You are on page 1of 11

WH I TE PA PER

Maximizing Spectral Efficiency with


Optical Engine and Line System
Innovations

Coherent optical engine technology is evolving along three vectors


related to performance, space and power, and the integration of systems-
level functions. Regarding the performance vector, much of the focus has
been on wavelength capacity-reach, which has a direct impact on DWDM
interface cost per bit, watts per bit, and rack units per bit. However, for many
applications, spectral efficiency is a more important, or at least as important,
performance criterion. This white paper focuses on why spectral efficiency
matters, how the relationship between wavelength capacity-reach and
spectral efficiency has changed, and the three key factors that determine
spectral efficiency.
W H I T E PA P E R

What Is Spectral Efficiency?


1. Raw Bits Per Symbol

2. Overhead 2. Overhead
(FEC, Framing, etc.) (FEC, Framing, etc.)

3. Wasted Spectrum

Data Data
Payload Payload

Spectrum

Figure 1: The three factors that determine spectral efficiency

Spectral efficiency describes how many bits per second of effective data payload can be transmitted in a
given amount of spectrum for a given set of reach conditions. As shown in Figure 1, spectral efficiency is
determined by three things. First is the number of raw bits per symbol, which is directly proportional to the
number of raw bits per second per Hz. The second factor is the overhead efficiency, that is the percentage
of these bits that can be used for the data payload as opposed to the overhead for functions such as
forward error correction (FEC), framing, performance monitoring, and in-band management. The third factor
is how much spectrum is “wasted,” which includes both the unused spectrum between wavelengths and
spectrum wasted by the wavelength itself due to its shape relative to an ideal wavelength where the symbol
rate in Gbaud exactly matches its spectrum in GHz (i.e., 0% roll-off), as illustrated in Figure 2.

Actual Wavelength Actual Wavelength

Wavelength Spectrum Waste


(Roll-off)
Ideal Ideal
Wavelength Wavelength
Unused Spectrum

Total Wasted Spectrum

Spectrum

Figure 2: Wasted spectrum

Maximizing Spectral Efficiency with Optical Engine and Line System Innovations 2
W H I T E PA P E R

Why (and When) Spectral Efficiency Matters


Spectral efficiency can be very valuable. For example, with 4,800 GHz (4.8 THz) of C-band spectrum on a
fiber pair between two locations 1,000 km apart, say Chicago and New York City, if solution A can deliver 4
bits/s/Hz while solution B can deliver 5 bits/s/Hz, then solution B can deliver 4.8 Tb/s of additional capacity
(24 Tb/s vs. 19.2 Tb/s), equivalent to an additional 48 x 100 GbE, that can be used for revenue-generating
services. However, the relative importance of spectral efficiency varies with the application, the availability of
additional fibers, and the cost to light those fibers. And while optical performance is not the only purchasing
criterion, spectral efficiency is typically the most important performance criterion in submarine networks
and a key performance criterion in long-haul networks, as shown in Table 1. In metro networks, ROADM
cascadeability tends to take priority as performance criteria unless fiber availability is constrained and
bandwidth requirements are very high. In metro DCI applications, wavelength capacity-reach tends to be the
top performance criterion unless fiber availability is constrained.

Spectral Efficiency Wavelength Capacity-Reach ROADM Cascadeability

Submarine ✓✓✓ ✓✓ (Landing to POP)

Long-haul ✓✓ ✓✓ ✓

Metro ✓ ✓ (if fiber constrained) ✓ ✓✓✓

P2P Metro DCI ✓ ✓ (if fiber constrained) ✓✓✓ x

Table 1: Typical optical engine performance priorities by application

Spectral Efficiency and Submarine Networks


Spectral efficiency is typically the key optical engine metric for submarine networks. One reason for this is
the very high cost to deploy a submarine cable system, for example, several hundred million dollars for a
transoceanic submarine cable system. A second reason is the five to seven years it typically takes to plan
and deploy such a cable system. A third reason is that the number of fiber pairs in a submarine cable is
limited by the need to provide submarine repeaters (i.e., amplifiers) and, most critically, to provide electrical
power to these repeaters. Transoceanic submarine cables have evolved from two to between four and eight
fiber pairs per cable following a philosophy of maximizing spectral efficiency (i.e., capacity per fiber pair).
New space-division multiplexing (SDM) cable systems are increasing the number of fiber pairs in the cable
to between 12 and 24, and possibly even more in the future, with each fiber operating with less total power
for amplification and therefore a lower individual fiber capacity, but with the additional fibers increasing the
total bandwidth capacity of the cable system. Spectral efficiency can also be very important in spectrum
sharing scenarios, that is where an operator acquires a chunk of spectrum, say 500 GHz, on the submarine
fiber pair. This operator will then typically want to maximize the data capacity they can achieve from that
spectrum.

Spectral Efficiency and Terrestrial Networks


In terrestrial networks, maximizing fiber capacity and spectral efficiency becomes an issue as the C-band
spectrum of a fiber is maxed out. How big an issue it is depends on whether the fiber is owned or leased,
whether additional fibers are available, the cost of these additional fibers, and the cost to light additional
fibers with optical line systems equipment such as mux/demux filters, ROADMs, and amplifiers. Another
factor is the length of the fiber path, with a longer path incurring more cost and longer delays.

Maximizing Spectral Efficiency with Optical Engine and Line System Innovations 3
W H I T E PA P E R

This is illustrated by the following four scenarios:

Scenario 1: Operator owns cable/duct with spare fibers


If the operator owns the fibers/ducts with spare unlit fibers, then there is an opportunity cost to the fiber – it
could have been leased to a third party as dark fiber – in addition to the cost to light the fiber. There will also
be a cost in terms of time, as fiber characterization may be needed and line system equipment may need
to be ordered, installed, and commissioned. In this scenario, the value of increased fiber capacity/spectral
efficiency will be lower relative to the other scenarios, at least for shorter distances. However, with longer
fiber routes, the higher cost of lighting the fiber makes increased fiber capacity/spectral efficiency more
valuable.

Scenario 2: Operator leases fiber and the cost of additional fibers is low
Depending on the regulatory regime, location, and number of competitive dark fiber providers, the
availability of additional fiber may be high and the cost low. This is often the case between data centers in
major metropolitan areas in Europe and North America. For example, dark fiber is typically readily available
and very cost-effective between the 20+ data centers of London’s Docklands. In this scenario, fiber capacity/
spectral efficiency is less critical relative to scenarios 3 and 4.

Scenario 3: Operator leases fiber and the cost of additional fibers is high
As the cost of leasing additional fibers increases with a less competitive environment or longer distances,
the ability of increased spectral efficiency to avoid the cost and delay of new fibers becomes very valuable.

Scenario 4: Additional fiber is not available


Spectral efficiency will be most valuable if additional fiber is just not available for lease or an operator that
owns fibers/ducts has used up all the fibers. In this case, one option may be to put in additional fibers/ducts,
with high costs and delays related to planning, materials, rights of way, and installation. However, this may
not even be an option due to factors such as the regulatory regime, an inability to secure rights of way, or
financial constraints.

Wavelength Capacity-Reach and Spectral Efficiency


Bits/s Per Hz (Bits Per Symbol)

Better Spectral Efficiency (Same Reach)

Better Wavelength Capacity-Reach

500 Gb/s 500 Gb/s 500 Gb/s 500 Gb/s

800 Gb/s 800 Gb/s

Spectrum

Figure 3: Wavelength capacity-reach vs. spectral efficiency (for the same reach)

Wavelength capacity-reach and spectral efficiency are not the same thing. Wavelength capacity-reach
describes how much capacity you can get out of the optical engine for a given reach requirement, or more
specifically the optical penalties, such as amplified spontaneous emission (ASE) noise from optical amplifiers
and fiber nonlinearities such as cross-phase modulation (XPM) and self-phase modulation (SPM) that
accumulate as the wavelength traverses its path through the network.

Maximizing Spectral Efficiency with Optical Engine and Line System Innovations 4
W H I T E PA P E R

As discussed previously, spectral efficiency is the amount of data payload capacity you can get out of a
given amount of spectrum for a given reach requirement. For the same reach requirement, it is possible
to have better wavelength capacity but worse spectral efficiency, as shown on the left of Figure 3, or lower
capacity for individual wavelengths but higher spectral efficiency, as shown on the right of Figure 3.

The Changing Relationship Between Wavelength Capacity-Reach and Spectral


Efficiency
In the past, wavelength capacity and spectral efficiency moved in tandem: when you increased the
wavelength’s capacity you also increased the spectral efficiency. This was primarily because with fixed-grid
line systems and limited baud rates, increasing the capacity of the wavelength did not have any impact on
the number of wavelengths that could be supported on the fiber. For example, with a 50 GHz grid, going
from 2.5G wavelengths to 10G wavelengths increased the total capacity in 4,000 GHz (4 THz) from 200 Gb/s
(80 x 2.5 Gb/s) to 800 Gb/s (80 x 10 Gb/s).

Bits Per Symbol (Bits/S/Hz)

50 GHz 50 GHz 50 GHz 50 GHz 75 GHz 100 GHz

800G

200G 400G

100G

2.5G 10G

Spectrum

Figure 4: Spectrum and bits per symbol: 2.5G to 800G examples

2.5G 10G 100G

Symbol Rate (Example) 2.7 Gbaud 11.1 Gbaud 30 Gbaud

Grid 50 GHz 50 Ghz 50 GHz

Typical Occupied Spectrum (Example) 5 GHz 20 GHz 35 GHz

Unused Spectrum in 50 GHz 45 GHz 30 GHz 15 GHz

Spectrum of Ideal Wavelength 2.7 GHz 11.1 GHz 30 GHz

Wavelength Spectrum Waste 2.3 GHz 8.9 GHz 5 GHz

Total Wasted Spectrum 47.3 GHz 38.9 GHz 20 GHz

% Total Wasted Spectrum 94.6% 77.8% 40%

Table 2: Total wasted spectrum with 50 GHz grid (2.5G, 10G, 100G)

Maximizing Spectral Efficiency with Optical Engine and Line System Innovations 5
W H I T E PA P E R

This increase in spectral efficiency came not from increasing the bits per symbol of the wavelength but
from wasting less spectrum, as shown with the example in Table 2. For example, an unshaped 2.7 Gbaud
2.5G wavelength that occupies 5 GHz actually wastes 94.6% of the spectrum compared to an ideal 2.7 GHz
wavelength, while an unshaped 11.1 Gbaud 10G wavelength that occupies 20 GHz actually wastes 77.8%
relative to the ideal 11.1 GHz 10G wavelength.

Likewise, going from 10G wavelengths to 100G wavelengths increased the capacity in 4,000 GHz (4 THz)
from 800 Gb/s to 8,000 Gb/s, a tenfold increase. This increase came from both increasing the number of
raw bits per symbol from 1 to 4, and from reducing the amount of total wasted spectrum from 77.8% to 40%,
as shown in Table 2.

Wavelength Spectral Efficiency


From To
Capacity (and Fiber Capacity)
2.5 Gb/s (NRZ, 2.7 Gbaud) 10 Gb/s (NRZ, 11.1 Gbaud) 4x 4x

10 Gb/s (NRZ, 11.1 Gbaud) 100 Gb/s (PM-QPSK, 30 Gbaud) 10x 10x

100 Gb/s (PM-QPSK, 30 Gbaud) 200 Gb/s (PM-16QAM, 30 Gbaud) 2x 2x

200 Gb/s (PM-16QAM, 30 Gbaud) 400 Gb/s (PM-16QAM, 60 Gbaud) 2x No Gain

200 Gb/s (PM-16QAM, 30 Gbaud) 800 Gb/s (PM-64QAM, 84 Gbaud) 4x 1.5x

Table 3: Generational transitions: wavelength capacity and spectral efficiency gains

However, with higher baud rates and flexible-grid ROADMs, the relationship between wavelength capacity-
reach and spectral efficiency has changed. For example, as shown in Table 3, increasing the wavelength
data rate from 200 Gb/s with 30 Gbaud (~35 GHz) to 800 Gb/s with 84 Gbaud (~90 GHz) increases the
spectral efficiency by approximately 50% rather than fourfold, as with the data rate. This is because the
800 Gb/s wavelength requires a little under three times the spectrum and we can no longer count on big
reductions in spectral waste to boost spectral efficiency with more bits per symbol now the primary source
of spectral efficiency gains.

Wavelength Capacity-Reach and Spectral Efficiency Trade-offs


With higher baud rates, there is also some trade-off between wavelength capacity-reach and spectral
efficiency. Higher baud rates indirectly enable better wavelength capacity-reach as fewer bits per symbol
are required for the same data rate. The wider spectrum of the high-baud-rate wavelength also enables
higher transmit power and therefore improved noise tolerance without increasing the power spectral
density of the wavelength and therefore the nonlinear penalties. These two factors outweigh the increased
sensitivity to noise of the increased baud rate itself. But to the extent that the higher baud rate itself
reduces the reach, the spectral efficiency is reduced when compared with a lower baud rate and higher
modulation (i.e., more bits per symbol) that could deliver the same reach. This is illustrated by the example
shown in Figure 5, where two 600G wavelengths in 225 GHz can deliver the same reach as two 550 Gb/s
wavelengths in 200 GHz, with the former having better wavelength capacity-reach and the latter having
better spectral efficiency.

Maximizing Spectral Efficiency with Optical Engine and Line System Innovations 6
W H I T E PA P E R

Bits/s Per Hz (Bits Per Symbol)

High Baud Rate, Lower Modulation Lower Baud Rate, Higher Modulation
225 GHz (=5.33 bits/s/Hz) 200 GHz (=5.5 bits/s/Hz)

600 Gb/s 600 Gb/s 550 Gb/s 550 Gb/s

Spectrum
Figure 5: Wavelength capacity-reach vs. spectral efficiency (same reach): baud rate

Figure 6 provides another example of this trade-off. A larger spectral gap between wavelengths typically
enables higher transmit power for the same nonlinear penalties, and higher transmit power enables a
higher optical signal-to-noise ratio (OSNR) with greater tolerance for ASE noise from the amplifiers in the
wavelength’s path. In this example, the choice is between maximizing wavelength capacity-reach with two
widely spaced 800 Gb/s wavelengths or getting better spectral efficiency with three 700 Gb/s wavelengths
crammed closer together.

Bits/s Per Hz (Bits Per Symbol)

High TX Power, Larger Gap


Lower TX Power, Smaller Gap

800 Gb/s 800 Gb/s 700 Gb/s 700 Gb/s 700 Gb/s

Spectrum

Figure 6: Wavelength capacity-reach vs. spectral efficiency (same reach): channel spacing

Spectral Efficiency Factors


As described previously, three key factors determine spectral efficiency. This section describes these three
factors in more detail, together with optical engine and line system features that enable superior spectral
efficiency.

Factor 1: More Raw Bits Per Symbol


The first factor that enables superior spectral efficiency is getting more bits per symbol for a given set of
reach conditions. At shorter distances, this is down to higher-order modulation.

Maximizing Spectral Efficiency with Optical Engine and Line System Innovations 7
W H I T E PA P E R

PM-QPSK PM-8QAM PM-16QAM PM-32QAM PM-64QAM


2 bits 3 bits 4 bits 5 bits 6 bits

2 bits 3 bits 4 bits 5 bits 6 bits

4 bits/symbol 6 bits/symbol 8 bits/symbol 10 bits/symbol 12 bits/symbol

Figure 7: Raw bits per symbol with conventional coherent modulation

As shown in Figure 7, PM-64QAM delivers 12 raw bits per symbol, PM-16QAM delivers 8 raw bits per
symbol, PM-8QAM delivers 6 raw bits per symbol, and PM-QPSK delivers 4 raw bits per symbol. Advanced
optical engine features that enable more bits per symbol for a given set of reach conditions include long-
codeword probabilistic constellation shaping (LC-PCS), Nyquist subcarriers, and a high modem signal-to-
noise ratio (SNR). Dynamic bandwidth allocation (DBA) combines LC-PCS and Nyquist subcarriers to enable
more bits per symbol on the inner subcarriers, thus increasing the aggregate number of bits per symbol
for the wavelength. Efficient FEC can also increase the number of bits per symbol for a given set of reach
conditions, with soft-decision FEC (SD-FEC) gain sharing enabling two wavelengths to share FEC gain, thus
increasing the number of raw bits per symbol on the second, more challenged wavelength.

Mode A Mode B Mode C Mode D Mode E

Higher Performance

Lower Performance

Spectrum

Figure 8: Multiple modes are required to optimize for each part of the spectrum in a submarine fiber

High levels of optical engine programmability in terms of both the baud rate and modulation (bits per
symbol) can also enable superior spectral efficiency, especially on submarine fibers where performance in
different parts of the spectrum varies with tilt, ripple, dispersion, and nonlinearities, as shown in Figure 8.
For example, the ICE6-powered CHM6 transponder for the GX G42 compact modular platform currently
supports over 200 combinations of baud rate and modulation bits per symbol.

Maximizing Spectral Efficiency with Optical Engine and Line System Innovations 8
W H I T E PA P E R

And while in theory PCS alone could provide the required performance granularity, the need to align to
useful bandwidth increments (i.e., 50 Gb/s or 100 Gb/s) makes the combination of modulation and baud rate
programmability highly beneficial.

Another factor that can maximize spectral efficiency in submarine networks is a super-Gaussian PCS
distribution that results in less variation in the power levels of the symbols and therefore lower nonlinear
impairments. This is ideal for dispersion-uncompensated large-effective-area fibers that typically operate
at high power levels. For legacy dispersion-managed submarine cables, specialized multi-dimensional
modulations that vary the number of bits per symbol by polarization, time slot, and/or subcarrier can be key
to maximizing spectral efficiency.

In terrestrial networks, the optical line system can also play a role in enabling more bits per symbol for a
given set of reach conditions. Low noise, high gain amplification, low filter narrowing, and advanced optical
link control that better optimizes wavelength and amplifier power levels can enable more bits per symbol.
Factor 2: Higher Overhead Efficiency
A second factor that drives superior spectral efficiency is overhead efficiency. That is, for a given number of
raw bits per symbol transmitted, how many of those bits are used for the data payload versus the overhead.
Efficient FEC with a high net coding gain is one example of a feature required for high overhead efficiency.
A second example of how overhead efficiency can be improved is ICE6’s Ethernet framing mode, which
reduces the framing overhead for Ethernet client traffic compared to OTN framing modes that can support
both Ethernet and OTN client types. Probabilistic constellation shaping is also a form of overhead, with
PCS-64QAM actually transmitting 12 bits per symbol, with the distribution matcher’s bits-to-symbols mapping
acting as an overhead. One future strategy for overhead optimization is optimally balancing FEC and PCS
overhead and gain.

Factor 3: Minimized Spectral Waste


A third factor that drives superior spectral efficiency is how tightly the wavelengths can be packed together,
or to put it another way, how little spectrum is wasted between wavelengths and by the wavelengths
themselves, as shown previously in Figure 2. One way to reduce the wasted spectrum is to have one very
high-baud-rate wavelength rather than lots of low-baud-rate wavelengths occupying the same spectrum,
each of which requires a gap between it and its neighbors. Another way is super-channels, packing multiple
wavelengths together in a single block of spectrum. Another factor is the wavelength’s roll-off, the amount
of additional spectrum required to accommodate the slopes and modes at the sides of the wavelength, with
a squarer frequency-domain shape, as shown on the right of Figure 9, reducing the spectrum wasted by the
wavelength itself and enabling wavelengths to be packed closer together.

Higher Roll-off Tighter Roll-off

Spectral Width in GHz = Baud Rate +30% Spectral Width in GHz = Baud Rate +6%

Figure 9: Tighter roll-off enables closer wavelength packing

Maximizing Spectral Efficiency with Optical Engine and Line System Innovations 9
W H I T E PA P E R

A final optical engine capability that can contribute to less spectral waste is a shared wavelocker. This
enables the lasers of two or more wavelengths to drift, for example, due to changes in temperature, in
tandem, thus reducing the amount of guard band that would be required to support uncorrelated frequency
shifts. In addition to these optical engine features, minimized spectral waste also requires flexible-
grid ROADMs with the ability to control both the width and center frequency of each channel with high
granularity.

Spectral Efficiency Limitations


The (Linear) Shannon Limit
A 1948 paper published by Claude Shannon, the mathematician, electrical engineer, and information theorist
who then worked at Bell Labs, established what became known as Shannon’s law, otherwise known as
the Shannon-Hartley theorem. This law/theorem puts a limit on the amount of information that can be
communicated over a channel with a given bandwidth and amount of noise. This is described by the famous
equation C/B = Log2 (1+SNR), where C is the maximum achievable channel capacity, B is the bandwidth,
and SNR is the ratio of signal power to noise power. At higher SNR values, where we can ignore the 1 in the
equation, increasing spectral efficiency by adding 1 bit per symbol to the spectral efficiency requires us to
double (+3 dB) the required SNR, thus halving the reach. This provides an upper bound on the maximum
spectral efficiency that can be achieved. Today’s state-of-the-art high-performance embedded optical
engines, such as Infinera’s ICE6, are typically between 1 and 2 dB from the Shannon limit, meaning that any
future spectral efficiency gains are likely to be incremental.

The Nonlinear “Shannon” Limit


Optical conference proceedings sometimes include papers referring to the “nonlinear Shannon limit,” with
bold claims of having exceeded this limit. This is not to be confused with the (linear) Shannon limit described
above, which places a hard upper bound on maximum spectral efficiency, and including Shannon in the
limit’s name is not without controversy. This lower bound is based on nonlinear penalties and is arrived at
by simulations. New nonlinear mitigation techniques can then beat this lower nonlinear “limit,” but they can
never beat the (linear) Shannon limit or the true, currently unknown, upper limit that would be a function of
both linear and nonlinear penalties.

Better Spectral Efficiency vs. More Spectrum


In addition to acquiring and lighting new fiber, which can be costly and subject to long delays, the other
option for increasing fiber capacity is to increase the spectrum on the fiber. While in the future additional
bands may become commercially viable, most probably the S-band, today we are limited to the C- and
L-bands. Modern DWDM systems typically leverage the 4,800 GHz (4.8 THz) of the extended C-band, with a
few commercial systems even extending this to the 6,000 GHz (6 THz) of the “super C-band.” C+L systems
also use the L-band to provide a total of 9,600 GHz (9.6 THz).

For operators planning to deploy a new line system, expanding the spectrum to the L-band (i.e., C+L), or at
least deploying a line system such as Infinera’s FlexILS that can be hitlessly and cost-effectively upgraded
from C-only to C+L, can be a good option. However, upgrading an existing C-band-only optical line system
can be an expensive and disruptive undertaking, leaving improved transceiver spectral efficiency a more
incrementally cost-effective and less disruptive option. That said, this is not an either/or. Operators that
upgrade to a C+L line system will typically also want to make the most of their spectrum with spectrally
efficient optical engine technology.

Maximizing Spectral Efficiency with Optical Engine and Line System Innovations 10
W H I T E PA P E R

Summary
Spectral efficiency is a top priority for submarine and long-haul networks and for fiber-constrained, high-
bandwidth metros. Three factors drive spectral efficiency for a given reach requirement: the raw bits per
symbol, overhead efficiency, and spectral waste. Maximizing spectral efficiency requires both optical
engine and line system innovations that can optimize these three factors. However, with today’s state-
of-the art optical engines getting close to the (linear) Shannon limit for spectral efficiency, other methods
for increasing capacity such as lighting new spectrum bands on existing fibers and SDM, including more
fibers per submarine cable and potentially even new multi-core or multi-mode fibers, are likely to become
increasingly important.

© 2022 Infinera Corporation. All Rights Reserved. Infinera and logos that contain Infinera are trademarks or registered trademarks of Infinera Corporation in the
United States and other countries. All other trademarks are the property of their respective owners. Statements herein may contain projections regarding future
products, features, or technology and resulting commercial or technical benefits, which are subject to risk and may or may not occur. This publication is subject to
change without notice and does not constitute legal obligation to deliver any material, code, or functionality and is not intended to modify or supplement any product
specifications or warranties. 0304-WP-RevA-0322

You might also like