contact button

NANOG 85 Tutorial: Optical Fiber Capacity Limits – Where Do We Go Next?

portrait of Geoff Bennett

October 6, 2022
By Geoff Bennett
Director, Solutions & Technology

Submarine Communication Cables

In this series of blogs, I’ve looked at fiber capacity limits and the options for increasing capacity on existing fiber as well as on new types of fiber. To finish the series, I’d like to look at the special case of submarine cables and start by answering the obvious question – why are they a special case?

Unlike the amplifier chains in terrestrial optical networks, submarine amplifier chains receive their power from the ends of the cable. So, this leads to several power-limiting conditions on the total number of fiber pairs in the cable. Note that these criteria are decided when the cable is designed – and once the cable is under the ocean they cannot be changed. These limits include:

  • How many amplifiers are in the total cable length?
    • The shorter the amp spacing the higher the total power consumption, but the better the fiber pair performance.
    • But the more amplifiers you have in a given cable, the fewer fiber pairs you will be able to support.
  • Are the amplifiers operating at high power levels in order to support high-spectral-efficiency modulation?
    • The higher the amp power, the more likely we can support higher QAM types or use more of the PCS constellation – in other words, the higher the spectral efficiency.
    • But the higher the power levels for a given length of cable, the fewer fiber pairs you can deploy.
  • Do these amplifiers have dedicated backup pump lasers (which has been the usual case historically), or do they employ some type of backup pump sharing (also known as pump farming, which is a new approach)?
    • Pump farming has no impact on optical performance – it is purely a way to conserve electrical power in the cable.
    • The use of N for M backup pumps is purely a matter of the industry accepting that this approach will deliver the required reliability levels – something that seems to be happening given the deployment of space-division multiplexing (SDM) cables today.

Characteristics of Multiple Generations of Transatlantic Submarine CablesFigure 1: Characteristics of multiple generations of trans-Atlantic submarine cables

Optimizing these factors allows wet plant designers to increase the number of fiber pairs for a given length of cable. In Figure 1 I show how this has happened over the past 20 years of submarine cable development. All four cables are about the same length (around 6,600 km) as they are classic trans-Atlantic routes. Let’s take a look at this table in more detail.

Dispersion-managed Cables

Apollo is an example of a dispersion-managed cable. These cables were designed long before coherent technology was developed and had to deal with the challenge that the 2.5G and 10G direct-detect transponders of the day could not tolerate high levels of chromatic dispersion in the cable. In these cables, the level of dispersion was controlled by alternating lengths of positive- and negative-dispersion fibers. For Apollo, this enabled transmission up to 10G and fiber pair capacity of around 650G before the development of coherent transponders.

By deploying the latest fifth-generation transponders on Apollo, we can increase fiber capacity to around 11 Tb/s – which is an amazing boost to the economic life of the cable considering it was not designed for these types of transponders. Note that the 11 Tb/s limit comes from a number of sources, including the fact that low dispersion levels and the lower effective area of negative-dispersion fibers will lead to higher nonlinear penalties. In addition, the amplifiers used in Apollo have lower bandwidth than those used in MAREA and later cables.

Uncompensated Cable Designs

When MAREA was designed, it was envisioned to be the perfect cable to support high-spectral-efficiency transmission using high-order QAM modulation such as 8QAM and even 16QAM. There were four primary design decisions that helped to achieve this performance level:

  • The use of extremely high-quality fiber with ultra-low attenuation (less than 0.15 dB/km) and very large effective area (150 square microns). Large effective areas lead to lower nonlinear penalties.
  • All this fiber is positive dispersion, and high dispersion along the cable length also contributes to a lower nonlinear penalty.
  • Short spacing between amplifiers – around 56 km.
  • High amplifier pump power and a wider amp bandwidth of around 4.8 THz.

In fact, MAREA was the first trans-Atlantic cable to be able to support 16QAM modulation – previous cables were typically limited to QPSK. When it came into service in 2018, MAREA alone offered as much capacity as all the existing trans-Atlantic cables combined. Over time we saw several record performance numbers published on MAREA, first using Infinera’s ICE4 to achieve 24 Tb/s of commercial capacity (including end-of-life operating margins), and more recently with ICE6 reaching 28 Tb/s of commercial capacity on a single MAREA fiber pair. Modern fifth-generation transponders like ICE6 are so successful at extracting high spectral efficiency from this type of cable that we are rapidly approaching the practical capacity limit for a single fiber pair. Despite the fact that MAREA does not use pump farming and has no compromises in amplifier power levels, advances in power feed equipment (PFE) such as higher voltages and dual-ended feed mean that it could support 8 fiber pairs – twice as many as Apollo.

C+L in Submarine Cables

At this point in the story, I’m going to divert from the structure of Table 1 and talk about the use of C+L in submarine cables. In Part 2 of this blog series, I discussed lighting other fiber bands – the O-, E-, S-, L-, and U-bands. Of these, the L-band is by far the most practical because attenuation and dispersion levels are still very favorable in the L-band, and we know how to build cost-effective L-band EDFAs. In fact, C+L systems are now the normal approach in terrestrial long-haul networks, where power can be supplied independently at each amplifier location.

As I mentioned at the start of the blog, submarine cables are different because they are constrained by electrical power. If we designed a cable to include C+L repeaters (such as the PLCN cable system), we could almost double the capacity per fiber pair. But we would also double the number of amplifiers in each fiber pair by adding L-band EDFAs. So, in a power-constrained cable, lighting the L-band does not increase the total cable capacity. In fact, the total capacity in a C+L submarine cable is slightly lower compared to an equivalent C-band-only cable because of the band coupling losses at each amp location. In the case of PLCN there was a logic to the C+L design because half the amount of fiber was used to achieve a given capacity. Note that PLCN is not included in the Table 1 comparison because it is 11,806 km long – almost twice the length of MAREA or Dunant, so such a direct capacity comparison would be misleading.

It is possible that the industry will revisit the idea of C+L in submarine cables at some point in the future if the wet plant industry finds a way to relax some of the power constraints in a submarine cable, but for the moment, the stated direction for scaling submarine cable capacity is SDM, and Dunant is the first example of a trans-Atlantic SDM cable.

The SDM Design Approach

In SDM, the focus is shifted away from maximum capacity per fiber pair toward maximizing total capacity per cable. To this end, the repeater spacing may be longer than in a cable like MAREA, the repeater power will be lower, and backup pumps will be shared rather than being dedicated one for one. But, of all these factors, it is the pump sharing that contributes most of the power saving.

As you can see from Figure 1, using the same ICE6 technology that yields 28 Tb/s per fiber pair on MAREA, we see around 26 Tb/s per fiber pair on Dunant. But there are 50% more fiber pairs on Dunant, so the total cable capacity can continue to scale.

One other effect of SDM is that the optical power levels used are well within the linear regime of the fiber, which means that very large-effective-area fibers are not necessary. This means that a cost-optimized fiber type can be used, typically with an effective area of around 80 square microns.

From Copper to Aluminum

Historically the conductor used in submarine cables has been copper. But if it is possible to switch to aluminum then there are significant advantages. Aluminum offers a lower voltage drop for unit distance compared to copper and it is far cheaper as a raw material. Dunant is a first-generation SDM architecture and as such it uses copper. The move to aluminum has to be backed up by extensive testing of the conductor in real submarine conditions in order to ensure the expected 25+ year engineering life of the cable.

Is There a Physical Limit of Fiber Pairs in a Submarine Cable?

The SDM scaling philosophy is based on increasing the number of fiber pairs in a given cable, but is there a physical limit? Deep water submarine cables typically have a diameter of 17 mm, but 20-mm cables are also available. The 17-mm option is preferred in order to optimize material cost and loaded cable weight on cable-laying ships. The standard fiber types for high-performance submarine operation have an outer diameter of 250 microns, and using this type of fiber up to 16 fiber pairs can be fitted into 17-mm cables, and 20 or more fiber pairs into a 20-mm cable. Note that trying to pack more fibers than this into the cable would most likely introduce microbending losses in the fibers and reduce optical performance significantly. One solution is to move to a smaller-diameter fiber.

In terrestrial networks it is already commonplace to use fiber with an outer diameter of 200 microns, and this diameter of fiber could also be used in submarine cables without significant loss in optical performance. The 20% reduction in fiber diameter leads to a 50% increase in the number of fibers that can fit into the cable without incurring microbending losses.

So, by moving to a 200-micron fiber, it would be possible to fit up to 24 fiber pairs into a standard 17-mm deep water cable without an increased risk of microbending losses, and this is the most likely approach for the next generation of SDM trans-Atlantic cable.

Fiber manufacturers like Corning have described how fiber thickness could be further reduced. This is partly by decreasing the fiber cladding diameter from 125 microns to 100 microns, and further decreasing the coating diameter to as low as 130-140 microns. Assuming that power optimization techniques will allow it, we can see the potential to scale to a 48-fiber pair trans-Atlantic cable using technologies that are well understood today.

At some point in the SDM roadmap, there will be a limit on the number of conventional fibers that can fit into existing cable designs, also taking into consideration the fact that cable repair times are increased with higher fiber count cables. At this point, and assuming there is a way to deliver more electrical power in the cable (using pump farming and aluminum conductors, for example), it may be time to revisit C+L transmission. But another option would be to look at novel fiber types.

Novel Fiber Types

In Part 3 of this series, I looked at the idea of novel fiber types, such as multi-core fiber (MCF) and hollow-core fiber (HCF) in terrestrial systems.

One of my conclusions was that there does not seem to be a compelling case for the use of MCF in terrestrial networks because the type of MCF available today solves a spatial efficiency problem, and that problem does not really exist in terrestrial networks.

Spatial efficiency will be a problem in future SDM designs, however, because of the cable diameter issue I just described. If we could increase the number of fiber cores without increasing the outer fiber diameter, that would open up a new dimension of cable capacity scaling. Existing experience with MCF seems to indicate that a two-core or four-core design should allow an uncoupled operation – which means there is minimal crosstalk between cores. If uncoupled operation applies it means that there is no need to develop special compensation algorithms in submarine transponders. This is an important issue because, if there really is no use case for MCF in terrestrial networks, it is unlikely that there is enough of a market in submarine “coupled MCF” implementations to justify the development of a submarine-specific MCF compensation ASIC.

HCF could be extremely interesting in the long term in submarine cables. HCF offers much lower latency than glass core fiber and, potentially, much wider low attenuation bandwidth. However, there are many questions to answer about HCF before these advantages can be exploited in future submarine cable architectures.

Both MCF and HCF will be considerably more expensive per kilometer than conventional fibers. And the most efficient and highest capacity amplification options for both types of fiber are still in the conceptual stage. While both approaches have the promise of further capacity scaling, it may be that they do not offer a significantly lower cost per bit because of these challenges.


SDM is the clear direction to scale submarine cable capacity for the foreseeable future. In fact, it seems possible that a 1 Pb/s trans-Atlantic cable could be deployed using technologies that are well understood today. Ultimately this familiarity may be the most important factor in submarine networks simply because of the expectation of extremely high levels of reliability over a 25-year service life for deployed cables.

In this series of blogs I’ve tried to outline the options for any optical network operator who is concerned about reaching “full capacity” on some or all of the links on their network today. I hope I’ve shown that there is a lot of capacity still to be tapped in existing fiber – first through tracking the coherent transponder evolution cycle, and then through upgrades such as C+L and even into other fiber wavebands.

When it comes to new types of fiber, the future is certainly promising, and we may well see a bifurcation of technologies, with hollow-core fibers finding applications in terrestrial networks first, and multi-core fibers being deployed in future submarine cable systems as an evolution of space-division multiplexing. A key point of reassurance is that, in almost all cases, current transponder technology can operate over these new systems without special developments.