Open Access
Add to BibSonomyAbstract
In order to mitigate the signal spectrum narrowing caused by optical filtering at nodes, an adequate guard band is needed between optical channels, which degrades the frequency utilization of optical fibers. In this study, we propose a grouped routing based network architecture that minimizes spectrum narrowing while greatly improving spectral efficiency. Coarse granular routing at GRE (grouped routing entity) level is employed at each ROADM node, but fine granular add/drop is adopted to retain high frequency utilization. Optical channels are packed densely in each GRE, and sufficient guard bands are inserted between GREs. As a result, signal spectrum narrowing is minimized and efficient spectrum utilization is achieved. Network design/control algorithms that support both static and dynamic traffic growth are developed. Extensive simulations demonstrate the effectiveness of the proposed architecture. To implement the scheme, current LCOS-based ROADMs are applied without any hardware changes; only the control schema are modified.
© 2016 Optical Society of America
1. Introduction
Due to the penetration of broadband access and the expansion of bandwidth-demanding video services including 4k high-definition video distribution services, Internet traffic continues to increase substantially. Intra-regional traffic, particularly metro traffic, is forecasted to grow about twice as fast as backbone traffic [1]. This is accelerating the wide deployment of ROADMs/OXCs in metro/regional networks, as they can eliminate expensive O/Es and E/Os from intermediate nodes.
The continuous traffic increase is driving the development of new technologies for fiber capacity enhancement. One example is the recent proposal of the elastic optical network [2]. Elastic optical networks enhance the frequency utilization of fibers by reducing the channel spacing, but the degree is limited due to the impairment caused by the optical filtering within the Wavelength Selective Switches (WSSs) at ROADM nodes [3,4 ]. The super Gaussian order of current LCOS WSSs, which can handle elastic optical paths, is approximately limited to 4-3 [5], and achieving higher orders will be difficult and costly. In order to improve spectral efficiency, suppression of the filter narrowing effect has been studied [6–14 ].
The traffic growth envisaged will further worsen the spectrum-narrowing problem. The increase in traffic demands the use of multiple fibers on each link, which will enlarge the necessary ROADM scale. Present ROADM port counts are capped at around 8, and when the port count demands exceed this limit, the commonly deployed architecture of ROADM, broadcast-and-select, in which optical couplers and WSSs are located at input and output sides (respectively), will need to be replaced by the route-and-select architecture (WSSs at both sides) [4], which will double the number of WSSs traversed at a ROADM node. The increased number of WSSs traversed demands broad guard bands between optical channels, which degrades spectral utilization. Moreover, when the ROADM degree exceeds that of available WSSs, currently limited to around 20 + , concatenation of WSSs will be needed to attain the necessary port count. This will further increase the number of WSSs traversed. Although ultra-high port-count WSSs have been intensively researched [15–17 ], problems such as cost and performance need to be solved before they are to become viable. Another important point to be considered for the future is that when ROADMs penetrate more widely in metro networks, the number of nodes traversed will become much larger than is true in core networks, since a metro network can have up to 200 nodes [3]. Thus, reducing the impairment imposed by optical filters is becoming a critical barrier to the maximization of fiber frequency utilization and the expansion of the applicable areas of wavelength routing networks given that larger port count route-and-select ROADMs will be needed.
In this paper, we propose the Grouped Routing [18,19 ] network architecture; it enables us to substantially reduce the guard band bandwidth necessary and the impairment caused by optical filtering at WSSs. The architecture combines coarse granular routing and fine granular (channel by channel) add/drop at ROADMs. The fiber frequency bandwidth is divided into several sub-bands called “Grouped Routing Entities”, or “GREs”. Routing at each ROADM node is done at the GRE level by setting a broad passband that completely covers each GRE. This routing scheme was originally developed for conventional fixed grid networks, where simple coarse granular routing devices, waveband selective switches (WBSS) that can be monolithically integrated on PLC chips [20], are utilized to implement Grouped Routing. The WBSS supports the allocation of interleaved channels on the ITU-T grid. The coarse granular routing concept applied in this study is limited to GREs consisting of continuously allocated channels. A sufficient guard band is inserted between GREs, while the optical channels within a GRE are densely packed. This achieves, simultaneously, minimization of filtering impairment and efficient spectrum utilization. One of the beauties of this scheme is that the hardware needed to implement this is basically the conventional LCOS-based WSS; only the control functions need be modified. The suppression of signal spectrum narrowing by using broad passband filters that cover multiple channels was studied in [10–13 ]. Our research [21] was the first to investigate and confirm the network-wide impact. One issue is the use of channel granular filtering operations at ROADM nodes to realize optical path drop, as optical paths adjacent to the dropped path are impaired by the operation.
We present here novel network design/control algorithms that can be applied to static network design and dynamic optical path control, where the impairment by adjacent path drop operations can be limited to an acceptable level. Numerical evaluations demonstrate that the proposed network can greatly improve frequency utilization efficiency and, as a result, fewer fibers are necessary to accommodate a certain amount of traffic.
The organization of this paper is as follows. Section 2 introduces the definitions used in describing the algorithms. Section 3 details the proposed optical filtering operation that reduces spectral narrowing. In Section 4, we propose two algorithms for a static network design scenario and a traffic growth model scenario. Section 5 demonstrates the effectiveness of the proposed architecture through various simulations. The proposed method can improve fiber utilization by 30% for 400Gbps DP-16QAM signals without any node hardware changes. Finally, we conclude this paper in Section 6. Parts of the preliminary versions of this study were presented at international conferences [21–23 ].
2. Preliminary
2.1 Grouped Routing network
The concept of Grouped Routing was first proposed and investigated for conventional fixed grid optical networks in [18,19 ]. Those studies divide the wavelengths in a fiber into several wavelength groups called “Grouped Routing Entities” (GREs). Routing is done at the GRE granularity level while add/drop operations at a node are processed at the wavelength granularity level as shown in Fig. 1 . GRE granularity routing defines virtual pipes (GRE pipes); wavelength paths can be added/dropped at any intermediate node on the route of a GRE pipe (See Fig. 2 ). GRE pipes differ from waveband paths (see Fig. 3(b) ) in conventional hierarchical optical path networks since GRE pipes do not offer path functions such as termination as defined by ITU-T [24]. This is a key difference from the conventional waveband “path”. It may form a closed loop as seen in Fig. 2, and a GRE can be regarded as a virtual fiber. Furthermore, GRE pipes differ from “super-channels” in that sub-carrier signal cannot be dropped(/added) from each super-channel along the route from source to destination node (See Fig. 3(a)).
The previous studies on Grouped Routing [18,19 ] focused on hardware scale reduction as mentioned before and the routing performance offset stemming from the coarse granular routing can be effectively mitigated by applying a novel network design algorithm [19]. Regarding hardware simplification, it is possible to utilize a waveband selective switch (WBSS) that can be monolithically integrated on a PLC chip, or 3D MEMS-based WSSs where the number of necessary mirrors can be reduced to 1/(number of channels in a GRE pipe). This does not hold true for LCOS-based WSSs, since the desirable optical filtering resolution is not changed from that of the conventional approach. Thus the grouped routing operation does not simplify LCOS-based WSS devices. On the other hand, the GRE concept using LCOS based WSSs can greatly reduce optical filtering impairment for each optical channel, as is verified in Section 3.
2.2 Spectrum narrowing
Since an ideal optical filter, i.e. a filter that has perfectly sharp cut-off, cannot be realized, the only solution available to conventional networks is to reserve sufficiently wide guard bands (see Fig. 4 ) and bound the impairment to a certain level. When the number of filtering operations increases or number of the node hops increases, the necessary guard band bandwidth increases, which worsens the fiber spectrum resource utilization efficiency. Figure 4 shows an example of how a WSS manages optical channels, where the frequency spectrum shape of each channel is rectangular (i.e. Nyquist WDM). If the frequency grid spacing is not broad enough, all channels will suffer from the spectrum narrowing effect whenever they traverse a WSS. The channel spacing should be broad enough to reduce the expected narrowing effect to a certain level. In other words, frequency utilization can be degraded. To resolve this, we must set a stringent filter shape requirement for WSSs; a higher order index for super Gaussian must be specified.
3. Grouped Routing that minimizes spectrum narrowing
In general, channel frequency bandwidth is several times that of rach guard band, however, the bandwidth wasted by guard bands is not negligible as discussed in the previous section. It is hard to reduce this overhead if we stick to the conventional routing mechanism. Our proposal is to use GRE pipes, each of which can accommodate several channels, and separate the pipes by guard bands as shown in Fig. 5 .
We assume hereafter, only for simplicity, that all GRE pipes set in a network occupy the same bandwidth. Our analysis below can be easily extended to unequal plural GRE bandwidths. Guard bands are inserted only between GRE pipes, and thus, the wasted bandwidth is substantially reduced. As you can imagine, the larger the GRE bandwidth, the higher the frequency utilization, however, the utilization of each GRE tends to be lessen. Thus, an optimal GRE bandwidth exists, as is discussed in Section 5.
The guard bands between GRE pipes protect the optical channels located at the GRE edges from the spectrum narrowing at routing nodes; each GRE is routed as a bundle. An optical channel in a GRE pipe incurs spectrum narrowing only when an adjacent optical channel is dropped at a node. In Fig. 6 , optical channel dropping is done using an optical coupler, however, the dropped channels need to be completly filtered out at the WSS in the ROADM so that none are routed to output fibers. Of course channel dropping can be done at the WSS in ROADM (in this case only a few WSS output ports are used for dropping). Adding signals at ROADMs doesn't cause any filtering effects because optical couplers can simply combine the optical signals. As to dropping, to reuse the bandwidth occupied by the dropped signals, a filter is applied to the corresponding bandwidth. Generally, longer optical paths tend to incur spectrum narrowing due to the possible multiple drop operations of adjacent optical paths. This, however, can be resolved by applying a wavelength (frequency slot for elastic optical networks) assignment strategy that can reduce frequency of filtering operations. For example, Fig. 7 shows three optical paths accommodated between node 1 to node 3: the wavelength assignments for the three paths are different in the upper and lower charts are different. With the upper wavelength assignment, spectrum narrowing of the green path occurs by adjacent red channel drop, while the lower wavelength assignment prevents this.
4. Design and control algorithm for Grouped Routing network with dense channel packing
The grouped routing scheme proposed in the previous section requires management of not only bundled path routing but also the frequency of adjacent path drop operations. In order to show that these requirements do not deteriorate the path routing performance and hence attain high frequency utilization efficiency, we describe a static network design algorithm that is an improved version of the one presented in [19], see Section 4.1. We also propose a path and GRE pipe control algorithm applicable for the traffic growth scenario. In the scenario, new path demands are added over time and hence the optimality of the path configuration, determined before the path demand increase, is lost. Thus a new control method needs to be developed that can maximally exploit the GRE. Section 4.2 verifies that our solution works well in practical situations.
4.1 Static network design
This algorithm defines two GRE pipe types: direct GRE pipes and aggregative GRE pipes. The former are filled with optical paths that have common source and destination nodes. The latter carry all other paths. The spectral narrowing effect caused by adjacent channel drop only occurrs in aggregative GRE pipes since the effect is triggered by optical paths in the pipes that have different source and destination node pairs. There is no filtering effect in direct GRE pipes The setting of these two GRE pipe types allows the algorithm to maximizes the fiber utilization efficiency and suppress the filtering effect.
< Design algorithm for Grouped Routing networks to mitigate spectral narrowing >
- Step1 Reserve optical paths to be carried by direct GRE pipes
Find and assign paths to direct GRE pipes. The number of reserved paths is a multiple of GRE capacity.
- Step2 Establish aggregative GRE pipes
From among the remaining paths, the algorithm finds the set of the optical paths that maximally fills up a GRE pipe and allocates them to an aggregative GRE pipe. When paths to be assigned to the pipe are determined, allocate each path a wavelength (slots) so as to minimize the frequency of adjacent path drop. This sorting reduces the number of drop operations of adjacent paths. Finally, establish the GRE pipe and allocate it to the network. Repeat these operations until all of non-reserved optical paths are assigned to the network.
- Step3 Establish the direct GRE pipes
Calculate the routes of the direct GRE pipes so as to efficiently use the fiber resources and allocate all direct GRE pipes in the network.
4.2 Dynamic network control
In dynamic network control, we restrict the number of adjacent path drop operations. Wavelengths assigned to already accommodated paths cannot be changed (re-optimized) at each demand increase operation, in order to prevent service disruption. Similarly, the signal impairment can be bounded by limiting the number of adjacent channel drops while the accommodated traffic is enhanced due to the dense packing within GRE pipes. The algorithm that manages newly arriving path demands is summarized below.
< Dynamic control algorithm of GRE-based network with adjacent path drop restriction >
- Step1 Try to accommodate a new demand within existing GRE pipes
If there is a GRE pipe with spare capacity that connects the source and destination nodes of the demand, the algorithm goes to Step2 and tries to accommodate the demand in that pipe. If no existing GRE pipe can accommodate the demand, go to Step3.
- Step2 Restrict the number of filtering operations
The number of adjacent channel drop operations is bounded, and the accommodation of the demand is denied if the bound is violated. If multiple pipes satisfy the requirements, the shortest GRE pipe is selected. If no existing GRE pipe can satisfy the requirements, go to Step3.
- Step3 Try to accommodate within a new GRE pipe
A new GRE pipe connecting the source and destination node is established and the demand is accommodated within it. If there is no fiber capacity left to establish the pipe, the demand is blocked.
5. Simulation
5.1 Calculation of fiber capacity
At first, we investigate, for each optical path, the number of filtering operations it will experience in the network to assess the filtering effect penalty and calculate fiber capacity. The topologies tested are the US long haul network [25] shown as Fig. 8(a) , Japan’s photonic network model [26] shown as Fig. 8(b), Telecom Italia network [27] shown as Fig. 8(c) and a 7x7 regular mesh network shown as Fig. 8(d). The traffic demand is randomly generated to follow a uniform distribution over the topology of interest. The average demand for each node pair is set at twenty optical paths. The number of paths accommodated in a fiber is set at 102 and 6 paths are bundled into a GRE. The benchmark is the corresponding conventional network where optical paths are independently routed. Only shortest hop count routes were considered in the conventional network to minimize the spectrum narrowing effect. Each ROADM node used WSSs with the route-and-select configuration.
Figure 9 shows a normalized distribution of the number of filtering operations when optical paths are individually routed, i.e. conventional routing method, and when GREs are utilized. It shows that using GREs can substantially reduce the number of filtering operations, both maximum and average values. The maximum number of hops is 12 for a 7x7 network and hence 22 WSSs (2 WSSs per intermediate node, 11 intermediate nodes) will be passed by the longest path in the conventional network. The proposal holds the maximum number of optical filtering operations to 2. Please note that the drop (add) portion may require additional WSSs if non-coherent detection is used, however it is not considered here for simplicity. The same assumption is used for the Grouped Routing case.
Figure 10 shows the OSNR penalties caused by the spectrum narrowing subject to the number of WSSs traversed. WSS filter shape was assumed to be m-th order super Gaussian. We investgated different modulation formats (DP-QPSK and DP-16QAM), channel spacing’s, and filter shapes. The baud rate of DP-QPSK (resp. 16QAM) is 28 (resp. 56 = 28*2) Gbaud and bit rate is 100 (resp. 400) Gbps. Note that increasing the FEC overhead is not always effective because the larger signal bandwidth is more susceptible to filter narrowing. Here, the filter bandwidths are set to be equal to the −10 dB down channel bandwidth. To measure the pure effect of filter narrowing, nonlinear degradation was not considered. For example, in the conventional optical path networks, if QPSK is used with 37.5GHz channel spacing, WSS filter order m is 3, and the number of hops is 5 (thus 8 WSSs are traversed), the power penalty will be 1.9 dB according to Fig. 10(a).
Based on the evaluations shown in Figs. 9 and 10 , we can determine the number of optical paths accommodated in a fiber. WSS filter order “m” was assumed to be 4. The narrowest channel spacings that limit the worst power penalties to around 1.0 dB were evaluated. In the evaluations, the maximum number of filtering times for conventional network ranged from 16 to 26 for each topology. According to Fig. 10, the conventional network needs 50GHz (resp. 87.5GHz) channel bandwidth for QPSK (resp. 16QAM) to hold the worst OSNR penalty under 1dB. The number of channels accommodated per fiber is 88 (resp. 50). On the other hand, the proposed network can use 37.5GHz (resp. 62.5GHz) channel spacing because the filtering number is limited to 2. Fiber capacity of Grouped Routing network varies with GRE capacity because 25 GHz guard band is inserted between adjacent GREs (See Fig. 11 ). The number of channels accommodated per fiber corresponding to GRE capacity is shown in Fig. 12 . These values are used in the rest of this section.
According to Fig. 12, fiber capacity enhancement was larger in 16QAM than QPSK. For example, when GRE capacity is set to 6, the number of channels per fiber can be increased 16% in QPSK and 32% in 16QAM. 16QAM signals are susceptible to filter narrowing and hence the reduction in channel spacing was larger. However, the transmission reach of 16QAM is limited. Therefore, QPSK is needed for large area networks such as those of the US and Japan.
5.2 Static network design simulation
Figure 13 shows the results of the static network design presented in Section 4.1. The average values of 20 simulations utilizing different traffic patterns are plotted. The horizontal axis plots the average number of paths between each node pair, i.e. the traffic volume. The vertical axis plots the number of fibers needed to accommodate given traffic into the network. The result of Grouped Routing is normalized by that of the conventional routing method.
Fig. 13 Necessary fiber numbers of Grouped Routing network normalized by those of conventional network.
Grouped Routing yields more empty channels in fibers than conventional routing methods due to its coarse granular routing, particularly when traffic demands are small, if we keep using the same channel spacing as conventional channel-by-channel routing networks, which has been evaluated in previous work [18,19 ] as explained in the Introduction. On the other hand, in the Grouped Routing networks discussed herein, channels are densely packed in each GRE as shown in Fig. 11, which enhances fiber utilization, particularly when traffic is sufficient to fill coarse granular pipes. Hence, GRE needs more fibers than conventional networks when the traffic volume (average number of paths) is very small, which reflects the inefficiency of course granular routing or GREs in smaller traffic conditions. However, substantially fewer fibers are needed when traffic volume is relatively large. Overall, the necessary fiber number is decreased over a wide traffic range because the dense channel packing in each GRE can greatly enhance fiber utilization
When the average number of paths between a node pair is 20, the necessary fibers can be reduced by 12.6% for the US long haul (with QPSK), 12.7% for Japan’s network (with QPSK), 22.8% for Italia’s network (with 16QAM) and 23.5% for the 7x7 regular mesh (with 16QAM) compared to the conventional network.
Figure 14 shows the result when different traffic distribution patterns are assumed. Here we examine Japan’s network model and QPSK signals. Population weighted, Tokyo-centric, and Tokyo-Osaka-centric traffic distributions were utilized in addition to the uniform traffic distribution. For all population-weighted models, the traffic volume between any two nodes is set so that it is proportional to the product of the population of each node. In the Tokyo-centric traffic model, all the paths from the other nodes are added/dropped at Tokyo, like the hub and spoke configuration, and no direct traffic between nodes other than to/from Tokyo is assumed. The traffic intensity between a node to/from Tokyo is set to be population weighted. It provides a good test as regards extremely non-uniform traffic patterns. The parameter settings were same as the simulation of Fig. 13. The Tokyo-Osaka-centric model considers population-weighted traffic between Tokyo/Osaka and other nodes, and between Tokyo and Osaka. According to Fig. 14, the performance of Grouped Routing network is improved with more centralized traffic conditions because paths are more effectively bundled, which enhances the usage of GREs.
Next, we investigated the effect of GRE capacity utilizing GRE capacity values shown in Fig. 12. The advantage of using wider GRE bandwidth is that the total guard band bandwidth can be reduced, which increases the number of paths accommodated in a fiber. As a result, higher spectral utilization is expected. On the other hand, more traffic is needed to fill each broad pipe. In other words, it suffers from low pipe utilization efficiency over a wide traffic range. For narrower GRE bandwidths, the discussion is opposite. This trade-off is presented in Figs. 15 .
Figure 15 shows the normalized necessary number of fibers to accommodate certain levels of traffic demands where GRE capacities are changed. The results are for the 7x7 regular mesh network and the modulation format is 16QAM. The larger the traffic volumes, the larger the relative fiber number reduction. The fiber number reduction, or the increase in the number of channels accommodated per fiber, is saturated when GRE pipe capacity becomes larger than a certain value (~6). Simulations were carried out for the other topologies and similar trends were observed. This shows that larger GRE capacity is preferable as traffic volume increases, and that an optimal GRE bandwidth exists for each traffic demand value. The optimal bandwidth increases with the traffic volume.
5.3 Traffic growth model senario
We conducted simulations using the traffic growth model where paths are added to simulate gradual traffic expansion (additional path demand during the next period is estimated at every network expansion cycle, for example, every half year).
We first applied the static network design algorithm presented in Section 4.1 for different traffic demand sets randomly generated for a given forecasted traffic increase. The number of fibers on each link is determined in this process. The fiber arrangement is utilized in the following dynamic simulation. The dynamic path setup requests between each node pair arrive randomly. The metric for benchmarking was the traffic volume at which the path blocking ratio exceeds 0.1%. The proposed method is compared with the conventional network (optical paths are independently routed) with exactly the same fiber resources. We conducted 300 dynamic control simulations for each of 20 fiber configuration patterns each of which was created utilizing different traffic matrix, and the ensemble average was used for the comparison. The traffic volume metric is the average number of paths between each node pair in the static network design stage, i.e. the forecasted traffic volume. The parameters related to the filtering penalty are the same as those used in Section 5.1. When we use QPSK (resp. 16QAM), the conventional network demands 50GHz (resp. 87.5GHz) channel bandwidth and the number of channels accommodated per fiber is 88 (resp. 50). The fiber capacity of the proposed network depends on GRE capacity and filtering performance. With QPSK, the channel spacing is 37.5GHz if the filtering number is limited to 11 or less. When 16QAM is used, the channel spacing is 62.5GHz if the filtering number is limited to 2 or less. When the number is between 3 and 9, 75GHz channel spacing is used. The filtering limitation value is chosen before each simulation run. The proposed algorithm controls optical paths so that the filtering number does not exceed the limit value.
The horizontal axis of Fig. 16 plots the configured bounds of filtering number and the vertical axis plots accommodated traffic volume normalized by that of the conventional network. Here, GRE bandwidth is set at 6 channels. The forecasted traffic volume is 12. The result for QSPK was largest when the filtering number is limited to 2 or less and the results are almost same when the maximum filtering number was over 3. For 16QAM, the accommodated traffic was largest when the filtering number limit was 2. The proposed routing strategy can accommodate about 20% greater traffic demand than the conventional network in each topology (Italia and 7x7). When the maximum filtering number exceeds 3 with 16QAM, GRE must adopt 75GHz channel spacing, which is still smaller than the 87.5GHz used for conventional networks, however, GRE pipe utilization is not high enough to attain larger overall accommodated traffic volumes since Grouped Routing deteriorates routing performance, i.e. the usage of each GRE pipe is not high enough.
Figure 17 shows the normalized accommodated traffic volume simulated using various traffic patterns. This simulation utilizes Japan’s network model as in Fig. 13 and the parameter settings follow those of Fig. 16. Figure 17 shows that the performance of Grouped Routing decreases when we use biased traffic patterns, however, the proposal can still accommodate more traffic than the conventional routing method (8-30%). In the case of population distribution, the tested traffic divergence is huge since the populations of Tokyo and Osaka are much higher than those of other cities, and hence the number of paths between Tokyo and Osaka can be more than 150 times that between any two other cities. In this extreme case, the usage of GREs connecting small cities is not high enough given the assumed traffic volume so the GRE approach has limited effectiveness.
Figure 18 shows the relative accommodated traffic volume with different GRE capacities. The network topology used was the 7x7 regular mesh. Modulation format was 16QAM and the maximum filtering number was limited to less than 3. The trend is similar to that of the static simulation case (Fig. 15). The results for large GRE capacities (>6) are worse than that of the static simulation because the routing capability limitation of GRE has a greater impact in the dynamic network control scenario. As expected, the relative accommodated traffic volumes and the optimal GRE capacity increase with the forecasted traffic increase. Please note that no hardware modification is required in changing the GRE capacity, only the LCOS WSS settings need be modified. Thus the proposed method is shown to be flexible enough to adapt to dynamic variations in traffic volume.
6. Conclusion
We evaluated the effectiveness of Grouped Routing networks in relation to the optical filtering impairment. Network design/control algorithms for both static and dynamic traffic growth models were developed. Numerical evaluations demonstrated that when the channel spacing is fixed, the WSS filter order needed can be relaxed or when the filter order is fixed, the applicable physical network scale (the maximum number of nodes traversed) can be substantially expanded or spectral efficiency can be increased which results in fewer fibers being needed. Please remember that the ROADM node architectural requirements are the same for Grouped Routing networks as for the conventional ones, only the control algorithm need be modified to implement the proposed scheme. We believe that the proposed technologies will be effective in creating large scale and frequency-efficient optical networks of the future.
Acknowledgments
This work was partly supported by NICT and KAKENHI (26220905).
References and links
1. Strategic White Paper by Bell Labs, “Metro Network Traffic Growth: An Architecture Impact Study,” http://resources.alcatel-lucent.com/asset/171568.
2. M. Jinno, H. Takara, B. Kozicki, Y. Tsukishima, Y. Sone, and S. Matuoka, “Spectrum-efficient and scalable elastic optical path network: architecture, benefits, and enabling technologies,” IEEE Commun. Mag. 47(11), 66–73 (2009). [CrossRef]
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. Y. Sakamaki, T. Kawai, T. Komukai, M. Fukutoku, and T. Kataoka, “Evaluation of optical filtering penalty in digital coherent detection system,” IEICE Commun. Express 1(2), 54–59 (2012). [CrossRef]
6. K. Harada, K. Shimizu, T. Kudou, and T. Ozeki, “Hierarchical Optical Path Cross-Connect Systems for Large Scale WDM Networks,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 1999), paper WM55. [CrossRef]
7. C. Blaizot, E. Dotaro, L. Noirie and A. Jourdan, “Multi-Granularity Optical Networks,” in Optical Network Design and Modeling, IEEE Technical Digest (CD) (Institute of Electrical and Electronics Engineers, 2000), invited paper page 1.
8. L. Noirie, C. Blaizot, and E. Dotaro, “Multi-Granularity Optical Cross-Connect,” in European Conference and Exhibition on Optical Communication, OSA Technical Digest (CD) (Optical Society of America 2000), paper 9.2.4.
9. A. Ghazisaeidi, P. Tran, P. Brindel, O. Bertran-pardo, J. Renaudier, G. Charlet, and S. Bigo, “Impact of Tight Optical Filtering on the Performance of 28 Gbaud Nyquist-WDM PDM-8QAM over 37.5 GHzGrid,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2013), paper OTu3B.6.
10. F. Paolucci, F. Fresi, A. Castro, L. Velasco, F. Cugini, N. Sambo, A. Giorgetti, L. Potí, and P. Castoldi, “Filter Optimization in SDN-based Flexgrid Networks,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2014), paper Th3l.4. [CrossRef]
11. A. Castro, F. Cugini, L. Velasco, and P. Castoldi, “On the benefits of differentiating the filter configurations in flexigrid optical networks,” in European Conference and Exhibition on Optical Communication, OSA Technical Digest (CD) (Optical Society of America 2014), paper P.6.16.
12. F. Paolucci, F. Cugini, F. Fresi, G. Meloni, A. Giorgetti, N. Sambo, L. Potí, A. Castro, L. Velasco, and P. Castoldi, “Superfilter Technique in SDN-Controlled Elastic Optical Networks [Invited],” J. Opt. Commun. Netw. 7(2), A285–A292 (2015). [CrossRef]
13. P. S. Khodashenas, J. M. Rivas-Moscoso, D. Klonidis, D. M. Marom, and I. Tomkos, “Evaluating the Optimum Filter Resolution and Sub-Channel Spectrum Granularity for Flexible Super-Channels”, in European Conference and Exhibition on Optical Communication, OSA Technical Digest (CD) (Optical Society of America 2014), paper W1I.5.
14. A. Morea, J. Renaudier, A. Ghazisaeidi, O. Bertran-Pardo, and T. Zami, “Impact of reducing channel spacing from 50 GHz to 37.5 GHz in fully transparent meshed networks,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2014), paper Th1E.4. [CrossRef]
15. X. Gu, K. Seno, H. Tanobe, and F. Koyama, “60-channel Wavelength Selective Switch on Bragg Reflector Waveguides Array with 125 Outputports,” in European Conference and Exhibition on Optical Communication, OSA Technical Digest (CD) (Optical Society of America 2014), paper We.3.5.3.
16. K. Suzuki, Y. Ikuma, E. Hashimoto, K. Yamaguchi, M. Itoh, and T. Takahashi, “Ultra-High Port Count Wavelength Selective Switch Employing Waveguide-Based I/O Frontend”, in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2015), paper Tu3A.7. [CrossRef]
17. M. Iwama, M. Takahashi, Y. Uchida, M. Kimura, R. Kawahara, S. Matsushita, and T. Mukaihara, “Low Loss 1x93 Wavelength Selective Switch Using PLC-based Spot Size Converter,” in European Conference and Exhibition on Optical Communication, OSA Technical Digest (CD) (Optical Society of America 2015), paper Mo.4.2.2.
18. Y. Taniguti, Y. Yamada, H. Hasegawa, and K. Sato, “A novel optical networking scheme utilizing coarse granular optical routing and fine granular add/drop,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2012), paper JW2A.2.
19. Y. Taniguti, Y. Yamada, H. Hasegawa, and K. Sato, “Coarse granular optical routing networks utilizing fine granular add/drop,” J. Opt. Commun. Netw. 5(7), 774–783 (2013). [CrossRef]
20. R. Hirako, K. Ishii, H. Hasegawa, K. Sato, H. Takahashi, and M. Okuno, “Development of Single PLC-Chip Waveband Selective Switch that Has Extra Ports for Grooming and Termination”, in Proceedings of the 16th Opto-Electronics and Communications Conference (2011), pp. 483–484.
21. Y. Terada, Y. Mori, H. Hasegawa, and K. Sato, “Enhancement of Fiber Frequency Utilization by Employing Grouped Optical Path Routing,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2014), paper W1C.6. [CrossRef]
22. Y. Terada, Y. Mori, H. Hasegawa, and K. Sato, “Verification of High Frequency Spectrum Utilization in Grouped Optical Path Routing Networks under Traffic Growth Scenario,” in European Conference and Exhibition on Optical Communication, OSA Technical Digest (CD) (Optical Society of America 2014), paper Mo.3.1.4. [CrossRef]
23. Y. Terada, Y. Mori, H. Hasegawa, K. Sato, and T. Tanaka, “Evaluation of Frequency Utilization Enhancement with Transmission-Characteristics-Aware Grouped Optical Path Routing,” in Photonics in Switching, IEEE Technical Digest (CD) (Institute of Electrical and Electronics Engineers, 2015), paper Poster.14.
24. ITU-T Recommendation G.783, Digital transmission systems.
25. A. Dwivedi and R. Wagner, “Traffic model for USA long-distance optimal network,” in Optical Fiber Communication Conference, OSA Technical Digest (CD) (Optical Society of America, 2000), paper TuK1.
26. T. Sakano, Y. Tsukishima, H. Hasegawa, T. Tsuritani, Y. Hirota, S. Arakawa, and H. Tode, “A study on a Photonic Network Model based on the regional characteristics of Japan,” in Proceeding of Technical Report of IEICE, PN2013–01 (2013), pp. 1–6 (in Japanese)
27. V. B. Allasia and M. Potenza, “Characteristics and trends of telecom italia transport networks,” Fiber Integr. Opt. 27(4), 183–193 (2008). [CrossRef]
