Spectrally efficient optical

orthogonal frequency division

royalsocietypublishing.org/journal/rsta
multiplexing
Arthur James Lowery
Review Department of Electrical and Computer Systems Engineering,
Monash University, Clayton, Victoria 3800, Australia
Cite this article: Lowery AJ. 2020 Spectrally
efficient optical orthogonal frequency division AJL, 0000-0001-7237-0121
multiplexing. Phil. Trans. R. Soc. A 378:
This paper charts the development of spectrally
20190180.
efficient forms of optical orthogonal frequency
http://dx.doi.org/10.1098/rsta.2019.0180 division multiplexing (OFDM) that are suited for
intensity-modulated direct detection systems, such
Accepted: 21 October 2019 as wireless optical communications. The journey
begins with systems using a DC-bias to ensure that
no parts of the signal that modulates the optical
One contribution of 17 to a theme issue source are negative in value, as negative optical
‘Optical wireless communication’. intensity is unphysical. As the DC-part of the optical
signal carries no information, it is wasteful in energy;
Subject Areas: thus asymmetrically clipped optical OFDM was
electrical engineering developed, removing any negative-going peaks
below the mean. Unfortunately, the clipping causes
Keywords: second-order distortion and intermodulation, so
optical communications, spectral efficiency, some subcarriers appear to be unusable, halving
spectral efficiency; this is similar for unipolar and
orthogonal frequency division multiplexing,
flipped optical OFDM. Thus, a considerable effort
intensity modulation, direct detection has been made to regain spectral efficiency, using
layered techniques where the clipping distortion is
Author for correspondence: mostly cancelled at the receiver, from a knowledge
Arthur James Lowery of one unpolluted layer, enabling one or more
extra ‘layers/paths/depths’ to be received on the
e-mail: [email protected]
previously unusable subcarriers. Importantly, for
a given optical power and high-order modulation,
layered methods offer the best spectral efficiencies
and need the lowest signal-to-noise ratios, especially
if diversity combining is used. Thus, they could
be important for high-bandwidth optical fibre
systems. Efficient methods of generating all layers
simultaneously, using fast Fourier transforms with
their partial calculations extracted, are discussed, as
are experimental demonstrations in both wireless
and short-haul communications links. A musical

2020 The Authors. Published by the Royal Society under the terms of the
Creative Commons Attribution License http://creativecommons.org/licenses/
by/4.0/, which permits unrestricted use, provided the original author and
source are credited.
 analogy is also provided, which may point to how orchestral and rock music is deciphered in 2
the brain.
This article is part of the theme issue ‘Optical wireless communication’.

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
1. Introduction
Communications systems have many metrics to quantify their performance, for example their
data rate (bit s−1 ) and transmission distance are key functional specifications. Added to these are
power consumption, reliability/availability, security and the ability to work with other systems
without interference. Mobile systems, including some optical wireless, also need to consider
coverage; that is, over what area will the link perform?
This paper is, however, about spectral efficiency, which has become important in optical
fibre communications as it indicates the ability to pack more wavelength-multiplexed channels
into a single optical fibre [1]. However, it is also a key parameter in wireless (fibre-less) optical
communications, as illustrated in figure 1, and single-wavelength systems that are used for short-
range interconnects. This is because each signalling rate of each wavelength of most optical
systems is ultimately limited by the electrical bandwidths of their optoelectronic components,
such as light-emitting diodes, lasers, optical modulators and photodiodes and of the associated
electronics, including analogue-to-digital and digital-to-analogue converters [2,3]. Thus, this
paper will hereon in refer to electrical spectral efficiency, defined as the bit rate per unit electrical
bandwidth (bit s−1 Hz−1 ). This efficiency can be increased by coding many bits per symbol, such
as by using higher order quadrature amplitude modulation (QAM) [4]; for example, 16-QAM will
carry log2 (16) = 4 bits symbol−1 . Using higher order QAM, say up to 1024 QAM, increases spectral
efficiency: it also increases the required signal-to-noise ratio (SNR) at the receiver substantially.
Thus, there is a trade-off between transmission distance and spectral efficiency for a given
transmitted power.
In optical wireless systems [5], and many fibre systems, there is also a limit on total optical
power. In the early days, this was usually due to the falling reliability of the laser or LED at
higher powers. More recently, this is due to safety standards, which are particularly stringent at
visible wavelengths as visible light is naturally focused onto the retina creating extremely high
power densities at the eyes’ photoreceptors. Thus, this paper will concentrate on the electrical
spectral efficiencies available for a given optical power, for a number of signalling schemes.
The paper is organized as follows. Firstly, optical orthogonal frequency division multiplexing
(OFDM) schemes will be introduced that reduce the unmodulated optical power, but lower the
spectral efficiency. Secondly, two- then three-layer schemes for regaining the spectral efficiency
will be discussed. Thirdly, the cost of increasing the spectral efficiency, in terms of the required
electrical SNR at the receiver, will be compared for various schemes all operating with the same
optical power; this section shows the advantage of layered schemes for spectral efficiencies
above 3 bit s−1 Hz−1 . Experimental demonstrations, both in wireless and fibre systems, will be
compared. The layered schemes require significantly more signal processing than conventional
transceivers; some methods of reducing this will be presented. Finally, other improvements will
be discussed before a conclusion is presented.

2. Optical orthogonal frequency division multiplexing

OFDM uses multiple subcarriers to transmit a given data channel [6]. The frequencies and
signalling rates of these subcarriers are arranged so that a receiver can separate them without
mutual interference. The simplest method is to arrange the frequency spacing to equal the inverse
of the duration of a raw OFDM symbol, which makes all subcarriers periodic within the symbol;
thus, they can be perfectly separated using a Fourier transform (FT). By ‘raw’, the OFDM symbol
 bias- bias wireless or
3
tee free-space photo-
OFDM link diode OFDM

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
transmitter receiver
ampl.
DAC ampl. ADC

data in LED or data out

–ve
laser optional lenses bias

Figure 1. Block diagram of a typical wireless (or free-space) optical link. The use of lenses increases the received power
substantially but reduces the area of transmitter coverage, or imposes directionality on the receiver. (Online version in colour.)

does not include cyclic extensions to the OFDM symbol, which are used in dispersive channels,
but are discarded before the FT. A detailed history of the early ‘radio’ implementations of OFDM
can be found in [7].
Early patents disclosed that OFDM could work over any frequency range of subcarriers,
even optical, though all-optical systems were not demonstrated until 2002 with the FT being
implemented by an optical interferometer [8]. More conventionally, an intensity-modulated direct
detection (IM-DD) system was shown to be able to carry OFDM subcarriers in 1996, with an
increased resistance to impulsive noise [9]. Similarly, hybrid-fibre coax systems [10] distribute
CATV signals, including data on subcarriers (data over cable service interface specification—
DOCSIS [11]). The key to these systems is to add a high DC-bias to the electrical signal before
it modulates the laser, so that the bipolar electrical signal modulates the light source above its
lowest light level; that is, there is always positive light intensity. This is equivalent to limiting
the optical modulation depth [10]. The bias has to be strong because OFDM signals have large
positive and negative peaks in their waveform, due to the phases of the subcarriers occasionally
aligning (depending on the data being sent).
The problem with adding a high bias is that it substantially increases the mean optical intensity,
especially if infrequent ‘clipping’ of the negative peaks is to be avoided. Any clipping adds to the
noise of the demultiplexed subcarriers and also reduces their amplitudes. This degradation is
especially troublesome for high-order QAM, where high-quality signals are required to enable
the separation (e.g. by thresholding, also known as slicing) of the received signal into sets of data
bits without error. A high mean optical intensity may cause reliability issues with the laser, or
safety issues, for example.
In 1996, Carruthers & Kahn [5] suggested methods to reduce the bias overhead for multi-
carrier modulation by clipping each subcarrier separately, then You & Kahn [12] proposed
block coding or variable bias in 2001; these required 3–7 dB more power than on–off keying
(OOK). In 2005, Lowery & Armstrong [13] and Armstrong & Lowery [14] provided two methods
based on clipping the sum of the subcarriers, rather than individual subcarriers; simulations
showed an advantage over OOK; both methods used clipping at exactly the mean level to
remove excursions below the mean level, and both also sacrificed some of the subcarriers,
to accommodate the substantial clipping-induced distortion, which was mostly second-order
harmonic and intermodulation distortion. Simulations identified that clipping exactly at the mean
level was optimal, in terms of the received signal quality for a given optical power. In [13], the
signal distortion caused by clipping mostly falls within a spectral gap with a bandwidth equal in
the signal bandwidth. In [14], only odd subcarriers are generated, and clipping of these created
distortion only at the frequencies of the (deleted) even subcarriers, as illustrated in figure 2.
Thus, the distortion could be completely rejected by the FT at the receiver. Both methods are
known as ‘asymmetrically clipped optical (ACO)-OFDM’, though the latter is subject of most
research for wireless systems. The ‘frequency gap’ idea is more suited to optical fibre systems
using modulation of the optical field rather than intensity (to enable electronic compensation of
 electrical DCO ACO DDO L-ACO 4
spectra
(1 symbol)

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
arrows are subcarriers circles are distortion frequency
spectral
efficiency
(4-QAM) 1.75 b s–1 Hz–1
2 b s–1 Hz–1 2 b s–1 Hz–1
1 b s–1 Hz–1 (3 layers)

DSNRreq
(10–3BER) reference
1.6 dB less 2.9 dB less
32-QAM
(5 bit s–1 Hz–1) diversity combining/ (interpolated)
noise cancellation
diversity combining/
noise cancellation
diversity combining/
mixtures of m-QAMs noise cancellation
needs to be 1024-QAM
to support 5 bit s–1 Hz–1

Figure 2. Evolution of optical OFDM systems from DC-biased (DCO), through asymmetrically clipped (ACO), then symmetrically
clipped and DC offset (ADO) to layered/enhanced (LACO). The top row is the electrical spectra for one-symbol; the middle row
is the relative spectral efficiency for 4-QAM constellations, and the bottom row shows the required electrical SNR relative to
DCO-OFDM for 5 bit s−1 Hz−1 spectral efficiencies. (Online version in colour.)

chromatic dispersion in the fibre dispersion [15]), where it is used to reject distortion caused by
direct detection of these field signals that leads to unwanted frequency-difference products [16].
Both ACO-OFDM solutions rely on ‘throwing away’ half of the subcarriers, which means that
the electrical bandwidths of the transmitter and receiver need to be doubled for the same data.
Similar methods, such as those relying on sending the negative-valued portions of the signal
in a second frame (Flip-OFDM [17–19], unipolar OFDM [20]) also suffer from halved spectral
efficiency. This has led to intense research on how to reclaim this bandwidth, for both the gapped
and odd subcarrier methods, and also for Flip and Unipolar OFDM.
The gapped method has proven more useful for fibre systems using optical field rather
than intensity modulation because when used in combination with single-sideband modulation
with carrier suppression, OFDM’s single-tap equalization can be used to compensate chromatic
dispersion, even with a single-photodiode receiver [15]. The photodiode’s square-law detection
causes an unwanted signal × signal beat product, but this fortuitously falls within the gap. Just
as with intensity-modulated systems, regaining spectral efficiency has been a major research
effort in field-modulated systems [21]; Kramers–Kronig receivers provided an interesting solution
based on the relationship between phase and amplitude of minimum-phase waveforms [22].
Interestingly, clipping of the received waveforms appears to be able to improve these systems [23].

3. Regaining spectral efficiency: two-layer schemes

The key to spectral reclamation techniques is that clipping distortion is deterministic, and so it
should be possible to calculate it from the received signal, and then subtract it to ‘clean-up’ the
unused portions of the spectrum, so that new subcarriers can be revealed. This also works for
Flip/U-OFDM signalling, but is due to inherent symmetries in the time domain signals.

(a) Asymmetrically clipped and DC-biased optical orthogonal frequency division

multiplexing
In 2011, Dissanayake et al. [24] proposed cancellation of clipping distortion and added DC-biased
optical OFDM (DCO-OFDM) subcarriers to all of the newly available ‘even’ subcarrier slots. As
illustrated in figure 3, the DCO-OFDM subcarriers were used because they create no distortion
themselves (they are not clipped before adding to the already clipped ACO-OFDM subcarriers).
 ACO-OFDM Rx. processing Rx.spectrum: both layers
5
S channel FFT
layer 1 data

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

select odds

...............................................................
slice
frequency – clip IFFT frequency
DCO-OFDM DCO-OFDM
circles
are clipping FFT
layer 2 data
distortion
select evens slice
frequency frequency

Figure 3. Processing of ADO-OFDM to clean the DCO-OFDM of ACO-OFDM interference. The select odds block separates out the
ACO-OFDM signal in the frequency domain. This is then clipped to recreate the distortion that fell on the evens at the transmitter.
This distortion is subtracted from the received signal to reveal the DCO-OFDM signal. (Online version in colour.)

Later work showed an advantage in electrical SNR for a given optical power over conventional
techniques for SEs above 3 bit s−1 Hz−1 [25]. At the receiver, the ACO-OFDM subcarriers were
isolated from the spectrum using an FT, then re-clipped to recreate the transmitted ACO-OFDM
signal, which was subtracted from the received signal to reveal the DCO-OFDM subcarriers. No
slicing nor thresholding was used during the recreation process; thus, noise from the ACO-OFDM
spectrum would degrade the DCO-OFDM channels.

(b) Hybrid asymmetrically clipped optical–orthogonal frequency division multiplexing

This 2014 method by Ranjha & Kavehrad uses QAM-modulated ACO-OFDM on the odd
subcarriers and clipped pulse-amplitude modulation discrete multi-tone (PAM-DMT) on the even
subcarriers [26] to achieve various spectral efficiencies depending on the mix of constellation
sizes and apportionment of power. As with ADO-OFDM, but with thresholding, the ACO-
OFDM waveform is recreated at the receiver and subtracted to reveal the PAM-DMT subcarriers.
Interestingly, if PAM-DMT subcarriers are clipped, the resultant distortion only falls on their
own (even) frequencies, so does not pollute the (odd) ACO-OFDM subcarriers; the distortion
falls in quadrature with the desired PAM-DMT signals, so can be discarded. A fundamental
disadvantage of PAM is that it requires higher SNRs than QAM for a given bits symbol−1 , because
it only uses one signal dimension and each additional bit requires a doubling of the number of
levels and approximately a doubling of power; this means that this system is optimal when the
PAM subcarriers transport fewer data than the QAM subcarriers.

(c) Asymmetrically and symmetrically clipping optical orthogonal frequency

division multiplexing
This 2015 scheme by Wu & Bar-Ness uses ACO-OFDM on the odd subcarriers and a
Flip/U-OFDM-type signal on the even subcarriers at only one-half of the baud rate of the
ACO-OFDM signal [27]. It follows from a 2012 paper improving the error rate performance of
ACO-OFDM [28]. Using mixes of constellation sizes on the two layers enables various spectral
efficiencies and generally requires lower optical powers than ADO-OFDM, because no bias is
required on either layer. Thresholding/slicing was not used when cancelling the ACO-ODFM
clipping distortion.

4. Regaining spectral efficiency: multi-layer schemes

In some techniques, more than two layers share the data load, and multiple stages of cancellation
are used at the receiver, to ‘peel-away’ the layers, like an onion. In these schemes, the first layer
must be unpolluted by the deeper layers, as it has to be decoded and reconstructed so that its
distortion can be removed from deeper layers. The sharing of data is usually performed with 50%
Rx signal T0 T1 T2 T3 T4 T5 T6 T7 6
frames
Layer 1 P11 N11 P12 N12 P13 N13 P14 N14

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
Layer 2 P21 P21 N21 N21 P22 P22 N22 N22
Layer 3 P31 P31 P31 P31 N31 N31 N31 N31
subtract T(n + 1) Rx signal T(n), T(n + 1)
from T(n)

P11
FFT slice recons. – add T(n), T(n + 1) pairs
0
N11 repeat for n = 0,2,4,6
data out

Layer 2 P21 N21 P22 N22

Layer 3 P31 P31 N31 N31

subtract T(n+2) and T(n+3)
from T(n) + T(n+1)
process L2,3 as above….

Figure 4. Processing of EU-OFDM frames T to extract layers. P indicates a sub-frame carrying the positive parts of the signal;
N indicates a frame carrying the inverted negative parts (so they are now positive). The fact that the frames of deeper layers
are repeated allows the deeper layers to be discarded by time domain subtraction of adjacent frames, and also a bipolar signal;
e.g. subtracting frame T1 from frame T0 provides Layer 1’s bipolar signal P11-N11. (Online version in colour.)

of the data of the equivalent DCO-OFDM on Layer 1, 25% on Layer 2, 12.5% on Layer 3, etc.,
which gives diminishing returns in spectral efficiency, at a cost of more processing every time a
layer is added. The advantage of using multiple layers is that all layers can use power-efficient
modulation, such as QAM with clipping; this leads to the lowest SNR requirements of all schemes.

(a) Enhanced unipolar orthogonal frequency division multiplexing

Enhanced unipolar orthogonal frequency division multiplexing (EU-OFDM) [29] uses the
superposition of time domain signals at various layers (called ‘depths’), where each U-OFDM
symbol (positive frame, P, followed by a negative frame, N, like Flip-OFDM) in the deeper layers
is transmitted more than once as a simple repeat. This is illustrated in figure 4. Obviously, the
data rate of each deeper layer is halved due to the repetition, and the effect of the repetition on its
subcarriers is to narrow their spectra. Experimental results showed a 2-dB electrical drive power
advantage for 16-QAM over DCO-OFDM for a BER = 10−3 at 80 Mbit s−1 [30]. The repetition of
the deeper layers allows them to be cancelled when a higher layer is decoded. Consider the initial
decoding of Layer 1, shown in figure 4. By subtracting pairs of frames, say T1 from T0, a bipolar
frame for Layer 1 will be revealed as it equals P11-N11. The adjacent frames of deeper layers at
this time cancel, e.g. P21–P21 = 0 and P31–P31 = 0. Thus, Layer 1 can be reconstructed without
interference from deeper layers. The reconstructed signal can be clipped, and then subtracted
from the complete set T0–T7, to reveal Layers 2 and 3. Next frames T3 and T4 can be subtracted
from T0 and T1, to obtain a bipolar signal for Layer 2, while also cancelling Layer 3.
EU-OFDM places the subcarriers of all of the layers at the same frequencies (albeit the
higher layers have narrower sinc-shaped spectra), which unfortunately means that any errors
when slicing the constellations of the higher layers results in strong error vectors at the deeper
layers, causing a cascade of bit errors. The optical and electrical spectra are not flat, because
some frequencies support multiple layers. This may reduce the performance of the system. A
further issue with Flip/U-OFDM is that the adjacent positive and negative signal blocks must
not interfere with one another in a bandwidth-limited channel. This either requires time domain
equalization before the processing of the blocks or additional cyclic prefixes between all frames.
 clipped individual layers Rx. processing Rx.spectrum: all layers
7
S channel FFT
Layer 1 data

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

select L1

...............................................................
slice
subcarriers
frequency – clip IFFT QAM frequency
Layers 2 and 3
FFT
Layer 2 data
select L2
slice
subcarriers
frequency frequency
– clip IFFT QAM Layer 3
circles
FFT
are clipping Layer 3 data
distortion select L3 slice
subcarriers
frequency frequency

Figure 5. Processing of LACO-OFDM to extract layers. Layer 1’s subcarriers are selected after the top-right FFT and after slicing
to extract data bits, and the clipped waveform of Layer 1 is reconstructed by a QAM modulator, inverse FFT (IFFT) and clipper.
This waveform is subtracted from the received signal to reveal Layers 2 and 3. Note that the reconstruction includes slicing (hard
decisions), to prevent noise from Layer 1 being passed to Layer 2’s processing (though hard errors would be propagated). (Online
version in colour.)

(b) Generalized EnhaNcEd unipolaR orthogonal frequency division multiplexing

Islim et al. have proposed that the spectral efficiency of layered schemes can be made to exactly
match that of DCO-OFDM simply by using different constellation sizes and transmit powers on
different layers [31]. The electrical SNR requirement was about 4 dB better than DCO-OFDM for
a BER of 10−3 .

(c) Layered/enhanced ACO-OFDM (LACO-OFDM or L/E-ACO-OFDM)

As shown in figure 5, LACO-OFDM successively fills the frequency slots with layer upon layer of
subcarriers. Unlike EU-OFDM, no slot carries more than one subcarrier. Layer 1 is conventional
ACO-OFDM, occupying only the odd frequency subcarriers (2n + 1); n integer including zero.
Layer 2 fills the even frequencies that are 2(2n + 1), for example 2, 6 and 10. Layer m fills
2m−1 (2n + 1). At the receiver, Layer 1 can be decoded as in conventional ACO-OFDM, because
it has no interference from higher layers. A facsimile of Layer 1’s transmitted signal can be
constructed and subtracted from the received signal, to reveal Layer 2 free of Layer 1’s distortion.
Importantly no clipping distortion from Layer (n + 1) falls on Layer n, so Layer n can be decoded
without error if the reconstruction of Layer (n − 1) is error free.
Layered/enhanced ACO-OFDM has been invented independently by several groups, via
various routes. In 2010, Chen et al. [32] published a successive decoding method at a conference,
which formed a small part of a later journal paper [33] and seems to have been missed by many
groups. In March 2015, a PCT patent application was published on EU-OFDM (Inventors: Tsonev
and Haas) but noting that ACO-OFDM could be used as a basis for the frames [34]. In May 2015,
Wang et al. [35] presented numerical results from the layered ACO-OFDM, and in December
2015, Islim et al. [36] published the PCT framed system using ACO-OFDM; the repetition of
frames in the higher layers causes the subcarriers of all layers to fall at the same frequencies,
albeit with narrower spectra. In October 2015, Lowery published a description based on musical
chords (each of odd harmonics of a fundamental) at octave intervals [37], each describing a layer,
and showed that the method offered the best sensitivities over other methods for higher order
QAM modulations. In contrast with Islim et al., each frequency slot was used only once, if at all.
In 2017, Mohammed et al. [38] added diversity combining to layered ACO-OFDM and reached
similar conclusions in performance comparisons. Diversity combining uses the information in
the clipping distortion as part of the recovery process and has been shown to improve errors rates
in both clipped [39,40] and flipped [17] OFDM systems.
 Errors in Layer n may not necessarily propagate to Layer (n + 1), because the error vectors are
8
small, as they are residues of the un-cancelled clipping distortion and their power is distributed
across many subcarriers. Inter-layer error correction reduces error propagation still further, as

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
proposed by Hanzo’s group [41,42] and also disclosed by Song’s PhD [43].

(d) Spectral and energy efficient orthogonal frequency division multiplexing

This 2014 scheme requires only one FT at the receiver, where the layers are called Paths [44]. The
top path is ACO-OFDM, and the lower layers are combinations of ACO-OFDM with repeated
frames that are flipped versions of the first frames. Unlike EU-OFDM, all layers have the same
OFDM symbol duration, which is the reason that only a single FT is required––an obvious
advantage. Slicing upon decoding was not used initially, but later versions use slicing and ACO-
OFDM on all layers [45]; thus, these require multiple FTs during the decoding process, as with
other layered ACO-OFDM methods. Using ACO-OFDM ensures that all of the subcarriers are
periodic within a common symbol duration, so advantageously share one common cyclic prefix
per symbol, thus lowering the CP overhead compared with EU-OFDM. At the receiver, the
channels are successively decoded, starting with the original odd subcarrier channel. Thus, for
reasonable SNRs, the penalty due to imperfect cancellation of a channel is very small, as there is
little noise on the cancellation waveform.

(e) Spectrally enhanced amplitude modulation discrete multi-tone and augmented

spectral efficiency discrete multi-tone
In 2015, Islim et al. [46] extended generalized EnhaNcEd unipolaR orthogonal frequency division
multiplexing (GREENER-OFDM) to enable multiple streams of PAM-DMT to be transmitted.
This used a frame structure similar to Flip-OFDM, with cyclic prefixes between each frame,
and considered optimum constellation sizes for each stream. A simplified frequency domain
generation method, similar to used in LACO, was then published which uses a FT per ‘depth’
[47]. The idea in both papers is that the clipping distortion of a PAM subcarrier on the quadrature
(sine) component of a complex signal will fall on the in-phase (cosine) component only. If this is
used for Depth/Layer 1, then the receiver is able to detect its quadrature component, reconstruct
the in-phase clipping distortion and then subtract it from the received signal. Thus, the in-phase
components can now also be used for signalling, providing that their distortion products do not
fall on the quadrature component. Fortunately, all clipping distortion of in-phase signals falls
upon them in-phase, so in the second paper, a similar arrangement to LACO-OFDM can be used
to ‘fill-up’ the in-phase subcarriers, starting from the odd subcarriers for Depth 2.

5. Signal-to-noise ratio cost of increasing spectral efficiency

The most common way of gaining spectral efficiency in communications systems is to use higher-
order quadrature amplitude modulation formats, denoted m-QAM, where m is the number of
distinct symbols in a two-dimensional constellation. The number of bits required to specify
the position of a symbol is log2 (m); for example, 256-QAM will carry eight bits of data for
each transmitted constellation pattern. If m is increased, then the separation of the constellation
points for a given transmitted power will decrease, making the impact of noise more likely to
misplace a constellation point and so cause a symbol error, which usually produces a bit error
if the constellation is Gray coded. Alternatively put, for a given noise level, higher-order QAM
constellations require more power.
Figure 6 compares the ‘cost’ of increasing the spectral efficiency of various OFDM schemes,
each for a number of m-QAMs [48]. The cost is defined as the increase in electrical SNR required
to support an increase over the reference system, which is 4-QAM ACO-OFDM. In every case,
the received mean optical power is set to be the same. Also included is PAM, where only one
 30 1024 9

SNR cost, dB (ref. 4-QAM ACO-OFDM)

O
DC

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
256

O
25

AC
PA
m-
4096 1024
16-PAM
20
64 CO
1024
256 LA
15 8-PAM 512
256 64
10 4-PAM
64
16
4 ADO-OFDM
5
4 16 SEE-OFDM
OOK
4 8
0
0 2 4 6 8 10

required spectral efficiency (bit s–1 Hz–1)

Figure 6. Comparisons of the costs (in electrical SNR) of increasing the spectral efficiency using various optical OFDM schemes,
with various constellation sizes. Note that results for noise cancellation or diversity combining are not shown, and these can
give around 2 dB further improvements for ACO and LACO. (Adapted with permission from [48]). (Online version in colour.)

Cartesian dimension is used; note that the actual bandwidth of PAM depends on the pulse
shaping used, and in the Nyquist limit (using sinc-shaped pulses), PAM could be twice as
spectrally efficient as shown; detailed comparisons for a range of modulator bandwidths are given
in the papers of Sharif et al. [2] and Perin et al. [3]. In these PAM simulations, the transmitted pulses
were unshaped, and the electrical receiver noise was band-limited by a fourth-order Bessel filter
with a bandwidth of 70% of the data rate. For IM-DD systems dominated by the thermal noise of
the receiver (typically systems without optical amplification), the electrical SNR will improve 2 dB
for each 1 dB increase in received optical power, due to the photodiode’s square-law detection
(IM/DD), where photocurrent is proportional to optical power. Thus, the ‘cost’ of increased
spectral efficiency, in terms of optical power (dB), is half that shown by figure 6. Although ACO-
OFDM provides the lowest SNR cost at low spectral efficiencies, its SNR cost rapidly increases
with spectral efficiency. This is because ACO-OFDM requires a constellation size of m-squared
compared with schemes that do not have a loss of half the subcarriers, e.g. 1024 constellation
points rather than 32 for schemes with full spectral efficiency. This fact makes layered schemes
very attractive for high spectral efficiencies because the extra data are carried by additional layers
rather than by increasing m; however, layered schemes have their own cost per layer, because
every layer contributes to the mean optical power, so for a constant optical power, each layer
must be allocated less power as the number of layers is increased.
For ease of comparisons, in figure 6, none of the schemes in the figure have diversity combining
nor noise cancellation as applied to layered systems by Wang et al. [49] and later by Mohammed
et al. [38], then Wang et al. [50] using soft successive interference cancellation. By taking QPSK
ACO-OFDM as the reference for SNR cost, Mohammed et al.’s fig. 5 can be used by subtracting
approximately 5 dB from their optical energy per bit, Eb(opt) /N0 , to obtain the electrical SNR cost
of figure 6. Note that their optical energy does not scale as the square-root of electrical SNR as
expected for an IM/DD system. This is because the optical power is derived from the expectation
value of the clipped OFDM waveform, which equals the standard deviation, σ , divided by
√ √
(2π ); this is then normalized for unity optical power, so σ = (2π ) [51]. The electrical power,
including signal and clipping components, is σ 2 /2, which thus equals π at unity optical power.
As the ratio of energies equals the ratio of powers, then Eb(opt) /N0 = (1/π)Eb(elec) /N0 [51]. If
only useful signals are included in the electrical power, then Eb(opt) /N0 = (2/π )Eb(elec) /N0 , which
10
suggests a 4.8-dB offset at a BER of 10−3 . Now, using Mohammed’s results including diversity
combining, we can expect 1.5–2.7-dB reduction in required electrical SNR for single-layer ACO-

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
OFDM (improving with constellation size); for the three-layer ACO-OFDM, the reduction is
between 1.25 and 1.5 dB. This improvement due to diversity combining makes layered schemes
even more attractive for high spectral efficiencies. Note that their results for DCO-OFDM use a
fixed bias of 10 or 13 dB, whereas figure 6 uses optimal biasing for each point, which is far lower
for small constellation sizes; for 256-QAM, the optimal bias is 10 dB, so the results converge.
It should be noted that different schemes produce different electrical and optical spectra. This
is particularly so for schemes using Flip-OFDM as a basis of each layer. This is because with Flip-
OFDM, the spectrum of each subcarrier is spread away from the allocated frequency, where there
is a null. This is because the flipping and repetition of the waveform to create the positive then
negative frames are functionally identical to repetition, then phase modulation (0, π ) followed by
clipping, so causes a null at the allocated frequency as would binary phase-shift keying. Also, the
higher layers only use evens and double-even subcarriers, coinciding with the lower layers. Thus,
the overall signal spectrum may be far from flat.
Sun et al. [52] have also compared the performance of various schemes, but from the point
of view of peak-to-average power ratio (PAPR) and a constraint of 1% of the signals’ duration
being out of limit. Although the conclusions were broadly similar, DCO’s performance was better
than in the above results, though still poorer than three-layer or more schemes. A later paper
on the analysis of LACO-OFDM used tone injection to give a further 5-dB reduction in PAPR
[53]. Zhou & Zhang [54] have also compared schemes based on the information rate, which
assumes unlimited coding to achieve zero error rates, and so is a more fundamental limit. Multi-
layer schemes were shown to approach the performance limit of intensity-modulated channels to
within 0.07 bits for power-limited optical channels.

6. Experimental demonstrations
Since its inception, there have been several experimental demonstrations of single-layer ACO-
OFDM for short-haul optical communications and wireless systems, where the low-cost of
intensity modulation (a directly modulated laser) combined with a single-photodiode receiver is
desirable. Azhar and O’Brien compared several systems, including a DC-biased ACO-OFDM that
reintroduced the bias, but kept the odd subcarrier allocation [55]. This helped mitigate baseline
wander (effectively the effect of having a high-pass filter) which caused ACO-OFDM to have a
penalty despite having a better SNR for a given optical power. Wang et al. [56] showed that ACO-
OFDM had lower BERs than DCO-OFDM in three-colour links of 10–60 cm for the same data
rate. Tahar et al. [57] showed that diversity combining provides a 2-dB improvement in 4-QAM
ACO-OFDM receiver sensitivity derived from the measured noise variance.
After Tsonev et al.’s [30] initial work, there are few experimental demonstrations of layered
techniques in wireless systems, though more in short-haul fibre systems where low cost is also
desirable. Most recently, Islim and Haas also demonstrated that DMT required 1–2 dB less optical
power than DCO-OFDM in 60-cm wireless systems using blue LED and infrared laser; a data rate
was not specified though the channel bandwidth was less than 48 MHz [58]. Chen et al. [59] have
transmitted ADO-OFDM over a 60-cm link using a laser with external modulator and Erbium-
doped fibre amplifier using 4-QAM and 16-QAM over a signal bandwidth of nearly 5 GHz. The
laser/modulator/amplifier combination is very bulky and costly compared with an LED. Zhang
et al. [60] have studied error propagation in a seven-layer 7-Gbit s−1 laser beam across an optical
bench, demonstrating a spectral efficiency of 7 bit s−1 Hz−1 , using bit and power loading that
minimized the effect of error propagation.
Very high data rate wireless systems that use slower electrical components can be built
using wavelength division multiplexing principles, as are ubiquitous in metro- and long-haul
communications systems. These use many electrical transmitters in parallel, each transmitting on
an optical wavelength. For example, six lasers, supporting 224 Gbit s−1 using a 3-m steered beam,
have been demonstrated [61]; this system used a coherent receiver to provide much improved
11
sensitivity over direct detection as the mixing of a local light oscillator and the incoming signal
upon photodetection provides gain [62]. More exotically, different channels can be encoded on

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
different modes of a light beam, such as using orbital angular momentum modes [63]. These
systems do not fit the mission of this paper, in that we are examining electrical spectral efficiency;
however, they should not be forgotten as a way of multiplying the capacity of optical wireless
systems, albeit at an exorbitant hardware cost.
Layered techniques have been demonstrated over short-haul (less than 20 km) fibre links,
which have a more complex optical channel than wireless links, due to the chromatic dispersion
of the fibre. In 2016, Song et al. [64] demonstrated a 4.375 Gbit s−1 three-layer QPSK system over
19.8 km of fibre using a directly modulated laser and Volterra equalization; a 2-dB improvement
in signal quality over DCO-OFDM was measured at the same optical power. Adopting LACO-
OFDM does enable the operating range and optimal bias of the laser to be optimized more
effectively. Later work using 16-QAM and 64-QAM at 3.5 Gbaud (14/28 Gbit s−1 ) showed a
0.7-dB advantage for 16 QAM, but no advantage for 64-QAM due to error propagation between
the layers, exacerbated by the combination of laser chirp and fibre chromatic dispersion [65];
pairwise-coding of good and bad subcarriers within each layer gave a 3.6-dB improvement in
signal quality at 20 Gbit s−1 mainly due to a reduction in error propagation [66]. Wang et al. [67]
also demonstrated augmented spectral efficiency discrete multi-tone (ASE-DMT) over fibre, using
a single fast Fourier transform (FFT) at the transmitter, as discussed in the next section.

7. Efficient digital processing

An important issue with most layered methods of improving spectral efficiency is that they
require many additional FTs at both the transmitter and receiver; although the transforms for
the higher layers can be reduced in length. Wang et al. have shown that a single FFT can be
modified to generate multiple layers within one transform. The arrangements proposed by Wang
et al. are shown in figure 7 for LACO [68] and ASE-DMT [67], which is based on layers of PAM.
Both techniques rely on extracting signals from within the core of the FFT for two reasons: (i) the
waveforms for the higher layers are available from the single transform and (ii) extracting these
waveforms is necessary so those subcarriers do not contribute to the final output of the transform.
The ASE-DMT method also makes use of the well-known result that one complex-valued-input
inverse FT can be used to generate two independent real-valued output streams [69]. This is
achieved by applying Hermitian and skew-Hermitian frequency domain inputs across positive
and negative frequencies, each producing an independent output after the inverse FFT, as shown
in the figure.

8. Other improvements
Yang et al. [70] have proposed an adaptive scheme where the number of layers is optimized for
a given required data rate, for either electrical or optical power constraints and SNRs. Using
simulations, they found that the number of layers should be constrained for poor SNRs. Their
conclusions support that single-layer ACO-OFDM should be used for poor SNRs, which also
means that the SE is poor.
Wang et al. [50] have proposed using simplified soft interference cancellation, followed by
a second stage of iterative or direct noise clipping, to obtain around 0.6-dB improvement over
Mohammed et al. [38].
Zhou et al. have reduced the PAPR of LACO-OFDM using single-frequency FDM [71],
which uses an FT before the transmitter’s IFFT, so that the transmitter creates a time domain
waveform, with a well-controlled spectrum [72]. A low PAPR is desirable to mitigate transmitter
nonlinearities, and also fibre nonlinearities in long-haul systems [73]. Bai et al. [74] have proposed
real and imaginary separation (RIS) to further reduce PAPR and improve BER in the presence of
LED nonlinearity.
 sum
(a) (b) 12
clip then clip then clip
repeat repeat
do not calculate

Layer 3 waveform
imaginary parts

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
A0, A1, A2, A3, from Layer 1’s PAM modl. all layers, a, b, c,
QAM inputs QAM inputs

are clipped

Layer 2 waveform
Layer 3

4 B1, B3, from Layer 2’s PAM modulator separately, then

C2, from Layer 3’s PAM modulator added together.

imaginary components
2

real comps. imag. comps.

green boxes split the c0, c4 b0
Layer 2

complex signals into c1, c5 b1

Layer 1 waveform
6 real and imaginary c2, c6 b2
parts. b3
c3, c7
b4
1 j( A0) b5
b6
j(–A0) b7
Layer 1 QAM
inputs

5 j( A2 + C2) a0
j(–A2 + C2) a1

real components
a2
3 j( A1 + B1) a3
j(–A3 + B3) a4
a5
7 j( A3 + B3) a6
j(–A1 + B1) a7

Figure 7. Methods of using a single inverse FFT to generate all layers at the transmitter (and regenerate them at the receiver):
(a) LACO-OFDM and (b) ASE-DMT. (Online version in colour.)

In 2013, Zhang & Hanzo proposed a multi-layer modulation scheme [75] similar to
superposition coding [76]. Rather than putting different layers on different subcarriers, the
layers are superimposed in the time domain with different amplitude weights; for example,
superimposing QPSK with a halved amplitude onto QPSK may lead to 16-QAM. The amplitude
weights were optimized by a genetic algorithm. Significant gains were achieved over ACO- and
DCO-OFDM. Similar algorithms could possibly improve the layered techniques outlined in this
paper.
There have been several papers on using multi-layer transmission to optimize the data
transmission at different dimming levels. Wang et al. [77] have shown that LACO-OFDM in
dimmable optical OFDM (DO-OFDM) is advantageous over DCO-OFDM at very high and very
low levels of illumination. The addition of pulse-width modulation gives many possibilities as
discussed in a recent paper by Li et al. [78].

9. Musical perception
The frequency allocation of LACO-OFDM subcarriers can be mapped onto a musical scale, as
illustrated in figure 8. Each layer has a series of notes that are odd multiples in frequency (i.e.
harmonics) of a fundamental tone, forming a chord. The higher layers start their chords at octaves
above the previous layer. The processing of LACO-OFDM starts with the lowest chord and cancels
it, and its distortion products, to reveal the next chord up.
It is interesting to speculate whether the brain uses similar processing to separate out
musical instruments in orchestral and other music, particularly rock music with heavily distorted
instruments. If the brain could first detect the lowest layers (such as a bass guitar), it would be able
to recreate the harmonic series of each note, so could cancel out the higher harmonics, to reveal
higher pitch instruments. This process could be repeated through the mid-range instruments, to
the highest pitch instruments or voice. In rock music, each instrument is usually quite distorted,
but importantly the distortion is created in a separate ‘guitar’ amplifier for each instrument, so is
harmonically related to each instrument, so may enable the brain to decipher the music. However,
if all instruments are played through a single distorting amplifier, then the sound becomes awful;
this was the experience of many teenage ‘garage’ bands who could only afford a single amplifier;
it may also be the reason why live music generally sounds better than recorded music.
 subcarrier 1 3 5 7 9 11 13 15
2 4 10 14 13
4 12
6

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
nearest C G¢ E≤ Bb≤ D≤¢ G b≤¢ B≤¢ Layer 1
musical Ab≤¢
C¢ G≤ E≤¢ Bb≤¢ Layer 2
note
C≤ G≤¢ Layer 3
C≤¢ Layer 4

Figure 8. Musical representation of LACO-OFDM as a series of chords at octave intervals, each conveying a layer. The chords
consist of odd harmonics of the fundamental note, C. (Online version in colour.)

10. Conclusion
The quest for power efficiency in intensity-modulated optical OFDM systems resulted in strongly
asymmetrically clipped waveforms that did not require additional DC-bias; unfortunately, these
halved the electrical spectral efficiency because many of the subcarrier frequency slots became
polluted by strong clipping distortion. The invention of layered techniques, in which the clipping
distortion is mostly cancelled by estimating the data on at least one unpolluted layer, then
subtracting this to reveal the wanted signals on deeper layers, has helped regain the spectral
efficiency. Importantly, layered systems using QAM theoretically require lower SNRs at the
receiver than conventional systems that use DC-bias or PAM that requires no bias. In addition,
diversity techniques can be used to gain further advantage because the clipping distortion also
contains information. Thus, layered systems should be adopted to maximize the performance of
intensity-modulated links.
A cost of layered systems is increased signal processing complexity at the transmitter and
receiver, to generate and then ‘peel back’ the layers, respectively. Some advances have been
made using ‘middle-out’ fast FTs, which are able to generate multiple layers with a single
transform. These reduce the processing at the transmitter, and also at the receiver because the
receiver requires ‘transmitter processing’ to regenerate the layers during the cancellation process.
Obviously, there is room for improvement in the receiver processing (except for spectral and
energy efficient-OFDM)––to reduce the number of transforms. There is also much work to be
done demonstrating the advantages of layered techniques in real systems with imperfections
(nonlinearities and memory) in the components, particularly in modulated lasers.
Data accessibility. This article has no additional data.
Competing interests. I serve as CTO of Ofidum Pty Ltd, Australia. Ofidium holds a number of patents in this area.
Funding. This work has been funded by the Australian Research Council under its Laureate Fellowship Scheme
(grant no. LF130100041).
Acknowledgements. I should like to thank my colleagues and students for their collegiality and contributions
to the development of optical OFDM over the years. The simulation results were created using
VPItransmissionMaker™, a product of VPIphotonics GmbH. I thank them for the licence to run this software.

References
1. Kahn JM, Keang-Po H. 2004 Spectral efficiency limits and modulation/detection techniques
for DWDM systems. IEEE J. Sel. Top. Quantum Electron. 10, 259–272. (doi:10.1109/
JSTQE.2004.826575)
2. Sharif M, Perin JK, Kahn JM. 2015 Modulation schemes for single-laser 100 Gbit/s links:
single-carrier. J. Lightwave Technol. 33, 4268–4277. (doi:10.1109/JLT.2015.2470523)
 3. Perin JK, Sharif M, Kahn JM. 2015 Modulation schemes for single-laser 100 Gb/s links:
14
multicarrier. J. Lightwave Technol 33, 5122–5132. (doi:10.1109/JLT.2015.2480700)
4. Winzer PJ, Essiambre RJ. 2006 Advanced optical modulation formats. Proc. IEEE 94, 952–985.

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
(doi:10.1109/JPROC.2006.873438)
5. Carruthers JB, Kahn JM. 1996 Multiple-subcarrier modulation for nondirected wireless
infrared communication. IEEE J. Sel. Areas Comms. 14, 538–546. (doi:10.1109/49.490239)
6. Chang RW. 1966 High-speed multichannel data transmission with bandlimited orthogonal
signals. Bell Syst. Tech. J. 45, 1775–1796. (doi:10.1002/j.1538-7305.1966.tb02435.x)
7. Weinstein SB. 2009 The history of orthogonal frequency-division multiplexing [history of
communications]. IEEE Commun. Mag. 47, 26–35. (doi:10.1109/MCOM.2009.5307460)
8. Sanjoh H, Yamada E, Yoshikuni Y. 2002 Optical orthogonal frequency division multiplexing
using frequency/time domain filtering for high spectral efficiency up to 1 bit/s/Hz. In Conf.
on Optical Fiber Communication, OFC, pp. 401–402. Anaheim, CA: OSA.
9. Qi P, Green RJ. 1996 Bit-error-rate performance of lightwave hybrid AM/OFDM systems with
comparison with AM/QAM systems in the presence of clipping impulse noise. IEEE Photonics
Technol. Letts. 8, 278–280. (doi:10.1109/68.484266)
10. Lu X, Bodeep GE, Darcie TE. 1995 Broad-band AM-VSB/64 QAM cable TV system over hybrid
fiber/coax network. IEEE Photonics Technol. Lett. 7, 330–332. (doi:10.1109/68.372762)
11. Fellows D, Jones D. 2001 DOCSIS/sup TM/cable modem technology. IEEE Commun. Mag. 39,
202–209. (doi:10.1109/35.910608)
12. You R, Kahn JM. 2001 Average power reduction techniques for multiple-subcarrier intensity-
modulated optical signals. IEEE Trans. Commun. 49, 2164–2171. (doi:10.1109/26.974263)
13. Lowery AJ, Armstrong J. 2005 10 Gbit/s multimode fiber link using power-efficient
orthogonal-frequency-division multiplexing. Opt. Express 13, 10 003–10 009. (doi:10.1364/
OPEX.13.010003)
14. Armstrong J, Lowery AJ. 2006 Power efficient optical OFDM. Electron. Letts 42, 370–371.
(doi:10.1049/el:20063636)
15. Lowery AJ, Armstrong J. 2006 Orthogonal-frequency-division multiplexing for dispersion
compensation of long-haul optical systems. Opt. Express 14, 2079–2084. (doi:10.1364/
OE.14.002079)
16. Li Z et al. 2016 Comparison of digital signal-signal beat interference compensation
techniques in direct-detection subcarrier modulation systems. Opt. Express 24, 29 176–29 189.
(doi:10.1364/OE.24.029176)
17. Fernando N, Yi H, Viterbo E. 2012 Flip-OFDM for unipolar communication systems. IEEE
Trans. Commun. 60, 3726–3733. (doi:10.1109/TCOMM.2012.082712.110812)
18. Fernando N, Hong Y, Viterbo E. 2011 Flip-OFDM for optical wireless communications. In
IEEE Inf. Theory Workshop, Paraty, Brazil, 16–20 October. Piscataway, NJ: IEEE.
19. Yong J. 2007 Modulation and demodulation apparatuses and methods for wired/wireless
communication. Korea Patent no. WOZ007/064165A.
20. Tsonev D, Sinanovic S, Haas H. 2012 Novel unipolar orthogonal frequency division
multiplexing (U-OFDM) for optical wireless. In 75th IEEE Vehicular Technology Conf. (VTC
Spring), Yokohama, Japan, 6–9 May. Piscataway, NJ: IEEE.
21. Lowery AJ, Du LB. 2011 Optical orthogonal frequency division multiplexing for long haul
optical communications: a review of the last five years. Opt. Fiber Technol. 17, 421–438.
(doi:10.1016/j.yofte.2011.07.009)
22. Mecozzi A, Antonelli C, Shtaif M. 2016 Kramers–Kronig coherent receiver. Optica 3, 1220–
1227. (doi:10.1364/OPTICA.3.001220)
23. Lowery AJ, Wang T, Corcoran B. 2019 Clipping-enhanced Kramers–Kronig receivers. In 2019
Optical Fiber Communications Conference and Exhibition (OFC), pp. 1–3, San Diego, CA, 3–7
March, paper M1H.2. Anaheim, CA: OSA.
24. Dissanayake SD, Panta K, Armstrong J. 2011 A novel technique to simultaneously transmit
ACO-OFDM and DCO-OFDM in IM/DD systems. In GLOBECOM Workshops (GC Wkshps),
Houston, TX, 5–9 December, pp. 782–786. Piscataway, NJ: IEEE.
25. Dissanayake SD, Armstrong J. 2013 Comparison of ACO-OFDM, DCO-OFDM and ADO-
OFDM in IM/DD systems. J. Lightwave Technol. 31, 1063–1072. (doi:10.1109/JLT.2013.2241731)
26. Ranjha B, Kavehrad M. 2014 Hybrid asymmetrically clipped OFDM-based IM/DD
optical wireless system. IEEE/OSA J. Opt. Commun. Networking 6, 387–396. (doi:10.1364/
JOCN.6.000387)
27. Wu N, Bar-Ness Y. 2015 A novel power-efficient scheme for asymmetrically and
15
symmetrically clipping optical (ASCOC)-OFDM for IM/DD optical systems. EURASIP J. Adv.
Sig. Proc. 2015, 1–10. (doi:10.1186/1687-6180-2015-1)

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
28. Katz E, Laufer A, Bar-Ness Y. 2012 A new improved-performance decoding technique for
Asymmetrically-Clipped Optical-OFDM. In 2012 46th Annual Conf. on Information Sciences and
Systems (CISS), Princeton, NJ, 21–23 March, pp. 1–6. Piscataway, NJ: IEEE.
29. Tsonev D, Haas H. 2014 Avoiding spectral efficiency loss in unipolar OFDM for optical
wireless communications. In Int. Conf. on Communications (ICC), Sydney, Australia, 10–14 June,
pp. 3336–3341. Piscataway, NJ: IEEE.
30. Tsonev D, Videv S, Haas H. 2015 Unlocking spectral efficiency in intensity modulation
and direct detection systems. IEEE J. Sel. Areas Commun. 33, 1758–1770. (doi:10.1109/
JSAC.2015.2432530)
31. Islim MS, Tsonev D, Haas H. 2015 A generalized solution to the spectral efficiency loss
in unipolar optical OFDM-based systems. In 2015 IEEE Int. Conf. on Communications (ICC),
London, UK, 8–12 June, pp. 5126–5131. Piscataway, NJ: IEEE.
32. Chen L, Krongold BS, Evans J. 2010 Successive decoding of anti-periodic OFDM signals in
IM/DD optical channel. In Presented at the IEEE Int. Conf. on Communications, ICC 2010, Cape
Town, South Africa, 23–27 May. Piscataway, NJ: IEEE.
33. Chen L, Krongold BS, Evans J. 2012 Performance analysis of optical OFDM transmission in
short-range IM/DD systems. J. Lightwave Technol. 30, 974–983. (doi:10.1109/JLT.2012.2185779)
34. Haas H, Tsonev D. 2019 Transmission scheme for communications systems, Patent no. US
10,236,993.
35. Wang Q, Qian C, Guo X, Wang Z, Cunningham DG, White IH. 2015 Layered ACO-
OFDM for intensity-modulated direct-detection optical wireless transmission. Opt. Express 23,
12 382–12 393. (doi:10.1364/OE.23.012382)
36. Islim MS, Tsonev D, Haas H. 2015 On the superposition modulation for OFDM based optical
wireless communication. In GlobalSIP, Orlando, FL, 14–16 December. Piscataway, NJ: IEEE.
37. Lowery AJ. Enhanced asymmetrically-clipped optical OFDM [Online]. See https://arxiv.org/
ftp/arxiv/papers/1510/1510.04771.pdf.
38. Mohammed MMA, He C, Armstrong J. 2017 Diversity combining in layered
asymmetrically clipped optical OFDM. J. Lightwave Technol. 35, 2078–2085. (doi:10.1109/
JLT.2017.2685591)
39. Chen L, Krongold B, Evans J. 2009 Diversity combining for asymmetrically clipped optical
OFDM in IM/DD channels. In Global Telecommunications Conf., 2009. GLOBECOM 2009. IEEE,
Honolulu, Hawaii, 30 November–4 December, pp. 1–6. Piscataway, NJ: IEEE.
40. Dissanayake SD, Armstrong J. 2011 Novel techniques for combating DC offset in
diversity combined ACO-OFDM. IEEE Commun. Lett. 15, 1237–1239. (doi:10.1109/LCOMM.
2011.092011.110709)
41. Zhang X, Babar Z, Zhang R, Chen S, Hanzo L. 2019 Multi-class coded layered asymmetrically
clipped optical OFDM. IEEE Trans. Commun. 67, 578–589. (doi:10.1109/TCOMM.
2018.2869821)
42. Babar Z, Zhang X, Botsinis P, Alanis D, Chandra D, Ng SX, Hanzo L. 2019 Near-capacity
multi-layered code design for LACO-OFDM-aided optical wireless systems. IEEE Trans. Veh.
Technol. 68, 4051–4054. (doi:10.1109/TVT.2019.2896764)
43. Song B. 2018 Short-Haul Optical Fibre Transmission Systems. PhD, Department of Electrical
and Computer Systems Engineering, Monash University, Melbourne.
44. Elgala H, Little TDC. 2014 SEE-OFDM: Spectral and energy efficient OFDM for optical IM/DD
systems. In IEEE 25th Annual Int. Symp. Personal, Indoor, and Mobile Radio Communication
(PIMRC), Washington, DC, 2–5 September, pp. 851–855. Piscataway, NJ: IEEE.
45. Lam E, Wilson SK, Elgala E, Little TDC. Spectrally and energy efficient OFDM (SEE-OFDM)
for intensity modulated optical wireless systems. arXiv.
46. Islim MS, Tsonev D, Haas H. 2015 Spectrally enhanced PAM-DMT for IM/DD optical wireless
communications. In Int. Symp. Personal, Indoor and Mobile Radio Communications (PIMRC),
30 August–2 September, pp. 927–932. Hong Kong. Piscataway, NJ: IEEE.
47. Islim MS, Haas H. 2016 Augmenting the spectral efficiency of enhanced PAM-DMT-based
optical wireless communications. Opt. Express 24, 11 932–11 949. (doi:10.1364/OE.24.011932)
48. Lowery AJ. 2016 Comparisons of spectrally-enhanced asymmetrically-clipped optical OFDM
systems. Opt. Express 24, 3950–3966. (doi:10.1364/OE.24.003950)
49. Wang Q, Wang Z, Guo X, Dai L. 2016 Improved receiver design for layered ACO-
16
OFDM in optical wireless communications. IEEE Photonics Technol. Lett. 28, 319–322.
(doi:10.1109/LPT.2015.2495320)

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
50. Wang TQ, Li H, Huang X. 2017 Diversity combining for layered asymmetrically clipped
optical OFDM using soft successive interference cancellation. IEEE Commun. Lett. 21, 1309–
1312. (doi:10.1109/LCOMM.2017.2668421)
51. Armstrong J, Schmidt BJC. 2008 Comparison of asymmetrically clipped optical OFDM
and DC-biased optical OFDM in AWGN. IEEE Commun. Letts. 12, 343–345. (doi:10.1109/
LCOMM.2008.080193)
52. Sun Y, Yang F, Gao J. 2017 Comparison of hybrid optical modulation schemes for visible light
communication. IEEE Photonics J. 9, 1–13. (doi:10.1109/jphot.2017.2705040)
53. Zhang X, Wang Q, Zhang R, Chen S, Hanzo L. 2017 Performance analysis of layered ACO-
OFDM. IEEE Access 5, 18 366–18 381. (doi:10.1109/ACCESS.2017.2748057)
54. Zhou J, Zhang W. 2018 A comparative study of unipolar OFDM schemes in Gaussian
optical intensity channel. IEEE Trans. Comms. 66, 1549–1564. (doi:10.1109/TCOMM.2017.
2785766)
55. Azhar AH, Brien DO. 2013 Experimental comparisons of optical OFDM approaches in visible
light communications. In 2013 IEEE Globecom Workshops (GC Wkshps), Atlanta, GA, 9–13
December, pp. 1076–1080. Piscataway, NJ: IEEE.
56. Wang Y, Wang Y, Chi N. 2014 Experimental verification of performance improvement for
a gigabit wavelength division multiplexing visible light communication system utilizing
asymmetrically clipped optical orthogonal frequency division multiplexing. Photonics Res. 2,
138–142. (doi:10.1364/PRJ.2.000138)
57. Tahar M, Wang TQ, Medina MFG, González O, Armstrong J. 2016 Experimental
demonstration of diversity combining for asymmetrically clipped optical OFDM. IEEE
Commun. Lett. 20, 906–909. (doi:10.1109/LCOMM.2016.2517003)
58. Islim MS, Haas H. 2019 An experimental demonstration of an energy efficient DMT technique
for LiFi systems. In Presented at the IEEE Int. Conf. on Communication Workshops (ICCC
Workshops), Shanghai, China, 20–24 May. Piscataway, NJ: IEEE.
59. Chen H, Ding J, Wu H, Bian S, Yan M, Liu W et al. 2018 Experimental demonstration of
free space optical card-to-card transmission architecture based on ADO-OFDM. In Advanced
Photonic Congress, Zurich, Switzerland, 2–5 July. Anaheim, CA: OSA.
60. Zhang Z, Chaaban A, Shen C, Elgala H, Ng TK, Ooi BS et al. 2018 Worst-case residual clipping
noise power model for bit loading in LACO-OFDM. In 2018 Global LIFI Congress (GLC), Paris,
France, 8–9 February, pp. 1–6. Piscataway, NJ: IEEE.
61. Gomez A, Shi K, Quintana C, Sato M, Faulkner G, Thomsen BC, O’Brien D. 2015 Beyond 100-
Gbit/s indoor wide field-of-view-optical wireless communications. IEEE Photo. Technol. Lett.
27, 367–370. (doi:10.1109/LPT.2014.2374995)
62. Ip E, Lau APT, Barros DJF, Kahn JM. 2008 Coherent detection in optical fiber systems. Opt.
Express 16, 753–791. (doi:10.1364/OE.16.000753)
63. Wang J, Liu J, Li S, Zhu L, Li A, Li C et al. 2015 Experimental demonstration of free-space
optical communications using OFDM-QPSK/16QAM-carrying fractional orbital angular
momentum (OAM) multiplexing. In Optical Fiber Communications (OFC), Los Angeles, CA,
22–26 March, p. M2F.5. Anaheim, CA: OSA.
64. Song B, Zhu C, Corcoran B, Wang Q, Zhuang L, Lowery AJ. 2016 Experimental
layered/enhanced ACO-OFDM short-haul optical fiber link. IEEE Photonics Technol. Lett. 28,
2815–2818. (doi:10.1109/LPT.2016.2623811)
65. Song B, Corcoran B, Wang Q, Lowery AJ. 2017 Field-trial of layered/enhanced ACO-OFDM.
In 2017 European Conf. on Optical Communication (ECOC), Gothenburg, Sweden, 17–21 September,
Ta.2.E.2, pp. 1–3. Piscataway, NJ: IEEE.
66. Song B, Corcoran B, Wang Q, Zhuang L, Lowery AJ. 2017 Subcarrier pairwise coding
for short-haul L/E-ACO-OFDM. IEEE Photonics Technol. Letts. 29, 1584–1587. (doi:10.1109/
LPT.2017.2737643)
67. Wang Q, Song B, Corcoran B, Zhuang L, Lowery AJ. 2017 Real-time demonstration of
augmented-spectral-efficiency DMT transmitter using a single IFFT. J. Lightwave Technol. 35,
4796–4803. (doi:10.1109/JLT.2017.2758401)
68. Wang Q, Song B, Corcoran B, Boland D, Zhu C, Zhuang L, Lowery AJ. 2017 Hardware-efficient
17
signal generation of layered/enhanced ACO-OFDM for short-haul fiber-optic links. Opt. Expr.
25, 13 359–13 371. (doi:10.1364/OE.25.013359)

royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 378: 20190180

...............................................................
69. Sorensen HV, Jones DL, Heideman M, Burrus CS. 1987 Real-valued fast Fourier
transform algorithms. IEEE Trans. Acoust., Speech Sig. Proc. 35, 849–863. (doi:10.1109/
TASSP.1987.1165220)
70. Yang F, Sun Y, Gao J. 2017 Adaptive LACO-OFDM with variable layer for visible light
communication. IEEE Photonics J. 9, 1–8. (doi:10.1109/jphot.2017.2768435)
71. Myung HG, Junsung L, Goodman D. 2006 Peak-to-average power ratio of single carrier
FDMA signals with pulse shaping. In 2006 IEEE 17th Int. Symp. Personal, Indoor and Mobile
Radio Communications, Helsinki, Finland, 11–14 September, pp. 1–5. Piscataway, NJ: IEEE.
72. Zhou J, Wang Q, Cheng Q, Guo M, Lu Y, Yang A, Qiao Y. 2018 Low-PAPR layered/enhanced
ACO-SCFDM for optical-wireless communications. IEEE Photonics Technol. Lett. 30, 165–168.
(doi:10.1109/LPT.2017.2779486)
73. Shieh W, Yan T, Krongold BS. 2010 DFT-spread OFDM for optical communications. In 2010
9th Int. Conf. on Optical Internet (COIN), Jeju, South Korea, 11–14 July, pp. 1–3. Piscataway, NJ:
IEEE.
74. Bai R, Wang Z, Jiang R, Cheng J. 2018 Interleaved DFT-spread layered/enhanced ACO-
OFDM for intensity-modulated direct-detection systems. J. Lightwave Technol. 36, 4713–4722.
(doi:10.1109/JLT.2018.2864275)
75. Zhang R, Hanzo L. 2013 Multi-layer modulation for intensity-modulated direct-detection
optical OFDM. IEEE/OSA J. Opt. Commun. Netw. 5, 1402–1412. (doi:10.1364/JOCN.5.001402)
76. Ma X, Li P. 2004 Coded modulation using superimposed binary codes. IEEE Trans. Inf. Theory
50, 3331–3343. (doi:10.1109/TIT.2004.838104)
77. Wang Q, Wang Z, Dai L, Quan J. 2016 Dimmable visible light communications based on
multilayer ACO-OFDM. IEEE Photonics J. 8, 1–11. (doi:10.1109/jphot.2016.2573040)
78. Li B, Xu W, Feng S, Li Z. 2019 Spectral-efficient reconstructed LACO-OFDM transmission
for dimming compatible visible light communications. IEEE Photonics J. 11, 1–14.
(doi:10.1109/jphot.2019.2892849)