196 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 26, NO. 1, JANUARY 1, 2008
Experimental Demonstrations of Electronic
Dispersion Compensation for Long-Haul
Transmission Using Direct-Detection
Optical OFDM
Brendon J. C. Schmidt, Arthur James Lowery, Senior Member, IEEE, and Jean Armstrong, Senior Member, IEEE
Abstract—We present experimental demonstrations using
direct-detection and optical-orthogonal frequency division mul-
tiplexing (DD-OOFDM) for the compensation of chromatic
dispersion in long-haul optical fiber links. Three transmitter
designs of varying electrical and optical complexity are used
for optical single sideband (OSSB) transmission and the theory
behind each design is discussed. The data rates achieved for the
three systems are 10, 12, and 20 Gbit/s for fiber distances between
320 and 400 km. A discussion of system overheads is provided
together with simulations of the required optical signal-to-noise
ratio (OSNR).
Index Terms—Chromatic dispersion, compensation, optical
single sideband (OSSB), orthogonal frequency division multi-
plexing (OFDM).
I. INTRODUCTION
O
RTHOGONAL frequency division multiplexing (OFDM)
has largely replaced serial modulation formats for new
broadband wireless communication systems, because it is a
simple solution to signal dispersion and because it scales well
when data rate increases result in intersymbol interference
that affects multiple symbol periods. Recently there has been
increasing interest in OFDM for optical fiber applications
because as data rates increase, the computational requirements
involved in electronic dispersion compensation (EDC) for
serial modulation formats become impractical [1]. Recent
research has shown that OFDM can be used for the electronic
compensation of chromatic dispersion and polarization mode
dispersion in single-mode optical fiber systems [2]–[5] and for
mode dispersion in multimode systems [6].
One major advantage of the use of EDC rather than disper-
sion-compensation fiber (DCF) to compensate for chromatic
dispersion in long haul optical links is that the amount of com-
pensation can be changed rapidly, simply by adjusting the pa-
rameters of the compensation algorithm; no changes are re-
quired to outside plant. The first installed long-haul EDC sys-
Manuscript received July 30, 2007; revised October 30, 2007. This work was
supported by the Australian Research Council’s Discovery funding scheme (DP
0772937).
The authors are with the Department of Electrical and Computer Systems
Engineering, Monash University, Melbourne, VIC 3800, Australia (e-mail:
Brendon.Schmidt@eng.monash.edu.au; Arthur.Lowery@eng.monash.edu.au;
Jean.Armstrong@eng.monash.edu.au).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/JLT.2007.913017
tems used electronic predistortion (EPD) [7], [8]. EPD operates
at the transmitter, and adjusts the transmitted optical waveform
so that the optical waveform at the receiver is optimized for di-
rect-detection. EPD requires a feedback path from receiver to
transmitter to optimize the transmitted waveform; so it cannot
cope with rapid channel variations and the signal cannot be re-
covered at points between the transmitter and receiver. EDC sys-
tems that use equalization in the receiver overcome these disad-
vantages.
EDC using OFDM uses receiver based equalization and is
scalable. A number of forms of optical OFDM have been pro-
posed recently. Coherent optical OFDM (CO-OFDM) is simply
conventional OFDM as used in radio systems with the radio
frequency carrier replaced by an optical carrier and a number
of patents and patent applications have recognized that conven-
tional OFDM can be applied to optical fiber communications
[9]–[13]. CO-OFDM requires coherent reception and because
of the sensitivity of OFDM to frequency offset [14] and phase
noise, very narrow linewidth lasers are required at the trans-
mitter and receiver and sophisticated tracking algorithms are
required to track laser frequency and phase [15]. To overcome
these disadvantages a number of new forms of OFDM tailored
to optical applications have been described [16]–[19]. Direct-
detection optical OFDM (DD-OOFDM) improves on previous
systems [20], [21] and has been demonstrated experimentally
and by simulation to provide a simple and effective solution to
chromatic dispersion for long-haul single-mode fiber applica-
tions [2], [22], [23].
This paper reports experimental demonstrations of three ver-
sions of DD-OOFDM. These have various degrees of optical
complexity at the transmitter, but all use a single photodiode di-
rect-detection photoreceiver. As well as compensating for chro-
matic dispersion, OFDM can correct for linear distortions in the
electrical components, including the optical modulator and pho-
todiode. This means DD-OOFDM systems are tolerant of linear
imperfections and the experimental systems were implemented
using commercial-off-the-shelf components.
This paper is organized as follows. Section II describes the
OFDM systems in detail. Section III presents the experimental
setup and results. Section IV discusses the stability of the
system. Section V discusses the overheads required to compen-
sate for chromatic dispersion. Section VI provides simulated
results for the optical signal-to-noise ratio (OSNR) for different
system configurations. Section VII summarizes the paper.
0733-8724/$25.00 © 2008 IEEE
Authorized licensed use limited to: Monash University. Downloaded on October 15, 2009 at 22:38 from IEEE Xplore. Restrictions apply.
SCHMIDT et al.: ELECTRONIC DISPERSION COMPENSATION FOR LONG-HAUL TRANSMISSION 197
Fig. 1. DD-OOFDM transmitters: (a) using Hermitian symmetry and an optical
filter; (b) using upconversion and an optical filter; (c) using a frequency domain
Hilbert transform.
II. DIRECT DETECTION OFDM SYSTEMS
The key advantage of DD-OOFDM is that it can be received
using a simple direct detection receiver: no laser is required at
the receiver. To achieve this, an optical single sideband (OSSB)
OFDM signal and a component at the optical carrier frequency
are transmitted. A frequency guard band separates the OFDM
signal from the optical carrier. The signal is received by de-
tecting the carrier
signal mixing products [22]. Chromatic
dispersion causes a frequency dependent phase shift of the
optical signal. For single sideband optical OFDM systems, with
linear field modulation, each OFDM subcarrier is represented
by a single optical frequency, so phase shifts in the optical
domain translate to phase shifts in the demodulated electrical
OFDM subcarriers. These phase shifts can be corrected simply
and without SNR penalty in the digital section of the receiver by
applying a single complex multiplication for each subcarrier.
A. DD-OOFDM Transmitters
Fig. 1 shows three different designs for DD-OOFDM trans-
mitters. The digital sections of the transmitter are conventional
OFDM systems: an IFFT is used to simultaneously modulate
and multiplex the OFDM subcarriers, a cyclic prefix is added
and then the signal is converted from parallel to serial format.
The digital sections of the transmitters differ only in the way the
input vector
is mapped onto the IFFT input vector .
Fig. 1(a) shows the details of the first transmitter design. A
single input optical modulator is used to generate a double side-
band optical signal and then one sideband is suppressed using
an optical filter. The electrical input to the optical modulator
is a real, baseband signal and only one digital-to-analog con-
verter (DAC) is required. In general, the output of an IFFT is
complex rather than real. In systems where a single real output
is required, the input vector to the IFFT,
, is constrained to
have Hermitian symmetry [6] so that the imaginary component
of the IFFT output is zero. Fig. 1(a) shows the input vector
being mapped to the IFFT inputs
(1)
where
denotes the complex conjugate of and is the
size of the IFFT. The inputs
and which correspond
to the dc and Nyquist frequencies are set to zero, as are the inputs
corresponding to guard band frequencies. The number of inde-
pendent complex values transmitted per OFDM symbol depends
on the width of the guard band,
. For the case where the
guard band is equal to
the bandwidth used for the OFDM
signal, then only
independent complex values can be trans-
mitted per OFDM symbol. The transmitter is simple to set up but
if a wavelength-agile (colorless) transmitter is required, a tun-
able optical filter must be used to track the laser.
The details of the second design are shown in Fig. 1(b). The
single DAC of the first design is replaced by two DACs and
an electrical RF upconversion stage. This allows the complex
baseband OFDM signal to be mixed with an RF carrier before
driving the single input optical modulator. The width of the
guard band is determined by the RF frequency (not by nulling
OFDM inputs) and so all subcarriers except the dc subcarrier
can be used to carry data. The analog upconversion allows flex-
ible placement of the signal spectrum relative to the optical car-
rier and the RF frequency is independent of the DAC sample
rate. As in the first design an optical filter is used to suppress
one sideband.
For a given data rate, this design requires a DAC sample rate
of approximately one quarter that of the first design. But the ad-
dition of analog mixers with such high frequency and bandwidth
requirements may cause problems with frequency synchroniza-
tion and inphase (
) and quadrature ( ) balance.
The third transmitter design, which is shown in Fig. 1(c), gen-
erates an optical single sideband by using a signal and its Hilbert
transform to drive an optical I/Q modulator. A Hilbert transform
can be generated simply in an OFDM transmitter, by setting half
of the IFFT inputs to zero. In this case the input vector to the
IFFT is given by
(2)
The IFFT output is then a single sided, analytic signal and its
real and imaginary components are used for the I and Q inputs of
the complex optical modulator for OSSB transmission without
an optical filter. The frequency guard band is created by set-
ting the corresponding inputs to zero. This design requires two
DACs, each with the same sample rate as the first design.
Authorized licensed use limited to: Monash University. Downloaded on October 15, 2009 at 22:38 from IEEE Xplore. Restrictions apply.
198 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 26, NO. 1, JANUARY 1, 2008
Fig. 2. DD-OOFDM direct detection receiver.
B. Biasing and Linear Field Modulation
The biasing of the optical modulator plays an important part
in system performance. To provide the optimum noise perfor-
mance, the transmitter optical modulator should be biased for
equal carrier and sideband powers [22] as this provides the peak
electrical SNR and lowest bit error rate (BER) for a given op-
tical SNR (OSNR). This optimal carrier to sideband power ratio
can be achieved by biasing the modulators just above their in-
tensity nulls. Using this bias point also means that the modulator
is operating in the linear field region where the optical field is
proportional to the input voltage. In the optical OFDM system,
this means that the discrete subcarrier frequencies in the base-
band electrical domain map to discrete frequencies in the op-
tical domain which after direct detection map back to single
electrical subcarrier frequencies. Most conventional optical sys-
tems bias the modulator at the quadrature point. But if this bias
point is used in an optical OFDM system, the discrete input
electrical frequencies map to multiple optical frequencies and
direct detection results in high levels of intermodulation distor-
tion (IMD). For OFDM signals which have a high peak to av-
erage power ratio (PAPR), the average input drive levels were
carefully adjusted to maintain linear field modulation and a rea-
sonable modulator output level.
C. DD-OOFDM Receiver
Fig. 2 shows the simple direct detection receiver required for
DD-OOFDM. As a result of the square law characteristic of the
photodiode, the received signal consists of a number of mixing
products. These can be classified into the useful components
from which the data is recovered, the unwanted components
which fall within band and limit the BER performance of the
system and unwanted components which fall out of band. The
useful components are the difference terms which result from
the mixing of the OFDM sideband and the optical carrier. The
unwanted inband terms are the result of carrier
noise and
signal
noise mixing. If a narrow optical filter is used, only
noise from one optical sideband will be detected, however if a
wider optical filter is used, noise from both sidebands will mix in
with a resulting 3-dB loss in SNR. The use of a frequency guard
band means that all of the results of mixing between OFDM sub-
carriers fall out of band and do not degrade performance. There
may also be other second-order effects: for example, nonlinear-
ities in the system or I/Q imbalance in the transmitter will result
in other components in the transmitted optical signal which may
create other mixing products which degrade performance. The
digital section of the receiver is a conventional OFDM system
in which the phase and amplitude of each received subcarrier is
corrected by the single tap equalizer.
III. E
XPERIMENTAL SETUP AND RESULTS
A series of experiments were performed to compare the
three transmitter designs. In all of the experiments, the OFDM
baseband digital signal processing was performed offline using
MATLAB. The signals to be transmitted were downloaded into
a Tektronix AWG7102 arbitrary waveform generator (AWG)
which generated the required analog baseband signals. The
received signals were captured using an Agilent 81004A digital
sampling oscilloscope (DSO). The OSNR was set by adding
a variable ASE noise source to the signal before the final
optical amplifier. All OSNR measurements include the entire
experimental setup. Standard off-the-shelf, commercial grade,
electrical and optical components were used. The main limi-
tation to the achievable data rates for the two digital systems
was the sampling rate of the AWG which can generate either
two independent 10-GS/s outputs each with a 5-GHz band-
width or a single 20-GS/s output with a 5.8-GHz bandwidth.
Chromatic dispersion was not a limitation, because as long
as the dispersion is less than the length of the cyclic prefix,
chromatic dispersion simply causes a rotation of each OFDM
subcarrier which can be corrected by the equalizer. For each
system the same QAM constellation was used on each data
carrying OFDM subcarrier. The size of constellation for each
system was chosen to give a bit error rate averaged over the
subcarriers of approximately 1
.
A. 12-Gbit/s Digital System
Fig. 3 shows the experimental configuration using the Hilbert
transform based colorless transmitter (third transmitter design).
The AWG is used in dual output mode to generate an analog
signal and its Hilbert transform and these are amplified and used
to drive the two inputs of the Sumitomo single sideband mod-
ulator (a cascaded triple Mach-Zehnder Interferometer modu-
lator). The Hilbert transform enables a single sideband optical
signal to be generated without the need of an optical filter. The
optical source for the modulator is a Photonetics tunable laser
and the modulated output is buffered and transmitted through
five spans of 80 km of Corning SMF-28e fiber. After each span,
the signal is amplified for a target power of 0 dBm. After the
final span, a Fiber Bragg Grating (FBG) is used as an ASE
filter with a 50-GHz bandwidth and the signal is detected by a
Discovery Semiconductors DSCR404 photoreceiver. The DSO
captures the signal which is later processed using MATLAB.
A 512-point IFFT with no oversampling is used in the trans-
mitter. With a guard band equal in width to the OFDM band-
width and with the nulling of subcarriers to create the Hilbert
transform, this design theoretically allows 128 independent data
carrying subcarriers. However in this experiment, a further five
subcarriers are not used due to their proximity to a side mode
of the external cavity laser. Of the three systems, this one is the
least affected by component imperfections and can support the
highest order QAM constellation. This is due to the simplicity
of the analog electrical and optical components and because the
AWG is used within specification. The available 123 subcarriers
are modulated with 32 QAM, giving a data rate of 12 Gbit/s. A
cyclic prefix of 32 samples is used in the two digital systems
and a cyclic prefix of 64 samples is used in the system using an
RF carrier.
Authorized licensed use limited to: Monash University. Downloaded on October 15, 2009 at 22:38 from IEEE Xplore. Restrictions apply.
SCHMIDT et al.: ELECTRONIC DISPERSION COMPENSATION FOR LONG-HAUL TRANSMISSION 199
Fig. 3. Schematic of the two digital baseband electrical systems: 12 and 20 Gbit/s.
Fig. 4. Results for 12-Gbit/s colorless system using digital Hilbert transform to generate single sideband signal. Unequalized constellation. Equalized constella-
tion. Channel response
E
=
N
and BER.
Three hundred OFDM symbols (184 500 bits) were trans-
mitted over a distance of 400 km and recovered with a BER of
3.03
at a measured OSNR of 25 dB. The un-equalized
and equalized constellations can be seen in Fig. 4. The single
tap equalizer function in the OFDM receiver corrects for the
amplitude distortions caused by frequency roll-off of the com-
ponents and the phase distortions caused by chromatic disper-
sion and OFDM frame timing offsets. The high QAM system
also illustrates the low levels of IMD that can be achieved
using a modulator biased in the linear optical field region. The
required OSNR for the 12-Gbit/s system was 22 dB for a BER
of 1
. Fig. 4 also shows the relative channel response of
the entire system, the received SNR in terms of
, and
the BER as a function of frequency. The figure illustrates the
falling response at high frequencies which causes a reduction
in the SNR and an increase in the BER on the higher frequency
subcarriers.
B. 20-Gbit/s Digital System
Fig. 3 also shows the experimental configuration for the
20 Gbit/s digital system which uses a single DAC and an
optical filter (first transmitter design). Because only one output
is required, the two channels of the AWG are interleaved to in-
crease the sampling rate to 20 GS/s. The I/Q optical modulator
of the first experiment is replaced by a Fujitsu MZI modulator.
One optical sideband is suppressed using a FBG in transmission
mode as an optical filter. Because of the loss in the FBG, the
length of the first fiber span is reduced to 30 km. The receiver is
the same as for the 12-Gbit/s system, however a tunable optical
filter (200-GHz FWHM) is used as the ASE filter.
Fig. 5. Results for 20 Gbit/s digital system. Equalized constellation channel
response,
E
=
N
and BER.
A 512-point, nonoversampled, IFFT was used which, al-
lowing for Hermitian symmetry and a guard band, gives
128 independent data carrying subcarriers. The output signal
requires a 5- to 10-GHz bandwidth which exceeds the 5.8-GHz
bandwidth limit specified for the AWG. This experiment is only
possible if the AWG DACs are set to produce return-to-zero
pulses for each sample rather than holding their output values
Authorized licensed use limited to: Monash University. Downloaded on October 15, 2009 at 22:38 from IEEE Xplore. Restrictions apply.
200 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 26, NO. 1, JANUARY 1, 2008
Fig. 6. 10 Gbit/s RF electrical upconverted system.
Fig. 7. Results for 10-Gbit/s system using RF upconversion. Equalized constellation.
E
=
N
versus Subcarrier – High frequencies in middle subcarriers. BER
versus OSNR.
for the duration of the sample. The specified bandwidth is still
5.8 GHz but the rate of roll-off above this limit is reduced and
a response at 10-GHz of minus 10 dB is achieved. However,
this mode introduces spurious noise and frequency response
aberrations. A constellation size of 16QAM was used for a data
rate of 20 Gbit/s. Two hundred symbols (102 000 bits) were
transmitted over a distance of 320 km with a BER of 2.5
at an OSNR of 26 dB.
Fig. 5 shows the equalized constellation, the signal to noise
ratio (
), the relative channel response and the BER versus
frequency. The frequency response ripples are due to the AWG
in return-to-zero mode and the bit errors can be seen to line
up with the frequency response dips. A direct, back to back,
electrical connection between the AWG and the DSO resulted
in a BER of 3
and an OSNR of 26 dB was required for
a system BER to 2.5
.
C. 10-Gbit/s Up-Converted System
Fig. 6 shows the experimental set up for the RF electrical
upconverted system which uses the second transmitter design.
The OFDM signal is modulated onto a 7.5-GHz electrical carrier
before driving the optical modulator. The main limitation in this
system is the bandwidth of the RF I/Q mixer.
A 512 subcarrier, 2
oversampled, trigonometrically inter-
polated [24] OFDM signal is generated by using a 1024 point
IFFT with zero padding at the center. In contrast with the pre-
vious digital systems, all 512 subcarriers are modulated. The
two output signals from the AWG each have a useful bandwidth
of 2.5 GHz. These are low pass filtered to remove images and
modulated onto a 7.5-GHz RF-carrier using an I/Q mixer. The
bandwidth after up conversion is 5 to 10 GHz. The signal is am-
plified to drive a single input modulator and an FBG in trans-
mission mode is used to suppress one sideband. Because of the
imperfections of the RF electronics, only 4-QAM modulation
could be supported and so the data rate was limited to 10 Gbit/s.
Five spans of 80 km of Corning SMF-28e fiber are used.
Fig. 7 shows the equalized constellation,
as a func-
tion of subcarrier and the BER versus OSNR. The dip in high
frequencies (the center of the channel response graph) increased
the required OSNR for a given BER. Nevertheless, because
4-QAM modulation was used, the required OSNR for a BER
of 1
was 17.2 dB; 5 dB lower than the 12-Gbit/s digital
system which used 32 QAM.
IV. C
HANNEL STABILITY
One of the advantages of DD-OOFDM compared with
CO-OFDM is the relative stability of the channel and the re-
duced overheads in terms of training symbols and pilots. In this
section we show that the channels in the experiments were very
stable and that as a result, only infrequent training symbols are
required.
The channel stability of the 12-Gbit/s digital system is dis-
played in Fig. 8, which shows the unequalized amplitude of all
data carrying subcarriers for all 300 received symbols as a func-
tion of frequency and time. The frequency axis shows the ampli-
tude response of the subcarriers in the 2.4 to 4.8 GHz spectrum
and the time axis shows the subcarriers amplitude response over
a time period of 15.36
, equivalent to the transmission of 300
Authorized licensed use limited to: Monash University. Downloaded on October 15, 2009 at 22:38 from IEEE Xplore. Restrictions apply.
