Open Access
Add to BibSonomyAbstract
We report a novel architecture that can be used to construct optical switch fabrics with very high port count and nanoseconds switching speed. It is well known that optical switch fabrics with very fast switching time and high port count are challenging to realize. Currently, one of the most promising solutions is based on a combination of wavelength-tunable lasers and the arrayed waveguide grating router (AWGR). To scale up the number of ports in such switches, a direct method is to use AWGRs with a high channel count. However, such AWGRs introduce very large crosstalk noise due to the close wavelength channel spacing. In this paper, we propose an architecture for realizing a high-port count optical switch fabric using a combination of low-port count AWGRs, optical ON-OFF gates and WDM couplers. Using this new methodology, we constructed a proof-of-concept experiment to demonstrate the feasibility of a 256×256 optical switch fabric. To our knowledge, this port count is the highest ever reported for switch fabrics of this type.
©2009 Optical Society of America
1. Introduction
The capacity of electronic routers is expected to grow strongly due to the Internet and the popularity of video-based applications [1]. Unfortunately, as router capacity increases, so does heat dissipation and power consumption. Designers of next generation routers need to develop new routers with higher capacity and lower power consumption to meet the end users’ expectations.
The electronic switch fabric is a critical component in modern routers, and its primary function is to perform space or wavelength switching. To facilitate the transmission of packets from one router to another, electronic data has to be converted to optical signals. But once they reach another router, these optical signals are converted back to electronic form again for header processing and switching, thus consuming more power in the process. Aggravating the situation further, high capacity routers usually occupy multiple racks and additional stages of optoelectronic conversions are required because optical fibers are often used for inter-rack communications [1].
As line rates increase beyond 100Gbit/s, optoelectronic conversions will become more challenging. To alleviate this problem, optical switch fabrics with rapid switching speeds have been proposed [1–3]. By using optical switch fabrics, the number of optoelectronic conversions in the router can be reduced. However it is unlikely that all-optical routers can become a commercial reality in the near future because of several challenges such as the lack of practical optical buffers [4]. Nevertheless, it is advantageous to keep the signal in optical form when it is physically switched within the router itself.
Optical switch fabrics with high port count are attractive as they can meet the requirements of high-capacity routers without requiring the use of multi-stage architecture [5]. Optical MEMs technology is an attractive technology for realizing such large switch fabrics, but its major drawback is the slow switching time [6]. Conversely, optical switch fabrics that operate in the nanoseconds regime are difficult to realize and the challenges become more severe if high port count is also desired.
The wavelength-assisted routing technique, comprising an AWGR and a high-speed wavelength-tunable laser, is currently the most promising solution for realizing optical switch fabrics with nanoseconds switching time [2, 7]. However, the research focus is currently on realizing faster switching time and increasing the bit-rate per wavelength channel [2]; so far, there has been little attempt to realize switches with higher port count using the wavelength-assisted routing technique.
To increase the port count of such switch fabrics, one of the most direct methods is to increase the number of channels of the AWGR [8]. Unfortunately, that leads to several problems such as more severe interchannel crosstalk due to a smaller wavelength channel spacing [9].
In this paper, we propose a novel optical switch architecture for realizing large port-count high-speed optical switch fabrics using a set of low port-count AWGRs. A 2MN×2MN switch fabric can be realized by using M number of N×N AWGRs. This architecture allows one to circumvent the design, manufacturing and crosstalk noise problems associated with the use of high port-count AWGRs.
2. The proposed architecture
Figure 1 shows a generic form of the proposed architecture with 2MN input and 2MN output ports. The switch fabric comprises M number of N×N AWGRs that are interconnected by optical splitters and combiners, high-speed optical ON-OFF gates, and WDM couplers. To aid the subsequent discussions, the architecture is divided into four stages.
Fig. 1. A generic architecture illustrating how an 2MN×2MN optical switch fabric can be constructed using M number of N×N AWGRs.
The first stage, Stage A, is where the input signals λxi are broadcasted to different AWGRs using a broadcast-and-select scheme. We define
where x, y are the port numbers at the input and output ports of the switch, respectively, and i, j are the wavelength channels at input and output ports, respectively.
Each input port of the switch accommodates one wavelength channel as depicted in Fig. 1. Optical splitters are used to replicate the signal M times, where M is total number of AWGRs in the architecture. Next, by controlling the switching states of the M optical ON-OFF gates, an input signal can be routed to the any of the M AWGRs. The semiconductor optical amplifier (SOA) is a good candidate for ON-OFF gate because it possesses nanoseconds switching speed, very high extinction ratio, and optical reamplification function [10, 11]. At this point, we do not consider broadcast operation and therefore at most one of the M ON-OFF gates will be turned ON at any one time.
At Stage B, optical power combiners are then used to interconnect the gates to the AWGR input ports. Each combiner has 2M input ports and 1 output port, and the following mapping functions determine how the switch inputs are connected to the AWGRs. Let Sm,n, and tm,n be the n-th input port and the n-th output port of the m-th AWGR, respectively. We define:
where m is the position of the AWGR in the switch (Fig. 1) and n is the input or output channel number of the AWGR.
To map the signals at the input ports of the switch to Sm,n, the n-th input port of the m-th AWGR, we use the following functions:
for each input port of the AWGR from n=0,..,N-1.
At Stage C, the optical signals get routed to different AWGR output ports tm,n, based on their wavelengths. Finally at Stage D, each output port of the AWGR is connected to a dual-window WDM coupler that is used to separate wavelengths belonging to different wavelength bands. The bandwidth of the two passbands of the WDM coupler should be the same as the free spectral range (FSR) of the AWGR. Commercially available WDM couplers usually separate the input signal into the C and L bands but custom-designed WDM couplers are also available. Without loss of generality, we assign the wavelength channels from the shorter wavelength band of the WDM couplers to the even numbered output ports of the switch and wavelength channels from the longer wavelength band to the odd numbered output ports. Figure 2 illustrates how wavelength channels from the WDM coupler are routed to different ports of the switch.
Fig. 2. A WDM coupler distributes the wavelength channels from the AWGR to the odd and even output ports of the switch fabric.
By using the cyclic property of the AWGR, the proposed architecture allows two different wavelength channels from the same AWGR output port to enter a WDM coupler simultaneously. The role of the WDM coupler is to separate the signals belonging to different FSRs of the AWGR to different output ports of the switch. Without the WDM coupler, there will a port contention because a normal photo receiver can only detect one signal at any time. Therefore, the WDM coupler has an important function of doubling the output port count of the switch.
3. Design considerations for practical implementation
A 16×16 switch fabric depicted in Fig. 3 is used to illustrate how the architecture described in Section 2 can be implemented.
From Fig. 3, it is evident that M=2, N=4, and the total number of input or output ports is given by 2MN or 16. The example also shows that by using just two 4×4 AWGRs, the total port count can be quadrupled. Undoubtedly M, the number of AWGRs, plays an important role in the architecture. If M is large, the insertion losses associated with the splitters and combiners in Stages A and B will be high. Nevertheless, even if M is kept low at say M=2, the input/output port count will double and it is a significant advantage gained for a tradeoff in increased insertion losses in the power splitters and combiners.
If M is small, the splitter and the ON-OFF gates can be replaced by high-speed 1×M optical switches [12]. The main advantage of using such planar lightwave circuit (PLC) switches is that they are usually transparent to the bit-rate of the signals but the disadvantages include small ON-OFF extinction ratios and smaller power budget in the switch fabric because unlike the SOA, there is no optical re-amplification. On the other hand, SOA-based ON-OFF gates offer excellent ON-OFF extinction ratio of more than 50dB and the optical signals are amplified at Stage A to compensate for any insertion losses in the later stages. With the advent of quantum dot-based SOAs [13, 14], the bit-rate scalability of such gates is promising as well.
4. Experimental setup
We conducted a proof-of-concept experiment to examine the feasibility of implementing a 256×256 switch fabric (M=4; N=32) constructed using the proposed switch architecture. As shown in Fig. 4, a 10Gbit/s pseudo-random bit sequence (PRBS) NRZ-OOK signal from a pulse pattern generator is first split into four parts by a 1:4 electronic splitter (SHF 10410) where the individual bit streams are delayed by different amounts of time. Then the signals are recombined using a 4:1 electronic multiplexer (SHF 404) and the resulting 40Gbit/s signal is then amplified (SHF 806E). A manually controlled tunable laser (TL) is used to characterize the static switching performance of the proposed fabric. However, in a practical system, this should be replaced with a wavelength-tunable laser with fast tuning speed.
The light from TL is modulated with the 40Gbit/s signal using a Mach-Zehnder LiNbO3 intensity modulator (MZM) and split into 4 branches. After passing through an SOA ON-OFF gate, the signal in each branch is amplified to 10dBm and then launched into corresponding input ports of four individual AWGRs. An optical attenuator (ATT) with a 9.6dB insertion loss is used to simulate the power loss in the 8:1 optical power combiner. The AWGR has a channel spacing of 100GHz, FSR of 3200GHz, and average insertion loss of 6dB. At each output port of AWGR, a C/L band WDM coupler is used to separate the wavelengths located in different FSRs of the AWGR. A 40Gbit/s O/E converter followed by a 1:4 electronic demux and a bit error rate tester is used to measure the BER performance of the signal before and after the switch fabric. The eye diagram is detected by a digital communication analyzer.
5. Experiment results and discussions
The optical spectra of the signal before and after it passes through the switch fabric are shown in Fig. 5(a). Here, the central wavelength of the laser is 1533.88nm and the difference between the input and output spectra can be attributed to the AWGR and SOA. To study the effect of the optical switch fabric on the system performance, the BER of input and output signals are measured (Fig. 5(b)). The receiver sensitivity at BER of 10-9 is -11.2dBm and - 10.05dBm for input signal and output signal, respectively. Thus a 1.15dB power penalty is introduced by this switch fabric. Switching to any output port has a similar BER performance and therefore only two of the output ports were investigated (Fig. 5(b)). The insets are the eye diagrams of the 40Gbit/s signal measured before and after the fabric.
As shown in Fig. 5(a), the optical bandwidth of the signal is larger than channel spacing of the AWGR. Since the adjacent channel isolation of the AWGR is measured to be -25dB, this introduces some inter-channel crosstalk if multiple signals in the AWGR are switched simultaneously. To characterize the effect of this crosstalk, two more optical signals occupying the wavelength channels to the left and right of 1533.88nm (the original wavelength channel) were added. The signals were also modulated using the same MZM and all three signals enter the switch fabric via the same input port. The BER performance with inter-channel crosstalk present is shown in Fig. 5(b) and it is evident that an extra 0.2dB power penalty is now present.
Fig. 5. (a). Optical spectra. (b) BER performance; ◊: input signal (back-to-back measurement without switch fabric), ▫: output signal from AWGR0, +: output signal from AWGR3, ×: signal from AWGR0 with inter-channel crosstalk.
Fig. 6. Spectra of the switched signal measured at 256 output ports. The input port is fixed and the wavelength of the input signal is varied. All output channels have similar characteristics, but due to space constraints only selected channels are shown.
To demonstrate the feasibility of the proposed fiber fabric to support 256 ports, the wavelength of the TL is varied to switch the input signal to different output ports while the input port number remains fixed. Using different combinations of wavelengths and SOA gates, the experimental results show that a signal at any given input port can be switched to any of the 256 output ports. The optical spectrum measured at each output port is shown in Fig. 6. In this experiment, due to the lack of a large number of high speed optical modulators, only a pair of input and output ports can be measured at any instance. However, because of the symmetry of the switch fabric design, the measured results in Fig. 6 should be similar for the other input ports. As a result, we believe this architecture can be used to realize a 256×256 switch fabric.
In the proposed architecture, a coarse WDM coupler is used to separate the wavelengths into different FSRs. However, commercially available C/L band WDM couplers cannot separate wavelengths channels located at edge of the filtering profile without introducing high crosstalk noise. In other words, there is a small wavelength range in-between the two passbands that cannot be used, and in our experiment this range is about 8nm. To avoid a loss in the number of ports because of the unusable wavelength channels, it is possible to remap some of these wavelength channels in the unusable range to another FSR. As shown in Fig. 7, 10 wavelength channels cannot be used as they fall in-between two passbands of the WDM-coupler. Instead of dropping these channels altogether, they are remapped into the next FSR and the total number of ports in the switch fabric will not be affected. However, now the wavelength tuning range of the tunable laser has to be increased.
The maximum port count of the proposed architecture depends primarily on the insertion loss (IL) of the switch fabric as well as the receiver sensitivity. The relationship between the number of AWGRs (M), the insertion loss of the fabric, and the power margin is shown in Table 1. Here the input power to the switch is 0dBm and it is assumed that the SOA gain and output saturation power are 20dB and 10dBm, respectively. These assumptions are based on actual measurements obtained from the experiment setup in Section 4. Considering a 40Gbit/s receiver with sensitivity of -9.85dBm such as the one shown in Fig. 5(b), the power margins are 6.25dB, 3.25dB and 0.25dB when M is 2, 4 and 8, respectively. Thus the maximum port count for the proposed scheme could be as high as (2×8×32)×(2×8×32) or 512×512. However, to ensure sufficient power budget, the insertion loss of each component should be further reduced, and this can be achieved by selecting better components and eliminating the connector losses through splicing the interconnecting fibers.
Table 1. Relationship between fabric insertion loss and the power margin (N=32)
6. Conclusions
We proposed an architecture for constructing high-speed optical switch fabrics with large port count. The architecture uses a set of mapping functions to interconnect a set of AWGRs, WDM couplers and optical ON-OFF gates as well as power splitters and combiners in a specific manner to achieve a highly scalable switch fabric design. This allows the switch designer to overcome the scalability challenges posed by conventional wavelength-assisted routing schemes that rely on only a single large port-count AWGR. By using M units of N×N AWGRs, the architecture allows a 2MN×2MN switch fabric to be realized. We conducted an experiment to determine the feasibility of implementing a 256×256 optical switch fabric using the proposed solution and experimental results demonstrated that 40GBit/s intensity modulated signals can be switched from any input port to any output port at the expense of a 1.35dB power penalty due to the switch fabric. Thus we believe that the proposed architecture can be potentially used to design very large port-count optical switch fabrics for future terabit/s routers.
Acknowledgement
This work is supported by Singapore’s Agency for Science, Technology, and Research (A*STAR) under SERC project Grant No. 0721010019.
References and links
1. D. T. Neilson, “Photonics for switching and routing,” IEEE J. Sel. Top. Quantum Electron. 12, 4, 669–678 (2006). [CrossRef]
2. J. Gripp, P. J. Winzer, G. Raybon, J.E. Simsarian, and C.R. Doerr, “107-Gb/s serial optical packet switching with 1-bit/s/Hz spectral efficiency for 100-GbE backplanes,” IEEE Photon. Technol. Lett. 19, 15, 1124–1126 (2007). [CrossRef]
3. Y. K. Yeo, J. Yu, and G. K. Chang, “A broadcast and multicast-enabled switch architecture utilizing a gateless channel selection scheme,” in Proceedings of Optical Fiber Commun. Conf. (OFC) 2006, OTuG7 (2006).
4. G. K. Chang, J. Yu, Y. K. Yeo, A. Chowdhury, and Z. Jia, “Enabling technologies for next-generation optical packet-switching networks,” Proc. of the IEEE 94, 5, 892–910 (2006). [CrossRef]
5. Jan Cheyns et al., “Clos lives on in optical packet switching,” IEEE Commun. Mag. , 114–121 (2004). [CrossRef]
6. C. H. Huang, H. F. Chou, J. E. Bowers, F. Toudeh-Fallah, and R. Gyurek, “Dynamically reconfigurable optical packet switch,” Opt. Express 14, 5, 12008–12014 (2006). [CrossRef] [PubMed]
7. H. Takahasi, K. Oda, H. Toba, and Y. Inoue, “Transmission characteristics of arrayed waveguide N×N wavelength multiplexer,” IEEE J. Lightwave Technol. 13, 3, 447–455 (1995). [CrossRef]
8. M. Ishii et al., “Low-loss fibre-pigtailed 256 channel arrayed waveguide grating multiplexer using cascaded laterally-tapered waveguide,” Electron. Lett. 37, 23, 1401–1402 (2001). [CrossRef]
9. P. Monnard, C. R. Doerr, C. Dragone, M. Cappuzzo, E. Laskowski, and A. Paunescu, “Large N×waveguide grating routers,”IEEE J. Lightwave Technol. 11, 1287–1295 (1993).
10. N. Sahri, D. Prieto, S. Silvestre, D. Keller, and et al., “A highly integrated 32-SOA gates optoelectronic module suitable for IP multi-terabit optical packet routers,” in Proceedings of Optical Fiber Commun. Conf. (OFC) 2001, PD32-1-3 (2001).
11. A. Ehrhardt et al., “Semiconductor laser amplifier as optical switching gate,” IEEE J. Lightwave Technol. 11, 1287–1295 (1993). [CrossRef]
12. http://www.eospace.com/Switches.htm.
13. D. Bimberg, “Quantum dot based nanophotonics and nanoelectronics,” Electron. Lett. 44, 3, 168–171 (2008). [CrossRef]
14. S. Dommers et al., ‘Complete ground state gain recovery after ultrashort double pulses in quantum dot based semiconductor optical amplifier,” Appl. Phys. Lett. 90, 033508 (2007). [CrossRef]
