Open Access
Add to BibSonomyAbstract
We demonstrated, for the first time, a field trial of 160 (16λ × 10) Gbit/s, fine granularity, DWDM-based optical packet switching and transmission by newly- eveloped burst-mode EDFAs and an optical packet switch prototype with multiple all-optical label processors. We achieved 64 km field transmission and switching of 160 (16λ × 10) Gbit/s DWDM-based optical packets encapsulating almost 10 Gbit/s IP packets with error-free operation (IP-packet-loss-rate <10-6and bit-error-rate <10-9).
©2008 Optical Society of America
1. Introduction
Optical packet switch (OPS) network is a promising candidate for a backbone network with high scalability, fine granularity, and efficient bandwidth utilization. Recently, while a 40 Gbit/s electronic router is available [1], many OPS systems have been proposed and demonstrated to forward higher-speed (over 40 Gbit/s) optical packets despite the relative immaturity of optical technologies [2–10]. Since the payload’s data path including switches and buffers consists of pure optical devices, OPS systems have transparency for optical signals of various bit-rate or formats. On the other hand, in [2–5], these systems rely on electronics for label processing, a function for recognizing a label of an optical packet and determining its designated port. However, electronic label processing requires memory access, an operation which will inevitably limit the potential of optics to switch an extremely large amount of traffic.
One feature of our proposed OPS system is the introduction of optical correlation techniques for label processing. Property unique to optical label processing is that label recognition is an analog operation entirely performed in the optical domain, and no logic operations are required. Therefore, the packet processing speed is limited only by the propagation delay in the optical device in this way, and optical-label-processing based OPS systems can provide larger scalability and ultrahigh speed hopping to networks. Previously, we have experimentally demonstrated OPS systems based on optical label (i.e., code) processing [8, 9]. We have also designed high-throughput electronic scheduling for optical buffering, confirmed the performance by simulation [11], and implemented it on FPGA for experimental demonstration [12]. In 2005, based on these technologies, our group developed a 160 Gbit/s/port OPS prototype with narrow-band optical label processing, optical switching, optical buffering, electrical scheduling and optical signal multiplexing/demultiplexing, and demonstrated 160 Gbit/s optical time division multiplexing (OTDM) based optical packet switching [13].
To provide high-speed (40 Gbit/s or higher) optical packets, OTDM technologies using ultra-short (several picoseconds) optical pulses have been introduced [7, 13]. However, since ultra-short optical pulses are likely to be distorted due to dispersion or nonlinear effects in long-haul fiber transmission, it is difficult to develop OTDM-based high-speed optical packet transmission systems compared to current 10 Gbit/s signal transmission systems. In addition, there is an interface gap between high-speed OPS networks and slow-speed (10 Gbit/s or slower) metro or access networks based on Internet Protocol (IP) technologies at edge nodes. To solve these problems, we proposed to introduce dense wavelength division multiplexing (DWDM) technologies into OPS systems [14]. Since a DWDM-based optical packet consists of plural 10 Gbit/s payloads with different wavelengths, it is compatible with the matured 10 Gbit/s signal transmission systems and the interface of slow-speed networks. Previously, we have developed novel network-interface devices: 10Gigabit Ethernet (10GbE) / 80 Gbit/s optical-packet (80GbOP) converters and arrayed burst-mode optical packet transmitters/receivers for connecting OPS networks with IP-technology-based metro or access networks [15]. The network-interface devices can encapsulate almost 10 Gbit/s IP packets into 80 (8λ × 10) Gbit/s WDM-based optical packets. We have demonstrated error free 80 (8λ × 10) Gbit/s DWDM based optical packet switching by the interface devices and our OPS prototype [16]. However, DWDM-based optical packet transmission has not been implemented.
In this paper, we demonstrate a field trial of DWDM-based optical packet switching and transmission. The payload’s data-rate of optical packets is increased from 80 Gbit/s to 160 Gbit/s. Using newly developed burst-mode erbium doped fiber amplifiers (EDFAs) and our OPS prototype based on multiple optical label processing, 64 km fiber transmission and switching of 160 (16λ × 10) Gbit/s DWDM-based optical packets encapsulating IP packets is implemented in a field environment. Error-free operation with an IP packet-loss-rate (IP-PLR) of less than 10-6 and a bit-error-rate (BER) of less than 10-9 is achieved. The recent papers ([17]) have shown a brief overview about the field trial, but this paper addresses key technologies and the technical issues in more details.
2. Architecture and key technologies of optical packet switch system
Fig. 3. (a) Multiple optical en/decoder. (b) Generated 16-chip PSK optical code and (c) output auto-correlation signal.
Fig. 4. (a) Developed burst-mode EDFA. (b) Amplified optical packet by commercially available EDFA and (c) our developed EDFA.
Figure 1(a) shows the concept of a proposed OPS network. The OPS network is connected to IP-technology-based metro or access networks. Namely, at an ingress edge node of the OPS network, the edge node system encapsulates an IP packet over Ethernet or SONET/SDH into an optical packet. Conversely, at an egress edge node, the IP packet is decapsulated from the optical packet.
In OPS networks, we introduce large-capacity colored optical packets using DWDM technology. As shown in Fig. 1(b), a DWDM-based optical packet consists of multiple 10 Gbit/s optical payloads of different N-wavelengths and an optical label. Thus, the data rate of multiple optical payloads in a DWDM-based optical packet is N × 10 Gbit/s. Since 10 Gbit/s transmission systems are mature and are easy to interface with electronic systems, we can develop an OPS system more easily compared to a system involving transmission of an OTDM-based high-speed optical packet consisting of a single, large-bandwidth wavelength. Theoretically, when a 10 Gbit/s IP packet is converted into the N × 10 Gbit/s DWDM-based optical packet, the length of the optical packet in time is about 1/N compared with that of the original IP packets. Because of its fine granularity of DWDM-based optical packets, we can fully use the bandwidth of OPS networks.
At core nodes, optical packets are forwarded to an appropriate edge node by OPS systems. Our proposed OPS system roughly consists of optical label processing, optical switching, optical buffering, electrical scheduling, and electrical routing, as shown in Fig. 2. An optical label processor controls optical switches to forward optical packets to appropriate output ports according to a packet label, and give arrival information to an electrical scheduler. In the label processing, the analysis of a label is performed based on parallel optical correlation in the time domain using optical codes (OC) as optical labels [8, 9]. A set of optical correlators acts as en/decoders, each storing an OC corresponding to a destination node in a routing table of the OPS network. The encoder generates an OC and the decoder outputs a high-intensity auto-correlation or low-intensity cross-correlation signal in the matched or unmatched cases, respectively. The payload is spread in the time domain. As a result, the label recognition can be achieved by intensity thresholding. In [13], we used en/decoders consisting of some passive devices, such as tunable taps, phase shifters, delays, and combiners. However, since one en/decoder handles only one OC, it is necessary to provide a number of en/decoders in order to handle many labels. This limits the scalability of the OPS system or edge node system. Here, we introduce multiple optical en/decoders into the OPS system [18]. This device has an arrayed waveguide grating (AWG) configuration as shown in Fig. 3(a) and behaves like a transversal filter. It simultaneously generates and processes 16-chip (M= 16) optical phase-shift keying (PSK) OCs with low latency. To generate a full set of OCs, we send a short laser pulse into one of the device input port, and at the device output ports we obtain sixteen different OCs. Each code is composed of 16 pulses (which are often termed chips in the literature) with a different phase that is a PSK code). The time interval between two consecutive chips is Δτ = 5 ps, so that the code chip-rate is 1/Δτ = 200 Gchip/s. The label processing speed is 1/(M - 1)Δτ = 13.3 × 109 packet/s. Figures 3(b) and 3(c) show one OC generated at an output port of the multiple optical encoder and a single auto-correlation signal from a matched port of the decoder, respectively.
Fig. 5. (a) Field trial configuration of 160 (16λ × 10) Gbit/s optical packet switching and area where field fiber lines are located. (b) Photograph of experimental setup. (c) One round-trip compensated dispersion of fiber line between Keihannna and Nara in JGN2 testbed networks, dispersion of 32.0 km SMF and 11.5 km DCF.
For optical switching, we use 1×2 LiNbO3 switches. Since the switching speed is < 50 ps, even a short-duration optical packet can be efficiency switched with a short-duration gap between packets. An electrical scheduler controls optical buffers to avoid packet collisions. Optical buffers, which consist of optical switches and fiber-delay-lines with different length, store packets for appropriate time [12, 13].
An optical amplifier such as an EDFA is useful to compensate for the power loss of optical signals caused by long-haul fiber transmission or optical processing. However, the transient response of an EDFA distorts the waveform even in the case of a short-duration packet of ~400 ns, despite the much longer lifetime of Er3+ (about 10 ms), which causes serious impairment depending on the link utilization [19]. In addition, when many EDFAs are used in a system, the cumulative transient response effect is very large. To compensate such the transient of EDFAs, automatic gain controlling (AGC) techniques have been proposed. Actually, the response time of AGC was 20 ns at the best [20]. However, the increase of bit rate beyond 40 Gbit/s or a parallel encoding technique such as DWDM technology [14] accelerate shortening of the duration of optical packets more and more. Then, it is difficult to compensate for the transient response during such a short time by electronic control. As an alternative, we developed a new class of EDFAs employing erbium doped fiber (EDF) with enhanced active erbium area as shown in Fig. 4(a) [21]. Figures 4(b) and 4(c) show the effect of the developed EDFA. Using our EDFA, we successfully suppressed the transient response as compared with a commercially available one. We therefore introduced this burst-mode EDFA for long-haul fiber transmission and optical packet switching.
3. Experiment
We conducted a field trial of long-haul transmission and switching of 160 (16λ × 10) Gbit/s optical packets accommodating IP packets. Figures 5(a) and 5(b) show the configuration of the field trial and a photograph of the experimental setup. We set up two ingress/egress edge nodes and a bufferless OPS prototype with two I/O ports as a core node. Ingress edge nodes #1 and #2 encapsulate asynchronous IP packets from end hosts #1 and #2, respectively, into 160 Gbit/s optical packets. The assigned optical labels are “A” at ingress edge node #1 and “B” at ingress edge node #2. Here, half of the 160 Gbit/s optical payload is dummy data generated, and the IP packet is encapsulated into other half of the optical payload. Optical packets are transmitted through JGN2 testbed networks [22]. The fiber lines are located between Keihanna in Kyoto Prefecture and Nara in Nara Prefecture with a loop-back configuration. The dispersion of field fibers is compensated by using dispersion compensating fibers (DCF). One round-trip fiber line between Keihanna and Nara is composed of 32.0 km single-mode fiber (SMF), installed in the field, and 11.5 km DCF. The one round-trip compensated dispersion of field fiber, dispersion of 32.0 km SMF and 11.5 km DCF in C-band is shown in Fig. 5(c). The round-trip loss is about 15 dB including losses at connectors. In the OPS prototype, transmitted 160 Gbit/s optical packets are switched to appropriate output ports according to their optical labels. The 160 Gbit/s optical packets are received in each egress edge node #3 and #4. Finally, the IP packet is decapsulated from the optical packet.
Figure 6 shows the setup of ingress edge nodes. The IP packets over 10GbE frames with different IP addresses were transmitted from network analyzers at the end hosts #1 and #2. The transmitted rates of the 10GbE-frames at the end hosts #1 and #2 were 5.3 Gbit/s and 3.2 Gbit/s, respectively. The frame length was 1500 bytes. At ingress edge node #1, an input IP packet on a 10GbE frame was segmented into eight electrical payloads each of which was 9.95328 Gbit/s by a 10GbE-80GbOP converter [15, 16]. Eight burst-mode optical packet transmitters (Tx.) at different wavelengths were arrayed and generated eight wavelength-channel 10 Gbit/s optical payloads. By multiplexing with a wavelength multi/demultiplexer (e.g., AWG), 80 (8λ × 10) Gbit/s optical payloads were generated with 100 GHz channel spacing (odd wavelength channels). In addition, we generated another 80 (8λ × 10) Gbit/s optical payloads with 100 GHz channel spacing (even wavelength channels) between odd wavelength channels by modulating eight wavelength continuous-wave optical signals with a LiNbO3 intensity modulator (LN-IM). These 80 Gbit/s payloads were multiplexed into 160 Gbit/s one. The wavelength of the sixteen-channel payloads was from 1547.72 nm to 1553.73 nm with 50 GHz channel spacing. In parallel, the 10GbE-80GbOP converter output an electrical label signal according to a look-up table using the destination address of an IP packet. A 10 GHz mode-locked laser diode (MLLD) with a pulse width of 1.9 ps and a center wavelength of 1558 nm was used as a label light source. The 10 GHz optical pulse train was modulated with a LN-IM controlled by the electrical label signal. Then, a 16-chip PSK code was generated as a label “A” by a multiple optical encoder (MOE) at 200 Gchip/s. By combining the optical label and payloads, a 160 (16λ × 10) Gbit/s optical packet was generated. At ingress edge node #2, the same operation was implemented with a different label (label “B”). Figures 6(a)–6(f) show the packet pattern measured by a 3 GHz real-time oscilloscope (Figs. 6(a) and 6(b)), the waveform measured by a sampling oscilloscope (Figs. 6(c) and 6(d)), and the spectra (Figs. 6(e) and 6(f)) of 160 (16λ × 10) Gbit/s optical packets with labels “A” and “B” at the ingress edge nodes #1 and #2, respectively. Figures 6(g) and 6(h) are the eye diagrams of the optical payload in one channel (arrows in Figs. 6(e) and 6(f)).
Figure 7 shows the setup of a bufferless OPS prototype and field-installed fibers. These packets from ingress edge nodes were launched into 32 km SMF, installed in the field, and 11.5 km of DCF. Figures 7(a) and 7(b) are the eye diagrams of the optical payloads at each input port of the OPS prototype in one channel after one round-trip field transmission. In the OPS prototype, we set MODs #1 and #2 to recognize only label “A”. Figures 7(c) and 7(d) show an auto-correlation signal from MOD #1 and cross-correlation signals from MOD #2 for controlling the optical switches. Then, an electrical gate signal was output from a gate signal generator (GSG) with an optical-to-electrical (O/E) converter to open and close an optical 1×2 LiNbO3 switch (LN-SW). Figures 7(e), 7(f), 7(g), and 7(h) show switched packets at output ports “A-1”, “A-2”, “B-1”, and “B-2”. It was confirmed that optical packets with label “A” from edge node #1 were switched to output port “A-1” and optical packets with label “B” from edge node #2 were switched to output port “B-2”.
Figure 8 shows the setup of egress edge nodes and field-installed fibers. Here, we applied one receiving set-up to both egress edge nodes #3 and #4, and measured optical packets alternately. Through field-installed fibers, each optical packet was de-multiplexed into 16 wavelength channels at egress edge nodes #3 and #4. Figures 8(a) and 8(b) show the eye diagram of the de-multiplexed optical payload in one channel at each edge node. The eight arrayed burst-mode optical packet receivers (Rx.) recovered 10 GHz clock signals and 10 Gbit/s electrical payloads in odd wavelength channels. Figures 8(c), 8(d), 8(e) and 8(f) show the waveform and the eye diagram of the recovered electrical payload in one channel at each edge node. Then, an 80GbOP-10GbE converter decapsulated and reconstructed the IP packet from the recovered eight 10 Gbit/s electrical payloads [15, 16]. At each end host (#3 and #4), we measured an IP packet-loss-rate (IP-PLR) with a network analyzer. Here, the IP-PLR from optical packets with labels “A” and “B” are defined as the ratio of the number of non-received IP packets at end hosts #3 and #4 to that of generated IP packets at end hosts #1 and #2, respectively. Figure 9(a) shows the IP-PLR from optical packets with labels “A” and “B” in the field trial, together with that of the back-to-back configuration. The power is that of the sum of eight optical payloads in odd wavelength channels. In addition, at each egress edge node, we measured the BER of eight optical payloads in even wavelength channels with a burst-mode optical packet receiver and a packet error detector (ED) that we developed. The packet ED can measure the BER of a random packet stream in real time [23]. Figures 9(b) and 9(c) show the BER of optical packets with labels “A” and “B” in the field trial, together with that of the back-to-back configuration. The power is that of one optical payload. Error-free operation (IP-PLR < 10-6 and BER ≤ 10-9) was achieved in both cases. The average power penalty of the BER was about 4 dB.
Fig. 9. (a) Packet-loss-rate. (b) Bit-error-rate of transmitted optical packets with label “A” and (c)with label “B”.
In DWDM-based optical packet switching and transmission, we should pay attention to the skew among multiple 10 Gbit/s optical payloads with different wavelength. In the 80GbOP-10GbE converter as an egress edge node, the acceptable maximum skew among the eight optical payloads in odd wavelength channels is about 12.8 ns. Therefore, the total skew among eight wavelength-channel optical payloads on transmission optical fibers should be suppressed below 12.8 ns. In this experiment, as a result of dispersion-compensating on testbed networks, the total dispersion of one round-trip fiber line is below 60 ps/nm in C-band as shown in Fig. 5(c). The wavelength range of eight optical payloads in odd wavelength channels is about 6.0 nm. Therefore, after two round-trip field transmission, the skew among eight wavelength-channel optical payloads in C-band is estimated at 720 ps, which is much lower than the acceptable maximum skew of the 80GOP-10GbE converter. In addition, since the bandwidth of one 10 Gbit/s optical payload is narrower than 100 GHz, the distortion of temporal waveform of the payload can be disregarded.
4. Conclusion
We demonstrated, for the first time, a field trial of 160 (16λ × 10) Gbit/s, fine granularity, DWDM-based optical packet switching and transmission. We developed novel burst-mode EDFAs and a 160 Gbit/s/port OPS prototype with multiple all-optical label encoders/decoders. We achieved 64 km field-transmission and switching of 160 (16λ × 10) Gbit/s optical packets encapsulating IP packets with error-free operation (IP-PLR ≤10-6 and BER ≤10-9). We intend to conduct various trials such as network demonstrations using the developed systems to advance the technology of OPS networks.
Acknowledgments
The authors would like to thank T. Makino, H. Sumimoto, and Y. Tomiyama of the National Institute of Information and Communications Technology for their support in the experiments.
References and links
1. Cisco systems Inc, “Cisco carrier routing system,” http://www.cisco.com/en/US/prod/collateral/routers/ps5763/prodbrochure0900aecd800f8118.pdf.
2. D. J. Blumenthal, A. Carea, L. Rau, V. Curri, and S. Humphries, “All-optical label swapping with wavelength conversion for WDM-IP networks with subcarrier multiplexed addressing,” IEEE Photon. Technol. Lett. 11, 1497–1499 (1999). [CrossRef]
3. K. Habara, H. Sanjo, H. Nishizawa, Y. Yamada, S. Hino, I. Ogawa, and Y. Suzuki, “Large-capacity photonic packet switch prototype using wavelength routing techniques,” IEICE Trans. on Commun. E83-B, 2304–2311 (2000).
4. S. B. Yoo “Optical-packet switching and optical-label switching technologies for the next generation optical Internet,” in Tech. Dig. Optical Fiber Communications Conf.2, no. FS5, pp. 797–798 (2003).
5. D. Klonidis, R. Nejabati, C. Politi, M. O’Mahony, and D. Simeonidou, “Demonstration of a fully functional and controlled asynchronous optical packet switch at 40Gb/s,” in Proc. 30th European Conference and Exhibition on Optical Communication no. PDP-Th4.5.5 (2004).
6. J. McGeehan, S. Kumar, and A. Willner, “Optical time-to-live decrementing and subsequent dropping of an optical packet,” in Tech. Dig. Optical Fiber Communications Conf. 2, no. FS6, pp. 798–801 (2003).
7. J. Herrera, O. Raz, E. Tangdiongga, Y. Liu, H. C. H. Mulvad, F. Ramos, J. Marti, G. Maxwell, A. Poustie, M. T. Hill, H. de Waardt, G. D. Khoe, A. M. J. Koonen, and H. J. S. Dorren, “160-Gb/s All-Optical Packet Switching Over a 110-km Field Installed Optical Fiber Link,” J. Lightwave Technol. 26, 176–182 (2008). [CrossRef]
8. K Kitayama and N. Wada, “Photonic IP routing,” IEEE Photon. Technol. Lett. 11, 1689–1691 (1999). [CrossRef]
9. N. Wada and H. Harai, “Photonic packet routing based on multi-wavelength label switch using multi-section fiber Bragg gratings,” Proc. SPIE 4872, 185–198 (2002). [CrossRef]
10. N. Wada, H. Harai, and F. Kubota, “40 Gbit/s interface, optical code based photonic packet switch prototype,” in Tech. Dig. Optical Fiber Communications Conf.2, no. FS7, pp.801–802 (2003).
11. H. Harai and M. Murata, “High-Speed Buffer Management for 40Gb/s-based Photonic Packet Switches,” IEEE/ACM Trans. Networking 14, 191–203 (2006). [CrossRef]
12. H. Furukawa, H. Harai, N. Wada, T. Miyazaki, N. Takezawa, and K. Nashimoto, “A 31-FDL Buffer Based on Trees of 1x8 PLZT Optical Switches,” in Proc. in Proc. 32nd European Conference and Exhibition on Optical Communication no. Tu4.6.5 (2006).
13. N. Wada, H. Furukawa, K. Fujinuma, T. Wada, and T. Miyazaki, “160Gbit/s/port optical packet switch prototype with variable-length packet BER and loss real-time measurement,” in Proc. 31st European Conference and Exhibition on Optical Communication3, no. We1.4.1 (2005).
14. H. Harai and N. Wada, “More than 10 Gbps photonic packet-switched networks using WDM-based packet compression,” in Proc. 8th OptoElectronics and Communications Conference (2003), pp. 703–704.
15. H. Furukawa, N. Wada, H. Harai, M. Naruse, H. Otsuki, M. Katsumoto, T. Miyazaki, K. Ikezawa, A. Toyama, N. Itou, H. Shimizu, H. Fujinuma, and H. Iiduka, “Novel Layer-3 IP Packet Switching between 10 Gbps Ethernet and 80 Gbps Optical Packet-Switched Networks,” in Proc. 2007 Workshop on High Performance Switching and Routing (2007), pp. 1–6.
16. H. Furukawa, N. Wada, H. Harai, M. Naruse, H. Otsuki, M. Katsumoto, T. Miyazaki, K. Ikezawa, A. Toyama, N. Itou, H. Shimizu, H. Fujinuma, H. Iiduka, G. Cincotti, and K. Kitayama, “All-optical multiple-label-processing based optical packet switch prototype and novel 10Gb Ethernet / 80 (8λ × 10) Gbps-wide colored optical packet converter with 8-channel array burst-mode packet transceiver,” in Tech. Dig. Optical Fiber Communications Conf. (2007), no. OWC5.
17. H. Furukawa, N. Wada, H. Harai, Yoshinari Awaji, M. Naruse, H. Otsuki, T. Miyazaki, K. Ikezawa, A. Toyama, N. Itou, H. Shimizu, H. Fujinuma, H. Iiduka, E. Kong, P. Chan, R. Man, G. Cincotti, and K. Kitayama, “Field trial of IP over 160 Gbit/s colored-optical-packet switching network with transient-response-suppressed EDFA and 320 Gbit/s through put optical packet switch demonstrator,” in Tech. Dig. Optical Fiber Communications Conf. (2007), no. PDP4.
18. G. Cincotti, N. Wada, and K. Kitayama, “Characterization of a Full Encoder/Decoder in the AWG Configuration for Code-Based Photonic Routers—Part I: Modeling and Design,” J. Lightwave Technol. 24, 103–112 (2006). [CrossRef]
19. Y. Awaji, H. Furukawa, and N. Wada, “Adaptive Regeneration of Optical Packet Suffering from Gain Transient of EDFA by Using NOLM Discriminator,” in Proc. Conf. on Lasers and Electro-Optics (2008), no. JWA73.
20. Y. Oikawa, Y. Horiuchi, Y. Tanaka, M. Shiga, and Nagaeda H. , “Super-Fast AGC-EDFA for the Burst-Mode Systems without Gain Excursion in 20-ns and 21-dB Ramped Input,” in Proc. Conf. on Lasers and Electro-Optics (2008), no. JThA15.
21. Y. Awaji, H. Furukawa, N. Wada, P. Chan, and R. Man, “Mitigation of Transient Response of Erbium-Doped Fiber Amplifier for Traffic of High Speed Optical Packets,” in Proc. Conf. on Lasers and Electro-Optics (2007), no. JTuA133.
22. Advanced Testbed Network for R&D, “ JGN2,” http://www.j gn.nict. go.jp/.
23. N. Wada, K. Fujinuma, T. Wada, H. Iiduka, and T. Kubota, “Pure Packet BER and Loss Real-time Measurement with Optical Label Switching and Preamble Free Optical Packet 3R,” in Proc. 30th European Conference and Exhibition on Optical Communication (2004), no. PDP-Th4.5.4.
