FROM AN EMAIL BY GREG FERRO:
I have done some research on cable lengths for Gigabit as per our discussions and came up with the following:
From the Nortel Networks Accelar manuals you can find the following information about the 1000BaseLX Laser Xmitter and receiver standards.
Laser Transmitter Characteristics:
Minimum Launch Power -9.5 dBm
Maximum Launch Power -3 dBm
Receiver Characteristics:
Minimum Receiver Sensitivity -20 dBm
Maximum Input Power -3 dBm
Now from this we can deduce that the worst case scenario would be when the launch power from the laser is -9.5dBm and the receiver sensitivity is -20dBM, this gives us a worst case of 11.5dBm of "headroom" for _guaranteed_ function. In any given circumstance, launch power is affected by the quality of the fibre connector and the laser installed, not much we can do there. Receiver Sensitivity is determined by the amount laser power received at the other end, which is degraded by patch leads, fibre length, fusion splice or mechanical splice, types of connectors etc.
The interesting thing is that even on circuits where it is conceivable that we have more 11.5dB of loss things may well work, most likely because we are getting good launch characteristics i.e. up to -3dBm of launch giving up to 6.5 dbM of headroom. Potentially, the receiver could work on an even lower input power. Thus the recommended length of 550m for 1000BaseLX on 62.5um multimode could be longer depending on how good the end to end performance of the cabling system is.
Note that for 1000BaseSX the figures are different, this is why the shorter cable length comes into play, less power budget to work with, but in a given situation we may well be able to exceed the standard and get away with it.
Laser Transmitter Characteristics:
Minimum Launch Power -10 dBm
Maximum Launch Power -4 dBm
Receiver Characteristics:
Minimum Receiver Sensitivity -17 dBm
Maximum Input Power 0 dBm
The trick here is that the overall power budget can fluctuate with temperature and other environmental issues, and this may cause intermittent errors. The gamble with overrunning the standard lengths is that we risk intermittent and indeterminate faults, often with no apparent cause.
Hope this helps.
Regards
Greg