For the PDF version of this article, click here.
Redundant power-conversion designs using multiple converters often require current sharing to both distribute the losses and maintain voltage regulation through a fault. Several popular methods of current sharing include master/slave, droop and democratic sharing. Democratic current sharing offers special benefits for redundant systems, particularly for high-current, low-voltage designs in which voltage regulation is required to be several percent of the regulated voltage.
If the control loop and method of disconnecting a faulty converter are optimally designed, a fault condition should appear as a simple load transient condition. This current step transient appears and the current in the faulty converter must suddenly now be shared among the remaining converters. Specifically, each operating converter sees an increase of current from Iout/(N+1) to Iout/N or Iout/[(N+1)*N].
Of all the current-sharing techniques, droop sharing is the simplest to implement. Droop sharing does not require a current-sharing “bus” and works by decreasing output voltage while increasing output current. The common connection point at the load determines the shared voltage, and a load line determines the current of each converter.
One advantage of droop sharing is its robust sharing stability, allowing it to work well over a wide set of operating conditions. It also has a unique property called “adaptive voltage positioning” that can improve the transient response from a large current step. However, in order for accurate current sharing, the droop method requires a precise voltage setpoint and droop slope. Moreover, if tight voltage regulation is required, this method might not be acceptable since the output voltage will vary with current.
A slightly more complicated sharing technique is master/slave. Master/slave sharing is accomplished by first electing a unit to be master. Typically, the unit with the highest output voltage becomes the master. This type of selection can be accomplished by combining the feedback signals of the units using diodes, active circuits or open drain connections. While the master continues to regulate the output voltage, the slaves move into a constant-current mode of operation. The master broadcasts its output current on a current-sharing bus, and the slaves attempt to match that current level.
For current-mode control converters, this share bus is simply the output of the master's voltage-feedback loop. This situation can make the master/slave sharing method very susceptible to noise. This is because the master/slave method requires the master to control an internal feedback node of units that might be several inches away.
Another drawback to the master/slave control is integrator windup. For redundant systems with tight regulation requirements, the failure of the bus master requires one of the slave controls to come out of control-loop saturation to take control of the sharing bus. The time lag for coming out of saturation creates a transient that generally exceeds the optimal sharing potential of Iout/[(N+1)*N] transient response.
Democratic Current-sharing Advantages
Of all the methods discussed here, democratic current sharing can be the most difficult to implement but it has several advantages. Democratic current sharing works by having each converter compare its current to the average of all units. The current-share line is used to broadcast this average current.
This method is symmetric in the sense that each unit contributes and cooperates equally. If a unit's current differs from the average current, it will adjust its output voltage setpoint accordingly. For example, it will raise its voltage if it needs to contribute more current. This method does not suffer from the drawback of the droop method; its output voltage does not vary. In fact, the resultant output voltage simply becomes the average of the different converters' voltage references (see “Derivation of Output Voltage”).
Since all controllers are always active and symmetric, there is no integrator windup problem as in the master/slave method. Furthermore, it doesn't have the noise issues that the master/slave method exhibits. This is because the democratic current-sharing method has two separate control loops — one for voltage and one for sharing. And since the sharing control loop can be a relatively low bandwidth, it tends not to respond to high-frequency noise on the current-share bus.
One difficulty of this type of control is that it requires more system analysis to guarantee loop stability. At a high level, the control design is performed in two steps. First, the voltage loop is optimized for transient response. Next, the resultant closed-loop output impedance is then fed into the current-sharing loop analysis. The current-sharing control loop consists of current-sense resistor, the current-share gain, the voltage loop response to a setpoint change, and finally, the output-current change derived from the closed-loop output impedance.
With a proper design, the democratic current technique can approach an optimal response. For example, we have demonstrated 3+1 redundancy at 1.5 V and 300 A using democratic sharing methods that allowed us to meet very tight regulation requirements with peak fault transients of less than 50 mV.
Derivation of Output Voltage
Four-way bidirectional democratic current sharing results in an output voltage that is the average of the converters' voltage references.
Where Vsense is the load voltage/remote sense voltage, Vref1, Vref2, Vref3 and Vref4 are the internal reference voltages of the converters. Gshare is the gain of the current share circuit. I1, I2, I3 and I4 are the converter currents, and Iave = (I1 + I2 + I3 + I4)/4.
Solving these equations results in: Vsense = (Vref1 + Vref2 + Vref3 + Vref4)/4.
So, the output voltage is the average of the reference voltages.