Open Access
Add to BibSonomyAbstract
We demonstrate optical burst add-drop multiplexing as a practical application of the optical burst switching technology in a wavelength-division-multiplexed ring network. To control optical bursts in the network, a burst identifier (BI) for delivering control information, and a BI processor for handling the BI, were designed. Optical bursts of 10- to 100-μs in length were optically multiplexed or demultiplexed in an intermediate node of the ring network. The demonstration shows that the optical burst add-drop multiplexing technique provides sub-wavelength granularity to a ring network.
©2007 Optical Society of America
1. Introduction
Packet-based switching has been considered and used in telecommunication networks as an ideal solution to handle bursty data traffic, which has increased exponentially over recent years due to explosive increases in Internet users, Internet protocol-based services, and digital multimedia contents. The increasing traffic has created a need for high-speed, large-capacity networks. To accommodate the large-volume of data traffic, an optical circuit switching (OCS) network was developed. As a typical example of an OCS, synchronous optical network/synchronous digital hierarchy (SONET/SDH)-based ring networks have been widely deployed for metro-network application [1]. A SONET/SDH ring network provides good granularity. However, it requires optical-electrical-optical (O-E-O) conversion and data processing at each node and shows limited scalability because the amount of through-traffic increases as the number of switching nodes increases. Recently, the SONET/SDH ring network has adopted a wavelength division multiplexing (WDM) technology to improve network capacity. The WDM SONET/SDH ring network connects nodes using a static wavelength, providing improved scalability. However, its granularity is too coarse since it cannot add or drop a packet at the sub-wavelength level. It also requires a number of transponders at each node, and such as has been considered an expensive solution. In addition, the OCS system is fundamentally inefficient for bursty traffic because of its high connection-setup overhead [2]. Naturally, an optical packet switching (OPS) technology has been attractive as an eventual solution enabling improvement of network efficiency under a bursty packet traffic environment [3]. OPS supports ultra-fast switching and label switching and can packetize bursty traffic, making it favorable for network utilization. However, implementation of a substantial OPS network has been considered as a very difficult task because key technologies such as a practical optical buffer, high-speed optical signal processing, and a fast optical switch are still under study [4,5].
Therefore, as a practical alternative for a direct migration from the prevailing OCS network to OPS network, an optical burst switching (OBS) technology was proposed and has been studied intensively [6–8]. The OBS technology can facilitate a reservation and allocation of resources of a switching node by transmission of an optical header in as much time prior to a payload as the predetermined offset time. In particular, it enables construction of an optical switching network with no need for optical buffers. The OBS has been studied for application mainly in wide area networks with mesh topologies. Recently, however, metro optical ring application of the OBS technology has been attempted as a migration from the prevailing SONET/SDH ring networks. With a proper media access control (MAC) protocol, OBS metro-ring networks can be more scalable and bandwidth-efficient as well as cost-effective by reducing costly O-E-O conversions and sub-wavelength multiplexing. Previous works have suggested several OBS ring architectures that utilize different MAC protocols [9,10] and have compared them with other optical ring architectures [11]. Quality-of-service features in an OBS ring have also been evaluated with logical optical burst add-drop multiplexing (OB-ADM) structures [10]. Howerver, most of these studies have been limited to theoretical or numerical analyses, and no physical implementation of a WDM-based OB-ADM ring network has been reported.
In this study, we demonstrate OB-ADM based on WDM technology as a practical application of the OBS-related technologies that have been previously studied. We physically construct an OB-ADM testbed with three nodes. For the construction, we design and embody BI, BI processor, and burst-mode optical transmitter and receiver. We also introduce a sub-microsecond electro-optic switch and a gain transient-controlled optical amplifier. We then demonstrate OB-ADM operation and transmission of optical bursts (OBs) at a 9.953 Gb/s rate controlled by the BI with a just-enough-time (JET) protocol [7]. And finally, we verify that the implemented OB-ADM techniques can provide sub-wavelength granularity to a ring network.
2. Testbed implementation
For the sake of OB-ADM, switches in an intermediate node of the ring network have to be controlled using proper control methods. Thus, we designed the BI as a burst control packet and implemented the BI processor to deal with the BIs. To achieve polarization-independent burstness of sub-microsecond speed, we introduced fast electro-optic switches with low polarization dependency. In a transmission experiment, an optical surge and gain transient was observed, so we introduced a couple of additional methods for suppressing the transient. The following list shows the key units and technologies for the testbed implementation.
- BI and BI processor
- Burst-mode optical transmitter and receiver
- Burst-mode optical amplification
- Electro-optic switch
2.1 BI and BI processor
Fig. 1. BI-based burst data transmission: (a) A structure of the BI. (b) BI transmission in continuous mode (lower) and a BI location indicator (upper). The BI indicator helps to find a meaningful BI among dummy BIs. The inset shows a fine view of the waveform of a 14-byte BI packet. (c) Measured electrical signals for BI indicators (top), laser diode-control signal (middle), and optical signal of an output OB channel (bottom). The BI indicator is used to measure the offset time between a BI and an OB.
To control the OB, a BI processor based on a field programmable gate array (FPGA, Mercury EP1M120: Altera) was embodied, which generates and processes a BI 14-byte in length at a 311.04 Mb/s rate. The BI is a control packet known as a burst control packet in the OBS technology. It has information of the OB, such as its source and destination addresses, size, offset time, and so on. It is used to reserve system resources and to add or drop the OB. In the BI processor, a number of dummy BIs were multiplexed with meaningful BIs by using a time-division multiplexing method to sustain optical power of the BI channel at a constant level. The constant power of the BI channel was used to suppress a gain transient induced by an optical amplifier. In addition, continuous-mode operation of the BI channel contributes to stable transmission of the BI itself in comparison with burst-mode transmission of the BI.
Figure 1(a) shows the structure of the BI. The BI consists of a preamble (32 bits), destination node address (8 bits), source node address (8 bits), channel identification (8 bits), offset time (12 bits), burst length (12 bits), and cyclic redundancy check (32 bits). Examples of its major parameters are presented in Table 1. Figure 1(b) shows BI indicators, a continuous-mode BI channel, and a 14-byte BI packet. The BI indicators were additionally generated by the BI processor to designate the exact position of a BI in the mixed stream. The stream of multiplexed meaningful and dummy BIs was generated by the BI processor, and it fed into a driving circuit that was designed and fabricated based on a burst-mode driver chipset (MAX3656: Maxim). The driver directly modulated a distributed feedback laser diode (DFB-LD, LC25EW: Bookham). Then, an optical BI channel with constant power was obtained. The inset of Fig. 1(b) magnifies a part of the BI channel stream and clearly shows a meaningful 14-byte BI packet. Figure 1(c) shows a 20-μs offset time between the generation of a BI and an OB, and the BI indicator proves BI generation.
Table 1. Selected Parameters in the BI
2.2 Burst-mode optical transmitter and receiver
The burst-mode optical transmitter (BM Tx) is made up of a laser, a driver, and an external modulator. To generate the envelope of an OB a few tens of micro-seconds in length, the DFB lasers were turned on and off in nano-second response time. A fast on- and off-time of the BM Tx is useful because it can reduce the loss of bits in the head and tail parts of an OB. The burst-mode driver chipset used in the BI driving circuit was exploited again here. The output from the DFB laser was modulated one more time at a 9.953 Gb/s rate by using an optical amplitude modulator. In the testbed, one OB channel was modulated, while for simplicity, the others had OB envelopes only without any data. In spite of its simplicity, this method had no severe effect on the final results in the transmission experiment, so we adopted it.
The burst-mode optical receiver (BM Rx) was made up of an avalanche photo diode (AT10GC: Bookham), a limiting amplifier (LXT13002: Intel), and a clock/data recovery (CDR) device (MTC5585: Multilink). The clock signal was recovered within 17 ns by using a direct filtering method embedded in the CDR. This fast recovery time enables us to adopt a short preamble length in the OB and to increase network efficiency.
2.3 Burst-mode optical amplification
Despite the powerful performance of Erbium-doped fiber amplifiers (EDFAs) in long haul transmission and WDM applications, they show some defects in burst-mode environments. When the power of an optical signal increases rapidly at an input of an EDFA, e.g., a rising edge of an OB, an optical surge could be generated because of the slow gain dynamics of the EDFA. In particular, the optical surge accumulates as the number of the EDFAs increases [12]. In our preliminary experiments, the optical surge increased up to 28 dB after a successive chain of four EDFAs (OAB1550: JDSU) and optical attenuators. In this case, the optical attenuators played a role of power loss produced by optical fiber links. In the experiments, the burst interval was 500 μs and burst duration was 100 μs, while the number and power of the OB channel were set to be four and 3 dBm, respectively. As verified by the pretests, the optical surge or power transient must be controlled to use the EDFA as an optical amplifier in an OBS network. Therefore, in the testbed, we utilized two techniques simultaneously to effectively suppress the unwanted transient effects in the amplifier chain.
One was an all-optical gain clamping (AGC) technique [13]. In general, the power of an amplified spontaneous emission (ASE) noise of an EDFA decreases as the input power increases so that the EDFA can generate an inversely proportional ASE signal with respect to the input signal. In the AGC method, a small portion of the reversed ASE noise of the first EDFA is fed back via an optical band-pass filter (OPBF). Then, the ASE signal is used to compensate the low power of idle intervals among the OBs. Thus, the fed-back ASE noise incorporated with the input OBs can supply a constant power input to the EDFA. Eventually, the optical gain of the EDFA can be sustained without any transient. Findings from a preliminary measurement of the power excursion show that the center wavelength of ASE lasing should be close to the wavelength of an OB in order to suppress the optical surge effectively, as shown in Fig. 2(a). In particular, for a power excursion of less than 2dB, the wavelength of an ASE signal should be within the range of - 5 nm to 10 nm from the wavelength of the OB channel. In the testbed implementation, the ASE signal was allocated at a wavelength of 2.0-nm less than the first OB channel and was fed back through an OBPF with a 3-dB bandwidth of 0.25 nm.
Fig. 2. Reduction of the transient and surge by (a) close allocation of an ASE feedback channel to an OB channel and (b) an increase of BI channel power.
The other technique used a constant power of the BI channel. When a signal with constant power is incorporated with an OB channel with fluctuating power, the signal plays a role of a direct current or baseband signal. Then, the EDFA is fed by a less fluctuating optical signal. Eventually, the optical surge in the OB can be suppressed. Figure 2(b) shows that higher optical power of a BI channel can further suppress the surge and transient of an OB. Our findings show that, to suppress a power excursion of less than 2 dB, output power of the BI channel should be adjusted to be higher than -3 dBm. In the testbed implementation, the BI channel power was set to be -2 dBm. To sustain the power of the BI channel constantly, as described in section 2.1, the BI channel operated in continuous mode, that is, dummy packets were added into the BI channel when the BI was not triggered.
2.4 Electro-optic switch
To drop the OB in the intermediate node of the ring network, 1×2 electro-optic (EO) switches (FX-UF: CIVCOM) were used. Polarization dependence of the EO switch was less than 0.2 dB. The rise- and fall-time of the EO switch were measured to be 267 ns and 180 ns, respectively. But the response times of the EO switch for a control signal were delayed by a driving circuit and measured up to 700 ns and 460 ns, respectively. Thus, a guard time of at least 700 ns was required to switch an OB safely. In our testbed, more than a 1-μs guard time was given before and behind the OB.
2.5. Testbed construction
Fig. 3. A schematic diagram and photo of the OB-ADM testbed. Where, MOD: external intensity modulator, DCF: dispersion compensating fiber (DCF), SMF: single-mode fiber, Mux: optical multiplexer, Demux: optical demultiplexer, SW: electro-optic switch, OB-tTx: tunable transmitter, and BER tester: bit-error-rate tester.
Figure 3 shows a schematic diagram of the constructed OB-ADM testbed. The testbed has a three-node structure; node A, B, and C. Node B represents an ADM node. Node A was only equipped with a BI processor and a Tx part, and node C was only equipped with a Rx part and a BI processor. The reason for this was that we had concluded that this configuration is enough to show full functionality of the physical layer of an OB-ADM ring network.
In node A, four DFB lasers operating at 1549.41 nm, 1551.01 nm, 1552.61 nm, and 1554.22 nm (λ 1, λ 2, λ 3, and λ 4) were used for OB generation. Among them, the output of the LD1 was modulated by a 231-1 pseudo random signal at a 9.953 Gb/s rate using an external intensity modulator. In the BI channel, a DFB laser operating at 1556.63 nm was used. The BI processor generated an electrical signal multiplexed with BI and dummy packets at a 311.04 Mb/s speed for modulation of the BI channel. An arrayed-waveguide grating multiplexer combined the modulated BI channel with the OB channels. And a fiber link consisting of a dispersion compensating fiber (DCF), a 50-km single-mode fiber (SMF), and two EDFAs delivered the combined optical signals. The gain of the first EDFA was maintained to be constant by using the AGC method and the technique of continuous operation of the BI channel. In node B, the BI processor extracted data from the received BI packets and sent control signals to the corresponding EO switches. Then, the optical switches dropped the OBs according to information from the BI processor. The BI processor generated new BIs for the new OBs and sent gating signals to the OB tunable transmitter (OB-tTx), which added one OB to a determined output channel. The OB-tTx was made up of four LDs, a 4×1 coupler, and an optical intensity modulator. The channel tuning time of the OB-tTx was about 40 ns. The second link delivered the output of node B to node C. In node C, an optical demultiplexer separated the BI and OB channels, and the BI processor extracted a gating signal from the BI for burst-mode bit-error-rate (BER) estimation.
So far, the key units and technologies for testbed implementation have been introduced. Testbed construction was also explained. Table 2 summarizes the design specifications of the constructed testbed.
Table 2. Design Specifications of the OB-ADM Testbed
3. OB-ADM operation and transmission experiment
In the case of bursty packets, if a network can supports fine granularity, the network could be attracted because small-size random packets are mainly generated in present telecommunication networks. The WDM SONET/SDH-based ring network switches the packets according to wavelength, requiring a costly O-E-O conversion to extract a specific packet multiplexed in a long data stream. Eventually, without O-E-O conversion, wavelength-level granularity is only available. However, in our testbed, OBs are directly extracted from or inserted into a wavelength channel by using the OB-ADM function. Therefore, sub-wavelength granularity can be achieved naturally.
For the purpose of OB-ADM operation, a BI was formed before an offset time of 20 μs when an OB was generated using a combination of pulse pattern generator and burst-mode LD driver. The combination emulates the output from a burst assembler. The length of the OB ranged from 10 to 100 μs, corresponding to tens of kilo-bytes to hundreds of kilo-bytes for a 10 Gb/s speed packet. The OB length was chosen to achieve high network utilization for 10 Gb/s Internet protocol traffic [14]. The BI was then sent to the next node. After the offset time, the corresponding OB was transmitted at an available wavelength. According to information in the BI, the BI processor in the ADM node determined whether the incoming OB was dropped or allowed to pass through the node. The BI processor had an OB scheduling function and communicated with the other BI processors of adjacent nodes. Thus, it can determine when and where a new OB must be added into the ring network. When a new OB was generated from the ADM node, the BI processor determined its available wavelength channel and time window for OB add-multiplex. A new BI with a predetermined offset time was also created for the new OB.
The OB-ADM testbed adopted a simple delayed reservation scheme based on the JET protocol [7] to implement the scheduling function. The BI processor determines when and where a newly generated OB is inserted based on the information in all incoming BIs from neighboring nodes prior to the OBs. When it finds an expected void in the OB channels, it reserves the section and add-multiplexes a new OB into the OB-ADM ring network according to the time schedule. However, because of the simplified scheme implemented in the MAC layer, the OB could be fragmented or might not be add-multiplexed when the length of the new OB is longer than the void. In our testbed operation, this did not happen due to the well-defined scenario. Studies on blocking probability and void filling are needed and will be carried out hereafter.
Fig. 4. Implemented OB-ADM functions. (a) Input optical OBs, (b) dropped OBs, (c) added OBs, (d) outputs from the ADM node.
Fig. 5. Spectra for (a) AGC operation at the booster amplifier of node A, (b) the received BI, (c) dropped λ1 channel, and (d) added λ1 channel in node B.
Figure 4 shows timing diagrams representing the OB-ADM operation implemented in the testbed. For the first channel, λ 1, all OBs were dropped and new OBs were added. For the third channel, λ 3, some OBs were dropped and others passed through the ADM node. All OBs in channel λ 2 and λ 4 passed through the ADM node. Figure 5 shows an optical spectrum for an output of the booster amplifier of node A and optical spectra for the received BI, dropped λ 1 channel, and added λ 1 channel in the ADM node. During the operation of the testbed, we observed all OB-ADM functions to be exactly controlled by the control signal generated by the BI processor on the basis of information from the BI. Though the testbed operated four OB channels, the number of channels can be increased. However, in this case, a technique for dynamic allocation of the offset time must be considered first to avoid collision between BIs. The dynamic allocation technique is also required to increase the number of the nodes.
To evaluate the transmission performance, the BER of the λ 1 channel was estimated by using a BER tester (MP1764A: Anritsu). BER curves and eye diagrams for the λ 1 channel from node A to B, from node B to C, and from node A to C are shown in Fig. 6 (a) ~ (c), and represent drop, add, and pass functions of the OB-ADM network, respectively. For BER estimation, a signal converter was fabricated, which generates gating signals controlling the BER tester and other peripheral equipments. The converter provided the tester with a continuous clock signal that was synthesized by a radio-frequency switch using two types of clock signals from the CDR and the pulse pattern generator. The clock signal was used for safe synchronization of the tester with the pulse pattern generator. As shown in Fig. 6, measured burst-mode penalties at a BER of 10-9 of the network were less than 1.1 dB for each span when the burst-mode BER results were compared with those of continuous-mode transmission cases. The penalty is mainly attributed to burst-mode operation of the optical receiver. In the transmission experiment, the back-to-back BER results had little difference with those of one-span transmission for all cases due to full dispersion compensation, so those are not presented in Fig. 6. The findings show that transmission of the OBs over a greater increased number of spans is possible. However, if the node number approaches or exceeds ten, adoption of this OB-ADM network has to be considered carefully. This is because the dynamic range of the optical receiver used in the testbed is limited to 11 dB. Also, the induced power penalty cannot be guaranteed to increase linearly as the number of spans increases. Thus, when the size of a ring network is smaller than ten-spans, the proposed scheme can be considered.
Fig. 6. BER curves and eye diagrams for (a) A to B node, (b) B to C node, and (c) A to C node transmission in continuous mode and burst mode.
In the transmission experiment, the OBs delivered a random bit sequence. But, for a more practical examination, a detailed structure of the OB must be designed and implemented as in the BI in this testbed. A fast burst assembler is also required to generate the OBs in real time. In the case of the BI, the FPGA chip was able to carry this out successfully. However, the speed of the FPGA chip was lower than 1 GHz. Thus, a more powerful processor such as a network processor is needed to deal with the OB at a 10 Gb/s rate or more. This requires the introduction of a newly designed device. A high-speed optical interface must also be considered to transmit the assembled OBs. This need could be met by exploiting techniques studied in the development of 10 Gb Ethernet transceivers.
The testbed in this paper presented physical layer functions only. If it is linked with control and management layers based on a generalized multi-protocol label switching protocol and a general switch management protocol, respectively, then it will be able to support more varied and high-level functions of an OBS network [15,16].
4. Summary and conclusions
We demonstrated the OB-ADM functions in a ring network using a testbed. In the testbed, the optical bursts were controlled by the BI and the OBs with a duration of several tens of microseconds were transmitted with acceptable penalties of less than 1.1 dB per hop by burst-mode transmission, receipt, and amplification functions. In the experimental findings from the OB-ADM operation and link transmission, the proposed OB-ADM ring network showed the possibilities of providing sub-wavelength granularity and of being a replacement of a legacy WDM ring network based on SONET/SDH technologies.
Acknowledgments
This work was supported by the Korea Science and Engineering Foundation (KOSEF) grant funded by the Korean government (MOST) (No. R11-2000-074-03002-0)
References and links
1. P. Tomsu and C. Schmutzer, Next Generation Optical Networks, (Prentice-Hall, NJ, 2002).
2. T. Battestilli and H. Perros, “An Introduction to Optical Burst Switching,” IEEE Communications Mag. 41, S10–S15 (2003). [CrossRef]
3. D. K. Hunter and I. Andonovic, “Approaches to Optical Internet Packet Switching,” IEEE Communications Mag. 38, 116–122 (2000). [CrossRef]
4. J. Bowers, E. Burmeister, and D. Blumenthal, “Optical Buffering and Switching for Optical Packet Switching,” in Proceedings of International Conference on Photonics in Switching, (Institute of Electrical and Electronics Engineers, New York, 2006), pp. 203–205.
5. Y. Liu, E. Tangdiongga, Z. Li, S. Zhang, M. Hill, J. van Zantvoort, F. Huijskens, H. de Waardt, M. Smit, A. Koonen, G. Khoe, and H. Dorren, “Ultrafast all-optical signal processing: toward optical packet switching,” Proc. SPIE 6353, 635312-1–635312-12 (2006).
6. Y. Chen and J. S. Turner, “WDM burst switching for petabit capacity routers,” in Proceedings of IEEE Military Communications Conference (Institute of Electrical and Electronics Engineers, New York, 1999), pp. 968–973.
7. M. Yoo and C. Qiao, “A novel switching paradigm for buffer-less WDM networks,” in Proceedings of IEEE Optical Fiber Communication Conference / International Conference on Integrated Optics and Optical Fiber Communication (Institute of Electrical and Electronics Engineers, New York, 1999), pp. 177–179.
8. C. Qiao, “Labeled Optical Burst Switching for IP-over-WDM Integration,” IEEE Communications Mag. 38, 104–114 (2000). [CrossRef]
9. L Xu, H. G. Perros, and G. N. Rouskas, “A simulation study of optical burst switching and access protocols for WDM ring networks,” Computer Networks 41, 143–160 (2003). [CrossRef]
10. A. Fumagalli and P. Krishnamoorthy, “A Low-Latency and Bandwidth-Efficient Distributed Optical Burst Switching Architecture for Metro Ring,” in Conference Records of IEEE International Conference on Communications (Institute of Electrical and Electronics Engineers, New York, 2003), pp. 1340–1344.
11. A. Zapata, I. de Miguel, M. Duser, J. Spencer, P. Bayvel, D. Breuer, N. Hanik, and A. Gladisch, “Performance Comparison of Static and Dynamic Optical Metro Ring Network Architectures,” in Proceedings of European Conference on Optical Communication (Institution of Electrical Engineers, London, 2003), pp. 374–375.
12. M. Zirngibl, “Gain control in erbium-doped fiber amplifiers by and all-optical feedback loop,” IEE Electron. Lett. 27, 560–561 (1991). [CrossRef]
13. M. Karásek, A. Bononi, L. A. Rusch, and M. Menif, “Gain Stabilization in Gain Clamped EDFA Cascades Fed by WDM Burst-Mode Packet Traffic,” J. Lightwave Technol. 18, 308–313 (2000). [CrossRef]
14. B. Feng, N. Stol, E. Zouganeli, R. Ø. Andreassen, A. Solem, R. B. Haugen, A. Sudbø, and B. E. Helvik, “Direct comparison between optical circuit and optical packet/burst switching using simulations,” in Proceedings of Optoelectronics and Communications Conference / Conference on Optical Internet (Institute of Electronics, Information and Communication Engineers, Japan, 2004), pp. 394–395.
15. H. Jeong, J. Cho, M. Kang, H. Park, J. Choi, and Y. Won, “GMPLS over OBS Architecture for an Optical Packet Internet,” in Proceedings of Conference on Optical Internet - Photonics in Switching (Korean Institute of Communication Sciences, Seoul, 2002), pp. 84–86.
16. J. K. Choi and Y. W. Cha, “Network management in the GSMP open interface,” in Proceedings of Conference on Optical Internet - Photonics in Switching (Korean Institute of Communication Sciences, Seoul, 2002), pp. 203–205.
