Open Access
Add to BibSonomyAbstract
We propose a cost-effective and scalable OXC/ROADM that consists of a subsystem-modular express switch part and a transponder-bank-based add/drop part. The effectiveness of the proposed architecture is verified via a hardware scale evaluation, network performance simulations, and transmission experiments. The architecture enables large throughput and offers significant hardware-scale reductions with marginal fiber-utilization penalty against the conventional architectures. A part of the OXC/ROADM designed to accommodate 35x35 express fiber ports and 2,800 transponders for add/drop is constructed. Its net throughput reaches 314 Tbps using 80 channels of 120-Gbps signal (30-Gbaud dual-polarization quadrature phase-shift-keying signals with 7% overhead are assumed).
© 2015 Optical Society of America
1. Introduction
Optical cross-connects/Reconfigurable optical add-drop multiplexers (OXC/ROADM, which are used interchangeably in this paper) must cost-effectively implement the express switch and add/drop of optical paths simultaneously. To enhance network flexibility, future OXCs must have colorless, directionless, and contentionless (C/D/C) capability in both the express switch part and the add/drop part [1]. The C/D/C functionalities mean that an OXC can express/add/drop signals without any restriction. The number of fibers between nodes and that of transponders in nodes need to be expanded to meet the ever-increasing Internet traffic. As a result, OXCs must have scalability to accommodate large numbers of express fiber ports and add/drop ports in a cost-effective way.
The present C/D/C express switch part of an OXC mostly uses wavelength-selective switches (WSSs). The port count of the practical WSSs is, however, limited to 20 + and hence the express switch part fiber port count is limited to around 20. Although a large-port-count WSS can be created by cascading multiple small-port-count WSSs, the number of costly WSSs can explode. Moreover, we cannot expand the scale of the present broadcast-and-select (splitters and WSSs) or route-and-combine (WSSs and splitters) architectures due to the increase in splitter loss; therefore, the route-and-select (WSSs and WSSs) architecture needs be employed, which doubles the number of needed WSSs. To reduce the hardware cost of the express switch part, we recently have proposed a subsystem-modular optical cross-connect architecture in which multiple small OXCs, called ‘subsystems,’ are interconnected with a limited number of fibers [2]. Since the use of small OXCs can avoid the costly route-and-select architecture and WSS concatenation, the number of necessary WSSs and that of the traversed WSSs can be greatly reduced. This also mitigates signal impairments such as insertion loss and filter narrowing induced in nodes [3–6]. This architecture can, however, cause contention at the interconnection fibers between subsystems, however, its performance was proven to be almost the same as that of the contentionless express architecture when the restriction-aware routing and wavelength assignment (RWA) algorithm is employed [2, 5].
The C/D/C add/drop part can be realized with matrix switches or a combination of splitter-switch and tunable filters [7–9]. The hardware scale of the matrix switches expands with the square of the number of dropped channels. As a result, the matrix-switch-based add/drop configuration is not applicable to future large-scale OXCs. In contrast, for the second architecture, the numbers of splitter ports and tunable filters, and the switch degree for incoming-fiber selection, increase linearly. Allowable splitter loss is still restricted to maintain signal quality, and hence this architecture necessitates multi-stage splitters and a large number of erbium-doped fiber amplifiers (EDFAs) behind or before the splitter to compensate the splitter loss in the add/drop part. Furthermore, the switch scale necessary to select any target incoming/outgoing fiber can be very large given the rapid increase in incoming/outgoing fiber numbers. To overcome these issues and to attain scalability, we have proposed the transponder-bank based add/drop architecture, which emphasizes the subsystem-modular express part. The transponders are grouped into ‘transponder banks’ and each transponder bank, consisting of a fixed number of transponders, is connected to a limited number of incoming/outgoing fibers. Thanks to this architecture, the number of EDFAs and the switch scale necessary in the add/drop part can be substantially reduced because the splitter scale and the number of selectable outgoing/incoming fibers are reduced. Although this arrangement does not attain the directionless capability because of limits placed on the incoming/outgoing fibers that can be selected, the performance has been shown to be almost the same as that of the full-C/D/C add/drop configuration [10].
In this paper, we comprehensively evaluate the OXC that adopts the subsystem-modular express switch part and the transponder-bank-based add/drop part, simultaneously. First, we assess the required hardware cost in terms of the numbers of WSSs in the express part, and the number of EDFAs and the scale of switches in the add/drop part. We demonstrated that for a 35x35 OXC with 100% drop ratio, the proposal achieves a 68% reduction in WSS number in the express switch part, and a 77% reduction in EDFA number and switch scale in the add/drop part. Second, we calculate the fiber-utilization efficiency under a dynamic network control scenario and show that the proposed OXC architecture can attain almost the same performance as that obtained by the C/D/C architecture. Third, we configure a part of the OXC that can accommodate 35x35 express fibers and an add/drop part with 2,800-transponders. Our newly developed wavelength-tunable filters [11] are applied to the drop part. Utilizing the constructed OXC system, we measure the transmission performance of 80-channel 30-GBaud dual-polarization quadrature-phase-shift-keying (DP-QPSK) signals. The results verify the good performance of the proposed OXC. We assumed 7% overhead for forward-error correction (FEC), and the net throughput of the fully implemented OXC reaches 314 Tbps, which is the largest capacity to the best of our knowledge.
The organization of this paper is as follows. Chapter 2 describes our proposed ROADM architecture and assesses its hardware cost. Chapter 3 evaluates the network performance through intensive computer simulations. In Chapter 4, we confirm the effectiveness of our OXC architecture through transmission experiments. Finally, we conclude this paper in Chapter 5.
2. OXC architecture
2.1 Full C/D/C OXC architecture
Figure 1 depicts a full C/D/C OXC based on the route-and-select express switch part and the splitter-switch-filter based add/drop part. Note that the add part is almost the same as the drop part though tunable filters are unnecessary, and so detailed explanations of the add part are omitted hereinafter. The OXC with M incoming/outgoing fibers consists of 2M 1x(M + 1) WSSs, where M ports are used for the express switch part and one port is assigned for the drop part. Hence, M is the total number of drop ports. It is notable that multiple WSS concatenation is necessary when M is large; such a configuration increases node cost and induces signal degradation due to insertion loss and filter narrowing.
First, the incoming signals are input to a WSS and selectively routed to express ports and drop ports. The express signals are cross-connected in a C/D/C manner with this route-and-select architecture provided each wavelength is not already in use on the target output port. As for the drop signals, multiple drop channels are distributed by an optical-power splitter and then amplified by EDFAs. Note that such a split-and-amplification process should be divided into multiple stages so as to minimize signal degradation when the number of drop channels (splitter ports) is large. Next, a desired incoming fiber is selected by a switch and a target drop channel is finally extracted by a tunable filter. In this way, each receiver can access an arbitrary channel in an arbitrary fiber. This drop part can be configured using MxQ multicast switches (MCSs) as illustrated in the right side of Fig. 2.
To construct the large-scale express switch part, this scheme concatenates multiple WSSs when the port count of an available WSS is smaller than the number of incoming/outgoing fibers, and uses the route-and-select architecture to avoid the unacceptable splitter loss, which results in a significant increase in the number of necessary WSSs. In the drop part, large-degree splitters are needed to distribute signals to all the transponders, which demands a large number of EDFAs be implemented to compensate the huge splitter loss. Furthermore, switch scale becomes large with the increase in the number of incoming/outgoing fibers. Accordingly, this OXC configuration lacks any significant scalability. As mentioned before, we did not consider using space switches for the add/drop part here since the hardware scale becomes prohibitively large.
2.2 OXC based on subsystem-modular express and transponder-bank add/drop
Figure 3 shows our proposed OXC architecture that consists of multiple OXC subsystems [2] and multiple transponder-banks in the add/drop part. With this architecture, a large-scale MxM OXC is created by interconnecting multiple small-scale mxm OXCs, i.e., ‘subsystems’. The use of small-OXC subsystems allows us to utilize the route-and-combine architecture instead of the costly route-and-select architecture without any WSS concatenation. As a result, the number of necessary WSSs can be reduced by more than 50% [2,5]. In the transponder-bank based add/drop part, all R transponders are divided into r-transponder groups called ‘transponder banks’ where d add/drop fibers are connected to each transponder bank, that is, each transponder can be connected with d (out of M) outgoing/incoming fibers through the splitter-switch stage. This scheme reduces the splitter degree and hence the splitter loss by bounding the number of connectable receivers per bank. As a result, we can reduce the number of split-and-amplification stages and the number of EDFAs. In addition, switch scale in the splitter-switch part is reduced from Mx1 (see Fig. 1) to dx1 (see Fig. 3). The transponder-bank based add/drop part necessitates a relatively large number of WSS ports connected to the add/drop part. The subsystem-modular architecture can offer a certain number of ports even if commercially available WSSs are used. For example, if m = 9 and we use 1x20 WSSs, 11 WSS ports can be used as drop ports for the incoming fiber. The combination of the subsystem-modular express part and the transponder-bank add/drop part yields a very effective configuration. The transponder-bank configuration can also be applied to the full C/D/C express part shown in Fig. 1. However, the combination necessitates WSS cascading to create WSS ports connected to the add/drop part, which increases system complexity as noted in Section 2.1. The proposed OXC architecture introduces a routing restriction both in the express switch part and in the add/drop part. In return, significant reductions in the numbers of WSSs and EDFAs are obtained. The performance deterioration stemming from the restriction can be minimized by applying the routing-and-wavelength assignment strategies that are aware of the restriction [2, 5, 6], which will be detailed in Section 3. Note that our scheme is applicable to flexible-grid networks. The frequency-slot resolution of the system is determined by those of WSSs and tunable filters.
2.3 Assessment of hardware requirement
Table 1 summarizes the numbers of key components necessary for the express switch part and drop part of the proposed subsystem-modular OXC. The designed OXC accommodates 35 or 42 incoming/outgoing fibers with a 100% drop ratio. Each subsystem has the scale of 9, the number of wavelengths per fiber is 80, the WSS used has scale of 1x20, and EDFA gain is 20 dB. Note that for the drop part, the number shows the value for each channel (transponder), except for the total splitter loss value. Each transponder bank accommodates 128 transponders.
Table 1. Hardware elements necessary for ROADM architecture.
The reductions in the necessary number of 1x20 WSSs, EDFAs, and switch degree are 68%, 77%, and 77%, respectively. The proposed OXC thus substantially reduces the necessary hardware scale compared to the single large scale OXC that adopts the route-and-select configuration and full-C/D/C capabilities. The tunable filter can be removed if coherent detection is used, but when the channel number is large, a tunable filter is necessary to ensure that the optical power level of the desired channel is under the level that triggers OSNR deterioration. Note that the necessary hardware in the add part is almost the same as that in the drop part, but the tunable filters can be excluded.
3. Simulations
In order to evaluate the fiber-utilization efficiency attained with the proposed OXC, we analyzed the traffic possible under the following conditions. The network topology tested is the 26-node Pan-European network (COST266) as shown in Fig. 4. Each fiber accommodates up to 80 wavelengths. The traffic demand is uniformly and randomly distributed. The generation of connection requests follows a Poisson process and the holding time for each connection a negative exponential distribution. Each connection request is associated with a randomly selected source-destination transponder pair, and traffic intensity is represented as the average number of wavelength paths between each node pair, which equals the average path-arrival rate multiplied by the average holding time [12]. The subsystem size is 9x9 and each subsystem consists of nine 1x20 WSSs and nine 20x1 splitters, where 7, 11, and 2 output/input ports of each WSS/splitter are assigned for the express switch part, drop/add parts, and subsystem interconnection, respectively. Each transponder bank has 128 transponders and is connected with OXC subsystems through 8 add/drop fibers in a round-robin fashion. Note that in each transponder bank, transponders can be added one by one as necessary, not all transponders need be placed at the outset.
Figure 5 shows accepted traffic demands normalized by those of the full C/D/C configuration (no restriction at both express switch part and add/drop part) where the numbers of incoming/outgoing fibers are changed. The performance of the proposed OXC almost matches that of the full C/D/C configuration even though it greatly reduces total OXC hardware scale. The accommodated traffic demand offset is just 2% even when the number of incoming/outgoing fibers is 42. Moreover, in the drop part, the proposed scheme requires a switch scale (8x1) that is determined by the number of input/output fibers to/from a transponder bank, whereas the required switch scale in the full C/D/C scenario increases with OXC incoming/outgoing fiber numbers. In Fig. 5, the switch scale is evaluated by the number of element 1x2 switches, which means that a 1xn switch requires n-1 component 1x2 switches. It is verified that our proposed OXC reduces the hardware scale (the network performance offset is marginal), and enables highly-scalable modular growth both in the express switch part (adding subsystems) and in the add/drop part (adding transponder banks).
4. Experiments
To confirm the transmission performance of the proposed OXC, we conducted experiments using the setup depicted in Fig. 6. We emulated a part of the OXC with 35x35 input/outgoing fibers and 22 transponder banks, each of which accommodates 128 transponders; up to 100% drop is possible. At the transmitter side, external-cavity lasers (ECLs) produced 80 wavelengths with 50-GHz spacing. Then, an IQ modulator (IQM) created 30-GBaud QPSK signals. The QPSK signals were polarization multiplexed with a polarization-division-multiplexing (PDM) emulator in a split-delay-combine manner, resulting in 120 Gbps per channel. The odd and even wavelength channels were decorrelated with a WSS, a 1-m fiber delay, and coupled with a 2x1 splitter. The optical signal-to-noise ratio (OSNR) of the 80 wavelength multiplexed signals was controlled with an amplified-spontaneous-emission (ASE) noise source and a variable optical attenuator (VOA). The signals thus obtained were launched into the proposed OXC. In the OXC, the signal traversed a 1x20 WSS in the express switch part, and a 1x4 splitter, an EDFA with 20-dB gain, and a 1x32 splitter in this order in the drop part. This arrangement was designed so as to minimize the number of necessary EDFAs while maintaining needed signal quality. The signals were input to an 8x1 switch and delivered to a coherent receiver through a tunable filter. The outputs from the coherent receiver passed through 23-GHz analog low-pass filters and were sampled at 50 Gsample/s by a four-channel analog-to-digital converter with eight-bit resolution. The digitized data were sent to an offline digital signal processing (DSP) circuit which performed polarization recovery, carrier recovery, adaptive equalization, and symbol decoding [13].
For this setup, we utilized newly developed optical tunable filters based on arrayed waveguide gratings (AWGs) [11] so as to maximize the target-channel power input to the coherent receiver. Each tunable filter consists of a 1x7 switch, a 7x14 AWG, and a 14x1 switch as shown in Fig. 7. Wavelength extraction was performed as follows: up to 96 channels were input to the 1x7 switch and sent to one of the input ports of the 7x14 AWG. The 7x14 AWG divides the 96 channels according to wavelength and the input port used. The 14x1 switch selects one of the output ports of the AWG and the target channel is finally extracted. As shown in Fig. 6, two tunable filters were monolithically implemented on a 37x84.4-mm2 planar-lightwave-circuit chip. Figures 8(a), 8(b), 8(c), and 8(d) show, for the first time, measured values of 3-dB bandwidth, insertion loss, adjacent crosstalk, and polarization-dependent loss (PDL), of the developed filter, respectively. The minimum 3-dB bandwidth was 34 GHz, hence a 30-Gbaud DP-QPSK signal can pass through the tunable filter. The worst insertion loss, adjacent-channel crosstalk, and PDL were 8.4 dB, −22.3 dB, and 0.43 dB, respectively. These results indicate that the fabricated tunable filter hardly impairs the signal quality.
Fig. 8 Measured transmission characteristics of the tunable filter, (a) 3-dB bandwidth, (b) insertion loss, (c) adjacent-channel crosstalk, and (d) PDL.
Figure 9 plots bit-error ratios (BERs) measured as a function of OSNR. We achieved BERs below the 7% FEC threshold (BER = 4.4x10−3) at OSNR = 15.5 dB and those below the 20% FEC threshold (BER = 1.1x10−2) at OSNR = 14 dB. Thus, the transmission experiment confirms the good performance of our proposed OXC. If 7% FEC overhead is assumed, the total OXC throughput reaches 314 Tbps, (i.e., 120 Gbps/wavelength, 80 wavelengths/fiber, 35 fibers, 1/1.07 FEC rate). To the best of our knowledge, the proposed OXC attains the highest capacity yet reported.
5. Conclusion
We comprehensively investigated a novel OXC architecture that consists of a subsystem-modular express switch part and a transponder-bank add/drop part. The proposed architecture achieves significant reductions in the necessary number of WSSs, EDFAs, and switch degree. Numerical evaluations proved that the proposed OXC can offer virtually the same fiber-utilization efficiency as the conventional full-C/D/C OXC. We developed a part of the OXC to support 35x35 express fibers and 2,800 transponders for add/drop, and confirmed its good performance. The net throughput of 314 Tbps was successfully demonstrated through transmission experiments using 80-channel 120-Gbps DP-QPSK signals with 7% FEC overhead.
Acknowledgment
This work was partly supported by NICT λ-reach project and KAKENHI (26220905). We are grateful to Dr. M. Okuno and Mr. Y. Jinnouchi of NTT Electronics Co., for their useful discussions.
References and links
1. V. Shukla, D. Brown, C. J. Hunt, T. Mueller, and E. Varma, “Next generation optical network - enabling dynamic bandwidth services,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2007), paper NWB3.
2. Y. Iwai, H. Hasegawa, and K. Sato, “A large-scale photonic node architecture that utilizes interconnected OXC subsystems,” Opt. Express 21(1), 478–487 (2013). [CrossRef] [PubMed]
3. R. Younce, J. Larikova, and Y. Wang, “Engineering 400G for colorless-directionless-contentionless architecture in metro/regional networks,” J. Opt. Commun. Netw. 5(10), A267–A273 (2013). [CrossRef]
4. M. Filer and S. Tibuleac, “N-degree ROADM architecture comparison: broadcast-and-select versus route-and-select in 120 Gb/s DP-QPSK transmission systems,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2014), paper Th1I.2. [CrossRef]
5. H. Hasegawa, Y. Tanaka, K.-I. Sato, and Y. Iwai, “Subsystem modular OXC architecture that achieves disruption free port count expansion,” in European Conference and Exhibition on Optical Communication, Technical Digest (CD), paper Th.2.E.4, 2013. [CrossRef]
6. Y. Tanaka, H. Hasegawa, and K. Sato, “Performance evaluations of large-scale OXC that achieves modular and hitless expansion,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2014), paper W2A.50. [CrossRef]
7. R. A. Jensen, “Colourless, directionless, contentionless ROADM architecture using low-loss optical matrix switches,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2011), paper OThR3.
8. T. Watanabe, K. Suzuki, T. Goh, K. Hattori, A. Mori, T. Takahashi, T. Sakamoto, K. Morita, S. Sohma, and S. Kamei, “Compact PLC-based transponder aggregator for colorless and directionless ROADM,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2011), paper OTuD3. [CrossRef]
9. W. I. Way, “Optimum architecture for M×N multicast switch-based colorless, directionless, contentionless, and flexible-grid ROADM,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2012), paper NW3F.5.
10. H. Ishida, H. Hasegawa, and K. Sato, “Highly scalable subsystem modular OXC nodes that host tailored add/drop mechanism,” presented at Optical Fiber Communication Conference, paper Th2A.47, 22–26 Mar. 2015. [CrossRef]
11. S. Takashina, Y. Mori, H. Hasegawa, K. Sato, and T. Watanabe, “Low-loss and low-power-consumption wavelength tunable filter enabling colorless/directionless/contentionless optical drop in ROADMs,” in European Conference and Exhibition on Optical Communication, IEEE Technical Digest (CD) (IEEE, 2014), paper We.3.5.2. [CrossRef]
12. Y. Tanaka, H. Hasegawa, and K. Sato, “Performance analysis of large-scale OXC that enables dynamic modular growth,” Opt. Express 23(5), 5994–6006 (2015). [CrossRef] [PubMed]
13. S. J. Savory, “Digital filters for coherent optical receivers,” Opt. Express 16(2), 804–817 (2008). [CrossRef] [PubMed]
