General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright
owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
You may not further distribute the material or use it for any profit-making activity or commercial gain
You may freely distribute the URL identifying the publication in the public portal
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
Downloaded from orbit.dtu.dk on: Aug 09, 2022
All-optical wavelength conversion and signal regeneration using an electroabsorption
modulator
Højfeldt, Sune; Bischoff, Svend; Mørk, Jesper
Published in:
Journal of Lightwave Technology
Link to article, DOI:
10.1109/50.857758
Publication date:
2000
Document Version
Publisher's PDF, also known as Version of record
Link back to DTU Orbit
Citation (APA):
Højfeldt, S., Bischoff, S., & Mørk, J. (2000). All-optical wavelength conversion and signal regeneration using an
electroabsorption modulator. Journal of Lightwave Technology, 18(8), 1121-1127.
https://doi.org/10.1109/50.857758
JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 8, AUGUST 2000 1121
All-Optical Wavelength Conversion and Signal
Regeneration Using an Electroabsorption Modulator
Sune Højfeldt, Svend Bischoff, and Jesper Mørk
Abstract—All-optical wavelength conversion and signal regen-
eration based on cross-absorption modulation in an InGaAsP
quantum well electroabsorption modulator (EAM) is studied
at different bit rates. We present theoretical results showing
wavelength conversion efficiency in agreement with existing
experimental results, and the signal regeneration capability of
the device is investigated. In particular, we demonstrate the
dependence of the extinction ratio of both the converted signal
and the control signal on the device length and on the power level
of the control signal. We also show how the sweep-out dynamics
influences the results.
Index Terms—All-optical wavelength conversion, electroabsorp-
tion modulator (EAM), optical communication, semiconductor de-
vices, signal regeneration, ultrafast signal processing.
I. INTRODUCTION
I
N the last decade, the electroabsorption modulator (EAM)
has found a wider and wider range of applications within
optical communications, although so far mostly at the research
level. Around the early 1990s, functionalities such as pulse gen-
eration [1]–[4] (which has now found commercial application in
integrated distributed feedback laser/EAM modules), demulti-
plexing [5] and signal regeneration [6] were demonstrated using
an EAM. These functionalities are important for constructing
ultrahigh-speed all-optical networks [7].
With the quite recent achievements of fast demultiplexing
from 100 to 10 Gb/s [8], widely tunable pulse generation using
an EAM [9], all-optical wavelength conversion and signal re-
generation [10], [11], the usefulness of the EAM as an integrated
component for signal processing has been further substantiated.
The EAM is also interesting because its range of operation is
quite large, as can be seen in, e.g., [6] where an on-off ratio of
20 dB is obtained from 1.53 to 1.57
m with reverse biases less
than 5 V.
The above applications show that the EAM is capable of pro-
viding high on–off ratios, either by electrical modulation of the
absorption (electroabsorption) or by optical means (cross-ab-
sorption modulation), which includes all absorption changes in-
duced by photo-generated carriers.
In this paper, we focus on the all-optical properties of the
EAM where a fixed dc bias is applied to adjust the band edge
as well as for ensuring a fast sweep-out of the photogenerated
carriers. All-optical wavelength conversion up to 40 Gb/s
Manuscript received December 7, 1999; revised May 2, 2000. This work
was supported by the Danish Technical Research Council (STVF) through the
SCOOP program.
The authors are with Research Center COM, Technical University of Den-
mark, Lyngby DK-2800, Denmark (e-mail: sh@com.dtu.dk).
Publisher Item Identifier S 0733-8724(00)06476-8.
was recently carried out, with good performance up to 20
Gb/s [10]–[12]. The all-optical wavelength conversion in an
EAM relies on the saturable absorption characteristic, through
cross-absorption modulation (XAM). Note that this method
avoids the signal inversion which results when using cross-gain
modulation (XGM) in semiconductor optical amplifiers [13].
Native signal regeneration (where the incoming signal itself is
improved and not converted to a new, locally generated signal)
at 10 Gb/s was also demonstrated [11], [12]. The regenerative
capability of the EAM relies on saturation of the absorption by
the incoming signal itself.
We present theoretical results that are in good agreement with
available experimental results, and investigate the influence of
device length and input power level of the control signal on the
conversion and regeneration results. We also show the influence
of the sweep-out dynamics on the quality of the results.
The paper is organized as follows: In Section II we describe
the model used in the simulations and the configuration used for
performing the wavelength conversion and signal regeneration.
In Section III the simulation results are presented. First, the non-
linearity in the absorption as function of input pulse energy is
described. We then describe wavelength conversion and signal
regeneration at 10, 20, and 40 Gb/s, and show the influence of
various parameters on the results. Finally, Section IV summa-
rizes and concludes the paper.
II. M
ODEL AND SETUP
The model used for the reverse-biased quantum well absorber
is a large-signal model originally developed for studying col-
liding-pulse mode-locked lasers [14]. The model includes prop-
agation effects, and is based on a detailed gain model, derived
using the density matrix formalism. Ultrafast effects such as
spectral hole-burning and carrier heating are taken into account,
while excitonic effects are not included in the model. We con-
sider a device with five quantum wells, and restrict our attention
to transverse electric (TE) polarized light.
Further, a simple carrier density dependent sweep-out time is
assumed. In the literature, sweep-out times on the order of sev-
eral tens of picoseconds in multiple quantum well (MQW) In-
GaAsP and AlGaAs structures have been reported, even at rela-
tively high reverse biases, around
V [15], [16]. Elsewhere,
saturable absorbers have displayed faster recovery times, down
to 8 ps [11], [17]. We have implemented a simple model, where
the sweep-out time varies as a function of the carrier density,
from 8 ps at low densities to 25 ps at transparency, based on
results in [10]. The mechanism that leads to longer sweep-out
times at higher carrier densities is screening of the applied field
by photogenerated carriers, see, e.g., [15], [16].
0733–8724/00$10.00 © 2000 IEEE
Authorized licensed use limited to: Danmarks Tekniske Informationscenter. Downloaded on February 9, 2010 at 02:31 from IEEE Xplore. Restrictions apply.
1122 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 8, AUGUST 2000
Fig. 1. The setup in the simulations. A data signal at wavelength
and a
CW signal at
are injected into an EAM from the same end (copropagation
scheme). Through XAM the information from the control signal at
is
transferred to the CW signal at
.
Fig. 1 shows a schematic of the wavelength conversion setup.
A control signal at wavelength
containing a bit sequence,
and a continuous-wave (CW) signal at
are launched into
the EAM. In our simulations we have mainly considered the
co-propagation scheme, in which the two signals enter the de-
vice from the same end as shown in Fig. 1. At the output end
of the EAM, the bit-sequence represented by the control signal
has been transferred to the CW signal through XAM.
III. R
ESULTS
In the following simulations, we investigate the absorption
nonlinearity of our device, as well as the use of this nonlinearity
for performing wavelength conversion and signal regeneration.
In all the results presented in this paper, the pulse full-width at
half-maximum (FWHM) is set to 8 ps.
For the remaining part of the paper, all power and energy
levels refer to in-chip levels. The results of course depend on
the confinement factor of the optical field in the chip.
A. The Absorption Nonlinearity
In Fig. 2, the calculated transmitted pulse energy is shown as
a function of input pulse energy for various wavelengths corre-
sponding to energies above the bandgap. For small pulse ener-
gies, the absorption corresponds to an essentially empty conduc-
tion band, and hence the absorption due to promotion of carriers
to the conduction band is high. When the carrier density comes
close to the transparency density (i.e., when the pulse energy is
large), the stimulated absorption is very low and the total ab-
sorption is approximately equal to the total internal loss. The
total internal loss is made up of losses in the active region (such
as free-carrier absorption), absorption in the cladding, and scat-
tering losses [18], and is in this paper taken to be 40 cm
. The
results are quite insensitive to the exact value of the internal loss
because the total absorption is dominated by stimulated absorp-
tion.
Fig. 2 also indicates another important issue, namely that
the output energy is severely reduced if one operates at shorter
wavelengths.This is also expected, since the absorption is larger
at shorter wavelengths. The saturation energy also increases:
At shorter wavelengths the photogenerated (quasi-equilibrium)
carrier density leads to a smaller change of the absorption. This
means that saturation of the absorption at shorter wavelengths
requires much more energetic pulses.
If one relies entirely on the change in absorption obtained
through phase-space filling, it is not desirable to have
(the
CW signal wavelength) at the longer wavelengths (close to the
band edge). This is because the change in absorption here is
relatively small, due to the low density of states. On the other
hand, the absorption increases at shorter wavelengths, resulting
Fig. 2. The graph show the output pulse energy as function of input pulse
energy. The pulses propagate through a 150-
m-long device. The results are
shown for four different wavelengths, and illustrate the nonlinear absorption
property of the EAM.
in a reduced output power. Hence, a compromise must be made
to obtain both a reasonable output power (pulling to the longer
wavelength side), and an efficient modulation of the carrier den-
sity (pulling to the shorter wavelength side).
By the same token, the controlling signal at
should not be
too close to the band edge, or its overall influence on the carrier
density will be small. Its influence on the absorption at the CW
wavelength
will correspondingly be small. If one wishes to
regenerate the signal at
, this wavelength should not be too
far into the band, again since the output power decreases with
decreasing wavelength.
B. Wavelength Conversion and Signal Regeneration at 10 Gb/s
To demonstrate the wavelength conversion and signal regen-
eration capability of the EAM, we propagate various bit pat-
terns consisting of 8-ps wide return-to-zero (RZ) pulses with
an extinction ratio (ER) of 10 dB (peak-to-floor) through the
device. The bit-patterns used were chosen such that the worst
possible eye was obtained. The 10-dB extinction ratio (ER) is
introduced by adding a constant background to the bit sequence
at the wavelength
. The background allows us to simulate the
presence of unwanted signals. This could for instance be an op-
tical time-division multiplexing (OTDM) channel that was not
perfectly cleared, or it could be spontaneous emission added by,
e.g., signal generators and amplifiers (and other active compo-
nents) previously encountered in the system.
Fig. 3 shows a simulation where a 10-dBm CW signal (“CW”
in Fig. 3) at
nm and an 11.5-dBm average pulse
power bit-stream signal (“signal” in Fig. 3) at
nm
are launched into the EAM from the same end (copropagation
scheme, cf., Fig. 1). Note that the quoted average pulse power
includes the contribution from the background, which is rela-
tively more important at low bit rates, where the duty-cycle is
low, than at higher bit rates. In practice, the two signals would
be separated at the output of the device using a spectral filter.
The figure shows the power levels for mark and space as a func-
tion of device length. For increasing device length, the ER in-
creases both for the control signal and for the converted signal
Authorized licensed use limited to: Danmarks Tekniske Informationscenter. Downloaded on February 9, 2010 at 02:31 from IEEE Xplore. Restrictions apply.
HØJFELDT et al.: ALL-OPTICAL WAVELENGTH CONVERSION AND SIGNAL REGENERATION USING AN EAM 1123
Fig. 3. Power levels for space (“0”) and mark (“1”) as function of length for
then controlsignal (“signal”) andthe CW signal (“CW”). The bit rate is 10 Gb/s.
The input power of the control signal was 11.5 dBm, while the power in the CW
signal was 10 dBm. The ER was extracted from worst-case eye-diagrams.
(the latter with an ER which is “0” at the input), demonstrating
the wavelength conversion and signal regeneration capabilities
of the device. When the control signal power level becomes too
low to significantly influence the absorption (through the car-
rier density), the ER values become constant. The power level,
however, keeps decreasing due to the internal loss and stimu-
lated absorption.
Fig. 3 clearly shows that the improvement in ER comes about
due to the nonlinear saturation properties: The energy in a “0”
decreases more rapidly than the energy in a “1” due to the higher
absorption for smaller input energy levels. To ensure a certain
power level at the output (e.g., 10 dB above the amplified spon-
taneous emission (ASE) level in an amplifier), the length of the
device must be kept below a certain value, which of course de-
pends on the input power levels.
C. Pulse Energy Dependence
The obtainable ER depends on the input power level of the
control signal. Fig. 4 showsthe ER ofa converted 10-Gb/s signal
as a function of device length for various average control pulse
energies. Corresponding for example to an average power of
11.5 dBm (average pulse energy of 2.8 pJ), an ER of
12 can be
obtained on the converted signal. Fig. 4 also shows that the ER
obtained below, say 150
m, for the highest input power level
shown (11.2 pJ average pulse energy) is smaller than that ob-
tained using a range of lower input power levels. This is because
the saturation of the absorption in the first part of the devicesup-
presses efficient carrier density modulation, and hence mark and
space will experiencealmost the same absorption. This indicates
that for a given device length there is an optimum input power
level giving the highest ER at the output. This is demonstrated
in Fig. 5, where the ER at the output of a 110-
m-long device
is shown as a function of the average control pulse energy. A
maximum ER at the output is found in this case around an av-
erage pulse energy of 4 pJ. For a longer device, the maximum
ER increases, and the peak occurs at a higher input energy.
Fig. 4. ER of a converted 10-Gb/s signal as function of length with average
control pulse energy as a parameter. The ER was extracted from worst-case
eye-diagrams. The CW power level was
P
=
10
dBm.
Fig. 5. Extinction ratio for the converted signal as function of average input
energy of the control signal for a 110-
m-long device. The bit rate was 10 Gb/s.
The ER was extracted from worst-case eye-diagrams.
D. Wavelength Conversion up to 40 Gb/s
To investigate the wavelength conversion capability at higher
bit rates, we modeled the propagation of 20 and 40 Gb/s pulse-
trains through the EAM. Again, we chose pulse-trains that en-
sured the worst possible eye-distortion.
To facilitate the understanding of the effects that are impor-
tant, we display the simulated eye diagrams. Note that the dia-
grams include the background introduced in Section III-B, re-
sulting in the before mentioned input extinction ratio of 10 dB.
We assumed that fixedand well-definedpowerlevelsentered the
device. This is of course an idealization, and any noise on the
input signals (amplitude or timing jitter) will reduce the quality
of both the converted signal and the control signal. However,
these are small effects compared to the distortions due to pat-
terning, that we find even without noise. Noise on the input will
tend to smear out the rather discrete levels found in the figures.
Eye-diagrams for different EAM lengths are shown for dif-
ferent power levels in Figs. 6 and 7. The control signal power
Authorized licensed use limited to: Danmarks Tekniske Informationscenter. Downloaded on February 9, 2010 at 02:31 from IEEE Xplore. Restrictions apply.
1124 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 8, AUGUST 2000
Fig. 6. Converted 10, 20, and 40 Gb/s signal at the output of a 125-
m long
device (left) and a 200-
m long device (right). The CW power was 10 dBm and
the signal power power is 11 dBm at 10 Gb/s, 16 dBm at 20 Gb/s, and 19 dBm
at 40 Gb/s. We explicitly point out that the levels indicated in the left figure (and
similar levels in the following figures) are all marks. The temporal window is
50 ps for all bit rates.
levels for the 10, 20, and 40 Gb/s signals in Fig. 6 were 13, 16,
and 19 dBm, respectively, and 16, 19, and 22 dBm, respectively,
in Fig. 7. The results are shown for 125 and 200-
m long de-
vices.
In Fig. 6, the quality of the eye for the 10 Gb/s signal hardly
improves from 125 to 200
m, because the remaining power
level at
m is too small to change significantly the
absorption (see also Fig. 3). Note that the width of the pulses
making up the converted signal decreases slightly with length.
The eye-quality in Fig. 6 for the 20 Gb/s signal also hardly im-
provesas the length of the EAM is increased, although some am-
plitude variation now occurs in the marks. This variation comes
about because the available time per bit is short enough that a
total recovery of the absorption is not achieved between bits. Fi-
nally, in the 40 Gb/s case the converted signal eye for the 125
m long device shows a large range of both the space and the
mark levels (only the mark levels can be clearly distinguished
in the diagrams), resulting again because the absorption is not
allowed to recover from one bit to the next. As expected, the
distortion is more severe at 40 Gb/s compared to 20 Gb/s. The
distortion seen in the eye agrees well with the experimental re-
sults in [10].
In Fig. 7, the power level for the control signal at each in-
dividual bit rate is twice as large as in Fig. 6. At 10 Gb/s, this
results in a better ER at both 125
m and 200 m. This is a man-
ifestation of the same dependence that was also shown in Fig. 4
in terms of the ER, namely that a moderately higher power level
can result in a better ER at the output. At 20 Gb/s there is some
distortion, again because the absorption has not completely re-
covered from one bit to the next. The distortion especially at
m is larger than in the corresponding eye in Fig. 6.
The reason is that the higher power level for the control signal
in Fig. 7 results in a longer sweep-out time. At the same time,
however, the background is absorbed overa longer length due to
the higher power level, resulting in a better ER. For the 40 Gb/s
case in Fig. 7, the distortion encountered in Fig. 6, where the
mark splits up, is at first slightly suppressed because each pulse
saturates the absorber more, but the background is also higher.
Fig. 7. Similar to Fig. 6, but the power level for control signal is 16 dBm at
10 Gb/s, 19 dBm at 20 Gb/s, and 22 dBm at 40 Gb/s. The temporal window is
again 50 ps for all bit rates.
At m, the overall result is worsened compared to the
corresponding eye in Fig. 6, just as for the 20 Gb/s case.
Another interesting and important feature is seen when com-
paring for instance the 10-Gb/s converted signal in Fig. 6 with
that of Fig. 7: The converted signal at the output of a 125
m
long device is broader in Fig. 7 than in Fig. 6. However, at
m, the situation is reversed. In general, this is due to
the fact that the conversion happens over a longer distance for
the more energetic signal. Initially the higher power level will
give broader converted pulses due to a longer sweep-out time,
but eventually (with length) the higher power levels give more
narrow pulses due to the longer conversion length.
To summarize, the converted signal at 10 Gb/s is improved
by doubling the power level. For the 20-Gb/s signal, doubling
the power gives a slightly worse ER for the shorter device, but
for the longer device (
m) the ER is improved and the
width of the converted pulses is more narrow, but at the expense
of some variation in the value of the converted marks. For the
40-Gb/s case, the higher power level initially helps to suppress
amplitude variations, but lifts the background. For the longer
device, the converted signal shows slight improvements in the
ER for the higher power level, but as for the 20 Gb/s case this
happens at the expense of more amplitude variation in the mark,
and even more so at 40 Gb/s than at 20 Gb/s.
Note that the converted pulses thus generated are somewhat
broader than the control pulses. This is a general property of this
type of conversion scheme. It is especially the trailing edges of
the converted pulses which are longer than those of the con-
trol pulses because the sweep-out dynamics “keeps the device
open,” thereby giving the converted pulses an asymmetric pulse
shape. This asymmetry increases for longer sweep-out times,
again demonstrating the importance of keeping the sweep-out
time short.
E. Signal Regeneration
The nonlinearity used for performing wavelength conversion
of course also allows a similar improvement in the ER for the
control signal itself.
It has been shown that an incoming signal with a reminiscent
pulse in the time-slot can be cleaned up [12]. We show in the
Authorized licensed use limited to: Danmarks Tekniske Informationscenter. Downloaded on February 9, 2010 at 02:31 from IEEE Xplore. Restrictions apply.
