1
Space Division Multiplexing in Optical Fibres
D. J. Richardson
1
, J. M. Fini
2
and L E. Nelson
3
1
Optoelectronics Research Centre, University of Southampton, Highfield, Southampton, SO17 1BJ, UK.
2
OFS Laboratories,19 Schoolhouse Road, Somerset, New Jersey 08873, USA.
3
AT&T Labs - Research, 200 S. Laurel Avenue, Middletown, New Jersey 07747, USA.
Optical communications technology has made enormous and steady progress for several
decades, providing the key resource in our increasingly information-driven society and
economy. Much of this progress has been in finding innovative ways to increase the data
carrying capacity of a single optical fibre. In this search, researchers have explored (and close
to maximally exploited) every available degree of freedom, and even commercial systems now
utilize multiplexing in time, wavelength, polarization, and phase to speed more information
through the fibre infrastructure. Conspicuously, one potentially enormous source of
improvement has however been left untapped in these systems: fibres can easily support
hundreds of spatial modes, but today’s commercial systems (single-mode or multi-mode)
make no attempt to use these as parallel channels for independent signals.
The notion of increasing fibre capacity with Space Division Multiplexing (SDM) is almost as old as
fibre communications itself, with the fabrication of fibres containing multiple cores, the first and
most obvious approach to SDM, reported as far back as 1979
1
. Yet only recently has serious
attention been given to building a complete networking platform as needed to make use of this
multicore fibre (MCF) approach. The alternative approach of using modes within a multimode
fibre (MMF) as a means to define separate spatially distinct channels also dates back to that era
2
.
The current frenzied progress in SDM is occurring now because of a convergence of enabling
technological capabilities and a rapidly emerging need. On the one hand, SDM draws on the
accumulated progress of fibre research. This includes subtle improvements in traditional fibres
3
,
and the fantastically precise fabrication methods developed to produce hollow-core and other
complex microstructure fibres
4-7
. Sophisticated mode control
8
and analysis
9
methods along with
tapered devices
10
can be borrowed from high-power fibre laser research, which itself has needed to
develop means to better exploit the spatial domain in the drive to achieve ever higher power
levels
11
. Photonic lantern
12
and endoscope devices
13
are available from their development for
imaging.
Today’s SDM research is also occurring as coherent detection and digital compensation are capable
of overcoming complex impairments (such as polarization mode dispersion (PMD)) and are
accepted as a standard part of high-performance systems. This is crucial: since SDM packs spatial
channels tightly into each fibre, crosstalk between channels is an obvious potential disadvantage
and needs to be addressed. The addition of significant crosstalk to a transmission line would have
been particularly unattractive a few years ago, before coherent-detection systems offered hope of
subtracting out crosstalk electronically at the receiver.
2
Figure 1 | The evolution of transmission capacity in optical fibres as evidenced by state of the art laboratory
transmission demonstrations over the years. The data points shown represent the highest capacity transmission
numbers (all transmission distances considered) as reported in the Postdeadline Session of the Optical Fiber
Communications conference held each year in the USA. The transmission capacity of a single fibre strand is seen to
have increased by approximately a factor of 10 every 4 years. Key previous technological breakthroughs include the
development of low-loss single-mode fibres, the Erbium Doped Fibre Amplifier (EDFA), Wavelength Division
Multiplexing (WDM) and more recently high-spectral efficiency coding via DSP-enabled coherent transmission. The
data points for Space Division Multiplexing also include results from the Postdeadline Session at the annual European
Conference on Optical Communications (ECOC). As can be seen SDM appears poised to provide the next step change
in transmission capacity.
These enabling technologies have made SDM a viable strategy just as a severe need for innovation
emerges. Over the past forty years, a series of technological breakthroughs have allowed the
capacity-per-fibre to increase around 10x every four years, as illustrated in Figure 1. Transmission
technology has therefore thus far been able to keep up with the relentless, exponential growth of
capacity demand. The cost of transmitting exponentially more data was also manageable, in large
part because more data was transmitted over the same fibre by upgrading equipment at the fibre
ends. But in the coming decade or so, an increasing number of fibres in real networks will reach
their capacity limit
14
. Keeping up with demand will therefore mean lighting new fibres and
installing new cables - potentially also at an exponentially increasing rate. Further, this fibre
capacity limit is not specific to a particular modulation format or transponder standard - it is
fundamental and can be derived from a straightforward extension of the fundamental Shannon
capacity limit to a nonlinear fibre channel under quite broad assumptions
15
. It says that standard
single mode fibre (SMF) can carry no more than around 100Tbit/s of data, corresponding to filling
the C and L amplification bands of the erbium doped fibre amplifier (EDFA) at a spectral efficiency
of ~10 bits/s/Hz.
3
The upcoming potential “capacity crunch”, then, is an era of unfavourable cost scaling. For some
carriers who have access to a limited number of dark fibres, very expensive installation of new
cables will be the only alternative as the capacity of existing fibres is filled. “Fibre-rich” carriers
who attempted to future-proof their fibre plant by including large numbers of premium fibres in
each cable (thus putting off the need for subsequent new cables) will be forced to overbuild, i.e.
deploy multiple systems over parallel fibres, to keep up with demand. However, multiple systems
over parallel fibres suggest that transmission costs and power consumption will scale linearly with
growing capacity. The fear is that, without further innovation to lower the cost-per-bit, the capacity
crunch will apply pressure to constrain growth, and we will finally reach the end of the seemingly
boundless connectivity that drives our economy and enriches our experiences.
The anticipated promise of SDM is not only that it will provide the next leap in capacity-per-fibre,
as shown in Figure 1, but that this will concurrently enable large reductions in cost-per-bit and
improved energy efficiency
16
. This is a formidable challenge. SDM is very different from
wavelength division multiplexing (WDM) which inherently allows the sharing of key components:
e.g., an EDFA and dispersion compensation module can easily be shared by many WDM channels
with minimal added complexity. The benefits of SDM are more speculative, and assume that many
system components can be eventually integrated and engineered to support this potentially
disruptive new platform.
Given this emerging need, major research effort has been mobilized around the world to explore
and establish the viability of SDM
17
. Exciting recent results show that a wide array of new tools are
now being focused on probing the potential benefits of SDM, and chipping away at the many
engineering problems obscuring these benefits.
Technical approaches to SDM
The term SDM is nowadays taken to refer to multiplexing techniques that establish multiple
spatially distinguishable data pathways through the same fibre, although in earlier days the same
terminology was previously applied to describe the case of multiple parallel fibre systems: the
benchmark that needs to be beaten on a cost-per-bit perspective if any of the SDM approaches
currently under investigation are ever to be commercially deployed. The primary technical
challenge given the more intimate proximity of the pathways is management of cross-talk.
In the case of multicore fibre (MCF) in which the distinguishable pathways are defined by an array
of physically-distinct single-mode cores (Figure 2(b)) the simplest way to limit cross-talk is to keep
the fibre cores well-spaced. Small variations in core properties, either deliberately imposed across
the fibre cross-section
18
, or due to fabrication/cabling
19
, can also reduce cross-coupling along the
fibre length. As will be discussed later, to date, the highest capacities and longest transmission
distances demonstrated in SDM system experiments have all utilized such “uncoupled” MCFs. A
study of the tolerance of various advanced modulation format signals to in-band accumulated cross-
talk (to include contributions from signal multiplexing and demultiplexing, amplification, splicing
and distributed-coupling along the fibre length) showed that < -25dB cross-talk levels are typically
required to avoid significant transmission penalties
20
.
4
Using trench-type core refractive index profiles matched to standard SMF (Figure 2(g)), to better
confine the mode, it has proved possible to reduce core-to-core coupling to impressively low levels
(<-90dB/km) for a spacing of around 40µm, enabling transmission over multi-1000km length
scales
21
. However, fibre reliability issues, in particular susceptibility to fracture, mean that MCF
diameters beyond 200µm are not considered practical, placing a fairly firm bound on the number of
(f)
(g)
Figure 2 | Different approaches to SDM: (a) Fibre-bundles composed of physically-independent,
single-mode fibres of reduced cladding dimension could provide for increased core packing densities
relative to current fibre cables, however “in-fibre” SDM will be needed to achieve the higher core densities
and levels of integration ultimately desired. (b) MCF comprising multiple independent cores sufficiently
spaced to limit cross-talk. Fibres with up to 19 cores have so far been demonstrated for long haul
transmission – higher core counts are possible for short haul applications (e.g. datacomms) which can
tolerate higher levels of cross-talk per unit length. (c) FMF with a core dimension/numerical aperture set to
guide a restricted number of modes – so far typically 6-12 distinct modes (including all degeneracies and
polarisations). To date work has focussed primarily on using the first few LP-fibre modes; however, work
is now beginning on using other modal basis sets that exploit the true vector modes of the fibre – in
particular on modes that carry orbital angular momentum and which may provide benefits in terms of
reduced mode-coupling and associated DSP requirements
42
. (d) Coupled-core fibres support supermodes
that allow for higher spatial mode densities than isolated-core fibres. MIMO processing is essential to
address the inherent mode-coupling. (e) Photonic Band Gap fibres
4,5
guide light in an air-core and thus have
ultra-low optical nonlinearity, offer the potential for lower losses than solid core fibres (albeit at longer
transmission wavelengths around 2µm
rather than 1.55µm
43,44
). Work is underway to understand whether
such fibres can support MDM and to establish their practicality for high capacity communications. (f)
Refractive index profile of a GI core design providing low DMGD and low mode-coupling for long haul
FMF transmission
29
. (g) Core refractive index design incorporating a trench profile to reduce cross-talk and
thus allow closer core separations in MCF
21
.
5
cores that can be incorporated in MCFs for long-haul transmission. Most fibres to date have used a
hexagonal arrangement of 7-cores for which the central core, with six nearest-neighbours as
opposed to three for cores in the outer ring, suffers the highest level of cross-talk. More recent
work
22
has used 12-cores arranged in a ring geometry such that each core has just two nearest-
neighbours and experiences nominally the same level of cross-talk (-57 dB/km in this case). A 19-
core fibre of 200µm outer diameter has also been reported; however the cross-talk was already
substantially higher and limited the useful transmission distance to ~10km
23
.
The situation is quite different for mode division multiplexed (MDM) transmission in MMF where
the distinguishable pathways have significant spatial overlap and, as a consequence, signals are
prone to couple randomly between the modes during propagation. In general the modes will exhibit
differential mode group delays (DMGD) and also differential modal loss or gain. The energy of a
given data symbol launched into a particular mode spreads out into adjacent symbol time slots as a
result of mode-coupling, rapidly compromising successful reception of the information it carries.
Crosstalk occurs when light is coupled from one mode to another and remains there upon detection.
Inter-symbol interference occurs when the crosstalk is coupled back to the original mode after
propagation in a mode with different group velocity. As in wireless systems, equalization utilizing
multiple-input multiple-output (MIMO) techniques
24
is required at the receivers to mitigate these
linear impairments.
MIMO signal processing is already widely used in current coherent optical transmission
systems with polarization division multiplexing (PDM) over standard single-mode fibres. A 2x2
realization with four finite impulse response (FIR) filters recovers the signals on the two
polarizations and compensates for PMD in the link
25
. For an MDM system with M modes, the
respective algorithms would need to be scaled to 2Mx2M MIMO, requiring 4M
2
adaptive FIR
filters. (By way of comparison, the same capacity carried on M uncoupled SDM waveguides would
require 4M adaptive FIR filters.) Thus, if we assume an equal number of taps per adaptive FIR filter
and equal complexity of the adaptation algorithm, comparing a 2M×2M MIMO system on M
coupled waveguides to M uncoupled SDM waveguides using PDM results in a complexity scaling
26
as 4M
2
/(4M) = M . To compensate DMGD and mode cross-talk completely, the equalization filter
length should be larger than the impulse response spread. The computational complexity of FIR
filters implemented as time-domain equalizers (TDE) increases linearly with the total DMGD of the
link
25
, which can make TDE unfeasible for long-haul MDM transmission. Common equalizer
algorithms were studied and orthogonal frequency division multiplexing (OFDM) was found to
achieve the lowest complexity
27
. However, for OFDM the DMGD to be compensated (and thus the
reach) is limited by the length of the cyclic prefix. In other work aimed at lowering the DSP
complexity, single-carrier adaptive frequency-domain equalization (SC-FDE) for MDM
transmission has recently been proposed
28
, where the complexity of SC-FDE scales logarithmically
with the total DMGD.
Conventional MMFs with core/cladding diameters of 50/125 and 62.5/125um support more
than 100 modes and have large DMGDs, and thus are not suitable for long-haul transmission
because the DSP complexity would be too high. Recent advancements have led to fibres supporting
a small number of modes, the so-called “few-mode fibres” (FMFs), with low DMGD (see Figure
2(c)). The most significant research demonstrations have so far concentrated on the simplest FMF,
which supports three modes, the LP
01
and degenerate LP
11
modes, for a total of 6 polarization and
