Wavelength division multiplexing was originally constrained by the tyranny of a fixed grid, first at 100GHz, and as lasers improved at 50GHz, imposed by the inflexible nature of fixed filter technologies. A fibre could not mix schemes, which hindered the deployment and efficient utilisation of new transmission technologies. 50GHz was often too narrow, while 100GHz was wastefully wide.
The introduction of flexible grid technologies about a decade ago, like liquid crystal on silicon, enabled customising channels with varying widths all along a fibre, typically with 6.25GHz granularity. Most greenfield deployments today use flexible grid.
Flexible grid provides the ability to tailor the width of each channel on a fibre to provide a “best fit” for the transmission needs of the signal it carries. This depends on multiple transmission factors like desired line rate, modulation scheme, baud rate, and target distance. For example, a 200G coherent signal transmitting a few hundred kilometers may require only a 50GHz channel, but this may need to be widened to 75GHz to traverse several thousand kilometers.
In theory, flexible grid allows continuously optimising channel widths across the entire spectrum on a fibre. In practice, however, flexible grid is generally used in a “set and forget” fashion for operational reasons - keeping track of changes along multiple hops between ROADM-powered network nodes is onerous.
Spectrum re-assigned to a channel on one hop may not be available on the next, creating a need for complex re-shuffling across the entire network that becomes exceedingly difficult to manage. Ultimately, spectrum “defragmentation” becomes so complex that it can lead to abandoned, and thus wasted, spectrum.
In fact, there is a risk that we are heading to this situation as we move to a new generation of high-performance coherent transmission technologies. The culprit is 800G wavelengths.
800G wavelengths are a milestone achievement and have been the recent transmission grabbing the headlines. However, what the headlines do not say is that these 800G signals require a strange channel bandwidth of 112.5GHz. Yes, 112.5GHz, non-negotiable, dictated by immutable laws of physics. While this is deployable with flexible spectrum, one might see how it may cause problems down the road.
There is, however, a better way, which is to use flexible grid to create a spectrum channelization plan based on multiples of 75GHz. This makes sense for the following reasons:
Starting in 2021, the workhorse transceiver for 100GbE and 400GbE client transport in metro ROADM-based mesh optical networks will be CFP2 DCO pluggables based on an OpenROADM MSA. These cost-power optimised pluggables will be available from multiple suppliers and use a 75GHz channel width.
For applications where maximum performance is required – such as for dense metro traffic or long haul – merchant technology has been commercially available for more than 18 months that pairs 600G wavelengths in a single 150GHz super-channel.
That merchant technology will follow the silicon cycle and progress in the coming 18 months to deliver the same performance as dual 600G wavelengths in a single 1.2T wavelength, also within a 150GHz channel width.
The chart below compares these optical technologies for the 400GbE client transport that is on its way to become the predominant client traffic. (100GbE client transport has the same pattern but with four times as many intermediary steps.)
Noteworthy on the chart is that dual 600G wavelengths today provide superior capacity and distance performance than 800G wavelengths. Moreover, they use a channel plan that is 100% compatible with achieving the same performance using a single 1.2T wavelength some 18 months from now.
The spectrum implications below demonstrate that we can use an Mx75GHz channel plan to build a Nx400G optical highway combining both performance-optimized optical transmission (using proprietary modules) and cost-power optimised transmission (using pluggable modules). We can start building this today, and it is future proof to slot it 1.2T wavelengths.
Therefore, if you really want to deploy 800G wavelengths, it is probably better to assign these 150GHz channels, and not need to deal with spectrum fragmentation in the future.
