Electronic Design

Evolving PoE Promises Rich Landscape Of Power Innovation

Developments surrounding Power over Ethernet include a number of unexpected apps, many spec-compliant products from chips to midspan hubs, and efforts to boost the maximum current that's deliverable via Ethernet cable.

Developments surrounding Power over Ethernet include a number of unexpected apps, many spec-compliant products from chips to midspan hubs, and efforts to boost the maximum current that's deliverable via Ethernet cable.

It has been barely a year since the Power-over-Ethernet (PoE) IEEE 802.3af standard, DTE Power via MDI, was formally approved. However, the birth of the standard effort for PoE, a backward-compatible replacement for previous proprietary power-over-LAN (local-area network) schemes, occurred back in 1999, with 3Com, Intel, PowerDsine, Nortel, Mitel, and National Semiconductor as its proponents. Originally conceived as a means of powering Voice-over-Internet-Protocol (VoIP) phones, the number of PoE applications expanded dramatically once the standard was approved (Fig. 1). These range from the predictable (wireless access points and RFID tag readers) to the revolutionary (PoE and the garage band). (For a guide to terminology, see "IEEE 802.3AF Voltage, Power, And Equipment Designations," p. 68.)

PoE uses standard Ethernet cable, which has four twisted pairs. However, only two of those pairs are used for 10BaseT or 100BaseT. Because Ethernet data pairs are transformer-coupled at each end of the cable, either the spare pairs or the data pairs can be used to power powered-device (PD) equipment.

The PD must be able to accept power from either distribution configuration, though it may use only one at a time. At the power-source end of the cable, the power-source equipment (PSE) may apply power to either the spare pairs or the data pairs in a cable, but not to both simultaneously. Also, the PSE may not apply power to non-PoE Ethernet devices if they're connected to the cable.

The reasoning behind whether the power is carried over the spare pair or the data pair takes some explaining. According to Linear Technology design manager David Dwelley, some existing Ethernet installations don't have continuity through the spare pairs. So, the IEEE 802.3af committee specified that midspan PSE equipment could only power the spare pairs, while endpoint equipment could power either pair. (In practice, all endpoints power the data pair.)

That may seem counterproductive at this point in PoE deployment, because midspan PSE can't be used to power legacy infrastructure if it has no spare-pair continuity. However, says Dwelley, the committee thought that in the long term, most PoE would be endpoint powered, so endpoints should be the type of equipment that could handle any kind of infrastructure, including legacy sites with unconnected spare pairs.

When using the spare pairs, pins 4 and 5 are paralleled for one side of the dc supply, and pins 7 and 8 are paralleled for the other side. When using the data pairs, the PSE applies dc power to the center tap of each isolation transformer so that pins 3 and 6 supply one side of the dc and pins 1 and 2 supply the other. At the PD, data-pair power is recovered via center taps on each of its transformers. In either case, the sense of the dc voltage applied to the pairs doesn't matter, because the standard requires a diode bridge ahead of the dc-dc converter in the PD.

Ultimately, using a single cable for data and power provides simplicity, safety, and economic advantages (Fig. 2). There's no need to install an ac outlet next to a LAN socket, and the low-voltage LAN cable can be installed by individuals with fewer skills than electricians with regular electrical wiring experience. In addition, PoE's ­48 V is designated as a Safety Extra-Low Voltage (SELV), providing an additional safety factor.

Forty-eight volts is also the standard backplane bus in telephone central offices and data centers, so a standard uninterruptible power system can keep PoE appliances running in the event of a power failure. Furthermore, there's a large selection of dc-dc power converters designed for 48-V inputs. On top of that, power can be controlled remotely from the Ethernet switch (e.g., from a system controller via I2C) while Simple Network Management Protocol (SNMP) commands from the switch or from further upstream offer additional PD-control options.

A disadvantage of PoE may be its 13-W load-power limitation. However, this wasn't anticipated to be a liability in 1999 when the standard was seen primarily as a means of powering VoIP enterprise telephone sets.

The first PoE killer app to show up after enterprise phones was surveillance cameras. It's far easier to put a camera up on a pole in the company parking lot if the only wire needed to run to it is a CAT5 Ethernet cable. On the other hand, notes PowerDsine president Rich Bauer, cameras also challenge PoE's power limit when users want motors for pan, tilt, and zoom functions. "A basic camera may use five to seven watts," he says. "Add some motors and you quickly come up against that power limitation."

Less susceptible to the 13-W blues are the two wireless PoE killer apps—Wi-Fi access points and RFID tag readers. PoE lets a company put access points in lobbies, lunchrooms, conference rooms, and even in outdoor dining areas more or less ad hoc. If a particular location doesn't provide adequate coverage, it's relatively simple to shift to another without a great deal of fuss.

A UHF (15- to 20-foot range) RFID tag reader with a 1-W RF output draws only around 120 mA at 48 V. Shorter-range low-frequency readers draw even less, turning tag readers into another hot candidate for PoE implementation. High unit prices will probably see readers implemented first in the enterprise for supply-chain management, but they will eventually be competitive with alternative loss-control methods in retail stores. Long term, their combined advantages in loss-control and inventory management should make their use widespread.

To take advantage of PoE, PSEs must be able to detect the presence of a PD at the end of any Ethernet cable connected to it. PD appliances must assert their PoE compatibility, and they may assert their maximum power requirements.

Powering up a PD works like this: PoE-enabled Ethernet appliances identify themselves by means of a nominal 25 k(omega) (23.5 to 26.25 k(omega) with less than 0.1 µF in parallel) resistance across their power input.

The PSE first applies a voltage between ­2.8 and ­10 V on the cable. Because there's a bridge rectifier ahead of the resistor, the PSE must detect this resistance using a slope-impedance method, which involves making at least two operating point measurements. This is the discovery phase.

If the resistor is present, the PSE may move on to the optional classification phase, in which the PD may tell the PSE its maximum power requirements. If the PSE determines that the additional load of this PD will exceed its capacity, it won't apply power to that Ethernet cable.

For classification, the PSE presents a fixed voltage between ­15.5 and 20 V (limited to 100 mA) to the PD, which in turn asserts its load current on the line. The PSE then measures this signature current to determine which of a set of four power classifications the PD falls into (Class 0 = 0.44 to 12.95 W, Class 1 = 0.44 to 3.84 W, Class 2 = 3.84 to 6.49 W, and Class 3 = 6.49 to 12.95 W, measured at the PD). After successful discovery and, if necessary, classification, the PD is powered up.

IEEE 802.3af requires a PD to be powered up within one second. That breaks down into 500 ms(max) for detection, 10 to 75 ms for classification, and 400 ms for power turn-on. If the PD drops off the line, the PSE notes the absence of current draw, and the ­48 V is removed until a 25-k(omega) load is again discovered.

System designers needn't completely understand the intimate details of discovery and classification. Several chip makers offer products that make it easier to design PoE endpoint, midspan, and powered-device products. These include Supertex's HV110, TI's TPS2370, and Linear Technology's LTC4267 (Fig. 3), which incorporates a PD interface controller and a current-mode flyback switching regulator. National Semiconductor's LM5070 likewise consists of a PD controller and regulator, as do Maxim's MAX5941 and MAX5942. (The '41 is recommended for isolated forward and flyback converters, and the '42 is for non-isolated designs.)

Designs that use external buck regulators have a wider selection of PD controllers to choose from, including Linear's LTC 4358 and 4259 four-port controller/drivers and Power-Dsine's PD64012 12-port controller/driver. Beyond that, there's a wide selection of ­48-V supplies and center-tapped Ethernet data-pair transformers.

Already-configured PSE equipment ranges from Ethernet switches to midspans to single-port injectors. Cisco provides its Catalyst 4500 and 6500 48- and 96-port lines of modular switches, daughtercards, and line cards. On a smaller scale, it offers the stackable Catalyst 3560 and 3750 24- and 48-port switches. (In its 48-port configurations, the latter are limited to 7.7 W output per port.) The SuperStack3 4400 PWR switches from 3Com support 24 or 48 ports. Avaya's stackable P333T-PWR switches support up to 24 ports.

Midspan hubs include PowerDsine's 6000 family, with versions for 24 and 12 ports, along with Aldous Systems' POH-0850TX (eight ports), Alpha Telecom's P6012S (12 ports), and System Engineering International's 12- and 24-port PH12 and PH24 "Juiceboxes."

Many companies, including Axis, Hyperlink, MicroTik, MiLAN, Phihong, PowerDsine, SeaCom, Symbol Technologies, and 3Com, offer one-port and/or six-port midspan injectors for budget installations.

Not surprisingly, there's a renewed clamor among those in the industry concerning how much power a PSE can supply. In November, there will be a "Call for Interest" before the 802.3 committee as a whole. (A Call for Interest is the official way of launching a new Working Group). Assuming the Call for Interest is successful, a new working group (designated 802.3a-something) will be formed.

The new 802.3 subcommittee will hear two proposals for "high power over Ethernet." Not only are there pan/tilt/zoom security cameras that cry out for more power, but conference-room speaker-phones—part of the VoIP phone concept that spawned PoE—could use additional power for extra volume in large venues.

A proposal from PowerDsine, already implemented in its 8000-series midspans, involves powering all four pairs simultaneously. Spare pair pins 4 and 5 and data pins 3 and 6 would carry the positive side of the PoE supply; data pins 1 and 2 and spare pins 7 and 8 would carry the negative side. The spare pairs and the data pairs would each source and sink the same 350 mA as today's PoE. This would permit PSEs to provide up to 39.5 W per LAN cable. On the PD end, 25.9 W would theoretically be available.

An alternate proposal involves simply increasing the allowable maximum current per pair in the existing scheme. Although the gauge of the wire in CAT5 cable pairs is only 24 AWG (0.51 mm dia.), Linear Technology's David Dwelley says the major concern is the current capacity of an RJ45 connector.

On paper, he says, the standard Ethernet connector has a higher maximum current spec than CAT5 cable, but in tests-to-failure in Linear's lab, the connectors always fail first. However, those same tests show that cable and connectors together have a capacity for much more current than the present PoE limit. Empirically, those power-over-LAN security cameras said to run an amp-and-a-half tell their own story. (For a more conservative story, check the Handbook of Electronic Tables and Formulas, which recommends a maximum of 577 mA per AWG 24 wire, or 1.15 A for a pair.)

At first glance, the PowerDsine approach would seem like a winner. Why not use both pairs? But there are two drawbacks to the idea. One is the inevitable doubling of bill-of-materials cost for PoE components on both ends of a cable. The other is incompatibility with legacy installations that have open spare pairs. Today, the PoE standard is backward-compatible with any installed Ethernet plant. The four-pair approach would not be.

On the other hand, PowerDsine, clearly a leader in PoE, is already shipping hardware based on the proposal. The only required change for the standard would be to make it possible to power both sets of pairs at present current levels.

Yet following through on the proposal for increasing current levels would require additional research and experimentation, not to mention coordination with the various authorities that set fire and building codes. No one can predict which alternative the new committee will choose.

TAGS: Intel
Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.