The Netgear FS116P is advertised as being an 802.3af power source for use with POE (power-over-ethernet) gear. it is a 16-port home-type switch with provisions to supply power from 8 of the ports. it was selected by WSCICC for our first large installation of Soekris gear at a site we call TGR (top of Gold Ridge).
we have powered Soekris kit and Senao CB3-Deluxe bridges with power injectors home-built by Jack Christensen (using 12VDC from a computer power supply), with a modified hub where 12VDC is fed to the unused pairs in a similar manner, and with the AC-powered single midspan power unit sold by Pacific Wireless, a 48-volt unit advertised as being 802.3af compliant.
we had assumed the Netgear would "just work" but a posting on the Soekris-tech mailing list from John Bellardo who found the Soekris didn't work with an HP powered switch caused us to try our Netgear switch on the bench without waiting to go to the remote site. the Soekris failed to work with the Netgear. here's a direct link to John's report which involves modification of the Soekris board.
i like Netgear's "Installation Overview" detailed instructions:
Estimated Time: 5-10 minutes.the reason is that the power in both the HP switch and the Netgear switch is supplied on the data pairs (actually from the center-taps of the isolation transformers for the data pairs), while the Soekris expects power on the unused pairs in a CAT5 cable. the solution offered in the other posting is not a good idea for us because we will be installing 6 Soekris boards at this location and don't want to modify that many units or have non-interchangeable gear at different locations.
but perhaps the biggest reason is that the Netgear ties each of the unused pairs together and has a small surface-mounted 75-ohm resistor to earth from each pair. so this value would draw about 2/3rds amp, exceeding the power available from the powered port and also exceed the capability of the resistors and would burn them out if power were to be supplied at 48 volts.
the POE chipset chosen by Netgear is the PowerDsine 64004 chip. (update on 18 Jul 2006 - PowerDsine now part of MicroSemi - link to FAQ updated) check out this FAQ from the chip manufacturer's support pages. in addition PowerDsine also sells mid-span power units which work the way we want and power the Soekris boards, and all their data sheets recommend using the unused pairs rather than the center-taps. here is the page for the actual chip used by Netgear.
fortunately the above mentioned 75-ohm resistors can be removed and then it's easy to add jumpers to bring power to the pads that connect to the unused CAT-5 pairs making the Netgear supply power to both types of POE gear.
here you can see the 8 pads from ethernet port 1 at the upper left corner of the board and note you can visually follow the traces from the top of R231 to pin 8 and from the top of R232 to pin 4. (internally there is a trace from pin 8 to pin 7 and from pin 4 to pin 5). if you carefully look there is a via hole between the lower ends of R231 and R232 which connects these ends of the resistors to earth. you can also follow traces from pins 1 and 2 back to two vias near the screened legend #6 of R216 which go thru to the transformer winding on the other side of the board, and likewise traces from pins 3 and 6 going a bit farther down the line just to the left of the #7 in legend R217. and note the two traces from C212 and C213 (the physical parts, not their legends) which go inbetween the data leads from the ethernet connector
a group of 8 transformers is contained in a surface-mount package of 48 pins. our ethernet port connector #1 is associated with the transformer with 12 pins, (pins 19 thru 30), with the screened legend T1. the connections to the windings is in this order: two pins for winding 1, a pin for center tap 1, center tap 2, and the two pins for winding 2.
C212, R215 and C213, R216 are just terminating components for the power supplied to the center taps of the transformers, and so make a convenient location to access the power from the PD64004 POE controller chip.
this pin-out from the transformer is different than the transformer used by Soekris where the center-taps are between the end-points of their respective windings, more like you'd visualize in a schematic diagram.
so all we have to do is remove R231 and add a jumper from the near end of C212 to the pad connecting to pins 7 and 8 of the ethernet port, and remove R232 and add a jumper from the near end of C213 to the pad connecting to pins 4 and 5 of the ethernet port, as in the picture above. tried this out with a Soekris board and it worked great (while power is still applied to the data pairs, the center taps on the Soekris end are not connected). this photo shows only one port modified for the initial testing.
a close-up of the two jumpers (unfortunately suffering from motion-blur) shows the detail of the connections. i used 28 gauge tinned copper wire with thin-wall teflon tubing (Alpha TFT-200) for insulation. i choose to remove the original solder from the pads of the capapacitors and add Sn67 electronic solder, then connect along the side of the capacitor with the jumper wire. on the connector end of the jumper note that the teflon has to insulate us from the nearer pads, as they are connected to earth. on that end i bent the jumper lead 90 degrees, pointing straight down, and soldered with fresh Sn67 electronic solder to the pads. to ensure reliable connections, i always freshly tin the leads, add the appropriate amount of solder to the pads and then "sweat" solder to make the attachment. then pull with some force after the connection has cooled to make sure the joint is mechanically sound.
continue to remove the 75-ohm resistors and add jumpers to modify the remaining 3 ports of the first group. note that on the jumper from C218 to the pad of R233 i didn't go thru the square pins, but arced the jumper an a sweeping angle. avoid using teflon tubing with any tension near a sharp corner, as teflon will flow and over time cause a short circuit. teflon is great as it doesn't melt or even shrink at soldering temperatures, but cold-flow against sharp edges is its Achilles heel.
our jumpers for the first group:
R231 to C212
R232 to C213
R247 to C214
R248 to C215
R255 to C216
R256 to C217
R233 to C218
R234 to C219
continue for the second group:
R235 to C220
R236 to C221
R257 to C222
R258 to C223
R259 to C224
R260 to C225
R237 to C226
R238 to C227
on the right side of the photo note that R239 and R240 are for port 9 of the switch and are the start of the non-POE ports so don't remove these terminating resistors.
here's the finished view. 16 jumpers and our power will work with both types of POE powered devices: either center taps on the data leads or spare pairs in CAT-5. if the powered device has terminating resistors, then you can have a problem, but because of the PowerDsine managed power device, each port is independently monitored and fails out. so one non-working device won't take out the others.
POE is handy, POE is good, POE is a standard, but there are some issues, especially when transitioning from pre-standard to true 802.3af compliance. and since Sep 2005 the follow-on 802.3at committee has been working on increasing the maximum power available to essentially double the limit specified in af.
1) we got here by people using the spare pairs in CAT-5 installations
to carry DC power thru the RJ-45 connectors with "passive" injection which
merely placed DC on the lines. often the power is applied without appropriate
current limits so you can burn out equipment that is wired with different
thinking, as well as no standard of which polarity and voltage to expect
from the "passive" line. so setting the standard placed emphasis on
protecting prior equipment when plugged into a "standard POE jack" as well
as protecting the wiring from excessive currents to remove fire hazard.
2) "passive" injection (and where appropriate "splitting") is handy since it can accomodate existing equipment that may not even have center-tapped transformers.
3) transformer center-tap "phantom" power has the advantage that it works over CAT3 cabling which utilizes already installed telephone wiring with only two pairs specified for the data leads...there are no "spare" pairs. this makes it easy to take out existing EKT (electronic key telephone) desk-sets and just plug in VoIP desk-sets reusing the existing wiring in the walls.
4) there are a lot of considerations in POE especially when considering powering Wi-Fi devices that use external antennas that are grounded. thus the 802.3af standard requires that these devices have no galvanic connection between the power supply (either supplied thru center-tap or on spare pairs) and the antenna connector. this can only be accomplished with extra expense of a power transformer which the Soekris boards do not have. however, the IEEE seems to recognise that the power is "earthed" as well as the antenna mast so that when the powered device is supplied from the same power company tranformer this need for isolation is perhaps less critical. in fact, we see no problems in most cases, but did run into an issue with powering at our WSCICC thorn site running with antennas grounded to tall water-tank structures that were not connected to the same earth as the power. (this was with a VIA small form factor computer board with two PCMCIA cards and two Senao access ports, not Soekris gear, but would apply the same if Soekris gear were being used)
5) so far, i don't see any reason that the modification performed on the Netgear FS116P does not conform to the 802.3af standard, powering both the center-taps and the spare pairs. which makes me wonder why Netgear didn't choose to do so. would be interested in learning if anyone can point out any justification for powering only center-taps (realizing that the 802.3af standard does allow powering either center-taps or spare wires or both).
-ron wickersham