Hacker News Reader

Comment by centimeter | Hacker News Reader

Yeah, you periodically need amplification/signal conditioning hardware in the cable. If it was just fibers you could put a ton in there.

The amplification is obviously there for losses in the cable, but you also need to do clock recovery and thresholding because the light pulse gets spread out as it travels.

Higher data rate => More analog bandwidth (distribution of frequencies from the laser)

More analog bandwidth => Worse chromatic dispersion

You need to either:

* Increase the frequency of the laser (so BW/center frequency is smaller, hopefully reducing dispersion), which might not be possible with current tech or be very expensive

* Decrease the dispersion of the fiber (more expensive or lossy fiber)

* Have periodic signal conditioners

View on HN

It sounds like you are describing direct detect systems. These are no longer deployed in subsea.

Due to the advent of coherent optical transmission and DSP, dispersion is no longer an issue for long distance optical communication. The limitation becomes nonlinear “crosstalk” between wavelengths and lower dispersion makes the nonlinear crosstalk worse as the phase matching condition for nonlinear processes can be reached more easily. To avoid this most new subsea cables use a high dispersion design as this helps mitigate nonlinear effects.

Additionally coherent systems have a higher spectral efficiency and are transmitting >2bits/symbol because the information is encoded in the electric field’s phase and not just intensity. Depending on the reach, the number of bits per symbol can be increased. The advent of probabilistic shaping, allows more bits per symbol to be transmitted in higher order constellations. Probabilistic shaping produces a more ‘random-like’ coding that gets us closer to the Shannon Limit and also reduces the optical intensity rms further reducing nonlinear penalties. In fact in optical communications there is something called the nonlinear Shannon limit that takes into account the additional “noise” produced by nonlinearities in the fiber. See:

https://arxiv.org/abs/1802.09120

I consider myself fairly technical but this read like something out of Silicon Valley (Weissman score!) to me. Hats off!

Wow this is the kind of comments I'm visiting this site for. This includes your parent comment as it made you write the comment :).

Dispersion is no longer an issue but lower dispersion increases crosstalk so new subsea cables are high dispersion? High compared to early days when dispersion was a problem or just high compared to low dispersion cables which suffer from crosstalk? So there is the perfect amount of dispersion now?

It never occurred to me that fiber optic communication in glass actually used intensity to transmit bits. This sounds like AM radio albeit digital I guess the laser could be on or off. How long has it worked like that?

I always assumed it was like CDMA radio where like you say phase shifts correspond to a point in a constellation which represents more than 1 bit.

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.

TIL! Thanks for sharing.

Reply on news.ycombinator.com