Nonlinear Optical Signal Processing in Optical Packet Switching ...
978
IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 18, NO. 2, MARCH/APRIL 2012
Nonlinear Optical Signal Processing in Optical Packet Switching Systems Junya Kurumida, Member, IEEE, and S. J. Ben Yoo, Fellow, IEEE (Invited Paper)
Abstract—This paper discusses nonlinear optical signal processing employed in optical packet switching systems. Nonlinear optical signal processing provides optical label (header) recognition, optical switching, wavelength conversion, and time buffering with typically higher capacity, lower latency, and lower power consumption than electronic counterparts. In order to provide diverse signal processing functions, large-scale integration of nonlinear optical signal processing devices is essential. We discuss possible future directions in optical packet switching involving nonlinear optical signal processing of optical packets with advanced data and label modulation formats. Index Terms—Internet, nonlinear optics, optical burst switching, optical circuit switching, optical flow switching, optical packet switching, optical-label switching, photonic switching.
I. INTRODUCTION HE FUTURE Internet expects to demand protocol-agile and high capacity networking in support of a variety of applications including 3-D multimedia entertainment, telemedicine, and cloud computing. The Internet Protocol (IP) supports all applications on any physical layer platforms, following the hour-glass model [1]. While the generality and heterogeneity of the ‘‘Internet hourglass” are the critical strengths of the Internet that made it so successful, new multimedia, data services, and cloud computing are driving the needs for high performance, high utilization, and secure networks. A unified networking platform in support of voice, data, and multimedia applications are attractive especially on a high-capacity optical layer. Since IP uses packets as the unit of transport and switching, the capability to switch and transport packets with low latency and high throughput can enhance the performance of the IP network. As IP packets are most naturally accommodated in the form of packet switching, optical packet switching may be the technology best suited for the future Internet providing both high-capacity and high-utilization, aggregation [2] of such packets at the edge into bursts, flows, or circuits can further improve energy efficiency and reduce the complexity of the control plane. In particular, optical-label switching (OLS) [3]
T
Manuscript received February 19, 2011; accepted April 4, 2011. Date of publication July 29, 2011; date of current version March 2, 2012. J. Kurumida is with the National Institute of Advanced Industrial Scienece and Technology, Tsukuba, Ibaraki, 305-8568 Japan (e-mail: [email protected]). S. J. B. Yoo is with the Department of Electrical and Computer Engineering, University of California, Davis, CA 95616 USA (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSTQE.2011.2143390
Fig. 1. Optical packet switching architecture with (a) synchronous and fixedlength packet forwarding, and (b) asynchronous and variable-length packet forwarding.
can utilize a unified control plane in support of optical packet switching (OPS), optical burst switching (OBS), optical flow switching (OFS), and optical circuit switching (OCS) [4]–[6]. The unified control plane [7], [8] of OLS also supports applications’ differing degrees [9] of quality of service (QoS), class of service (CoS), and type of service (ToS) requirements by mapping them onto the optical labels [4]–[6]. The minimum requirement of the OLS system is to support OPS with low latency, high throughput, and high scalability. Fig. 1 illustrates a simplified schematic of an OPS system for (a) synchronous and fixed-length packet forwarding, and (b) asynchronous and variable-length packet forwarding. In both cases, the OPS system includes a switch controller, an input controller, and an output controller in the control plane, and an input interface, a switch fabric, and an output interface in the data plane. In the synchronous OPS, the input interface must synchronize the packets, whereas in the latter, no synchronization is necessary. The input interface in both cases extracts the headers; the input control detects the headers and sends the header content information to the switch control where forwarding table lookup, contention resolution, arbitration, and forwarding decision take place. The switch controller then sends the new switching state command to the switching fabric and also sends new header information to the output control so that the switching fabric will switch and the output interface can conduct
1077-260X/$26.00 © 2011 IEEE
KURUMIDA AND YOO: NONLINEAR OPTICAL SIGNAL PROCESSING IN OPTICAL PACKET SWITCHING SYSTEMS
header replacement (and possibly signal regeneration). While the asynchronous, variable-length packet switching relieves the need for packet synchronization or segmentation processes, the contention probability is higher in this case, so that more effective contention resolution schemes are necessary. While the conventional electronic packet switching used electronic signal processing in both control and data planes, future OPS systems can utilize all-optical processing for many such processes to exploit high-speed, parallelism, and potentially low power consumption compared to electronic counterparts. In particular, nonlinear optical (NLO) signal processing can exploit nonlinear transfer functions essential for achieving packet switching. This paper reviews and compares the NLO signal processing in OPS systems.
979
Fig. 2. Block diagram of an optoelectronic packet-switching system using time-to-wavelength conversion (courtesy of [26]).
II. SIGNAL PROCESSING NEEDS IN OPTICAL PACKET SWITCHING SYSTEMS As discussed earlier, optical signal processing can reside in the control plane and the data plane of OPS systems. The control plane of OPS systems included label (header) processing such as optical label (header) extraction, lookup, rewriting (reinsertion). The data plane of OPS systems included optical signal regeneration and optical switching in time, space, and wavelength domains. A. Optical Signal Processing in the Control Plane (Label Processing) Optical labels (headers) should be easily detached from and attached to the optical data payload in OPS systems using optical techniques in order to avoid high-speed electronic processing in the data plane. Hence, the format of the optical label (header) has important implications for the OPS system architecture. Once extracted, the label can be all-optically or electronically recognized and the new label can be generated based on optical or electronic label processing. Various label encoding schemes based on serial or parallel processing methods have been demonstrated [10]. Examples of parallel optical labeling methods are: subcarrier multiplexing [11]–[15], wavelength multiplexing [16], orthogonal modulations [17], and optical code based optical labels [10], [18], [19]. Serial optical labeling methods utilize time domain multiplexed (TDM) labels [20]. Koch et al., [21] showed high-speed payload envelope detection and time domain switching utilizing nonlinear optical transfer functions of lasers and cross-gain modulation (XGM) semiconductor optical amplifiers (SOAs). While optoelectronic types (O/E-E/O) of label processors have been demonstrated using serial-to-parallel conversion techniques at 100 Gb/s [22], the advantages of all-optical label processing can potentially achieve lower power, higher bit rates, and lower latency for label detection. The main disadvantage of the all-optical label detection is the scalability in the number of labels it can detect and process compared to the optoelectronic counterparts supported by DRAMs that can support larger than 8 GByte label space. The current research direction is mostly in all-optical label processing combining a hybrid-configuration for future photonic routers [23], [24]. This section discusses signal pro-
cessing methods and the processing speed for OPS systems from linear and nonlinear all-optical signal processing. 1) Optical Logic Gates in the Control Plane: One of the most straightforward methods for optical labels is bit serial processing combined with electronic logic functions. While serial label processing requires relatively strict control of the timing between the label and the payload, it can still support optical transparency for the data payload. Typically, hybrid optical signal processing including O/E header processing is utilized. Since the large signal integrated (LSI) circuits do not support clock speed beyond 10 GHz, they typically employ serializer/deserializer (SERDES) for label recognition functions while paying the penalty in latency, which strictly depends on synchronization time between clock and signal. On the other hand, all-optical label processing methods for bit serial processing methods can achieve low power and high bit-rate label processing, for example, by using multistage switches [25]. Photonic integration circuits (PICs) can greatly reduce latency and power, and support high-speed serial label processing. An example of hybrid optical label signal processing is to combine optical and electronic processing method including alloptical serial-to-parallel conversion by the time-to-wavelength mapping method. Fig. 2 shows a schematic diagram of such a system by Teimoori et al. [26]. A combination of the time-towavelength all-optical processor and the electronics (ADC and FPGA) forms a label processor within the optoelectronic packet switching system. In this example, the time-to-wavelength converter converts the time domain label to a parallel bit stream, which will be detected by the photo detectors (PDs) and the low voltage differential signaling (LVDS) circuits. There are several types of label recognition circuits based on optical logic gate functions in the optoelectronic packet-switch [18], [26], [27]. The optical logic gates combined with serial to parallel conversion can support rapid decision processes based on the label content. An example of such optical logic gates includes cascaded semiconductor optical amplifier- Mach-Zehnder interferometers (SOA-MZI) [28]. By exploiting the nonlinear transfer function of the SOA-MZI, one can construct an optical XOR logic
980
IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 18, NO. 2, MARCH/APRIL 2012
gate. It is then possible to construct a programmable logic gate consisting of many such XOR logic gates for label detection and processing. While the level of photonic integration available today for such SOA-MZIs is still far below that of the electronic logic gates, the optical logic gates can conduct high-speed (>10 Gb/s) label processing for several byte labels. Another example of optical devices with nonlinear optical transfer functions to support can support logic functions is an ultrafast nonlinear interferometer (UNI) [29], [30]. UNI is a single-arm interferometer, and can be realized a stable highspeed logic processor. Because pump and signal path in the nonlinear material has an identical path, unstable refractive index changes of the nonlinear device can be eliminated. For making several bits label recognition, number of UNI logic gate is required [27]. Again, photonic integration of the UNI circuits is a must for UNI to be useful in the control plane of OPS systems. 2) Optical Label Pattern Recognition in the Control Plane: Optical labels using optical code-division-multiple-access (O-CDMA) techniques support all-optical label encoding, decoding, and conversion [31]–[33]. O-CDMA research investigated optical codes applied in various combinations of wavelength and time domains [34]. Further, O-CDMA can utilize encoding in the phase or the amplitude domains, and spectral phase encoding provides greater cardinality compared to the amplitude counterpart [35], [36]. On the other hand, decoding requires nonlinear processing either by using optical or electronic methods since decoded optical codes will exhibit differing peak power levels with identical integrated energy [37]. Fig. 3 shows the conceptual setup for a correlation technique based on fiber Bragg gratings (FBGs) and the nonlinear peak detection using photodiodes (PDs). Here, the time domain wavelength label encoded by three different colors is split into three separate label detection circuits each including an FBG, a PD, and an optical circulator. The label detection circuit with the correct color sequence encoded on the FBG will yield the high peak power level at the PD. Then the label detection circuit will activate the corresponding optical gate switch to perform optical packet switching based on the label content. Instead of using PDs and electronic circuits, all-optical detection using nonlinear optical circuits can overcome the latency associated with electronic signal processing. As an example, parallel optical correlation in the time domain using an optical code is distinctive way to realize ultrafast label processing [38] at the expense of requiring multiple O-CDMA decoder optical circuits in parallel. Wada et al. [38] describes 160-Gb/s OTDM data payload on eight wavelengths with optical code-division multiple-access (O-CDMA) labels detected in FBG-based label. Further, Furukawa et al. [39] introduced multiple optical code label processing using arrayed waveguide grating (AWG) based multiencoder and 10 GEthernet to 80 Gb/s packet converter. Fig. 4 shows the architecture and the optical code encoder/decoder. B. Optical Switching in the Data Plane Optical switching functions in OPS systems are enabled by the optical switch fabric and the control interface to the control
Fig. 3. Conceptual setup for a correlation technique based on fiber Bragg gratings.
Fig. 4. (a) Architecture of OPS system. (b) Multiple optical label processing (courtesy of [39]).
unit. In this section, we will consider switching in the space domain and consider wavelength conversion and time buffering in later sections. In particular, a combination of tunable wavelength conversion and spectral deflection can provide space switching (and wavelength switching). Since optical packet signals are considered to be in very short durations with typical packet sizes in the 10 ns∼1 μs range, OPS systems require rapid switching typically in the nanosecond timescale. While all-optical switching technologies typically offer high data rate, scalability, and low latency switching, their characteristics are strongly dependent on the control plane. The control interfaces and the data plane architecture strongly affect the system performance. The following shows several examples of switches using linear or nonlinear optical transfer functions. 1) Linear Optical Switching Methods: a) Multimode-Interference (MMI) Switches: Optical switch fabric architectures often involve switching units of size 1 × 2 or 2 × 2. Takeda et al. [40] describe all optical switches and flip-flops based on a multimode-interference (MMI) bistable laser diode (BLD) with distributed Bragg reflectors (DBRs). Fig. 5 shows the proposed device architecture. The structure works as a single flip-flop (FF) logic element. This type of device has wavelength tuning function
KURUMIDA AND YOO: NONLINEAR OPTICAL SIGNAL PROCESSING IN OPTICAL PACKET SWITCHING SYSTEMS
Fig. 7.
981
Configuration of multiformat OPS [44].
Fig. 5. Multimode-interference bistable laser diode with distributed Bragg reflectors [40].
Fig. 6.
Schematic of 1×5 InP–InGaAsP optical phased-array switch [42].
by the band-filling effect and the free-carrier plasma effect with carrier injections to the DBRs. A large-scale switch can incorporate them in a PIC by monolithic integration. Based on this structure, the dynamic FF operation of the device was investigated. An experimental result shows fast switching time 280 and 228 ps rising and falling times, respectively. The switching speed has an enough capability for OPS system, and it has compatibility to the waveguide devices. b) Waveguide Switches: A semiconductor waveguide type optical switch has an advantage of high-speed switching. Sato et al. [41] proposed an X-shaped optical waveguide switch which consists of GaAs-based double-heterostructure ridge waveguides. It operates at 40 Gb/s and achieves less than 1.4 ns switching time by electronic triggering. The structure of the switch facilitates the fabrication of the number of switches on the single wafer. Tanemura et al. [42] proposed and demonstrated an integrated optical phased-array switch. Fig. 6 shows schematic of 1 × 5 InP-InGaAsP optical phased-array switch [42]. Wideband operation in C-band (1520–1580 nm) and fast switching response below 10 ns are reported. The proposed switch has been utilized for OPS system, 160-Gb/s optical time-domain-multiplexed packets were dynamically switched to 16 outputs with power penalties of 0.7 dB [43] implying possible applications in high-throughput OPS systems.
2) Nonlinear Optical Switching Methods: a) Arrayed Waveguide Grating Router (AWGR) and Tunable Wavelength Converters: A combination of tunable wavelength conversion and wavelength routing devices can provide space and wavelength switching functions. A typical configuration is to use AWGR together with a tunable wavelength converter at the input stage and a fixed wavelength converter at the output stage. However, all-optical wavelength converter based on an SOA-MZI can support digital transparency. Therefore, OPS routers with optical-label switching technology have an opportunity to transparently support multiple data rates. On the other hand, wavelength conversion technologies based on parametric nonlinear optical processes can support any data format and protocols. Kurumida et al. [44] reported multiformat OPS using four-wave-mixing (FWM). Fig. 7 shows the diagram of an OPS router for multiformat switching. It consists of function blocks, called label extractor (LE) and label rewriter (LR) for subcarrier multiplexing (SCM) techniques. The key function blocks for the switching router, are the tunable wavelength converter (T-WC) and the fixed wavelength converter (F-WC). These blocks must provide format independent conversion for multiformat packets. For more wavelength controllability, dualpump FWM wavelength conversion based packet switching system was also demonstrated [45]. A wavelength control method would be an important issue including a suppression method of the pump wavelength. Wavelength conversion techniques will be discussed in the next section. b) Ultrafast Nonlinear Interferometer (UNI) Gates: Ultrafast nonlinear interferometers (UNI) [30] can be used as an all-optical high-speed gate for OPS systems. UNI provides several picosecond switching window by utilizing optically induced gain and index nonlinearities in a semiconductor amplifier. As mentioned in Section II-2, this type of device can have a stable logic function, however, configuration requires a relatively strict control of the timing for gate open/close control. This characteristic is suitable for OTDM technologies, but a gate switching method is applicable to label/payload (L/P) separation function in OPS systems. An all-optical L/P separation method using UNI AND gate with clock recover has been demonstrated [46]. The switching technique is able to be in principle extendable to higher rates (>100 Gb/s) with appropriate clock recovery sub-systems. On the other hand, photonic integration of UNI gates to a large fabric can face challenges due to polarization characteristics.
982
IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 18, NO. 2, MARCH/APRIL 2012
C. Wavelength Conversion Wavelength conversion allows flexible signal wavelength assignments in the network [47]. All-optical wavelength conversion technologies have a potential advantage over the optoelectric counterpart in realizing relatively lower packaging costs and dimensions. This section discusses wavelength conversion techniques for OPS system. 1) Four-Wave Mixing (FWM): Nonlinear optical wavemixing supports a transparent condition in optical networks preserveing phase and amplitude information simultaneously. This feature performs format independent wavelength conversion. Conversion efficiency of the wavelength in the fiber is the most critical issues for the system designs; therefore, highly nonlinear fibers (HNLFs) with improvements in the conversion efficiency have been pursued even for relatively short (
IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 18, NO. 2, MARCH/APRIL 2012
Nonlinear Optical Signal Processing in Optical Packet Switching Systems Junya Kurumida, Member, IEEE, and S. J. Ben Yoo, Fellow, IEEE (Invited Paper)
Abstract—This paper discusses nonlinear optical signal processing employed in optical packet switching systems. Nonlinear optical signal processing provides optical label (header) recognition, optical switching, wavelength conversion, and time buffering with typically higher capacity, lower latency, and lower power consumption than electronic counterparts. In order to provide diverse signal processing functions, large-scale integration of nonlinear optical signal processing devices is essential. We discuss possible future directions in optical packet switching involving nonlinear optical signal processing of optical packets with advanced data and label modulation formats. Index Terms—Internet, nonlinear optics, optical burst switching, optical circuit switching, optical flow switching, optical packet switching, optical-label switching, photonic switching.
I. INTRODUCTION HE FUTURE Internet expects to demand protocol-agile and high capacity networking in support of a variety of applications including 3-D multimedia entertainment, telemedicine, and cloud computing. The Internet Protocol (IP) supports all applications on any physical layer platforms, following the hour-glass model [1]. While the generality and heterogeneity of the ‘‘Internet hourglass” are the critical strengths of the Internet that made it so successful, new multimedia, data services, and cloud computing are driving the needs for high performance, high utilization, and secure networks. A unified networking platform in support of voice, data, and multimedia applications are attractive especially on a high-capacity optical layer. Since IP uses packets as the unit of transport and switching, the capability to switch and transport packets with low latency and high throughput can enhance the performance of the IP network. As IP packets are most naturally accommodated in the form of packet switching, optical packet switching may be the technology best suited for the future Internet providing both high-capacity and high-utilization, aggregation [2] of such packets at the edge into bursts, flows, or circuits can further improve energy efficiency and reduce the complexity of the control plane. In particular, optical-label switching (OLS) [3]
T
Manuscript received February 19, 2011; accepted April 4, 2011. Date of publication July 29, 2011; date of current version March 2, 2012. J. Kurumida is with the National Institute of Advanced Industrial Scienece and Technology, Tsukuba, Ibaraki, 305-8568 Japan (e-mail: [email protected]). S. J. B. Yoo is with the Department of Electrical and Computer Engineering, University of California, Davis, CA 95616 USA (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSTQE.2011.2143390
Fig. 1. Optical packet switching architecture with (a) synchronous and fixedlength packet forwarding, and (b) asynchronous and variable-length packet forwarding.
can utilize a unified control plane in support of optical packet switching (OPS), optical burst switching (OBS), optical flow switching (OFS), and optical circuit switching (OCS) [4]–[6]. The unified control plane [7], [8] of OLS also supports applications’ differing degrees [9] of quality of service (QoS), class of service (CoS), and type of service (ToS) requirements by mapping them onto the optical labels [4]–[6]. The minimum requirement of the OLS system is to support OPS with low latency, high throughput, and high scalability. Fig. 1 illustrates a simplified schematic of an OPS system for (a) synchronous and fixed-length packet forwarding, and (b) asynchronous and variable-length packet forwarding. In both cases, the OPS system includes a switch controller, an input controller, and an output controller in the control plane, and an input interface, a switch fabric, and an output interface in the data plane. In the synchronous OPS, the input interface must synchronize the packets, whereas in the latter, no synchronization is necessary. The input interface in both cases extracts the headers; the input control detects the headers and sends the header content information to the switch control where forwarding table lookup, contention resolution, arbitration, and forwarding decision take place. The switch controller then sends the new switching state command to the switching fabric and also sends new header information to the output control so that the switching fabric will switch and the output interface can conduct
1077-260X/$26.00 © 2011 IEEE
KURUMIDA AND YOO: NONLINEAR OPTICAL SIGNAL PROCESSING IN OPTICAL PACKET SWITCHING SYSTEMS
header replacement (and possibly signal regeneration). While the asynchronous, variable-length packet switching relieves the need for packet synchronization or segmentation processes, the contention probability is higher in this case, so that more effective contention resolution schemes are necessary. While the conventional electronic packet switching used electronic signal processing in both control and data planes, future OPS systems can utilize all-optical processing for many such processes to exploit high-speed, parallelism, and potentially low power consumption compared to electronic counterparts. In particular, nonlinear optical (NLO) signal processing can exploit nonlinear transfer functions essential for achieving packet switching. This paper reviews and compares the NLO signal processing in OPS systems.
979
Fig. 2. Block diagram of an optoelectronic packet-switching system using time-to-wavelength conversion (courtesy of [26]).
II. SIGNAL PROCESSING NEEDS IN OPTICAL PACKET SWITCHING SYSTEMS As discussed earlier, optical signal processing can reside in the control plane and the data plane of OPS systems. The control plane of OPS systems included label (header) processing such as optical label (header) extraction, lookup, rewriting (reinsertion). The data plane of OPS systems included optical signal regeneration and optical switching in time, space, and wavelength domains. A. Optical Signal Processing in the Control Plane (Label Processing) Optical labels (headers) should be easily detached from and attached to the optical data payload in OPS systems using optical techniques in order to avoid high-speed electronic processing in the data plane. Hence, the format of the optical label (header) has important implications for the OPS system architecture. Once extracted, the label can be all-optically or electronically recognized and the new label can be generated based on optical or electronic label processing. Various label encoding schemes based on serial or parallel processing methods have been demonstrated [10]. Examples of parallel optical labeling methods are: subcarrier multiplexing [11]–[15], wavelength multiplexing [16], orthogonal modulations [17], and optical code based optical labels [10], [18], [19]. Serial optical labeling methods utilize time domain multiplexed (TDM) labels [20]. Koch et al., [21] showed high-speed payload envelope detection and time domain switching utilizing nonlinear optical transfer functions of lasers and cross-gain modulation (XGM) semiconductor optical amplifiers (SOAs). While optoelectronic types (O/E-E/O) of label processors have been demonstrated using serial-to-parallel conversion techniques at 100 Gb/s [22], the advantages of all-optical label processing can potentially achieve lower power, higher bit rates, and lower latency for label detection. The main disadvantage of the all-optical label detection is the scalability in the number of labels it can detect and process compared to the optoelectronic counterparts supported by DRAMs that can support larger than 8 GByte label space. The current research direction is mostly in all-optical label processing combining a hybrid-configuration for future photonic routers [23], [24]. This section discusses signal pro-
cessing methods and the processing speed for OPS systems from linear and nonlinear all-optical signal processing. 1) Optical Logic Gates in the Control Plane: One of the most straightforward methods for optical labels is bit serial processing combined with electronic logic functions. While serial label processing requires relatively strict control of the timing between the label and the payload, it can still support optical transparency for the data payload. Typically, hybrid optical signal processing including O/E header processing is utilized. Since the large signal integrated (LSI) circuits do not support clock speed beyond 10 GHz, they typically employ serializer/deserializer (SERDES) for label recognition functions while paying the penalty in latency, which strictly depends on synchronization time between clock and signal. On the other hand, all-optical label processing methods for bit serial processing methods can achieve low power and high bit-rate label processing, for example, by using multistage switches [25]. Photonic integration circuits (PICs) can greatly reduce latency and power, and support high-speed serial label processing. An example of hybrid optical label signal processing is to combine optical and electronic processing method including alloptical serial-to-parallel conversion by the time-to-wavelength mapping method. Fig. 2 shows a schematic diagram of such a system by Teimoori et al. [26]. A combination of the time-towavelength all-optical processor and the electronics (ADC and FPGA) forms a label processor within the optoelectronic packet switching system. In this example, the time-to-wavelength converter converts the time domain label to a parallel bit stream, which will be detected by the photo detectors (PDs) and the low voltage differential signaling (LVDS) circuits. There are several types of label recognition circuits based on optical logic gate functions in the optoelectronic packet-switch [18], [26], [27]. The optical logic gates combined with serial to parallel conversion can support rapid decision processes based on the label content. An example of such optical logic gates includes cascaded semiconductor optical amplifier- Mach-Zehnder interferometers (SOA-MZI) [28]. By exploiting the nonlinear transfer function of the SOA-MZI, one can construct an optical XOR logic
980
IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 18, NO. 2, MARCH/APRIL 2012
gate. It is then possible to construct a programmable logic gate consisting of many such XOR logic gates for label detection and processing. While the level of photonic integration available today for such SOA-MZIs is still far below that of the electronic logic gates, the optical logic gates can conduct high-speed (>10 Gb/s) label processing for several byte labels. Another example of optical devices with nonlinear optical transfer functions to support can support logic functions is an ultrafast nonlinear interferometer (UNI) [29], [30]. UNI is a single-arm interferometer, and can be realized a stable highspeed logic processor. Because pump and signal path in the nonlinear material has an identical path, unstable refractive index changes of the nonlinear device can be eliminated. For making several bits label recognition, number of UNI logic gate is required [27]. Again, photonic integration of the UNI circuits is a must for UNI to be useful in the control plane of OPS systems. 2) Optical Label Pattern Recognition in the Control Plane: Optical labels using optical code-division-multiple-access (O-CDMA) techniques support all-optical label encoding, decoding, and conversion [31]–[33]. O-CDMA research investigated optical codes applied in various combinations of wavelength and time domains [34]. Further, O-CDMA can utilize encoding in the phase or the amplitude domains, and spectral phase encoding provides greater cardinality compared to the amplitude counterpart [35], [36]. On the other hand, decoding requires nonlinear processing either by using optical or electronic methods since decoded optical codes will exhibit differing peak power levels with identical integrated energy [37]. Fig. 3 shows the conceptual setup for a correlation technique based on fiber Bragg gratings (FBGs) and the nonlinear peak detection using photodiodes (PDs). Here, the time domain wavelength label encoded by three different colors is split into three separate label detection circuits each including an FBG, a PD, and an optical circulator. The label detection circuit with the correct color sequence encoded on the FBG will yield the high peak power level at the PD. Then the label detection circuit will activate the corresponding optical gate switch to perform optical packet switching based on the label content. Instead of using PDs and electronic circuits, all-optical detection using nonlinear optical circuits can overcome the latency associated with electronic signal processing. As an example, parallel optical correlation in the time domain using an optical code is distinctive way to realize ultrafast label processing [38] at the expense of requiring multiple O-CDMA decoder optical circuits in parallel. Wada et al. [38] describes 160-Gb/s OTDM data payload on eight wavelengths with optical code-division multiple-access (O-CDMA) labels detected in FBG-based label. Further, Furukawa et al. [39] introduced multiple optical code label processing using arrayed waveguide grating (AWG) based multiencoder and 10 GEthernet to 80 Gb/s packet converter. Fig. 4 shows the architecture and the optical code encoder/decoder. B. Optical Switching in the Data Plane Optical switching functions in OPS systems are enabled by the optical switch fabric and the control interface to the control
Fig. 3. Conceptual setup for a correlation technique based on fiber Bragg gratings.
Fig. 4. (a) Architecture of OPS system. (b) Multiple optical label processing (courtesy of [39]).
unit. In this section, we will consider switching in the space domain and consider wavelength conversion and time buffering in later sections. In particular, a combination of tunable wavelength conversion and spectral deflection can provide space switching (and wavelength switching). Since optical packet signals are considered to be in very short durations with typical packet sizes in the 10 ns∼1 μs range, OPS systems require rapid switching typically in the nanosecond timescale. While all-optical switching technologies typically offer high data rate, scalability, and low latency switching, their characteristics are strongly dependent on the control plane. The control interfaces and the data plane architecture strongly affect the system performance. The following shows several examples of switches using linear or nonlinear optical transfer functions. 1) Linear Optical Switching Methods: a) Multimode-Interference (MMI) Switches: Optical switch fabric architectures often involve switching units of size 1 × 2 or 2 × 2. Takeda et al. [40] describe all optical switches and flip-flops based on a multimode-interference (MMI) bistable laser diode (BLD) with distributed Bragg reflectors (DBRs). Fig. 5 shows the proposed device architecture. The structure works as a single flip-flop (FF) logic element. This type of device has wavelength tuning function
KURUMIDA AND YOO: NONLINEAR OPTICAL SIGNAL PROCESSING IN OPTICAL PACKET SWITCHING SYSTEMS
Fig. 7.
981
Configuration of multiformat OPS [44].
Fig. 5. Multimode-interference bistable laser diode with distributed Bragg reflectors [40].
Fig. 6.
Schematic of 1×5 InP–InGaAsP optical phased-array switch [42].
by the band-filling effect and the free-carrier plasma effect with carrier injections to the DBRs. A large-scale switch can incorporate them in a PIC by monolithic integration. Based on this structure, the dynamic FF operation of the device was investigated. An experimental result shows fast switching time 280 and 228 ps rising and falling times, respectively. The switching speed has an enough capability for OPS system, and it has compatibility to the waveguide devices. b) Waveguide Switches: A semiconductor waveguide type optical switch has an advantage of high-speed switching. Sato et al. [41] proposed an X-shaped optical waveguide switch which consists of GaAs-based double-heterostructure ridge waveguides. It operates at 40 Gb/s and achieves less than 1.4 ns switching time by electronic triggering. The structure of the switch facilitates the fabrication of the number of switches on the single wafer. Tanemura et al. [42] proposed and demonstrated an integrated optical phased-array switch. Fig. 6 shows schematic of 1 × 5 InP-InGaAsP optical phased-array switch [42]. Wideband operation in C-band (1520–1580 nm) and fast switching response below 10 ns are reported. The proposed switch has been utilized for OPS system, 160-Gb/s optical time-domain-multiplexed packets were dynamically switched to 16 outputs with power penalties of 0.7 dB [43] implying possible applications in high-throughput OPS systems.
2) Nonlinear Optical Switching Methods: a) Arrayed Waveguide Grating Router (AWGR) and Tunable Wavelength Converters: A combination of tunable wavelength conversion and wavelength routing devices can provide space and wavelength switching functions. A typical configuration is to use AWGR together with a tunable wavelength converter at the input stage and a fixed wavelength converter at the output stage. However, all-optical wavelength converter based on an SOA-MZI can support digital transparency. Therefore, OPS routers with optical-label switching technology have an opportunity to transparently support multiple data rates. On the other hand, wavelength conversion technologies based on parametric nonlinear optical processes can support any data format and protocols. Kurumida et al. [44] reported multiformat OPS using four-wave-mixing (FWM). Fig. 7 shows the diagram of an OPS router for multiformat switching. It consists of function blocks, called label extractor (LE) and label rewriter (LR) for subcarrier multiplexing (SCM) techniques. The key function blocks for the switching router, are the tunable wavelength converter (T-WC) and the fixed wavelength converter (F-WC). These blocks must provide format independent conversion for multiformat packets. For more wavelength controllability, dualpump FWM wavelength conversion based packet switching system was also demonstrated [45]. A wavelength control method would be an important issue including a suppression method of the pump wavelength. Wavelength conversion techniques will be discussed in the next section. b) Ultrafast Nonlinear Interferometer (UNI) Gates: Ultrafast nonlinear interferometers (UNI) [30] can be used as an all-optical high-speed gate for OPS systems. UNI provides several picosecond switching window by utilizing optically induced gain and index nonlinearities in a semiconductor amplifier. As mentioned in Section II-2, this type of device can have a stable logic function, however, configuration requires a relatively strict control of the timing for gate open/close control. This characteristic is suitable for OTDM technologies, but a gate switching method is applicable to label/payload (L/P) separation function in OPS systems. An all-optical L/P separation method using UNI AND gate with clock recover has been demonstrated [46]. The switching technique is able to be in principle extendable to higher rates (>100 Gb/s) with appropriate clock recovery sub-systems. On the other hand, photonic integration of UNI gates to a large fabric can face challenges due to polarization characteristics.
982
IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 18, NO. 2, MARCH/APRIL 2012
C. Wavelength Conversion Wavelength conversion allows flexible signal wavelength assignments in the network [47]. All-optical wavelength conversion technologies have a potential advantage over the optoelectric counterpart in realizing relatively lower packaging costs and dimensions. This section discusses wavelength conversion techniques for OPS system. 1) Four-Wave Mixing (FWM): Nonlinear optical wavemixing supports a transparent condition in optical networks preserveing phase and amplitude information simultaneously. This feature performs format independent wavelength conversion. Conversion efficiency of the wavelength in the fiber is the most critical issues for the system designs; therefore, highly nonlinear fibers (HNLFs) with improvements in the conversion efficiency have been pursued even for relatively short (
