TO DO

POWER OVER ETHERNET

Published in Circuit Cellar Magazine

TITLE

Power Over Ethernet, The Basics

Strap

Powering devices over Ethernet cabling should be easy, but there is a bit more to it than meets the eye, Eddie Insam explains how it all works under the covers

Introduction

OK, so you have designed your brand new Ethernet based appliance, maybe a clock, a weather sensor or some industrial controller device. It will soon be proudly hanging from your wall and connected to a RJ45 wall socket. But how are you going to power it? Where will it get its juice from? Surely you are not going to disgrace your design with a brick wart. There must be a better way!

Why not feed power over the CAT 5 cable? Well, it doesn't take much guessing to realise we won't be the first people to think of this. Standard CAT5 cable has 4 pairs, and only two are used for data in a typical 10 or 100Mbps installation (Fig 1a). So it sounds obvious to stick a few DC volts down the spare pairs. Oh yes, but hang on, life is never so simple. This is technology remember? There has to be a catch somewhere. So sit down and relax, we have the story.

Fig 1 Caption: Standard Ethernet 10Mbps or 100Mbps devices use just two of the four pairs available. The spare wires could be used to transmit power to the remote. Two possible methods are shown in b) and c). But watch out! The power source must be smart enough to detect shorts, overloads and avoid damaging components at the far end.

It may not come as a surprise that the wise men at the IEEE have been thinking about this for a while, and came up with a standard (IEEE 802.3af). This standard has been around since 1999, but progress has been relatively slow and only started to take off recently, mainly because of the availability of cheap specialist components. Tom Cantrell's article (Circuit Cellar 165, April 2004) and more recently Jeff Bachiochi's (Circuit Cellar 187, Feb 2006) discussed some of the available components and modules. A wide range of parts are now available, including dedicated switching transistors, isolation transformers and high quality non-saturating magnetics, making Power over Ethernet a practical proposition.

The Technicalities

The IEEE document discusses two main methods for sending power down the CAT5 wire: (a) by using the spare pairs, and (b) by sharing with the existing data lines using centre tapped transformers (Fig 1b and 1c). The latter method beneficial for situations where no spare cable capacity is available.

Method (a) allows a decent amount of current to be drawn, as the two spare pairs are paralleled together to increase capacity by reducing the total DC resistance. The present IEEE specifications allow up to 13W of power to be downloaded this way. This may not be enough for some heavy-duty devices, but it is quite acceptable for medium and small items such as TV cameras, VOIP phones and other smaller appliances. An updated "PoePlus" standard is currently being considered. This will allow for up to 30W capacity, while still remaining backwards compatible.

Transmission of power using the centre tapped transformers option is rather more limited. Pulse transformers and other magnetics in the Ethernet controller must be designed to take the full DC power load current without saturating, not an easy task for miniature surface mounted components. The advantage of this alternative is that it leaves the extra pairs alone, an essential consideration in higher speed Gigabit Ethernet, which requires all four pairs to carry data.

Why can't I just stick any old power supply across the spare wires?

Because we do not know what is at the remote end, and we may run the risk of blowing up sensitive equipment. If you do not believe me, look at Fig 2. This shows your typical Ethernet terminator. This kind of circuitry is sometimes contained within a single metal enclosure called a magjack.

Fig2 Caption: A typical Ethernet termination. The resistors strapped to the spare data pins and to the center taps are there to balance the line and to reduce noise, but can quickly flash to smithereens in true Harry Potter style if any unmanaged DC power is placed on the cable.

Note the two 50 ohm resistors R3 R4 across the center taps of transformers T3 T4. They are branched in series to form an effective 150 ohm DC load across the input lines. Also note the two 50 ohm resistors R1 R2 right across pins 7/8 and 4/5. The purpose of these is to present a controlled impedance load to the otherwise non-terminated wires, and are added for robustness and noise reduction. This hook-up is sometimes known as a Bob Smith termination.

Anybody who places a 48 volt DC raw supply into the above socket will be placing a good third of an amp across these tiny resistors, guaranteed to vaporise them to kingdom come. Tiny SMD resistors were not built for such treatment.

Admittedly, some terminators and magjacks have extra series capacitors to protect the resistors, and not all Ethernet devices use such extra networks. However, we want to ensure that the power supply will not embarrass us by blowing those other devices that have them. There are other potential problems, which can be blamed on bad design or to pure accident. For example, a wireman could accidentally short or swap the CAT5 pairs. All possibilities have to be considered, and many are mentioned in a recent IEEE report "DTE power problem set and methodology" See resources at the end of this article for the website.

Needless to say, the good people at the IEEE have devised cunning schemes to pre-empt the above "challenges" In simple terms, the power supply is made smart, and is able to figure out what is happening at the load end. It does this by taking a number of graded impedance measurements before applying full power. These "impedance signatures" tell the supply whether it is safe to apply full power or not. Full power is applied only when it is safe to do so. Furthermore, the load is regularly monitored during normal operation to ensure nothing drastic will have happened since. This allows the supply to turn the wick off if it detects any suspicious problems, when the load fails, or when it is disconnected. This arrangement of course, needs co-operating equipment at the load end to provide the right dummy impedances at the right time.

Apart from the safety factor, the IEEE standard helps in reducing overall energy loss, as only those sockets that have a valid load can be programmed with power. During sensing, the supply knows the range of power loading taken by a load and ensures the correct amount of current is delivered (within a reasonable range) No more, no less.

So the power supply needs to be computer controlled?

Well yes, but what isn't nowadays. The operating algorithm is relatively straightforward, and even the tiniest of microprocessors can handle it. You just need a power supply that can deliver a programmable voltage between 2 and 48 volts, means of sensing the load current, and means of measuring its output voltage, from which we can compute the load impedance and various other parameters. The rest is just software. Mind you, and as you would suspect, the IEEE standard is not that straightforward, and many options are included to cater for all eventualities. For example, there are options for sensing an AC load as well as the DC load, but many of these are just optional enhancements, and you can get away by just sensing a plain DC resistive load. Fig 5 shows conceptually what the supply looks like.

How about the load end?

The power source does its validation by sensing the impedance of the load at different source voltage levels. So while this is taking place, the load needs to behave a bit like a non-linear resistor, otherwise called a Signature Impedance, see Fig 3. The circuitry to do this is relatively simple, and already there are a number of ICs that will do the job for you. The basic circuit can best be described in terms of discrete components, shown in Fig 4, showing the basic principle.

Tell me how it works

First, we have to introduce some jargon. Don't forget we are talking IEEE standards, so the use of jaw churning techno-speak is essential. The Power Sourcing Equipment (PSE) is another word for the source end, or power supply. The Powered Device (PD) is the equipment at the user end or load. An Endpoint Feed describes the arrangement or situation where the power supply is fitted inside the source box e.g. inside an Ethernet Router, so only one cable link is needed between the Router and the PD. A Midspan Feed Unit (MFU) is a separate box that is added somewhere between the router and the PD to provide the power. This necessitates two CAT5 links, one between the router and the MFU, and another between the MFU and the PD. You would need to buy an MFU if you already had a router that did not provide Power Over Ethernet. If you were to start from scratch, you may prefer to buy a router with a built in Endpoint Feed. Are you still with us? Don't go away, there's more. The voltage level at the power supply is specified as between 44 and 57 volts, whereas at the user end this is widened to 36 and 57 volts to allow for reasonable ohmic drop down the CAT5 cable. The PSE is allowed to supply up to 15.4W of power with a maximum current limit of 350mA. The maximum power consumption at the PD is about 13Watts, which corresponds to a nominal current of 270mA at 48 volts. CAT5 runs can be considerably long, and a lot of ohmic loss can be expected. This is one of the reasons why the standards suggest that pairs 4,5 and 7,8 should be paralleled together to halve the cable's resistance.

Although the specifications define which pin should be positive and negative, the load must not assume anything (Murphy's Law!). The PD must also ensure that the internal supply is floating with respect to the input power feed. So it needs to include a bridge rectifier on the input plus a floating transformer-isolated power converter.

OK so far, how does it work?

Lets take it in stages. Look at Fig 3. When there is no load applied, i.e. when the user end PD is disconnected, or during first power on, the source (PSE) repeatedly sits in a short loop sensing the line for an ohmic signature. This stage is called the Detection Phase. It does this by placing at least two spot voltage levels between 2 and 10 volts, and measuring the line currents drawn at these points. The current difference is taken rather than the absolute values as this makes for a more precise derivation of the signature impedance and also compensates for fixed losses such as diode drops. A current limiter on the line ensures the load can draw no more than 5mA, just in case there is a short or similar problem. The two test voltages are changed relatively slowly to avoid any glitches, the specifications suggest between 2mS and 500mS between readings. During the Detection Phase, the load has to present a 25.4K resistive component in parallel with a 0.1uF capacitor. This is not a real component value, as you cannot buy 24.5K resistors in the shops, but a theoretical average. To be more precise, any load between 23.75K and 26.25K is considered valid. Loads below 15K or above 33k are considered invalid. Loads outside these two ranges are in a no-mans land and may or may not indicate the presence of a (possibly faulty) PD. If this all sounds confusing, it is because this is the way Standards tend to specify things that need to lie in ranges. Mere mortals like us only need to know that the resistance needs to be about 25k. The capacitor is required for an optional alternative AC load sensing method, this will be discussed later.

Fig 3 Caption: Three distinct phases. The simplest of loads will present a 24.5k resistance until the input voltage rises above 30v, at which point the actual driven circuit will be switched into operation

When the 24.5k resistor is detected, the PSE proceeds to the next stage, the Classification Phase. If at any point, the load measures too low or too high, the PSE assumes there is no valid termination and removes the power altogether. It then waits a couple of seconds, and starts again from the Detection Phase, repeating the cycle forever. In the worst case, an incompatible or bad PD will see a maximum of 10 volts or 5mA applied across it, and no harm will be done. This is somewhat preferable than being hit with 48volts at full current!

The purpose of the Classification Phase is to determine the range of load currents the user device will need. In other words, the PD tells the PSE how much current it is going to need. The use of limited power ranges could be useful for loads that need critical monitoring, or to avoid users connecting unauthorised devices to certain sockets. A main application for this is to allow limited resource PSEs to allocate different power levels to different outlets, or to allow the PSE to enable only certain PDs in case of emergency or other priority, although in practice this may create more problems than it can solve. The following chart shows the options available.

Class I load by PD Usage Power Range

0 0-4mA Default 0.44 to 12.95W (full range)

1 9-12mA Optional 0.44 to 3.84W

2 17-20mA Optional 3.84 to 6.49W

3 26-30mA Optional 6.49 to 12.95W

During the Classification Phase, the PSE applies two or more voltages between 15.5 and 20.5V (current limited to 100mA), and measures the new signature impedance. The PD recognises these new voltage levels, and switches in a suitable load resistor according to its expected needs. Note that if the PD retains the original 25.4k resistor, it will be classified as Class 0, and default to full power range, which is very convenient. In other words, the simple do-nothing option will give us the full power range. Who says committees never come up with practical ideas?

The PSE will have a further chance of detecting improper loads or shorts during this stage, and remove the power altogether if anything feels suspicious.

Having passed the Classification Phase, the PSE can now slowly ramp up to full power, so the voltage now goes up to the 48 volt/300mA current limit. At the same time the PD will connect the line to its internal circuits powering the user electronics. After this new stage, and while providing full power, the PSE will constantly monitor the load for current drawn. The PD will guarantee to sink a defined Maintain Power Signature (MPS). In other words If the load current rises above 400mA or drops below 10mA (at less than 20% of the time), the PSE will assume the load has gone funny, kill the supply and revert to its Detection Phase as before. There is a defined back-off period of 2 seconds to avoid the whole thing going into wild oscillations. A well-known scenario to be avoided is where a valid PD device has just been unplugged from an Ethernet wall socket and a legacy device is plugged in immediately after. If the PSE does not recognise this situation quickly, it can damage the legacy device since full power is still being applied to the line. This is where the alternative AC sensing method scores. A 500Hz AC common mode signal is superimposed on the DC, any AC disconnection can be detected immediately, whereas a DC disconnection has to rely on slow voltage decays before it can be correctly detected. Note that the supply can optionally use either AC or DC sensing, but the load must include methods for supporting both. In practice this is just a 100nF capacitor in parallel with our beloved 24.5K resistor.

The Job of the Load

During initialisation, the PD presents a variable impedance to the supply depending on the input voltage across its input pins. In summary: between zero and ten volts, the load looks like a 25.4K resistor (plus the voltage drop effects of the bridge rectifier). Between ten and twenty volts it can still be a simple resistor, but calculated to give the current load specified in the chart above. Alternatively it can keep the same 25.4K to query for Class 0 and the full power range. As the input voltage ramps up between 30 and 42 volts, the user load is switched in. If during full power, the input falls below 36V, the PD disconnects itself from the supply. This is known as Under Voltage Lock Out (UVLO).

It is the responsibility of the PD to ensure the load does not take more than the rated power, or less than a minimum threshold current to ensure it does not get turned off. This minimum current is specified as 10mA for at least 75mS in every 325mS. Unplugging the PD can then be easily recognised by the PSE as it sees the current drop below 10mA. One disadvantaged product group in this scheme are very low power devices that will need to include a bleed resistor just to ensure the minimum current threshold is met. So much for energy conservation!

A Typical PD

Fig 4 shows a most basic PD. It has been divided in sections to show the relative responsibilities. Section (A) shows a bridge rectifier. As already mentioned it is always good practice to use a bridge in case the wires have been swapped around. A PD can make use of both alternative sources by having two bridges, each connected to the two power options shown above in fig 1b and 1c. Section (B) shows the main 24.5k signature sensing resistor, plus a 100nf capacitor to provide an AC signature load, There is also a 60v zener to provide some sort of overall protection, (an extra fuse connected between this and the input line would not come amiss). In this simplified circuit, the classification phase is also managed with the same 24.5K resistor, classifying the unit as Class 0. Section (C) is a simple gated switch that turns the load on when the input voltage reaches 30v or so. Section (D) is mainly here to denote that the load has to sink at least 10mA. Section (E) represents a 36-42 volt converter, which must be floating (e.g. transformer isolated) A typical example of such modules is the RECOM Econoline Series. These are small potted modules, e.g. RS4805 that takes 36-72 dc input and 5v at 200mA output, all in a small SIL footprint.

Fig 4 Caption: The operation of a typical PD in stages. Each is described in the text.

A Typical PSE

Fig 5 shows a basic design for a PSE. Of course, most of the time you would not be making your own PSE but obtaining one ready made. The details here are given mainly for information. The supply consists of a decent 48V DC power supply, and a series regulator controlled by the D/A output from a microprocessor (possibly via a PWM output). The series resistor emulates current limit and a place to take a sample of the current drawn. One such controller is needed for each Ethernet line or RJ45 outlet. The design is pretty straightforward as accuracy is not primordial. One tricky part of the design is the wide-ranging metering of the output current, which needs to cover a range of 100uA to over 300mA. This necessitates either a high resolution (14 bit) A/D converter or means of switching in different shunt resistors for the different ranges. Note that the series pass transistor will not need much heat sinking, as it will normally be operating either fully on, when delivering full power, or at limited current during the initialisation phases. The software consists of a simple timed loop, covering the Detection, Classification and Power Delivery phases, one at a time. The IEEE 802.af document describes procedures for implementing a version of this flowchart if you have the time and inclination to decipher the gripping notation and methodologies used.

Fig 5 Caption: A power supply will include a micro in a standard design configuration such as this to sense load current and generate output voltage levels accordingly.

And here is one I cooked earlier

Of course you may not be interested in making your own circuits as there are plenty of ready-made chip and module solutions available out there to make it all easier. But understanding the principles involved will ensure you won't get caught in many gotchas! The current line up includes:

MAX5940/1: Among the first kids in the block these chips provide all of the 802.3af interface detection, classification and switching facilities. The chip is normally used in conjunction with a separate Maxim 48volt switching down regulator (MAX 5014) to provide a complete power supply function.

National Semiconductor's LM5070, LM5071 and LM5072 are typical of the all-in one-chip solutions. They integrate a current mode DC-to-DC controller, user programmable under-voltage threshold, a fault current control loop, and many other functions. The 5071/2 can also accept power from an external AC/DC adapter (wall-wart).

Texas TPS2370 and the latest TPS23750, and TPS23770 are also big contenders. They combine the functionality of the older TPS2375 controllers and need a minimum number of external components,

Similar devices are also available from Linear Technology (LTC4257) and Supertex (HV110).

Chip solutions are also available for the PSE end. Some of these have multiple controllers, which allow four, eight or even twelve power supply controllers from one chip. Current devices are Maxim MAX5945, Texas TPS2383, Linear Technology LTC4258 and the PowerDSine PD640xx. series

Instant satisfaction "complete modules" include PowerDsine 3001 (a single port Mid-Span supply) and the corresponding DWL-P50 end load adapter. The latter comprises of a floating supply that can generate either 5v or 12v dc at the flick of a switch (Fig 6) this pair can provide a relatively cheap and complete instant solution to small POE needs. Similar products also available from other suppliers such as Hyperlink Technologies.

Be sure to see many more Power over Ethernet solutions in VOIP phones, CCTV cameras, and industrial Ethernet applications. Integrating a POE supply into a module will be commonplace.

Fig 6 Typical example of a ready to go module. Ethernet in, Ethernet out, and a choice between 12V and 5V dc outputs.

The Author

Eddie Insam lives next to the Thames in southern England. He has been designing specialist signal processing and telecomms systems for over 20 years. You can reach him at edinsam@eix.co.uk.

Resources:

IEEE "802.3af-2003" IEE standard Part 3 Amendment: Data Terminal Equipment (DTE) Power via Media Dependent Interface (MDI). Free from the IEEE website. ISBN 0-7381-3697-2 SS95132

Tom Cantrell, "Powered Points" Circuit Cellar Issue 165, April 2004

Jeff Bachiochi "FROM THE BENCH Power Over Ethernet Primer" Circuit Cellar Issue 187, February 2006

DTE power problem set and methodology

http://www.ieee802.org/3/power_study/public/nov99/mccormack_1_1199.pdf

Maxim MAX45940 POE chips

http://pdfserv.maxim-ic.com/en/ds/MAX5941A-MAX5941B.pdf

Supertex HV110 POE chip

http://www.supertex.com/pdf/datasheets/HV110.pdf

RECOM power modules

www.recom-international.com

National Semiconductors LM5070/5071

http://www.national.com/pf/LM/LM5070.html

http://www.national.com/pf/LM/LM5071.html

http://www.national.com/pf/LM/LM5072.html

Texas TPS2370

http://focus.ti.com/docs/pr/pressrelease.jhtml?prelId=sc04044

Linear Technologies LTC4257

http://www.linear.com/pc/productDetail.do?navId=H0,C1,C1003,C1006,C1125,P2282

Dlink DWL50

http://www.dlink.com/products/?sec=0&pid=368

HyperLink Technologies

http://www.hyperlinktech.com/web/poe.php