Project

General

Profile

Feature #3314

design simple SFP experimentation board with LVDS transceiver

Added by laforge 2 months ago. Updated 15 days ago.

Status:
In Progress
Priority:
Normal
Assignee:
Target version:
-
Start date:
06/04/2018
Due date:
06/04/2018
% Done:

50%

Estimated time:

Description

This is basically a follow-up to #3313, but with a LVDS transceiver chip onboard.

So we'll have a single SMA? connector for Rx and Tx, which lead to the LVDS transceiver which then connects to the SFP slot.

The purpose here is to try to have some boards using which we can experiment how easy (or not) it is to use SFP transceivers with custom protocols/signals/systems that are not 8B10B encoded Ethernet, Fiberchannel, OC-* or CPRI.

https://www.digikey.de/product-detail/en/texas-instruments/SN65LVDS180PWRG4Q1/296-23882-1-ND/1966622 might be a part to use. It's fine for "lower speed" use cases up to 400Mbps but won't be able to do any higher speeds - while the SFP modules typically can do at least 1250 Mbit, up to 10GBit. So it's a bit of a waste of potential, but then I don't think we'd be playing with those kind of data rates any time soon...

SFP-xcvr-schema-detail.jpg View SFP-xcvr-schema-detail.jpg 34.3 KB mschramm, 06/07/2018 03:17 PM
SFP-xcvr-xairwires-detail.jpg View SFP-xcvr-xairwires-detail.jpg 65.9 KB mschramm, 06/07/2018 03:17 PM
20180607-tnt-sfp-test.png View 20180607-tnt-sfp-test.png 45.3 KB scope plot of running 10MHz via tnt's breadboard SFP laforge, 06/08/2018 07:56 AM
20180607_230547.jpg View 20180607_230547.jpg 379 KB breadboard SFP receiver by tnt laforge, 06/08/2018 07:58 AM
SFP-xcvr-schema2-detail.jpg View SFP-xcvr-schema2-detail.jpg 26.7 KB mschramm, 06/08/2018 08:50 PM
SFP-xcvr-xairwires2-detail.jpg View SFP-xcvr-xairwires2-detail.jpg 67.1 KB mschramm, 06/08/2018 08:50 PM
3200
3201
3205
3206
3207
3208

Related issues

Follows Misc Hardware Projects - Feature #3313: design simple SFP experimentation boardIn Progress2018-06-03

History

#1 Updated by laforge 2 months ago

  • Due date set to 06/04/2018
  • Start date changed from 06/03/2018 to 06/04/2018
  • Follows Feature #3313: design simple SFP experimentation board added

#2 Updated by tnt 2 months ago

Does it have to be a separate board ?

I would think some simple 0R links to switch between external LVDS lines or the ones generated by the TI LVDS converter would be enough and only have one PCB.
This would also potentially allow to easily monitor the LVDS lines externally by mounting 470R resistors instead of the 0R, then connect the SMA to a scope in 50ohm mode (creating sort of a 1:10 probe).

#3 Updated by laforge 2 months ago

On Wed, Jun 06, 2018 at 06:41:15AM +0000, tnt [REDMINE] wrote:

Does it have to be a separate board ?

not really, but given the low cost, I think it's the easier way.

I would think some simple 0R links to switch between external LVDS lines or the ones generated by the TI LVDS converter would be enough and only have one PCB.

but then you complicate the layout of the sensitive differential traces,
likely have to introudce stubs/vias, ...

That's why I thought let's keep it simple. But then, it's not set in
stone (erm... FR4) yet.

#4 Updated by mschramm 2 months ago

  • Status changed from New to In Progress

#5 Updated by mschramm 2 months ago

3200
3201

After creating the footprint for the SN65LVDS180 and placing it close to the SFP, it reveals that this would mean crossed Rx/Tx diff pairs (see images) which is not desirable. Those wires should go straight on top layer into the transceiver. I've committed that xcvr lib anyway, but will continue likely with a pair of SN65LVDS1 and SN65LVDT2.

#6 Updated by tnt 2 months ago

Huh ... don't you have TX/RX swapped on the SFP there ?!?

RD pins are the SFP receiver pins i.e. they are outputs.

See https://www.embrionix.com/storage/app/media/resources/HOST%20RT-nonnonMSA.gif for instance.

#7 Updated by mschramm 2 months ago

tnt wrote:

Huh ... don't you have TX/RX swapped on the SFP there ?!?

RD pins are the SFP receiver pins i.e. they are outputs.

You are right... sorry for the noise!

Changed this, then Rx and Tx pairs are not crossed anymore but the diff pairs swap P and N. Just tried it with SN65LVDT2 / SN65LVDS1, and no twisting takes place. A 'discrete' solution (separate Rx/Tx ICs) would also give the advantage of easier placing them close to the SMA and hence reducing the single-ended path length.

#8 Updated by laforge 2 months ago

3205
3206

Attaching two pictures of experiments dony by tnt yesterday

#9 Updated by mschramm 2 months ago

3207
3208

Committed libs for SN65LVDS1 and SN65LVDT2 (single port LVDS tx /rx) and will continue to design with them.

#10 Updated by mschramm about 2 months ago

  • % Done changed from 20 to 50

committed a first proposal for an SFP experimenter PCBA.

remarks:

  • the width of the differential signal traces and the two single-ended towards the SMA will change until production. The 'edge-coupled coplanar waveguide w/ GND' calculation will change it:
  • to achieve reasonable width values but stay with a 2-layer PCB, we might want to take a different FR4 thickness, e.g. 1mm or 0.8mm. Drawback on this is: different types of SMA edge conn are needed for those thin PCBs
  • the decoup caps on the driver/receiver might change to 0402 and probably move closer to the respective IC's supply pins
  • the selected pinout for driver and receiver in compatible with other manufacturer's types (PI90LVB01 / PI90LV02 , FIN1001/2 ...) I think
  • X3 and X5 won't be placed here ('residue' for/from #3313)
  • some receivers already integrate the 100R termination while others don't. An R footprint in front of IC3's inputs should be added which can stay unpopulated depending on the receiver type.
  • I don't insist on the LDO (TLV1117; loss would be about 1/2W) ;)

#11 Updated by laforge 17 days ago

13:25 < tnt> Any "real" design using the SFP would be 4 layer, so I guess it's an argument
             to make the experimentation board as close to what a final product would use.

so let's go four-layer. mschramm: please change layer stack-up, add gnd/vcc planes, and adjust trace widths for "usual" four-layer stacking.

#12 Updated by tnt 17 days ago

A couple of comments :

  • Add pin headers for power input/output as well (both before and after the regulator in case tapping the 3.3v can be useful to power something 'else').
  • Add leds for TX fault and LOS (with transistor to drive them, not directly from the signal of course)
  • Make sure a ground pin is available next to the broken out 'TX_DISABLE' pin. It's internally pulled up and so having a GND next to it would allow to easily use a jumper to force it low and force enable TX.
  • Having a couple of GND points to tie scope ground would be neat as well :p

#13 Updated by laforge 15 days ago

Also available in: Atom PDF

Add picture from clipboard (Maximum size: 48.8 MB)