It’s Always Good to Share – The Value of SD-FEC Gain Sharing

As children we are taught from an early age that it is good to share. It helps to build trust and relationships while demonstrating our capacity for kindness. It also leads to reciprocity, with other parties sharing with us when we are in need. Somewhat surprisingly, this same principle can also apply to DWDM transport.

Figure 1: Amplifier tilt and ripple

Even along the exact same transmission path (including fibers, amplifiers, ROADMs, etc.) between the same two endpoints, different wavelengths are likely to experience different impairments. Erbium-doped fiber amplifiers (EDFAs) do not amplify each wavelength identically due to tilt and ripple across the transmission band, as shown in Figure 1. Other impairments also differ from wavelength to wavelength.

For example, polarization-dependent loss (PDL) and polarization mode dispersion (PMD), which occur due to asymmetries in the fiber, are dependent on the exact angle at which the two polarizations of each wavelength enter the fiber, as shown in Figure 2, which will be different for each wavelength. Chromatic dispersion also varies with wavelength, which can be a particular challenge for dispersion-managed submarine cables, where chromatic dispersion may be very low at some wavelengths and very high at other wavelengths.

Figure 2: PDL and PMD depend on the exact angles of polarization

So now that we know all wavelengths are not treated equally, what can we do about this variation? Well, forward error correction (FEC) has long been used to correct errors that occur in noisy transmission environments, with a portion of the transmitted bits used to detect and correct errors, as shown in Figure 3.

Figure 3: Forward error correction

However, each wavelength traditionally has its own FEC, so any unused gain on one wavelength cannot be used to help another, more challenged wavelength. An Infinera innovation, SD-FEC gain sharing, changes all that. Leveraging a dual-wavelength coherent digital signal processor (DSP), the FEC gain is shared across two wavelengths, enabling a more challenged wavelength to be paired with a less challenged wavelength. But how does this work exactly? How does the DSP know how much gain to shift from one wavelength to the other?

Figure 4: SD-FEC gain sharing actually “shares” errors across the two frames

The answer is that it does not need to know.  What is really being shared are the pre-FEC errors. Let me explain. Suppose we have two frames, A and B, with A traveling over a less challenged wavelength and B traveling over a more challenged wavelength. B’s FEC has more “work” to do, and it may experience too many pre-FEC errors for it to be able to correct them all.

What SD-FEC gain sharing does is interleave the two frames, payload and FEC overhead, so that half of each frame goes over each wavelength, and therefore they each experience a statistically identical number of pre-FEC errors. Now, when the original frames are reassembled, the FEC decoders have the same amount of work to do, and the gain is the same. No need to work out how much “gain” or FEC overhead to transfer from one wavelength to the other.

So that’s how it works – but what is the economic benefit of SD-FEC gain sharing? Well, one scenario could be where even at the lowest modulation wavelength, B could not operate error free between the two endpoints, while wavelength A’s FEC decoder could tolerate a lot more pre-FEC errors before exceeding its maximum error correction rate.

A more common scenario is that the modulation on wavelength B could be downgraded, say from 8QAM (i.e., 150 Gb/s) to QPSK (i.e., 100 Gb/s) thus reducing the capacity of the two wavelengths to 250 Gb/s. With SD-FEC gain sharing, both wavelengths could run at the higher modulation, 8QAM, increasing the capacity of the two wavelengths by 50 Gb/s to 300 Gb/s (20%), as shown in Figure 5.

Figure 5: SD-FEC gain sharing enables higher capacity

So, when will this ingenious and valuable feature be available, I hear you ask? Well, the good news is that this feature has been available since the introduction of Infinera’s original 500G super-channels back in 2012, was carried forward into the record-breaking fourth-generation Infinite Capacity Engine (ICE4) optical engine in 2017, and will also be available in Infinera’s ICE6 optical engine currently under development, all of which leverage the dual-wavelength DSPs required for this feature.