![RSNR [dB] at BER = 10 −3 vs. the baud rate. Black curve shows the case without DPD and red with. Here we analyzed 200Gb/s 16QAM with a DAC nominal-3dB BW of 20.1 GHz and ENob = 5.5.](https://www.researchgate.net/profile/Antonio-Napoli/publication/268195919/figure/fig5/AS:668771534524420@1536458967575/RSNR-dB-at-BER-10-3-vs-the-baud-rate-Black-curve-shows-the-case-without-DPD-and_Q320.jpg)
Content uploaded by Antonio Napoli
Author content
All content in this area was uploaded by Antonio Napoli on Mar 18, 2015
Content may be subject to copyright.
On the next generation bandwidth variable
transponders for future flexible optical systems
Antonio Napoli∗, Markus N¨
olle†, Danish Rafique∗, Johannes K. Fischer†, Bernhard Spinnler∗,
Talha Rahman‡, Mahdi M. Mezghanni§, and Marc Bohn∗
∗Coriant R&D GmbH, Munich, Germany, antonio.napoli@coriant.com
†Fraunhofer Institute, Heinrich Hertz Institute, Einsteinufer 37, Berlin, Germany, markus.noelle@hhi.fraunhofer.de
‡Eindhoven University of Technology, Eindhoven, The Netherlands, t.rahman@tue.nl
§Technische Universitat Munchen, Munich, Germany, mahdi.mezghanni@googlemail.com
Abstract—Elastic optical networks represent one of the most
promising candidates for the imminent upgrade of current fixed-
grid based optical systems. This novel architecture based on high
efficient spectrum allocation (i.e. higher capacity), is necessary to
cope with foreseen future exponential increase of Internet traffic.
This work discusses the influence of hardware components on
the transmission performance of recently proposed bandwidth
variable transponders employing digital signal processing algo-
rithms. Both are key elements for the realization of future elastic
optical networks, and only their successful interplay with coherent
detection can enable the transmission of different modulation
formats at variable symbol and data-rates over flexible links.
We evaluate the performance of such transponders when different
modulation schemes are generated employing ideal and realistic
values for some key hardware components. Finally, we briefly
present a couple of examples of mitigation techniques: namely
digital pre-distortion and digital back-propagation.
I. INTRODUCTION
During the last decade telecommunication networks wit-
nessed an unprecedented exponential growth of bandwidth
demand [1]. The roots are to be found in the explosion
of bandwidth-hungry web-based applications such as video
streaming and cloud storage. Moreover, high definition mul-
timedia contents will soon stress even further current optical
backhauls. If these forecasts come true, optical networks will
certainly saturate their bandwidth capability, heading toward a
capacity crunch as predicted in [2]. To cope with this imminent
scenario several strategies to increase the spectral efficiency
×maximum length product (SE×L) have been investigated
and among them it is worth to mention the following three:
(a) the development of high-performing fibers and hybrid
amplification schemes; (b) multi-mode or -core propagation;
and finally (c) flexible optical systems.
There is common agreement that (c) is the preferred approach
because: (I) it can be almost immediately deployed; (II) it
optimizes the optical spectrum allocation; and (III) its overall
costs are affordable since only two network elements, the
transponder and the optical node, need to be replaced.
The idea of flexibility stems from mobile and microwave
communications, and recently has been proposed for optics to
efficiently utilize the available optical spectrum. The two novel
elements, namely an advanced reconfigurable optical add-drop
multiplexer (ROADM) and a bandwidth variable transponder
(BVT), should possess the characteristics described in [3]. For
example, the node must be capable to address the current dy-
namic traffic by employing novel solutions such as architecture
on demand [4] and switchless elastic rate node [5]; while the
transponder is designed to enhance the flexibility in terms
of modulation formats, data-rate and transmission medium
scenarios [6]. This article mainly focuses on the influence of
realistic hardware components (e.g. digital-to-analog converter
(DAC) and modulator) on the physical layer performance of
a BVT. Such a transponder cannot operate without advanced
digital signal processing (DSP) techniques (blind [7] or data-
aided [8]), developed to compensate for transponder inherited
degradation and to mitigate fiber propagation impairments [9].
The paper is structured as follows: Sec. II introduces the
high-level description of a BVT, explaining the technical
motivations for its implementation; Sec. III reports a first pre-
selection of the most suitable modulation formats for given
transmission applications; Sec. IV assesses the dependence
of performance on components’ quality providing recommen-
dations on their characteristics according to the scenarios of
interest; finally Sec. V shows two examples of impairments
mitigation, namely digital pre-distortion (DPD) of a bandwidth
limited high-speed DAC board and compensation of fiber non-
linear propagation effects (e.g. self-phase modulation) through
digital back-propagation [10]. Sec. VI draws the conclusions.
All analyses are carried out for Nyquist WDM only and
propagating dual polarization signals.
Fig. 1. Block diagram of a BVT: upper part (transmitter), lower part
(receiver).
II. NE XT G EN ERATI ON BVTS
Next generation transponders will operate through flexible
signaling (in terms of modulation formats, symbol rate (RB)
and code) to be propagated over different distances and fiber
types. These functionalities must be realized by utilizing iden-
tical hardware components and transparent DSP algorithms.
Fig. 1 displays a general high-level block diagram of a BVT
as proposed in [6], where:
•The upper part describes the transmitter, consisting
of two main blocks: an application-specific integrated
circuit (ASIC) (with a logical unit plus the digital
pre-distortion and a DAC) followed by two hardware
modules: the radio-frequency power amplifier (RF-PA)
and the Mach-Zehnder modulator (MZM).
•The lower part presents the optical front-end (with a
polarization beam splitter (PBS), an optical 90 degree
hybrid, a local oscillator and a balanced photo-detector
(BPD)). After the analog-to-digital converter (ADC),
the receiver is followed by a second ASIC to mitigate
for channel and components’ distortions. The BVT
ends with the de-mapping and decoding.
In general, a BVT must be able to: (I) generate a large
set of modulation formats spanning from binary phase shift
keying (BPSK) up to (at least) 16-quadrature amplitude
modulation (QAM); (II) vary data-, symbol-, and code-rate
according to actual traffic conditions and (III) propagate over
different networks and distances. For example, a point-to-
point connection at high data- rate over a short distance is
optimized by employing high-order modulation formats (i.e.
maximizing the spectral efficiency), or, on the contrary, the
need for a high-capacity transoceanic link, can be performed
by robust modulation formats such as BPSK, but with lower
spectral efficiency. Starting from these assumptions, it is clear
that a BVT requires high-quality components to satisfy reach
requirements and advanced DSP to mitigate fiber propagation
effects.
III. PRE -SELECTION OF MODULATION FORMATS FOR
BANDWIDTH VARIABLE TRANSPONDERS
This section aims at pre-selecting the most suitable modu-
lation schemes for future BVTs operating in mixed networks
scenarios, from regional to long-haul size. To do that, we
evaluate the system performance of the modulation formats
listed in Table I by considering as discriminating factor the
maximum reachable distance in km, calculated as the in-
tersection between required and available optical signal to
noise ratio (OSNR) over 0.1 nm. The forward error correction
(FEC) overhead was set at 15% for a BER = 10−3. The
maximum reach (Lmax) estimation was analytically based on
the Gaussian noise (GN) model described in [11]. Being 1 Tb/s
our most challenging target for future BVTs, we consider this
data-rate only.
The pre-selection is carried out based on the results of Table I
where: the second column defines the symbol rate in [GBaud];
the third the required OSNR [dB] in back-to-back (b2b); the
fourth identifies the number of transmitted sub-carriers for each
super-channel; and the fifth is the line-rate. Within this analysis
we assumed the physical values of standard single mode fiber
(SSMF) as reported in Table III for a span length equals to 70
km. (Note: LEAF stands for large effective area fiber and TWC
for true-wave classic). The total simulated system bandwidth
was 3 THz for an input power / channel of -2 dBm. The EDFAs
have a noise figure of 5 dB.
As expected, the transmission reach (last column of Table
I) decreases when increasing the modulation order, while
the required bandwidth decreases (i.e. the spectral efficiency
increases). Consequently, some of the options of Table I can
be already excluded because they do not satisfy the distance
requirements of a typical European network size (e.g. 64QAM
and 128QAM reach only 320 km and 80 km respectively) or
because their spectral efficiency is limited (e.g. BPSK). We
TABLE I. CONSIDERED REALIZATION OPTIONS FOR 1 TB/S.
Mod. format RBROSNR carriers line-rate Lmax
[GBaud] [dB] [Tb/s] [km]
BPSK 30 10.6 20 1.2 10960
QPSK 30 13.6 10 1.2 5520
8QAM 30 17.2 7 1.26 2400
16QAM 30 20.4 5 1.2 1120
32QAM 30 23.4 4 1.2 560
64QAM 30 26.4 4 1.44 320
128QAM 18 27.1 5 1.26 80
conclude, that only QPSK, 8QAM, and 16QAM are suitable
for our purpose and these modulation formats will be evaluated
in the next section also when transmitting at lower data-rates
(e.g. 100 Gb/s), in order to explore all possible cases of interest
for a BVTs.
TABLE II. COMPONENT VALUES USED THROUGHOUT THE
NU MER IC AL L INK S IM UL ATIO NS
Component ideal available in 3/4 years
DAC / ADC resolution (bits) ideal 5.5 8
DAC / ADC bandwidth at −3dB [GHz] ideal 0.4·RB0.7·RB
MZM extinction Ratio [dB] ∞24 30
LASER line-width [kHz] 0 100 50
TABLE III. PH YSI CA L VALUE S FO R TH E OPT IC AL FI BER S UN DE R TES T.
Parameters SSMF LEAF TWC
Attenuation α[dB/km] 0.21 0.225 0.225
Dispersion Coefficient Dat 1550 nm [ps/(nm·km)] 16.8 4.2 2.8
Dispersion Slope S [ps/nm2/km] 0.058 0.086 0.068
Nonlinear Coefficient γ[1/(W·km)] 1.14 1.3 2
IV. INFLUENCE OF COMPONENT QUALITY ON REACH
PERFORMANCE AND NYQUIST DWDM
The quality of hardware components is of fundamental
importance to fulfill the requirements needed by next genera-
tion optical networks. In this section we numerically evaluate
the b2b and link performance for the modulation formats
selected in previous Sec. III, when realistic values for some
key components are considered as summarized in Table II.
The chosen parameters identify devices such as DAC, MZM,
and LASER. For example: the DAC bandwidth limitation at
−3dB and a realistic resolution, together with the quality of
the employed LASERs, in terms of line-width and stability,
will significantly reduce the maximum reachable distance.
Last but not the least, this limitation grows as the order of
the modulation format increases. One of the goals of this
section is to establish the required values to achieve the pre-
defined distance targets. The analysis is based on following
four scenarios:
•assessment of b2b performance in terms of ROSNR
•evaluation of theoretical Lmax by using the GN model
•evaluation of link performance in case of ideal compo-
nents (only fiber propagation effects are considered).
•link simulations for available (i.e. current commercial
values) and realistic in 3-4 years.
TABLE IV. N UM ER IC ALLY E VALUATE D B2B PERFORMANCE OF ALL
INVESTIGATED CONFIGURATIONS.
Gross RB∆f∆P∆P∆P
Mod. format & rate [GBaud] [GHz] [dB] [dB] [dB]
ideal now 3/4 yrs
100 Gb/s QPSK 28.75 33.06 0.2 0.7 0.3
200 Gb/s 16QAM 28.75 33.06 0.8 2 1.2
200 Gb/s 8QAM 38.34 44.09 0.6 1.3 0.7
400 Gb/s 2x200 16QAM 28.75 33.06 0.8 2 1.2
400 Gb/s 2x200 8QAM 38.34 44.09 0.6 1.3 0.7
400 Gb/s 4x100 QPSK 28.75 33.06 0.2 0.7 0.3
1Tb/s 4x250 16QAM 35.94 41.33 0.9 2 1.1
1Tb/s 5x200 8QAM 38.34 44.09 0.6 1.3 0.7
1Tb/s 10x100 QPSK 28.75 33.06 0.2 0.7 0.3
Back-to-back: The b2b performance is reported in Table IV.
Fig. 2. Optical spectra of 40×200 Gb/s 16QAM system with channel spacing
equal to the RB.
Where the second column indicates the gross RBand the third
the spacing between the channels. The results are contained in
the last three columns, in terms of OSNR penalty (∆P[dB])
at BER = 10−3with respect to the theoretical values as in [12].
The comparison is carried out in case of an AWGN channel
and matched filtering for all configurations and component
scenarios respectively. The DSP employed throughout this
section makes use of the aforementioned commercial DSP
receivers as in [13] and it consists of the re-sampling of the
signal, the correction for imperfections of the optical coherent
frontend [14], and an optional compensation of the CD, an
FFT-based correction of the offset frequency between signal
and local oscillator LASER [15], an adaptive, blind equalizer
stage for polarization separation and filtering and a carrier-
phase recovery for correction of the remaining phase noise of
the LASERs [16].
It is clear that a ROSNR penalty ∈[0.2,0.9] [dB], depending
on the modulation format, is already present in case of ideal
system components. When assuming the specifications of
available components, the penalty grows up to at most 2 dB. In
case we consider possible values that these components might
attain in 4 yrs-time from now, the ROSNR penalty decreases
considerably, and it approaches ideal conditions.
Link: In terms of maximum reach we carried out extensive
numerical simulations by propagating the signal with the same
assumptions of Sec. III. Due to the different constellation
Fig. 3. Lmax at BER <10−3as a function of the fiber input power /
channel for a 200 Gb/s 16QAM system.
orders, we assess the performance by evaluating SE×L. During
this analysis only standard DSP algorithms are used [7] (i.e. no
pre-distortion or nonlinear mitigation). To establish the influ-
ence of the key hardware components on the link performance
we propagated an ultra-dense WDM (UDWDM) signal when
considering the following two setups: (I) channel spacing equal
to 1.15·RBand a Nyquist pulse shape with a roll-off (RO) =
0.2; and (II) a channel spacing = RBand a RO = 0.01. Fig.
2 displays exemplary transmitted spectra of the transmitted
signal for the system configuration of a 40×200 Gb/s 16QAM
UDWDM with a channel spacing = RBand RO = 0.01. The
bandwidth limitation of the individual transmitters in case of
realistic components is clearly visible moving from the upper
spectrum to the lower one. Fig. 3 reports an example of results
showing the maximum transmittable distance as function of
the optical power at BER = 10−3with a channel spacing =
1.15·RBand RO = 0.2.
In Fig. 3 the black curve shows the analytical results obtained
Fig. 4. Lmax at BER <10−3as a function of the fiber input power /
channel for a 200 Gb/s 16QAM system, open symbols represent a channel
spacing of 1.15·RB, while solid symbols are systems with a channel spacing
=RB.
by applying the GN model, while the blue by considering ideal
components, which proofs the correctness of such model. Note:
please consider that the numerically evaluated ROSNR penalty
for the b2b case is already included in this graph. Further,
including realistic components in this analysis we obtain the
red curve, which presents a considerable reduction in terms
of maximum transmission reach. It should be noted, that the
reduction in transmission reach is almost exclusively caused
by the linear penalty as already seen in the b2b simulations.
Lastly, the green curve clearly shows that if the hardware
components will approach the expected characteristics, per-
formance comparable to the ideal ones can be achieved for
the considered 1 Tb/s transmission. Summarizing: The trans-
TABLE V. M AX IM UM N ET SE×L [B/S/H Z·KM ].
BWmin SE×L SE×L SE×L
Mod. format & rate [GHz]
ideal now 3/4 yrs
100 Gb/s QPSK 33.1 20325 18420 20325
200 Gb/s 16QAM 33.1 8892 6352 8469
200 Gb/s 8QAM 44.1 13974 12068 13656
400 Gb/s 2x200 16QAM 66.1 8892 6352 8469
400 Gb/s 2x200 8QAM 88.2 13974 12068 13656
400 Gb/s 4x100 QPSK 132.3 20325 18420 20325
1Tb/s 4x250 16QAM 165.3 8469 6352 8045
1Tb/s 5x200 8QAM 220.5 13974 12068 13656
1Tb/s 10x100 QPSK 330.6 20325 18420 20325
mission results reported in Table V represent the net spectral
efficiencies, meaning the net bit-rate divided by the channel
spacing. A first outcome of this study is that QPSK systems
show by far the highest SE×L due to the disproportionately
high transmission reaches. However, for targeted transmission
reaches around 2000 km to 3000 km, also 8QAM is attractive.
Please also note, that no joint DSP for the different sub-carriers
in a super-channel was applied (i.e. each sub-carrier has been
processed individually). The exploitation of the information
from neighboring channels within one super-channel for equal-
ization would probably lead to an increased performance of
these super-channel configurations. To conclude, the same
overall WDM bandwidth of 1.23 THz is assumed for all the
simulations and therefore the number of super-channels varies
according to the different modulation formats. This is also the
reason, for equal SE×L products of systems using the same
modutlation format and symbol rate (e.g. for 100 Gb/s QPSK,
400 Gb/s QPSK and 1 Tb/s QPSK). For these systems only
the number agragated subcarriers per superchannel differs.
Therefore, concidering the same overall WDM bandwidth and
not considering joint processing of mulitple subcarriers, the
SE×L is the same for all of these systems, independent of the
number of aggregated subcarriers.
When comparing the results of the systems with different
channel spacing, clearly the larger the channel spacing is, the
stronger is the system against nonlinear impairments. As an
example the results of the numerical simulations for 200 Gb/s
16QAM systems, considering the two different roll-off and
channel spacing, is reported in Fig. 4. The optimum fiber
input power, as well as the maximum transmission reach,
independent of the used components, increases for the systems
with larger channel spacing. However, the spectral efficiency
of course is reduced, so that this larger tolerance toward
nonlinearities does not pay off in terms of a higher spectral
efficiency times distance product in most cases (see Table V).
V. EXAMPLE OF MITIGATIONS OF DISTORTIONS
This final section focuses on the mitigation of system
impairments presenting two techniques: the first digitally pre-
distorts the DAC bandwidth limitation (DAC DPD) at the
transmitter; the second mitigates the intra-channel nonlinear
fiber effects (e.g. self-modulation effect (SPM)).
A. Example of digital pre-distortion for bandwidth limited
DACs
As showed in Fig. 1 we apply DPD at the transmitter
immediately before the DAC. Currently, it is possible to
transmit data-rates ≥400 Gb/s only by means of high-
order modulation formats (e.g. 8 or 16QAM) together with
multi-carrier transmission. Here the limitation produced by
the DAC bandwidth is a major problem because multi-level
signaling requires high quality hardware components in terms
of transmitter bandwidth and ENob at both sides of the BVT.
With respect to Table II, we consider the specific case of
resolution and bandwidth (BW−3dB ) limited DAC together
with several high-order modulation formats need, since DPD
may enable the utilization of such modulation schemes paving
the way to high-speed optical networks.
Although the importance of lost information cannot be recov-
ered, but only attenuated as showed in [17], the bandwidth
limitation can be partially recovered by using DPD just in front
of the DAC. In this analysis we placed the DPD as shown in
Fig. 1, and we determine its transfer function as
Pcurrent(f) = Ddesired (f)
DDAC(f) + c0
,(1)
where Pcurrent(f)is the DPD filter, Ddesired(f)is the desired
transfer function, and DDAC(f)is the DAC transfer function
(here assumed Gaussian). Finally c0is a constant heuristically
optimized.
The performance of the system employing a DPD are reported
in Fig. 5 and Fig. 6. The first reports on the usage of DPD to
improve the performance of a 16QAM with a DAC nominal
bandwidth at −3dB of 20.1 GHz and ENob = 5.5, reporting the
RSNR at BER = 10−3vs. the baud rate. Clearly, the case with
DPD outperforms the one without. In the second, we fixed the
RSNR penalty to 0.5 dB and we show the maximum allowable
baud rate for a set of modulation format still with ENob =
5.5. The results show, as one would expect, that QPSK is the
scheme with the larger benefit, being the Euclidean distance
of its constellation point the highest.
30 35 40 45 50 55 60 65
10
15
20
25
Baudrate [GBaud]
RSNR (@ 1e−3) [dB]
w/o DPD
w/ DPD
Fig. 5. RSNR [dB] at BER = 10−3vs. the baud rate. Black curve shows
the case without DPD and red with. Here we analyzed 200Gb/s 16QAM with
a DAC nominal -3dB BW of 20.1 GHz and ENob = 5.5.
B. Nonlinear mitigation for BVT over different fiber types
This last subsection briefly discusses the performance of
digital back-propagation (DBP), a highly investigated non-
linear method placed at the receiver as showed in Fig. 1.
The analysis considered high-order modulation and multi-
carrier transmission. More precisely, we carried our numerical
analyses for 400 Gb/s and 1 Tb/s, over the three fiber types
described in Table III. The performance is reported in Fig. 7,
QPSK 8−QAM 16−QAM 32−QAM 64−QAM
30
35
40
45
50
55
60
65
modulation format
Max Baud rate [GBaud]
w/o DPD
with DPD
Fig. 6. Maximum allowable baud rate given a RSNR penalty = 0.5 dB for
different modulation format with ENob = 5.5. Black curve shows the case
without DPD and red with.
where on the left it shows 400 Gb/s (2×200 Gb/s) 16QAM
and on the right 1 Tb/s (4x250 Gb/s) 16QAM. In all cases the
blu curves indicate the BVT with DBP on and red when it is
OFF [10]. The plots report the Lmax for a BER = 2·10−2,
employing a FEC with 20% overhead.
From our extensive simulations it is clear that DBP algorithm
provides a considerable improvement (∼15%) only at low
data-rate transmission (≤400 Gb/s), while its efficiency sig-
nificantly decreases at 1 Tb/s (∼9%). We conjuncture that this
is mainly due to the fact that DBP has been developed as in
[10], [18] so that it compensates only for SPM, leaving cross-
phase modulation uncompensated. We conclude that based on
the conservative assumptions made on the DBP algorithm (em-
ploying adaptive single-carrier DBP), a high benefit provided
by this algorithm is observed only for a low number of carrier
transmission (≤2), while the gain vanishes for high number
of carriers (e.g. 1 Tb/s). Moreover, in general the benefit, in
low data-rate regime, increases when moving from fiber with
low nonlinear coefficient towards ones with higher values [10].
This tendency is valid only till inter-channel penalties play a
little role (single-carrier transmission), in fact as soon as we
move towards higher data-rate, the efficiency of A-SCDBP
with high nonlinear fiber decreases with respect to the one
with SSMF.
−6 −4 −2 0 2 4
0
500
1000
1500
2000
2500
3000
3500
4000
Pin [dBm]
LMAX [km]
SSMF LEAF
TWC
−4 −2 0 2 4 6 8
0
500
1000
1500
2000
2500
3000
3500
4000
Pin [dBm]
LMAX [km]
SSMF
TWC
LEAF
Fig. 7. 400 Gb/s (2×200 Gb/s) 16QAM (left) and 1 Tb/s (4x250 Gb/s)
16QAM (right) over the same set of fibers as reported in Table III. Red circles
indicate case without A-SCDBP, while blue triangles are with A-SCDBP. The
average gain for 400 Gb/s 16QAM is of ∼15%, while for 1 Tb/s 16QAM is
∼9%.
VI. CONCLUSIONS
We extensively investigated the performance of BVTs
under different working conditions. Firstly, we carried out pre-
liminary analyses to determine the optimal modulation format,
for given scenarios. Secondly, we analyzed the influence of the
component quality, in case of Nyquist DWDM, showing that,
for the case of 1T b/s, reasonable predicted values in 4-years
time will lead to quasi ideal performance. Finally, we presented
two examples of distortion mitigations: namely the impact of
DAC bandwidth limitation and nonlinear propagation effects.
Herewith we report that the solely DAC bandwidth limitation
can be significantly improved by applying a linear filter and
that currently investigated nonlinear compensation techniques
are effective as long as intra-channel effects are dominant.
ACKNOWLEDGMENT
This work was supported by the European Union FP-7
IDEALIST project under grant agreement number 317999.
REFERENCES
[1] Amsterdam Data Exchange, [Online] http://www.ams-ix.net/.
[2] D. van den Borne, M. S. Alfiad, and S. L. Jansen, The Capacity Crunch
Challenge: How to Design the Next Generation of Ultra-High Capacity
Transmission Systems, Signal Processing in Photonic Communications,
2010.
[3] M. Bohn et al., Elastic Optical Networks: the Vision of the ICT Project
IDEALIST, in Proceedings of Future Network & Mobile Summit (FU-
NEMS) 2013.
[4] N. Amaya, G. Zervas, and D. Simeonidou, Architecture on Demand for
Transparent Optical Networks, in Proceedings of ICTON, 2011.
[5] A. Stavdas et al., A Novel Architecture for Highly Virtualised Software-
Defined Optical Clouds, ECOC, 2013.
[6] J. K. Fischer, S. Alreesh, R. Elschner, F. Frey, M. Nolle, C. Schubert,
Bandwidth-Variable Transceivers Based on 4D Modulation Formats for
Future Flexible Networks, in Proceedings of ECOC, 2013.
[7] M. Kuschnerov, F. N. Hauske, K. Piyawanno, B. Spinnler, M. S. Alfiad,
A. Napoli, and B. Lankl, DSP for coherent single-carrier receivers, IEEE
/ OSA Lightwave Technology, Journal of, 2009.
[8] M. Kuschnerov, M. Chouayakh, K. Piyawanno, B. Spinnler, E. De Man,
P. Kainzmaier, M. S. Alfiad, A. Napoli, and B. Lankl, Data-aided versus
blind single-carrier coherent receivers, Photonics Journal, IEEE, 2010
[9] A. Napoli, Z. Maalej, V. Sleiffer, M. Kuschnerov, D. Rafique, E.
Timmers, B. Spinnler, T. Rahman, L. Coelho, and N. Hanik, Reduced
Complexity Digital Back-Propagation Methods for Optical Communica-
tion Systems, IEEE / OSA Journal of Lightwave Technology, 2013.
[10] A. Napoli, D. Rafique, B. Spinnler, M. Kuschnerov, M. Noelle, and
M. Bohn, Performance Dependence of Single-Carrier Digital Back-
Propagation on Fiber Types and Data Rates, in Proceedings of OFC
2014
[11] P. Poggiolini, A. Carena, V. Curri, G. Bosco, F. Forghieri, Analytical
Modeling of Non-Linear Propagation in Uncompensated Optical Trans-
mission Links, IEEE PTL, Vol. 23, No. 11, June 1, 2011.
[12] J. Proakis, M. Salehi, Digital communication, 5th edition, McGraw-Hill
Science, 2007.
[13] Seb J. Savory, Digital coherent optical receivers: algorithms and
subsystems, Selected Topics in Quantum Electronics, IEEE Journal of,
2010.
[14] I. Fatadin, S. J. Savory, and D. Ives Compensation of quadrature
imbalance in an optical QPSK coherent receiver, Photonics Technology
Letters, IEEE, 2008.
[15] M. Selmi, Y. Jaouen, and P. Ciblat, Accurate digital frequency offset
estimator for coherent PolMux QAM transmission systems, ECOC, 2009
[16] X. Zhou, An Improved Feed-Forward Carrier Recovery Algorithm for
Coherent Receivers With M-QAM Modulation Format, IEEE PTL, 2010.
[17] M. N¨
olle, F. Frey, R. Elschner, C. Schmidt-Langhorst, A. Napoli, and
C. Schubert, Performance Comparison of Different 8QAM Constellations
for the Use in Flexible Optical Networks, in Proceedings of OFC 2014
[18] C.-Y. Lin, A. Napoli, B. Spinnler, V. Sleiffer, D. Rafique, M.
Kuschnerov, M. Bohn, and B. Schmauss, Adaptive Digital Back-
Propagation for Optical Communication Systems, in Proceedings of OFC
2014
