Encoding information in the phase and amplitude of the light and recovering it with a local oscillator and digital signal processing - which is what makes very long distances possible.

Conventional optical transmission switches the light on and off. It is simple and works well over shorter distances, but chromatic and polarisation dispersion accumulate over long fibre runs and eventually make the signal unrecoverable.

Coherent detection mixes the received light with a local laser, preserving phase information, and digital signal processing then compensates for dispersion electronically rather than requiring optical compensation in the fibre path.

That allows far greater reach, higher capacity per wavelength through advanced modulation, and simpler line systems.

The cost is electronics: the processing is substantial and consumes significant power, which is exactly why coherent modules were large. As process technology improved, coherent optics moved into smaller formats - but the underlying trade between reach, capacity and power remains.

When the application needs more power and thermal capacity than a QSFP cage can provide - long-haul coherent transmission being the main case.

QSFP28 and QSFP-DD cover the data centre and campus comprehensively, and coherent optics in QSFP-DD form now cover a useful range of metropolitan distances.

CFP2 remains where the requirement exceeds that: longer reaches, higher capacity per wavelength, more complex modulation, or transport platforms designed around the format.

The practical answer is usually determined by the equipment. Transport platforms specify which module formats their line cards accept, and the choice is made when the platform is selected.

For an organisation buying a data centre interconnect service or building one, the question to ask is what the platform supports and whether the reach and capacity required can be met by a smaller format - because if it can, the smaller format will be cheaper, denser and easier to cool.

Successive reductions in size as the electronics shrank, each with less power capacity and more density than its predecessor.

The original CFP was physically large with a correspondingly large power budget, suited to first-generation 100 Gb and coherent applications.

CFP2 is roughly half the size with a reduced power envelope, and became the mainstream format for coherent transport - the generation most widely deployed in dense wavelength systems.

CFP4 is smaller again, approaching QSFP28 density, and suited to applications whose power requirement had fallen accordingly.

They are not interchangeable: each has its own connector and cage, and a host supports one.

In practice the format is dictated by the transport platform's line cards. What matters when specifying is matching the module generation, the reach and modulation capability, and the tuning range to what the platform and the optical line system require - and confirming the platform's compatibility list rather than assuming generic interoperability.

Coherent CFP2 modules generally are, and tunability is essential to how dense wavelength systems are operated.

A dense wavelength system carries many channels on one fibre, each at a precise frequency. A tunable module is set to its channel by configuration rather than by part number.

That matters for two reasons. Sparing: one module type covers every channel, rather than stocking one part per channel across a network. And operations: channels can be reassigned, moved around faults, or reconfigured for capacity without physical work.

Modern coherent modules often also allow the modulation format and line rate to be selected, trading reach against capacity - a longer link at a lower rate, or a shorter link at a higher one, from the same hardware.

The operational requirement is disciplined channel management. Setting a module to an occupied wavelength disrupts a live service, so assignment must be controlled and recorded in the optical line system's documentation.

Substantial and engineered - these are among the highest-dissipation pluggable modules made, and the host platform is designed around them.

A coherent CFP2 can dissipate many times what a short-reach QSFP28 does. The cage, the heatsink arrangement and the line card's airflow are designed for that specific figure.

The practical implications are for the installation rather than the module. Confirm the platform's stated inlet temperature limits and airflow requirements, ensure the cabinet provides them, and do not block the airflow path with cable management.

Population rules apply on some platforms - a line card may not support every slot filled with the highest-power module, or may require specific slot ordering.

Monitor module temperature through the platform's diagnostics. A module running near its limit will show rising error correction activity before it fails, which is a usable warning if anybody is watching for it.

As the transport layer between sites, carrying many wavelengths over a fibre pair across distances that switch optics cannot reach.

Data centre interconnect joins facilities tens or hundreds of kilometres apart, typically over leased fibre where capacity per strand matters commercially.

Coherent modules in a transport platform put many high-capacity wavelengths on one fibre pair, so a single leased pair carries a great deal of traffic. That is the economic argument.

The alternative for shorter distances is direct connection using long-reach switch optics, which is simpler and cheaper where the distance and capacity allow - and increasingly viable as coherent optics appear in QSFP-DD form.

So the design decision is whether the requirement justifies a transport layer at all. A single link over a few kilometres does not; multiple high-capacity links over a hundred kilometres on leased fibre certainly does, and the CFP-class module is what makes it work.

Progressively, as coherent electronics shrink - but they remain where the reach and capacity requirement exceeds what smaller cages can cool.

Coherent optics in QSFP-DD form now cover metropolitan and regional distances that previously required CFP2, and they do so at data centre switch density. That has absorbed a meaningful share of the applications.

What remains with CFP-class modules is the long-haul and high-capacity end, where the processing and optical performance still need the power envelope.

The direction is clear and follows the same pattern as every previous generation: as process technology improves, function moves into smaller packages, and the larger format retreats to the demanding edge of the application space.

For anyone operating CFP-based transport today, the practical guidance is the same as for any long-lived platform: know what is installed, hold spares for the specific variants, and plan replacement around the platform's support lifecycle rather than around the module format.