Produce a 100MHz clock signal locally, and distribute this to 16 50Ohm inputs. Or input a clock signal (5-200MHz) and output this to the same 16 channels. Has a jitter of less than 200fs. Vpk-pk = 0.4V min.
200fs is probably pushing it even for ECLinPS. See
for what you can do with reasonably off-the-shelf parts.
ON semiconductor also does GigaComm parts, which I've not used, but John Larkin has
It seems that they are appreciably faster (and more expensive). They were available off the shelf from Digikey but today Digikey didn't recognise the part number NBSG16VS - Newark did, but indicated zero stock and a 64 day lead time.
What are the physical distances involved? If the distrib is box-to-box through cables, the 200 fs is unrealistic, although local pll regeneration at each destination would help. 200 fs is pretty thin air.
Even if the cables were RG-405 semi-rigid cable or Nexans Quickform 141 conformable cable with PTFE (Teflon) dielectric the attentuation for the high frequency components of your signal would be pretty high - Qickform 141 has an attentuation of 1.95dB per metre at 20GHz (corresponding to a transistion time of 8 psec).
You might do better with super-conducting cable, where the decreasing skin depth won't be quite as lethal, but it sounds very hairy.
Would it be possible to use a VGA (or any other type of amplifier), and feed the output to a power splitter (eg. those ones from MiniCircuits)? Would the amp's noise add significantly to the jitter? How could I calculate this?