PECL is a low impedance output, and is designed to drive 50 ohm loads, to allow it to drive unbalanced terminated 50 ohm interconnects.
LVDS uses a 100ohms balanced termination resistance at the receiver, which for a differential signal is equivalent to each line having a 50ohm unbalanced load.
The series impedance of C139x is completely negligible, and the shunt 500ohm impedance of two R115x in parallel is only 10% of the 50ohm load, so fairly negligible.
This adds up to your 800mV swing being essentially unaltered at the LVDS input.
I would have cautioned against using AC coupling to connect digital data like this, but I notice the signals are labelled SYS_CLK2_300. Continuously running clocks are OK with AC coupling.
LVDS usually specifies a much smaller swing. I notice your FPGA appears to have a 1.2v rail. While this potential divider does place the signal in the middle of its voltage range, it may be worth checking whether the FPGA clock input is OK with this large swing, or whether it will cause any damage or malfunction.
If you do need to reduce the swing, small resistors in series with the PECL output is the right way to do it, they will pot down the signal into the transmission line (if any) that's terminated with the LVDS termination resistor. It's important to have them in the order of PECL driver, series resistors, 100ohm balanced interconnect tracks, LVDS receiver with termination. The Rs and Cs you have drawn are sufficiently different impedance from 50ohms to go anywhere along that path.
The first thing you should do is specify the acceptable timing error. The solutions to sync within 100ns or 10ps are very different. If you just need to trigger an interrupt within a few cycles in each processors, then differential signaling is unnecessary, a coaxial cable will do the trick.
The two GPIOs solution is risky. If you cannot guaranty that they will start their transitions at the same time, then they will have the same output and the difference will be 0. If the difference is 0, the output of the line receiver is undetermined, and could be rapid oscillation due to noise.
The line driver / receiver will give you much better timing performance, if required. You don't have to be paranoid, differential signals can be very robust. for exemple, LVDS is used extensively in the automotive industry. However, differential signals can tolerate only a certain amount of voltage shift, so if your ground is very noisy you could still have issues.
For synchronization down to a few nanoseconds, over a short distance, and with frequency below 1MHz, an LVCMOS signal over coaxial cable can give good results:
simulate this circuit – Schematic created using CircuitLab
You can add other protections as needed.
You also need to look at where you are going to input that signal to the microprocessor if you need synchronization better than what interrupts can give you.
If you are expecting ground loops, maybe work at the root of the problem instead and try to minimize the area between your ground connections.
Best Answer
That is a very odd datasheet. Normally when we talk about "absolute maximum" ratings, we mean this table from page 4 in the datasheet:
These are requirements for how you must use the chip to avoid damaging it.
The spec you are referring to is in a table labelled "AC Characteristics". Normally a "characteristic" spec is a promise from the chipmaker about the behavior of the chip, not a requirement for how you should use the chip.
As I read it, they are promising that the output differential voltage will absolutely not exceed 1150 mV. A figure on page 8 tries to clarify what they mean by the spec:
I can't say how this relates to the requirements for your other chip without seeing the datasheet for the other chip.