Content uploaded by Agnieszka Konczykowska
Author content
All content in this area was uploaded by Agnieszka Konczykowska on Dec 16, 2015
Content may be subject to copyright.
824 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 32, NO. 4, FEBRUARY 15, 2014
High Symbol Rate Coherent Optical Transmission
Systems: 80 and 107 Gbaud
Gregory Raybon, Andrew Adamiecki, Peter J. Winzer, Sebastian Randel, Luis Salamanca, A. Konczykowska,
Filip Jorge, Jean-Yves Dupuy, Larry L. Buhl, Sethumadhavan Chandrashekhar, Chongin Xie,
Steve Draving, Marty Grove, Kenneth Rush, and Ruediger Urbanke
Abstract—We demonstrate high speed optical transmission sys-
tems using digital coherent detection at all-electronically mul-
tiplexed symbol rates of 80 and 107 Gbaud. At 107 Gbaud,
we demonstrate a single-carrier polarization division multiplexed
quadrature phase shift keyed (PDM-QPSK) line rate of 428 Gb/s.
At 80 Gbaud, we achieve a single-carrier line rate of 640 Gb/s using
PDM 16-ary quadrature amplitude modulation (16-QAM). Using
two optical subcarriers, we demonstrate a 1-Tb/s optical interface
and conduct long-haul wavelength-division multiplexed (WDM)
transmission on a 200-GHz grid over 3200 km of ultra-large effec-
tive area fiber.
Index Terms—Coherent communications, optical communica-
tions.
I. INTRODUCTION
THE ever-increasing demand for bandwidth is fueling re-
search and development in high-capacity optical transmis-
sion systems using high-speed per-channel optical interfaces.
Commercial 100-Gb/s transponders employ single-carrier po-
larization division multiplexed quadrature phase shift keying
(PDM-QPSK) with symbol rates around 30 Gbaud [1]–[4]. Ap-
proaches to achieve 400-Gb/s and 1-Tb/s channels range from
single-carrier [5] and dual-carrier [6]–[8] systems to orthogo-
nal frequency division multiplexing (OFDM) and optical super-
channels [9], [10] and recently also to spatial superchannels in
the context of spatial multiplexing systems [11], [12]. Terabit
superchannels have been demonstrated in research using, for
Manuscript received July 25, 2013; revised September 19, 2013; accepted
October 14, 2013. Date of publication November 5, 2013; date of current
version January 10, 2014.
G. Raybon, A. Adamiecki, P. J. Winzer, S. Randel, S. Chandrasekhar,
L. L. Buhl, and C. Xie are with the Bell Laboratories, Alcatel-Lucent,
Holmdel, NJ 07733 USA (e-mail: greg.raybon@alcatel-lucent.com; andrew.
adamiecki@alcatel-lucent.com; peter.winzer@alcatel-lucent.com; sebastian.
randel@alcatel-lucent.com; Chandra.Sethumadhavan@alcatel-lucent.com;
lawrence.buhl@alcatel-lucent.com; chongjin.xie@alcatel-lucent.com).
A. Konczykowska, F. Jorge, and J.-Y. Dupuy are with the III–V Labs Route
de Nozay, Marcoussis 91460, France (e-mail: agnieszka.konczykowska@
3-5lab.fr; filipe.jorge@3-5lab.fr; jean-yves.dupuy@3-5lab.fr).
S. Draving, M. Grove, and K. Rush, are with the Agilent Technologies,
Colorado Springs, CO 80907 USA (e-mail; steve_draving@agilent.com;
marty_grove@agilent.com; kenneth_rush@non.agilent.com).
L. Salamanca is with the Department of Signal Theory and Communications,
University of Sevilla, 41004 Seville, Spain (e-mail: l.s.mino@gmail.com).
R. Urbanke is with the LTHC, IC, EPFL, Lausanne 1015, Switzerland
(e-mail: ruediger.urbanke@epfl.ch).
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.2013.2286963
example, 24 [13], 8 [14], 4 [15], and 2 [8] optical subcarri-
ers. The trade-off in designing (spectral or spatial) superchan-
nel transponders is between electronic multiplexing and opto-
electronic modulation/detection speed [16], [17]. Since opti-
cal components typically dominate transponder costs, choosing
the highest economically feasible electronic multiplexing rates
with the lowest possible number of superchannel subcarriers is
usually optimal; hence the desire to increase single-carrier bit
rates as much as possible. In the context of advanced modula-
tion formats and digital coherent detection, the highest single-
carrier line rates have been achieved using polarization-division
multiplexed 16-ary quadrature amplitude modulation (PDM-16-
QAM) [5], [8], [18].
In this paper, we present recent experimental results of very
high symbol rate coherent optical transmission systems based
on electronic time-division multiplexing (ETDM) to 80 and
107 Gbaud to achieve single- and dual-carrier serial interface
rates from 400 Gb/s to 1 Tb/s. Specifically, we demonstrate
(1) the generation, detection, and 4,800-km transmission of 10
single-carrier channels of 428-Gb/s using 107-Gbaud PDM-
QPSK, and (2) the generation, detection, and 3,200-km trans-
mission of 5 dual-carrier channels carrying 1.28 Tb/s using 80-
Gbaud PDM-16-QAM. These experiments currently represent
the highest optical interface rates using electronic multiplexing
with the lowest numbers of optical carriers. Spectral efficien-
cies of 3.3 b/s/Hz (net bit rate of 400 Gb/s on a 120-GHz grid)
and 5 b/s/Hz (net bit rate of 1 Tb/s on a 200-GHz grid) are
demonstrated for each system, respectively. The transmission
distances were achieved assuming 7% overhead hard decision
forward error correction (HD-FEC) in the 400-Gb/s experiment,
and implementing 23% overhead soft-decision FEC (SD-FEC)
in the 1-Tb/s experiment.
II. TRENDS IN SERIAL INTERFACE RATES
Fig. 1 reviews the evolution of serial single-carrier interface
rates as a function of time for optical transmission systems as
represented by research milestones. The solid green line repre-
sents single-carrier line rates and the dashed red line represents
the corresponding electronic symbol rates. Since binary on/off
keying (OOK) in combination with direct detection was used in
all these record experiments up until the first 100-Gb/s demon-
strations, symbol rates and line rates coincide prior to 2005
and increase steadily from 8 Gb/s in 1986 [19] to 100 Gb/s in
2005 [20]. The first 100-Gb/s transmitter that did not employ
the well-known technique of optical time-division multiplex-
ing (OTDM) [21] but instead used all-electronic multiplexing
0733-8724 © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
RAYBON et al.: HIGH SYMBOL RATE COHERENT OPTICAL TRANSMISSION SYSTEMS: 80 AND 107 Gbaud 825
Fig. 1. Evolution of serial interface rates for optical systems.
was demonstrated in 2005 [20]. This was soon followed by the
first 100-Gbaud ETDM receivers [22], [23], the first ETDM
transmitter and receiver combinations [24] and the first fully
ETDM 100G transmission field experiment [25]. In order to
meet the demand for 100-Gb/s serial interfaces at higher spec-
tral efficiencies than supported by OOK, quaternary modulation
schemes were soon applied. This relaxed the bandwidth require-
ments of the optoelectronic components and enabled 100-Gb/s
to be obtained through 50-Gb/s electronics and QPSK [26], [27].
Soon after, digital coherent detection inherently using PDM and
coupled with digital signal processing (DSP) allowed a further
reduction in symbol rates by another factor of two, establish-
ing 100-Gb/s transmission at 28 Gbaud [1]. In Fig. 1 the dashed
line, representing symbol rates, takes an abrupt step down at this
point in time. More complex modulation formats subsequently
permitted single-polarization line rates that were more than a
factor of 4 higher than the electronically generated symbol rates
and allowed for high-speed transmission at even higher spectral
efficiencies. Modern coherent systems are limited by the avail-
ability of high-speed and high-resolution ADCs and DACs, with
a linear bit rate increase with symbol rate but only a logarith-
mic bit rate increase with constellation size, which is coupled
to the required converter resolution [17]. Thus, it took a few
years for symbol rates to climb back up to the 100-Gbaud level
that was obtained through OOK in 2005. While today’s com-
mercial systems using real-time DSP operate at symbol rates
of about 30 Gbaud, using ADCs and DACs sampling at around
60 Gsamples/s [28], most research beyond these rates relies on
high-speed active [29] or passive [5], [8] DACs at the transmitter
as well as on high-speed ADCs at the receiver, typically imple-
mented as real-time sampling oscilloscopes [30]–[32] followed
by off-line DSP. At 56 Gbaud [5], [31] and 80 Gbaud [33], [34],
PDM-QPSK and PDM-16-QAM have been reported, while only
PDM-QPSK has been reported at 107 Gbaud [35] as of now.
III. 10 ×107-GBAUD (428-GB/S) PDM-QPSK
A. 107-Gbaud QPSK Transmitter
Fig. 2 shows our ETDM 107-Gbaud PDM-QPSK transmit-
ter set-up. The in-phase (I)and quadrature (Q)107-Gb/s bit
sequences are generated through two stages of electronic mul-
tiplexing. The first stage accepts 4 delay decorrelated copies of
a2
15–1 pseudorandom bit sequence (PRBS) at 13.375 Gb/s
Fig. 2. 10 ×107-Gbaud (428-Gb/s) 10-channel PDM-QPSK transmitter.
(a) 107-Gb/s electrical eye diagram. (b) Optical 107-Gbaud QPSK eye diagram.
(c) Optical spectrum of signal immediately following the LiNbO3modulator.
and combines them to provide differential 53.5-Gb/s data
sequences. The delay between copies of the 13.375-Gb/s signals
ranges from 5 to 26 bits. The two complementary outputs are
also delay decorrelated and multiplexed to 107 Gb/s using 2:1
multiplexers fabricated in high-speed Indium Phosphide het-
erojunction bipolar transistor (InP-HBT) technology [36]. Inset
a) of Fig. 2 shows the electrical eye diagram at the output of
the multiplexers with a peak-to-peak voltage swing of 500 mV.
The 107-Gb/s signal is again delay decorrelated by 7 and 13
bits and applied, without the use of driver amplifiers, to the
in-phase (I)and quadrature (Q)arms of an integrated LiNbO3
dual-drive double-nested Mach–Zehnder modulator (MZM). At
107 Gb/s, the modulator is operated well above its 3-dB band-
width of 37 GHz and well below the optimal operating voltage
swing (2Vπ=7 V), resulting in an additional insertion loss of
∼25 dB. An optical QPSK intensity eye is shown in inset (b) of
Fig. 2, measured using a 70-GHz photodiode and a high-speed
sampling oscilloscope with 70-GHz bandwidth. The signal opti-
cal spectrum is shown in inset (c) of Fig. 2. After modulation, the
optical signal is pre-filtered using a programmable optical filter
(OF), which is set to sharply suppress any spectral content out-
side a 110-GHz passband. Within its passband, the OF equalizes
for the limited bandwidth of opto-electronic transmit and receive
components by emphasizing the high-frequency portions of the
signal spectrum [37]. The joint optimization of transmit-side
pre-emphasis and receive-side digital equalization is described
in detail in [33]. The OF characteristic is shown in Fig. 2 and the
resulting optical signal spectrum in Fig. 3, inset b). To generate
10 densely spaced WDM channels with only a single available
107-Gbaud transmitter, we first form a 240-GHz spaced mul-
tiplex of 4 distributed feedback (DFB) lasers and one tunable
external cavity laser (ECL, for the signal under test), cf. Fig. 2.
After modulation, each of the 5 optical signals is converted into
two frequency-shifted copies at ±60 GHz from the original laser
frequency using the dual-tone generator shown in Fig. 3, imple-
mented by a LiNbO3MZM biased at the null and sinusoidally
driven at 60 GHz. Spurious tones are suppressed by >30 dB, as
shown in inset a) of Fig. 3, measured with only a CW signal at
the input to the tone generator. Owing to the sharp pre-filtering,
the shifted spectra do not noticeably overlap and hence do not
produce any measurable crosstalk. This fact, together with the
rapid channel walk-off due to fiber dispersion at the high un-
derlying symbol rates, permits the use of a single modulator
826 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 32, NO. 4, FEBRUARY 15, 2014
Fig. 3. Frequency shifting scheme using dual-tone generation in a LiNbO3
MZM producing channels spaced on a 120-GHz grid. (a) Example of tone
generation when a CW signal is input to the modulator. (b) 107-Gbaud QPSK
spectrum at the input to the frequency shifter. (c) Frequency shifted output.
(d) 10-channel WDM spectrum.
Fig. 4. 107-Gbaud coherent intradyne receiver. LO: local oscillator, Var ATT:
variable attenuator, DG: delay generator.
in this experiment. The maximum optical signal-to-noise ratio
(OSNR) per channel is ∼32 dB, referenced to 0.1 nm resolution
bandwidth and taking into account noise in both polarizations.
The resulting 2 WDM channels generated from a single input
signal are shown in inset c) of Fig. 3. Inset d) of Fig. 3 shows the
10-channel WDM spectrum generated from 5 original modu-
lated signals. After polarization multiplexing using a standard
PDM emulator with a symbol delay of 15 ns, we obtain an ag-
gregate line rate of 4.28 Tb/s (10 ×428 Gb/s). Assuming 7%
overhead (for FEC and framing), this results in ten 400-Gb/s
single-carrier channels a net spectral efficiency of 3.3 b/s/Hz.
B. 107-Gbaud QPSK Receiver
The coherent intradyne receiver shown in Fig. 4, consists of
a noise-loading stage to set the received OSNR, followed by a
125-GHz optical filter to select each of the wavelength channels
to be tested. The signal is combined with an ECL local oscilla-
tor (∼100-kHz linewidth) in a polarization diversity 90-degree
optical hybrid. The signal outputs from the hybrid are detected
using balanced detectors with 39-GHz bandwidth. For analog-
to-digital conversion, two high-speed Agilent 90000 Q-series
oscilloscopes, whose two 63-GHz bandwidth inputs are sam-
pling at 160 GS/s, are used. To measure the PDM signals, the
two scopes are co-triggered using a logic circuit that precisely
triggers both scopes at the same time [5] and are also syn-
chronized with a 10-MHz clock reference. Multiple blocks of
1 million samples are recorded for offline DSP.
At 160 GS/s, the 107-Gbaud signal is oversampled by a fac-
tor of about 1.5; the oscilloscope front-end limits the signal
spectrum to ∼63 GHz, which is equivalent to a sharp 126-GHz
optical bandpass filter. The off-line intradyne receiver algorithm
corrects for front-end skews and hybrid phase errors before ap-
plying bulk chromatic dispersion compensation in the frequency
domain. The algorithm then oversamples a portion of the signal
by a factor of 3 using zero-padding in the frequency domain
and extracts the clock tone according to the nonlinear spec-
tral line method. Using the recovered clock, it synchronously
upsamples the signal from 1.5 to 2 [38]. We then use stan-
dard blind intradyne QPSK signal processing, including 16-tap
T/2-spaced finite impulse response (FIR) filters in a butter-
fly structure, adapted by the constant-modulus algorithm [39],
fourth-power frequency and phase recovery [40], and differ-
ential decoding to avoid potential phase slips. The FIR filter’s
8-symbol impulse response is short enough to avoid correlation
artifacts in the BER [41].
C. System performance and 10 ×400G WDM transmission
Transmission testing is performed using a recirculating loop
consisting of four 100-km spans of ultra-large effective-area
fiber with a chromatic dispersion at 1550 nm of 20 ps/nm/km,
an attenuation of 0.185 dB/km, an effective area of 120 μm2and
a nonlinearity coefficient of 0.81 W−1m−1. Raman/EDFA am-
plification is used with a total gain of 25 dB. The launch power
per channel is controlled at the beginning of each span with vari-
able optical attenuators and is optimum near +5.0 dBm/channel
as shown below. Two additional EDFAs are used at the input
and output of the loop to account for extra losses in the loop
switches and couplers. A gain equalizing filter is employed to
flatten the gain response of the amplifiers in the loop to main-
tain equal powers in all WDM channels. No in-line dispersion
compensation is employed. The back-to-back 107-Gbaud BER
measurements for the 428-Gb/s PDM-QPSK signals are shown
in Fig. 5 (for a PRBS length of 215–1 at the 13.375-Gb/s pattern
source). Data is plotted for the seed wavelength after spectral
shaping and for the two resultant ±60-GHz copies after fre-
quency shifting. At BER =3.8 ×10−3, which corresponds to
the threshold of a typical 7%-overhead HD-FEC [42], the re-
quired OSNR is 21 dB. The theoretical BER for differentially
decoded QPSK is also shown. The implementation penalty at
the FEC limit is ∼2 dB. The measured back-to-back BER for
the tightly filtered channel and two ±60-GHz copies perform
the same, confirming the absence of crosstalk and other artifacts
in our WDM generation scheme. Examples of the constella-
tion diagrams are shown for the two polarizations of one WDM
channel at high (27-dB) OSNR.
Fig. 6 shows the Q-factor as calculated from the measured
BER versus launch power for a central channel (Ch4) at dif-
ferent transmission distances. The launch power is controlled
RAYBON et al.: HIGH SYMBOL RATE COHERENT OPTICAL TRANSMISSION SYSTEMS: 80 AND 107 Gbaud 827
Fig. 5. Back-to-back BER measurements for 107-Gbaud PDM-QPSK signals.
BER is shown for the seed wavelength and two resulting shifted channels
(±60 GHz). Inset: QPSK constellations for one channel and both polarizations
at 27-dB OSNR.
Fig. 6. Q-factor versus launch power per channel for channel 4 (107-GBaud
PDM QPSK) at transmission distances of 3200, 4000, 5600, and 6400 km.
at the input to each span and is optimum near +5dBm.Only
4 channels were transmitted during this measurement in order
to achieve the necessary high launch powers due to limitations
of the optical amplifiers. At power levels above +5dBm,the
BER degrades due to nonlinear transmission impairments. We
observe that after transmission over 5,600 km, a Q-factor above
the HD-FEC threshold (8.5 dB) is achieved for a range of launch
powers from +3to+6 dBm per channel.
Figs. 7 and 8 show the Q-factor vs. transmission distance
for all ten 400-Gb/s channels. In the 10-channel transmission
experiment, the per-channel launch power is limited to ∼3to
4 dBm, which is less than the optimum as determined in the
measurement shown in Fig. 6, resulting in a slightly shortened
transmission reach. The BER for each PDM-QPSK channel is
measured at distances from 3,200 km to 7,200 km at 800-km
increments and is plotted in terms of Q-factor. (The missing data
point for channel 10 at 3,200 km is due to a file capture error
Fig. 7. Transmission performance for ten 400-Gb/s channels versus distance
(107-GBaud PDM-QPSK).
Fig 8. Measured Q-factor for each channel at transmission distances from
3,200 to 7,200 km (107-GBaud PDM-QPSK).
during the measurement.) Only channel 4 is measured below
3,200 km, as plotted in Fig. 7. The variation in Q-factor from
channel to channel is largely due to imperfect gain equalization
in the recirculating loop resulting in variations of the delivered
OSNR of each channel. Constellation diagrams for each polar-
ization of channel 4 are shown in Fig. 7 after transmission over
3,200 and 7,200 km. All channels achieve a BER above the
HD-FEC threshold at 4,800 km.
IV. 1.0-TB/SDUAL-CARRIER 80-GBAUD PDM-16-QAM
In earlier work, we reported an 80-Gbaud PDM-QPSK
(320-Gb/s) experiment, demonstrating 8-channel WDM trans-
mission over 5,600 km in a recirculating loop experiment similar
to the one described in section III [43]. In that work, a spectral ef-
ficiency of 2.9 b/s/Hz (at 10.3% overhead), with WDM channels
on a 100-GHz grid was achieved, assuming an HD-FEC limit of
3.8 ×10−3. In this section, we describe the generation of a PDM-
16-QAM signal at 80 Gbaud and its dual-carrier superchannel
extension to achieve a per-channel interface rate of 1.28 Tb/s
(640 Gb/s per subcarrier). WDM transmission of five such
1-Tb/s channels at a spectral efficiency of 5.2 b/s/Hz (at 23%
OH) is demonstrated over 3,200 km using coherent detection,
offline digital signal processing, and SD-FEC.
828 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 32, NO. 4, FEBRUARY 15, 2014
Fig. 9. 1.28-Tb/s 80-Gbaud dual carrier PDM-16- QAM transmitter. Insets:
(a) 4-level 80 Gbaud electrical eye diagram and (b) optical filter response
characteristic, and (c) 5 dual-carrier channel spectra.
A. 80-Gbaud Dual-Carrier PDM-16-QAM Transmitter
The schematic of the dual-carrier PDM-16-QAM transmitter
is shown in Fig. 9. The 16-QAM signal is generated using a
nested LiNbO3MZM driven by a 4-level electrical signal at
80 Gbaud, derived from two high-speed InP-HBT multiplexers
that combine two delay-decorrelated copies of 40-Gb/s PRBS of
length 215–1 originating from separate PRBS generators. Two
3-dB attenuators and a control voltage are used to obtain two
sets of signals with a 2x amplitude difference. These are then
combined in a 65-GHz resistive combiner to form two 4-level
80-Gbaud electrical waveforms for in-phase (I)and quadrature
(Q)modulation (cf. inset) with an I/Q delay of 200 ps (16 sym-
bols). Similar to the 107-Gbaud QPSK system described above,
the optical signal is pre-filtered using an OF whose bandwidth
is 95 GHz and whose passband has a sinusoidal shape with a
10-dB optically equalizing dip to compensate for the limited
opto-electronic bandwidth of various transmit and receive sys-
tem components. The OF transfer characteristic is shown in
inset b) in Fig. 9. The 320-Gb/s (80-Gbaud 16-QAM) signal is
then shifted by ±50 GHz using the same technique as shown in
Fig. 3, except now the LiNbO3MZM for dual-carrier generation
is driven by a 50-GHz sine wave to produce sub-carriers with
100-GHz spacing and >35 dB suppression of spurious tones.
The dual-carrier channel is then polarization multiplexed to a
total channel rate of 1.28 Tb/s by a standard PDM emulator
(with a delay of 1,200 symbols). Finally, the two subcarriers are
separated using a 100/200-GHz interleaver, delay decorrelated
by ∼2,400 symbols, and recombined in a matching interleaver.
In WDM operation, 5 lasers separated by 200 GHz are simul-
taneously input to the modulator to produce 5 such dual-carrier
channels which at this point in the system may also be viewed as
10 modulated 80-Gbaud PDM-16-QAM signals on a 100-GHz
grid. The optical spectrum is shown in inset c) of Fig. 9.
B. 80-Gbaud Dual-Carrier Receiver, DSP, and SD-FEC
For back-to-back characterization and transmission testing
we used a coherent intradyne receiver similar to the one shown
in Fig. 4, and implemented off-line DSP and SD-FEC decod-
ing. The underlying DSP algorithm of this experiment [44]
first compensates for receiver front-end imperfections and bulk
chromatic dispersion. It then applies a T/2-spaced feed-forward
adaptive butterfly equalizer with 30 taps per filter. This filter
Fig. 10. Proposed schemes for the decoder: in (a) the scheme implement an
LDPC as SD code, as already stated in the literature, and in (b) we consider
LDPCC codes as the inner SD-FEC.
length ensures the absence of any pattern correlation artifacts
(the 16-symbol I/Q decorrelation delay being the limiting fac-
tor) [41]. Coefficient acquisition is obtained through a data-
aided least-mean squares (LMS) algorithm based on 40,000
symbols before switching to tracking mode based on a decision-
directed LMS. Coarse carrier frequency estimation is done dur-
ing the data-aided acquisition phase, and the carrier phase is
tracked blindly using a maximum likelihood scheme with 30
test-angles [38], [45]. We evaluate the raw bit error ratio (BER)
over 400,000 symbols (3.2 million bits per subcarrier), but also
pass the DSP-processed noisy recovered signal constellations
on to the SD-FEC for decoding.
In order to compensate for possible transmission performance
differences between the two carriers and the two polarizations
making up our dual-carrier Terabit signal, we implement two
levels of interleaving: First, we interleave xand ypolarizations
of each subcarrier, and then we interleave the two subcarriers
making up a Terabit channel. In other words, the aggregate Tb/s
transmit symbol sequence {s1,s
2,s
3,s
4,s
5,s
6,s
7,s
8,...}is
mapped as {s1x
1,s
1y
2,s
2x
3,s
2y
4,s
1x
5,s
1y
6,s
2x
7,s
2y
8,...}, where the
subscript denotes the temporal symbol number and the super-
script denotes the subcarrier and polarization that the respec-
tive symbol is mapped to. It is at this important point that the
dual-carrier interface differs from two independently modulated
500-Gb/s channels. Once the received signal is digitally pro-
cessed, the receiver undoes both stages of interleaving. Then,
the algorithm calculates, for each soft received symbol value,
the 16 probabilities by which that particular value is caused by
each of the 16 possible transmit symbols. These calculations
are based on the estimated transition probabilities, which the
receiver obtains during a training phase by estimating the indi-
vidual received symbol clouds’ means and their (common) vari-
ance. Combining the 16 estimated probabilities for the transmit
symbol with the underlying bit-to-symbol mapping, the algo-
rithm calculates the probability with which each of the 4 bits
corresponding to the received symbol value is a 0 and a 1. These
bit-wise probabilities are then used by the subsequent SD-FEC
decoding block shown in Fig. 10, which we will explain in more
detail below for the two particular coding schemes compared in
RAYBON et al.: HIGH SYMBOL RATE COHERENT OPTICAL TRANSMISSION SYSTEMS: 80 AND 107 Gbaud 829
this work. As illustrated in Fig. 10, we use a concatenated cod-
ing scheme where an inner SD-FEC lowers the error rate and an
outer HD-FEC eliminates the residual error floor. The HD-FEC
scheme itself implements an inner and an outer section [42].
As shown by the feedback loop in Fig. 10, in order to improve
the decoding threshold of the concatenated HD-FEC, the HD
decoder iterates between the inner and the outer sections of the
code, i.e., it first decodes the inner code and the output is then
decoded in the outer section, whose output is fed back again
to the inner section and so on. A detailed description of both
HD-FECs can be found in [42].
We first use a quasi-cyclic low-density parity-check (QC-
LDPC) code to implement the SD-FEC, as already studied in
earlier work [46] and as shown in Fig. 10(a). The code is con-
structed with the progressive edge-growth algorithm [47], which
ensures a minimum girth of 8. The QC-LDPC code has a degree
distribution of (3, 24), which leads to an overhead of 14%. The
soft decoder runs the belief propagation algorithm over binary
codewords of length 32,760.
Alternatively, and as shown in Fig. 10(b), we propose to
use LDPC convolutional (LDPCC) codes as the SD-FEC.
LDPCC codes are a novel coding scheme that promises to
achieve near-Shannon limited performance [48]. We choose a
degree distribution of (4, 32), which leads to an overhead of
14%, as discussed above. Although large codewords are re-
quired, we implement a window-sliding scheme [49] for the
belief propagation algorithm, which allows us to match the de-
coding complexities between LDPC and LDPCC codes. This
way, we are able to perform a fair comparison in terms of de-
coding complexity between the two implemented schemes. Out
of the 85 blocks of 4,680 bits that form each LDPCC code-
word, only 7 are considered at the same time by the decoder,
which results in a practical code-length of 32,760, as used for
the QC-LDPC described above.
After the SD-FEC, we implement a 7% overhead HD-FEC
code, finally leading to a total overhead of 23%. Specifically,
we use for the scheme of Fig. 10 (a) the concatenated HD-FEC
invoked in previous sections of this paper, with a correction
threshold of 3.8 ×10−3[42]. Conversely, since the exact inner
structure of that HD-FEC is not publically known but its knowl-
edge is essential for the feedback path introduced in Fig. 10(b),
we use the concatenated-BCH I.3 proposed in the G.975.1 rec-
ommendation [45] for the LDPCC scheme. This code has a
threshold of 3.1 ×10−3[42], as measured at the output of
the SD-FEC decoder, and is chosen due to the great correction
capability of its inner section.
In the LDPCC scheme represented in Fig. 10(b), after each
block of 4,680 is decoded, the window slides one block for-
ward and the decoded bits are used to change the states of the
SD decoder [49]. In order to improve the performance of the
whole system, instead of directly using the output bits from the
SD decoder, we feed these bits to the inner section of the HD
decoder. The output of this block is then provided to the SD-
FEC decoder, as illustrated in Fig. 10(b) with the dashed arrow,
to change the states in the SD-FEC decoder before continuing
the decoding. Note that this feedback does not imply additional
iterations.
Fig. 11. Measured back-to-back BER versus. OSNR for dual-carrier Tb/s
interface. Pre-FEC and post-FEC performance using soft-FEC is shown (SP:
Single polarization; SC: Subcarrier).
Fig. 12. Zoom of back-to-back BER versus OSNR. Constellation diagrams
for 16-QAM signals for each polarization and for each subcarrier at high OSNR
are shown.
C. 80-Gbaud Dual-Carrier PDM-16-QAM Back-to-Back BER
The back-to-back BER before and after SD-FEC is plotted in
Fig. 11 as a function of the OSNR. The OSNR values reported
in Fig. 11 only account for the ASE loaded at the receiver but
neglect any residual ASE due to transmit-side optical ampli-
fiers. The maximum OSNR available at the transmitter before
filtering is ∼32 dB as measured on the signal before subcarrier
generation. The single-polarization raw BER for the unshifted
80-Gbaud signal directly after the data modulator (black aster-
isks) is shown together with subcarrier 2 (SC2) at the transmitter
output (black crosses). At a BER of 10−2, the implementation
penalty referred to theory for non-differential single polarization
16-QAM is 3 dB, with an additional 1 dB due to dual-carrier
generation and interleaver-based decorrelation. The raw BER
(pre-FEC) for PDM-16-QAM are represented by red and blue
circles for SC1 and SC2 in Fig. 11, and we observe an expected
increase in the required OSNR of 3 dB. The corrected BER is
also plotted for each subcarrier, for both LDPC and LDPCC
codes (squares, diamonds). Fig. 12 examines more closely the
SD-FEC correction for the BER data corresponding to the PDM
830 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 32, NO. 4, FEBRUARY 15, 2014
Fig. 13. Q-factors for all 5 WDM Terabit channels after transmission. Pre-
FEC BER is shown for 2,000, 2,400, 2,800, and 3,200 km. Post-FEC BER using
LDPC and LDPCC is shown for all channels at 3,200 km.
16-QAM subcarriers in the range of OSNRs between 21 and
23 dB as indicated by the dashed box in Fig. 11. The post-SD-
FEC BERs using QC-LDPC (red and blue squares) and LDPCC
(red and blue diamonds) are plotted for both subcarriers. The
threshold of the outer HD-FECs, 3.1 ×10−3and 3.8 ×10−3,are
reached at a required OSNR of ∼22 and 22.4 dB, respectively,
for the two different SD-FECs.
D. 5-Tb/s WDM Transmission
The five 1.28-Tb/s channels are then transmitted in a WDM
setup using a recirculating loop as described in Section III-C.
The optimum launch power per Tb/s channel was +6dBm.
Fig. 13 shows the Q-factors (converted from BER results) for
all five 1-Tb/s channels and for different transmission distances.
Pre-SD-FEC BERs are shown for 2,000, 2,400, 2,800, and
3,200 km, while the post-SD-FEC BERs for LDPC and LDPCC
are only shown at 3,200 km. The correction capabilities of the
LDPC code (open diamonds) are insufficient and the Q-factor
therefore does not reach the HD-FEC threshold of 3.8 ×10−3.
On the other hand, the LDPCC code (open squares) success-
fully provides a Q-factor for all channels above the G.975.1.I.3
HD-FEC threshold of 3.1 ×10−3. Hence, the LDPCC imple-
mentation is capable of providing error-free transmission for up
to 3,200 km.
V. CONCLUSION
We reviewed recent high-speed all-electronically multi-
plexed digital coherent transmission experiments at 80 and
107 Gbaud. The reported Terabit dual-carrier interface uses
all-electronically multiplexed 80-Gbaud PDM-16-QAM sub-
carriers, capable of transporting ten 100-Gb/s tributaries within
one WDM channel over 3,200 km at a spectral efficiency of
5.2 b/s/Hz. This result is achieved through the use of high speed
electronic components in combination with an optical equalizer
at the transmitter and receiver and through the implementation
of advanced SD-FEC. The reported single carrier 400-Gb/s in-
terface is based on 107-Gbaud PDM-QPSK and transmission
over 4,800 km using state-of-the-art HD-FEC is demonstrated.
REFERENCES
[1] C. R. S. Fludger, T. Duthel, D. van den Borne, C. Schulien, E-D. Schmidt,
T. Wuth, E. de Man, G. D. Khoe, and H. de Waardt, “10 ×111 Gbit/s,
50 GHz spaced, POLMUX-RZ-DQPSK transmission over 2375 km em-
ploying coherent equalization,” presented at the Optical Fiber Communi-
cation Conf., Anaheim, CA, USA, 2007, Paper PDP22.
[2] K. Roberts, D. Beckett,D. Boertjes, J. Berthold, and C. Laperle, “100G and
Beyond with digital coherent signal processing,” IEEE Commun. Mag.,
vol. 48, no. 7, pp. 62–69, Jul. 2010.
[3] (2013). [Online]. Available: http://www3.alcatel-lucent.com/features/
100 g_era/
[4] L. E. Nelson, G. Zhang, M. Birk, C. Skolnick, R. Isaac, Y. Pan,
C. Rasmussen, G. Pendock, and B. Mikkelsen, “A robust real-time 100G
transceiver with soft-decision forward error correction,” J. Opt. Commun.,
vol. 4, pp. 131–140, 2012.
[5] P. J. Winzer, A. H. Gnauck, S. Chandrasekhar, S. Draving, J. Evangelista,
and B. Zhu, “Generation and 1,200-km transmission of 448-Gb/s ETDM
56-Gbaud PDM 16-QAM using a Single I/Q Modulator,” in Proc. Eur.
Conf. Opt. Commun., 2010, Paper PDP2.2.
[6] F. Buchali, K. Schuh, D. R¨
osener, E. Lach, R. Dischler, W. Idler,
L. Schmalen, A. Leven, R. Peter Braun, A. Ehrhardt, C. Gerlach, and
L. Sch¨
urer, “512-Gb/s DP-16-QAM field trial over 734 km installed SSMF
with Co-Propagating 10 Gb/s NRZ neighbors incorporating soft-FEC de-
coding,” in Proc. Opt. Fiber Commun. Conf, 2012, Paper OW4 C.4.
[7] (2013). [Online]. Available:http://www3.alcatel-lucent.com/400 g-pse/
[8] G. Raybon, S. Randel, A. Adamiecki, P. J. Winzer, L. Salamanca,
R. Urbanke, S. Chandrasekhar, A. Konczykowska, F. Jorge, J.-Y. Dupuy,
L. L. Buhl, S. Draving, M. Grove, and K. Rush, “1-Tb/s dual-carrier
80-GBaud PDM-16QAM WDM transmission at 5.2 b/s/Hz over
3200 km,” in Proc. IEEE Photon. Conf., 2012, Paper PD1.2.
[9] S. Chandrasekhar and X. Liu, “OFDM based superchannel transmission
technology,” J.n. Lightw. Technol., vol. 30, pp. 3816–3823, 2012.
[10] R. Freund, M. N¨
olle, C. Schmidt-Langhorst, R. Ludwig, C. Schubert, G.
Bosco, A. Carena, P. Poggiolini, L. Oxenløwe, M. Galili, H. C. Hansen
Mulvad, M. Winter, D. Hillerkuss, R. Schmogrow, W. Freude, J. Leuthold,
A. D. Ellis, F. C. Garcia Gunning, J. Zhao, P. Frascella, S. K. Ibrahim, and
N. Mac Suibhne, “Single- and multi-carrier techniques to build up Tb/s
per channel transmission systems,” in Proc. Int. Conf. Transparent Opt.
Netw., 2010, Paper Tu.D1.4.
[11] M. Feuer, L. E. Nelson, X. Zhou, S. L. Woodward, R. Isaac, B. Zhu, T. F.
Taunay, M. Fishteyn, J. Michael Fini, and M. F. Yan, “Joint digital signal
processing receivers for spatial superchannels,” Photon. Technol. Lett.,
vol. 24, pp. 1957–1960, 2012.
[12] S. Randel, “Space-division multiplexed transmission,” in Proc. Opt. Fiber
Commun. Conf., 2012, Paper OW4 F.1.
[13] Chandrasekhar, X. Liu, B. Zhu, and D. Peckham, “Transmission of a
1.2-Tb/s 24-carrier no-guard-interval coherent OFDM superchannel over
7200-km of ultra-large area fiber,” in Proc. Eur. Conf. Opt. Commun.,
2009, Paper PD2.6.
[14] X. Liu, S. Chandrasekhar, P. J. Winzer, T. Lotz, J. Carlson, J. Yang,
G. Cheren, and S. Zederbaum, “1.5-Tb/s guard-banded superchannel
transmission over 56×100-km (5600-km) ULAF using 30-Gbaud pilot-
free OFDM-16QAM signals with 5.75-b/s/Hz net spectral efficiency,”
presented at the European Conf. Optical Communication, Amsterdam,
The Netherlands, 2012, Paper PD 3.C.5.
[15] J. Renaudier, O. Bertran-Pardo, H. Mardoyan, P. Tran, G. Charlet,
S. Bigo, A. Konczykowska, J.-Y. Dupuy, F. Jorge, M. Riet, and J. Godin,
“Spectrally efficient long-haul transmission of 22-Tb/s using 40-Gbaud
PDM-16QAM with coherent detection,” in Proc. Opt. Fiber Commun.
Conf., 2012, Paper OW4 C.2.
[16] P. J. Winzer, “High-spectral-efficiency optical modulation formats,” J.
Lightw. Technol., vol. 30, no. 24, pp. 3824–3835, 2012.
[17] R. H. Walden, “Analog-to-digital converter survey and analysis,” IEEE J.
Sel. Areas Commun., vol. 17, no. 4, pp. 539–540, Apr. 1999.
[18] G. Raybon, A. Adamiecki, P. J. Winzer, M. Montoliu, S. Randel, A.
Umbach, M. Margraf, J. Stephan, S. Draving, M. Grove,and K. Rush, “All-
ETDM 107-Gbaud PDM-16QAM (856-Gb/s) transmitter and coherent
receiver,” in Proc. Eur. Conf. Opt. Commun., 2013, Paper PD 2.D.3.
RAYBON et al.: HIGH SYMBOL RATE COHERENT OPTICAL TRANSMISSION SYSTEMS: 80 AND 107 Gbaud 831
[19] A. H. Gnauck, J. E. Bowers, and J. C. Campbell, “8 Gbit/s transmission
over 30 Km of optical fobre,” Electron. Lett., vol. 22, pp. 600–602, 1986.
[20] P. J. Winzer, G. Raybon, and M. Duelk, “107-Gb/s optical ETDM trans-
mitter for 100G Ethernet transport,” presented at the European. Conf.
Optical Communication Conf., Glasgow, U.K., 2005, Paper Th4.1.1.
[21] R. S. Tucker, G. Eisenstein, and S. K. Korotky, “Optical time-division
multiplexing for very high bit-rate transmission,” J. Lightw. Technol.,
vol. 6, pp. 1737–1749, 1988.
[22] R. H. Derksen, G. Lehmann, C.-J. Weiske, C. Schubert, R. Ludwig,
S. Ferber, C. Schmidt-Langhorst, M. Moller, and J. Lutz, “Integrated
100 Gbit/s ETDM receiver in a transmission experiment over 480 km
DMF,” in Proc. Opt. Fiber Commun. Conf., 2006, Paper PDP37.
[23] J. H. Sinsky, A. L. Adamiecki, L. L. Buhl, G. Raybon, P. J. Winzer,
O. Wohlgemuth, M. Duelk, C. R. Doerr, A. Umbach, and H. G. Bach,
“107-Gbit/s Opto-electronic receiver with hybrid integrated photodetec-
tor and demultiplexer,” in Proc. Opt. Fiber Commun. Conf., 2007, Paper
PDP30.
[24] K. Schuh, E. Lach, and B. Junginger, “100 Gbit/s ETDM transmission
system based on electronic multiplexing transmitter and demultiplexing
receiver,” in Proc. Eur. Conf. Opt. Commun., 2006, Paper We3.P.124.
[25] S. L. Jansen, R. H. Derksen, C. Schubert, X. Zhou, M. Birk, C.- J. Weiske,
M. Bohn, D. van den Borne, P. M. Krummrich, M. M ¨
oller, F. Horst,
B.-J. Offrein, H. de Waardt, G.-D. Khoe, and A. Kirst¨
adter, “107-Gb/s
full-ETDM transmission over field installed fiber using vestigial side-
band modulation,” in Proc. Opt. Fiber Commun. Conf., Mar. 2007, Paper
OWE3.
[26] M. Daikoku, I. Morita, H. Taga, H. Tanaka, T. Kawanishi, T. Sakamoto,
T. Miyazaki, and T. Fujita, “100 Gbit/s DQPSK transmission experiment
without OTDM for 100G Ethernet Transport,” in Proc. Opt. Fiber Com-
mun. Conf., 2006, Paper PDP36.
[27] P. J. Winzer, G. Raybon, H. Song, A. L. Adamiecki, S. Corteselli, A. H.
Gnauck, D. A. Fishman, C. R. Doerr, S. Chandrasekhar, L. L. Buhl,
T. J. Xia, G. Wellbrock, W. Lee, B. Basch, T. Kawanishi, K. Higuma,
and Y. Painchaudet, “100-Gb/s DQPSK transmission: From laboratory
experiments to field trials,” J. Lightw. Technol., vol. 26, no. 20, pp. 3388–
3402, Oct. 2008.
[28] (2013). [Online]. Available: http://www.fujitsu.com/emea/services/
microelectronics/dataconverters/chais/
[29] A. Ko ´
nczykowska, J.-Y. Dupuy, F. Jorge, M. Riet, J. Moulu, V. Nodjiadjim,
P. Berdaguer, and J. Godin, “42 Gbaud 3-bit power-DAC for optical com-
munications with advanced modulation formats in InP DHBT,” IET Elec-
tron. Lett., vol. 47, no. 6, pp. 389–390, 2011.
[30] P. J. Pupalaikis and M. Schnecker, “A 30 GHz bandwidth, 80 GS/s sample
rate real-time waveform digitizing system,” in Proc. Opt. Fiber Commun.
Conf., 2010, Paper JThA52.
[31] A. H. Gnauck, P. J. Winzer, G. Raybon, M. Schnecker, and P. J. Pupalaikis,
“10 224-Gb/s WDM transmission of 56-Gbaud PDM-QPSK signals over
1890 km of Fiber,” Photon. Technol. Lett., vol. 22, no. 13, pp. 954–956,
2010.
[32] G. Raybon, A. Adamiecki, P. J. Winzer, C. Xie, A. Konczykowska,
F.Jorge, J.-Y. Dupuy, L. L. Buhl, S. Chandrasekhar, S. Draving, M. Grove,
K. Rush, B. Zhu, and D. W. Peckham, “All-ETDM 107-Gbaud (214-Gb/s)
single-polarization QPSK transmitter and coherent receiver,” in Proc. Eur.
Conf. Opt. Commun., 2012, Paper We3 A.2.
[33] G. Raybon, P. J. Winzer, A. Adamiecki, A. H. Gnauck, A. Konczykowska,
F. Jorge, J.- Y. Dupuy, L. L. Buhl, C. R. Doerr, R. Delbue, and
P. J. Pupalaikis, “All-ETDM 80-Gbaud (160-Gb/s) QPSK generation and
coherent detection,” Photon. Technol. Lett., vol. 23, no. 22, pp. 1667–1669,
2011.
[34] G. Raybon, S. Randel, A. Adamiecki, P. J. Winzer, L. Salamanca,
R. Urbanke, S. Chandrasekhar, A. Konczykowska, F. Jorge, J.-Y. Dupuy,
L. L. Buhl, S. Draving, M. Grove, and K. Rush, “All-ETDM 80-Gbaud
(640-Gb/s) PDM 16-QAM generation and coherent detection,” Photon.
Technol. Lett., vol. 24, pp. 1328–1330, 2012.
[35] G. Raybon, A. Adamiecki, P. J. Winzer, C. Xie, A. Konczykowska,
F.Jorge, J.-Y. Dupuy, L. L. Buhl, S. Chandrasekhar, S. Draving, M. Grove,
and K. Rush, “Single-carrier 400G interface and 10-channel WDM trans-
mission over 4800 km using all-ETDM 107-Gbaud PDM-QPSK,” in Proc.
Opt. Fiber Commun.Conf, 2013, Paper PDP5 A.5.
[36] J. Godin, V. Nodjiadjim, M. Riet, P. Berdaguer, O. Drisse, E. Derouin,
A. Konczykowska, J. Moulu, J.-Y. Dupuy, F. Jorge, J.-L. Gentner, and A.
Scavennec, in Proc. Compound Semiconduc. Integ. Circuit Symp., Mon-
terey, CA, USA, 2008, pp. 109–112.
[37] C. R. Doerr, S. Chandrasekhar, P. J. Winzer, A. R. Chraplyvy, A. H.
Gnauck, L. W. Stulz, R. Pafchek, E. Burrows, “Simple Multichannel Opti-
cal Equalizer Mitigating Intersymbol Interference for 40-Gb/s Nonreturn-
To-Zero Signals,” J. Lightw. Technol., vol. 22, pp. 249–256, 2004.
[38] P. J. Winzer, A. H. Gnauck, C. R. Doerr, M. Magarini, and L. L. Buhl,
“Spectrally efficient long-haul optical networking using 112-Gb/s
polarization-multiplexed 16-QAM,” J. Lightw. Technol., vol. 28, pp. 547–
556, 2010.
[39] S. J. Savory, “Digital filters for coherent optical receivers,” Opt. Exp.,
vol. 16, no. 2, pp. 804–817, 2008.
[40] A. Leven, N. Kaneda, A. Klein, U.-V. Koc, and Y.-K. Chen, “Real-time
implementation of 4.4 Gbit/s QPSK intradyne receiver using field pro-
grammable gate array,” Electron. Lett., vol. 42, no. 24, pp. 1421–1422,
2006.
[41] A. Sierra, S. Randel, P. J. Winzer, R. Ryf, A. H. Gnauck, and
R.-J. Essiambre, “On the use of delay-decorrelated I/Q test sequences for
QPSK and QAM signals,” Photon. Technol. Lett., vol. 24, pp. 1000–1003,
2012.
[42] ITU-T Rec. G.975.1 Appendix I.9, 2004.
[43] G. Raybon, P. J. Winzer, A Adamiecki, A. H. Gnauck, A. Konczykowska,
F. Jorge, J. Y. Dupuy, A. Sureka, C. Scholz, R. Delbue, P. J. Pupalaikis,
L. L. Buhl, C. R. Doerr, S. Chandrasekhar, B. Zhu, and D. W. Peckham,
“8 ×320-Gb/s transmission over 5600 km using all-ETDM 80-Gbaud po-
larization multiplexed QPSK transmitter and coherent receiver,” in Proc.
Opt. Fiber Commun. Conf., 2012, Paper OTu2 A.1.
[44] S. Randel, R. Ryf, A. Sierra, P. J. Winzer, A. H. Gnauck, C. A. Bolle,
R.-J. Essiambre, D. W. Peckham, A. McCurdy, and R. Lingle, “6 ×56-
Gb/s mode-division multiplexed transmission over 33-km few-mode fiber
enabled by 6 ×6 MIMO equalization,” Opt. Exp., vol. 19, pp. 16697–
16707, 2011.
[45] T. Pfau, S. Hoffmann, and R. No´
e, “Hardware-efficient coherent digital
receiver concept with feed forward carrier recovery for M-QAM constel-
lations,” J. Lightw. Technol., vol. 27, no. 8, pp. 989–999, 2009.
[46] I. Djordjevic, S. Sankaranarayanan, S. Chilappagari, and B. Vasic, “Low-
density parity-check codes for 40-Gb/s optical transmission systems,”
IEEE J. Sel. Topics Quantum Electron., vol. 12, no. 4, pp. 555–562, Jul.
2006.
[47] X.-Y. Hu, E. Eleftheriou, and D. M. Arnold, “Regular and irregular pro-
gressive edge-growth tanner graphs,” IEEE Trans. Inform. Theory, vol. 51,
no. 1, pp. 386–398, Jan. 2005.
[48] M. Lentmaier, G. P. Fettweis, K. S. Zigangirov, and D. J. Costello, “Ap-
proaching capacity with asymptotically regular LDPC codes,” in Proc.
IEEE Inf. Theory Appl. Workshop, 2009, pp. 173–177.
[49] A. R. Iyengar, M. Papaleo, P. H. Siegel, J. K. Wolf, A. Vanelli-Coralli, and
G. E. Corazza, “Windowed decoding of protograph-based LDPC convo-
lutional codes over erasure channels,” IEEE Trans. Inf. Theory, vol. 58,
no. 4, pp. 2303–2320, 2012.
Authors’ biographies not available at the time of publication.
