多波長環境におけるSOAスイッチのダイナミックレンジ評価

社団法人 電子情報通信学会
THE INSTITUTE OF ELECTRONICS,
INFORMATION AND COMMUNICATION ENGINEERS
信学技報
TECHNICAL REPORT OF IEICE.
多波長環境における SOA スイッチのダイナミックレンジ評価実験
李
慧†
今泉
英明†
種村
中野 義昭†
森川
博之†
拓夫†
† 東京大学
〒 153–8904 東京都目黒区駒場 4–6–1
E-mail: †[email protected]
あらまし
本稿では，SOA 型光スイッチに基づいて多波長光パケット交換方式を実現する上で問題となる，SOA 型光
スイッチの非線形効果による波長数制約に関して特性評価を行う．超広帯域インターネットを実現する上で，低消費
電力で動作可能な光パケット交換技術の実現が必須となる．その中でも特に，多波長光パケット交換技術は，1 つの
パケットを複数の波長に分割・符号化し，波長依存性の低い光スイッチにより一括で交換することで，光パケット交
換ノードを構成する光デバイス数を劇的に減少させる．そのため，単波長光パケット交換に比べて飛躍的に消費電力
を低減できる可能性がある．高速応答性と高消光比の観点から，我々は SOA 型光スイッチを用いた多波長光パケット
交換技術を検討している．SOA 型光スイッチは，大規模化が容易なブロードキャストアンドセレクト方式と高い親和
性を備えるため，光パケット交換に適していると考えられる．しかし一方では，入力信号強度が強くなると非線形光
学効果によって光信号雑音比 (OSNR) が劣化するため，入力される波長数や波長間隔に制約を与える．本稿では，光
信号雑音と利用する波長数，波長間隔，SOA 型光スイッチのゲインなどの要素間の関係を明らかにするために，入力
信号強度のダイナミックレンジ評価を行った．その結果，SOA 型光スイッチのゲインが 6dB のとき，100GHz 間隔の
15 波長でダイナミックレンジが 15dB であり，ゲインを更に小さくすれば十分な波長数が確保できることが分かった．
キーワード
多波長光パケット交換，SOA 型光スイッチ, 入力パワーダイナミックレンジ
Experimental Evaluation on Dynamic Range of SOA Switch for
Multi-Wavelength Optical Packet Switching
Hui LI† , Hideaki IMAIZUMI† , Takuo TANEMURA† ,
Yoshiaki NAKANO† , and Hiroyuki MORIKAWA†
† The University of Tokyo
Komaba 4–6–1, Meguro-ku, Tokyo, 153–8904 Japan
E-mail: †[email protected]
Abstract In this paper, we aim to investigate the way for wavelength scalability guarantee while applying SOA switch to
MW-OPS, considering the influence caused by the non-linear phenomenon of SOA switch. Optical packet switching (OPS)
with low power consumption is required for realizing ultra-high bandwidth Internet. Especially, MW-OPS has been researched
recently as a promising technology to provide high capacity with less number of optical devices due to its property that wavelength-multiplexed optical packets are switched with wide-band optical switches, and SOA switches are viable for MW-OPS
in terms of ultra-fast switching speed and high extinction ratio. Although SOA switches have high affinity to scalable broadcast-and-selection architecture, optical signal to noise ratio (OSNR) becomes wore due to the non-linear phenomenon for high
power input signals. In this paper, in order to investigate the relationship between OSNR and the parameters such as the number of wavelengths utilized, wavelengths intervals, and the optical gain provided by SOA switch, we experimental evaluate the
dynamic range of input power into SOA switch for MW-OPS. The results show that sufficient number of wavelengths could
be utilized with lower gain.
Key words multi-wavelength optical packet switching (MW-OPS), semiconductor optical amplifier (SOA) switch, dynamic range
—1—
1. Introduction
In recent years, with the popularity of the broadband access technology, the rate of Internet traffic amount increasing
concludes the paper.
2. SOA switch for Multi-Wavelength Optical
Packet Switching
is higher than Moore’s law and is being accelerated. The average volume of traffic data per month in the Internet grew
In this section, we first introduce the MW-OPS and then
from 75.25 million to 11.72 billion gigabytes between the
discuss the optical switches appropriate to the MW-OPS.
years 2000 and 2009 [1] while the use of Internet has grown
SOA switches are viable for MW-OPS in terms of ultra-fast
an impressive 380.3% in the period 2000 − 2009 [2]. This
switching speed, high extinction ratio and low crosstalk.
means high-capacity network is badly in need.
In terms of transport, WDM technology has promise for
2. 1 Multi-Wavelength Optical Packet Switching
accommodating such enormous traffic due to its tremendous
As mentioned above, optical packet switching (OPS) has
capacity which can exceed 30Tb/s nowadays [3]. However, if
been researched with the aim of efficiently utilizing the broad
we simply expand the current routers based on electric pro-
bandwidth with low power consumption and high bandwidth
cessing trying to support the huge bandwidth required in fu-
utilization. In a network based on ordinary packet OPS ar-
ture Internet, total power consumption would grow out of re-
chitectures, an optical packet consisting of a header and a
alistic quantity [4].
payload is generally encoded into a single wavelength. An
In order to address this issue, optical packet switching
independent optical packet is, therefore, encoded into each
(OPS) has been researched in these two decades [5]. OPS
wavelength available within a single fiber. This property
networks can support the packet switching-based services
makes several kinds of devices necessary for packet forward-
without O/E conversion with low power consumption. Espe-
ing (e.g., header processing and contention resolution de-
cially, multi-wavelength optical packet switching (MW-OPS)
vices) proportional to W × P , where W is the number of
has been researched recently as a promising technology to
wavelengths available in the WDM network and P is the
provide high capacity with less number of optical devices due
number of input ports in the core node. The increase of num-
to its property that wavelength-multiplexed optical packets
ber of these devices could make the node implementation un-
are switched with wide-band optical switches [6], [7].
realistic in terms of physical size and cost especially for the
SOA switches have been shown to be high extinction, low
crosstalk, and fast switching speed [8] and have been al-
future WDM networks where the number of available wavelengths in a single fiber could be over a thousand [9].
ready applied to several demonstrations of MW-OPS [10]∼
Multi-Wavelength OPS (MW-OPS) has been researched as
[13]. With the SOA gate switches and couplers, packet
one solution to reduce the number of these devices. Core
switching based on broadcast-and-selection mechanism can
nodes based on MW-OPS use optical switches with lower
be achieved at large scale easily. However, some effect of
wavelength dependency and forward optical packets encoded
the non-linear optical phenomenon such as four-wave mix-
into multiple wavelengths. For instance, header data is en-
ing (FWM) can make the optical signal noise ratio (OSNR)
coded into a wavelength and payload data is encoded into
worse for high power input signals. That means, investigat-
other wavelengths (Fig. 1). This property makes the neces-
ing the performance of SOA switches for MW-OPS is very
sary number of optical devices processing incoming packets
important and is not necessarily enough.
in a core node proportional to P as compared to ordinary
In this paper, we experimentally evaluate dynamic range
OPS architecture. In addition to that, the control complex-
into SOA switches for MW-OPS in order to investigate the
ity in MW-OPS nodes could be lower than that in traditional
relationship between OSNR and parameters such as the num-
OPS nodes (Fig. 2). Due to these characteristics, MW-OPS
ber of wavelengths utilized, wavelengths intervals, and the
is considered to be a more scalable architecture.
optical gain provided by the SOA switches.
The remainder of this paper is organized as follows: Sec-
2. 2 SOA switch for MW-OPS
tion 2 introduces MW-OPS and SOA switches. Section 3
The optical switches applied to MW-OPS are required with
explains the details of our experiment. Section 4 shows ex-
higher performance than those for the ordinary OPS, due to
perimental results and provides some discussions. Section 5
the fact that the lengths of Multi-Wavelength optical packets
(MW-packets) in time domain get multiple times shorter than
—2—
λ
H1
λ
P1
H2
H : Header
P : Payload
P2
H3
In fact, the performance of SOA switch has been studied
H1 H2 H3
P1
P2
λ
H Payload
H
P
P3
P
P
P3
(a) Ordinary 2
×
t
2 OPS Node
(b) 2
×
for single-wavelength optical packet switching by measuring
the dynamic range of input power into SOA switches [14].
t
t
2 MW-OPS Node
Fig. 1 The ordinary and Multi-Wavelength optical packet
According to the paper, low input power causes an increase
in power penalty because of the SOA patten effect which degrades the optical signal to noise ratio (OSNR). As the input
power is increased, the power penalty reaches a minimum.
For high input power, the saturation distorts the signal and results in an increased power penalty. However, for MW-OPS,
FWM could be a more significant factor than saturation while
the input power increases. Herein, we particularly focus on
these nonlinear effects of SOA switches in MW-OPS appli-
Fig. 2 The ordinary OPS node and MW-OPS node
cations and investigate the dynamic range of input signal for
various cases.
those of single-wavelength optical packets, depending on
the number of available wavelengths for MW-packets. The
switching speed of optical switches for MW-OPS should be
3. Experiment and Performance Measurement
faster (e.g., less than 10 ns). In addition, as the MW-packets
The experimental setup designed to measure the dynamic
are switched in a wide-band, the optical switch for MW-OPS
range is depicted in Fig. 3. In this experiment, at most 15
should be non-wavelength dependency and low polarization
wavelengths (1550.92 nm∼1562.23 nm) with 100GHz spac-
dependency.
ing which are modulated to 10 Gb/s PRBS 231 − 1 signal by
The PLZT ((Pb,La)(Zr,Ti)O3) switch and SOA switch,
the LiN bO3 (LN) modulator with PPG (Pulse Pattern Gen-
both of which have high enough switching speed, have al-
erator) are used as the data signal. 2 AWGs and 15 FDLs of
ready been showed to be appropriate for MW-OPS [7], [10].
different lengths remove bit-level correlations among the 15
However as a conventional electro-optic switches on pla-
wavelengths. All channels are aligned to a same polarization
nar light wave circuit based on ferroelectric material, PLZT
state by using a polarizer. With an EDFA and a Variable Op-
switch suffers from increasing optical insertion loss while
tical Attenuator (VOA), the input power of the SOA switch
scaling to large switching ports. While with the SOA gate
becomes adjustable. We evaluate the performance of single-
switch and couplers, broadcast and selection switching can
stage SOA switch this time.
be easily achieved and can easily become large scale. As a
result, SOA switch is considered to be more appropriate to
MW-OPS.
To the best of authors’ knowledge, several demonstration
of applying SOA switches to MW-OPS have already been
reported [10]∼[13]. However, experimental parameters in
these demonstrations are all different to each other and the
maximum number of wavelengths used for MW-OPS among
Fig. 3 Experimental Setup of Input Power Dynamic Range Evaluation of
SOA Switch
these reports is 10, which is far from enough compared to the
number of available wavelengths in a single fiber provided by
nowadays WDM technology. Also, it is important to evaluate
Num. of wavelengths
1,2,4,8,15(Maximum)
the performance of SOA switch for MW-OPS, due to the rea-
Interval of wavelengths (GHz)
100, 200, 300, 400
son that non-linear optical phenomenon like four-wave mixing will show up easily depending on the parameters such
SOA gain (dB)
0, 6, 12
Table. 1 Experimental Parameters of Input Power Dynamic Range Evaluation of SOA Switch
as the input signal power, wavelength intervals, the optical
gain provided by SOA switch and modulation rate while using SOA switch.
The focus of the experimental work is on clarifying the way
for wavelength scalability guarantee when applying SOA
—3—
switch to MW-OPS. We compare the power penalties by
7
changing the total number of wavelengths, the wavelength
interval and the SOA optical gain, to find the relationship be-
)
B
d(
yt
la
ne
Pr
e
w
oP
tween SOA performance and the parameters.
The experimental parameters we considered are referred to
the tabel 1. Bit error rates (BERs) are measured for a range
of input powers into the SOA switch. Power penalty is mea−9
sured against BER at 10
through the SOA switch. The
dynamic range achieving power penalty under 2 dB are mea-
SOA gain= 0dB, 1 wavelength
6
SOA gain= 6dB, 1 wavelength
SOA gain= 12dB, 1 wavelength
5
4
3
2
1
sured by varying the input power into the SOA switch.
0
-35 -30 -25 -20 -15 -10
-1
4. Experimental Results and discussion
-5
0
5
10
Input Power of SOA@Ch07: 1555.75nm(dBm)
The BER is assessed for a range of input conditions, and
the value of input power is recorded which will give an error
Fig. 4 Dynamic Range of 1 Wavelength with Different SOA Gain
rate of 10−9 . We compared the input power dynamic range
7
with different parameters. The results are described as fol-
6
lows. In all cases, the power penalty is plotted against the
input power of the single channel (CH07:1555.75 nm) for
fair comparison.
4. 1 Dynamic Range with Different SOA Gain
In this subsection, we focus on the relationship between
OSNR and the optical gain provided by SOA switch. Fig. 4
5
)
B
d(
yt
la
ne
Pr
e
w
oP
4
3
2
1
shows the dynamic range results compared with different optical gain provided by SOA switch when only one wavelength
is used as data signal. Focusing on each single curve, we
can see that the power penalty is limited by the pattern effect at low input powers. At higher powers, the penalty be-
SOA gain = 0dB, 15 wavelengths
SOA gain = 6dB, 15 wavelengths
SOA gain = 12dB, 15 wavelengths
-35
0
-1
-30
-25
-20
-15
-10
-5
0
Input Power of SOA@Ch07: 1555.75nm(dBm)
Fig. 5 Dynamic Range of 15 Wavelength with Different SOA Gain
comes dominated by distortion in the amplifiers. The results
is conformity to the results showed in [14]. While the op-
the role of distortion becomes more dominant as an order of
tical gain of the SOA amplifier increasing, we can see the
magnitude increase in aggregate input power is required to
dynamic range becomes narrower because the distortion be-
achieve the same power levels at the receiver. The results
comes more Non-negligible. We can get a similar trend while
with 200GHz interval shown in Fig. 7 have the same trend.
increasing the wavelength channel number to 15 (Fig. 5), al-
We can achieve a dynamic range of approximately 20dB
though here, the role of distortion becomes more dominant
while the number of wavelength channels is 8, with the wave-
as an order of magnitude increase in aggregate input power
length interval of 200GHz and approximately 15dB dynamic
is required to achieve the same power levels at the receiver.
range with 15 wavelength channels of 100GHz spacing. The
optical gain provided by the SOA is fixed to 6dB. During
4. 2 Dynamic Range with Different Number of Wavelengths
the comparison results in 4. 1, we believe that wider dynamic
We compare the input power dynamic range results with
range could be achieved if we decrease the SOA optical gain.
different number of wavelength channels with the same
wavelength interval and the optical gain provided by SOA
4. 3 Dynamic Range with Different Wavelength Intervals
switch is fixed to 6dB. Fig. 6 shows the results when wave-
In this subsection, we compare the dynamic range results
length interval is 100GHz. We can see that as the wavelength
in order to clarify the efficient of the wavelength intervals.
channels increased, the dynamic range becomes narrower.
We compared BER curves with different wavelength inter-
This trend could be explained by the analysis in 4. 1 that the
—4—
7
6
)
B
d(
tyl
an
eP
re
w
oP
5
6
6dB-1wave
6dB-2wave-100G
6dB-4wave-100G
6dB-8wave-100G
6dB-15wave-100G
4wave-100G-6dB
4wave-200G-6dB
4wave-300G-6dB
4wave-400G-6dB
5
) 4
B
d(
yt
la 3
ne
Pr
e 2
w
oP
4
3
2
1
1
0
-35 -30 -25 -20 -15 -10 -5
-1
0
5
0
-35 -30 -25 -20 -15 -10 -5
0
5
Input Power of SOA@Ch07: 1555.75nm(dBm)
Input Power of SOA@Ch07: 1555.75nm(dBm)
Fig. 6 Dynamic Range of Different Num. of Wavelengths (100GHz Spac-
Fig. 8 Dynamic Range of Different Wavelengths Interval (4 Wavelengths,
ing, Gain = 6dB)
7
8
6dB-1wave
6dB-2wave-200G
6dB-4wave-200G
6dB-8wave-200G
6
5
)
B
d(
tyl
an
eP
re
w
oP
Gain = 6dB)
7
)
B
d(
yt
la
ne
Pr
e
w
oP
4
3
2
1
5
4
3
2
1
0
-35
6
4wave-100G-12dB
4wave-200G-12dB
4wave-300G-12dB
4wave-400G-12dB
-30
-25
-20
-15
-10
-5
0
5
-1
Fig. 7 Dynamic Range of Different Num. of Wavelengths (200GHz Spacing, Gain = 6dB)
vals such as 100GHz, 200GHz, 300GHz, and 400GHz. Due
to the limitation of the experimental devices, with the interval of 400GHz, we can only use as much as 4 wavelength
channels. Fig. 8 shows the comparison results with 4 wavelengths under the optical gain of 6dB and Fig. 9 shows the results with 4 wavelengths under the optical gain of 12dB. We
can figure out that when the wavelength interval is increased
from 100GHz to 200GHz, dynamic range becomes wider obviously. This is because the FWM becomes smaller while the
wavelength interval increasing. While the interval increased
continuously, e.g., from 200GHz to 300GHz or 400GHz, the
increasing of dynamic range decelerated sharply. This can be
explained as that the effect of FWM phenomenon becomes
small enough when the wavelength interval is 200GHz.
0
-35 -30 -25 -20 -15 -10
-5
0
Input Power of SOA@Ch07: 1555.75nm(dBm)
Fig. 9 Dynamic Range of Different Wavelengths Interval (4 Wavelengths,
Gain = 12dB)
4. 4 Discussion
By comparing the BER curves for the dynamic range
measurement, we find the trend that as the number of utilized wavelengths increasing, the input power range becomes
worse. And the dynamic range becomes wider if we decrease
the SOA optical gain or increase the interval between wavelength channels.
We achieve approximately 15dB dynamic range for 15
wavelengths with the interval of 100GHz at an SOA optical gain of 6dB and approximately 20dB for 8 wavelength
channels with 200GHz spacing under the SOA optical gain
of 6dB. With this result we confirm that at least 15 wavelengths is usable for MW-OPS. Moreover, during the comparison results, reducing optical gain or increasing wavelength interval can achieve wider dynamic range. Especially,
input power dynamic range obviously becomes wider when
—5—
optical gain is decreased. Obviously wider dynamic range
is also achieved when adjusting wavelength interval from
100GHz to 200GHz, while little effect on dynamic range
[7]
is observed when continuously increasing wavelength interval from 200GHz to 300GHz or more. We believe that
[8]
wide enough dynamic range could be achieved by much
more number of wavelength channels with 200GHz spacing
[9]
and the optical gain lower than 6dB, while the loss through
the transmission could be compensated by applying Erbium
[10]
doped fiber amplifier (EDFA).
[11]
5. Conclusion
[12]
In this paper we investigated the relationship between
OSNR and several SOA-specific parameters such as the number of wavelength utilized, wavelength interval, the opti-
[13]
cal gain, trying to clarify the way for wavelength scalability guarantee when applying SOA switch to MW-OPS.
[14]
Bandwidth Colored Optical Packet Switching with Polarization Independent High-speed Switch and All-optical Hierarchical Label Processing ”, ECOC2007, PDS3.1, (2007).
K. Watabe, et al.“ 320Gb/s Multi-wavelength Optical Packet Switching with Contention Resolution Mechanism using PLZT Switches ”,
OFC/NFOEC2008, OThA5, (2008).
E.F.Burmeister, et al., “ Integrated gate matrix switch for optical
packet buffering ”, IEEE Photon. Technol. Lett., vol.18, no.1, pp.
103-105, (2006).
H. Takara, et al,“ Field demonstration of over 1000-channel DWDM
transmission with supercontinuum multi-carrier source, ” Electron.
Lett., vol. 41, pp. 270-271 (2005).
H. Onaka, et al. “ WDM Optical Packet Interconnection using
Multi-Gate SOA Switch Architecture for Peta-Flops Ultra-HighPerformance Computing Systems ”,ECOC2006, Tu4.6.6, (2006).
E.T.Aw, et al.,“ Multi-Stage SOA Switch Fabrics: 4x40Gb/s Packet
Switching and Fault Tolerance”, OFC/NFOEC 2007, OThF2, (2007).
N. Kataoka, et al., “ Experimental Demonstration of Multicastcapable Variable Bandwidth Colored Packet Switching using SOA
Switch and Stacked OC Label Processing ”, OFC/NFOEC 2008,
OThA3, (2008).
C.P.Lai, et al.,“ Demonstration of Programmable Broadband Packet
Multicasting in an Optical Switching Fabric Test-bed”, OFC/NFOEC
2009, OTuA5, (2009).
J.P.Mack, et al.,“ 40 Gb/s Autonomous Optical Packet Synchronizer ”, OFC/NFOEC 2008, OTuD3, (2008).
Due to the limitation of our experimental environment, the
maximum number of wavelength was 15 channels with
100GHz spacing in C-band and each channel was modulated
at 10Gbps in this experiment. The results showed that reducing optical gain can most effectively increase the input
power range and the effect of wavelength interval on dynamic
range was obvious only when it is adjusted from 100GHz to
200GHz. The results suggests large wavelength scalability
could be achieved by carefully adjusting the two parameters
and introducing EDFAs for optical amplification instead of
SOA.
Acknowledgements
This paper is supported by National Institute of Information and Communication Technology(NICT).
References
[1] Cisco Visual Networking Index Forecast, “ Hyperconnectivity and the Approaching Zettabyte Era ”, [Online], Available:
http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns53
7/ns705/ns827/VNI Hyperconnectivity WP.html (2009)
[2] Internet World Stats,“INTERNET USAGE STATISTICS”, [Online].
Available: http://www.internetworldstats.com/stats.htm
[3] X.Zhou, et al.“ 32Tb/s (320x114Gb/s) PDMRZ8QAM Transmission over 580km of SMF28 UltraLowLossFiber,” OFC/NFOEC2009,
PDPB4, (2009).
[4] T. Asami, et. al.,“ Energy Consumption Targets for Network Systems, ”ECOC2008, Tu.4.A.3, (2008).
[5] C.Guillemot, et al.,“ Transparent Optical Packet Switching: The European ACTS KEOPS Project Approach”, J.Lightwave Tech., vol.16,
no.12, pp. 2117 - 2134 (1998).
[6] N.Wada, et al.“ Field Demonstration of 1.28T bit/s/port, Ultra-wide
—6—