User Tools

Site Tools


m0dul-ethernet

Descritption

Ethernet m0dul. Makes the r0ket speak IP. Currently in planning.

Idea

Main reason to do this is to get the received Openbeacon data from the r0ket to a central location in a nice and easy way.

Alternate Options

There are of course other options to do this:

We currently get the data off via USB-Serial, this requires one USB port per r0ket – due to the length limits of USB cables about one Computer (Dockstar/etc.) per 2 r0kets. This requires more Hardware and more configuration for roughly the same effect.

We could get the data off via rf - i.e. Wlan or Xbee.

  • Long-Range Xbee is rather expensive, we currently have no receivers and no deep knowledge of it.
  • Wifi: Own Wlan would clash in crowded locations and might not reach far enough. Normal Participant wlan is possibly not stable enough. Also: running Wlan from r0ket sounds complicated unless there are all-in-one chips.

Hardware

Plans on how to do it on the hardware side.

Ethernet Chip

Two main options.

Use a “simple”/“stupid” Ethernet chip, and do IP in the r0ket.

  • Pro: cheaper, cool, we can tweak it, we know exactly what it does.
  • Con: limited space means we will probably only get UDP, and maybe DHCP. Enough for our logging stuff, but difficult for other projects, only 10baseT
  • ENC28J60 2,60€ - 2,95€ bei Reichelt
    • Please do not use this chip. It's a pain to work with, doesn't do Autonegotiation, loses packets and has weird unfixed erratas. –equinox
    • considerably simpler buffer/queue management than the ENC28J60
    • supports 100base-TX & Autonegotiation
    • supports Auto-MDIX (Crossover)
    • QFN32 package
    • ca. 3€

Use a “smart” Ethernet chip which has a full TCP/IP implementation.

  • Pro: can do other cool stuff, supports 100baseT
  • Con: more expensive, possibly bugs in the IP stack. No IPv6.
  • W5100 4,03€ bei Mouser

→ So far it looks like the “smart” variant is the way to go.

Power

It would be nice to be able to (optionally) power this via POE.

Again two options here:

  • “simple” homegrown POE (just use the unused pairs to send Power)
    • homegrown/uncontrolled PoE will be instantly refused by the NOC. Frying an 20k€ Switch by misplugging a cable is not an acceptable risk. This applies for all uses that touch the building infrastructure – even if it's not supposed to be linked to a “big” switch you can always misplug something.
  • “normal” POE (802.3af or 802.3at) (.af sounds simple - provides for ~ 200mA @ 48V - enough for us)

according to Wikipedia, the injector decides wether to provide power via the “unused” or the “data” wires, which means more work for us…

→ Still undecided.

RJ45 Connector

As far as I can tell, we must decouple the ethernet lines from our ciruit. Doing this sounds like no fun[tm].

There are RJ45 Connectors with “integrated magnectics” which do this for us. But they then need to be POE-Capable which appears to be quite expensive.

1-6605834-1 is 6,08€ at mouser (http://www.te.com/catalog/pn/en/1-6605834-1)

“simple” RJ45 Connectors appear to be at about 0,90€

→ Still unclear if there are no other options.

Discrete Magnetics:

Recommended reading for magnetics selection: http://www.ceetus.com.cn/paper/VSC8201.pdf (Vitesse magnetics design note, cannot find original on Vitesse's website anymore)

Connection to r0ket

Over m0dulbus. All the Ethernet chips above do SPI.

Software

Nothing done yet. Need to decide on the Hardware first

Please implement LLDP and you will make the NOC happy :) (Possible on all of the chips, simple ethernet protocol.)

Openbeacon Reader

The openbeacon reader protocol we want to emulate is fairly known. Check tracking or reader for more details.

m0dul-ethernet.txt · Last modified: 2012/06/09 21:32 by equinox