When we did intensity modulation, there was a time when some cables, particularly in Japan used NZDSF (Non Zero dispersion shifted fiber) type fiber. This was useful when we used OOK (on-off keying) and direct detection. With direct detection, we didn’t have access to the phase and could not compensate for dispersion all that well. So we used a DSF (dispersion shifted fiber) that has the opposite slope (sign) and cancels out the dispersion. This was imperfect and DSF was a source of nonlinearity due to its narrower core (nonlinearity goes as the time derivative of the intensity per unit area).

There is no perfect amount of dispersion. However allowing the dispersion to increase in modern cables also allows us to make the fiber core a little bigger. This reduces the intensity per unit area, reducing the fiber nonlinearity. Additionally there’s some benefit for highly dispersed signals as they look like noise.

For long distances we encode the information on the electric field phase and amplitude as us done in wireless. In actuality most of what’s done in optical from a modulation point of view and most of the theory comes from wireless originally. A difference is that we use a laser as a local oscillator to shift the data back down to baseband just like in wireless. We did intensity modulation for a long time because the signal speeds are very high and the semiconductor technology wasn’t there to do ADC and DACs at the rates we needed at low enough power and cost. This changed around 2008 when the first coherent systems became practical due to the availability of 45nm CMOS.

Now most everything above 25Gb/s and 80km is coherent. Below that we still do intensity modulation and inside data centers it’s all intensity modulation. This is due to cost and power reasons. Even in 7nm, coherent is still higher power on a bit/s basis. It’s also more expensive due to complexity of the transmitter and receiver as well as the need for better quality lasers.