yes, it can be done, quad zeon servers have these, redundant psu's, you can take a look here:http://www.procooling.com/articles/html/sharing_the_load_between_psu_s1.php
I believe it would be more cost effective to just buy one higher power PC supply.
The outputs of most power supplies are regulated voltage sources. These voltages can vary with temperature, but usually not much. Modern SMPSs use precision voltage references and within the same lot (manufacturing process run) they are usually close in value. However, from lot to lot they will usually vary to much to allow power supplies to share current when paralleled. Another potential problem with some SMPSs, is that if you remove the load or lighten the load on one output, this can cause other output voltages to change value. This depends on the design.
The outputs of most power supplies are regulated voltage sources. These voltages can vary with temperature, but usually not much. Modern SMPSs use precision voltage references and within the same lot (manufacturing process run) they are usually close in value. However, from lot to lot they will usually vary to much to allow power supplies to share current when paralleled. Another potential problem with some SMPSs, is that if you remove the load or lighten the load on one output, this can cause other output voltages to change value. This depends on the design.
Connecting AT or ATX SMPS in paralell is a bad idea since there will be almost no current sharing
This means that one supply may be sourcing all the current of one rail until it enters current limiting allowing other units to start sourcing current
This phenomena may happen for different supplies at different outputs, ie: one unit may be sourcing all the +3.3V and +12V current, and the other may be sourcing all the +5V current, but this is almost random because it depends on the exact output voltages of each unit [5-10% tolerance]
+12V sharing may be easier since most +12V ouputs are only regulated indirectly through the +5V output, relying on the cross-coupling provided by the coupled buck inductor
To get proper current sharing, units should be modified to work in current control mode and the current control signal should be linked on all units, but the fact of having various outputs cross-regulated makes this hard to achieve [peak/average current sensing on the primary plus some extra op-amps and the right frequency compensation for stability could be enough to do the trick]
This means that one supply may be sourcing all the current of one rail until it enters current limiting allowing other units to start sourcing current
This phenomena may happen for different supplies at different outputs, ie: one unit may be sourcing all the +3.3V and +12V current, and the other may be sourcing all the +5V current, but this is almost random because it depends on the exact output voltages of each unit [5-10% tolerance]
+12V sharing may be easier since most +12V ouputs are only regulated indirectly through the +5V output, relying on the cross-coupling provided by the coupled buck inductor
To get proper current sharing, units should be modified to work in current control mode and the current control signal should be linked on all units, but the fact of having various outputs cross-regulated makes this hard to achieve [peak/average current sensing on the primary plus some extra op-amps and the right frequency compensation for stability could be enough to do the trick]
Hi Eva!
I agree that normally they will not share the currents
in an equal way. But who cares?
As long as you have units that can only source, you have a
good chance to get it running in acceptable manner.
As long as one unit can deliver the required current, it does
not hurt that the second does not deliver anything.
As soon as the load exceeds the level which can be delivered by the first, the voltage will drop a little bit and the second unit starts to
bring its part.
If you you want more symmetric current sharing, then it might
help to parallel the outputs via resistors.
My concerns would more be the following:
-No load stableness of the used smps?
-Tendency for ringing?
-What happens if only one unit is powered up?
All this may depend very much on the specific design of the
used smps. Modern flybacks might be able to handle all
this....
Good luck to hugefathead.
Markus
I agree that normally they will not share the currents
in an equal way. But who cares?
As long as you have units that can only source, you have a
good chance to get it running in acceptable manner.
As long as one unit can deliver the required current, it does
not hurt that the second does not deliver anything.
As soon as the load exceeds the level which can be delivered by the first, the voltage will drop a little bit and the second unit starts to
bring its part.
If you you want more symmetric current sharing, then it might
help to parallel the outputs via resistors.
My concerns would more be the following:
-No load stableness of the used smps?
-Tendency for ringing?
-What happens if only one unit is powered up?
All this may depend very much on the specific design of the
used smps. Modern flybacks might be able to handle all
this....
Good luck to hugefathead.
Markus
Most units lose their voltage control-loop stability and operate in discontinuous mode when current-limiting is triggered, some will even shut down after a second or two of current limiting
Working in discontinuous mode means huge ripple on the output, loss of cross-regulation, higher stress to the switching transistors and lots of audible noise coming from magnetic components
PC power supplies are definitely not designed to work in current limiting mode
In the other hand, current is sensed only at the primary for AT and ATX units, so a 200W unit may start limiting at 250W or more *total* power. If this unit is delivering all the +5V current, this means you will need more than 50A to activate the current limiting. Output inductor or diodes will blow in these conditions
Working in discontinuous mode means huge ripple on the output, loss of cross-regulation, higher stress to the switching transistors and lots of audible noise coming from magnetic components
PC power supplies are definitely not designed to work in current limiting mode
In the other hand, current is sensed only at the primary for AT and ATX units, so a 200W unit may start limiting at 250W or more *total* power. If this unit is delivering all the +5V current, this means you will need more than 50A to activate the current limiting. Output inductor or diodes will blow in these conditions
It would be possible to parallel almost any regulated power supply if you isolate each output with a series diode. Obviously you'll lose a diode drop in voltage. The amount of current sharing you get depends on how close the regulated voltages are to each other and how much compliance there is in the diodes.
Without isolation diodes, the higher voltage unit will prevail and the others will shut down. Whether any damage results from this depends on the design.
Without isolation diodes, the higher voltage unit will prevail and the others will shut down. Whether any damage results from this depends on the design.
Using diodes the higher voltage will also prevail
To get decent current sharing, the power supplies have to be operated in current mode, using the same error-current signal for all units and only one voltage amplifier sensing near the load
Current mode is not so complicated, it uses two error amplifiers. First error-amp senses output voltage, compares it to a reference and generates a error-current signal proportional to the difference. Second error-amp senses output current [usually at the primary side], compares it to the error-current and generates a duty cycle signal.
Each error amplifier has its own frequency compensation and this control method provides better transient response, higher line regulation and lower output impedance [higher load regulation]
As I've said, the current-error signal from the error-voltage amplifier [unique for all units] could be used to fed multiple error-current amps of multiple units and get almost perfect current sharing. Modular stackable industrial SMPS work this way but their prices are prohibitive
EDIT : With the usual voltage-mode control circuit present in most PC supplies, using the same duty cycle signal for all units and only one voltage amplifier may provide some degree of current sharing, maybe enough to get decent results, altough 12V/5V cross-regulation may suffer [It won't work for +3.3V outputs since these have dedicated error-amps and regulation schemes]
To get decent current sharing, the power supplies have to be operated in current mode, using the same error-current signal for all units and only one voltage amplifier sensing near the load
Current mode is not so complicated, it uses two error amplifiers. First error-amp senses output voltage, compares it to a reference and generates a error-current signal proportional to the difference. Second error-amp senses output current [usually at the primary side], compares it to the error-current and generates a duty cycle signal.
Each error amplifier has its own frequency compensation and this control method provides better transient response, higher line regulation and lower output impedance [higher load regulation]
As I've said, the current-error signal from the error-voltage amplifier [unique for all units] could be used to fed multiple error-current amps of multiple units and get almost perfect current sharing. Modular stackable industrial SMPS work this way but their prices are prohibitive
EDIT : With the usual voltage-mode control circuit present in most PC supplies, using the same duty cycle signal for all units and only one voltage amplifier may provide some degree of current sharing, maybe enough to get decent results, altough 12V/5V cross-regulation may suffer [It won't work for +3.3V outputs since these have dedicated error-amps and regulation schemes]
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.