EEVblog #504 - UPS Tutorial & Teardown

[EEVblog]

Dave explains the three main methods of Uninterruptible Power Supply design - Offline, Online, and Line Interactive. And then opens an APC 2200XL rack mount UPS to see how well it matches the classic block diagram theory. Will there be any surprises?

Schematic: http://eevblog.com/files/apc_smart-ups_su2200_3000.pdf

Patent: Method and apparatus for providing battery charging in a backup power system
http://www.google.com/patents/US5302858

[NickS]

Wow you managed to find the schematic? I wouldn't have thought APC would have released them.

I've always been curious how they get the 'pure' sine wave. Is it always high frequency PWM filtered and smoothed?

[EEVblog]

Wow you managed to find the schematic? I wouldn't have thought APC would have released them.

Someone on Youtube found it. I didn't find it on first look, would have been handy for the teardown.

I've always been curious how they get the 'pure' sine wave. Is it always high frequency PWM filtered and smoothed?

Usually, yes. The parasitics in the transformer do a lot of the filtering of those higher frequencies.

[komet]

The schematic isn't of the unit you tore down, although clearly similar.

APC have been reusing their product numbers for years.

[Wytnucls]

I have one of the popular Back-UPS RS 500VA, which was sold with a design fault in the battery charging circuit.
In case someone else was affected by the same problem, here is the repair procedure:
http://www.heime.org/Lists/Posts/Post.aspx?ID=17

http://www.apc.com/resource/include/techspec_index.cfm?base_sku=br500i

[moemoe]

what is this? some idiots embedding the youtube video inside their own page and linking to it on youtube with a short trailer.

[madires]

Typically two different modules are inserted into the module slot. One is a port expander with three serial ports. IIRC those include some switched output pins for dumb devices and a 3-wire RS232. The other module is a SNMP agent with ethernet port for your network management.

[Phoenix]

There is a bit more going on here than at first glance me thinks. I'll cover some of my initial thoughts.

I'm pretty sure it has a low PWM frequency. The steel core transformers will not only have poor magnetic performance at high frequencies, but the interwinding capacitance will also cause problems for power quality and EMI. They also must have substantial leakage inductance for the boost active rectifer to operate properly (and/or they are driven well into saturation). This means the leakage inductance AND the 2x4.7uF big blue caps form the bulk of the output filter. The small (yes small) single turn common mode choke also suggests the switching frequency is low as otherwise there would be substantially more high frequency common mode that would need rejection.

During the charging stage the H-bridge essentially acts as two anti-parallel boost converters. One operates in the positive AC half-cycle, the other operates in the negative half-cycle. It can't be a dumb diode bridge as it needs boost functionality as the DC battery voltage MUST be greater than the peak sinusoidal voltage. This is actually very similar in practice to a boost PFC topology (but most importantly 4-quadrant/bi-directional operation). I can provide more detail here, but it would require effort and pictures  .

The two CTs are also a little odd - very small and clearly not measuring any high power/current circuitry. CT1 measures the capacitor current - the only thing that comes to my mind is that it is being used as part of the feedback control to actively damp the LC resonance of the leakage inductance + filter capacitance and/or helps with high speed control feedback as voltage feedback alone is quite slow (when talking about recovering from a transient event like load change). It does feed into an IC along side a measurement of the output AC voltage and sinusoidal reference which obviously is the feedback controller. CT2 though has me beat - it is a short circuit primary turn, kinda just branching off the main output power trace... (anyone?)

As for all those relays - the first relay provides isolation for mains and neutral to prevent backfeed into the grid. The second relay is the second isolation relay (I believe required by the standards) and has weld open/shut detection (again, standards). No second relay or weld detection on the neutral though, mustn't be required. The next 2 relays appear to be transformer tap selection/bypass/backfeed - need to know the transformer details to work that out properly. The final relay looks like a simple AC output enable.

The schematic in the OP appears quite accurate. Even many of the header names are matching.

Cheers
Stewart

[envisionelec]

CT2 though has me beat - it is a short circuit primary turn, kinda just branching off the main output power trace... (anyone?)
Cheers
Stewart

Zero-Cross Detection?

Also of note are the Schmitt Inverters at the H-Bridge gate drive being used as propagation delay matching elements (note the Vdd Net Name: 12V-DLY (delay)); Presumably to counteract the unequal timing encountered in the ASIC or to reduce cross-conduction.

[rollatorwieltje]

This is what happens when something goes wrong:


This was a slightly older 3000VA model. It doesn't like moisture

[SeanB]

This is fried silicon......... Some had a hole blown in the top of the steel cases. With a 48V lead acid battery and 55000uF of smoothing there was enough energy. Was around 100kg of scrap steel and copper at the scrapyard.

[ConnorGames]

You kept mentioning P-channel MOSFETS. It seems as if it would make sense to use N-channel for the high side too with this kind of power. Did they really use P-channel? It seems the schematic you linked uses all N-channel MOSFETS.

[envisionelec]

You kept mentioning P-channel MOSFETS. It seems as if it would make sense to use N-channel for the high side too with this kind of power. Did they really use P-channel? It seems the schematic you linked uses all N-channel MOSFETS.

That's correct. With a high side driver, it is correct to use N Channel devices. P Channel suffer from too many drawbacks to be a good fit for this function.

Dave also mentions that bad placement of the filter caps because of heat, but the reason they're placed close to the MosFETs is to reduce dv/dt due to trace inductance. Such is life! The fan will move plenty of air for the duty cycle.

[Monkeh]

You mention that it's not isolated, as it's line interactive. This may be true of the model you have, however, it is not true of all units.

[EEVblog]

You mention that it's not isolated, as it's line interactive. This may be true of the model you have, however, it is not true of all units.

Of course, nothing I said is 100% true in all circumstances for all model, there are countless variations.

[EEVblog]

You kept mentioning P-channel MOSFETS. It seems as if it would make sense to use N-channel for the high side too with this kind of power. Did they really use P-channel? It seems the schematic you linked uses all N-channel MOSFETS.

According to the schematic, yes, it's all N-channel.

[Computeruser]

I have a nice APC 750 XL in the basement. This thread reminds me that I should get replacement batteries for it soon.

It runs my network (4 devices) and keeps my desktop computer running smoothly.

It is a long way from my model railroad and the output is not part of the house wiring, but I keep dreaming about putting the DCC (Digital Command Control) power supply and DCC distribution network on UPS.

... C

[SeanB]

Try Esrviceinfo.com for it, they have a lot of APC UPS diagrams there.

[RupertGo]

A long time ago, I ran a comparative review project for UPSs for PC Magazine. We got in everything on the market, and subjected them to a serious battery (sorry) of tests - including quality of output waveform, power factor issues, injected transient resilience and efficiency goodness. Hard to believe that we did all that for a mainstream newsstand publication, but we could afford a proper lab with proper test equipment and the time to go to the real experts to get the methodology right. What's worse is that we came to the conclusion that the UK electricity supply was so reliable, you'd really have to spend serious money to get higher reliability through UPS than just not bothering (cheap UPSs fail more often than the mains does).

However, something we did learn was that UPS design back then was seriously old-fashioned. APC was the market leader, but even then (in the mid 90s) it was basically 70s ideas with a few more modern bits slapped on top for monitoring and network integration. Given that high speed switching was available and DSP could do a lot of the heavy lifting, it was a bit disappointing to find quite so much heavy iron and old thinking. From what I've read about solar power even today, the lack of clever design at this end of the market is a limiting factor.

[crisr]

I have a couple of APC Back-UPS RS 1500 that both developed the same fault: after a power fault they do not detect when AC comes back on - if I disconnect them from mains they begin supplying AC from the batteries, but if I reconnect the mains they stay on inverter mode, until the batteries go low and they automatically switch off; only then they "see" that there is AC input and turn back on supplying power from mains... strange behavior, so far I did not have the patience to open them up and check. Anyone knows something about that with this model or similar ones? Any suggestion on where the problem might be or if it would be easily correctable?

[tesla500]

Here's an in depth look at these true sine wave UPSs that I did awhile back:


And an overview of the UPS:

[NickS]

If the difference between modified sine and 'true' sine is just driving the H-bridge with PWM, why are the true sine wave converters so much more expensive?

Also wouldn't that make it technically possible to modify a modified sine inverter to true sine just with a little microcontroller?
Get the micro to take the modified sine pulses for synchronisation and the code would make a sine ramp up/down for the PWM outputs.

Not possible on these types that use the mosfet diodes to charge the batteries of course, but cheaper ones are usually simpler anyway.

[NiHaoMike]

If the difference between modified sine and 'true' sine is just driving the H-bridge with PWM, why are the true sine wave converters so much more expensive?

Also wouldn't that make it technically possible to modify a modified sine inverter to true sine just with a little microcontroller?
Get the micro to take the modified sine pulses for synchronisation and the code would make a sine ramp up/down for the PWM outputs.

The gate driver needs to switch faster (source/sink more current) in order to get a good efficiency with PWM. There is RL-PDM (Rebecca Liu Pulse Density Modulation) that synthesizes a sine wave with less transitions per cycle. But as it turns out, most electronics are perfectly happy running from square waves (actually, square waves mean less peak current and the rectifier will run cooler) and motors don't really mind running from square waves apart from more vibration and noise. The sharp rise/fall times can cause EMI and other issues, but a filter (even more important with PWM or RL-PDM) takes care of that.

Even really large inverter motors like electric car motors are driven with square waves at high speed since it's more efficient.
http://techno-fandom.org/~hobbit/cars/ginv/VH.html

I know from some other observations that the inverter drive signals for MG2 move from a multiple-kilohertz PWM regime into simply switching 3-phase square waves at the motor's native electrical rotation speed, because it's more efficient and the motor is turning fast enough to smooth out any torque ripple that would produce.  But overall applied motor current can still be regulated smoothly!  How?  By using a variable boost voltage.  In this speed range I see VH rising and falling corresponding to my go-pedal demand, with its lowest baseline creeping up a bit as I head toward 40 MPH and MG2's own peak output rises sufficiently.  It's almost like having the switching behave like a brush commutator, simply leading the electrical rotation angle by 60 or more degrees, and regulating motor speed via applied voltage like it was a big ol' toy-train rheostat.

What I would like to see is an off the shelf 12V input inverter that has built in V/Hz control. It can dial back to reduce power usage of motors as well as soft start them.

[Cabwood]

OK Dave. I never doubted that you were a good electronic engineer, but your call on that APC UPS battery charging system was pretty damn inspired. Just from the absent charging circuit you deduced that it must be charging "backwards" through the inverter H-bridge. And just to rub it in you went and found the patent showing that you were right.

I am super-impressed.

[mfeinstein]

Hey Dave, 

I think I can add some information about the HUGE capacitors you found inside the UPS... I have been involved in motor controllers for combat robot in the past years, and motor controllers and UPS usually look much very alike (Battery + H-Bridge + Inductive Load).

What happens is, when there is a power failure and the battery switches on, the load at which the unit is connected might need a huge power burst, and batteries are usually horrible for delivering high amounts of current in a short time (this is measured by the batteries C rate), and capacitors are very good at it...

...BUT the real problem in high power applications is the stray inductance of the connections, wire, PCBs etc.. since there is a H-Bridge switching the battery's voltage, the turn-on and turn-off time of the MOSFETs are really short, dealing with high ammounts of current at the same time... this will cause high voltage spikes in the battery that can really damage it AND completely destroy the MOSFETs. The capacitors are there to smooth and absorb the voltage spikes in the MOSFETs and in the battery, preventing a total destruction of the system due to the inductive load of the transformer.

(begginers, remeber, V = L di/dt, so high current being switched at low time will generate a really high voltage spike! And most MOSFETs in the market wont survive this spikes)

As to the battery charging patent, this is very much the same as vehicles use as "regenerative braking", where the vehicle uses its own movement, due to inertia or some external force, like descending a hill, to charge the batteries using the internal diodes of the MOSFETs (usually there are dedicated schottky diodes as well, since this MOSFET diodes have high voltage drops in the order of 1.5V and heat a lot the MOSFET).

After years learning in the EEVBlog, It feels great to be able to contribute a bit too