If you want to ensure isolated power from your Soekris boards, can always get a DC Regulator Board, with dual 5V/3.3V output... similar to the MIKROE-192 that itscraig has, and feed it either from a wall-wart or a dedicated power supply for both voltages...
Or, you can take the 3.3v from the Soekris board:
Took a while to figure out what on earth is going on with the coax input.. then it made sense. Soekris decided to use the FPGA's native LVDS receivers instead of an external one, which is an unusual step and a bit brave since the noise doesn't go through a filtered buffer first but only relies on the transformer's bandwidth filtering. However, the FPGA has blazingly fast LVDS receivers so it's a sensible step. The 1.2V biasing via the 1k resistor is since that's the centre voltage of LVDS, capacitor is just to clean it up a bit. The 330R termination resistor is because the LVDS receiver has integrated 100R resistors, and the parallel combo gives the required 75ohms.
Personally, I would consider the following changes:
- Use a centre-tapped transformer and apply the bias to the centre-tap instead
- Buffer the bias voltage with a low-noise opamp since the 1.2V rail is not exactly clean
- Add a DC-block to the input since this purely relies on the transmitter to be 100% DC-free (which it admittedly is most of the time, but pulse transformers saturate easily from even small DC currents)
- Perhaps experiment with an external LVDS repeater chip
- Add more filtering to the optical input, either with just a small chip inductor or a dedicated small LDO