(in charge or discharge direction).

Such a scheme will self synchronize any number of inverters, even of varying capacities, into a stable grid, and they will all be the same percentage loaded. The grid can even have a few dumb gas generators on too and remain stable. No fancy comms needed.

[marcosdumay]

That's not simple to create with inverters alone.

Yep, inverters can easily read the frequency change to adapt their output. But a device that reads the network load and translates it into a frequency change is really not trivial. Grids do this because mechanical generators do that translation, but the easiest way to add it to a non-mechanical grid is adding a mechanical device.

Lines exclusively fed by inverters will naturally get undervoltage, instead of frequency shifts.

[londons_explore]

> But a device that reads the network load and translates it into a frequency change is really not trivial.

Nah - it's pretty simple. Have your inverter measure it's own load - for example by measuring the current through the main switching element. Compare that to the max current that switching element can handle. That is your percentage load that you are under. (negative if current goes backwards)

Then use that to decide what frequency you will output, using the formula.

There is one extra element to this... At startup, to avoid large transient currents, you need to start with the 'max switching element current' for the formula at zero, and then increase it smoothly only while the calculated load figure is under 100%. Within a few seconds you will get sync with any existing network. Such a scheme also works for blackstarting an isolated grid, or with many other inverters with the same strategy.

This scheme has a big benefit that there is no need to measure the grid frequency precisely - which is actually a fairly hard thing to do on noisy small grids.

And it turns out thats all thats required to maintain grid sync. As long as supply capacity exceeds demand at all times, this grid will be stable.

[amluto]

Do you have a reference?

If I followed this right, the frequency goes up with load (which seems backwards). So a 50%-loaded grid will be at 60.05 Hz. A newly started inverter will try to output 60 Hz (and fail, since it is unable to single-handedly reduce the frequency). But it will surely end up with nonzero current.

Don’t real spinning synchronous machines work the other way? If they are overloaded, they spin slower, and if they are spinning a bit slower than the grid when connected, the grid will speed them up?

[londons_explore]

By the way, the reason big national grids don't allow this scheme is because it is too stable. Ie. if your house, containing some inverters like this, is disconnected from the national grid, it will continue independantly. Your house and the national grid will continue on their own paths, with slightly differing frequencies.

Reconnecting two grids at differing frequencies is hard.

[mschuster91]

> Reconnecting two grids at differing frequencies is hard.

Indeed it is. The last time this happened in 2021 here in Europe, South-Eastern Europe split off, and it took a good hour to get everything balanced again [1].

[1] https://www.entsoe.eu/news/2021/01/26/system-separation-in-t...

[ilyt]

> if your house, containing some inverters like this, is disconnected from the national grid, it will continue independantly. Your house and the national grid will continue on their own paths, with slightly differing frequencies.

That's not the main reason.

The main reason is that if grid is down, workers need to work on it safely and so they need it de-energized, not some random solar installation trying to feed power into it.

Some inverters for that reason have 2 outputs, one for grid, other one for so called EPS (emergency power supply). When grid goes down, usually relay disconnects the two and inverter only feeds power to EPS, and re-connects both outputs only when grid is back up and synced. So you'd shove your important loads onto EPS output and have it grid-independent.

[amluto]

> The main reason is that if grid is down, workers need to work on it safely and so they need it de-energized, not some random solar installation trying to feed power into it.

I’m sure this has motivated someone who makes rules, but I once had occasion to ask some actual line workers replacing equipment serving me, and they laughed. They said that they always assumed the lines were hot at both ends, and if they needed a line to be de-energized for safety, they would deliberately short it out. A pesky little residential inverter was not in the slightest bit concerning to them, anti-islanding or no.

I do believe that closing a switch between the grid and a small residential island could be unpleasant for the island or maybe even for the switch. And closing a large switch connecting an entire neighborhood that somehow formed a functional island and connecting it to the grid without synchronization might genuinely go poorly.

But mostly I think the most important current reason for anti-islanding is that a small residential or light commercial inverter is unlikely to have anywhere near the capacity to power an entire secondary circuit, nor does its owner want it to.

[ilyt]

It might be pretty complex code wise but not cost wise, you just need to probe current and voltage, which you do anyway

[amluto]

I think this works for discharge, possibly with some extra cleverness needed if the batteries are at different SoCs. But I’m not sure how it could handle being tied to the grid intelligently or how charging could be managed without some additional controls.

Also, can a pure frequency-based scheme handle the case where a whole bunch of inverters are in parallel and a load that’s much larger than any one of them can handle individually starts up?

[londons_explore]

> handle the case where a whole bunch of inverters are in parallel and a load that’s much larger than any one of them can handle individually

Yes, but with caveats... The grid will startup, smoothly (ie. gradually), but during startup every inverter will be at its maximum configured load (with the maximum load config ramped during startup). When the maximum load is hit, the frequency sits at the minimum design frequency (59.9 Hz in this case), and voltage drops instead.

So your large device will see a sine wave at 59.9 Hz that starts at zero volts, and increases in amplitude to 230 volts gradually over many seconds, and as soon as it hits 230 volts, the frequency will almost immediately become 60 Hz.

For large motors, thats a problem. Large motors typically don't like line frequency at a reduced voltage - they can end up not having enough torque to turn, and will just rapidly heat up. They would do better without the gradual startup ramp - but that startup ramp is necessary to ensure the inverters stay in sync.

In reality, this scheme works for AC type motors rated up to about 30% of the capacity of the inverter set, or up to 100% as long as another motor is already running on the same grid.

[londons_explore]

> extra cleverness needed if the batteries are at different SoCs

The scheme as presented keeps the inverters under equal fraction load, but does not guarantee the batteries end up at equal state of charge. The formula can easily be modified to over time achieve this too:

X = 60 + (fraction load of inverter)*0.1 + per_battery_factor

The per_battery_factor would be set based on the batteries state of charge and/or any user instructions from buttons to charge or discharge a specific battery.

[ilyt]

> I have to admit that I really dislike using ~12V batteries for high power applications like this. I say this having built a ~400A ~14V system. It’s miserable.

The schematic looks to be pretty adaptable to higher driving voltage, just need separate 12V for control board. There is even one in datasheet for 24-36V operation

>2: batteries with microinverters and a civilized way to share current. A manufacturer could make a single package with a 1kWh battery, a BMS, a low voltage, low current DC auxiliary output, and a ground-fault and overcurrent-protected 110-250V AC input/output. And an RS485 or 10BASE-T1S or CAN connection so that they can coordinate their I-V characteristics to appropriate distribute charge or discharge current.

> Now you can connect as many microinverter-batteries as you like in parallel, using #14 wire, to one ordinary circuit breaker per battery plus (depending on the overall arrangement) one big breaker to protect the common bus.

You can build it right now. AC coupled batteries exist; here is some random one that scales up: https://www.fortresspower.com/ac-coupled/

The problem is that you generally want batteries when you want renewables and in that case just having one big box handling batteries and solar panels is more economical than microinverters everywhere

48V battery pack + BMS is significantly cheaper than same thing with microinverter, and when you scale up one big inverter is cheaper than a bunch of smaller ones.

So yeah, it is "best" but also most expensive way. And frankly, the hardest to develop, which is probably why there is little to no open designs for that.

[h2odragon]

Depending on the load, you can often skimp on the conductors and get away with it. Someone else mentioned undersized starter power cables. There's really only a few loads that are full draw, full duty cycle.

Old car stereo trick, to power big amplifiers in the trunk: use a second battery and big power lead to it, small power leads from the main battery to the secondary. That secondary can be backed or replaced by big capacitors, too; with commensurate increases in cost and possible risks when things go wrongs. But you can provide rich chunky amps and use skinnier cable than you'd think on the long run to do it.

[mschuster91]

Never forget to put an override relay (normal open, closing when the engine has run for a set amount of time or when manually bypassed) and a fuse in the path though (and that also applies for campers). There are a number of things to consider, and if done badly, also serious risks:

- you don't want a permanent connection between the main and aux battery to avoid accidentally sucking both batteries dry on the aux-battery side, and to avoid the starter overloading the cable between the two batteries

- you don't want to risk an empty aux battery charging itself on the main battery and engine with more current than the cable supports, hence the fuse

- you do want to be able to connect both batteries in a scenario where you accidentally drained the main battery (e.g. a light left on) to "self-start"

[h2odragon]

All very good point, tx. I liked diodes as well as fuses, especially on the inputs of capacitors. Automotive power is never clean.

[RetpolineDrama]

> 1A: a series arrangement of batteries at a civilized 48V

And here I am mad that home-storage server rack batteries are all 48V it seems, but for the same reasons (huge 400+ amp cables required to get decent wattages). When each car charger can do ~14.4kw you need a lot of fat cables running to battery banks

[numpad0]

That's because 48-50V is threshold for "High Voltage", what THAT sign actually signifies, and licenses are required!

[stephen_g]

Technically not, actually. By the IEC standard, 50V (RMS AC) is about the highest you can go for 'Extra-low voltage'. IEC actually has DC up to 120 V as ELV but often regional standards will have a lower value (like EU seems to be 75V and Australia/New Zealand at 60V)

Only above 1000V RMS AC or 1500 V DC is considered 'high voltage'.

That's why a correct sticker on domestic equipment will usually use the wording 'Dangerous voltage' or 'Hazardous voltage' or the like, not 'high voltage'. An actual correctly placed 'Danger - High Voltage' sign tends to mean something like 11kV AC