The IEEE 802.3af standard for Power over Ethernet (PoE) introduces a new facet to Ethernet networking. Most additions to the 802.3 standard are in the vein of improving Ethernet's data capabilities. PoE doesn't add to Ethernet's data capabilities, it delivers dc power in tandem with 10/100/1000 Mbps data. PoE brings with it a unique set of problems and new ways to thinking that may be unfamiliar to engineers with experience designing Ethernet equipment.
This article highlights many challenges inherent in PoE and guides the reader over some of the pitfalls, potholes and gaps in the 802.3af standard. Covering these issues leaves little room for background on the 802.3af standard (available at www.ieee.org). Briefly, the PoE link allows a powered device (PD) to draw up to 12.95 W from the power-sourcing equipment (PSE). The PoE link or port is controlled by the PSE, which identifies PDs via detection and classification before powering and monitoring the port (ICUT, ILIM and disconnect).
Much of the burden of PoE rests on the PSE. It must perform detection and disconnect flawlessly or legacy devices will be damaged. If the PSE does not adequately perform classification, power delivery and monitoring, intermittent failures and instabilities result. The PSE cannot control everything; when it applies power, it trusts the PD to follow the standard, turn on without oscillating and not draw more power than requested. Because both types of devices must cooperate, PD and PSE designers should consider the following issues from the perspective of both devices. Even with the availability of PoE specific ICs, system designers can't ignore these issues. PD and PSE interface controller ICs must be selected with care, because the ICs can't solve all the myriad board and system level challenges.
Before we examine the complexities of the PoE protocol and subtleties that can cause trouble, here are some simple mistakes to avoid:
The common-mode termination must be ac-coupled on both the PSE and PD or it will interfere with PoE detection (Figs. 1 and 2). Terminations may use 0.1-µF capacitors rated at 200 V, but the -3 dB point where this network comes into effect is near 47 kHz as opposed to 10 kHz.
With PoE, wires and circuit board traces, which used to only carry data, are now transferring power with up to 10 times the current. Consider placing components closer together and using wider traces.
Some Ethernet magnetics may not to be compatible with PoE even though they bring out center taps, because they cannot handle the dc current of PoE and magnetics can saturate or overheat.
As shown in Fig. 2, often the common-mode voltages, and hence the PoE port, is accessed through center taps on the data transformers. If the current flow is unequal, the transformer can saturate, blocking data transmissions. Careful winding of the auto transformer in Fig. 1 can encourage both wires of a pair to share current. Making the choke and data transformer higher resistance reduces the dc current flowing through them.
A common-mode choke in series with the common-mode termination blocks the termination. In Fig. 2, signals on the cable see the high impedance of the choke's inductance, rendering the common-mode termination network ineffective. The arrangement in Fig. 1 solves this problem.
Sharing a common-mode choke between two ports couples the ports together and doesn't achieve the goal of limiting each port's common-mode current. Instead, the choke only controls the sum of both ports' common-mode current and acts as a transformer between the ports' common modes. PoE relies on common-mode current to transfer power. In addition, to turn-on and turn-off transients, fault conditions can cause larger transients. Sharing a common-mode choke between two ports couples all of these transients from one port to the other.
Detection prevents legacy devices from being damaged by PoE's 48-V output and is the primary step in establishing a PoE connection. Without proper detection, all other aspects of a PSE are useless. The controller IC's detection circuits shouldn't be fooled into discovering a PD by noise, offsets, loose connections or complex impedances. Tolerating and rejecting external noise is critical because the 25-kΩ signature resistance is measured through up to 100 m of cable without the advantage of rejecting common-mode noise that normally comes with twisted pair cabling. Inductive coupling onto the cable can cause noise amplitudes approaching that of the measured signal.
Classification and Power Allocation
During classification, the PSE must maintain an output between 15.5 V and 20.5 V, so it can measure the PD's classification signature and determine how much power the PD needs. A bad classification measurement or a faulty power allocation scheme can overload the PSE and bring the whole PoE network down.
Some PSE controllers reuse the MOSFET that switches power to the port in a low drop out (LDO) configuration regulating the classification voltage from the PSE's 48-V supply.
However, LDOs are extremely sensitive to their load impedance and can be difficult to compensate. Classification is a particularly tricky LDO circuit to stabilize, because the Ethernet cable inserts inductance between the LDO and its load. Furthermore, bypass capacitance must be less than 0.52 µF. When a PSE controller IC takes this approach, systems designers must follow the IC vendor's recommendations carefully. However, this still doesn't ensure stability.
Perhaps the biggest enemy of classification stability is the PD. The 802.3af standard only requires the PD to draw current to meet one of the five classes. A PD may draw little current until the port voltage enters the 14.5-V to 20.5-V classification range. At this point, the PD can throw a switch and instantly load the port with up to 44 mA (Class 4). This load step can quickly cause the port voltage to leave the classification range, leading to oscillations, overshoot and other undesirable behavior. PD designers should limit the rate of current increase when the PD applies its classification current. A PD with an I-V characteristic similar to that shown in Fig. 3 will exhibit benevolent behavior. The standard lays responsibility for classification stability at the foot of the PSE. PSE designers must take up this mantle to ensure their devices provide stable outputs, regardless of PD behavior.
In addition to the slew rate or di/dt when the PD turns on its classification current, large sections of the PD's I-V curve are also unspecified, shown with implied limits in Fig. 3. The standard says nothing about PD current between detection (10.1 V) and classification (14.5 V) and between classification (20.5 V) and power on (30 V). A PD that draws no current in these ranges may take a long time to be detected, because the PSE isn't required to actively discharge the port and only leakages will discharge the port in these ranges.
In these two undefined regions, the PDs should roughly stay within these limits: PDs should draw more current than a 25-kΩ resistor and less current than their classification signature (Fig. 3). What does this mean for a Class 0 PD where the classification signature may be 0 mA? The answer is simple: Don't build a Class 0 PD.
Classification is optional. However, customers will expect classification because devices without classification waste power and thus waste their money. Classification provides three power levels that PDs in the corresponding classes are guaranteed never to exceed. With the knowledge that Class 2 PDs will never need more than 7 W, a 24-port PSE can be designed to power 24 Class 2 PDs from a 180-W supply. Without classification, the same supply is only good for 180 ÷ 15.4 = 11 PDs.
While classification makes better use of the PSE's available power, the 802.3af standard still emphasizes reliability over efficiency. A PSE can never allow itself to become overloaded. To this end, the standard mandates a power accounting and reservation scheme called power allocation, which serves as the primary method of overload prevention. When a PSE applies power to a PD, it guarantees that power, no matter what happens on the PSE's other ports. The PSE reserves the class-appropriate power for all the PDs it powers and keeps the total reserved power below the capabilities of its power supply. Thus, the PSE always has enough power to meet the PD's demands.
Turning power on is the real test of a PD design, as the PD must prevent port voltage oscillations. At a port voltage between 30 V and 42 V, the PD must turn on, which generally means connecting a hefty input bypass capacitor to the port. The combination of the large capacitor and the PSE's current limit can cause the port voltage to drop below 30 V, even to 0 V. If the PD turns off because of a drop in the port voltage, it will begin toggling between on and off. A PD design can guard against motor boating by limiting its current to keep the port voltage in the powering range or the PD can allow the voltage to drop but maintain its powered-on state for a limited amount of time, regardless of the port voltage.
For example, Linear's Technology's LTC4257 implements a 350-mA (typ) current limit that's lower than the PSE's 400-mA to 450-mA output and separates its turn-on and turn-off thresholds by 9 V. The current limit prevents the port voltage from dropping significantly while the 9-V hysteresis ensures the PD will stay on during smaller voltage drops.
The LTC4257 also includes a “power good” feature that indicates when the load is charged to the port voltage. Using a power good indicator is particularly important with dc-dc converters, as they tend to draw more current the lower their input voltage and could prevent the bypass cap from ever charging.
For the PSE, applying power tests the power-handling capabilities of the MOSFET it uses to control port current. This MOSFET spends the majority of its time either completely on or off, dissipating little or no power and tempting designers to select small devices. However, during the 50 ms to 75 ms that MOSFET limits the port current, dissipation can be as high as 25.7 W. Keeping the MOSFET within its safe operating area (SOA) can be a challenge.
PSEs can use foldback to reduce MOSFET heating. The 802.3af standard allows a port current limit as low as 60 mA when the port voltage is below 30 V (Fig. 4). This cuts MOSFET power to 3.4 W at 0 V on the port and 12.2 W at 30 V on the port so PSEs can use smaller, less expensive MOSFETs.
As with all power supplies, capacitors are a crucial element of a PSE: large value electrolytic capacitors to handle surges in load current and small value, low equivalent series resistance (ESR) capacitors to squash higher-frequency disturbances. The 802.3af standard has specific ripple requirements that every PSE output must meet. PSEs must comply with these requirements regardless of the behavior of PDs attached to other ports. Depending on the response time of its 48-V supply, a PSE will need 50 µF to 300 µF per port. Each port should have 0.1 µF to 0.52 µF of capacitance at its output. To ensure high-frequency stability, this local bypass should have low ESR. Be aware that all capacitors are not created equal, and tradeoffs are inevitable when high capacitance is squeezed into a tiny package.
Sensing the removal of the PD and turning off power, what the IEEE calls “disconnect,” is as important as the decision to apply power. In either case, the consequences of a mistake are the same: damaged equipment. DC disconnect, which relies on measuring the port current to determine the PD's continued presence, is simple in concept and implementation. AC disconnect senses the presence of the PD with a low-frequency ac measurement of the port impedance.
AC disconnect must sense the impedance of a PD at the far end of the Ethernet cable while simultaneously providing a stable output voltage to power the PD. PSE designers juggle these opposing constraints by putting a diode in series with the PSE's output. If the ac disconnect circuitry is not designed properly, microamp-level leakages will be sufficient to keep the port on. If your system uses ac disconnect, be sure to check its sensitivity to leakage currents and temperature.
Fault and Abnormal Conditions
In addition to turning the port off when the PD is unplugged, the PSE must remove power due to fault conditions such as excessive port current. The PSE also actively limits its output current to keep faults and power-hungry PDs under control. While the standard allows PSEs a leisurely 1 ms to bring their outputs into the 400-mA to 450-mA ILIM current limit, carefully designed PSEs will operate much faster, quickly limiting the energy during a fault condition.
If a short occurs at the end of a 1-m to 10-m cable, the cable inductance does little to slow the rapid increase in current, which can exceed 10 A within 100 µs, storing an appreciable amount of energy in the cable's magnetic field. When the PSE attempts to limit the current, the cable responds to reduced current by unleashing its stored energy and applying what could become kilovolts to the PSE. At the very least, PoE devices should include unidirectional surge suppressors on each port. Additional features, beyond the scope of the 802.3af standard, can aid in managing fault conditions. The fast pull-down circuitry included in Linear Technology's LTC4258 and LTC4259A PSE controllers turns off the port in less than 2 µs if the current exceeds 600 mA.
PDs also should be equipped with transient suppressors, as they can experience high voltages during turn on and turn off. Even with protection, some PD interface ICs are damaged by successive plugging and unplugging of the cable.
The standard is clear about what the PSE should do in response to fault conditions. However, the standard does not cover abnormal conditions, such as changes in the PSE's output voltage. A PSE's output voltage might change by 10 V when PDs are plugged or unplugged or the PSE switches to a redundant 48-V supply. These output changes are permitted as long as the slew rate is kept under control.
When the PSE's output changes dramatically, attached PDs can momentarily draw too much or too little current and be disconnected by no fault of their own. Consequently, when powering a PD, PSE outputs should not change by more than a few volts.
The 802.3 standard requires all Ethernet ports to be electrically isolated from any other conductor that is user accessible. This obviously includes the device's metal chassis, other connectors, a PSE's connection to ac line power and any alternate power connector on a PD. Items that may not instantly spring to mind, such as touch screens, EMI gaskets, switches, screws and even the metal shield that surrounds most RJ-45 jacks, must also be isolated from the pins of the RJ-45.
Some PD applications such as wireless access points with built-in antennas don't need user-accessible connectors beyond the RJ-45. For PDs that need other external connectors, the isolation problem is solved by dc-dc converters without a galvanic connection from input to output. Forward and flyback switcher topologies use a transformer to isolate the PD's PoE interface from the rest of its circuitry while also converting the PoE input down to voltages suitable for powering the PD's circuitry.
Both PSEs and PDs may have multiple RJ-45 jacks to support more than one Ethernet connection. The 802.3af standard includes additional requirements for isolation between Ethernet ports on devices with multiple ports. These requirements are illustrated with PSE examples but are equally applicable to PDs.
Environment B, the stricter isolation requirement, is for PSEs attached to Ethernet cables that cross an ac power distribution boundary. That is, electrical equipment near at least one cable connected to the PSE is wired to a different earth ground than the PSE or equipment near another cable. To prevent hazardous shocks due to different ground potentials at the end of each cable, Environment B PSEs have electrical isolation between all of their PoE ports; each pin in every RJ-45 jack must be electrically isolated from the PSE's chassis and the pins in all the other RJ-45 jacks on the PSE.
Environment A isolation addresses devices connected to Ethernet cables that do not cross over ac power distribution boundaries or leave a single building. All the ac outlets in the area served by the PSE are wired to the same earth ground so the PSE does not need to isolate its ports from each other. It can use one isolated 48-V supply to power all the ports. If one Environment A PSE switched its positive output and another PSE switched the negative polarity, their switched outputs would have a dangerous 90 V to 114 V between them. The 802.3af solves this problem by requiring Environment A PSEs to switch their negative outputs.
For more information on this article, CIRCLE 344 on Reader Service Card