Received 21 August 2020; revised 12 October 2020; accepted 19 October 2020.

Date of publication 27 October 2020; date of current version 23 November 2020.

Digital Object Identifier 10.1109/OJCOMS.2020.3034204

Optimization and Comparison of M-PAM and

Optical OFDM Modulation for Optical
Wireless Communication
SHOKOUFEH MARDANIKORANI 1 (Student Member, IEEE), XIONG DENG 1,2 (Member, IEEE),
AND JEAN-PAUL M. G. LINNARTZ 1,3 (Fellow, IEEE)
1 Department of Electrical Engineering, Eindhoven University of Technology, 5600 Eindhoven, The Netherlands

2 Academy of Advanced Optoelectronics, South China Normal University, Guangzhou 510006, China

3 Reserach Department, Signify (Philips Lighting) Research, 5656 Eindhoven, The Netherlands

CORRESPONDING AUTHOR: S. MARDANIKORANI (e-mail: s.mardanikorani@tue.nl)

This work was supported in part by the EU H2020 Project under Grant ELIOT 825651,
and in part by the National Natural Science Foundation of China under Grant 62001174.

ABSTRACT Illumination LEDs, but also infrared LEDs have limited bandwidth. To achieve high through-
put, one needs to modulate the LED significantly above its 3 dB bandwidth. Orthogonal Frequency Division
Multiplexing (OFDM) is a popular modulation technique to cope with the frequency selectivity of the
LED channel. In this article, we challenge whether its large Peak-to-Average-Power Ratio (PAPR) and
resulting large DC bias are justified. We compare systems using the same power and derive how PAM and
OFDM variants reach their optimum throughput at different bandwidths and differently shaped spectral
densities, thus at very different Signal to Noise Ratio (SNR) profiles but nonetheless the same transmit
power.When corrected for the path loss and normalized to the noise power in the 3 dB bandwidth of the
LED, we call this the Normalized Power Budget (NPB). OFDM can exploit the low-pass LED response
using a waterfilling approach. This is attractive if the NPB exceeds 60 dB. OFDM will then have to spread
its signal over more than ten times the LED bandwidth and requires a DC bias of more than 4 times the
rms modulation depth. Second-order distortion and LED droop may then become a limitation, if not com-
pensated. At lower power (NPB between 30 and 60 dB), DCO-OFDM outperforms PAM, provided that
it significantly reduces its bias and only if it uses an appropriate adaptive bit and power loading. Without
adaptive bit loading, thus using a frequency–constant modulation order, for instance made feasible by a
pre-emphasis, OFDM always shows lower performance than PAM; about 2.5 dB at a NPB around 60 dB.
Below 30 dB of NPB, even waterfilling cannot outweigh the need for a larger bias in OFDM, and PAM
should be preferred. We argue that a mobile system that has to operate seamlessly in wide coverage and
short–range high–throughput regimes, needs to adapt not only its bandwidth and its bit–loading profile,
but also its DCO-OFDM modulation depth, and preferably falls back from OFDM to PAM.
INDEX TERMS LED, VLC, IR, PAM, OFDM, waterfilling, pre-emphasis, optical wireless communication.

I. INTRODUCTION both academia and industry to provide an alternative tech-

T HE RAPID growth of bandwidth-intensive mobile

applications combined with the emerging Internet-of-
Things (IoT) services are putting immense pressure on the
nology for the currently predominantly RF-based connec-
tivity [1]–[4]. OWC, employing visible light (denoted as
Visible Light Communication, VLC) or the Infrared spec-
Radio Frequency (RF) spectrum. Recently, Optical Wireless trum (denoted as IR communication), offers unique features,
Communication (OWC) has gained research attention from such as free access to huge amounts of unregulated but
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

VOLUME 1, 2020 1721

MARDANIKORANI et al.: OPTIMIZATION AND COMPARISON OF M-PAM AND OPTICAL OFDM MODULATION FOR OPTICAL WIRELESS COMMUNICATION

TABLE 1. Current work in comparison with the previously published works.

largely interference–free bandwidth, a high degree of spatial costs extra electrical power that can deteriorate the over-
reuse, secure connectivity, and absence of electromagnetic all system efficiency and has to be limited. Thus, for VLC,
interference. extra consumed power is the key constraint, rather than total
The output optical flux of commercial Light Emitting electrical power [13]. Particularly for IR, human eye safety
Diodes (LEDs), illumination or IR LEDs, is modulated in can limit the average optical power to be transmitted by the
an Intensity Modulation Direct Detection (IM/DD) OWC LED [14].
system. This optical channel, typically, exhibits a low–pass To have a fair comparison of PAM and DC-biased Optical
frequency response with a 3 dB bandwidth that is dominated OFDM (DCO-OFDM) under certain constraints, one needs
by LED properties which are not optimized for commu- to operate both systems at their particular optimum. A proper
nication purposes. This low–pass nature, in particular the framework includes for OFDM:
LED junction capacitance attenuates higher frequencies in • optimum sub-carrier–dependent bit and power loading,
the intensity-modulated spectrum [3], [5]–[8]. In this respect, • optimized total bandwidth, and
line-of-sight OWC differs from Rayleigh or Rician distri- • optimum bias current and modulation depth, in relation
butions in radio communication where frequency-selective to the optimally tolerated clipping level, considering
fades are sufficiently narrow to be overcome by coding a realistic non-linear LED model (clipping, static and
and interleaving, employed in IEEE 802.11a/g standard. In dynamic higher–order terms),
OWC, excessive attenuation occurs in too wide portions of includes for PAM:
the bandwidth to rely on coding.
• pre-emphasis, with associated back-off to adhere to the
To handle the low-pass nature of LEDs, Orthogonal
power constraint and
Frequency-Division Multiplexing (OFDM) yet adapted for
• optimum bandwidth and modulation order as, in con-
optical applications (denoted as Optical OFDM, O-OFDM)
trast to non-dispersive AWGN channels adhering to
is popular [9], [10]. There is a persistent debate on whether
Shannon limits, we see that for the LED channel, the
multi-carrier OFDM outperforms carrier–free modulation,
optimum does not necessarily lies at the smallest con-
such as Pulse Amplitude Modulation (PAM) over an OWC
stellation (e.g., 2-PAM) and using the corresponding
low-pass channel. In fact, O-OFDM allows one to optimize
large bandwidth,
the distribution of the available modulation power among
the sub-carriers and to select the bit load independently on and for both modulation methods addresses
every sub-carrier to maximize the data rate [5]. However, the • the type of (extra) electrical or optical power constraint
OFDM composition of multiple frequency components has imposed by the application, and
a high Peak to Average Power Ratio (PAPR) that increases • the low-pass LED response.
the power consumption. A large DC bias needs to accommo- The comparison of different modulation schemes was stud-
date peaks in the signal. OFDM also requires highly linear ied extensively. For instance, [11], [12], [15] and [16] com-
amplifiers, which are inefficient. In an OWC link, a pre- pare OFDM variants with a PAM scheme, while [17], [18]
emphasis filter can be used in front of the LED to flatten address OFDM variants. In fact, with respect to the above
the channel frequency response. In this case, OFDM might listed aspects, previous papers known to us lack at least
no longer be needed. In such a flattened channel, using one aspect or do not generalize their findings into generic
the simpler PAM modulation with lower PAPR reduces the expressions that extend outside the simulation range. We
biasing power waste [11], [12]. A comparison involves con- summarize the comparison between prior art and this work
sideration of many aspects, which we further extend in this in Table 1.
article. DCO-PAM, thus level-shifted, non-negative PAM was
Depending on the application, the constraint on the chan- found in [11] to outperform all variants of OFDM in terms
nel differs. For VLC, the DC power is already available for of optical power efficiency (including DC bias power) over a
the illumination and the modern LEDs are designed to have range of spectral efficiencies. In [11], a Decision Feedback
a high wall-plug-to-lumen efficiency. However, modulation Equalizer (DFE) was used to combat the LED low-pass

1722 VOLUME 1, 2020

nature and the optimization of the bit loading in OFDM (non-optimized) bandwidth causes pronounced, abrupt dis-
could not revert this finding. However, the DFE has a continuities in the throughput, versus changes in the SNR [5].
high complexity, while we show that already with a simple That is, e.g., if the receiver gradually moves away from the
pre-emphasis filter, PAM can become attractive, provided transmitter, there will be a stepwise, non-graceful cut-off in
that also the constellation size is optimized for the LED throughput. In [18], throughput achieved by OFDM-based
response, in particular allowing M = 8, .. for high Signal to schemes were discussed. The clipping noise as well as the
Noise Ratio (SNR). In fact, [11] used 2 and 4-PAM only, distortion introduced by the LED are modelled. However,
presumably because of the DFE restrictions. the results of [18] did not include the frequency selectivity
Numerical optimization for a constrained peak opti- of the LED channel.
cal power in [12] showed that in a limited bandwidth, To optimize OFDM for frequency selective LED channels,
DCO-PAM performs better. However, in contrast to RF, different power and bit loading strategies have been discussed
the bandwidth in unregulated OWC is a design freedom in the literature, e.g., [5], [20]–[25]. Waterfilling and uniform
that preferably is not a priori restricted. We see that for bit loading (also known as pre-emphasized power loading)
the same transmit power constraint, different modulation are the two well-known strategies. Waterfilling is known as
methods and different constellations each have a different the optimum strategy that results in the maximum through-
optimum bandwidth, and that it leads to different SNR pro- put in a frequency selective communication channel [20].
files along the frequency axis. Hence SNR is not a preferred However, it requires a (relatively) complex algorithm [5],
benchmark. [21], [22]. The existing ITU g.9991 standard [26] simplifies
In DCO-OFDM, the modulation depth, relative to the this into assigning the same power level to all sub-carriers,
DC-bias determines the amount of clipping. This may but adapts the constellation per sub-carrier. Forcing a uni-
prohibit the use of larger modulation orders. In [12], clip- form constellation on all sub-carriers would further simplify
ping noise was assumed to have a flat spectrum at the the implementation to a great extent [25]. This is also con-
receiver over all FFT outputs regardless of the actual sig- sidered in the current standardization of IEEE 802.11bb [27],
nal bandwidth. However, we show that the clipping noise as it can reuse approaches designed earlier for RF channels.
predominantly depends on the modulation bandwidth. That In this work both waterfilling and pre-emphasis strategies
is, one cannot arbitrarily spread clipping noise outside are considered.
the signal bandwidth by using faster, oversized FFT pro- The main contributions of this work include the following:
cessing at the receiver. Moreover, the clipping artefacts • In many other communication channels, using a higher
further are subject to the low–pass LED frequency response. bandwidth enhances throughput. In contrast to this,
As shown in Table 1, this was simplified in previous we show that for an LED there exists an optimum
works. modulation bandwidth beyond which the throughput
The work in [15] compares single–carrier (but frequency– reduces. Moreover, OWC standards that fix bandwidth,
domain equalized) M-PAM modulation to multiple OFDM as radio standards typically do, abruptly fail to sustain
variants, with a main focus on multi-path dispersion of the a weakening link.
OWC propagation channel. M-PAM appeared to require a • To make a fair comparison among systems that optimize
lower SNR to achieve the same Bit Error Rate (BER). Both their transmit bandwidth, we introduce the Normalized
LED clipping and low-pass memory effects are covered in Power Budget (NPB), defined as transmit power cor-
numerical simulations. However, no further optimizations for rected for path loss, normalized to the noise in the
modulation bandwidth nor for a (frequency–adaptive) mod- 3 dB bandwidth of the LED. In fact, we cannot use the
ulation order are discussed. On-Off Keying (OOK) shows a bandwidth of transmit signal as different modulation
better optical power efficiency than DCO-OFDM and unipo- strategies optimize differently.
lar Asymmetric Clipped Optical OFDM (ACO-OFDM) in • We derive mathematical expressions for the throughput
single-mode fiber systems [16], where the DC bias and and the preferred modulation bandwidth for DCO-PAM
bandwidth optimizations were carried out by numerical sim- and DCO-OFDM. Using the now commonly reported
ulations, considering clipping for low biases. In fact, at exponential OWC channel frequency response [28],
low available transmit power, ACO-OFDM can become [29], we capture these in new expressions. Hitherto,
more attractive than DCO-OFDM [11], [19]. However, the comparisons were mostly limited to simulations
Section VII-B2 shows that PAM reaches higher throughput for specific settings, thereby did not give generic
for the same power. expressions for other settings. Furthermore, we derive
In [17], OFDM has been studied for VLC in flat expressions for the optimum modulation bandwidth for
and dispersive channels, addressing also clipping noise DCO-OFDM and for (DCO-) PAM, considering discrete
while optimizing the DC offset of OFDM. However, the modulation orders and optimizing for the LED low-pass
practical limitations of a discrete modulation order and response. Our optimization includes the impact of lim-
an optimization of the modulation bandwidth for OFDM iting the DC bias for an OFDM signal, by allowing
were not discussed. On a pre-emphasized channel, the clipping and by making a trade off with the resulting
use of a fixed number of constellation bits over a fixed clipping noise.

VOLUME 1, 2020 1723

MARDANIKORANI et al.: OPTIMIZATION AND COMPARISON OF M-PAM AND OPTICAL OFDM MODULATION FOR OPTICAL WIRELESS COMMUNICATION

• We quantify clipping for DCO-OFDM as it raises the LED channel model in Section II. Section III presents the
perceived noise floor and thereby limits the usable DCO-PAM model, its performance over an OWC channel
modulation order, even in a further noise–free channel. and the DC penalty required. DCO-OFDM is discussed
Following arguments in [5], [11], [17], [30], [31], we in Section IV. Both the continuous (for theoretical pur-
conclude that for modern LEDs, a saturation peak limit poses) and discrete (practical case) modulation orders are
does not accurately model the behavior. We use and discussed. This section also presents the optimum waterfill-
extend the clipping noise model of [17] which con- ing approach results for the comparison. The DC penalty and
sidered one-sided clipping of the LED current. This the clipping noise associated with DCO-OFDM is discussed
extends our previous bit loading evaluations in [5], in Section V. In Section VI a proper measure is given to
which assumed clipping-free DCO-OFDM, leading to choose a proper DC bias for the LED based on the modula-
more complete, realistic model. tion order. Furthermore, this section includes the distortion
• We compare constrained optical power (related to the power due to clipping (to reduce the DC penalty) of the LED
average LED current), the extra electrical power (related current in the throughput and modulation bandwidth require-
to the variance of the current caused by modulation) ment. Section VII compares DCO-OFDM and DCO-PAM
and the total electrical power (related to a combina- in three different contexts, VLC, IR and average–optical–
tion of DC current and AC variance, weighted by the power constrained channels. The computational complexity
LED (say, bandgap) voltage and the dynamic resis- of DCO-OFDM and PAM is discussed in Section VIII.
tance, respectively). While previously published works, Finally, conclusions are drawn in Section IX.
e.g., [11], [12], [15], and [16], often report outspoken
preferences for the choice of modulation, we conclude
II. OWC POWER CONSTRAINTS
that there is not always simple unique answer to the
Various LED models are used in scientific literature. This
question whether OFDM and PAM is performing better,
section elaborates on our LED model, that considers non-
depending on which constraint applies.
negativity, junction voltage and LED junction capacitance
• We show that in a VLC context, where the extra power
and resistances. So, in fact following [5], we consider LED
needs to be far below the illumination power, there is
low–pass nature and one–sided clipping. We model that elec-
no difference in performance between pre-emphasized
trical power consumption not only grows with the DC bias,
DCO-OFDM and a DCO-PAM. However, DCO-OFDM
but also with the modulation variance. In contrast to this,
with waterfilling outperforms DCO-PAM.
the average optical power only relates to the biasing, while
• For IR, where the bias or the mean DC light has to be
modulation comes for free, in the sense that DC-free mod-
paid for from the communication power budget, PAM
ulation does not affect the average current. We denote the
with an appropriate high-boost and a carefully cho-
LED current to consist of ILED (t) = ILED + iled (t), where
sen bit rate and bandwidth outperforms pre-emphasized
iled (t) is the zero–mean (AC) modulation current and ILED
OFDM. Our model of the impact of clipping artefacts
is the DC current of the LED to ensure ILED (t) ≥ 0. The
allows us to optimize the choice of the modulation
DC power consumption of the LED is, PDC = VLED ILED .
depth for OFDM. In fact, one can intuitively interpret
Here, the DC voltage VLED can be expressed as
our results as a quantification that the power penalty
incurred for the DC bias in pre-emphasized DCO- VLED ≈ V0 + RLED ILED , (1)
OFDM is not compensated by the ability to adaptively
load sub-carriers over a certain NPB range. For high where V0 can be interpreted as the turn-on limit and RLED is
power budgets, say NPB above 30 dB, however, OFDM the dynamic plus parasitic resistance of the LED [32], [33].
with waterfilling and optimum choice of LED bias cur- So the total electrical power consumed by the LED is,
rent outperforms PAM. Here, OFDM can fully exploit
1
the adaptive bit and power loading. For high power Ptot = V0 ILED + RLED I2LED + RLED σmod
2
, (2)
budgets one can afford a large back-off of the modula- η
tion depth to avoid clipping of the OFDM signal, the where σmod
2 is the variance of LED AC current iled (t) and
latter conclusion disagrees with [12]. We show that the where η is the modulation LED power amplifier efficiency,
crossover point where OFDM with waterfilling outper- used in a Bias-T setting [34]. An extensive study [34] into
forms PAM moves to higher power budget values when the power efficiency of a series transistor modulator revealed
LED is biased at higher currents. If, instead, more LEDs a total power consumption of Ptot ≈ (V0 + 2RLED ILED )ILED
were used to boost coverage, this would not happen. where factor 2 is due to an extra voltage headroom
• We propose a simple rule of thumb and an algorithm RLED {max ILED (t)} required to operate the modulating series
to optimize the modulation order and the modulation transistor [34]. More generically, a versatile power constraint
bandwidth of M-PAM, which works for both VLC and is the weighted sum of moments of the probability of the
IR applications. signal
The rest of the paper is organized as follows. We start
with a short introduction to the OWC link and the realistic Ptot = PDC + Pext = β1 ILED + β2 σmod
2
+ β3 I2LED , (3)

1724 VOLUME 1, 2020

where Pext is the extra power on top of the DC power con- B. CHANNEL MODEL
sumed by the LED due to modulation and the β-weights The low-pass frequency response of the LED channel from
may also depend on the electronic topology and the use LED current to photodiode current can be modeled as a
case (VLC vs IR). In fact, various papers take different low-pass filter [5], [28], [29],
interpretations of β1 , β2 and β3 , as we will discuss in the
|H( f )|2 = H02 2−f /f0 , (4)
next sections.
where H0 and f0 are the low frequency channel gain and the
A. POWER CONSTRAINT 3 dB cut-off frequency, respectively.
For IR and VLC communication, the power can be con- We focus on Line-of-Sight (LoS) channels. In fact, we
strained either in the optical or the electrical domain, which increasingly see the creation of beam steering emitters and
may lead to different optimizations. of angular diversity receivers. In such case, each resolved
1) OPTICAL POWER CONSTRAINT angular path is not likely to be subject to a significant
delay spread. Hence, we believe that the reflection-free
Optimizations for the optical domain, for instance dictated
LoS assumption remains relevant. If nonetheless long delay
by eye-safety in IR or illumination level in VLC, basically
spreads occur, a linear time–domain equalizer can become
limit the average (or DC-bias) LED current ILED : (β1 > 0,
complex for PAM, and frequency–domain equalization may
β3 = 0), but do not impose a power penalty for modulation
be preferred, as in OFDM [15].
(β2 = 0). As we will quantify in Section V, the DC current
ILED needs to accommodate the LED input current AC excur-
C. NORMALIZED POWER BUDGET (NPB) DEFINITION
sions. Hence, it nonetheless becomes an indirect function of
Often, systems are compared based on the (frequency–
σmod
2 to ensure a sufficiently low clipping distortion.
average) SNR at the receiver, for a particular choice of the
2) ELECTRICAL POWER CONSTRAINT IN IR modulation bandwidth. However, this leads to an intrinsically
The associated electrical power also depends on the vari- unfair comparison as PAM and OFDM benefit differently
ance via β2 = RLED . For a constrained total electrical power, from expanding the modulation bandwidth further beyond
in (2), β1 = V0 , β2 = β3 = RLED . In fact, the non-linear the LED 3 dB bandwidth. In fact, bandwidth is a parameter
current-voltage curve, approximated in (1), was further sim- subject to modulation–specific optimization constrained by
plified by omitting the photonic junction voltage (V0 = 0) transmit power (see for instance Figure. 1 and Figure. 4).
in [17], taking Ptot = RLED (I2LED + σmod 2 ), thus β = β This prohibits us to compare two systems just with the same
2 3
and β1 = 0. Yet, V0 dominates the voltage across the LED bandwidth or with the same SNR. Although, it seemingly
(V0 > RLED ILED ). Hence, OWC sees a large β1 , so that complicates the number of variables, we must restrict a com-
Ptot ≈ V0 ILED may be reasonable as a first-order estimate, parison to essential parameters that are not a design freedom.
particularly if V0 is adjusted for typical biased VLED voltages. We use H0 , f0 , the modulation rms σmod
2 and the noise spec-
tral density N0 , represented in A /Hz, referenced to currents
2
3) EXTRA POWER CONSTRAINT IN VLC through the photodiode detector at the receiver and we define
The primary function of VLC is illumination, so the DC the NPB γ as
current of the LED is determined by the target illumi- σmod
2 H2
nation level [18], and is not subject to a communication γ = 0
. (5)
N0 f0
optimization (β1 = 0, β3 = 0). As communication is a
secondary function, the illumination system is expected to In fact, normalizing to the LED bandwidth f0 and not to
deliver a high lumen-per-wall-socket-watt. Consequently, any signal bandwidth fmax allows us to plot generic curves for
additional consumption of power just for modulation dete- throughput. To optimally cope with the frequency-dependent
riorates the energy efficiency and may even jeopardize the channel response H( f ), we take the freedom to optimize
’green’ certification of the LED lighting product. Hence, an the emitted spectral density Sx ( f ) and the total bandwidth.
important VLC design objective is to get the highest pos- The subscript x indicates the modulation strategy; PAM for
sible throughput for the least amount of extra power, in a DCO-PAM, p for DCO-OFDM with pre-emphasis and w
regime where DC bias is not the dominant scarce resource. for DCO-OFDM with waterfilling. The noise bandwidth is
In fact, modulating the LED current consumes extra power subject to dynamic adaptations and the SNR is frequency
Pext ≈ RLED σmod
2 /η, as reflected in β = R
2 LED /η [5], [19].
dependent:
An (in-) efficiency of the amplifier (LED driver) can be Sx ( f )|H( f )|2
reflected in η. If we are only interested in the LED power, we SNR( f ) = . (6)
N0
take η = 1. However, inefficient (linear) modulator ampli-
We denote the frequency-domain spectral density of iled (t)
fiers in VLC make the overall LED lighting product less
by Sx ( f ), expressed in A2 /Hz. Over the signal bandwidth,
efficient, even to the extend that it fails lighting energy con-
Sx ( f ) integrates to σmod
2 . That is,
servation regulations [5]. Hence, VLC optimizations on Pext 
are highly relevant (thus with β1 , β3 = 0). In this article, we
Sx ( f )df = σmod 2
. (7)
evaluate systems limited by extra power in Section VII-A. f

VOLUME 1, 2020 1725

MARDANIKORANI et al.: OPTIMIZATION AND COMPARISON OF M-PAM AND OPTICAL OFDM MODULATION FOR OPTICAL WIRELESS COMMUNICATION

TABLE 2. Required received average energy per dimension normalized to N0 for

Here, σmod
2 and γ address effective signal powers thus allow different constellation size (M) and the minimum normalized bias requirement z at
BER = 10−4 .
the calculation of link performance, but ignore DC-biasing
power. We relate these to consumed power later when we
invoke β weight factors.
For a pre-emphasized spectrum, the received modulation
spectrum after the photodiode, Sx ( f )|H( f )|2 , is flat over
frequency. To achieve this, Sx ( f ) inverts H( f ) according to

σmod
2
Sx ( f ) = κ 2f /f0 , (8)
fx The BER is expressed as [37]
   
where κ is the pre-emphasis back-off to satisfy the con- 2 M−1 6N
BERM = Q
. (14)
straint (7) and fx is the modulation bandwidth over which log2 M M M 2 − 1 N0
the Sx ( f ) is spread. Inserting (8) into (7), the coefficient κ
is calculated as Thus, the average energy requirement of a M-PAM scheme,
normalized to N0 and denoted as X(M) for a pre-determined
ln(2)fx /f0
κ= . (9) BERM is
2fx /f0 − 1   2
M2 − 1 −1 M log2 M
If, for PAM, instead of a pre-filter, a linear post-equalizer X(M) = Q BERM . (15)
6 2(M − 1)
is used, the transmit current density is uniform, or fully
determined by the pulse shaping. However, the receive filter We list X(M) in Table 2 and extend it to M 2 -QAM by inter-
will then boost the noise in every sample by κ. That is, the preting QAM as just a 2D variant of M-PAM. We explicitly
SNR for every PAM sample is the same for either a pre or use a different symbol X(M) to represent a fixed system
post equalization (κ applies). property, while N /N0 is a property of the incoming sig-
nal, subject to optimization and may even be frequency
dependent.
III. PULSE AMPLITUDE MODULATION (PAM)
Within a Nyquist bandwidth of fPAM , a system reaches a
PAM requires a flat frequency response for Inter–Symbol throughput RPAM of
Interference (ISI) free communication. To repair the low–
pass LED frequency response, as in (4), a linear equalizer RPAM = 2fPAM log2 M. (16)
can be used to boost high frequency components [35], [36].
According to Nyquist theory, a baseband PAM signal with A. PAM BIAS PENALTY
a bandwidth fPAM can accommodate 2TfPAM symbol dimen- For PAM as in (11), a DC-bias of at least (M − 1)dM is
sions in a time interval T. For a symbol duration Ts needed to make the LED signal non-negative. We define a
(Ts = 1/(2fPAM )), we multiply the numerator of the SNR parameter z to be the ratio of the bias current over the LED
in (6) by 2Ts fPAM , thus by unity, to get rms current. For PAM,

2SPAM ( f )|H( f )|2 fPAM Ts ILED (M − 1)dM M−1
SNRPAM ( f ) = =
2N
. (10) z= =  = 3 , (17)
N0 N0 σmod M −1
2 M+1
3 dM

where N is the average received symbol energy per dimen- where the variance of the modulation σmod can be calculated
sion. For PAM, the energy per symbol s equals N , while from (11), as σmod = dM (M 2 − 1)/3. For such DCO-PAM,
for two-dimensional QAM, as used in OFDM, s = 2 N . the parameter z depends on the modulation√ order M. It equals
In (bi-polar) M-PAM, input data are mapped into a zero- z = 1 for M = 2 and approaches z = 3 for M → ∞. We
mean sequence of symbols chosen from M discrete levels, will use this parameter in the later sections to compare PAM
uniformly spaced by distance 2dM , so with OFDM.

sm = mdM , m ∈ {±(M − 1), ±(M − 3), . . . , ±1}. (11) B. THROUGHPUT OF DCO-PAM OVER LOW-PASS
CHANNEL
The average energy per symbol (at the receiver), s , is
Inserting (8) into (6) with κ given in (9) and the channel
2 
M/2 model (4), the SNR( f ) for PAM becomes
2dM M2 − 1 2
s = N = m2 = dM . (12) σmod
2 H2
ln(2)
M 3 SNRPAM ( f ) = 0
· f /f . (18)
m=1
N0 f0 2 PAM 0 −1

The distance dM can then be expressed as a function of N To benchmark our results, we also relate it to the NPB
as follows: defined in (5),

3N ln(2)
dM = . (13) SNRPAM ( f ) = γ f /f . (19)
M2 − 1 2 PAM 0 −1

1726 VOLUME 1, 2020

 2-PAM, by increasing the bit rate while adhering to the trans-
mit power constant. If it turns out that for this throughput, the
corresponding fmaxPAM exceeds 3.85 f0 , then the algorithm
adopts 4-PAM, and searches for the new highest sustain-
able bit rate by scaling down fmax . The limits of fmaxPAM for
which 8-PAM and 16-PAM are appropriate appear to be 5.8
f0 and 7.7 f0 , respectively. For a total–power limited chan-
nel, similar numbers apply. When a communication link is
operational, one preferably uses receiver feedback to change
the symbol rate while keeping M fixed, but only switch up
FIGURE 1. (a) (Optimum) normalized modulation bandwidth (fmaxPAM /f0 ,
fmaxp /f0 and fmaxw /f0 for PAM, pre-emphasized OFDM and waterfilling, or down M at specific threshold symbol rates. Figure. 1(b)
respectively) and (b) normalized throughput versus NPB used for modulation, γ , shows that the penalty for sticking to suboptimal M can be
ignoring DC-bias power (VLC scenario). Dashed-red lines represent the performance
for various constellation sizes M (for PAM and pre-emphasized DCO-OFDM) with the substantial. At higher NPB, sticking to 2-PAM or 4-PAM is
solid red being the choice of M optimized for maximum throughput. Solid blue and not attractive. Similarly, sticking to a pre-configured, pos-
black lines represent the performance of OFDM with pre-emphasis and waterfilling,
respectively, for continuous modulation order. For all plots BERM = 10−4 . sibly sub-optimum fmax , thus only adapting M, can have a
significant penalty and leads to a full collapse of the link at
In (10), we derived an equivalent expression for the SNR as some low γ .
a function of N . From (10) and (19), the achieved N /N0 , IV. OFDM
expressed in terms of the NPB and the bandwidth in PAM OFDM can naturally handle the frequency selective LED
modulation is behavior by dividing the input information over multiple
N ln 2 sub-carriers, with a aggregate bandwidth that can be multiple
= γ f /f
. (20)
N0 2 2 PAM 0 − 1 times the channel 3-dB bandwidth. As each sub-carrier only
occupies a small fraction of the modulation bandwidth, it
In order to support a constellation M, the N /N0 must exceed sees a (relatively) flat channel frequency response. A sub-
X(M) (given in Table 2 and defined in (15)). So, we require carrier at frequency f with a bandwidth f can accommodate
2X(M)  fPAM /f0  Tf two dimensional M 2 -QAM symbols in a time dura-
γ ≥ 2 −1 . (21)
ln 2 tion T. The duration of one OFDM block is Ts = 1/f .
This allows us, for any NPB γ and M, to find the modulation As the symbol energy equals s ( f ) = Sx ( f )|H( f )|2 fTs =
bandwidth, Sx ( f )|H( f )|2 , we can rewrite the SNR as
  s ( f ) 2N ( f )
ln 2 γ SNROFDM ( f ) = = . (23)
fPAM ≤ f0 log2 +1 . (22) N0 N0
2 X(M)
For OFDM, each sub-carrier symbol is received with a dif-
For any M, we fully utilize the available power when fPAM ferent energy, thus preferably it is loaded with its optimized
is set to reach an equality. For BERM = 10−4 and M = constellation M( f ). Therefore, we explicitly write N ( f ) as
2, . . . , 32, we use the X(M) values of Table 2 to plot fPAM a function of frequency.
as a function of γPAM in Figure. 1(a), shown with dashed red
lines. We use (16) to plot the throughput in Figure. 1(b) for A. THROUGHPUT OF DCO-OFDM OVER LOW-PASS
various M as the function of γ . Normalization to f0 allows CHANNEL
us to plot generic curves, not specific for the bandwidth of We use BER formula (14) for (two-dimensional) M 2 −QAM
the chosen LED. by including the frequency dependency of N ( f ). Taking the
For each γ value, the optimum value of M is the one inverse of (14), the modulation order M( f ) of the sub-carrier
that gives the highest throughput, shown in Figure. 1(b) at frequency f can be expressed as

with a solid red line. The corresponding optimum (or maxi- 2N ( f )
mum) normalized modulation bandwidth fmaxPAM to achieve M( f ) = + 1, (24)
N0
the maximum throughput is also shown in Figure. 1(a) with
a solid red line. We learn from Figure. 1, that for a NPB (γ ) where
  2
up to 24.3 dB and for BERM = 10−4 , the optimum modula- M( f ) log2 M( f )
tion is OOK (2-PAM). In fact for a NPB below 24.3 dB, it is = 1/3 Q−1 BERM (25)
2(M( f ) − 1)
preferred to use a crude modulation method very far beyond
is the modulation gap. The gap is a slightly decreasing func-
the 3 dB bandwidth of the LED rather than to choose a
tion of the modulation order M [5]. For simplicity we use
higher constellation to stay within the LED bandwidth. This
the worst case of modulation order M = 2 which gives a
NPB also corresponds to a fmaxPAM = 3.85 f0 . This insight
maximum for the given BERM . This simplifies (25) into
can be the basis for a practical algorithm to find, adapt and   2
track the best compromise between bandwidth and M: ini- −1 BER
= 1/3 Q , (26)
tially search for the highest throughput that is possible with 2
VOLUME 1, 2020 1727
MARDANIKORANI et al.: OPTIMIZATION AND COMPARISON OF M-PAM AND OPTICAL OFDM MODULATION FOR OPTICAL WIRELESS COMMUNICATION

where BER is the total bit error rate, BER ≈ 2BERM . The by (32), respectively. Practical algorithms such as Hughes-
number of bits b( f ) per dimension that can be delivered is Hartogs (HH) [21], [22] provide an iterative, discretized
  algorithm to calculate the optimum bit and power loading
1 2N ( f )
b( f ) = log2 (M( f )) = log2 1 + . (27) distribution. In [5], a good match between the theoret-
2 N0
ical throughput and the throughput achieved by discrete
Inserting (23) and (6) results in constellations using HH is shown. It optimizes the through-
  put, however, with high complexity and large overhead in
1 1 Sx ( f )|H( f )|2
b( f ) = log2 1 + . (28) communicating the used constellation on all sub-carriers.
2 N0
The throughput1 over a modulation bandwidth [0, fx ] is C. OFDM WITH PRE-EMPHASIS
obtained by integrating all the rate contributions, given A simpler implementation is to pre-emphasize the chan-
by (28), nel and to use the same constellation for all sub-carriers.
 fx Pre-emphasizing implies a forced inversion of the chan-
R= 2b( f )df nel response at the transmitter to compensate its low–pass
0
 fx   behaviour. This is often referred to as a bandwidth exten-
1 Sx ( f )|H( f )|2 sion, but comes at a penalty. Such pre-emphasis tends to
= log2 1 + df . (29)
0 N0 defeat the advantage of OFDM to load every frequency bin
The factor 2 reflects the two dimensions per second per Hz optimally, thus is counterproductive. Nonetheless, we see
of QAM. This expression looks like a misused Shannon limit IEEE 802.11bb standardization proposals to reuse WiFi-like
for AWGN channels, which repeatedly was argued not to be OFDM schemes with constant constellations for OWC, to
valid for optical channels. However, here (28) and (29) come use existing IC designs. Our results will show that repairing
just as a consequence of inverting the BER expression. the frequency response to support a fixed constellation can
be reasonable in the lower NPB ranges, but the transmit
bandwidth needs to be made adaptive to the NPB.
B. OFDM WITH WATERFILLING
In practice, the constellation size M can only take values 1) ARBITRARY MODULATION CONSTELLATIONS:
from the discrete set {2, 4, 8, . . .}. However, for theoreti-
A filter inverts the LED low–pass response in the frequency
cal derivations it is convenient to assume that M can take
range [0, fp ]. The throughput Rp is derived from (29) and (8)
any arbitrary positive value, including a non-integer one.
with the back-off κ given in (9):
As argued in [5], regardless of the choice of β1,2,3 , any  
 
optimized power spectral loading is equivalent to applying Rp fp γ ln 2
constraint (7) to choose the transmitted Sx ( f ) to maximize = log2 1 + f /f
. (33)
f0 f0 2p 0 −1
the throughput (29). Lagrangian optimization leads to the
well-known waterfilling solution with Sx ( f ) adhering to [20] The optimum modulation bandwidth, denoted by fmaxp , to
 + maximize the throughput is calculated from dRp /dfmaxp = 0,
N0 N0 which depends only on γ , f0 and [5].
Sw ( f ) = − , (30)
|H(fmaxw )|2 |H( f )|2
2) DISCRETE MODULATION CONSTELLATIONS
where the subscript w refers to waterfilling and fmaxw is the
maximum modulation frequency for which Sw ( f ) is non- Using discrete M, in (33), we cannot get tractable expressions
zero. The optimal power allocation of (30) shows that low for the derivatives w.r.t. spectral density. As an alternative
frequency sub-carriers that experience a good channel quality optimization track, we exploit the fact that all sub-carriers
are assigned more power than those at higher frequencies. carry the same constellation size M. In the previous sub-
Substituting (30) into (7) and solving the integral relates the section, we implicitly assumed a continuous-valued M,
optimum modulation fmaxw to the NPB γ : but in this section, we assume an M 2 -QAM modulation
    that can only take integer values of an even power of 2
ln(2)fmaxw fmaxw /f0 (M = 2, 4, 8, ..) and identical on all sub-carriers. We use
γ = 1+ −1 2 . (31)
ln(2) f0 the relation (23) to express N /N0 in terms of SNR( f ), as
The maximum throughput is calculated by inserting (30) in (6) but with a pre-emphasized spectral density (8),
into (29) and integrating over [0, fmaxw ]: 2N ( f ) σ 2 H2 ln 2
     = mod 0 . (34)
Rw 1 fmaxw |H( f )|2 1 fmaxw 2 N0 (
N0 f0 2 p 0 ) − 1
f /f
= log2 df = . (32)
f0 f0 0 |H(fmax )|2 2 f0 Our optimization tests various M and for each M value, the
optimum modulation bandwidth fp is taken such that N /N0
For a given γ , the optimum modulation bandwidth and the just exceeds X(M). This results in
throughput are implicitly given by the inverse of (31) and  
fp γ ln(2)
= log2 +1 , (35)
1. Gross rate before coding. f0 2X(M)
1728 VOLUME 1, 2020
which is identical to (22). The throughput for pre-emphasized electron-hole pair is the highest. This is far below any
OFDM employing M 2 −QAM modulation scheme on all sub- clipping point or breakdown rating. At higher currents,
carriers is calculated from the LED efficiency only gradually reduces (LED droop).
Rp fp This justifies a single-sided (non-negative) clipping
= · log2 M 2 , (36) model [5], [11], [17], [31]. Similarly, many practi-
f0 f0
cal electronic drivers do not allow a negative current
which reduces to (16). In conclusion, for the same NPB through the LED.
γ , thus not yet considering the bias penalty on a pre- • Droop: Above their most efficient point, the LED
emphasized channel, both PAM and pre-emphasized OFDM becomes somewhat less efficient. This ‘droop’ leads to
schemes demand the same optimum modulation bandwidth invertible second–order distortion, inherent to non-linear
and provide identical throughput and, therefore, the modu- photon generation rates [38]–[40].
lation bandwidth and throughput plots of Figure. 1 are also
In this article we focus on the second model, but we also
applicable for DCO-OFDM employing M 2 -QAM.
discuss the consequences of droop, as in the third model. In
Figure. 1 also includes the required modulation bandwidth
OFDM, the LED AC current, iled (t), has in good approxima-
and the throughput for pre-emphasized OFDM (blue lines)
tion a Gaussian probability density. It has rms modulation
and for waterfilling (black lines) with continuous modulation
depth σmod . To ensure that the signal remains in the lin-
order M at BER = 10−4 . As expected, waterfilling pro-
ear region, a DC bias ILED is needed for the LED. Further,
vides the maximum throughput. Pre-emphasis comes with a
the LED imposes a low-pass nature, but studying memory
penalty in throughput, which increases with NPB but is small
effects in distortion is beyond the scope of this article.
for low NPB. However, pre-emphasis requires less band-
width. This can reduce the sampling rate, hence it consumes
less power in analog-to-digital conversion and in digital sig- A. CURRENT CLIPPING
nal processing. Furthermore, pre-emphasis avoids the need The choice of z (defined in (17)) needs to ensure that the clip-
to exchange the bit loading profile, thus it reduces signalling ping noise stays below the maximum tolerable noise floor.
overhead. From arguments in [5], [17], [31], we conclude that modern
In Figure. 1(b), we see a small artefact due to simplifying LEDs clip negative currents but are not peak limited in their
: OFDM with discrete M (red line) cannot outperform operational range. The clipping noise per sample is zero
OFDM with continuous M (blue line). This artefact is small. if the signal iled (t) ≥ −zσmod (or ILED ≥ 0) and equal to
Comparing the maximum normalized modulation bandwidth, iled + zσmod otherwise. Using a Gaussian pdf for ILED with
continuous M does not show any jump in the optimized mean value zσmod and variance σmod 2 and integrating over
modulation bandwidth, which was also observed in [5]. ξ = iLED − zσmod , the i-th moment of the clipping is
 −zσ  
Fixing the bandwidth means operating on a point on a (ξ + zσmod )i ξ2
horizontal line in Figure. 1(a). For operational points on μi = √ exp − 2 dξ. (37)
this line, the link collapses if it is above the curves of the −∞ 2π σmod 2σmod
calculated maximum supportable fmax . As an example, if a The effective noise variance of the distortion is σD2 = μ2 −μ21
system with an LED of f0 = 10 MHz fixes the transmit and is calculated as
bandwidth to 40 MHz, it operates on the horizontal line of  
a normalized modulation bandwidth of 4. Below an NPB of σD2
= z 2
+ 1 Q(z) − zg(z) − (g(z) − zQ(z))2 , (38)
about 25 dB, it uses a bandwidth broader than what PAM or σmod
2

pre-emphasized OFDM can support (the point of operation where Q(.) and g(.) are the tail distribution function and pdf
is above the plotted curves). Nonetheless, a well-performing of the standard normal distribution, respectively. For ease of
link would be feasible if the system were allowed to scale notation, we introduce cz = σD /σmod .
back the bandwidth, rather than to aggressively push symbol Clipping also attenuates the signal, particularly if z < 2.
rates beyond the 3 dB LED bandwidth. Below z = 1, where the signal level is multiplied by az =
0.84 [17], the effect becomes pronounced. While we refer
V. CLIPPING AND DISTORTION MODEL the reader to [17] for expressions that relate z and az , we
The modelling of clipping and distortion is subject to use az in following throughput equations.
improving insights [31]. In the following we discuss three We argue that the clipping spectrum is limited to fx and
models does not significantly spill over to empty sub-carriers far
• Double sided clipping: In the early days, LEDs had above fx : A signal spectrum limited to fx , creates time–
to be designed for maximum power output. Above a domain signals that are highly correlated in a period fx−1 /2.
certain current level, the LED would thermally break Every clipping event causes an error signal that has a typical
down. This justified a model in which the LED current duration of about fx−1 /2. By virtue of properties of Fourier
is both non-negative and peak-limited [18]. Transforms and as we confirm by simulation, this leads to
• Clipping of the current: Today’s LEDs are operated at a clipping noise spectrum that is mainly restricted to (0, fx ).
a set point where the photon output per recombining Oversampling, and using an oversized FFT with broader

VOLUME 1, 2020 1729

MARDANIKORANI et al.: OPTIMIZATION AND COMPARISON OF M-PAM AND OPTICAL OFDM MODULATION FOR OPTICAL WIRELESS COMMUNICATION

FIGURE 2. PSD of an OFDM signal (black) and clipping noise (gray), for bias ratio
of 0.5, 1 and 2. LED low-pass response not included. FIGURE 3. Bias ratio z versus number of bits b = log2 M per sub-carrier in one
dimension. Noise-free (r = 1) and leaving a 3 and 6 dB power margin (r = 2 and r = 4,
respectively) to operate over a noisy channel. Solid line: clipping limit. Dashed line:
invertible distortion limit. For distortion–limited z, we used r = 1.
bandwidth (fs fmax ) sees clipping artefacts that span
multiple time samples, but oversampling does not increase
their bandwidth. Multiple independent clipping events add where σ2D 2 is the variance of the second–order distortion

incoherently on a particular victim sub-carrier. Here, we and we used that, for a Gaussian distribution, E{i4led } =
refine the clipping noise model of [12], [17] that considers 3(E{i2led })2 = 3σmod2 and inserted z2 = I2LED /σmod 2 . Based on

low-pass filtering of flat (spectrally white) clipping artefacts our observations, the second–order distortion is the dominant
in the LED. Figure. 2 shows the PSD of 64-QAM (M = 8) distortion in LEDs for z > 2, hence we can neglect the term
on the 64 lower sub-carriers in an OFDM system with 128 α3 and the distortion caused by the third order non-linearity.
sub-carriers thus with an IFFT size of 256. The PSD of The distortion i2led is uncorrelated with the LED modulation
the clipping noise is shown in Figure. 2 for z = 0.5 (overly current iled , i.e., E{i2led · iled } = 0 . Its spectrum, N2D ( f ) can
aggressive clipping), z = 1 and z = 2. This plot confirms our be calculated by the convolution of the modulation spectrum
argument that the clipping noise is mostly confined within of iled by itself.
the modulation bandwidth of the signal where it may have
two or three dB variations. Also, the clipping PSD raises VI. EFFECT OF CLIPPING AND DISTORTION ON OFDM
with lowering z. For the signal in Figure. 2, z ≥ 2.2 is In the following, we discuss two different approaches to
required to achieve a simulated BER of < 10−4 . handle the clipping noise.
As clipping noise raises the noise floor, we model
N0 − → N0 + ND ( f ). We approximate the simulated clip-
A. CONSERVATIVELY CHOOSING LOW
ping spectra by a rectangular function within the modulation
MODULATION DEPTH
bandwidth fx :
A pragmatic (but not optimum) approach is to ensure the
σ2 c2 σ 2 |H( f )|2 clipping noise spectrum falls below the receiver noise level.
ND ( f ) ≈ D |H( f )|2 = z mod . (39) This can be translated into a requirement on the Signal-to-
fx fx
Distortion Ratio (SDR),
B. INVERTIBLE DISTORTION MODEL 2N a2
The hard clipping model of the LED needs refinement as SDR = = 2z ≥ 2rX(M), (41)
ND cz
other (invertible) non-linearities may dominate for high z.
Electrons and holes recombine at a rate governed by the for all f , where r is a design (power) margin. This, with (38)
ABC formula [38]–[40]. For a brief discussion here, we gives the maximum modulation order M that can be used
simplify the dynamic model [30], [31], [39] by describing for a given z. Thus, for a target modulation order M (for
the light output φ as a function of LED current, M 2 -QAM), it specifies the minimum required LED bias.
Figure. 3 shows the minimum z as a function of number of
φ = α1 ILED + α2 ILED
2
+ α3 ILED
3
. bits per sub-carrier in one dimension for margins r = 1, 2
Modulating with ILED = ILED +iled , the signal φ sees second– and 4. It can be seen that for a typical modulation order
order distortion with a relative strength of 64-QAM (M = 8), z ≥ 2.15 (compared to the simulated
  z ≥ 2.2 in Section V) and z ≥ 2.4 are needed for r = 1 and
σ2D2 (α2 + 3α3 ILED )2 E i4led r = 2, respectively.
=
2  
σmod
2
α1 + 2α2 ILED + 3α3 I2 LED E i2led The optimum modulation bandwidth and the throughput
 2 follow from (35) and (36), if the distortion can be assumed
α2 3α3
α1 + α1 ILED I LED
2
3 to be negligible compared to receiver noise. This requires the
= 2 2 (40) modulation depth and constellation size to satisfy (41) for
z 3α3 2
1 + 2α 2
I
α1 LED + I
α1 LED the given z with an adequate margin factor r ≥ 1. However,

1730 VOLUME 1, 2020

choosing the distortion power just below the noise level can achieve a normalized throughput of
(r = 1) may not be adequate, as the distortion raises the ⎛ ⎞
 fp /f0
noise level by 3 dB. Since the distortion also has a low–pass Rp 1 γ ln 2 a2z
= log2 ⎝1 + · ⎠dx. (44)
spectrum response, this affects mainly the lower sub-carriers. f0 0 2fp /f0 − 1 1 + c2z γ 2−x
fp /f0
Nonetheless, to avoid that clipping affects the BER at any
sub-carrier, a margin r ≥ 1 is needed. The above integral has a closed form solution,
Considering a channel limited by second–order distortion, Rp Rp (z → ∞) 1
thus clipping– and noise–free channels, (41) can be written = +
f0 f0 (ln 2)2
as     
−c2zn γ −f /f −c2zn γ 1
σmod
2 × Li2 2 p 0 − Li2 −
SDR = ≥ 2rX(M). fp /f0 fp /f0 (ln 2)2
σ2D
2     
−c2z γ −fp /f0 −c2z γ
× Li2 2 − Li2 , (45)
Dashed lines in Figure. 3 also show the minimum required z fp /f0 fp /f0
for margin r = 1 for two values of α2 /α1 when ILED = 0.3
where Rp (z → ∞) is the throughput for the case of no
A. It can be seen that for modulation order of M ≤ 16,
clipping noise, given in (33), Li2 (.) is the Spence function
thus 256 QAM, the minimum z (for this specific exam-
defined as
ple) is dominated by the clipping noise and the distortion
z
is negligible. Values in the range of a Signal–to–Distortion–  ln(1 − u)
and–Noise Ratio (SNDR) around 40 dB are achieved in Li2 (z) = du, (46)
−u
commercial ITU G.9991 systems, allowing up to 1024-QAM 0
(M = 32), or 4096-QAM (M = 64) at maximum. The and
steep dashed curves confirm the practical experience that cz
modulation orders above M = 64 are hard to achieve at czn =  . (47)
a2z γ ln 2
reasonable z. In future systems, the distortion may be over- 1+ 2fp /f0 −1
come by a pre or post-distortion compensation method, such
as in [8]. Therefore, we do not elaborate on invertible distor- The optimum modulation bandwidth for pre-emphasis,
tion as limiting the throughput, so we focus on non-invertible fmaxp , is normalized to f0 to create versatile, generically appli-
clipping. cable curves in Figure. 4(a) with blue lines for z → ∞ (solid
blue) and z = 2.5 (dashed blue). The corresponding normal-
ized rates are shown in Figure. 4(b). Reducing z from ∞
B. OPTIMIZING FOR THROUGHPUT (thus allowing unbounded biasing power) to 2.5 at a constant
In this section, we include the clipping distortion power γ , increases the optimum modulation bandwidth to leverage
in our optimization of the modulation bandwidth and the the better SNDR at higher frequencies. Nevertheless, the
throughput. Recalling (23), the received QAM symbol energy throughput is experiencing a considerable penalty, which is
to noise plus distortion ratio (SNDR) is, using (39), about 25% for a γ of 50 dB. Increasing z from 2.5 to 3
reduces the penalty to about 10%.
2N a2z Sx ( f )|H( f )|2 More of practical use is a discrete constellation size M for
SNDROFDM = = . (42)
N0 + ND ( f ) c2 σ 2
N0 + z fxmod |H( f )|2 M 2 -QAM modulation. The calculated SNDR is an increasing
function of frequency, that is, the minimum SNDR occurs
where x stands for pre-emphasis (p) or waterfilling (w). at low frequencies. For the communication link to use the
same constellation on all sub-carriers with the target BER,
1) THROUGHPUT OF PRE-EMPHASIS WITH DISTORTION the choice of z needs to ensure that the required X(M) can
be satisfied at low frequencies,
Inserting Sp ( f ) and κ from (8) and (9), respectively,
N γ ln 2 a2z
γ ln 2 a2z ≥ X(M) → f /f · ≥ 2X(M).
SNDROFDM = · . (43) N0 + ND (0) 2 p 0 − 1 1 + c2z γ
2fp /f0 − 1 1 + c2z γ 2−f /f0 fp /f0
fp /f0 (48)
For z → ∞, cz → 0, az → 1, and (43) reduces to (34) For a given γ and modulation order M, the optimum mod-
which was derived for clipping–free modulation. The above ulation bandwidth is the maximum fp that satisfies (48).
equations (42) and (43) are based on the effective energy Unfortunately, a closed form expression for the optimum
emitted per symbol, thus do not reflect that with increasing bandwidth cannot be derived. In the limiting case of
z, more bias power is needed to achieve N . clipping–free communication, our result reduces to (35). For
For a continuous modulation order M, one can replace a given γ and a given choice of z, the modulation bandwidth
the SNDR into (29) and solve the integral numerically for fp is optimized from (48) as a function of M, so the through-
different fp choices to optimize fp for a given z. Pre-emphasis put follows from (36). The optimum modulation order M is

VOLUME 1, 2020 1731

MARDANIKORANI et al.: OPTIMIZATION AND COMPARISON OF M-PAM AND OPTICAL OFDM MODULATION FOR OPTICAL WIRELESS COMMUNICATION

invertible distortion, the modulation bandwidth converges to

a constant. Having γ → ∞ in (48),
(ln 2)a2z fp /f0
≥ 2X(M), (49)
c2z (2fp /f0 − 1)
shows that fp /f0 only depends on M, irrespective of the NPB
γ . The throughput in (36), which only depends on M and
the modulation bandwidth, is also approaching to a constant
at large γ values.

2) WATERFILLING WITH DISTORTION

In [5], it is shown that the presence of clipping noise does
not affect the modulation bandwidth fmaxw that optimizes
the throughput. Hence the modulation bandwidth versus
NPB (31) also holds when there is clipping noise provided
that the signal power is corrected for the attenuation fac-
tor a2z . Based on equations in this article, we quantify the
throughput penalty due to distortion as
 
FIGURE 4. Normalized maximum modulation bandwidth and throughput for (a,b) Rw Rw (z → ∞) fmaxw c2z γ −fmaxw /f0
= + log2 1 + 2
waterfilling (W.F.), pre-emphasis (P.E.) with continuous (Cont.) and discrete (Disc.)
f0 f0 f0 fmaxw /f0
modulation order M and (c,d) for pre-emphasis with discrete M, and different values of
z as a function of γ . The BER is fixed at 10−4 .
    
1 −c2z γ −fmax /f0 −c2z γ
+ Li2 2 w − Li2 .
(ln 2)2 fmaxw /f0 fmaxw /f0
the one that gives the highest throughput and the correspond- (50)
ing normalized modulation bandwidth is the optimum, fmaxp .
The throughput and the associated optimum modulation
The throughput and fmaxp are shown in Figures. 4(a) and (b)
bandwidth are shown in Figures. 4(a) and (b) as a function
with red lines for z = 2.5 (dashed-red) and z → ∞ (solid
of the NPB γ . Waterfilling provides a better performance
red) and in Figures. 4(c) and (d) for different z values.
compared to pre-emphasis but uses a larger modulation band-
The difference between the continuous and discrete con-
width, for both clipping–free and clipped communication.
stellation size M was discussed in Section IV for distortion–
Choosing z = 2.5 reduces the throughput of waterfilling
free modulation (z → ∞). Considering distortion with
approach by a gap that increases with γ and that is about
z = 2.5, as in Figure. 4(b), shows a considerable cut in
18% at γ of 50 dB compared to z → ∞.
throughput when using a discrete M, compared to a non-
practical non-integer modulation order M. The throughput
shows a reduction of about 40% at γ of 50 dB (see dashed VII. COMPARISON OF DCO-PAM AND DCO-OFDM
red and dashed-blue lines). We compare the two modulation schemes, DCO-PAM and
We see a very substantial throughput penalty if one has to DCO-OFDM, for different power constraints at the trans-
stick to discrete constellations M that are a power of 2, which mitter side. For various power constraints at the transmitter,
is understood from the discussions that led to (48). In fact, we calculate the portion of the power that contributes to
while pre-emphasis equalizes the SNR (derived from (43) for the throughput versus biasing power. We redefine the NPB
cz = 0), it does not generically equalize the SNDR, which, parameter that allows for a fair comparison of the schemes,
tends to be worse at lower frequencies. considering that a particular σmod leads to different consumed
To mitigate this gap while still using a common equal con- powers.
stellation M, the transmitter can adjust (lower) the power for
the sub-carriers at higher frequencies with a better SNDR. A. EXTRA–POWER LIMITED CHANNEL
This approach, however, requires an adaptive power loading For VLC links, the illumination power is available already
algorithm which increases the complexity. Another approach (β1 = 0) and only the extra power which is needed for
to recover the throughput of discrete M modulation scheme modulation is of interest. Extra power was shown to be
(compared to the theoretical dashed-blue line of Figure. 4(b)) directly related to the LED current variance through the
is to use a higher DC current. Figure. 4(d) shows that increas- factor RLED . The LED resistance RLED consists of two parts,
ing z from 2.5 to 3 can recover a big fraction of the loss; RLED = VT /ILED + Rs , where the dynamic part is an inverse
the penalty of using discrete M compared to continuous M function of ILED and the second part is the constant parasitic
is about 20% and compared to z → ∞ is about 30%. resistance Rs . For LEDs biased at a typical current of ILED =
Figures. 4(c) and (d) show that for a fixed z and large 0.35 A and with VT being 25 mV, the dynamic resistance
γ , thus when distortion dominates over the noise and over becomes approximately 70 m which is negligible compared

1732 VOLUME 1, 2020

to the Rs which is typically in the order of 1 − 2  [33]. As
a result, identical extra power for both modulation schemes
is translated into identical σmod , hence the same NPB for
both schemes. In Section IV and Figure. 1, it was shown
that for the same NPB, thus ignoring biasing power and
taking adequate z (no significant distortion), both schemes
achieve the same throughput in a pre-emphasized channel.
In fact, one may interpret the FFT with Hermitian symmetry,
as used in OFDM, as just a unitary rotation of the PAM time
signals along the time-frequency domains. OFDM preserves
the number of dimensions and the distances in the signal
space, thus in a pre-emphasized channel has equal spectrum
efficiency and BER as PAM.

B. OPTICAL–POWER LIMITED CHANNEL

Optical power limitations can be induced for instance in VLC
where illumination dictates the light level or in IR where eye-
safety needs to be guaranteed. The average optical power of
an LED can be written as [8]
 
Ep FIGURE 5. (a,c) Optimum (normalized) modulation bandwidth and (b,d) throughput
Popt = ILED , (51) versus optical NPB γopt for PAM (red lines) and OFDM (with pre-emphasis and with
q waterfilling) with different z choices. For all plots BER = 10−4 .

where
Ep  is the average energy of the photons transmitted
1) HIGH NORMALIZED POWER BUDGETS
by the LED and q is the unit electron charge. According
to (51), constraining the average optical power is equivalent As an example, for an LED with f0 = 10 MHz bandwidth,
to constraining the LED DC current via β1 (β2 = 0, β3 = 0). to reach a throughput near a gigabit (Rp /f0 = 100), z = 4
As we compare DCO-PAM and DCO-OFDM for the same is needed, but that significantly jeopardizes the through-
LED DC current, their variances differ. The variance σmod 2 put for more distant receivers (with lower available NPBs)
is related to ILED via z in (17). To reflect this, we use γopt where z < 3 needs to guarantee range. In another exam-
as a variant of γ that addresses the optical power limit: ple, to provide a throughput of 60f0 , DCO-PAM requires
an about 2.5 dB lower NPB compared to pre-emphasized
q2 P2opt H 2 DCO-OFDM while z = 4 is used for OFDM. Keeping the
γopt =  2 · 0 . (52) bias ratio of OFDM at z = 4, at a lower throughput of 10f0 ,
Ep N0 f0
the NPB difference between DCO-PAM and pre-emphasized
Then from (51), (52) and using the definition of z, the optical DCO-OFDM increases to about 5 dB while a lower z, e.g.,
NPB relates to γ via z = 2.5 shows only 1.6 dB NPB difference. We acknowl-
edge that if pulse shaping of PAM is needed, the advantage
γopt = z2 γ . (53) shrinks, as z rises.
Interestingly, DCO-PAM also outperforms DCO-OFDM
DCO-PAM has a lower PAPR, thus allows a smaller z than
with waterfilling at low optical NPBs. Waterfilling performs
DCO-OFDM, hence gets a better γ for the same γopt . This
better when the optical NPB increases, say γopt above 32 dB
implies a horizontal shift that differs per modulation set-
for z = 2.5 (equivalent to γ more than 24 dB) and above
ting. This changes the cross-over points for the choice of
50 dB for z = 4 (equivalent to a NPB γ of more than
modulation that performs best for a given NPB. Using (17)
38 dB2 ). The cross-over point for waterfilled DCO-OFDM
for M-PAM with M = 4, 8, 16 and 32, 1/z2 is equivalent to
to outperform PAM moves to higher NPBs when a higher z is
horizontal shifts of 2.55, 3.68, 4.23, and 4.5 dB, respectively.
selected. However, at large NPBs of 50 dB, the theoretically
For OFDM, the bias ratio z is subject to optimization.
optimum modulation bandwidth for DCO-PAM is around
We see in Figure. 5 that for pre-emphasized OFDM with
7f0 . In practice, these large bandwidth extensions impose
a fixed z the throughput converges to a constant for large
difficulties in the implementation.
γopt , thus when clipping dominates over the noise floor. On
the other hand for small γopt , when distortion is negligible, 2) LOW NORMALIZED POWER BUDGETS
increasing z just leads to a reduction in the received SNR.
At low NPBs, it may be attractive to use dedicated non-
Hence, at low γopt , the throughput curves of pre-emphasized
negative OFDM variants, such as ACO-OFDM or Flip
DCO-OFDM are horizontally shifted copies of each other;
the distance between the curves for z = 2.5 and z = 4 is 2. which is equivalent to a received electrical SNR of 23 dB in the
significant: 4 dB. modulation bandwidth of fmaxw ≈ 7.5f0 [5].

VOLUME 1, 2020 1733

MARDANIKORANI et al.: OPTIMIZATION AND COMPARISON OF M-PAM AND OPTICAL OFDM MODULATION FOR OPTICAL WIRELESS COMMUNICATION

TABLE 3. Quick comparison of the linear and quadratic terms in the power
consumption and a penalty on the SNR for flat channel (low NPB, small modulation For PAM, however, z can be as low as 1 (for M = 2) and
bandwidth).
the approximation β2 σmod
2 ≈ 0 results in about 10% error
(0.46 dB) in the total electrical power. The total electrical
power can reasonably be approximated by the LED DC
power consumption:

Ptot ≈ VLED ILED = VLED zσmod . (54)

OFDM, to avoid the power losses in the DC bias. Flip To acknowledge that z2 σmod
2 rather than σmod
2 itself is con-
OFDM carries the signal with variance σmod 2 , however, sam- strained, let us compare systems for the total NPB γtot
ples with positive polarity are transmitted in a first block, including bias losses as
negative samples are transmitted in flipped polarity in a
H02
second block. This ensures that a signal sample is always γtot = z2 σmod
2
· . (55)
transmitted, thus it retains σmod
2 , but the transmission time N0 f0
doubles. During reception, two blocked are folded back into This γtot = z2 γ is identical to the definition of (52). In this
one block to recover the full signal. It has been noticed [19], case, the curves of Figure. 5 also apply to electrical-power
[41]–[43], that this operation collects noise from two blocks, limited channel if the x-axis is read as γtot axis. Alternatively,
thus reduces the SNR by one half. This, to a large extent, it can be shown that, the electrical power model by [17],
defeats the gain obtained from trying to avoid the DC-bias. taking β2 = β3 and β1 = 0 would lead to γtot = (z2 + 1)γ
At high NPBs, these non-negative OFDM variants are which we do not consider in this work.
outperformed by DCO-OFDM, also because at high SNR, a
spectrum efficiency loss is incurred in Flip-OFDM by trans-
1) DESIGN CHOICE FOR z
mitting a second block: This demands higher constellations
to squeeze more bits into fewer dimensions [19]. At low At constant total power, lowering z boosts the signal σmod2 ,

NPBs, where LED bandwidth is adequate to carry a low- thus enhances N and γ , but it also increases distortion.
rate signal, the lower mean value of Flip-OFDM appears For example, systems optimized for large coverage spread
beneficial [19]. The signal in the collapsed block has an their optical power over a large area, thus often have to
effective symbol energy jointly equal to σmod 2 but is pro- operate with relatively small γtot , say of about 30 dB. Then
cessed over a single block time. The mean value of the z = 2.5 is more attractive than z = 4. The latter can improve
√ the throughput for short range or for systems with nar-
signal is 2/π σmod ≈ 0.80σmod . Table 3 lists the resulting
linear and quadratic factors in the power consumption (3). row beams by 65% (from 4.2f0 to 6.9f0 ) and 50% (from
For optical–power limited channels, we take β2 = 0. Flip R = 4.2f0 to 6.2f0 ) improvement for waterfilling and pre-
OFDM3 provides the maximum available σmod within a con- emphasis, respectively. The point where higher z (e.g., z = 4
strained β1 . Despite the 50% drop in the SNR of ACO/Flip to avoid clipping) preforms better than boosting the signal
OFDM, these appear to be slightly more attractive than strength (say, z = 2.5) is around a γtot of 46 dB for pre-
PAM for large M: The FFT shapes the almost uniform 2D emphasis and of 62 dB for waterfilling. For a high speed
PAM signal probability density into a one-sided Gaussian, link (several hundreds of Mbit/sec or several Gbit/sec) with
which appears to be beneficial. However, large modulation an LED with a typical 3 dB bandwidth of f0 ≈ 10 MHz, a
order are not suitable for weak links, which demand small large γtot (e.g., more than 70 dB) is needed. In this range, a
M. For small and moderate M, straight PAM appears bet- large fraction of the electrical power is burnt in DC biasing
ter than ACO-OFDM. From Figure. 3, we further see that to limit the distortion. From Figures. 5(b) and (d), we learn
DCO-OFDM performs comparably; by choosing a very low that a z above 4 will be required to achieve a transmission
z. It severely clips, but 4-QAM (M = 2) DCO-OFDM is rate of more than 80f0 . Moreover, mitigating second-order
nonetheless feasible. distortion also becomes critical (see Figure. 3).

C. ELECTRICAL–POWER LIMITED CHANNEL 2) A TYPICAL EXAMPLE

Often, the total electrical power (3) consumed is relevant. Consider an OWC system limited by total power, with the
For a bias-T modulator, β1 = V0 and β3 ≈ Rs dominate (3), channel frequency response given in (4). At 1m distance,
while β2 σmod
2 is much smaller. In fact, for a typical LED a gain-to-noise ratio of 70 dB in a 1 MHz sub-carrier
bias current of ILED = 0.35 A, V0 = 2.5 V and Rs = 1 , bandwidth requires
in the total electrical power equation (3), β1 ILED = 0.875,  
β2 σmod
2 = β2 I2LED /z2 = 0.1225/z2 and β3 I2LED =0.1225. For H02 −19
= 107
→ N0 ≈ 10 V 2
/Hz .
OFDM, typically z > 2, hence the term β2 σmod2 is negligible. N0 × 106
A 450 nm LXML-PB02-0023 blue LED was measured. It has
3. Alternatively one many argue that clipping halves the ACO-OFDM
power by 50%. So, to compare for the same signal power, we boost the a 3-dB bandwidth around f0 ≈ 10 MHz at ILED = 350 mA
2 .
ACO power by a factor of two, and also arrive at full σmod bias current [5] with V0 ≈ 2.5 V. Since the dominant term in

1734 VOLUME 1, 2020

 to be compensated. One can repair ISI at the receiver more
effectively by using a DFE equalizer [16]. The latter can
simultaneously handle channel multipath, if it occurs, and
avoids too large noise enhancements, but at the cost of a
complex Viterbi algorithm. Also frequency–domain, equal-
izers have been proposed, that place both an FFT and IFFT
at the receiver. However, one may argue that the complexity
of FFTs typically is small compared to other signal process-
ing, such that the use of an FFT is not prohibitive. Possibly,
the complexity of the signalling protocol, its over-head, and
FIGURE 6. Throughput versus distance for PAM (red), pre-emphasized OFDM with the number of memory operations in an OFDM system can
z = 2.5 (dashed-blue) and z = 4 (solid-blue) and waterfilling with z = 2.5 be of concern. In this respect, waterfilling or uniform power
(dashed-black) and z = 4 (solid-black). The total electrical power is limited to 1 W.
loading may be less attractive as it places a different mod-
ulation order per sub-carrier, which needs to be negotiated
the total power consumption equation (3) is the DC power, between receiver and transmitter.
from (1) we have
VLED = 2.5V + (1) × (0.35A) = 2.85V, IX. CONCLUSION
The two popular OWC modulation schemes, namely
and
Orthogonal Frequency Division Multiplexing (OFDM) and
Ptot ≈ PDC = (2.85V) × (0.35A) ≈ 1W. Pulse Amplitude Modulation (PAM) were compared for use
in an IM/DD system using LEDs, considering the minimally
The total NPB is calculated from (54) and (55) to be 1.6 × required DC biasing to ensure the non-negativity of driving
105 , thus approximately 52 dB. For the 52 dB of γtot , the LED current. To cope with the LED channel response, two
throughput can be found in Figures. 5(b) and (d) for DCO- well-known OFDM power loading strategies were discussed,
PAM and DCO-OFDM using waterfilling or pre-emphasis namely, waterfilling and the correction of the attenuation of
strategies. Figure. 6 shows the throughput versus the distance higher frequencies by a pre-emphasis.
between the transmitter and the receiver for different z values. We derived mathematical expressions for the throughput
To include the impact of distance d, we used the 4th power and the optimum modulation bandwidth to be used. Using
law (“40 log d”) path loss model of [44]. With γtot,dB = a suitable Normalized Power Budget (NPB) definition and
10 log10 (γtot ), a normalization to the LED 3 dB bandwidth, generic results
  could be derived. It was shown that for the same extra mod-
d
γtot,dB (d) = 52 − 40 log10 ulation power, which is a suitable metric for VLC where the
1m
DC bias is already available for illumination, pre-emphasized
to ensure that at 1 m distance, γtot is 52 dB. Several relevant OFDM and PAM at a reduced modulation depth showed no
observations can be made. Waterfilling marginally outper- difference in throughput and in required modulation band-
forms DCO-PAM at distances below 1 m, while operating width. Waterfilling, which is the optimum power allocation
beyond 3 m, DCO-PAM provides the better performance. strategy, outperforms pre-emphasized systems, but occupies
At a close distance (below 1 m for waterfilling and below a larger required bandwidth.
1.5 m for pre-emphasis), the received signal is sufficiently The conclusions and optimally recommended choices,
strong to focus merely on distortion. Therefore, a large z however, differ for channels that are limited by their opti-
(e.g., z = 4 rather than a small z = 2.5) is required to pro- cal power or by their electrical power. Optical power can
vide the optimum performance. On the other hand, when be confined by limits to the illumination level in VLC or
the distance increases, the receiver noise floor becomes the by eye safety precautions in IR. In IR communication, par-
dominant design concern and the transmitter has to boost ticularly with battery–powered devices, the total available
the modulation depth, thereby compromising z and tolerating electrical power may be limited. Here, the DC bias can
more clipping. be minimized, just to carry the data signal in an undis-
torted manner. OFDM suffers from a large peak-to-average
VIII. COMPUTATIONAL COMPLEXITY ratio. The non-negativity constraint forces the use of an
Another important aspect for the comparison is the com- unattractively large bias. Compromising for a practical bias
putational complexity of modulation at the transmitter and current for OFDM, peaks in the current have to be clipped
detection in the receiver. The complexity in the OFDM trans- before being applied to the LED. We quantified and mod-
mit Inverse Fast Fourier Transform (IFFT) and in the receive eled the resulting distortion and its impact on performance,
(FFT) of size N is in the order of 4N log2 N per block. For which allows for an optimization of the modulation depth
PAM, the use of simple pre-emphasis eliminates the need depending on, for instance, NPB. In this article, we gener-
for equalization if only the low-pass LED response needs alized derivations for OFDM, both for waterfilling and for

VOLUME 1, 2020 1735

MARDANIKORANI et al.: OPTIMIZATION AND COMPARISON OF M-PAM AND OPTICAL OFDM MODULATION FOR OPTICAL WIRELESS COMMUNICATION

pre-emphasis, by including the clipping noise in the through- [16] D. J. F. Barros and J. M. Kahn, “Comparison of orthogonal frequency-
put and bandwidth optimization. We showed that for an division multiplexing and on-off keying in amplified direct-detection
single-mode fiber systems,” J. Lightw. Technol., vol. 28, no. 12,
IR channel, more precisely, for optical–power limited chan- pp. 1811–1820, Jun. 2010.
nels, under moderate modulation bandwidth, M-PAM with [17] X. Ling, J. Wang, X. Liang, Z. Ding, and C. Zhao, “Offset and
a linear high-boost filter is able to provide a higher data power optimization for DCO-OFDM in visible light communication
systems,” IEEE Trans. Signal Process., vol. 64, no. 2, pp. 349–363,
transmission rate than any sub-carrier loading scheme, opti- Jan. 2016.
mized for DCO-OFDM. When a large NPB is available, [18] S. Dimitrov and H. Haas, “Information rate of OFDM-based optical
OFDM preferably with bit loading that follows waterfilling wireless communication systems with nonlinear distortion,” J. Lightw.
Technol., vol. 31, no. 6, pp. 918–929, Mar. 2013.
principles outperforms M-PAM. [19] X. Deng, S. Mardanikorani, G. Zhou, and J.-P. M. G. Linnartz, “DC-
The best LED bias setting depends on the NPB. Moreover, bias for optical OFDM in visible light communications,” IEEE Access,
the cross point for the NPB at which waterfilling DCO- vol. 7, pp. 98319–98330, 2019.
OFDM starts to outperform PAM moves towards higher NPB [20] R. G. Gallager, Information Theory and Reliable Communication. New
York, NY, USA: Wiley, 1968.
values when a higher bias current of the LED is selected. [21] D. Hughes-Hartogs, “Ensemble modem structure for imperfect trans-
Therefore, an OFDM system with a fixed LED bias current mission media,” U.S. Patent 4 679 227, Jul. 1987.
which is designed to operate for a range of NPBs might [22] D. Hughes-Hartogs, “The capacity of a degraded spectral Gaussian
broadcast channel,” Ph.D. dissertation, Inf. Syst. Lab., Center Syst.
underperform compared to PAM, if OFDM is optimized for Res., Stanford Univ., Stanford, CA, USA, Jul. 1995.
low NPB range or for large coverage. Preferably, an adaptive [23] J. A. C. Bingham, “Multicarrier modulation for data transmission: An
setting of the LED bias current, optimized for the NPB is idea whose time has come,” IEEE Commun. Mag., vol. 28, no. 5,
pp. 5–14, May 1990.
used to yield the highest DCO-OFDM throughput. [24] B. Cardiff, M. F. Flanagan, F. Smyth, L. P. Barry, and A.
D. Fagan, “On Bit and Power Loading for OFDM Over SI-
POF,” J. Lightw. Technol., vol. 29, no. 10, pp. 1547–1554, May,
2011.
REFERENCES
[25] L. Goldfeld, V. Lyandres, and D. Wulich, “Minimum BER power
[1] “Visible light communication (VLC)—A potential solution to the loading for OFDM in fading channel,” IEEE Trans. Commun.,
global wireless spectrum shortage,” GBI Research, London, U.K., Rep. vol. 50, no. 11, pp. 1729–1733, Nov. 2002
GBISC017MR, 2011. [Online]. Available: http://www.gbiresearch. [26] “G.vlc: Draft,” Int. Telecommun. Union, Geneva, Switzerland,
com/ Recommendation ITU-T G.9991, Sep. 2018. [Online]. Available:
[2] H. Elgala, R. Mesleh, and H. Haas, “Indoor optical wireless commu- https://www.itu.int/md/T17-SG15-181008-TD-PLEN-0291
nication: Potential and state-of-the-art,” IEEE Commun. Mag., vol. 49, [27] Status of IEEE 802.11 Light Communication TG. Accessed: Jul.
no. 9, pp. 56–62, Sep. 2011. 2020. [Online]. Available: http://www.ieee802.org/11/Reports/tgbb-
[3] A. Jovicic, J. Li, and T. Richardson, “Visible light communication: update.htm
Opportunities, challenges and the path to market,” IEEE Commun. [28] H. L. Minh et al., “100-Mb/s NRZ visible light communications
Mag., vol. 51, no. 12, pp. 26–32, Dec. 2013. using a postequalized white LED,” IEEE Photon. Technol. Lett.,
[4] L. Grobe et al., “High-speed visible light communication systems,” vol. 21, no. 15, pp. 1063–1065, Aug. 2009.
IEEE Commun. Mag., vol. 51, no. 12, pp. 60–66, Dec. 2013. [29] C. Chen, D. A. Basnayaka, and H. Haas, “Downlink performance
[5] S. Mardanikorani, X. Deng, and J.-P. M. G. Linnartz, “Sub-carrier of optical attocell networks,” J. Lightw. Technol., vol. 34, no. 1,
loading strategies for DCO-OFDM LED communication,” IEEE Trans. pp. 137–156, Jan. 2016.
Commun., vol. 68, no. 2, pp. 1101–1117, Feb. 2020. [30] S. Mardani, A. Alexeev, and J.-P. Linnartz, “Modeling and com-
[6] D. Tsonev, S. Sinanovic, and H. Haas, “Complete modeling of non- pensating dynamic nonlinearities in LED photon-emission rates to
linear distortion in OFDM-based optical wireless communication,” J. enhance OWC,” in Proc. SPIE Light-Emitting Devices Materials
Lightwave Tech., vol. 31, no. 18, pp. 3064–3076, Sep. 2013. Appl., vol. 10940. San Francisco, CA, USA, Mar. 2019. [Online].
[7] K. Ying, Z. Yu, R. J. Baxley, H. Qian, G.-K. Chang, and G. T. Zhou, Available: https://doi.org/10.1117/12.2511099
“Nonlinear distortion mitigation in visible light communications,” [31] J. P. Linnartz, X. Deng, A. Alexeev, and S. Mardani, “Wireless
IEEE Wireless Commun., vol. 22, no. 2, pp. 36–45, Apr. 2015. communication over an LED channel,” IEEE Commun. Mag., to be
[8] X. Deng et al., “Mitigating LED nonlinearity to enhance visi- published, 2020.
ble light communications,” IEEE Trans. Commun., vol. 66, no. 11, [32] X. Deng, K. Arulandu, Y. Wu, S. Mardanikorani, G. Zhou,
pp. 5593–5607, Nov. 2018. and J.-P. M. G. Linnartz, “Modeling and analysis of trans-
[9] J. Armstrong, “OFDM for Optical Communications,” J. Lightw. mitter performance in visible light communications,” IEEE
Technol., vol. 27, no. 3, pp. 189–204, Feb. 2009. Trans. Veh. Technol., vol. 68, no. 3, pp. 2316–2331, Mar.
[10] S. C. J. Lee, “Discrete multitone modulation for short-range optical 2019.
communications,” Ph.D. dissertation, Dept. Electr. Eng., Technische [33] “Advance electrical design LED model,” Lumileds, San Jose, CA,
Universiteit Eindhoven, Eindhoven, The Netherlands, 2009. USA, Rep. AB20-3A, 2002.
[11] D. J. F. Barros, S. K. Wilson, and J. M. Kahn, “Comparison of orthog- [34] K. Arulandu, J.-P. M. G. Linnartz, and X. Deng, “Enhanced visible
onal frequency-division multiplexing and pulse-amplitude modulation light communication modulator with dual feedback control,” IEEE
in indoor optical wireless links,” IEEE Trans. Commun., vol. 60, no. 1, Trans. Emerg. Sel. Topics Power Electron., early access, Dec. 30,
pp. 153–163, Jan. 2012. 2019, doi: 10.1109/JESTPE.2019.2962999.
[12] J. Lian, M. Noshad, and M. Brandt-Pearce, “Comparison of optical [35] S. Yoo, D. Yun, B. Song, J. Burm, J. Chung, and J. H. Chun, “A
OFDM and M-PAM for LED-based communication systems,” IEEE 10 Gb/s 4-PAM transceiver with adaptive pre-emphasis,” in Proc.
Commun. Lett., vol. 23, no. 3, pp. 430–433, Mar. 2019. ISIC, Singapore, 2011, pp. 258–261.
[13] S. Dimitrov, S. Sinanovic, and H. Haas, “Signal shaping and modula- [36] X. Huang, J. Shi, J. Li, Y. Wang, and N. Chi, “A Gb/s VLC
tion for optical wireless communication,” J. Lightw. Technol., vol. 30, transmission using hardware preequalization circuit,” IEEE Photon.
no. 9, pp. 1319–1328, May 2012. Technol. Lett., vol. 27, no. 18, pp. 1915–1918, Sep. 2015.
[14] Photobiological Safety of Lamps and Lamp Systems, British Standards [37] S. Haykin, Communication Systems, 3rd ed. New York, NY, USA:
BS EN 62471, Sep. 2008. Wiley, 1994.
[15] A. Nuwanpriya, S.-W. Ho, J. A. Zhang, A. J. Grant, and L. Luo, “PAM- [38] J. Piprek, “Efficiency droop in nitride-based light-emitting
SCFDE for optical wireless communications,” J. Lightw. Technol., diodes,” Phys. Status Solidi A, vol. 207, no. 10, pp. 2217–2225,
vol. 33, no. 14, pp. 2938–2949, Jul. 2015. Oct. 2010.

1736 VOLUME 1, 2020

[39] Q. Dai et al., “On the symmetry of efficiency-versus-carrier XIONG DENG (Member, IEEE) received the
concentration curves in GaInN/GaN light-emitting diodes and relation M.Eng. degree in communication and information
to droop-causing mechanisms,” Appl. Phys. Lett., vol. 98, no. 3, engineering from the University of Electronic
2011, Art. no. 033506. Science and Technology of China, in 2013,
[40] H. Zhao, G. Liu, J. Zhang, R. A. Arif, and N. Tansu, “Analysis and the Ph.D. degree in optical wireless com-
of internal quantum efficiency and current injection efficiency in munications from the Eindhoven University of
III-nitride light-emitting diodes,” J. Display Technol., vol. 9, no. 4, Technology, Eindhoven, The Netherlands, in 2018.
pp. 212–225, Apr. 2013 In 2013, he was a Researcher with the Terahertz
[41] N. Fernando, Y. Hong, and E. Viterbo, “Flip-OFDM for optical wire- Science and Technology Research Center, China
less communications,” in Proc. IEEE Trans. Inf. Theory Workshop, Academy of Engineering Physics, where he
Paraty, Brazil, 2011, pp. 5–9. was involved in the integrated terahertz com-
[42] N. Fernando, Y. Hong, and E. Viterbo, “Flip-OFDM for unipolar munication and imaging system. He was a Guest Researcher with
communication systems,” IEEE Trans. Commun., vol. 60, no. 12, Signify (Philips Lighting) Research, where he was involved in light
pp. 3726–3733, Dec. 2012. fidelity. He is currently a Postdoctoral Researcher with the Eindhoven
[43] Z. Yu, R. J. Baxley, and G. T. Zhou, “Achievable data rate analysis University of Technology. His research interests include the system
of clipped FLIP-OFDM in optical wireless communication,” in modeling, digital signal processing, circuits for intelligent lighting, mil-
Proc. IEEE Globecom Workshops, Anaheim, CA, USA, 2012, limeter wave, radio over fiber, and optical wireless communications.
pp. 1203–1207. He serves as a Reviewer for multiple IEEE/OSA journals, including
[44] A. Tsiatmas, C. P. M. J. Baggen, F. M. J. Willems, J.-P. M. G. Linnartz, IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, IEEE JOURNAL
and J. W. M. Bergmans, “An illumination perspective on visible light OF E MERGING AND S ELECTED TOPICS IN P OWER E LECTRONICS , IEEE
communications,” IEEE Commun. Mag., vol. 52, no. 7, pp. 64–71, TRANSACTIONS ON COMPUTERS, IEEE TRANSACTIONS ON VEHICULAR
Jul. 2014. TECHNOLOGY, IEEE JOURNAL OF LIGHTWAVE TECHNOLOGY, IEEE
TRANSACTIONS ON COGNITIVE COMMUNICATIONS AND NETWORKING,
IEEE COMMUNICATIONS LETTERS, IEEE PHOTONICS JOURNAL.

JEAN-PAUL M. G. LINNARTZ (Fellow, IEEE)

received the M.Sc. degree (cum laude) from
the Eindhoven University of Technology (TU/e),
Eindhoven, The Netherlands, in 1986, and
the Ph.D. degree (cum laude) from the Delft
University of Technology (TU Delft), The
Netherlands, in 1991. He was a Senior Director
with the Philips Research, Eindhoven, where he
SHOKOUFEH MARDANIKORANI (Student headed Security, Connectivity, and IC Design
Member, IEEE) received the B.Sc. degree in Research Groups. He initiated research on Coded
electrical engineering from the University of Light, to allow the embedding of identifiers in light
Shahrekord, Shahrekord, Iran, in 2012, and sources, which is now being used in many office and retail facilities. He is
the M.Sc. degree in communication systems currently a part-time Professor with the TU/e, addressing intelligent lighting
engineering from the Sharif University of systems and optical wireless communication, and a Research Fellow with
Technology, Tehran, in 2014. She is currently Signify (Philips Lighting) Research. His inventions led to over 75 granted
pursuing the Ph.D. degree in electrical engineer- patent families and have been a basis for three ventures. His papers have
ing with the Signal Processing Systems Group, been cited more than 11 000 times and his H-index is 53 (GS). From 1992
Eindhoven University of Technology, Eindhoven, to 1995, he was an Assistant Professor with the University of California
The Netherlands. Her main research interests at Berkeley. In 1994, he was an Associate Professor with TU Delft. From
include the application of information theory, system modeling and signal 1988 to 1991, he was an Assistant Professor with the TU Delft. He is a
processing in optical wireless, and visible light communications. Fellow of the IEEE for his leadership in Security with Noisy Data.

VOLUME 1, 2020 1737