Higher power POE standards are presently being defined
This article is a collection of power supply circuits that explain how to draw power from two or more Ethernet ports. An overview of each circuit is provided as well as some of the design issues faced with each implementation.
By Brian King and Robert Kollman, Texas Instruments
Power over Ethernet (PoE) has become a popular concept and is being used in products such as Internet telephones, security cameras and point-of-sale terminals. Power over Ethernet distributes power through an Ethernet connection. In a network providing POE, power is supplied by the power source equipment (PSE), which generates a 44-57V output on the Ethernet connection. At the other end of the Ethernet connection, power is consumed by the powered device (PD). Although higher power POE standards are presently being defined, the power available to the PD currently is limited to about 13W on a single Ethernet connection. Unfortunately, this often is not enough power for complex applications. Consequently, some high power PD designs need to convert power from multiple ports into usable voltages with galvanic isolation from the 48V input. Several techniques exist for providing isolated power conversion from multiple input sources.
One technique for paralleling DC/DC power supplies commonly is referred to as the droop method. Paralleled power supplies will share current if their output voltage decreases with increasing load currents. This requires no communication between the supplies and eliminates potential single fault failures. Minimal additional parts are needed to implement this technique. If current mode control is used, you can simply limit the DC gain of the control loop to introduce output voltage droop that is proportional to load current. If more accuracy is desired, the circuit can be implemented as shown in Figure 1. This circuit measures the output current with differential amplifier, U1B, and injects an error into the regulation loop at the compensation amplifier, U1A. Only a few resistors and a single amplifier need to be added to get autonomous current sharing.
Unfortunately, droop sharing is not very accurate. Figure 2 shows worst case variations with one percent resistor tolerances, 1.5 percent reference tolerance and 10 percent total droop. The design has a nominal set point of 5V and a variation of ±5 percent droop. The minimum and maximum curves show component tolerances at their extremes. If you put these three power supplies in parallel, at no load, the supply with the highest output tends to regulate the output voltage. If the supplies used diode regulation as in Figure 1, the supply with the lowest output voltage will not output any current. As the load current increases, the output voltage starts to fall. The supply with the highest output voltage sources all the current until its output falls to 5.25V. Then the supply with the second highest output starts to source current. With this set of assumed worst case tolerances, the first supply provides almost 70 percent of its output power before the supply with the lowest output voltage starts to contribute. This is undesirable because it is unreliable; however, in some instances it could be acceptable. As the load current is further increased, the first supply may go into current limit. Further current increases are handled by the two remaining supplies, thus allowing full-rated power operation.
Power supply topologies with synchronous rectification allow the power supply to either source or sink output current, which is extremely problematic for this control scheme. In the extreme case, one supply may try to regulate to the high end and the other at the low end. When this happens at no load, some supplies will source current to the output while others will sink current from the output. This pulls power from one source and feeds it back to the second with no power delivered to the load. For this reason, disabling the synchronous rectifiers at zero amps is recommended.
Interleaving provides another technique for balancing the power drawn from multiple inputs. Just like the droop method, interleaving uses a separate power stage for each input and supplies power to a common output. Unlike the droop method, the interleaved power stages, also referred to as phases, share a common single primary side controller. This provides a reduction in cost, and allows each power stage to be synchronized out of phase. Synchronizing reduces the ripple current in the output capacitors, and results in a smaller output filter. Interleaving requires all power inputs to share a common return, which may prohibit this approach from being used in some applications.
Many PWM controllers are designed specifically for interleaving. If only two phases are required, significant cost savings can be realized by using a push-pull controller to perform the interleaving. Figure 3 shows a schematic for a two-phase interleaved flyback supply using a push-pull controller like the UCC2808. This chip limits the duty cycle of each phase to fifty percent and switches the two power stages 180 degrees out of phase. This push-pull controller uses peak-current mode control, which keeps the peak current of both phases close to the same value. In a discontinuous flyback, the output power (per phase) is proportional to the square of the peak primary current. As a result, the power drawn is naturally balanced from the two inputs. This technique equalizes the power drawn from the two input sources to less than five percent. Switching delays on the primary MOSFETs are the main source of imbalance, which is worst when the two input voltages are unequal. The peak current limit provided by the controller limits the maximum power drawn from either input, and the duty cycle clamp limits the input current during undervoltage and fault conditions.
Power Share Using a Secondary Side Load Share Controller
A third method for sharing power among multiple inputs is provided by a secondary side load-sharing ICs. Using this method, any number of independent power supplies with power supplies with remote sensing capability can share a common output. Load share ICs are commonly used with power supply modules. An example is shown in Figure 4. A shunt resistor is used to measure the current supplied by each converter. Due to tolerances and parasitic impedances, one of the power supplies will deliver more current than the others. This supply acts as the master and will set the voltage on the load share (LS) bus. Slave units use this load share bus voltage as an input reference to control their output currents. Slave units are adjusted by injecting a voltage on the remote sense leads of the slave converters. This allows the master to control the output voltage to the load, which ensures good load regulation. This master/slave approach results in very good current sharing accuracy, which typically is better than three percent at full load.
Since one load share controller plus external discrete components are required for each paralleled supply, component count and cost with this method are slightly higher versus the droop or interleaved approach. Additionally, load share controllers are not recommended for use with synchronous rectifiers, which may cause problems during startup or when adding or removing individual supplies.
Master/Slave Isolated Primary-Side Current Share
Another technique that can be used to parallel power supplies is to sense the primary current of one (master) and compare it to another (slave). Using either optocouplers or current transformers provides a means of communicating current information between supplies while maintaining isolation. Current transformers represent the best choice because good performance can be achieved at the lowest price. In addition, current transformers have good accuracy when compared to optocouplers. Their accuracy is set by the turns ratio tolerance, which is better than two percent, and resistor tolerance, which is typically one percent. Optocouplers rely on the tolerance of their current transfer ratio, which is 30 percent at the very best.
Table 1 shows a comparison of the four load share methods. The droop method is the simplest and one of the cheapest methods, but results in the worst performance. It is also tolerant to single failures. Usually, the best performing technique, the load share controller, results in the most expensive solution. Using either the interleaved primary controllers or the optocoupler/current transformer technique provides a compromise between cost and performance. Additional factors, such as the use of synchronous rectifiers, the number of PoE inputs, and whether the PoE inputs must be isolated from each other need to be considered before selecting an approach. Using the appropriate technique for your application will ensure that you get the maximum power from your PoE.