1. Introduction

Recent developments in elastic optical path networks are substantially enhancing the capacity of optical fibers by employing fine frequency slot granularity bandwidth assignment [1,2]. Deployment of higher bitrate channels including 400Gbps is on the horizon and the target of 1Tbps has been set [3–5]. Further spectral efficiency improvement can be achieved by adopting distance-adaptive modulation or Nyquist WDM [6–11]. However, the traffic volume is still continuously increasing and the rate is so high that it exceeds that of fiber capacity enhancement. This trend leads to the inevitable increase in the number of fibers laid on each link, and as a result, the number of fibers connected to a ROADM/OXC (Rearrangeable Optical Add-Drop Multiplexers/Optical Cross-Connect) node will be large; i.e. its fiber degree will be higher. Current ROADMs/OXCs consist of WSSs (Wavelength Selective Switches) and optical couplers bridged in the broadcast-and-select manner. Increasing ROADM/OXC degree is not cost-effective, because the necessary degree should not exceed that of the WSSs although the degree (of commercial units) is limited to around 20 or so. Higher degree WSSs seem to be impossible to produce cost-effectively because the aerial beam manipulation needed becomes excessive and the adjustments necessary for all pairs of ports and wavelengths. Available ROADM/OXC degree is limited by that of the WSSs, currently around 30 + , and no technique that offers truly cost-effective expansion is on the horizon. Some studies on higher port count WSS development have been presented, for example [12] adopts a PLC (planar light wave circuit) based beam spot size converter in combination with spatial optics, however, the use of small port count WSSs remains essential in the development of cost-effective transport systems, since the reduced WSS port count relaxes the precision requirements and fiber array density in the spatial optics. Another point to be noted in using large port count WSSs is the explosion in interconnect fiber number between WSSs; the route and select 1x35 WSS configuration needs 35² = 1225 WSS interconnection fibers per system. Thus small port count WSSs remain the only truly practical solution and work is already underway on realizing large scale and cost-effective ROADM/OXCs.

The path hierarchy was introduced in conventional networks such as SDH/SONET (Synchronous Digital Hierarchy/Synchronous Optical NETwork) and OTN (Optical Transport Network). The higher order paths, coarser granular routing entities, can simplify the cross-connection. In optical networks, the introduction of optical path hierarchy has been studied [13–15], and enabling devices including waveband demultiplexers [16,17] and waveband selective switches [18] have been developed. All of them can be monolithically integrated on small PLC chips. However, the compactness of these devices depends on certain regular structures derived from the ITU-T fixed grid. In other words, we need more a flexible definition of coarse granular switching that offers path elasticity in combination with the development of a compact OXC architecture.

Based on this observation, we proposed a coarse granular routing scheme named flexible waveband routing [19]. The architecture is composed of flexgrid WSSs at input/output ports and optical delivery-coupling type switches (DCSWs) which bridge the WSSs. Flexgrid WSSs at input ports are used as demultiplexers for the grouping of any combination of incoming optical paths, named flexible wavebands, and the NxN DCSWs, which consist of 1xN switches and Nx1 optical couplers, are used for routing these wavebands to the desired output ports. Thus the WSSs at output ports work as multiplexers to combine the routed flexible wavebands. The maximum number of groups for each input defines the necessary WSS degree and the number of DCSWs. The target input/output port count of the proposed OXCs is twenty to thirty considering the continuous traffic increase of 30% a year. The number of adjacent nodes is up to eight to nine for real networks (ex. Pan-European topology), and if the fibers are fully utilized, the required port count of an OXC will reach this target size in 8-9 years (1.3⁹ = 10.6), as multiple fibers are laid on each link. It has been verified that the maximum number of groups can be bound to a small value, for example 4, at the cost of only a slight (3-5%) increase in fiber number in a network, or conversely if the fiber number is the same, the network using our proposed node can accommodate 3-5% less traffic. However, each port would need a 1x4 WSS while a conventional NxN OXC demands a 1xN WSS at each port. If we consider case of the current traffic growth rate (~30%/year) and year-by-year network resource (fiber) expansion, the above 3-5% traffic penalty quickens the expansion by just two months (i.e. ${1.3}^{2/12}~1.045$).

Although the scalability issue of WSSs is resolved with the introduction of flexible wavebands, DCSW degree is equal to or larger than OXC degree. Unfortunately, it is hard to manufacture such large scale DCSWs. Moreover, the optical couplers in DCSWs impose optical coupling loss, which increases with DCSW degree. Moreover, for monolithic implementation as optical integrated circuit chips, the increased number of waveguide crossings on the chip makes manufacture problematic. On the contrary, recent advancements in silicon photonics [20,21] will lead to the compact integration of small-scale switches. Thus, it will be critical to develop an OXC node architecture for flexible waveband routing networks that use small port count WSSs and switches. A tailor-made network design algorithm for the architecture will also be a key to making the proposal effective.

In this paper, we propose a compact OXC node architecture for flexible waveband routing networks that combines WSSs with small-scale DCSWs. Each large DCSW of the original flexible waveband routing OXC is divided into small-scale DCSWs arrayed in parallel to take advantage of the routing redundancy in the original architecture; for example, a waveband can be routed by any of the DCSWs. These divided switches are interconnected to WSSs at input/output ports of an OXC so the interconnections for each waveband are differentiated. This differentiated interconnection contributes to eliminating routing redundancy among all waveband indexes and maximizing routing flexibility. This leads to the suppression of performance degradation caused by switch splitting. A network design algorithm that is aware of the switch splitting and the differentiated interconnection is also proposed in this paper. Numerical and transmission experiments demonstrate the effectiveness of the proposed OXC architecture from several aspects. First, the optical loss of the proposed OXC node architecture is analyzed and the locations of EDFAs are specified, and then the amount of hardware needed for any given OXC degree is evaluated. Second, evaluations in a network demonstrate that the increment in the number of fibers over the conventional flexgrid network will generally be less than 5%, with appropriate WSS and DCSW degree. The number of WSSs necessary can be significantly reduced, while the routing performance degradation bounds to an acceptable level. Furthermore, we develop a prototype using monolithically integrated arrayed 3x3 DCSWs on a PLC chip. Transmission experiments on the prototype demonstrate its excellent transmission characteristic. Preliminary versions of this study were presented at international conferences [22, 23] where the first one [22] proposed an initial version of proposed compact OXC and the second one [23] shows a prototype development and a transmission experiment on the prototype. This study includes more detail on the architecture and the algorithm and more experimental results as well.

2. Flexible waveband routing optical networks

Throughout this paper, elastic optical path networks are assumed where center frequencies of optical paths lie on a 6.25 GHz spaced grid (193.1THz + 6.25i GHz (i: integer)) and path bandwidths are multiples of 12.5GHz. Since a path always occupies an even number of grid spacings, hereafter we assume that the center frequency offset value, i, of all paths are all even (or all odd) integers so as to avoid useless 6.25GHz gaps between adjacent paths. With this assumption, all paths are located on 12.5GHz spacing grid with bandwidth of 12.5xn GHz (n: positive integer). We call the 12.5GHz width unit bandwidth a frequency slot. The methodology discussed in this paper can be also applied to conventional fixed grid networks whose channel center frequencies are located on a fixed spacing grid (193.1THz + {12.5,25,50,100}i GHz (i: integer)) and path bandwidths are equally set to the selected grid spacing.

Figure 1 shows a WSS-based OXC node architecture for elastic optical networks. Flexgrid WSSs (for example LCOS (liquid crystal on silicon) based WSSs) enable optical paths to be switched with fine granularity ($\le$12.5GHz) [24]. Here for add/drop operations are realized by the dedicated assignment of 1x2 WSSs, however they can be implemented in the WSS in the OXC. The integration can be done in the same way for both proposed and conventional OXC nodes, and does not materially alter the hardware scale comparison given in this paper (required WSS port number increases 1 for each architecture).

 

Fig. 1 WSS based route and select OXC node architecture.

Download Full Size | PDF

In the figure, 1xn WSSs need to be cascaded so as to satisfy the port count requirement since the fiber degree N is higher than that of each WSS (i.e. $n^2\ge N>n$). Here we assume two-stage cascading which enables adequate scaling of OXC nodes; WSS cascading substantially increases the number of WSSs needed, and optical loss/impairment is enhanced by the traversal of multiple WSSs in each node. The impairment can reach unacceptable levels even if just two stage cascading is adopted, see details in Section 4.1 which considers the number of EDFAs necessary. In addition, the number of needed WSS explodes as evaluated also in Section 4.2, which raises system cost to an unacceptable level.

Figure 2 shows the original OXC node architecture for flexible waveband routing [19]. An NxN OXC based on the architecture consists of 1xB flexgrid WSSs at its input/output ports and NxN DCSWs bridging these WSSs. Optical paths from an incoming fiber are first divided into a limited number ($\le$B) of groups, and then these groups are distributed to up to B output ports of the WSS assigned to the corresponding input port as shown in Fig. 2. The input WSSs are used as a dynamic waveband demultiplexer, while WSSs are also responsible for path routing in the conventional OXC node architecture. At each output port of each switch, arriving flexible wavebands from all input ports are merged. This operation can be realized, for example, by DCSWs as shown in Fig. 2. The degree of this switching port is NxN and hence, we call it the NxN DCSW hereafter. Their integration on a PLC (Planar Lightwave Circuit) chip has already been demonstrated [25] and commercial units are now available. This switch can also be realized by using individual 1xN optical switches [26] and 1xN optical couplers. Finally, at each output port of the OXC, path groups from corresponding outputs of DCSWs are combined by the 1xB WSS assigned to the output. The WSS at the output side can be replaced by an optical coupler if B is sufficiently small (e.g., B = 4). Thanks to the dynamic and fine granular bandwidth control capability of WSSs, a flexible waveband can be created with any set of elastic optical paths and offers dynamic path addition and removal to/from that waveband. On the other hand, an additional constraint is triggered such that no pair of wavebands routed to an output fiber can overlap as shown in Fig. 3, which is named the waveband collision constraint.

 

Fig. 2 OXC node architecture for flexible waveband routing in [19].

Download Full Size | PDF

 

Fig. 3 Collision between flexible wavebands.

Download Full Size | PDF

The number of necessary 1xB WSSs to realize a conventional NxN (N > B) route and select type OXC node is 2N x ⌈(N − 1)/(B – 1)⌉ (B > 1), which is almost proportional to $N^2$, while an NxN proposed node only needs 2N in addition to B NxN DCSWs. Here ⌈x⌉ stands for the ceiling function that assigns the minimum integer not less than the real number x. Our previous works verified that B would be a small number (e.g. B = 4) and the use of such small degree WSSs would enhance overall cost effectiveness in [19]. Indeed, up to 50% reduction in WSS number was verified on the Pan-European network topology [27]. Moreover, reducing the number of WSSs traversed in a node reduces the optical signal filtering impairment caused by WSSs (Fig. 4).

 

Fig. 4 Relationship between fiber, flexible waveband, and elastic optical paths.

Download Full Size | PDF

Unfortunately, the scalability of the proposed OXC node architecture is still limited due to the difficulty in manufacturing high degree DCSWs. The unacceptably high optical loss due to high degree optical coupling is also a critical barrier. Therefore, the ideal way to realize large scale flexible waveband routing OXC nodes is replacing these DCSWs with small degree ones.

3. Switch decomposition in flexible waveband routing networks

The original flexible-waveband OXC node architecture explained in the previous section bundles elastic optical paths into several groups/wavebands. The number of wavebands is less than the maximum number of optical paths in a fiber, and hence, at the cost of some degradation in routing capability, a substantial reduction in WSS number can be achieved. A similar strategy for DCSW scale reduction is adopted in this paper; determine the routing capability that can be omitted with no severe penalty. Wavebands are routed independently by large scale DCSWs, which implies redundancy in routing. As a first example, suppose that paths using ${\lambda }_1$ and ${\lambda }_2$ are carried by wavebands #1 and #B, respectively, and both wavebands are routed to the same output port (Fig. 5(a)). This configuration is clearly inefficient since these wavelength paths can be bundled into the same waveband. That is, different wavebands need not to be routed to the same output port and we can omit such switching cases. As a second example, consider paths using ${\lambda }_1$ and ${\lambda }_2$ are also carried by wavebands #1 and #B, respectively, and these wavebands are routed to different output ports #1 and #2 (Fig. 5(b)). Exactly the same routing is possible by placing and into wavebands #B and #1, respectively, and these wavebands are routed to output ports #1 and #2 (Fig. 5(c)). These examples indicate that there is some redundancy in routing capability.

 

Fig. 5 Redundancy in path/waveband routing with OXC architecture in [18].

Download Full Size | PDF

Based on the above observations, we propose to use, for each waveband, small DCSWs arrayed in parallel instead of one large DCSW (Fig. 6). An interconnection example is shown in Fig. 6. The use of small switches should allow differentiation for all wavebands. This differentiated configuration bounds the routing performance degradation to an acceptable level while decreasing the routing redundancy, as elucidated later in the numerical experiments. There are various interconnection patterns between input and output port pairs of DCSWs. Once the degree of an OXC and the number of wavebands per fiber are determined, the interconnection design can be formulated as an optimization problem, for example, minimizing the disconnected input and output port pairs and so formulated as an ILP (Integer Linear Programming) problem. More complex optimization is possible in theory; for example, a tailor-made intra-node WSS-DCSW interconnection optimization approach that utilizes the switch splitting information of adjacent nodes and fiber interconnection with these nodes. Figure 7 shows blocking ratio variations for the proposed OXC where a network with original OXCs is designed first and then the DCSWs are split into 8x8/4x4 so that the fiber number increment is minimized. There is no fiber number increment if 8x8 DCSWs are adopted. However, such optimization of splitting DCSWs and interconnections between small DCSWs for each node is not practical. Another advantage given by switch splitting is modular growth capability, where a small port count node is expanded in response to traffic growth just by installing additional small DCSWs. For this sake, switch splitting and intra-node WSS-DCSW interconnection must not be specific to a given condition. Therefore, regular switch splitting and regular interconnections between switches will be necessary to make OXC manufacture easy and permit verification of routing performance in numerical experiments.

 

Fig. 6 Proposed OXC architecture with small DCSWs arrayed in parallel.

Download Full Size | PDF

 

Fig. 7 Blocking ratio variation for split switches with optimized interconnection to WSSs on a 5x5 mesh network.

Download Full Size | PDF

The following round-robin based selection of combination/grouping could be used for interconnection differentiation. Suppose that k small (m x m) switches are used to replace one large (km x km) switch (for one waveband). Let the input/output port indexes be {$i_1,i_2,\dots ,i_{km}$} and {$o_1,o_2,\dots ,o_{km}$}, respectively. Let the subsets of input and output ports be, respectively, $I_l=$ {$i_{(l-1)m+1},\dots ,i_{lm}$} and $O_l=${$o_{(l-1)m+1},\dots ,o_{lm}$} (l = 1,2,…,k). For DCSW#n (n = 1,2,…,k) in WB#b (b = 1,2,…,B), connect the input/output ports respectively to $I_n$ and $O_{(n+b-1)\mathrm{mod}k+1}$ where mod is the modulus operator. The hit to routing performance caused by switch decomposition can be compensated by increasing the number of waveband layers B, although the number of DCSWs necessary is proportional to B. In numerical experiments, we evaluate the trade-offs between routing performance and switch scale reduction.

We propose a design algorithm for our flexible waveband OXC node that is aware of DCSW decomposition. The proposed algorithm is divided into multiple stages. The first stage, which estimates OXC scale, is based on our network design algorithm for flexible waveband routing OXC nodes without switch decomposition [18]. The second stage implements switch splitting. Finally, the third stage implements path routing while being aware of flexible waveband path bundling and switch splitting. The algorithm and its flow chart are shown below and in Fig. 8.

 

Fig. 8 Flowchart of proposed design algorithm.

Download Full Size | PDF

<Design algorithm of Flexible waveband routing networks that utilize small scale DCSWs>

Step 0. Initialization

Sort all path establishment requests of given traffic demands in descending order of minimum hop counts or shortest distances between source and destination nodes. The order of path sets with the same metric value is determined randomly. Calculate a set of route candidates for each node pair by using the k-shortest path algorithm [28].

Step 1. OXC Node Size Estimation

Assume that all nodes consist of original flexible waveband OXC node as shown in Sec II; i.e. no scale restriction on DCSWs. Sequentially assign route and frequency slot sets to each path demand in the order determined in Step 0 to minimize the fiber number increment at each assignment. Count the number of fibers on each link and that connected to each node, and then remove all fibers.

Step 2. DCSW decomposition and Interconnection

Let the size of available DCSWs be m x m. Suppose that the degree of a node in Step 2 is M x N. Replace each of the B M x N DCSWs with ⌈max{M, N} / m⌉ m x m small DCSWs. Connect these m x m switches to WSSs at input/output sides by adopting the interconnection method described in above.

Step 3. Network Design That is Aware of Switch Decomposition

In the order specified in Step 0, accommodate paths one by one. For each path assignment, calculate the following cost function for all route candidates connecting the source and destination nodes of the path establishment request

where $r$ and $S$ are a sequence of fibers, i.e., route, and a contiguous slot set that accommodates the demand, $\alpha$ and $\beta$ are weights, $d(r)$ is hop count or distance of $r$, $WS{S}_{new}(r,S)$ is the number of WSSs whose output ports are newly reserved on the route, and $fibe{r}_{new}(r)$ is the number of fibers that need to be newly established. Establish a path on route $r$ and frequency slot set $S$ so that the cost function is minimized. Repeat this procedure until all path establishment requests are processed. If all paths are established, terminate.

Remark: In our preliminary study [22], a different interconnection between WSSs and DCSWs was used. The previous scheme is also based on round-robin arrangement, however port index interleaving (even/odd) was also applied. In this paper, we adopt a simpler interconnection scheme as interleaving provides negligible improvement in routing performance.

4. Numerical evaluations and transmission experiments

4.1 Hardware scale analysis for flexible WB routing OXCs

In this sub-section, the necessary number of key components in an NxN OXC is evaluated for hardware scale estimation. Suppose that both conventional and proposed OXCs have pre/post amplifiers (EDFAs) at all input/output ports and consideration of the add/drop part is omitted for simplicity. All WSSs are assumed to be 1x9/1x20. We assume that the WSS and the m x m DCSW have losses of 6.5dB and 10 ${\log }_{10}$(m) + 2 dB, respectively. Let the power budget be 20 + dB. Whenever a higher degree WSS is necessary, multiple 1x9/1x20 WSSs are cascaded so that the port count requirement is satisfied. For conventional OXCs, a path can traverse up to three WSSs without using EDFAs, and thus, cascading WSSs at input and output sides need additional EDFAs to compensate the power loss (see Fig. 9(a)). In this paper, EDFAs are located between all cascaded WSSs at the input side of the OXC. On the other hand, for flexible waveband routing OXCs, additional EDFAs are inserted just after WSSs at the input side (see Fig. 9(b)) if a DCSW has unacceptable loss. For example, 8x8 DCSWs need signal amplification as shown in Fig. 9(b) whereas 3x3 DCSWs do not. These small DCSWs can be also compactly integrated with the PLC technology.

 

Fig. 9 Additional EDFAs for conventional and proposed OXCs.

Download Full Size | PDF

Based on the above, we evaluate the number of WSSs, EDFAs necessary for NxN conventional OXCs and proposed OXCs with various combinations of WSS and DCSW. Figures 10 and 11 plot the necessary number of WSSs and EDFAs per OXC versus OXC size, respectively. For conventional OXCs, 1x9/1x20 WSSs need to be cascaded to create a large scale OXC when N > 9 or N > 20. The necessary number of WSSs is given by $\lceil (N-1)/(B-1)\rceil$ x 2N, where n is WSS degree. On the other hand, for the proposed flexible waveband routing OXCs, the necessary number of WSSs is always twice OXC size. Thus the proposal holds the increases in WSSs to linear on N and thus smaller than the conventional one as shown in Fig. 10. Moreover, based on the above observation, the necessary number of EDFAs is shown in Fig. 11. For conventional OXCs, the number of EDFAs is given by ($\lceil (N-1)/(B-1)\rceil$ - 1) x N + 2N, where the second term stands for the number of essential pre/post amplifiers. On the other hand, for the proposed OXCs that need additional EDFAs to compensate loss at DCSWs, the number of EDFAs is given by NB + 2N. Thus flexible waveband routing OXCs can reduce the number of EDFAs when the OXC size is large (e.g. N $\ge$ 42). If an additional EDFA is not necessary due to the use of small DCSWs, then the number of EDFAs will be 2N; only pre/post amplifiers are needed. Thus, the reduction is substantial, especially when small degree DCSWs are used (e.g. 3x3 DCSWs are used).

 

Fig. 10 Number of WSSs per OXC versus OXC node size.

Download Full Size | PDF

 

Fig. 11 Number of EDFAs per OXC versus OXC node size.

Download Full Size | PDF

4.2 Routing performance evaluations

In this subsection, we evaluate the routing performance of our proposed flexible waveband OXC node with split DCSWs in terms of the number of fibers and WSSs necessary.

Tested topologies were a 5x5 grid network, Japanese national network topology [29], Pan-European network topology [27] and Telecom Italia network [30] (Fig. 12). The characteristics of these topologies are also shown in Table 1. For example, in the Pan-European network, the highest node degree is 8, and the fiber degree N increases to twenty or thirty as multiple fibers are laid on each link. The explosive increase in WSS number in the conventional OXC is evaluated. Available frequency range is the C-band, i.e. 4,400GHz bandwidth, and 352 frequency slots whose bandwidth is 12.5 GHz are available. Three channel capacities and modulation formats are set to 40Gbps DP-QPSK/100Gbps DP-QPSK/400Gbps DP-16QAM and these channels require 3/4/7 frequency slots, respectively. Transparent optical transmission reach for different channel capacities without considering the filter narrowing effect are discussed in [31] using the values of 3500km and 4000km for 40Gbps and 100Gbps DP-QPSK, respectively. The traffic demand was generated as path connection requests whose source and destination nodes are randomly selected following a uniform distribution. Traffic intensity in a network is denoted by the average number of path establishment requests between each node pair. The capacity of each path is also determined randomly where the request probabilities for the three capacities are the same (i.e., 1/3 for each). When calculating route candidates, detours from the shortest hop routes, up to 2 hops, are allowed for all topologies. We adopt the conventional OXC nodes created by cascading 1x9 or 1x20 WSSs to realize fine granular routing, and the original flexible waveband OXC node shown in [19] without split DCSWs (denoted as “m = $\infty$ ” hereafter). Ten trials were performed for each parameter value setting and the averaged results are shown.

 

Fig. 12 Normalized number of fibers relative to conventional networks.

Download Full Size | PDF

Table 1. Network topologies and their characteristics.

View Table

Figure 12 shows the relative ratios of fiber numbers needed in a network. Note that the results are normalized to the values yielded by conventional fine granular wavelength routing. Labels “B” and “m” stand for WSS degree, B, and split small DCSW size, m x m. As shown in Fig. 13, increasing B offsets the reduction in DCSW size m. The fiber increment for the proposed design with 1x8 WSSs and split switches using 3x3 DCSWs in parallel is generally less than 5% over the conventional flexgrid network design for all cases in tested topologies. The shrinkage of DCSW size contributes to not only compact DCSW implementation but also the elimination of amplifiers. Note that the results also depend on node degree of the network topology (Fig. 13). This is because the proposed OXC nodes have the routing constraint such that a path from an input fiber can be only routed to up to B output fibers.

 

Fig. 13 Normalized number of WSSs relative to conventional networks with 1x9 WSSs.

Download Full Size | PDF

Figures 13 and 14 show the relative ratio of the normalized number of WSSs needed in a network. Note that the conventional network uses 1x9 WSSs or 1x20 WSSs, respectively. For all parameters of WSS degree and DCSW size, almost the same trends were observed, and thus only selected results are presented. The maximum OXC scale degree reaches forty or fifty in the tested topologies as the number of fibers in each link increases (e.g. it reaches around 60 in the Pan-European network). Compared to the conventional networks created using 1x9 WSSs, the number of WSSs necessary is substantially reduced, by up to 75%, regardless of WSS degree and DCSW size, and the average number of paths between each node pair is 20. On the other hand, compared to the conventional networks with 1x20WSSs, the number of WSSs is reduced by up to 50%. Regardless of the case, the proposed OXC can substantially reduce the number of WSSs. Based on sub-sections, 4-1 and 4-2, B = 8 and m = 3 were adopted in developing the prototype as explained in the following sub-section.

 

Fig. 14 Normalized number of WSSs relative to conventional networks with 1x20 WSSs.

Download Full Size | PDF

Remark: We also evaluated the performance with different distributions of path capacity such that the ratios of 40/100/400 Gpbs requests were set at 1:1:1, 2:1:1 and 1:1:2, respectively. As shown in Fig. 15(a), the routing performance does not depend on the distribution variation of path capacity. source. On the other hand, the reduction in the number of WSSs becomes large as large capacity traffic (i.e., 400 Gbps) increases since the conventional OXCs need more WSSs to create large scale OXCs (Fig. 15(b)). This is because 400Gbps requires more spectral resources and hence increasing the ratio of 400Gpbs (i.e. the 1:1:2 case) corresponds to the higher traffic intensity case presented in the main part of Sec. 4.2. Figure 16 shows the relative ratios of the normalized number of fibers and WSSs. Note that only the results of proposed OXC nodes with B = 8 and m = 3 are shown.

 

Fig. 15 The impact of distribution variation of each path request capacity.

Download Full Size | PDF

 

Fig. 16 Experimental setup.

Download Full Size | PDF

4.3 Transmission experiments

In order to confirm the feasibility of our OXC architecture, we built a flexible waveband routing OXC prototype with (B, m) = (8, 3). The experimental setup is shown in Fig. 16. At the transmitter side, 80-wavelength 30-GBaud QPSK signals were formed with an IQ modulator (IQM) driven by a 2-channel pulse-pattern generator (PPG). Then, dual-polarization QPSK signals were emulated in the split-delay-combine manner and their optical signal-to-noise ratio (OSNR) was controlled with an amplified-spontaneous-emission (ASE) noise source. The signals thus obtained were incident on the prototype. The signals were amplified with an EDFA whose output power was set to 20 dBm. After that, the signals traversed a 1x9 WSS, 3x3 DCSW, and 9x1 WSS in this order. The 3x3 DCSW was developed using PLC technology and five 3x3 DCSWs were monolithically integrated on a chip with size of 13 x 58 $m{m}^2$(Fig. 17). The losses of WSS and 3x3 DCSW were around 6.5 dB and 7.0 dB, respectively. In other words, the total loss of our node was around 20 dB. After the signals passed through N nodes, the target signal was dropped by a WSS and delivered to an offline digital coherent receiver. After the demodulation circuit, bit-error ratios (BERs) according to OSNR were counted.

 

Fig. 17 5-arrayed 3x3 DCSW PLC chip.

Download Full Size | PDF

Figures 18 and 19 show the bit error ratio and the OSNR penalty measured at the forward-error-correction (FEC) threshold (BER = 10-2), where performance of the conventional 8x8 route-and-select OXC was used as a baseline. We observe that the OSNR penalty due to insertion of 3x3 DCSW was marginal, less than 0.2 dB even after traversing 4 proposed OXCs (i.e. 4 hops); our scheme can realize large scale OXCs without increasing the degrees of WSSs and DCSWs.

 

Fig. 18 Bit-error ratio versus OSNR in 0.1nm.

Download Full Size | PDF

 

Fig. 19 OSNR penalty versus the number of nodes traversed (BER = 10-2).

Download Full Size | PDF

5. Conclusion

We have proposed a compact OXC node architecture that combines WSSs with small-scale switches. Its network design algorithm, which is aware of switch decomposition and differentiated interconnection between WSSs and switches, was also developed. Numerical experiments showed that the proposed OXC and design algorithm can achieve significant reductions in the number of amplifiers and WSSs while keeping the routing performance degradation to acceptable levels. We also developed our proposed OXC prototype using monolithically integrated arrayed switches on a PLC chip. Transmission experiments on the prototype confirmed the technical feasibility of the proposed OXC.