Does anyone know what the final stage of a switiching PC power supply looks like? I'm trying to create a redundant power supply using two (or three) ordindary ATX PC power supplies. Just combining the outputs seems to work. I tried adding a failed supply + a good one and it still worked fine (accomplishing the purpose). However, I'm not sure of all the failures modes. Does anyone have a schematic for a PC power supply or know any reason why this wouldn't work? Are the enterprise grade hot swappable redundant power supplies really any different in the final stage? Any insight appreciated.

 

Some research here may help you in your quest:
http://www.smps.com/

There are also circuits that are usually used to switch over to battery power when mains power is lost and vice versa.. You could use the same type of circuit to switch to an "active" supply once one of the 3v3, 5v +-12v supplies drop below spec.

A simple zener based circuit can do this quite effectively.

 

The output is usually filters, diodes and caps.

Problems can occur if you connect one supply to another. Say one supply is at 12.1V and the other is at 12.2V, and you tie them together. Assuming the wire has a resistance of 10 mΩ (milli-ohms), you now have 10 amps flowing from one PSU to another. Think of what troubles that might cause.

One way you could get this to work is by using diode OR gates. Use low forward drop schottky diodes and tie all outputs through the diodes. But, the output will be 0.3V less. And you will need to heatsink the diodes.

It will usually be ok to parallel supplies as long as outputs are not connected together. You can tie together the common rails. The way this usually works is you might have one PSU for the HDDs and graphics card, and one for the motherboard and peripherals. However, this is not a redundant supply.

 

The output is usually filters, diodes and caps.

Problems can occur if you connect one supply to another. Say one supply is at 12.1V and the other is at 12.2V, and you tie them together. Assuming the wire has a resistance of 10 mΩ (milli-ohms), you now have 10 amps flowing from one PSU to another. Think of what troubles that might cause.

.

Why would there be current between the supplies? Wouldn't the voltage from the higher voltage supply simply charge the filter on the lower voltage supply to the higher voltage and via the voltage monitor circuit in the lower voltage supply, shut down the switching of the lower voltage supply?? The rectifier diodes would be back biased on the lower voltage supply, wouldn't they? Installing isolation diodes would result in a 0.6 to 0.7 volt reduction in power to the external load, which may or may not be acceptable.

On edit: If one supply failed due to shorted diodes or shorted filter caps, then YES, there would be massive current from the good supply trying to bring the buss voltage up.

 

Does anyone know what the final stage of a switiching PC power supply looks like?

You may find this helpful:


http://www.smpspowersupply.com/atx-power-supply-pfc-schematic.pdf

 

As loosely regulated as some of these are most will balance the voltage out, depending on using them for redundancy would cause a problem if you had a diode for one of the output rails short out as the failure.

 

Thanks for the replies. After some investigation I've decided to build a 'voter' circuit using the PWR_OK output from the three supplies. According to the ATX power supply guidelines (http://www.formfactors.org/developer%5Cspecs%5CATX12V_1_3dg.pdf), the power supply should monitor all of the rails and only assert PWR_OK if they're all above the minimum limits. My 'voter' circuit is simply logic AND gates controlling SSR's for the five rails. The problem I'm having now is that some power supplies have multiple 12V rails, which would require multiple SSR's per rail. Four SSR's per supply was enough, but with multiple rails that could ballon to seven. Here's a good thread on multiple power supply rails:

http://www.jonnyguru.com/forums/showthread.php?t=3990

 

How about some DPDT relays (or more, I think you can get 4PST/4PDT relays.) Only problem I could see is the transition is momentary and might cause a power glitch. If you had some capacitors after the PSU out then you might be okay.

 

How important is the redundancy aspect for you really?

If you are really wanting high availability then you need to do a system appraisal of failure rate, repair time and SPOF.

If the appraisal indicated that you needed to get some more nines into your powering availability then I suggest you should be able to find high spec level PC supplies specifically designed to provide diode isolated outputs, and fail contacts. They would then be used along with independant battery backed UPS, and maintenance alarming with spare parts, and controlled shutdown/handover software (for when the UPS battery autonomy is finished). Doing anything less is imho playing around at the mickey mouse end of redundant powering.

 

The goal is to use ordinary PSU's (yes, they're UPS backed). Server power supplies are noisy (no fanless options), expensive, and not as efficient (only 80+ Bronze units available now are unfavorable, whereas consumer PSU's have already reached 80+ Gold -- which is 90% eff at 1/2-load). Anyway, for those still following the thread I've found *exactly* what I was looking for:

http://focus.ti.com/lit/ug/slvu181/slvu181.pdf

or more in-depth read:

http://www.eetimes.com/design/power...undant-power-techniques-for-servers-explained

The Texas Instruments TPS2410 allows you to diode OR power supplies using a MOSFET, rather than an actual diode. Thus, there's no voltage drop ... but still the simplicity. The IC was designed intentionally for the purpose of making low voltage high current power supplies redundant.

 

For interest:

Are you paralleling 5V rails, and also 12V rails?

Are you powering multiple loads from those rails, and are you A-B powering the loads?

Do the ordinary PSU's have power fail contacts?

 

That's the problem with the higher wattage PC PSU's ... they can have multiple rails per voltage output. Particulary for +12V, which means you need a TPS2410 & MOSFET for each rail of each voltage: +5, +12, -12, +3.3 (I'm planning on using an actual diode for +5VSB). The good news is the TPS2410 and MOSFET only cost around $4 total.

 

The final assembly cost of a batch of TPS2410 dual diode circuits will likely be much more than the price of the PSU's supplying the DC rails.

I guess your 'load' PC motherboard is not set up for A-B powering. Have you tried to find A-B powered 'loads', as that puts SPOFs closer to the final parts using the power (ie. past the power connector on the load, and not subject to the influence of multiple 'loads' if you were using some form of distribution).

 

Not sure I follow your math. The power supplies cost >$150 each. I've breadboarded the setup and its working well. I'll post the schematic and pictures after I finish the board layout for anyone who's following this post (now or in the future).

 

Please do. Im trying to visualize what you are doing.. Since you have it working, Ill just wait for the schematic.

 

Instance:
Did you get your setup running?
I'm very interested in doing something like this.

 

Instance:
Did you get your setup running?
I'm very interested in doing something like this.

After two years, I should hope so.

...but, I don't think Instance is coming back

 

If for a PC, it's much simpler and more stable to buy a system with dual or triple supplies. Then they are hot swappable in the event of failure.

They are also designed for load balancing (in cooperation with motherboard).

I don't think a DIY dual supply would be as stable as a quality single supply over-rated by 50% or so for the load (750W supply for a 500W system/board, for example).

 

Have you seen how much hot swappable PSUs cost?
A shitload.
http://www.directron.com/redundant.html

I usually buy power supplies with 80+ ratings for about $50.

I'm going to build something like this.
Its really not that complicated.

My controller will have hookups using ATX female connectors like this
So hot swapping is easy. Just unplug the PSU that has failed, and plug in another one.

A nice controller can maybe show you what percentage of the time the one PSU is providing power vs the other.
It will be a fun experiment.
Its strange how this is a forum about electronics but people here seem to discourage each other from getting their hands dirty. "Just be a consumer, even though it costs many times more."
There would never be any innovation in this world if everyone thought like that.

 

If it is about a critical system, where uptime is a premium, there isn't room for a hack.

When downtime costs thousands of dollars per minute, the price of a power supply, hot swap RAID, etc. is insignificant.

If you are a casual home user, then go ahead and tinker.