230 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 24, NO. 1, JANUARY 2006
Error-Free All-Optical Wavelength Conversion at
160 Gb/s Using a Semiconductor Optical Amplifier
and an Optical Bandpass Filter
Y. L i u , Member, IEEE, E. Tangdiongga, Z. Li, Shaoxian Zhang, Huug de Waardt,
G. D. Khoe, Fellow, IEEE, and H. J. S. Dorren, Member, IEEE
Abstract—Error-free and pattern-independent wavelength con-
version at 160 Gb/s is demonstrated. The wavelength converter
utilizes a semiconductor optical amplifier (SOA) with a recovery
time greater than 90 ps and an optical bandpass filter (OBF)
placed at the amplifier output. This paper shows that an OBF
with a central wavelength that is blue shifted compared to the
central wavelength of the converted signal shortens the recovery
time of the wavelength converter to 3 ps. The wavelength converter
is constructed by using commercially available fiber-pigtailed
components. It has a simple configuration and allows photonic
integration.
Index Terms—All-optical switch, nonlinear optics, semicon-
ductor optical amplifier (SOA), ultrafast, wavelength conversion.
I. INTRODUCTION
A
LL-OPTICAL wavelength converters (AOWCs) are con-
sidered as important building blocks in wavelength-
division-multiplexed networks [1], [2]. AOWCs that utilize
nonlinearities in semiconductor optical amplifiers (SOAs) have
attracted considerable research interest due to their integration
potential and their power efficiency [3], [4]. A number of SOA-
based AOWCs have been demonstrated [5]–[7]; however, the
slow SOA recovery, determined by the electron–hole recombi-
nation time (typically several tens to hundreds of picoseconds),
can cause unwanted pattern effects in the converted signal. This
sets a limit to the maximum operation speed of the wavelength
converter.
Many approaches have been tried to increase the operation
speed of SOA-based AOWCs. Some approaches focus on re-
ducing the carrier recovery time of the SOA. It has been shown
in [8] that SOAs with a long active region recover faster com-
pared to SOAs with a short active region. The SOA recovery can
also be reduced by saturating the SOA using an external holding
beam. The holding beam can be either continuous-wave (CW)
Manuscript received July 18, 2005; revised October 24, 2005. This work
was supported by the Netherlands Organization for Scientific Research (NWO),
the Technology Foundation Stichting Technologie Wetenschap (STW), and
the Ministry of Economic Affairs through, respectively, the National Research
Combination (NRC) Photonics Grant and the Innovational Research Incentives
Scheme (IRIS) program. This work has been supported in part by the Euro-
pean Commission through the project Information Society Technologies—all-
optical LAbel SwApping employing optical logic Gates in NEtwork nodes
(IST-LASAGNE) (FP6-507509).
The authors are with the COBRA Research Institute, Eindhoven University
of Technology, Eindhoven, MB 5600, The Netherlands (e-mail: Y.Liu@tue.nl).
Digital Object Identifier 10.1109/JLT.2005.861136
light [9]–[11], or modulated light [12]. However, it is difficult
to achieve full recovery of the SOA in the order of 10 ps.
Some methods have been proposed to increase the frequency
response of an SOA-based AOWC. An increased operation
speed has been achieved by employing a fiber Bragg grating
(FBG) [13], or a waveguide filter [14]. Wavelength conversion
at 100 Gb/s has been demonstrated by using a long SOA
(2 mm) in combination with an FBG [7]. In [15], a switch
using a differential Mach–Zehnder interferometer with SOAs
in both arms has been introduced. The latter configuration
allows the creation of a short switching window (several
picoseconds), although the SOA in each arm exhibits a slow
recovery. A delayed interferometric wavelength converter, in
which only one SOA has been implemented, is presented in
[6]. The operation speed of this wavelength converter can reach
160 Gb/s [6] and this approach also allows photonic inte-
gration [16]. This concept has been analyzed theoretically in
[17]. The delayed interferometer also acts as an optical filter.
Nielsen and Mørk [18] present a theoretical study that reveals
how optical filtering can increase the modulation bandwidth
of SOA-based switches.
Optical filtering of chirped wavelength-converted output
light of an SOA has been utilized to achieve polarity-preserved
wavelength conversion at 40 Gb/s [19]–[21]. It has been shown
in [19] that the red-chirped component of the converted output
light, filtered by an optical step filter (with a sharp frequency
response), can be used to obtain noninverted wavelength con-
version. Similarly, Nielsen et al. [20] shows that filtering of
the blue-chirped part of the converted output pulse can lead
to noninverted wavelength conversion. In [21], both the blue-
and red-chirped components of the converted signals are filtered
by a pulse reformatting optical filter to achieve noninverted
wavelength conversion.
In this paper, we show that it is possible to make the
wavelength converter fully recover in 3 ps, while the employed
SOA has a recovery time greater than 90 ps. Speeding up of
the AOWC recovery is realized by placing an optical bandpass
filter (OBF) at the SOA output. The central wavelength of
the OBF is blue shifted with respect to the wavelength of the
converted signal. This concept can be used to achieve error-free
and pattern-independent wavelength conversion at a bit rate of
160 Gb/s, with a low power penalty [22]. The signal filtered
by the bandpass filter is injected into a delayed interferometer,
which is used to convert the inverted signal into a noninverted
0733-8724/$20.00 © 2006 IEEE
LIU et al.: ERROR-FREE ALL-OPTICAL WAVELENGTH CONVERSION AT 160 Gb/s USING A SOA AND AN OBF 231
Fig. 1. (a) Setup for testing the SOA gain recovery. (b) Input pump pulse.
(c) SOA gain recovery measured by an optical sampling scope. BPF is (optical)
bandpass filter.
signal. It should be noted that, in contrast to [5] and [6],
differential operation in the delayed interferometer is not essen-
tial for realizing error-free 160-Gb/s operation, since the input
of the delayed interferometer already shows a clear open eye.
This wavelength converter allows high-speed wavelength
conversion without using sophisticated custom-made devices
such as differentially operated SOA-Mach–Zehnder interfer-
ometers. This concept also has the potential to be used in
other all-optical switches, such as in optical demultiplexers.
We obtained our results using commercially available fiber-
pigtailed components. The system is easy to implement since
the configuration is very simple. Moreover, the concept allows
photonic integration.
The paper is organized as follows. In Section II, the operation
principle for ultrafast recovery is explained. Experimental re-
sults are presented in Section III. Finally, conclusions are given.
II. O
PERATION PRINCIPLE
A setup for measuring the recovery time of the SOA is
depicted in Fig. 1(a). A pulsed pump signal is combined with a
CW probe signal, and the signals are simultaneously launched
into the SOA. At the output of the SOA, an OBF is used to select
the probe light, and to block the pump signal. The probe light
that outputs the OBF is monitored by an optical sampling scope
(Agilent 86119A) with an optical bandwidth up to 700 GHz.
The pulsewidth of pump signal is 2.2 ps, as shown in Fig. 1(b).
We first use the optical sampling scope to measure the SOA
recovery if the central wavelength of the OBF is located at
the center wavelength. The OBF in this measurement has a
3-dB bandwidth of 10 nm. The experimental result is shown
in Fig. 1(c). Fig. 1 summarizes some well-known results. The
Fig. 2. SOA gain recovery. Solid line: experimental result. Dashed line:
simulation result.
input optical pump pulse causes stimulated emission and leads
to a reduced SOA gain. The SOA gain saturation time is
determined by the pulse duration; the SOA gain approximately
reaches its minimum at the time that the input pump pulse
reaches its maximum intensity.
Roughly speaking, the SOA recovers on three different
timescales. Ultrafast gain recovery, driven by carrier–carrier
scattering takes place at subpicosecond timescales [23]. Fur-
thermore, carrier–phonon interactions contribute to the recov-
ery of the amplifier on a timescale of a few picoseconds [23].
Finally, on a nanosecond timescale, there is a contribution
driven by electron–hole interactions. We have utilized a model
that accounts for all these effects [24]. A simulation result for
the SOA recovery is given by the dashed line in Fig. 2, which
shows good agreement with the experimental curve (solid line).
The injected pulses not only modulate the gain of the SOA,
but also modulate the refractive index. This results in chirp on
the output signal. The leading edges of the (inverted) converted
probe pulses are red shifted, whereas the trailing edges are blue
shifted [19]–[21]. We also used our model to simulate the chirp
on the converted pulse. The simulation result is presented in the
lower panel of Fig. 3(a), whereas for reasons of comparison, the
gain recovery is shown in the upper panel. A similar result can
be found in [21].
If the central wavelength of the OBF is blue shifted with
respect to the central wavelength of the probe beam [Fig. 3(b)],
the converted signal recovers much faster compared to the
case that the central wavelengths of the filter and the probe
beam coincide. The operation of the wavelength converter is
schematically presented in Fig. 3(c). The dotted and dashed
lines in Fig. 3(c) are the SOA gain and chirp, respectively, and
also shown in Fig. 3(a). The fast recovery of the wavelength
converter can be explained as follows. When the pulse appears
at point A, the SOA carriers deplete and the gain drops, reach-
ing its minimum at point B. The SOA gain saturates during
timeslot A–B. Furthermore, in timeslot A–B, the wavelength
of the probe light moves to a longer wavelength (red chirp)
and thus receives more attenuation by the filter. As a result, the
transmittance of the probe light through the filter is reduced. At
point B, the chirp becomes 0, and the SOA starts to recover.
From this point onwards, the wavelength of the probe light is
blue shifted, leading to an increased transmittance. If the OBF is
232 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 24, NO. 1, JANUARY 2006
Fig. 3. Operation principle of the wavelength converter. (a) SOA gain (upper
panel) and chirp response (lower panel) as a function of time. (b) Schematic
of the optical spectrum of the input probe light and the filter characteristic.
(c) Transmittance through the OBF as a function of time. The dotted and dashed
lines are the SOA gain and chirp, respectively, the same as the curves in (a).
properly selected (the slope of the OBF is especially essential),
the enhancement of transmittance due to the blue chirp can
compensate the gain saturation. Thus, the transmittance at
point C is equal to the transmittance at point A. From points C
to D, the wavelength of the probe light slowly moves back to the
probe carrier wavelength, leading to a decreased transmittance.
However, the SOA gain starts to recover, leading to an increased
amplification of the probe light. These two effects take place
on the same timescale and cancel each other out. As a result,
the net intensity at the filter output is constant. This means that
the system effectively recovers much faster than the SOA gain.
It should be noted that these effects can be utilized with low-
energy optical pulses. In our experiment, the input pump pulse
energy was about 60 fJ. We will show that an OBF is sufficient
to utilize these ultrafast effects for wavelength conversion at
160 Gb/s.
III. E
XPERIMENT AND RESULTS
The experimental setup for 160-Gb/s wavelength conversion
is shown in Fig. 4(a). The setup was constructed by using com-
mercially available fiber-pigtailed components. A 10-Gb/s data
stream with 1.9-ps-wide optical pulses, generated by an actively
mode-locked fiber ring laser (MLFRL), is modulated by an
external modulator at 10 Gb/s to form a 2
7
− 1 return-to-zero
(RZ) pseudorandom binary sequence (PRBS). This 10-Gb/s
RZ-PRBS data stream is multiplexed to 160 Gb/s by using a
passive fiber-based pulse interleaver. The 160-Gb/s data signal
is combined with a CW probe light and fed into an AOWC
via a 3-dB coupler. As shown in the dashed box of Fig. 4(a),
the AOWC is composed out of an SOA, a 1.4-nm OBF, and a
delayed interferometer. The delayed interferometer consists of
two polarization controllers (PCs), a polarization maintaining
fiber (PMF) with 2-ps differential delay, and a polarization
beam splitter (PBS). The operation principle of the delayed
interferometer can be found in [5], [6], and [16]. Note that the
delayed interferometer allows photonic integration [16]. The
SOA in the AOWC is a commercial product from Kamelian
and is designed as an optical preamplifier. The SOA is pumped
with 250 mA of current. The center wavelengths of the
160-Gb/s data signal and the CW probe beam are at 1549.98
and 1560.77 nm, respectively. The average optical power of the
160-Gb/s data signal is 4.8 and 2.6 mW for CW probe light,
measured at the pigtail at the SOA input.
The injected 160-Gb/s data signal modulates the SOA car-
riers, and thus, the SOA gain. As a result, the CW probe light
is modulated via cross-gain modulation, causing inverted wave-
length conversion. Moreover, the injected data signal modulates
the refractive index of the SOA, resulting in a chirped converted
signal. As discussed before, the leading edges of the (inverted)
converted probe light are red shifted, whereas the trailing edges
are blue shifted. Thus, the spectrum of the probe light at the
SOA output is broadened as shown in Fig. 5(a) (measured by an
optical spectrum analyzer with 0.02-nm resolution). A 1.4-nm
OBF, which is placed at the SOA output, selects the blue-shifted
sideband of the probe light. The OBF characteristic is indicated
by the dashed line (experimental result) in Fig. 5(a). The center
wavelength of the OBF is detuned 1.23 nm to the blue side with
respect to the probe carrier wavelength. The insertion loss of
the detuned OBF is about 13 dB. Fig. 5(b) shows the optical
spectrum of the probe light at the output of the OBF.
The converted probe light is monitored by using the
optical sampling scope. Fig. 4(c) shows that an inverted
160-Gb/s signal with a clear open-eye pattern is obtained,
which indicates that the wavelength converter recovers in less
than 3 ps. This ensures pattern-independent wavelength conver-
sion at 160 Gb/s.
The inverted 160-Gb/s signal is subsequently injected into
the delayed interferometer, where the polarity of converted
signal is changed, i.e., the inverted signal is changed into a
noninverted signal. It is noted that differential operation in the
delayed interferometer is not essential for realizing 160-Gb/s
operation because the input (inverted) pulses have already been
fully recovered within 3 ps.
The transmittance of the delayed interferometer is presented
in Fig. 5(c), which shows that the delayed interferometer
operates as a “notch” filter. Since the role of the delayed
interferometer is to change the polarity of converted signal, we
align the wavelength of the notch with the center wavelength
LIU et al.: ERROR-FREE ALL-OPTICAL WAVELENGTH CONVERSION AT 160 Gb/s USING A SOA AND AN OBF 233
Fig. 4. (a) 160-Gb/s all-optical wavelength-conversion setup. (b) Eye diagram of the 160-Gb/s input pump signal. (c) Eye diagram of the converted light at the
output of the bandpass filter. (d) Eye diagram of the converted light at the output of the delayed interferometer. PC: polarization controller; MLFRL: mode-locked
fiber ring laser; BPF: (optical) bandpass filter; PBS, polarizing beam splitter; PMF: polarization maintaining fiber; EDFA: erbium-doped fiber amplifier.
of the converted light. This ensures a high attenuation of the
dc component corresponding to the “1” level in the inverted
signal [see Fig. 4(c)] and a larger transmittance of the “0” level.
Hence, the polarity of the signal that outputs the SOA is inverted
[Fig. 4(d)]. The optical spectrum at the output of the delayed
interferometer is shown in Fig. 5(d).
After wavelength conversion, the converted signal is demulti-
plexed from 160 to 10 Gb/s by using a gain transparent ultrafast
nonlinear interferometer [25]. The demultiplexed 10-Gb/s data
signal is fed into a 10-Gb/s receiver and a bit error rate (BER)
tester. Fig. 6 shows BER measurements of the 160-Gb/s input
signal and the converted signal. All the 16 10-Gb/s tributaries
are presented in Fig. 6(a). In addition, the 10-Gb/s basic chan-
nels that are multiplexed to the 160-Gb/s data stream is also
presented. It can be observed that the average sensitivity penalty
for wavelength conversion at a BER =10
−9
is about 2.5 dB
with respect to that of the original 160-Gb/s signal. The input
dynamic range is about 6 dB to keep BER values under 10
−9
.
Moreover, it is visible that no error floor is observed, which
indicates excellent performance of the proposed wavelength
converter.
The polarization dependence of our 160-Gb/s wavelength
conversion has also been investigated. We find that the wave-
length converter is polarization insensitive for the optical pump
signal because the SOA is polarization independent (< 0.3 dB).
However, we need to precisely control the polarization of the
CW input light. The main reason for this is that our scheme
requires exact control of the detuning between the center
wavelength of the OBF and the CW carrier frequency. The
OBF used in the experiment has a polarization-dependent trans-
mittance, resulting in a polarization-dependent performance of
the system. Thus, this wavelength-conversion concept can, in
principle, be made polarization independent if a polarization-
independent OBF is used.
It should be noted that the SOA used in the experiment has
a recovery time of more than 90 ps. In wavelength conversion
based on cross-gain/phase modulation, an SOA with a recovery
time of 90 ps allows operation at bit rates less than 40 Gb/s. We
have achieved error-free wavelength conversion at 160 Gb/s.
Simulations show that operation at higher bit rates is feasible.
As a last point, we discuss the speed of this wavelength
converter. Fig. 7 shows a numerical simulation of the SOA gain
saturation and chirp if an optical pulse is injected in the SOA.
The dashed line shows the result for the case that the pulse
had duration of 2.2 ps [full-width at half-maximum (FWHM)]
and the solid line represents the case for which the pulse has
duration of 1 ps. In the latter case, the peak power is reduced
such that the maximum gain saturation introduced by both
pulses is almost the same. It is visible that in both cases the
SOA recovery has a fast and a slow component. Simulations
indicate that the fast component of the recovery is determined
by the duration of the pulse, whereas the slow component
is determined by the SOA carrier dynamics. This means that
for pulses with duration as short as 1 ps, the recovery of the
wavelength converter is limited by the pulsewidth and not by
the carrier dynamics in the SOA. Simulations also show that
if the pulsewidth is decreased further (to about 200 fs), ulti-
mately, the recovery of the wavelength converter is limited by
234 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 24, NO. 1, JANUARY 2006
Fig. 5. Optical spectra and the 1.4-nm OBF shape. (a) Optical spectrum of
the probe light at the input and output of the SOA; the dashed line shows
the shape of the 1.4-nm OBF. (b) Spectrum after filtering the SOA output.
(c) Transmission characteristic of the delayed interferometer. DI: delayed
interferometer. (d) Spectrum at the output of the delayed interferometer.
Fig. 6. (a) BER performance of 160-Gb/s wavelength conversion. (b) and (c)
Eye diagrams of the demultiplexed 10-Gb/s input and wavelength-converted
signal, respectively.
ultrafast carrier dynamics in the SOA (carrier–carrier scattering
and carrier–phonon interactions) [22], [23]. These simulations
suggest that this approach can be potentially used to achieve
wavelength conversion at bit rates higher than 160 Gb/s.
Fig. 7. Simulated gain response (upper panel) and the chirp response (lower
panel) of probe light versus time in the case of the input pump pulses with
different pulsewidth. Solid line: 1.0-ps pump pulse; dashed line: 2.2-ps pump
pulse. Note that all the input pump pulses have the same optical peak power.
IV. CONCLUSION
We have demonstrated pattern-independent wavelength con-
version at 160 Gb/s with a low power penalty by employing
an SOA with a gain recovery time greater than 90 ps. The
essential point in our approach is to employ an OBF with
a center wavelength that is blue shifted with respect to the
center wavelength of the probe light. We have explained how
detuning of the OBF can be utilized to speed up the recovery
time of the wavelength converter. In our approach, the gain
recovery of SOA is still very slow, but chirp dynamics in the
SOA is very fast. The detuned OBF is used to extract the fast
chirp dynamics, which leads to high-speed wavelength con-
version. Numerical simulations indicate that by using shorter
optical pulses, the response time of the wavelength converter
can be further increased. Ultimately, the response time of the
wavelength converter is limited by ultrafast carrier dynamics
in the SOA, i.e., carrier–carrier interaction and carrier–phonon
interactions. In our experiments, the wavelength converter has
been demonstrated by using low-power optical pulses and
commercially available fiber-pigtailed components. The wave-
length converter has a simple configuration and allows photonic
integration.
Our approach does not constrain to a specific device. Similar
results have been obtained by using other SOAs. However, the
performance of the wavelength converter is better if the SOA
produces more chirp (this is an SOA with a large linewidth
enhancement factor). We observe a similar performance if
the wavelength of the probe light is changed. However, it is
desirable to use a probe light with a wavelength for which
the linewidth enhancement factor is large, which means the
probe wavelength is located at the “longer” wavelength side of
the SOA bandwidth. In fact, in this wavelength area, the gain
saturation is also smaller. Therefore, the required detuning of
the OBF is smaller, which leads to a better optical-signal-to-
noise ratio (OSNR). However, if the probe wavelength is too
