All-optical signal processing for optical packet switching networks

Transcription

1 Invited Paper All-optical signal processing for optical packet switching networks Y. Liu, M.T. Hill, N. Calabretta, E. Tangdiongga, R. Geldenhuys, S. Zhang, Z. Li, H. De Waardt, G.D. Khoe and H.J.S. Dorren COBRA Research Institute, Eindhoven University of Technology, P. O. Box 513, 5600 MB, Eindhoven, the Netherlands. ABSTRACT We discuss how all-optical signal processing might play a role in future all-optical packet switched networks. We introduce a concept of optical packet switches that employ entirely all-optical signal processing technology. The optical packet switch is made out of three functional blocks: the optical header processing block, the optical memory block and the wavelength conversion block. The operation principle of the optical packet switch is explained. We show that these three functional blocks can be realized by using the nonlinearities of semiconductor optical amplifiers. Some technologies in these three functional blocks are described. The header processor is realized using a Terahertz Optical Asymmetric Demultiplexer. We also describe a header pre-processor to improve the extinction ratio of the header processor output. In the optical memory block, we show that an all-optical memory can be obtained by using twocoupled lasers that form a master-slave configuration. The state of the optical memory is distinguished by the wavelength of the master laser. We extend the concept to an optical memory can have multiple states. In the wavelength conversion block, we demonstrate a 160 Gbit/s wavelength conversion using a single semiconductor optical amplifier in combination with a well-designed optical bandpass filter. The semiconductor optical amplifier has a gain recovery time greater than 90 ps, which corresponds to a less than 20 GHz bandwidth for conventional wavelength conversion. We show that by properly using the optical bandpass filter, ultrafast dynamics in the semiconductor optical amplifier can be employed for wavelength conversion at ultrahigh bit-rates. Keywords: Optical fiber communication, optical cross-connect, optical packet switching, optical signal processing, optical header processing, optical memory, optical wavelength conversion 1. INTRODUCTION In optical networks, the bandwidth mismatch between optical transmission and electronic routers has led to the development of various optical signal processing techniques and the investigation of optical packet switching [1-3]. Figure 1 shows a schematic diagram of a generic optical packet switched cross-connect. The main functions are synchronization, switching and buffering, regeneration [3]. The synchronization functional block is to synchronize the incoming packets so that the packets are in a specific time slot when the packets are being processed. The second functional block is the switching and buffering, which is used to route the packets and to solve packet contention. The regeneration block is utilized to regenerate the signal for consistent quality as the packet passes through multiple nodes. Figure 2 presents a concept of our all-optical packet switch, which can be realized by using entirely all-optical signal processing. The all-optical packet switch is made out of three functional blocks: the all-optical header processing block, the all-optical flip-flop memory block, and the wavelength conversion block. The packets that we used have a fixed duration and consist of an optical header and optical payload. The header contains the routing information of the packet while the payload contains the information content. Both the header and the payload consist of amplitude modulated data bits. When an optical packet arrives at the optical packet switch, the optical power of the packet is split into two parts. Half of the optical power of the packet is delayed and injected into a wavelength converter. Some delay is required to compensate for the time taken to carry out the header processing function. The header processing function block translates the optical header pattern into an optical pulse. This pulse triggers the flip-flop memory to generate Photonic Devices and Algorithms for Computing VII, edited by Khan M. Iftekharuddin, Abdul A. Awwal, Proc. of SPIE Vol. 5907, 59070J, (2005) X/05/$15 doi: / Proc. of SPIE Vol J-1

2 continuous wave (CW) light with a specific wavelength. Hence, different header information forces the optical memory to output CW light with a different wavelength. The output light from the optical memory is fed into the wavelength converter to convert the packet into the desired wavelength. Afterwards, a demultiplexer is used to route the packet into a specific port, depending on the wavelength of the packet. Thus the routing of the packet is determined by the wavelength of the optical memory output light, in turn, determined by the header information. Thus an optical packet switch is realized. 1 N Synchronisation Switch fabric Header recognition Wavelength conversion and signal regeneration Contention resolution Wavelength conversion and signal regeneration 1 N All-optical switch and buffer control Figure 1: Generic optical packet switched cross-connect structure. The three most important sections are synchronization, switching and buffering, regeneration. Wavelength conversion is used to facilitate some of these functions. Packet Input Header Processing Flip-Flop Memory Continuous Wave λ 1 or λ 2 Wavelength Conversion λ 1 λ 2 Payload Header Packet Format Fiber Delay Line Demultiplexer Figure 2: System concept for a 1 2 all-optical packet switch. ThreshQldfuflCtlQn nm SOA liii nm Header Processing nm EDFA (\\ T5ts5ts PacketFormat 35ts 175 s a a) header 0 payload between[ Figure 3: Experimental set-up to demonstrate the 1 2 all-optical packet switching. Traffic from the network is coupled in the packet switch at the input. The packet format is given. SOA: Semiconductor Optical Amplifier, FBG: Fiber Bragg Grating, EDFA: Erbium Doped Fiber Amplifier, ISO: Isolator and PHASAR: Phased array demultiplexer. Proc. of SPIE Vol J-2

3 Figure 3 presents an experimental demonstration of all-optical packet switch that allows routing of data-packets without electronic control [4]. This packet switch utilizes several optical functionalities such as an optical header processor, an optical flip-flop memory, and a wavelength converter. In this paper we show that these three optical functionalities can be realized by using the nonlinearities of semiconductor optical amplifiers. The optical header processor utilizes a Terahertz Optical Asymmetric Demultiplexer. We also describe a header pre-processor to improve the extinction ratio of the header processor output. The optical memory is based on two-coupled lasers that form a master-slave configuration. The state of the optical memory is determined by the wavelength of the master laser. We extend this concept to a three-state optical memory. For wavelength conversion, we propose a method to increase operating speed of wavelength conversion in a semiconductor optical amplifier. 160 Gbit/s wavelength conversion is demonstrated in a semiconductor optical amplifier that has a gain recovery time of more than 90 ps, which corresponds to a less than 20 GHz bandwidth for conventional wavelength conversion. In advanced lightwave systems, several optical elements are required to function together. Monolithic photonic integrated circuits provide the photonic functionality as well as the inexpensive, robust on-chip interconnection of devices necessary to build integrated subsystems on a chip. The all-optical functions discussed in this paper all share the advantage that they have the potential to be photonically integrated. 2. ALL-OPTICAL HEADER PROCESSING All-optical header processing has been investigated using several different methods. In [5], an all-optical method for processing packet headers is presented that uses tunable fiber Bragg gratings. Ultra fast all-optical header recognition has been reported in [6], and by using four-wave mixing in a Semiconductor Optical Amplifier (SOA) [7], and by using Terahertz Optical Asymmetric Demultiplexers (TOADs) [8]. The TOAD configuration has also been used to demonstrate all-optical ultra fast switching [9]. The ultra fast all-optical header recognition methods reported in [6] and [7] require a form of optical clock recovery that introduces additional complexity in the switching system. In [10] a header recognition method is described, it does not require optical clock recovery, but it needs a Manchester encoded packet payload and also depends on the SOA recovery time. In [11], an asynchronous multioutput all-optical header processing technique based on the two-pulse correlation principle in a semiconductor laser amplifier in a loop optical mirror (SLALOM) configuration is presented. This concept was employed in an all-optical packet switch [4]. This header processing technique does not require a synchronous control pulse, but the processing speed of the SLALOMbased header processor is determined by the semiconductor optical amplifier (SOA) recovery time (typically 1 ns). Moreover, the SOA has to be placed offset to the center of the loop with a distance that is larger than (10 cm), where is the group velocity of light (100 m/ps). Thus, a disadvantage of the header processing method presented in [11] is that the SLALOM configuration is too large to allow photonic integration. In this paper we describe a TOAD-based header recognition [12], which can function asynchronously and operate at low power. Using a TOAD-based header processing technique, together with a header preprocessor, results in a system that can distinguish a large number of header patterns and allows asynchronous operation and photonic integration. Another advantage is that TOAD operation guarantees ultra fast header processing at low power. Although the system described here was demonstrated at 10 Gbit/s [12], optical switching has been demonstrated using a TOAD at 250 Gbit/s [8, 9]. Furthermore, the header processing system as a whole operates asynchronously and the system can be extended to have multiple output ports. The optical header processing system is schematically presented in Figure 4. It consists of a header pre-processor (HPP) based on the principle of self-induced polarization rotation in an SOA [13], and two TOADs that are placed in parallel. The function of the HPP is twofold: it separates the packet header from the packet payload, and it also creates the control signal that is required for TOAD operation. Each of the TOADs is designed to recognize a specific header pattern. An essential part of the header processing concept is that the address information is encoded using the difference in time between the leading edges of two header pulses. The space between the header pulses is filled with a sequence of alternating NRZ 0 and 1 bits at the same bit rate as the data payload, which ensures that the SOA remains saturated while the packet header passes through [12, 13]. A series of 0 s with a duration that is longer than the SOA recovery time τ is placed before the second header bit, to allow the amplifier to recover before the second header Proc. of SPIE Vol J-3

4 pulse arrives at the SOA. Similarly, the guard time in between the header section and the payload section is filled with a sequence of alternating 0 and 1 bits to keep the amplifier saturated when the packet passes by. Finally, the packet payload (at 10 Gbit/s) is Manchester encoded to avoid repetition of the header pattern (at 2.5 Gbit/s) in the packet payload. Laser Input Signal Tp A Output Control Pattern generator 10 Gbit/s Input Port 50:50 Header 2 processor 50:50 x SOA PBS 90:10 Header section t h_1 t h_2 B C A B Header Pre-Processor SOA PBS Modulator :20 Header 1 processor x EDFA SOA t c_1 PBS 90:10 50:50 Input Port 50:50 t c_2 Payload Output port 1 Output port 2 Manchester encoded (10 Gb/s) Manchester encoded (10 Gb/s) Control Signal Figure 4: Experimental setup to demonstrate the header processing system and packet structure. : Polarization Controller, EDFA: Erbium Doped Fiber Amplifier, : Band Pass Filter, PBS: Polarizing Beam Splitter. Input packets [V] Output port 2 [mv] Output port 1 [mv] Output HPP [V] a) " b) " c) Header 1 Payload Header as d) 21 I a 7 " Time [ns] Figure 5: Measured oscilloscope traces. (a) The optical data packets that input the header processor. Packets with two different headers are shown. The inset shows details of the packet header. (b) The output of the HPP. It is clearly visible that only pulses synchronized with the leading edge of the header bits output the HPP. (c) Output at Port 1. It is visible that only Packet 1 is recognized. (d) Similar to (c), but now for Packet 2. The operation of the TOAD has been described in [14]. The TOAD can function as a header processor as follows. The HPP output pulses form the TOAD input signal and are split by an 80:20 coupler into a low power clockwise (CW) and counter clockwise (CCW) propagating data signal, and a high power control signal. The CCW propagating pulses arrive at the SOA after the CW propagating pulses. If no control is present the data signal is reflected back to the TOAD input. The first pulse of the high power control signal is used to saturate the SOA. Header recognition can be implemented by accurate timing of the control pulses. A pulse is only output from the TOAD if the time between the two header pulses matches the timing of the control pulse. Figure 5 presents the experiment results. The packets with two different headers are shown in Figure 5a. The output of HPP is presented in Figure 5b, which clearly illustrates that the HPP generates a pulse at the leading edges of the header bits. Figure 5c and 5d shows the outputs of the header processor. It is clearly visible that indeed a pulse is formed at output Port 1 only for packets with Header 1 while no pulse is formed for packets with Header 2. We can observe that a pulse is formed at output Port 2 only for packets with Header 2 while no pulse is formed for packet with Header 1. In contrast to header recognition based on two-pulse correlation in a SLALOM configuration, this header recognition concept does not critically depend on the SOA recovery time. Hence, the offset of the SOA with respect to the center of the loop and the delays for the control pulses can be made sufficiently small so that this header processing concept allows photonic integration. Proc. of SPIE Vol J-4

5 3. OPTICAL MEMORY All-optical memories have many potential applications in optical communication systems and optical computing [15]. Optical memories based on bistable laser diodes (BLD) have been extensively studied, and has been reported in [16] as a review paper. In the normal BLD, the bistabilty operation is performed in the S-shaped hysteretic region and needs trigger pulses to change the memory state. In all-optical set and reset operation, one difficulty is that the optical reset operation needs a negative optical pulse to switch the laser off. To overcome this problem, BLDs with separate optical set and reset inputs are proposed [17, 18]. The output states of BLDs can be distinguished by different output power or different polarization states, but in BLDs it is difficult to have output states that have a large wavelength range, and can be used in WDM system. In [19, 20], a bistable semiconductor fiber ring laser that has optical spectral bistability is proposed. In one of the bistable states, one wavelength is dominant. However, in another state, two wavelengths are dominant, including the wavelength that was dominant in the previous state. This makes the optical memory in [19, 20] difficult to be implemented in all-optical packet switches. In this paper, we described an optical flip-flop that is based on two-coupled lasers [21]. Each laser outputs at a specific wavelength. The output power of one laser is partly coupled into another laser, thus these two laser couple each other and form a bi-stability system. The memory has two states. In each state, only one laser lasing and another laser is suppressed. Therefore, in each state, the memory is dominated by one laser, and the memory s state is distinguished by a specific wavelength. 3.1 Operation principle of two-state optical memory The configuration is depicted in Figure 6. The optical memory is made out of two coupled lasers with separate laser cavities. Reset ISO 80/20 Set SOA 60/40 80/20 SOA Output ISO FPF λ 2 Figure 6: Laser implementation of the optical flip-flop memory. λ 1 FPF Output power (mw) Power of external injection (mw) Figure 7: the output power of a laser versus the external optical input power. As shown in Figure 6, in each single laser, a Semiconductor Optical Amplifier (SOA) acts as the laser gain medium. The wavelengths are selected by Fabry-Perot filters (FPFs). The operation principle of a single laser is described in [22]. If the gain of the SOA is higher than the threshold of the laser, lasing starts. However, the SOA can be saturated by injection of high-intensity external light, which causes the gain at the lasing wavelength to be reduced. If the reduced gain at the lasing wavelength is below the threshold value, lasing stops. Figure 7 shows the typical (experimental) laser output as a function of the intensity of the saturating external light. It is clearly visible that external light with sufficient intensity can suppress lasing. Two identical lasers can be coupled to make an all-optical flip-flop memory [21]. The first laser, Laser 1, outputs continuous wave (CW) light at wavelength λ 1. The second laser, Laser 2, outputs CW light at wavelength λ 2. The output of Laser 1 is coupled into Laser 2, and the output of Laser 2 is coupled into Laser 1. In such a configuration one laser can act as the master laser, which suppresses lasing in the other laser (the slave). Due to the symmetry of this masterslave configuration, the role of master and slave can be interchanged. Thus two states are possible. In State 1, Laser 1 is Proc. of SPIE Vol J-5

6 lasing and Laser 2 is suppressed. Therefore, in this state the optical flip-flop memory emits CW light at wavelength λ 1. Conversely, in State 2, Laser 2 is lasing and Laser 1 is suppressed, then the optical memory outputs CW light at wavelength λ 2. To change the states, lasing of the dominant laser can be stopped by injecting external light, which has a different wavelength from the dominant laser, into the dominant laser cavity [21]. As a consequence, the laser that was suppressed become to dominate the memory output even if the external light is removed, the flip-flop remains in the new state. An optical flip-flop based on gain quenching offers a number of advantages: it can provide high contrast ratios between states. There is no difference in the mechanisms that change from State 1 to State 2 and vice-versa, permitting symmetric set reset operation. The wavelength range of the input light, and the choice of output wavelengths, can be quite large. Furthermore, the flip-flop has controllable and predictable switching thresholds. The flip-flop is not tied to a specific structure or technology, and does not rely on second-order laser effects such as refractive index changes or nonlinear gain. Thus, the flip-flop can be implemented with a wide range of laser and interconnection types. Statel -20 State Wavelength (nm) Time (gs) Figure 8: (a) Spectral output of two states of the optical flip-flop memory is presented. The solid curve represents the state in which Laser 1 is lasing, and the dotted curve represents the state in which Laser 2 is lasing. (b) Oscilloscope traces showing power of Laser 1 and Laser 2. The regular toggling between the states can be clearly seen. 3.2 Experiment and results of two-state optical memory The experimental set-up for an experiment to demonstrate the operation of the optical flip-flop memory is presented in Figure 6. The setup is constructed by using commercial available fiber-pigtailed components. In this particular set-up a ring-laser configuration is used. The wavelengths of the lasers are λ 1 = nm and λ 2 = nm respectively. The SOAs were manufactured by JDS Uniphase and employ a strained bulk active region. The SOA injection currents are set in such a way that the system is symmetric. We have used Fabry-Perot filters with a bandwidth of 0.18 nm as wavelength selective elements. The SOA 1 were pumped with 168 ma of current SOA 2 was pumped with and 190 ma of current. The pulses that were used to set and reset the flip-flop had a power of 2 mw. The optical spectrum of the flip-flops output states is presented in Figure 8a. It is clearly visible that the difference in output power between the two states is over 45 db. The switching characteristics of the optical flip-flop are presented in Figure 8b. It can be observed from Figure 8b that the system changes states if sufficient external pulse is coupled in the flip-flop. Furthermore, the system remains at the new state even if the injection of external pulse is stopped. 3.3 Three-state optical memory The concept in [21] can be extended to a three-state optical memory [23]. In Figure 9, three identical ring lasers are coupled to each other to construct a three-state optical memory. The output of each ring laser has to be coupled into the other two lasers, but not into its own cavity. To realize this, the outputs of each ring laser are firstly combined by using a multiplexer. 10% of the combined light is coupled out of the system by using a 90/10 coupler. This is the memory output. The other 90% of the combined light is firstly fed into SOA 4 to be amplified and then fed back into each ring laser through a system of Fiber Bragg Gratings (FBGs). FBGs are used to prevent light at the lasing wavelength reentering the ring laser cavity. Thus, the output of each laser is coupled into the other lasers but not into its own cavity due to the FBGs. The amplification by SOA 4 ensures that the light injected into each laser is sufficient to suppress the lasing mode. Since the system is symmetric, all the lasers can suppress lasing in the other lasers, and thus each laser can become dominant. Therefore, the memory has three possible states, depending on which laser is lasing. The state of the Proc. of SPIE Vol J-6

7 memory is determined by the wavelength of the memory s output light. In each state, only one laser lases and the other lasers are suppressed, thus only one wavelength is dominant. In State 1, Laser 1 dominates and suppresses the lasing in the other lasers, thus λ 1 is dominant. In State 2, Laser 2 is dominant and suppresses lasing in the other two lasers, therefore, λ 2 is dominant. In State 3, only laser 3 is lasing and suppresses Laser 1 and Laser 2, thus λ 3 is dominant. The experimental results are presented in Figure 10, where each state of the memory is clearly shown [23]. Set1 Set2 Set3 MUX λ 1 A B λ 2 λ 3 90/10 FPF ISO Laser1: λ 1 SOA1 Laser2: λ 2 Memory output λ 1 FBG SOA2 λ C 2 Laser3: λ 3 SOA3 λ 3 SOA4 Splitter Figure 9: The configuration of the three-state all-optical memory based on coupled ring lasers. ISO: isolator, FPF: Fabry-Perot filter, : polarization controller, MUX: optical multiplexer, FBG: fiber Bragg grating Power(dBm) (a) Wavelength (nm) Figure 10: Spectral output at each state of the memory. (a) represents State 1 (λ 1 dominant), (b) represents State 2 (λ 2 dominant), (c) represents State 3 (λ 3 dominant). (b) (c) 4. WAVELENGTH CONVERTER All-optical wavelength converters based on nonlinearities in SOAs are considered as important building blocks for wavelength-division- multiplexed (WDM) networks [24, 25], because of the optical power efficiency and its advantage for integration. In current SOA-based wavelength converters, the relatively slow recovery time of the carrier density (typically several hundreds picoseconds), limits the maximum operating speed. Several technologies have been used to improve the bandwidth of SOA-based wavelength converters. A Fiber Bragg Grating (FBG) [26] and a waveguide filter [27] have been used to increase the frequency response of an SOA. Wavelength conversion at 100 Gbit/s has been achieved by using a long SOA (2 mm) in combination with a FBG [28]. In [29], a differential Mach-Zehnder Interferometer scheme with SOAs in both arms has been proposed. The latter configuration allows the creation of a short switching window (several picoseconds), although the SOA in each arm has a slow recovery of the carrier density. In [30, 31], a delay-interferometric wavelength converter has been demonstrated in which only one SOA is utilized. Ref. [32, 33] shows that optical filtering chirped component of converted light in the SOA can be utilized for 40 Gbit/s wavelength conversion. In this paper, we present a very simple approach that can dramatically increase the operation speed of wavelength converter [34, 35]. Assisted by a narrow optical bandpass filter, the operating speed has been increased from less than 10 Gbit/s to 160 Gbit/s. 4.1 Wavelength conversion results at 40Gbit/s The experimental set-up for testing recovery time of an SOA-based system is depicted in Figure 11, which is made out of one SOA and a tunable optical bandpass filter. The SOA employed is a commercially available strained bulk device, which can provide 20 db fiber-to-fiber gain. The measured result shows that the SOA has recovery time of more than 90 picoseconds, corresponding to a less than 10 Gbit/s operating speed for conventional wavelength conversion. The eye-diagram of output converted 40 Gbit/s signal is shown in Figure 12. If the center wavelength of the filter is coincide with the peak of the spectrum of the converted probe light, we observed a closed eye-diagram. A strong pattern-dependent effect is observed in the oscilloscope traces of the converted signal, as shown in Figure 12a. The reason is that the recovery time of the SOA is Proc. of SPIE Vol J-7

8 over 90 ps, which is much larger than 25 ps (40 G bit-rate). We have tuned the filter to select the blue-sideband of the probe light, a clear 40 Gbit/s eye-pattern is shown, indicating that error-free conversion is possible. The oscilloscope traces of the 40 Gbit/s converted signal is shown in the right column of Figure 12b. No pattern-dependent effect is observed. This clearly verifies that the optical bandpass filter can speed up the operating speed of the SOA-based system. 10 GHz Mode-locked fiber laser nm MOD 10 to 40 Gbit/s Multiplexer 3dB SOA Oscilloscope CW Laser nm Figure 11: Experimental setup for testing recovery time of an SOA-based system at 40 Gbit/s. MOD: extern modulator, : optical bandpass filter. (a) (b) Time base: 20 ps/div Time base: 50 ps/div Figure 12: Experimental results of the converted 40 Gbit/s signal. Left Column: eye-diagram, Right column: oscilloscope traces of the converted signal. (a) The optical bandpass filter is located the peak of converted signal spectrum, (b) the optical bandpass filter selects the blue sideband of the converted signal. 4.2 Wavelength conversion results at 160 Gbit/s This concept is applied for a 160 Gbit/s all-optical wavelength converter (AOWC). The experimental setup is depicted in Figure 13a. The 160 Gbit/s wavelength conversion setup was constructed by using commercial available fiberpigtailed components. A 10 Gbit/s data stream with 1.9 ps-wide optical pulses, is modulated by an external modulator (MOD) at 10 Gbit/s to form a RZ-PRBS signal. This data stream is multiplexed to 160 Gbit/s. The 160 Gbit/s RZ- PRBS data signal is combined with a CW probe light and fed into an AOWC via a 3 db coupler. The AOWC is made out of an SOA, a 1.4 nm optical bandpass filter () and a delayed-interferometer (DI). The 1.4 nm is detuned 1.23 nm to the blue side of the probe carrier wavelength. The DI consists of two polarization controllers (s), a polarization maintaining fiber (PMF) with 2 ps differential delay, and a polarization beam splitter (PBS). The SOA (manufactured by Kamelian) is pumped with 250 ma. The average optical power of the 160 Gbit/s data stream is 4.8 mw and 2.6 mw for the CW probe light. At the output of the 1.4 nm, the converted probe light is monitored by using an optical sampling scope, the result is shown in Figure 13b. An inverted 160 Gbit/s signal with a clear open eye-pattern is obtained, which clearly shows that a 3 ps recovery is achieved. The inverted 160 Gbit/s converted signal is subsequently injected into the DI, where the inverted signal is converted into a non-inverted signal. The result is presented in Figure 13c. It is noted that the differential operation in the DI is not essential for realizing 160 Proc. of SPIE Vol J-8

9 Gbit/s operation because the input (inverted) pulses have already been fully recovered within 3 ps, as shown in Figure 13b. 1.9 ps 10 Gbit/s MLFRL nm Tunable Laser nm Power (A.U.) CW 160 Gbit/s transmitter MOD 10 Gbit/s PRBS ps MuX 1 EDFA (b) Output of 1.4 nm 5nm 3dB Zero All-optical wavelength converter (AOWC) SOA 1.4nm 2 Figure 13: (a) 160 Gbit/s all-optical wavelength conversion setup. (b) and (c) are 160 Gbit/s eye-diagrams at various positions in the all-optical wavelength converter. The eye diagrams are measured by an optical sampling scope. (d) BER performance of 160 Gbit/s wavelength conversion. After wavelength conversion, the converted signal is demultiplexed from 160 Gbit/s to 10 Gbit/s and then analysed. Figure 13d shows BER measurement. All the sixteen 10 Gbit/s tributaries are presented. In addition, the 10 Gbit/s basic channel that is multiplexed to 160 Gbit/s is also presented. It can be observed that the average sensitivity penalty of wavelength conversion at a BER=10 9 is about 2.5 db. Moreover, it is visible that no error-floor is observed. The more detailed experimental results can be found in [35]. We have demonstrated pattern-independent wavelength conversion at 160 Gbit/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 optical bandpass filter to select the blue-shifted sideband of the probe light. The detuning optical bandpass filter selects an ultra-fast dynamics of chirped component in the converted probe light, and leads to 3 ps recovery in this SOA-based wavelength converter. Our wavelength converter has been demonstrated by using commercially available fiber pigtailed components. This wavelength converter has a simple configuration and allows photonic integration 2 ps PMF 3 PBS (a) Experimental setup 6.25 ps (c) Output of AOWC EDFA Zero 5nm Log (Bit Error Rate) Gbit/s receiver DeMuX P rec 10 Gbit/s Receiver Optical sampling scope 16x10Gbit/s λ-converted 16x10Gbit/s Original 10Gbit/s (d) Received power [dbm] 5. CONCLUSION This paper has presented some functionalities, such as an all-optical header recognizer, an all-optical memory and optical wavelength conversion, which form essential building blocks for all-optical packet switches. These functionalities enable all-optical switching of data-packets. It should be noted however, that in fully optically controlled packet switched cross-connects, buffering and packet-synchronization functionalities are also required. Moreover, it is desired that all these signal processing functions allow photonic integration, so that eventually monolithically integrated optical packet switched cross-connects emerge. The TOAD-based header recognizer that was discussed in section 2 allows photonic integration. To avoid the use of polarization beam splitters and polarization controllers (which introduce an additional complication for photonic Proc. of SPIE Vol J-9

10 integration), it should be remarked that it is desirable to operate the TOAD with a control signal at a different wavelength instead of a different polarization. A control signal at a different wavelength requires an additional wavelength conversion stage in the header recognizer. Alternatively, one could use an optical correlator based on nonlinear polarization rotation in an SOA for header recognition. From the point of view of photonic integration, such technology is challenging since it requires functionalities such as integrated polarization controllers and an integrated polarization beam splitter. One should realize, however, that the concept of nonlinear cross-polarization rotation shows many similarities with the concept of cross-phase modulation [36], so that the optical correlation can be implemented by using a Mach-Zehnder interferometer. One of the largest challenges on the road towards the application of optical packet switch technology in optical networks is undoubtedly related to the realization of optical packet buffers and packet synchronizers, due to the lack of the optical random memory. All-optical synchronization and buffering would be possible if integrated optical shiftregisters would be available. All-optical flip-flops could act as a fundamental building block for an optical shift-register. Such optical flip-flop memories should have fast (optical) set and reset times, operate at low power, have a high contrast ratio and should have sufficiently small dimensions. An integrated optical flip-flop memory based on laser operation is presented in [17, 18], but the power consumption, the size, and the switching speed of these devices remain an issue, which makes it difficult to couple them in large quantities as required in optical shift registers. We are investigating flip-flop concepts which address the issues of power consumption, size and switching speed [37]. Ideally, a flip-flop concept should have the potential to achieve dimensions in the order of the wavelength of light, a switching speed of a picosecond and a switching energy below a femtojoule. If one succeeds to interconnect these flip-flops, densely integrated digital optical logic operating at high speed and low power can be realized. ACKNOWLEDGEMENTS This work was supported by the Netherlands Organization for Scientific Research (NWO), the Technology Foundation STW and the Ministry of Economic Affairs through respectively the NRC Photonics grant and the Innovational Research Incentives Scheme program. This work was partly supported by the European project IST-LASAGNE. REFERENCES 1. D.J. Blumenthal, J.E. Bowers, L.Rau, et al., Optical Signal Processing for Optical Packet Switching Networks, IEEE Opt. Comm., S23-S29, (2003). 2. Carena, M.D. Vaughn, R. Gaudino, M. Shell and D.J. Blumenthal, OPERA: An optical packet experimental routing architecture with label swapping capability, J. Lightwave Technology 16, , (1998). 3. C. Guillemot et al, Transparent Optical Packet Switching: The European ACTS KEOPS Project Approach, J. Lightwave Technol. 16, (1998). 4. M.T. Hill, A. Srivatsa, N. Calabretta, Y. Liu, H. de Waardt, G.D. Khoe, H.J.S. Dorren, 1 2 optical packet switch using all-optical header processing, Electron. Lett. 37, (2001) 5. M.C. Cardakli, S. Lee, A.E. Willner, V. Grubsky, D. Starodubov, and J. Feinberg, Reconfigurable optical packet header recognition and routing using time-to-wavelength mapping and tunable fiber Bragg gratings for correlation decoding, IEEE Phot. Technol. Lett. 12, (2000). 6. D. Cotter, J.K. Lucek, M. Shabeer, K. Smith, D.C. Rogers, D. Nesset and P. Gunning, Self-routing of 100 Gbit/s packets using 6-bit keyword address recognition, Electron. Lett. 31, (1995). 7. D. Nesset, M.C. Tatham, L.D. Westbrook, and D. Cotter, Degenerate wavelength operation of an ultrafast alloptical AND gate using four-wave mixing in a semi-conductor laser amplifier, Electron. Lett. 30, (1994). 8. I. Glesk, J.P. Solokoff and P.R. Prucnal, All-optical address recognition and self-routing in a 250-Gbit/s packet switched network, Electron. Lett. 30, (1994). 9. I. Glesk, K.I. Kang, and P.R. Prucnal, Demonstration of Ultrafast all-optical packet routing, Electron. Lett. 33, (1997). Proc. of SPIE Vol J-10

11 10. H.J.S. Dorren, M.T. Hill, Y. Liu, N. Calabretta, A. Srivatsa, F.M. Huijskens, H. de Waardt and G.D. Khoe, Optical Packet Switching and Buffering by using all-optical Signal Processing Methods, J. Lightwave Techn. 21, 2-12 (2003). 11. N. Calabretta, Y. Liu, H. de Waardt, M.T. Hill, G. D. Khoe and H. J. S. Dorren, Multiple-output all-optical header processing technique based on two-pulse correlation principle, Electronics Letters, 37, pp , (2001). 12. N. Calabretta, H. de Waardt, G.D.Khoe, H.J.S. Dorren, Ultrafast Asynchronous Multioutput All-Optical Header Processor, IEEE Photon. Techn. Lett. 16, , (2004). 13. N. Calabretta, Y. Liu, F. M. Huijskens, M. T. Hill, H. de Waardt, G. D. Khoe and H. J. S. Dorren, Optical Signal Processing based on Self-induced Polarization Rotation in a Semiconductor Optical Amplifier, J. Lightwave Techn. 22, , (2004). 14. J. P. Sokoloff, P. R. Prucnal, I. Glesk, M. Kane, A terahertz optical asymmetric demultiplexer (TOAD), IEEE Photon. Techn. Lett., 5, , (1993). 15. H. Kawaguchi, Bistability and Nonlinearities in Laser Diodes. London: Artech House, Inc., ISBN X, H. Kawaguchi, Bistale laser diodes and their applications: state of the art, IEEE Journal on Selected Topics in Quantum Electronics, vol. 3, pp , Oct M. Takenaka and Y. Nakana, Realization of all-optical flip-flop using directionally coupled bistable laser diode, IEEE Photonics Technology Letters, vol.16, pp , Jan M. Takenaka and Y. Nakana, Multimode interference bistable laser diode, IEEE Photonics Technology Letters, vol.15, pp , Aug L. Xu, B. C. Wang, V. Baby, I. Glesk and P. R. Prucnal, Optical spectral bistability in a semiconductor fiber ring laser through gain saturation in an SOA, IEEE Photonics Technology Letters, vol. 14, pp , Feb B. C. Wang, L. Xu, V. Baby, I. Glesk and P. R. Prucnal, State selection of a bistability SOA ring laser for bit level optical memory applications, IEEE Photonics Technology Letters, vol. 14, pp , July M. T. Hill, H. de Waardt, G. D. Khoe and H. J. S. Dorren All-optical flip-flop based on coupled laser diodes, IEEE Journal of Quantum Electronics, vol. 37, pp , Mar M. T. Hill, E. E. E. Frietman, H. de Waardt, G.D. Khoe, H. J. S. Dorren, All fiber-optic neural network using coupled SOA based ring lasers, IEEE Transactions on Neural Networks, vol. 13, pp , Nov Y. Liu, M.T. Hill, N. Calabretta, H. de Waardt, G.D. Khoe and H.J.S. Dorren, Three-state all-optical memory based on coupled ring lasers, IEEE Photonics Technology Letters, vol. 15, pp , Oct D. Nesset, T. Kelly, and D. Marcenac, All-optical wavelength conversion using SOA nonlinearities, IEEE Commun. Mag. 36, 56 61, (1998). 25. T. Durhuus, B. Mikkelsen, C. Joergensen, S. L. Danielsen, and K. E. Stubkjaer, All-optical wavelength conversion by semiconductor optical amplifiers, J. Lightwave Technol. 14, , (1996). 26. H.Y. Yu, D. Mahgereftech, P.S. Cho, and J. Goldhar Optimization of the frequency response of a semiconductor optical amplifier wavelength converter using a fiber bragg grating, IEEE/OSA Journal of Lightwave Technology, 17, (2), pp , (1999) 27. Y. Dong, L. Lu, H. Wang, and S. Xie: Improving performance using waveguide filter and optimal probe and signal powers for all-optical wavelength conversion, Proceedings of OFC 2000, 5-10 March, 2000, Baltimore MD,USA,TuF2, pp.69-71, (2000) 28. A. D. Ellis, A. E. Kelly, D. Nesset, D. Pitcher, D. G. Moodie, and R. Kashyap, Error free 100 Gbit/s wavelength conversion using grating assisted cross-gain modulation in 2 mm long semiconductor amplifier, Electron. Lett. 34, , (1998). 29. K. Taijima, "All-optical switch with switch-off time unrestricted by carrier lifetime", Jpn. J. Appl. Phys., Part 1, vol. 32, L1726-L1749, Dec S. Nakamura, Y. Ueno and K. Tajima: 168-Gb/s all-optical wavelength conversion with a symmetric-mach- Zehnder-Type switch, IEEE Photo. Tech. Lett., Vol.13, , Oct., J. Leuthold, C. H. C. H. Joyner, B. Mikkelsen, G. Raybon, J. L. Pleumeekers, B. I. Miller, K. Dreyer, and C. A. Burrus, 100Gbit/s all-optical wavelength conversion with integrated SOA delayed-interference configuration, Electronics Letters, vol. 36, pp , June M.L. Nielsen, B. Lavigne and B. Dagens: Polarity-preserving SOA-based wavelength conversion at 40 Gbit/s using bandpass filtering, Electron. Lett., Vol.39, , Sept., J. Leuthold, D. M. Marom, S. Cabot, J. J. Jaques, R. Ryf and C. R. Giles: All-optical wavelength conversion using a pulse reformatting optical filter, J. of Lightwave Tech., Vol. 22, , Jan., Proc. of SPIE Vol J-11

12 34. Y. Liu, J.P. Turkiewicz, S. Zhang, E. Tangdiongga, E.J.M. Verdurmen, H. de Waardt, D. Lenstra, G.D. Khoe and H.J.S. Dorren, 40 Gbit/s SOA-based wavelength conversion assisted by a narrow optical bandpass filter, in Proceedings Symposium IEEE/LEOS Benelux Chapter, ISBN , pp , Ghent, Belgium, Dec Y.Liu, E. Tangdiongga, Z. Li, S. Zhang, H. de Waardt, G. D. Khoe and H.J.S. Dorren, 160 Gb/s SOA-based Wavelength Converter Assisted by an Optical Bandpass Filter, in Proceedings of 2005 OFC/NFOEC, postdeadline paper, PDP17, Anaheim, California, USA, March H.J.S. Dorren, D. Lenstra, Y.Liu, M.T. Hill, G.D. Khoe, Nonlinear polarization rotation in semiconductor optical amplifiers: theory and application to all-optical flip-flop memory, J. Quantum Electronics, 39, , (2003). 37. M.T. Hill, H.J.S. Dorren, T. de Vries, X.J.M. Leijtens, J.H. den Besten, B. Smallbrugge, S. Oei, J.J.M. Binsma, G.D. Khoe, M.K. Smit, A fast low power optical memory based on coupled micro-ring lasers, Nature 432, nr. 7014, pp , (2004). Proc. of SPIE Vol J-12