Now let's talk technical. Details.
What do we need to power our sub 200W machine?
Say we settled on ITX form factor.
We'd then need a DC-ATX psu. This is non-negotiable, unless you want to shove an entire UPS into your case.
Okay, a DC-ATX capable of 200W. What input then? We've got 12V, 19V and wide-input: 16-24V, 6-30V.
Let's omit 19V - cause then you're aiming for a specific voltage without any benefit other than accepting common laptop adaptors.
Let's also omit 6-30V - too wide, designed for rough environment like boat or car which assumes it gets connected to a generator. Too expensive for our needs.
So the choice is 12V or wide-input 16-24V
Both have merits:
Wide-input means your battery can just plug right in without regulation, as long as its voltage is in range.
12V needs regulation at input, but also means you are not constrained by battery voltage and your brick voltage, so long as your final regulation module can handle it. This means more choice in choosing packs, charging circuit and input power supply.
For batteries, we have a lot: UPSes uses lead acid, laptops use LMO (actually LiMnCO[sup]2[/sup], multiple similar chemistries exist) cells, later on they move to lipo (Li-ion Polymer, or as I'd like to remember it, li-ion in POuch) cells. RCs, drones and e-bikes, hoverboards use LiFePO[sup]4[/sup] cells. There's also NiMh, NiCad...
We are omitting obvious offenders against sff - lead acid for weight and NiMh/NiCad for volume.
Maybe we can omit lipo as well? Unless in pack form, it's usually made to shape and capacity as ordered by the big players unreachable to us.
Now let's leave these choices first, promise we'll get back to it..
The heart of every battery backup system is what you call "power path management" IC or module. For lead acid batteries, supply, load and charge can all be the same, parallel to the battery terminals so this can be omitted.
But in the case of Lithium, prolonged trickle charging is not good for longevity and safety, so a proper lithium battery backup system should charge the battery when needed and stop when it should, while maintaining seamless switchover to battery when mains fail or vice versa when mains returns.
You can, choose not to implement this and just use swappable packs and charge them separately.
For power path management, I found some ready made solutions:
OpenUPS and friends by mini-box.com
> all-included-solution limited to 120W, expensive, and rightly so - it's even got battery status reporting to windows via usb.
UPS-0528-11
> limited to 15A input/output, no output regulation (direct battery output or mains straight), no undervoltage and low battery start protection. Can be added though via daughterboard.
UPS-1228-12
> pretty much everything above, plus output regulation. Output is limited to 8.3A (at 12V) though.
DIY method with mini-box's Y-PWR, separate charging circuit, buck/boost circuits etc.
The choice is easy, I want as close to 200W as possible, so UPS-0528-11 it is.
The implication? Well with no regulation on output we either need a regulator or a wide-input dc-atx. Wide input 16 to 24V it is.
Now that we've decide on our output let's choose our battery chemistry and arrangement.
Battery cells can be arranged in series, parallel or both - pack. The resulting pack's voltage is the sum of the voltage of cells arranged in series, while its capacity is the sum of the capacity of cells arranged in parallel.
First we'll determine how many cells in series we need to be in the range of our dc-atx's input.
Say we go with LiFePO[sup]4[/sup], with nominal 3.2V, range 2.55V to 3.6V.
One configuration would be 6 cells in series, because then the entire pack's range would be 6*2.55 = 15.3V to 6*3.6 = 21.6V.
In fact the pack would never come below 16V, since our pico would shut down when it happened. So each cell will only reach 16Ć·6 = 2.67V.
7S (7 cells in series) would be 17.85 to 25.2V which means individual cells go from 2.55 to 3.43V (24Ć·7).
Both are good contender: 6S meaning we don't need to worry about undervoltage protection, 7S means we never charge them to 100%, which is beneficial for cell longevity.
Following the same math, for LMO cells, with nominal 3.7V, 3 - 4.2V:
5S: 16 - 21V, @3.2 - 4.2V
6S: 18 - 24V, @3 - 4V
For LFP summaried from above:
6S: 16 - 21.6V, @2.67 - 3.6V
7S: 17.85 - 24 , @2.55 - 3.43V
If we are after pack longevity, we'd choose configuration which limits our upper bound. But since (1) our chosen power path manager doesn't have UVP, and (2) its charging circuits requires our supply to be at least 1V higher than pack's maximum voltage, for simplicity's sake we are limited to 5S for LMO and 6S for LFP.
Now if you notice this means we never use the whole capacity of the cells. If the goal was to use every coulomb of charge in the cells, this is not the way to go.
But since lithium batteries are expensive to replace, this is a trade off that even Tesla is making. Idk how true they are, but online articles say Tesla only runs their cells at 70% capacity for its EVs.
And now, how do we decide how many to parallel?
We start at our needs. Maximum load is 200W but safe to say we won't design a pc without wiggle room for peaks and such, so let's settle on ~176W typical max load. That means 8A at 22V, all the way to 11A at 16V.
Yes I arbitrarily choose that number cause it gives nice, even results.
We can now talk about C-rating.
Batteries is rated by how many amperes they can provide in an hour. For bigger batteries it is simply Ah, smaller ones use mAh. 1C, for a certain cell, means a current flowing at a rate of this number for that cell. Different battery chemistry has different recommended rate of charge and discharge.
We'd actually need to consult the datasheet to find out this ratings, but generalization does exist for similar chemistry of batteries.
For most lithium batteries, 1C rate of charge is considered quick charge. Typical is 0.2 to 0.5C for cell longevity.
For LMO, discharge is typically 1C, max 2C.
LFP can go all the way to 25C. This is why they are used in RCs and stuff.
Looking at these, seems like LFP is the way to go isn't it? Well I do think so too, the fact that it's more stable is a bonus.
What's the catch then? Energy density. LFP still lags behind LMO for capacity per volume, a.k.a specific energy. More self-discharge and a little higher cost.
For example, LFP in 18650 form factor is currently limited to 1600mAh. Contrast this to current king of 18650 at 3400mAh.
Good, relatively cheap 18650 (a moniker for its size: 18 mm diameter and 650mm height) can be found in ~2600mAh variants.
This is important, because as I calculated above, we'd need the parallel cells to provide up to 11A. Say we run them at max 2C, this means we want 5.5Ah C rating. While LFP can prob handle that with one cell (flat in 10 mins maybe), LMO needs at least 3 cells in parallel. 3 parallel 2600mAh LMO can provide typical 2.6A*3 = 7.8A and max double that at 15.6A.
5 in Series, 3 in Parallel, 5S3P, 15 cells of 18650 LMO. So there we go. A quick rundown to how I came to this numbers.
While doing this, I realised why noone's doing it in bigger scale.
A properly designed battery backup system have to make a lot of trade offs. A reasonably small system needs to be carefully calculated for it can only serve an equally small range of hardware and uses.
If you are not constrained by weight and volume, UPS is still king. How do you beat a 12Ah 12V battery if you don't mind a 2L box weighing 4kg?
But everyone's got to start somewhere ^^
