Overview
Under the existing IEEE 802.3af standard, designers can supply up to 12.95 watts to the system's powered devices (PDs) over distances of up to 100 meters of CAT 5 Ethernet cable. The standard classifies PDs into four classes depending upon how much power they require for operation. Class 1 defines devices requiring small loads up to 3.84 watts such as print servers. Class 2 represents devices such as handheld computers or magnetic card readers that need between 3.85 and 6.49 watts. IP phones, security cameras and other devices that need anywhere from 6.5 watts to 12.95 watts fall into a Class 3 category. Finally, designers building low cost solutions can also use a general default class 0 designation which describes any PD requiring up to 12.95 watts.

PDs in an 802.3af-compliant network typically receive power from "endpoint" power sourcing equipment (PSE) such as a hub or switch, or via a midspan or midpoint PSE located between the switch or hub and patch panel. In both cases, power is delivered to the PD after a negotiation process determines that both a compliant PSE and PD are present. A PSE interrogates the PD by providing a voltage ramp (2.5 to 10 volts) to the receiving device. If the PSE detects the proper impedance signature (24.9 kilohms), indicating the presence of a PD, it proceeds to a classification stage where it applies a higher voltage ramp between 15 and 20 volts.

The PSE again measures current flow to determine the device's specific class. Until it detects a signature, the PSE delivers no energy. The power supply section of the PD is held inactive by an under-voltage lockout (UVLO) function, which isolates the switching stage until the signature and classification processes are complete. Once the process class is established, the PSE supplies full operating voltage to the PD which releases its UVLO to activate its DC/DC converter.

Enhancements
As PoE has taken off, demand has grown for support of applications, such as tilt/zoom/pan security cameras and POS terminals, which require more than 12.95 watts. To address this need, an IEEE Task Force has begun developing a higher power version of the standard. The IEEE802.3at, or PoEP, standard will support power levels up to 25 watts or more by using a new Class 4 category for PDs requiring more than 12.95 watts. While the standard is still in draft form and not scheduled for final publication until 2008 or 2009, we know the general guidelines upon which the new standard is being built.

Backward compatibility, for instance, is a requirement. IEEE802.3at-compliant PDs will recognize both IEE802.3af and IEEE802.3at-compliant PSEs. At the same time 802.3at-compliant PSEs will recognize PDs compliant with either standard. Also, IEEE802.3af-compliant PSEs will recognize new Class 4 IEEE802.3at-compliant PDs (Class 4 was included in the IEEE802.3af standard but reserved for future use). The primary challenge for those engineers writing the new standard was how to avoid initiation problems when a PD does not know if it is connected to a PSE compliant with either the older IEEE802.3af standard or the new IEEE802.3at standard. To avoid these issues, the task force added a mechanism in the new standard in which a PD requiring up to 25 watts and connected to an IEEE802.3af-compliant PSE can signal its system that it will not be able to access the power it needs.

This requirement necessitates a more sophisticated negotiation protocol. The recognition procedure (signature) in the new standard uses the same process as the IEEE802.3af standard. That is, it uses the same signature resistance and PSE voltage range for discovery. However, instead of sending a single voltage pulse to poll the PD, a PSE compliant with the new IEEE802.3at standard uses two classification-voltage pulses. After the initial pulse is sent, the PSE momentarily cuts off the classification voltage and allows the voltage level to drop (pulled down via the impedance of the PD). After a pre-determined delay, an IEEE802.3at-compliant PSE ramps the voltage back up to the classification point and effectively apply a second pulse.

If an IEEE802.3at compliant PD detects the voltage drop and the second class query, it recognizes that it is connected to an IEEE802.3at-compliant PSE. PDs compliant with the existing IEEE802.3af standard will not react to the second query signal because the duration of the classification cycle occurs within IEEE802.3af classification time limits. Instead they will respond to the second class query as it were an extension of the initial class query. The IEEE802.3at compliant PSE will be able to determine if it is connected to an IEEE802.3at PD because that device will respond to a classification query by presenting a Class 4 impedance. As the response to Class 4 is undefined in the IEEE802.3af standard an IEEE802.3af PSE may or may not power up an IEEE802.3at PD. To support higher power, the output range of an IEEE802.at-compliant PSE will be narrower than it is for an IEEE802.3af-compliant PSE.

Improved Power Distribution
The second key differentiator between standards is in PoEP's ability to control power distribution. Under the existing standard, once PSEs identify a PD's class, they simply provide the maximum power allowable under the class definition. One of the goals of the task force defining the new standard is to enable the transfer of additional information on the PD's power requirements as part of the classification process.

To reduce the complexity of the PD's hardware interface, the new standard will convey this information via layer 2 software instead of the layer 1 common mode power path. Accordingly, PDs requiring more than 12.95 watts under the 802.3at standard will support a layer 2 classification mechanism. By allowing each PD to determine peak and continuous power requirements, this new feature will allow designers to budget and match total system power capabilities more closely to actual PD load requirements.

In this way an IEEE802.3at PSE can allocate a minimal amount of power to a security camera when it is in a set position, but budget for more power when it is needed to operate a motor to pan or zoom (it is unlikely that multiple cameras will zoom at the same moment so the extra power could be part of a shared additional power budget). The task force has not yet completed the work on how this new software based description will be implemented in IEEE802.3at.

Tradeoffs—Flyback versus Forward
To take full advantage of the PoEP architecture, you must carefully consider your power supply design options to maximize system efficiency and drive down cost. Actually, the design rules are initially not too difficult since the voltage and current requirements of the PD will, in most cases, dictate the converter topology. Typically, designers building DC/DC converters for PDs in PoEP applications will choose between a flyback or forward topology.

If the load current remains under 6 amps, a flyback topology offers the least expensive option. Typically used in applications with output voltages greater than 2.5 volts, flyback converters offer excellent isolation between output and input, and eliminate the need for output inductor. However, as compared to other architectures this topology (Figure 1, below) offers limited efficiency and higher output ripple current.

If the output current requirements exceed 6 amps, forward converters offer a better option from a cost and efficiency standpoint. Often used in applications that require less than 12 volts, these converters deliver higher efficiency levels due to lower peak and RMS currents. They also exhibit lower output ripple current and can utilize synchronous rectification to provide an isolated output. A drawback, however, is that forward converters (Figure 2, below) incur the additional cost of an output inductor.

PoEP recognition process
As discussed previously, IEEE802.3af-compliant PDs typically receive power from the PSE after a negotiation process. Under the existing standard, PDs are categorized into any of four classes (0,1,2 or 3) depending upon their power requirements. A PSE determines the class of a PD by presenting a voltage ramp to the receiving device which rises from 2.5 volts to 10 volts. Once the PSE detects the proper impedance signature of the PD, it moves on to the classification stage where it applies a higher voltage between 15 and 20 volts to the PD and measures current flow to energize the cable.

The under-voltage lockout (UVLO) function block plays a key role in the process by holding the power supply section of the PD inactive until the signature and classification process is complete. Once that process concludes, the PSE supplies full operating voltage to the PD which will release the UVLO, connect its Pass-switch and activate its DC/DC converter.

The IEEE802.3at or PoEP standard extends the applicability of 802.3af by supporting power levels up to 25 watts or more with a "new" class 4 category (Class 4 is described in the IEEE802.3af standard). The recognition procedure in the new standard mimics the process used in the existing standard. Instead of sending a single voltage pulse to poll the PD, however, the PSE sends two pulses. Once the initial pulse is sent, the PSE cuts off the classification voltage and allows the voltage level to drop. The PD internally loads the rail and is able to detect the resulting voltage drop. After a pause, the PSE steps the voltage back up to the classification range. PDs compliant with the new standard will recognize the two pulse classification from the PSE and identify it as an IEEE802.3at compliant device. Since the duration of the classification cycle occurs within 802.3af time limits, PDs compliant with the existing standard will respond to the second query as if it were still the initial query (ignoring the two pulse message).

Discrete versus integrated
Given the nature of this recognition process and the power requirements of IEEE802.3at PDs as described above, the PD devices thus require a converter IC, power MOSFET switching, in-rush protection, UVLO, and signature and classification circuitry. Designers building power supplies for 802.3at-compliant PDs have two options available. Power semiconductor manufacturers have opted to take either an integrated or discrete approach. Each has its advantages and drawbacks.

The integrated approach combines the in-rush protection circuit and DC/DC controller with the UVLO, switch, signature, and classification functions. These circuits typically require an external MOSFET. Integrating the front-end PoE functions simplifies the designer's task and reduces parts count. However, it also limits the designer's options by forcing the development team to use a single device for all applications. This strategy will increase costs by 25 to 80 percent compared to a discrete approach.

Other vendors take a more discrete approach and integrates just the inrush protection, a DC/DC controller, a MOSFET, and a switch into a single CMOS IC (Fig. 3, below). The designer then uses discrete circuits to add signature and classification functions, ULVO, and a pass switch. This architecture offers generally good scalability across all PoE classes and gives the designer a higher degree of flexibility to fit a particular solution to a specific application.

When selecting their architecture, designers should consider a number of features that can increase design flexibility and, in the process, reduce cost and maximize power supply performance.

• Some devices support a maximum duty cycle (DCmax) and line feed-forward with DCmax reduction feature that maximizes support for both flyback and forward topologies.
• Higher switching frequencies allow the use of smaller size transformers and supply higher bandwidth for power supply loop control.
• Integrated soft-start functionality limits peak currents and voltages during startup and minimizes output overshoot.
• Cycle skipping operation at light load can help minimize standby power consumption.
• Line under-voltage capability helps eliminate glitches at power-up and power-down and can help designers meet system-level requirements common in DC/DC converters.
• Line overvoltage functionality protects the device from excessive input voltage and line surges.
• Some devices permit external current limit adjustments that allow the designer to set the current limit externally to a lower level near the operating peak current and adjust the level gradually as line voltage rises. This gives the designer greater flexibility when implementing overload power limiting.
• Hysteretic over-temperature shutdown allows the device to automatically recover from thermal faults.
• A wide range of package options help designers minimize cost and footprint.

About the author
Andrew Smith is responsible for product management for the DC/DC and AC/DC product families at Power Integrations in San Jose. Andrew worked for several power-focused companies in product management, marketing, technical management, and design. He started his career as a power supply design engineer at Advance Power (UK). Andrew holds a First-Class-Honors degree in Electrical and Electronic Engineering from Middlesex University.