Vous êtes sur la page 1sur 12

PERFORMANCE ANALYSIS OF DELAY

INTERFEROMETER-BASED SCHEME FOR


ULTRAFAST RETURN-TO-ZERO DATA
PATTERN EFFECT MITIGATION
IN SEMICONDUCTOR OPTICAL AMPLIFIER

K.E. Zoiros, T. Siarkos, C.S. Koukourlis


Democritus University of Thrace, School of Engineering
Departmemt of Electrical & Computer Engineering
12 Vas. Sofias Str., 67 100, Xanthi, Greece
kzoiros@ee.duth.gr

ABSTRACT
The feasibility of employing a delay interferometer (DI) to compensate for the pattern
effect that manifests in a stand-alone semiconductor optical amplifier (SOA) when it is
driven by return-to-zero (RZ) data being faster than its dynamic response is theoretically
investigated and demonstrated. This is achieved by means of a numerical model applied
to simulate and correlate the operation of these two elements in the context of their
cascaded configuration. The outcome of this process allows to analyze and evaluate the
impact of the input signal and SOA characteristics on the amplitude modulation of the
sequence exiting each one of these devices. The obtained results disclose that owing to
the utilization of the DI this quality metric is decreased compared to the case of a single
SOA only while the technical limitations imposed on the involved critical parameters are
significantly relaxed and their useful operational range is extended. These findings
underscore the merit of placing a DI after a SOA for successfully combatting the
undesirable pattern dependence and improving its performance when exploited purely for
direct amplification purposes.

Keywords: semiconductor optical amplifier, pattern effect, delay interferometer.

1 INTRODUCTION purposes in real-life optical communications systems.


In order to confront this problem and enable the
The semiconductor optical amplifier (SOA) has unobstructed exploitation of a stand-alone
become an indispensable element for the conventional SOA in its classical role several
implementation of modern lightwave networks methods have been proposed, which include the use
owing to the attractive properties of multi-functional of higher driving current and/or longer active region
capability, low power consumption, small footprint, length [4], external light injection [5-6],
broad gain bandwidth and integrability [1]. However, sophisticated line coding [7], bandpass filtering [8-9],
its use in linear amplification applications [2], such interferometric arrangements [10-11], cascading
as power booster, in-line amplifier and receiver SOAs [12] and new materials [13]. An alternative
preamplifier, is limited by the slow carrier lifetime, promising option is to place after the SOA a delay
which is typically of the same order as the launched interferometer (DI), whose suitability and
data modulation speeds. The negative consequence competence stems from the attractive features of [14-
of this drawback is that when the SOA is biased to 15] simplicity, compact size, affordable cost,
operate in the deep saturation regime, its dynamical stability to environmental perturbations, and the
behavior is dominated by the binary content of the potential to be integrated and co-packaged with the
incoming ultrafast random data stream. This SOA in a single module. Although this technique has
phenomenon, which is known as “pattern effect” [3], been shown in the laboratory to be capable of
causes the output of the SOA to depend on the values mitigating the pattern dependence on 10 Gb/s non
of the preceding input data bits instead of solely on return-to-zero (NRZ) pulses [14], in some practical
the current input bit. The negative by-product of this applications the return-to-zero (RZ) may be
fact is that the amplified output suffers from preferable [16] and is a format for which the research
amplitude fluctuations, which impede the interest is continuously growing [17-18]. However,
deployment of SOAs for direct signal amplification the success of the proposed scheme with the former

Ubiquitous Computing and Communication Journal 1


does not guarantee in advance that this will also
happen with the latter because in each case the
pattern effect manifests differently, namely it causes
a distortion between the leading and trailing edge of
the NRZ pulses [14] whilst it incurs fluctuations
between the peak amplitudes of the RZ pulses, as it
will be later shown. Hence the conclusion that the
employment of a DI after a SOA that amplifies NRZ
pulses is equally efficient for the RZ case is not à Figure 1: Layout of the theoretically investigated
priori assured unless it is reached through a pattern effect suppression scheme
comprehensive theoretical analysis and verified that
it holds for this format as well. Furthermore, in When the SOA is excited by a train of digitally
contrast to NRZ, a RZ pulse occupies by definition coded pulses that arrive at intervals shorter than its
only a fraction of the bit slot duration with inevitable carrier lifetime then the amplified profile is not
consequence that the requirement for the time delay uniform. This happens because the gain of a SOA
is expected to be stringent for the RZ coding. In that is subject to an ultrafast data stream is not
contrast it is more relaxed for the NRZ profile since altered in an ordinary fashion but in direct response
it is allowed to be much higher over the interval to the input binary content, namely it is dropped for a
between successive pulses, as it has been indeed logical‘1’ and partially recovered for a logical ‘0’.
confirmed from the corresponding experimental Moreover, continuous ‘0’s help the gain to rise
demonstration [14]. On the other hand, single further whilst continuous ‘1’s impede the gain to
channel data rates in optical communication systems recover. The result is that the gain and thus the
are progressively increased in order to satisfy the amplitude of a specific bit depends on the preceding
broadband needs created due to the massive use of bits, leading to amplitude distortion at the output of
Internet and multimedia applications. Thus for all the SOA. This effect is more intense when the
these combined reasons it would be practically logical content changes from ‘1’ to ‘0’, since a ‘1’
interesting to investigate if the specific pattern effect saturates the SOA and the ‘0’ that follows leaves
suppression concept can be efficiently extended and time for gain recovery so that the next ‘1’ exhibits a
adapted to such demanding requirements. This goal different gain than the previous ‘1’. This behavior is
is theoretically pursued in this paper by applying an verified from the variation of the SOA gain in
appropriate model to describe the operation of a response to the indicative pattern frame of 11010001
SOA when it is driven by a 20 Gb/s pseudorandom simulated according to the information provided in
binary sequence (PRBS) and link it with that of a DI. the next section and depicted in the following figure.
By undertaking a methodical approach based on
numerical simulation the impact of the involved
critical parameters on the resultant amplitude
modulation with and without the DI is thoroughly
investigated, revealing the pivotal role played by this
passive element in relaxing the technical restrictions
that would be imposed, if it were absent, on the
working conditions of a single SOA device in order
to suppress the undesirable pattern-dependent
wandering. Finally, useful guidelines are derived for
their proper selection and combination so as to
optimize the defined quality metric, which
determines decisively the performance of a SOA
when used for direct amplification purposes. The
outcome of this study may assist the design of more
complex SOA-based circuits and subsystems that
incorporate a DI for addressing the serious issue of Figure 2: Simulated SOA gain temporal evolution in
the pattern effect according to the suggested scheme. response to the pattern frame of 11010001
2 PRINCIPLE OF OPERATION As it can be clearly seen when two ‘1’s inside the
stream are separated by a ‘0’ then the trailing ‘1’
Fig. 1 illustrates the block diagram of the experiences a gain from a substantially recovered
configuration under study, which consists of a DI point that the leading ‘1’ did not have the time to
serially connected to a SOA. reach, and this situation is exacerbated for long
intervening strings of spaces. Similarly, in a row of
marks, the ‘1’ which follows just after the ‘1’ that

Ubiquitous Computing and Communication Journal 2


has been inserted first in the SOA (left-most bit in existence of ∆τ. Depending thus on how close this
the figure) and hence inevitably modified its optical quantity is to an even or an odd multipliciate of π,
properties from a higher level, suffers a reduced gain the recombined optical power is more or less
and this continues to happen for every next ‘1’ with transmitted at the crossed-coupled exit port,
respect to its preceding bit until a steady-state is respectively. The key thus for equalizing the peak
established or the sequence is broken by a ‘0’. power of the ‘1’s emerging from the SOA is to
Nevertheless, the specific gain variation, which compensate their phase change induced inside this
is the cause of the problem, can be suppressed by active element by the phase difference imparted
noting that it is always accompanied by a phase between their direct and delayed version in the DI.
change, with the two phenomena being coupled This can be realized by adjusting ∆τ so that the phase
through the SOA linewidth enhancement factor [19]. variations resulting from straightforward
This essentially implies that, for example, a mark amplification are smoothed at the output of the DI, as
which appears immediately after one or more spaces shown in the following figure obtained for a delay ∆τ
will undergo a stronger phase shift across its interval = 3 ps.
than a ‘1’ positioned ahead of a ‘0’ but after one or
more ‘1’s since, as explained, the gain acquired by
the former is larger. This fact is evident in the
following figure obtained for the fixed values of the
data and SOA critical parameters cited in section 4.

Figure 4: Difference between the directly modulated


phase and its delayed replica for ∆τ = 3 ps

Its comparison with Fig. 3 reveals that the temporal


offset in the DI creates a phase window across each
Figure 3: Stand-alone SOA modulated phase mark whose magnitude excursions occur around the
same level regardless of the exact ‘1’ position in the
Therefore, if this phase discrepancy could be driving pulse stream. In this manner the higher
converted to an amplitude modulation but of the amplified amplitudes can be clamped whilst the
opposite magnitude, the peaks of the output ‘1’s lower ones enhanced so that they are equalized and
would be balanced and consequently the pattern the pattern effect is effectively masked at the output
dependence would be alleviated. This can be of the interferometer.
achieved by exploiting the relative delay, ∆τ,
between the two branches of a DI formed by 3 MODELING
connecting together a pair of 50/50 couplers. A pulse
that comes out from the SOA is coupled into the The theoretical model that simulates the setup of
input port of the DI and then it is split in two beams the concatenated SOA and DI has been built by
of equal intensity, which propagate along the upper considering first the operation of each one of these
and lower arm, respectively, where without loss of units. More specifically, the SOA response, h(t), to
generality the former is the one that has a delay ∆τ.
As a result the corresponding halved constituent the inserted optical data pulses of power Pdata (t) is
reaches the second coupler later than its copy whose obtained if the partial coupled differential equations
path is shorter. This relative temporal difference is describing the gain, power and phase evolution of a
translated to the creation of a phase difference, strong optical signal that propagates and is amplified
namely ∆φ = 2πc∆τ/λ [20], where c is the speed of in a SOA [21] are properly manipulated following
light in vacuum and λ is the wavelength of the the procedure in [22] after the necessary valid
injected optical signal. Thus when these beams assumptions [23] have been made. In this manner it
collide and interfere with each other at the other end can be mathematically described in the time domain
of the DI the produced result depends periodically on by the formula
the amount of their differential phase due to the

Ubiquitous Computing and Communication Journal 3


dh(t) 1 equals 2kπ and (2k + 1)π, respectively, where k is an
= ×
dt 1+ ε exp[ h(t)]Pdata (t) integer. Since the SOA drives the input of the DI, the
combination of these elements can be modeled if
⎧ ln(G ss ) − h(t) − ε exp h(t) −1 dPdata (t) − ⎫
⎪⎪ Tcarrier [ ( ) ] dt ⎪ (1) Pin, DI (t) is replaced with Pout, SOA (t) from Eq. (2)

⎨ ⎬ and φin, DI (t) with φ out, SOA (t) from Eq. (3) so that
⎪−[exp( h(t) ) −1] Pdata (t)⎛⎜ ε + 1 ⎞⎟ ⎪
⎪⎩ Eq. (4) is modified to read
⎝ Tcarrier Esat ⎠ ⎪⎭
⎧ ⎫
where the SOA unsaturated small signal power gain, ⎪P (t) + Pout, SOA (t −∆τ) + ⎪
G ss , and carrier lifetime, Tcarrier , are treated as ⎪ out, SOA ⎪
1⎪ ⎪ (6)
Pout, DI (t) = ⎨+ 2 Pout, SOA (t)Pout, SOA (t −∆τ) ×⎬
varying parameters, whilst its nonlinear gain 4⎪ ⎪
compression factor due to the combined action of the ⎪× cos ⎡ − α ( h(t) − h(t − ∆τ) ) ⎤ ⎪
⎢⎣ 2 ⎥⎦
spectral hole burning and carrier heating intraband ⎩⎪ ⎭⎪
effects, ε, and its saturation energy, E sat , are set to
4 RESULTS
the representative values for bulk InGaAsP
semiconductor materials operating at 1550 nm of 0.2 The performance of the proposed scheme is
W-1 and 1000 fJ, respectively. Eq. (1) can be evaluated in terms of the amplitude modulation
numerically solved in a step-wise manner by
P
dividing the optical pulse in many small temporal (AM), which is defined as AM (dB) = 10 log max ,
segments, approximating the time derivative by a Pmin
finite difference [2] and applying the appropriate where Pmax is the maximum and Pmin the minimum
initial conditions at the end facets of a 1000 µm long peak power of the marks at the output of either the
SOA, according to the details stated in [24]. The SOA or the DI computed numerically through Eq.
input data signal used in this procedure is a RZ 20 (2) or (6), respectively. This metric characterizes in a
Gb/s PRBS of word length 27-1 whose Gaussian- quantitative manner the extent of the pattern effect in
shaped ‘1’ s have varying energy, E data , and full- a SOA as well as the degree of its suppression after
width at half-maximum (FWHM), T . the DI and although there is no absolute criterion to
FWHM
which it must comply it is widely accepted that it
The knowledge of h(t) allows to find the power should not exceed 1 dB [26]. However, the
and the phase shift acquired by each pulse at the achievement of this goal in a SOA that is not assisted
output of the SOA from the relationships [19] by a DI is not trivial because its performance is
susceptible to several factors, which include the
Pout, SOA (t) = Pdata (t)exp [ h(t)] (2) energy and FWHM of the data pulses as well as the
SOA carrier lifetime, small signal gain and linewidth
enhancement factor. For example, the values 300 fJ,
1
φ out, SOA (t) = − αh(t) (3) 7 ps, 80 ps, 25 dB and 6, respectively, used for the
2 scopes of the simulation are not appropriate since the
calculated AM of almost 1.2 dB is unacceptable.
respectively, where α is traditional linewidth This is inevitably reflected on the quality of the
enhancement factor. On the other hand, it is known amplified data stream, as it can be seen indeed in Fig.
that when a signal of initial power Pin, DI (t) and 5 for the representative 8-bit-long segment used
throughout the simulation for illustration purposes.
phase φin, DI (t) is inserted in a DI whose two Clearly, the amplitudes of the amplified pulses
constructing couplers have 50/50 splitting ratio then situated at the 2nd, 4th and 8th bit slots are not equal
its power at its crossed output port is given from [25] for the reasons explained in section 2. More precisely,
the peak of the last ‘1’ surpasses that of the two other
⎧ Pin, DI (t) + Pin, DI (t −∆τ) + ⎫ ‘1’s since it is preceded by a run of three zeros and
1 ⎪⎪ ⎪⎪
(4) hence encounters a gain that has had the time to
Pout, DI (t) = ⎨ + 2 Pin, DI (t)Pin, DI (t −∆τ) × ⎬
4⎪ recover to a raised level. Similarly, the third ‘1’

⎪⎩× cos[ ∆φ DI (t) ] ⎪⎭ senses a lower but still elevated gain as it is preceded
by a ‘0 and so its peak is higher than that of the
which indicates that it is maximized and minimized second ‘1’. The latter in turn enters the SOA after the
when the phase difference between the upper and first ‘1’ has strongly perturbed the carrier density
lower arms from an initial unsaturated state, and consequently it
perceives a gain exposed to a greater dynamic strain.
∆φ DI (t) = φin, DI (t) − φin, DI (t − ∆τ) ) (5)

Ubiquitous Computing and Communication Journal 4


significantly reduce the energy of the input signal,
which however is not an appropriate solution as it
suffers from the severe drawback of the output
signal-to-noise-ratio degradation [7], [10].
Nevertheless, with the increase of the temporal delay
offered by the DI the curve is progressively shifted
downwards, which is a highly desirable feature since
it indicates that thanks to this element the AM can be
levelled-off. Moreover, its slope is reversed and the
AM becomes smallest for a given ∆τ at the rightmost
side of the energy values. This behavior is attributed
to the fact that the more energetic data pulses
undergo a larger phase variation as they pass through
the SOA. This in turn is accompanied according to
Eq. (5) by the formation in the DI of a phase
Figure 5: Pattern-dependent amplified output after difference window of higher magnitude inside which
the SOA alone in response to 11010001 the required interference conditions can be met more
efficiently than for the case of a weaker input [27].
It is thus necessary to explore under which operating Furthermore, the AM can be reduced and stay below
conditions the desired requirement for the AM can 1 dB regardless of the exact amount of the pulse
be fulfilled in a single SOA and also to identify how energy inserted in the SOA and the associated
the addition of a DI can positively contribute to the saturation regime in which it lies, because the
accomplishment of the same target. For this purpose, suppression mechanism has a wider margin to be
each one of the critical parameters is scanned within properly established. For example the value ∆τ = 3
a realistic range of values, while the temporal delay ps drops the AM at the two energy extremes to 0.64
inserted by the DI is varied in incremental steps, so dB and 0.23 dB compared to 1.16 dB and 1.2 dB
as to interpret its AM curve. without the DI, respectively. This essentially means
The graph of the AM against the energy of the that the connection of a DI after a SOA, which
data pulses that are inserted in the SOA is presented directly amplifies RZ picosecond pulses,
in Fig. 6. compensates for the pattern dependent distortion on
the output signal in an increased input energy
dynamic range where operation is extended to the
saturation regime.
The calculated AM as a function of the FWHM
of the input data pulses is plotted in Fig. 7 bearing in
mind that it must be kept over the maximum ∆τ in
order for the pulse delayed in the upper arm of the DI
to meet at the output coupler the direct pulse from
the lower arm and interfere with it properly [28].

Figure 6: AM variation vs. energy of input data


pulses for different ∆τ

Initially, it is seen that when the SOA is alone the


examined metric is unacceptable in the entire span
across which the specific parameter is varied,
because as the latter is increased the gain excursions
between the recovered and saturated states are
enlarged aggravating the peak power fluctuations.
The net result is that the SOA cannot be driven by an Figure 7: AM dependence on FWHM of input data
optical signal of strong or even moderate energy pulses for different ∆τ
relative to its nominal saturation energy and the only
way of avoiding the pattern effect would be to Before proceeding with its interpretation, it must be

Ubiquitous Computing and Communication Journal 5


noted that for the NRZ pulses used in [14] this evidence that Tcarrier is the main constraining factor
parameter is of no significance since by definition for the amplification performance of a single SOA
the power content of a bit ‘1’ stays high throughout [2].
its entire time slot. On the contrary, for the RZ
format used in this work, the duration of the pulses is
shorter than the bit period by a fraction that is
determined by the duty cycle and subsequently the
FWHM, whose contribution must be thus definitely
taken into account. Initially it is observed that
although the AM is affected when the DI is
employed, however its dependence on ∆τ is weak
when the FWHM is comparatively much larger. This
is the case for pulse width values near the rightmost
side of the diagram where regardless of ∆τ the AM
remains inadmissible. From a physical standpoint
this situation is justified by noting that the intensity
content of a wider data pulse affects for a longer time
the SOA optical properties whose amendment is thus
protracted to a point where the associated gain
fluctuations between the marks cannot be
Figure 8: AM change against SOA carrier lifetime
compensated even with the use of a DI.
for different ∆τ
Nevertheless, with the decrease of the FWHM, the
gain transitions from the initial and high to the final
Nevertheless, Fig. 8 reveals the appealing feature
and smaller dynamic state become steeper and are
that thanks to the use of a DI it is not necessary for
thus averaged to the same level for all ‘1’s, which
the SOA to recover completely between successive
leads to an enhanced degree of amplitude
data pulses in order to equalize their different output
suppression. Therefore, in each one of the derived
amplitudes, as it would be definitely required in the
curves, the more the pulse width is reduced towards
absence of this passive element. Instead, the recovery
the corresponding temporal offset of the DI, the more
time can be slower and combat à posteriori the
the AM falls taking its minimum among them when
inevitable pattern dependencies that manifest in this
the FWHM comes closer to the upper bound of the
case at the SOA exit through the proper adjustment
considered ∆τ, i.e. for T = 5 ps . Interestingly of the temporal mismatch between the paths of the
FWHM
these findings are in qualitative agreement with those DI. This possibility is offered by noting that a higher
derived from the numerical study of an all-optical Tcarrier can be tolerated for a given AM if ∆τ is
packet self-synchronization scheme, which also larger, and inversely the AM is improved if ∆τ is
exploits a SOA in tandem with a DI [28]. However, increased while Tcarrier remains constant. These
although the aforementioned FWHM value that
observations suggest thus to choose the maximum ∆τ
yields the best performance can be practically
= 3 ps and then indeed the AM curve for ∆τ = 0 ps is
provided from state-of-the-art mode-locked
transferred under the critical limit of 1 dB across a
semiconductor ring lasers, yet this comes at the
range of Tcarrier that extends up to almost two times
expense of an increased operational complexity.
Here is where the use of a DI can be transformed into the repetition interval. Physically this happens
a technological advantage and allow to maintain the because the phase window that is built-up in the DI
AM less than 1 dB with a broader pulse, which can is wider for a larger ∆τ and therefore can tolerate a
be produced by a much simpler source, such as a greater wandering of the phase shifts of the ‘1’s
gain-switched distributed feedback semiconductor dispersed inside the incoming PRBS [27] so that the
diode laser after linear compression in a dispersion level of their amplified amplitudes can be controlled
compensating fiber of appropriate length. In fact, an and balanced by the DI despite the finite Tcarrier .
optical delay of ∆τ = 3 ps in combination with The feasibility thus of combating at 20 Gb/s the
T = 7 ps results in the more than adequate AM restrictive effect of the carrier lifetime by means of a
FWHM
practically realizable DI is very attractive as it allows
of 0.49 dB. to overcome several obstacles related to the
The dependence of the AM on the carrier photonics technology that deals directly with this
lifetime is displayed in Fig. 8. As it can be fundamental issue. More specifically, there is no
characteristically seen in case of a single SOA the imperative need for exploiting at this working rate
considered metric is very vulnerable to the variations gain recovery acceleration techniques [5-6], given
of the specific parameter since it is dramatically that their implementation is not straightforward and
aggravated when the latter deviates away from the easy but demands the careful adjustment of the
operating period. This fact confirms the experimental

Ubiquitous Computing and Communication Journal 6


power and the proper selection of the wavelength of signal gain that limits its usability can by overcome
the continuous-wave holding beam that is by the crosswise subtraction of a delayed copy of the
responsible for speeding-up the carrier lifetime. amplified signal from itself using the DI. This action
Furthermore, it is not necessary to raise the injection opens a differential window [27] whose height can
current of the SOA [4], thus avoiding the use of be regulated through the selection of ∆τ so that the
costly, complex and power consuming current phase excursions induced from those of the gain
sources with the associated driving electronics. inside the SOA are forced to occur around the same
Finally, it is not required to adopt the solution of a average level. In this manner the intensity
longer SOA [4], which is a favorable option fluctuations between the ‘1’s are eliminated
concerning the contribution of the amplified irrespectively of the order of their appearance in the
spontaneous emission (ASE) since then this effect is PRBS and thereby these logic symbols acquire a
weak and cannot harm the gain saturation behavior balanced peak power, which is translated into a
of the SOA or the signal-to-noise ratio of the diminished modulation depth and a more symmetric
amplified data pulses [4]. All the aforementioned output power profile. This is actually verified in Fig.
benefits highlight the catalytic role played by the DI 9 since as the time delay grows at a rate of 1 ps the
in neutralizing the impact of the slow carrier lifetime AM curve is gradually tilted towards the horizontal
and constitute the best proof of the merit of this axis becoming roughly independent of the single-
element when it is utilized in the proposed pattern pass power gain. Therefore, the specific temporal
suppression concept. condition in the DI widens the range of gains that the
The evolution of the AM versus the small signal SOA can provide and enhances its flexibility as
gain is portrayed in Fig. 9, which unveils that when amplifying element.
the SOA is on its own this metric is unacceptable The influence of the linewidth enhancement
across the whole span from 22 dB to 30 dB. factor on the AM is depicted in Fig. 10 and it can be
interpreted by recalling that this parameter links from
Eq. (3) the fluctuations of the phase of a data signal
inside a SOA with those of the gain from which they
have been provoked.

Figure 9: Evolution of AM as a function of SOA


small signal gain for different ∆τ

An insight to this comportment can be gained by


recalling that with a given pulse energy and width as
well as carrier lifetime the gain variation is more Figure 10: Influence of SOA linewidth enhancement
steep and intense for a higher G ss [29]. Accordingly, factor on AM for different ∆τ
the SOA properties are perceived within a broader This fact in combination with the principle of
dynamic area so that the discrepancy between the operation of the DI, which relies on the creation of
dropped and risen gains is expanded, which in turn the proper phase difference between its two arms,
exacerbates the dependence of the amplified output essentially means that the alpha-factor is expected to
on the value of the preceding input data bits and has have an impact on the overall performance only in
thus an adverse impact on the achievable AM. This the presence of this element, as it is indeed evident
essentially means that a SOA could be operated in its from Fig. 10 for ∆τ = 0 ps, where the AM remains
classical role without causing pattern-dependent degraded to a constant level over 1 dB no matter the
degradation only if it would be biased to operate in value of α. However, things change with the
its unsaturated state, where the gain is a linear introduction of an optical delay since for every ∆τ
function of the input power, or equivalently provide the AM is ameliorated with the increase of α. The
a small single pass power gain. However, this failure explanation for this improvement is that higher
of the SOA to be operated with a large nominal small

Ubiquitous Computing and Communication Journal 7


values of this parameter incur stronger phase shifts in the reduction of the AM is accompanied by an
the SOA, which in turn are followed by maximum amplification drop at the output PRBS from 8.67 dB
phase window excursions closer to the desired to 7.16 dB. Still, this can be practically tolerated
degree of phase matching at the output of the DI [27]. without compromising the contribution of the DI and
Conversely, when α is not varied the AM can be the significant improvement offered by its use, which
reduced by augmenting ∆τ, which is an action that is can become even more pronounced if the specific ∆τ
equivalent to controlling the width of opening or is combined with a selection of E data , T ,
FWHM
closure of the differential phase window and hence
the extent of pulse-to-pulse smoothing. In other Tcarrier , G ss and α, suggested according to Figs. 6-
words, there is an interplay between α and ∆τ, which 10. For example with the set of values 500 fJ, 8 ps,
allows to achieve a reasonable AM by altering either 100 ps, 28 dB and 8, respectively, the marks that
one of them while keeping the other fixed. For have been unevenly transmitted by the SOA are
example, an AM less than 1 dB can be obtained for equalized, as it is apparent in Fig. 12, and the AM is
the pairs α = 4, ∆τ = 2 ps and α = 8, ∆τ = 1 ps, whilst further reduced to approximately 0.22 dB.
almost full suppression is accomplished by utilizing
together the highest values of these parameters
within their examined range, namely α = 10 and ∆τ =
3 ps.
From the examination and analysis of Figs. 6-10
it can be deduced that the most appropriate amount
of the temporal offset between the two arms of the
interferometer that effectively satisfies the AM
requirement for all the critical parameters is 3 ps.
The suitability of this choice is proved indeed by the
reduction of the AM obtained for a single SOA and
the fixed values used throughout the simulation from
1.2 dB to 0.47 dB as well as by the accompanying
amelioration in the quality of the amplified data
stream shown in Fig. 11.

Figure 12: Elimination of the pattern effect in the


amplified data stream of Fig. 5 for ∆τ = 3 ps and
critical parameters selection suggested from Figs. 6-
10

In contrast, the achievement of the same AM using a


stand-alone SOA would be possible only at the cost
of stricter operational requirements, namely a much
lower injected energy, E data = 100 fJ , a narrower
pulse width, T = 5 ps , a four times smaller
FWHM
SOA small signal gain, G ss = 22 dB , and, above all,
a faster and shorter than the period of the 20 Gb/s
PRBS SOA carrier lifetime, Tcarrier = 35 ps .
Figure 11: Partial compensation of the pattern effect
in the amplified data train of Fig. 5 when a DI with 5 FREQUENCY DOMAIN APPROACH
∆τ = 3 ps is serially connected to the SOA exit
A useful approach that can provide a different
However, although the difference compared to the insight on the principle of operation of the proposed
significant lack of uniformity in Fig. 5 is apparent, pattern suppression technique and complement the
the price paid is that the data sequence receives less interpretation of the obtained results is to view the
amplification. This finding complies with other concept from the perspective of the frequency
reports on techniques aiming at alleviating the domain. In this case the relevant power transfer
problem governing straightforward SOA function of the delay interferometer at its crossed-
amplification, since they also have reported the coupled output port is [31]
inevitable existence of such a penalty in order to
attain the desired goal [10-11], [30]. In effect the 1
relevant comparison of Figs. 5 and 11 verifies that
2
H DI (ω) = [1 + cos(∆τω)] (7)
2

Ubiquitous Computing and Communication Journal 8


magnitude is comparatively smaller. The red chirp is
related to carrier depletion and gain compression
whilst the blue chirp is linked to carrier regeneration
and gain recovery [33]. This in turn means that their
magnitude is proportional to the rate of change of the
carrier concentration and subsequently of the gain.
For this reason it would be rationally expectable that
the pattern effect observed in the gain variation
would also be imposed on both of them. However, as
it can be seen in Fig. 14, this holds almost
exclusively for the red chirp whose peaks suffer
indeed from strong fluctuations unlike those of the
blue one. In the latter case an increase in the peaks
occurs only for successive ‘1’s, since then the carrier
recombination rate is enhanced due to the weaker
carrier depletion, but nevertheless it is so small that it
Figure 13: Power transfer function of DI
can hardly be noticed.
which is depicted in Fig. 13. Due to its cosinusoidal
form it can be realized that the device acts as a notch
filter with the null points of its spectrum situated at
± (2k+1)/(2∆τ), k ∈ Z , relative to the optical
carrier frequency, whereas the distance between the
transmission peaks, which defines the free spectral
range (FSR), is equal to the inverse of the inserted
delay [31]. Therefore, with the proper modification
of ∆τ the periodic comb-like profile of the DI can be
controlled to attenuate and hence suppress any
undesirable frequency components in its incident
signal, which is clearly an attractive property that
could be well exploited to alleviate the pattern effect
in an ultrafast data stream amplified by a SOA. This
is possible in the case of interest due to the frequency
chirp induced from the manifestation of self-phase Figure 14: Simulated chirp at the SOA output
modulation (SPM) in a pulse when its energy is
increased towards the saturation level of a SOA This pronounced discrepancy in the pattern
through which it is propagated [19]. This is dependence is attributed to the different time scale
mathematically described by the time derivative of on which the two chirping phenomena take place. In
the phase response according to the expression [19] fact the carrier depletion and the accompanying
phase increase in the leading edge of the pulse is
1 dφout, SOA (t) (8) rather faster compared to the much slower carrier
∆ν out, SOA (t) = −
2π dt replenishment and the concomitant phase decrease in
its trailing edge, which gives rise to an
which is plotted in Fig. 14 for the same transform- approximately constant amount of blue chirp.
limited input data pattern frame and parameter values Therefore the explanation of the pattern effect from
used in Figs. 2 and 3 while calculating φ out, SOA (t) the frequency domain standpoint can be given by
from Eq. (3) by means of the established numerical focusing on the red chirp form in Fig. 14 and
methodology. As it can be seen the ‘on’ pulses correlating it with the relevant gain change in Fig. 2
acquire a chirp that is significantly negative at their [34]. More specifically, the rising edge of the first ‘1’
leading edge, almost linear around their main part acquires the largest red chirp among all the marks
and somewhat positive at their trailing edge. These owing to the fact that for the same energy its gain is
characteristics confirm the experimental evidence dropped from the maximum possible level, i.e. that
available for the pure amplification of an ultrafast of the unsaturated state, and as a by-product the
RZ pulse train in a single SOA [32]. Moreover, they incurred gain difference is the highest. Similarly the
designate that the spectral components corresponding next peak is that of the last ‘1’ as it is preceded by a
to the leading edge are shifted enough to the low run of three ‘0’s. In contrast the lower the carrier
sideband (red chirp), whereas those corresponding to density, or equivalently the greater the extent of gain
the trailing edge remain closer to the position of the compression, the slower additional carriers are
input carrier frequency since their maximum depleted, thereby resulting in decreased gain

Ubiquitous Computing and Communication Journal 9


excursion and accordingly less red chirp, as it is the loss [35]. These implications designate that the
case for consecutive ‘1’s where the SOA cannot fully important benefits offered by the scheme are
recover in between them. In this context it is realized partially compromised by an inevitable amplification
that the root of the problem is the broadening of the reduction, which is a penalty that was also verified in
spectrum to the low frequency region and the the time domain.
unequal peaks of the red chirp, which in turn
suggests that if they could be suppressed then the
issue of patterning would be resolved. Clearly the
most direct means of achieving this goal is by
filtering out the red-shifted sideband while not
impairing the frequency of the inserted signal.
Concurrently the spectral component of the trailing
edge must be favored but without being converted
into amplitude information. Here is where the DI can
be employed to satisfy these requirements and
eliminate the distortion of the pulses exiting the SOA
by exploiting the frequency discrimination capability
that it exhibits thanks to the shape of its transfer
function. For this purpose its frequency at
transparency is positively detuned, through the
proper selection of the time delay, a bit higher, i.e.
by a small fraction of the FSR, than that at which the
introduced signal is centered [35]. This essentially
Figure 15: Simulated chirp at the DI output for the
implies that as the frequency decreases so does the
same ∆τ and parameter values as in Fig. 12
transmittance and accordingly the unwanted red
spectral components are attenuated in direct analogy
6 CONCLUSION
to the degree they have been shifted. In this manner
the variations in the red chirp amplitudes are
In summary, the feasibility of equalizing with
drastically clamped and in particular the more ∆τ is
the use of a DI the uneven pulse amplitudes
grown the more they are eliminated. This happens
occurring at the output of a stand-alone SOA when it
because then the filter’s response becomes steeper
is excited by a RZ repetitive data train that is faster
and the corresponding spectral components are
than its natural carrier lifetime has been theoretically
gradually pushed to fall around its notches, thus
explored and demonstrated by means of numerical
suffering greater loss and being blocked regardless of
simulation. Based on the interpretation of a set of
the exact mark location they stem from. The
graphical information for the amplitude modulation
observation of Figs. 6-10 confirms indeed this fact
in the presence or not of the DI it has been possible
since the augmentation of ∆τ for given involved
to evaluate the influence exerted by the involved
parameters improves the performance, while as it
critical parameters on this quality metric and figure-
was found among its scanned values the highest is
out how to achieve its improvement. In this manner
the preferred one in order to satisfy the targeted
it has been inferred that if the temporal delay inserted
criterion for the AM in a widened operational range.
by the DI is properly chosen the peak power
The beneficial action of the DI filtering on the chirp
fluctuations in the directly amplified data signal can
is illustrated in Fig. 15. This has been obtained by
be balanced and the profile of the latter become
numerically calculating, according to the definition
symmetrical. Furthermore, the requirements that
of this measure, the time derivative of the phase
must be fulfilled in order to avoid the consequences
difference from Eq. (5), which is a process reduced
of the deleterious pattern effect are relaxed and the
through Eq. (3) to the knowledge of both dh(t)/dt and
useful operational range of the involved key
of its delayed version from Eq. (1). Indeed the red
parameters is extended. With the aid thus of the DI
chirp is reduced and its fluctuations are strongly
the required performance criterion set for the AM
suppressed, as opposed to Fig. 14, but it must be
can be satisfied at 20 Gb/s by a SOA that operates
noted that inevitably the useful information
over an expanded saturation regime in terms of its
contained in these spectral components is lost. This
small signal gain or the energy of its driving data
rejection also concerns the blue ones since as it can
pulses, while the latter are simultaneously allowed to
be seen they are transferred nearly on the zero chirp
exhibit a wider FWHM. Furthermore, the intrinsic
level. This happens because for the increased ∆τ
characteristic of the slow carrier lifetime that
required to achieve better pattern-independent
ultimately limits the performance of conventional
operation they are forced to lie in a portion of the
SOAs no longer constitutes a technical barrier and
positive slope DI transfer function that can be
despite its existence the pattern dependencies can be
considered linear, thus suffering a greater insertion
mitigated by means of the DI. These results show

Ubiquitous Computing and Communication Journal 10


that the proposed scheme contributes decisively to pp. 12-14 (2002).
the cancellation of the pattern effect but at the [6] W. Ji, M. Zhang, P. Ye: Simulation of an All-
expense of a reduction in data signal amplification. Optical XOR Gate with a Semiconductor Optical
Nevertheless, this penalty is practically tolerable and Amplifier Mach-Zehnder Interferometer Sped
cannot impair the significance of the aforementioned Up by a Continuous-Wave Assistant Light,
operational benefits, which suggest that a DI together Journal of Optical Networking, Vol. 4, pp. 524-
with its simplicity of implementation, its integration 530 (2005).
potential and its cost-effectiveness can constitute a [7] K. Chan, C.-K. Chan, L.K. Chen, F. Tong:
powerful solution for combating the pattern effect in Mitigation of Pattern-Induced Degradation in
a SOA deployed in its classical amplifying role. The SOA-Based All-Optical OTDM Demultiplexers
contribution of this stage was also explained in the by Using RZ-DPSK Modulation Format,
frequency domain since the DI essentially is a Photonics Technology Letters, IEEE, Vol. 15, pp.
filtering element. In this context it was shown that if 1264-1266 (2003).
the minimum of its transfer function is suitably tuned [8] M.L. Nielsen, J. Mørk: Increasing the
to remove the fluctuations in the instant negative Modulation Bandwidth of Semiconductor-
frequency deviation associated with the phase Optical-Ampifier-Based Switches by Using
modulation at the SOA output then it can be usefully Optical Filtering, Journal of the Optical Society
exploited to restore the distorted profile of the of America B, OSA, Vol. 21, pp. 1606-1619
amplified data pulses. Moreover, the main (2004).
implications derived from this approach were found [9] Y. Liu, E. Tangdiongga, Z. Li, S. Zhang, H. de
to be consistent with those in the time domain, thus Waardt, G.D. Khoe, H.J.S. Dorren: Error-Free
allowing to understand the role of the DI in a All-Optical Wavelength Conversion at 160 Gb/s
spherical manner. The theoretical confirmation of Using a Semiconductor Optical Amplifier and an
the competence of the considered scheme to Optical Bandpass Filter, Journal of Lightwave
successfully handle RZ pulses arriving as twice as Technology, IEEE/OSA, Vol. 24, pp. 230-236
often than previously reported [36] constitutes in (2006).
addition a proof of its speed transparency up to the [10] Q. Xu. M. Yao, Y. Dong, W. Cai., J. Zhang:
specific extended bit rate. Therefore, the pattern-free Experimental Demonstration of Pattern Effect
performance of straightforward amplification in Compensation Using an Asymmetrical Mach-
single SOAs can be enhanced with direct by-product Zehnder Interferometer With SOAs, Photonics
the better exploitation of these devices for the Technology Letters, IEEE, Vol. 13, pp. 1325-
implementation of higher capacity photonic 1327 (2001).
communication systems and networks. [11] K. Chan, C.-K. Chan. W. Hung, F. Tong, L.K.
Chen: Waveform Restoration in Semiconductor
7 REFERENCES Optical Amplifier Using Fiber Loop Mirror,
Photonics Technology Letters, IEEE, Vol. 14, pp.
[1] J. Mørk, M.L. Nielsen, T.W. Berg: The 995-997 (2002).
Dynamics of Semiconductor Optical Amplifiers, [12] R.J. Manning, X. Yang, R.P. Webb, R. Giller,
Optics and Photonics News, OSA, Vol. 14, pp. F.C.G. Gunning, A.D. Ellis: The Turbo-Switch-
42-48 (2003). A Novel Technique to Increase the High-Speed
[2] M.J. Connelly: Semiconductor Optical Response of SOAs for Wavelength Conversion,
Amplifiers, Kluwer Academic Publishers, pp. Proceedings OFC, paper OWS8 (2006).
108-120 (2002). [13] A.V. Uskov, T.W. Berg, J. Mørk: Theory of
[3] R.J. Manning, A.D. Ellis, A.J. Poustie, K.J. Pulse-Train Amplification Without Patterning
Blow: Semiconductor Laser Amplifiers for Effects in Quantum-Dot Semiconductor Optical
Ultrafast All-Optical Signal Processing, Journal Amplifiers, Journal of Quantum Electronics,
of the Optical Society of America B, OSA, Vol. IEEE, Vol. 40, pp. 306-320 (2004).
14, pp. 3204-3216 (1997). [14] C.S. Wong, H.K. Tsang: Reduction of Bit-Patten
[4] L. Schares, C. Schubert, C. Schmidt, H.G. Dependent Errors from a Semiconductor Optical
Weber, L. Occhi, G. Guekos: Phase Dynamics of Amplifier Using an Optical Delay Interferometer,
Semiconductor Optical Amplifiers at 10-40 GHz, Optics Communications, Vol. 232, pp. 245-249
Journal of Quantum Electronics, IEEE, Vol. 39, (2004).
pp. 1394-1408 (2003). [15] T.-Y. Kim, M. Hanawa, S.-J. Kim, S. Hann, Y.H.
[5] J.L. Pleumeekers, M. Kauer, K. Dreyer, C. Kim, W.-T. Han, C.-S. Park: Optical Delay
Burrus, A.G. Dentai, S. Shunk, J. Leuthold, C.H. Interferometer Based on Phase Shifted Fiber
Joyner: Acceleration of Gain Recovery in Bragg Grating With Optically Controllable
Semiconductor Optical Amplifiers by Optical Phase Shifter, Optics Express, OSA, Vol. 14, pp.
Injection Near Transparency Wavelength, 4250-4255 (2006).
Photonics Technology Letters, IEEE, Vol. 14, [16] R. Ludwig, U. Feiste, E. Dietrich, H.G. Weber,

Ubiquitous Computing and Communication Journal 11


D. Breuer, M. Martin, F. Kuppers: Experimental [27] R. Gutiérrez-Castrejón, M. Dülk, St. Fischer, G.
Comparison of 40 Gbit/s RZ and NRZ Guekos: Novel Scheme for Optical Time-
Transmission Over Standard Singlemode Fiber, Division Demultiplexing Using a Delayed
Electronics Letters, IEE, Vol. 35, pp. 2216-2218 Interferometer, Optics Communications, Vol.
(1999). 192, pp. 245-254 (2001).
[17] C. G. Lee, Y.J. Kim, C. S. Park, H. J. Lee, C.-S. [28] X. Huang, P. Ye, M. Zhang, L. Wang: A Novel
Park: Experimental Demonstration of 10-Gb/s Self-Synchronization Scheme for All-Optical
Data Format Conversions Between NRZ and RZ Packet Networks, Photonics Technology Letters,
Using SOA-Loop-Mirror, Journal of Lightwave IEEE, Vol. 17, pp. 645-647 (2005).
Technology, IEEE/OSA, Vol. 23, pp. 834-841 [29] K.E. Zoiros, R. Chasioti, C.S. Koukourlis, T.
(2005). Houbavlis: On the Output Characteristics of a
[18] P. Ghelfi, M. Secondini, M. Scaffardi, F. Fresi, Semiconductor Optical Amplifier Driven by an
A. Bogoni, L. Potì: Impact of an Additional All- Ultrafast Optical Time Division Multiplexing
Optical Decision Element in Band-limited Pulse Train, Optik, Vol. 118, pp. 134-146 (2007).
Receivers for RZ Systems, Journal of Lightwave [30] R. Gutiérrez-Castrejón, A. Filios: Pattern-Effect
Technology, IEEE/OSA, Vol. 25, pp. 1728-1734 Reduction Using a Cross-Gain Modulated
(2007). Holding Beam in Semiconductor Optical In-Line
[19] G.P. Agrawal, N.A. Olsson: Self-Phase Amplifier, Journal of Lightwave Technology,
Modulation and Spectral Broadening of Optical IEEE/OSA, Vol. 24, pp. 4912- 4917 (2006).
Pulses in Semiconductor Laser Amplifiers, [31] X. Tang, N.Y. Kim, J.C. Cartledge: Noise
Journal of Quantum Electronics, IEEE, Vol. 25, Transfer Characteristics in a Semiconductor
pp. 2297-2306 (1989). Optical Amplifier With Application to
[20] H. Dong, G. Zhu, Q. Wang, H. Sun, N.K. Dutta, Wavelength Conversion Based on a Delay
J. Jaques, A.B. Piccirilli: Multiwavelength Fiber Interferometer, Journal of Lightwave
Ring Laser Source Based on a Delayed Technology, IEEE/OSA, Vol. 26, pp. 1715-1721
Interferometer, Photonics Technology Letters, (2008).
IEEE, Vol. 17, pp. 303-305 (2005). [32] M. Matsuura, N. Iwatsu, K. Kitamura, N. Kishi:
[21] J.M. Tang, K.A. Shore: Strong Picosecond Tine-Resolved Chirp Properties of SOAs
Optical Pulse Propagation in Semiconductor Measured With an Optical Bandpass Filter,
Optical Amplifiers at Transparency, Journal of Photonics Technology Letters, IEEE, Vol. 20, pp.
Quantum Electronics, IEEE, Vol. 34, pp. 1263- 2001-2003 (2008).
1269 (1998). [33] G. Girault, A.M. Clarke, D. Reid, C. Guignard,
[22] A. Mecozzi, J. Mørk: Saturation Induced by L. Brmerie, P. Anandarajah, L.P. Barry, J.-C.
Picosecond Pulses in Semiconductor Optical Simon, J Harvey: Analysis of Bit Rate
Amplifiers, Journal of the Optical Society of Dependence up to 80 Gbit/s of a Simple
America B, OSA, Vol. 14, pp. 761-770 (1997). Wavelength Converter Based on XPM in a SOA
[23] K.E. Zoiros, G. Papadopoulos, T. Houbavlis, and a Shifted Filtering, Optics Communications,
G.T. Kanellos: Theoretical Analysis and Vol. 281, pp. 5731-5738 (2008).
Performance Investigation of Ultrafast All- [34] J. Wang, A. Marculescu, J. Li, P. Vorreau, S.
Optical Boolean XOR Gate With Semiconductor Tzadok, S.B. Ezra, S. Tsadka, W. Freude, J.
Optical Amplifier-assisted Sagnac Leuthold: Pattern Effect Removal Techniue for
Interferometer, Optics Communications, Vol. Semiconductor Optical Amplifier Based
258, pp. 114-134 (2006). Wavelength Conversion, Photonics Technology
[24] C. Botsiaris, K.E. Zoiros, R. Chasioti, C.S. Letters, IEEE, Vol. 19, pp. 1955-1957 (2007).
Koukourlis: Q-factor Assessment of SOA-Based [35] T. Watanabe, H. Yasaka, N. Sakaida, M. Koga:
Ultrafast Nonlinear Interferometer, Optics Waveform Shaping of Chirp-Controlled Signal
Communications, Vol. 278, pp. 291-302 (2007). by Semiconductor Optical Amplfier Using
[25] N.Y. Kim, X. Tang, J.C. Cartledge, A.K. Atieh: Mach-Zehnder Frequency Discriminator,
Design and Performance of an All-Optical Photonics Technology Letters, IEEE, Vol. 10, pp.
Wavelength Converter Based on a 1422-1424 (1998).
Semiconductor Optical Amplifier and Delay [36] K.E. Zoiros, T. Siarkos, C.S. Koukourlis:
Interferometer, Journal of Lightwave Performance Investigation of Pattern Effect
Technology, IEEE/OSA, Vol. 25, pp. 3730-3738 Suppression in Semiconductor Optical Amplifier
(2007). Using Optical Delay Interferometer, Proceedings
[26] J.S. Vardakas, K.E. Zoiros: Performance CSNDSP, pp. 680-684 (2008).
Investigation of All-Optical Clock Recovery
Circuit Based on Fabry-Pérot Filter and SOA-
Assisted Sagnac Switch, Optical Engineering,
SPIE, Vol. 46, pp. 1-21 (2007).

Ubiquitous Computing and Communication Journal 12

Vous aimerez peut-être aussi