Open Access
Add to BibSonomy
Abstract
Photonic networks based on wavelength-selective switches (WSSs) can transport wavelength-division-multiplexed (WDM) signals in a cost-effective manner. To accommodate the ever-increasing network traffic, the spectral efficiency should be maximized by minimizing the bandwidth of the guardbands inserted between WDM signals. Quasi-Nyquist WDM systems are seen as offering the highest spectral efficiency in a feasible way. However, highly dense WDM systems suffer from the signal-spectrum narrowing induced by the non-rectangular passbands of WSSs. Furthermore, widely deployed WSSs cannot process quasi-Nyquist WDM signals since the signal-alignment granularity does not match the passband resolution of the WSSs. This paper proposes a network architecture that enables quasi-Nyquist WDM networking with widely deployed WSSs. Through intensive network analyses based on computer simulations, we confirm that it has 30.8% higher spectral efficiency than conventional networks. Its feasibility is verified by transmission experiments on 72-channel 32-Gbaud/400-Gbps dual-carrier dual-polarization 16-ary quadrature-amplitude-modulation signals aligned with 66.6-GHz spacing in the full C-band. The net fiber capacity of 28.8 Tbps, the transmission distance of 900 km, and the hop count of 9 are attained by our proposed quasi-Nyquist WDM networking scheme.
© 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
In photonic networks, wavelength-selective switches (WSSs) process wavelength-division-multiplexed (WDM) signals without power-consuming optical-to-electrical and electrical-to-optical conversion. Due to the rise of subscription-based streaming services and cloud-computing services, the amount of Internet traffic is exponentially increasing [1]. To cost-effectively accommodate the large amounts of expected traffic, the network capacity needs to be increased by enhancing the spectral efficiency without replacing currently deployed hardware. Setting smaller guardbands between WDM signals results in higher spectral efficiency. In such networks, however, the signal spectrum is narrowed with each WSS traversal [2–4]. Furthermore, the spectrum-narrowing effect is emphasized as the number of WSS traversals increases. Since the signal-to-noise ratio (SNR) is degraded by spectrum narrowing, employing smaller guardbands shortens the signal transmission reach.
The number of spectrum-narrowing events can be suppressed by introducing grouped routing; multiple optical paths bundled in a group are jointly managed at their intermediate nodes [5,6]. At a WSS, a passband is set on a group basis and a relatively broad guardband is adopted only between adjacent groups. Since the overall guardband bandwidth is reduced, the spectral efficiency of a fiber can be enhanced. However, wavelength assignment in the grouped-routing networks is restricted by passband granularity of the WSSs. As a result, some spectrum resources remain unassigned.
Quasi-Nyquist WDM can minimize the idle frequency resources in a feasible way and thus offers high spectral efficiency provided that a transmitter and receiver are connected in a point-to-point manner [7,8]. In photonic networks, however, quasi-Nyquist WDM signals suffer from severe spectrum narrowing. Furthermore, the bandwidth assigned for each quasi-Nyquist WDM signal does not match the WSS bandwidth resolution; hence, the signals cannot be processed on a path basis. Although finely tunable WSSs are available [9], installing them in networks or replacing existing ones would be impractical due to the general fact that finely tunable WSSs are much more costly than coarsely tunable ones. The currently deployed WSSs will continue to be used for a long time since mean time between WSS failures is over 20 years [10]. The aim of this paper is to show that highly dense WDM networks can be realized without relying upon state-of-the-art hardware.
In this paper, we demonstrate a photonic-network architecture that enables widely deployed WSSs to support quasi-Nyquist WDM transmission. Its effectiveness is confirmed through network simulations and transmission experiments. In order to complement the limited WSS-passband resolution, multiple channels consecutive in the frequency domain are bundled so that the aggregated bandwidth of each bundle matches the WSS-passband resolution. Additionally, the maximum number of spectrum-narrowing events for each path is controlled by our wavelength-assignment algorithm. Consequently, we can accommodate three 32-Gbaud/400-Gbps dual-carrier dual-polarization 16-ary quadrature-amplitude-modulation (DP-16QAM) signals aligned with 66.6-GHz spacing within a 200-GHz bandwidth even while complying with the two restrictions. Network analyses of 400-Gbps DP-16QAM signals confirm that the spectral efficiency is improved by 30.8%. Its feasibility is confirmed by transmission experiments on 72-channel 400-Gbps DP-16QAM signals aligned with 66.6-GHz spacing in 4.8 THz of the full C-band; the net fiber capacity of 28.8 Tbps is demonstrated over the transmission distance of 900 km and the hop count of 9. The preliminary edition of this paper was reported in the Optical Fiber Communication Conference 2019 [11]. We extend the explanation and discussion by adding new results of network simulations and transmission experiments.
The remainder of this paper is organized as follows. Section 2 describes the quasi-Nyquist WDM network architecture and its problems. In Section 3, our proposed quasi-Nyquist WDM network architecture is presented. Section 4 evaluates its spectral efficiency through intensive computer simulations. In Section 5, the feasibility of the quasi-Nyquist WDM network is experimentally proven. Finally, this paper is concluded in Section 6.
2. Quasi-Nyquist wavelength-division multiplexing
2.1. Architecture
To improve spectral efficiency of a fiber, a bandwidth assigned for each WDM signal needs to be reduced. In the OIF-recommended WDM systems, a 50-GHz bandwidth is assigned for a 32-Gbaud/100-Gbps dual-polarization quaternary-phase-shift-keying (DP-QPSK) signal whereas an 87.5-GHz bandwidth is assigned for a 32-Gbaud/400-Gbps dual-carrier DP-16QAM signal [12]. The spectral efficiency can be maximized by eliminating guardbands; however, such an approach is infeasible with real systems due to the frequency instability of lasers [13]. On the other hand, the quasi-Nyquist WDM system can attain high spectral efficiency with only moderate requirements on laser-frequency stability. For example, a 33.3-GHz bandwidth is assigned for a 100-Gbps DP-QPSK signal whereas a 66.6-GHz bandwidth is assigned for a 400-Gbps DP-16QAM signal. The theoretical improvements in the spectral efficiency are 50.0% for 100-Gbps signals (i.e. 50/33.3 ~1.500) and 31.3% for 400-Gbps signals (i.e. 87.5/66.6 ~1.313).
2.2. Problems
Many studies have shown the effectiveness of quasi-Nyquist WDM in point-to-point transmission systems [7,8]. However, its use for networking is hindered by two obstacles, i.e. the limited setting granularity of WSS bandwidth and the significant spectrum narrowing with each WSS traversal. As shown in Fig. 1, the bandwidth assigned for each quasi-Nyquist WDM signal is not a multiple of the WSS-passband resolution. In this example, signals with a 66.6-GHz spacing cannot be separated by currently deployed flexible-grid WSSs whose passband resolution is 12.5 GHz as standardized by ITU-T [14]. The other obstacle is spectrum narrowing as depicted in Fig. 2. When optical signals are filtered, the signal spectra are narrowed by WSS traversal. The spectrum narrowing can be avoided if the passband shape of the WSS is rectangular; however, actual WSS passbands have gradual cut-off characteristics [4], so spectrum narrowing is inevitable. The spectrum narrowing is intensified as the guardband bandwidth decreases. Obviously, quasi-Nyquist WDM signals suffer from extremely severe spectrum narrowing. Moreover, the signal degradation caused by spectrum narrowing further worsens with multiple WSS traversals. Consequently, signal transmission fails if the number of spectrum-narrowing events surpasses the limit.
Fig. 1 The limited WSS-passband resolution, where 32-Gbaud/400-Gbps dual-carrier DP-16QAM signals are depicted as an example.
3. Proposed wavelength assignment realizing quasi-Nyquist WDM networks
3.1. Path bundling to reconcile channel spacing and WSS-passband resolution
To process quasi-Nyquist WDM signals under the restriction of limited WSS-passband resolution, multiple paths are bundled and filtered as a bundle. The concept is shown in Fig. 3, where the 33.3-GHz and 66.6-GHz channel intervals depicted are just examples. Multiple channels are bundled so as to match the aggregated bandwidth with the WSS-passband resolution. For example, three 32-Gbaud/100-Gbps DP-QPSK signals aligned with 33.3-GHz spacing are bundled into a 100-GHz bandwidth, which is a multiple of 12.5 GHz. Bundled paths need to share the same sequence of optical fibers until the bundle is physically terminated by a WSS because paths once bundled cannot be separated by WSSs. On the other hand, with path bundling, quasi-Nyquist WDM signals can be processed with widely deployed WSS hardware. Thus, we can optically process quasi-Nyquist WDM signals though the selectable routes and wavelengths in the path-setup process are restricted due to path bundling. This scheme differs from the super-channel technique. A bundle may have vacant wavelengths whereas a super channel does not. In this scheme, multiple super channels can be bundled, e.g. three 32-Gbaud/400-Gbps dual-carrier DP-16QAM signals can be bundled into a 200-GHz bandwidth, which is a multiple of 12.5 GHz. Note that the frequency resolution of practical lasers is fine enough for quasi-Nyquist WDM systems; lasers with 1-MHz tuning capability are recommended by OIF for the integrable tunable laser assembly [12].
3.2. Wavelength assignment aware of signal-spectrum narrowing
To mitigate the impact of spectrum narrowing, we introduce a wavelength-assignment scheme that is aware of the number of spectrum-narrowing events. Since spectrum narrowing degrades signal quality as the optical path traverses each WSS, we need to control the number of spectrum-narrowing events so as not to exceed the prescribed limit. The limit is determined by the transmission characteristics, e.g. transmission distance, modulation format, and WSS attenuation slope. Since only adjacent paths cause spectrum narrowing, we can control the impact of spectrum narrowing by considering adjacent-path “add-drop” operations at nodes.
Hereafter, we assume that all path routes are calculated and fixed in advance and focus on the assignment of wavelengths to paths. An example of prior route determination can be found in the literature [15], which searches for a set of p-cycles that minimizes the number of cycles while covering all node pairs on the given network topology. As all paths are routed along the small number of cycles, high spectral efficiency can be achieved. In other words, the routing and wavelength assignment problem in a mesh network is reduced to wavelength assignment on ring-shaped sub-networks.
3.3. Filterless signal drop
The bundled-path wavelength assignment shown in Subsection 3.1 and the spectrum-narrowing-aware wavelength assignment explained in Subsection 3.2 impose restrictions in terms of wavelength assignment to optical paths. To relax the restrictions, we introduce filterless signal drop at nodes. The signals to be dropped are sent to the drop portion by using a splitter and detected by coherent receivers. Here, an arbitrary wavelength signal can be extracted by a combination of coherent detection, analog filtering, and digital filtering [16]. Since the drop operation does not necessitate optical filtering, the signal can be dropped with the path granularity irrespective of the WSS-passband resolution. Consequently, paths having different destination nodes can be bundled, and the restriction imposed by path bundling is relaxed. Here, the dropped signal residual in the express port should not be terminated unless another signal is newly added to the same wavelength; the spectrum narrowing is then induced only by the adjacent-path “add” operation, and as a result the restriction due to the spectrum narrowing can also be alleviated. Note that the filterless drop operation is realized by the broadcast-and-select (B&S) node architecture; the route-and-select (R&S) node architecture cannot exploit this benefit because the drop operation is performed by a WSS [17]. The above conditions related to wavelength assignment are summarized in Table 1. On the other hand, the impact of crosstalk is more serious in B&S nodes than in R&S nodes. Therefore, we need to consider the tradeoff between spectrum narrowing and crosstalk. For example, a reference reported that the B&S node is preferable if the port count is less than 9 and the channel configuration is 120-Gbps DP-QPSK signals with 50-GHz spacing [18].
Table 1. The conditions related to wavelength assignment
3.4. Wavelength-assignment algorithm
Figure 4 shows a flowchart of the proposed wavelength-assignment algorithm for quasi-Nyquist WDM networks using widely deployed WSSs [19]. We assume the use of B&S nodes but applying the proposed concept to the network using R&S nodes is straightforward. The procedure is detailed below;
- Step 0. Initialize parameters and load traffic demands.
- Step 1. Bundle optical paths.
- Step 2. Make a priority list.
- Make a list that defines the order of assigning bundles as follows.
- A bundle that shares the same source and destination nodes with the already-assigned adjacent-frequency bundle.
- A bundle that shares the same source node and more nodes with the already-assigned adjacent-frequency bundle.
- A bundle that shares more nodes with the already-assigned adjacent-frequency bundle.
- Step 3. Find a valid bundle.
- Assign the bundle tentatively in accordance with the priority list and count the number of spectrum-narrowing events of paths in all bundles involved.
- Iterate the above procedure until a valid bundle is found. The number of spectrum-narrowing events must be less than or equal to the limit number defined in Step 0.
- Step 4. Does a valid bundle exist?
- Step 4.5. Increase the target frequency slot number or fiber number.
- Step 6. Does an unassigned bundle exist?
In Step 0, the limit number of spectrum-narrowing events is determined by transmission simulations or experiments. Step 1 allows currently deployed WSSs to process quasi-Nyquist WDM signals; signals in a bundle are free from intra-bundle spectrum narrowing. Step 2, Step 3, and Step 4 jointly suppress the number of inter-bundle spectrum-narrowing events. Step 6 detects whether all of the traffic demands have been successfully accommodated in the network or not.
4. Network simulations
To evaluate the spectral efficiency of quasi-Nyquist WDM networks, we execute computer simulations of static-traffic scenarios; networks are built from scratch to accommodate the given traffic-demand set (i.e. traffic matrix) with the minimum number of fibers. Then, the resulting improvement in spectral efficiency is evaluated. The physical topology tested is an 18-node ring network. Each link comprises 100-km single-core single-mode fiber/s, which is consistent with the experimental assessments in Section 5. The frequency bandwidth of a fiber is 4.8 THz. All optical channels are assumed to be 32-Gbaud/400-Gbps dual-carrier DP-16QAM signals and path demands are given in this channel granularity through traffic/service aggregation in the electrical layer. The average number of path demands between each node pair, D, is changed from 4 to 20. A set of path demands are created based on the given traffic matrix that defines the number of traffic demands between each node pair. Two traffic-distribution patterns are examined. One is that the traffic demands are uniformly distributed. The other is that the traffic demands are distributed non-uniformly and one selected node has 5 times traffic larger than the other nodes. In both cases, the total number of traffic demands is same when the average number of traffic demands between each node pair is same. For example, the average number of fibers of each link is around 7 when D is 12. The maximum number of spectrum-narrowing events triggered by adjacent paths, Ns, is controlled. Table 2 compares the networking schemes examined. Scheme A corresponds to typical networks in which sufficient guardbands are inserted and thus the impact of spectrum narrowing is negligible. Scheme B enables denser WDM than Scheme A. Scheme C is our proposal; quasi-Nyquist WDM is attained in return for the wavelength-assignment restrictions due to the signal-spectrum narrowing and the WSS-passband resolution.
Table 2. Comparison of networking schemes
First, we analyze wavelength-assignment inflexibility in quasi-Nyquist WDM networks by means of spectral efficiency. The baseline is the spectral efficiency obtained with the ideal WSSs that have rectangular passbands and full frequency-setting flexibility. Figure 5 shows the spectral-efficiency penalty induced by limiting the number of spectrum-narrowing events, Ns, where we assume that spectrum narrowing is triggered (a) by adjacent-path add operation i.e. B&S nodes and (b) by adjacent-path add-drop operation i.e. R&S nodes. If Ns is limited to 2, the penalty is marginal in both cases; it is less than 0.2% when D is 20. When Ns is limited to 1, the impact of spectrum narrowing can be suppressed in the case of (a); in contrast, the penalty is over 10% if condition (b) is applied. In this way, thanks to the use of our impairment-aware wavelength assignment, the restriction caused by spectrum narrowing can be mitigated without any notable penalty in spectral efficiency even if the allowable number of spectrum-narrowing events is just 1 for the network using B&S nodes. If no adjacent filtering operation is permitted, the penalty is excessive in both cases and our method is of no avail. Figure 6 shows the spectral-efficiency penalty due to path bundling when the add/drop operation is performed (a) on a bundle/path basis i.e. B&S nodes and (b) on a bundle/bundle basis i.e. R&S nodes. In both cases, the penalty decreases as the number of traffic demands increases because path bundling is facilitated by increasing the number of traffic demands between node pairs. When D is 20, the penalty is less than 0.7% in case of (a); however, applying (b) results in a penalty of 4.7%.
Fig. 5 Spectral efficiency calculated as a function of the average number of wavelength path demands between each node pair, D. The acceptable maximum number of spectrum-narrowing events, Ns, is parameterized.
Fig. 6 Spectral efficiency calculated as a function of the average number of wavelength path demands between each node pair, D. The add/drop operation is performed (a) on a bundle/path basis and (b) on a bundle/bundle basis.
Finally, we evaluate the improvement in the spectral efficiency of quasi-Nyquist WDM networks. Figure 7 plots the spectral efficiency where the results are normalized by the counterparts of Scheme A shown in Table 2. The B&S node architecture outperforms the R&S node architecture for the same Ns. In the uniform-traffic scenario, when B&S nodes are utilized and Ns = 0, the improvement in spectral efficiency is 4.8% compared to Scheme A. In this case, Scheme C is not spectrally efficient compared to Scheme B whose spectral-efficiency improvement is 14.2%. Furthermore, Scheme C always yields shorter transmission distances than Scheme B due to its worse spectrum narrowing and larger cross-phase modulation. A similar argument is valid when R&S nodes are utilized and Ns ≤ 1. Consequently, Scheme C requires Ns to be larger than or equal to 1 for B&S-node-based networks and 2 for R&S-node-based networks so as to attain high spectral efficiency. When D = 20, the improvement in the B&S node network is 30.8% with Ns = 1; in the R&S node network, an improvement of 25.3% is observed if Ns = 2. We also observe the similar results for the centralized traffic distribution; the improvement in the B&S node network is 30.1% with Ns = 1 and that in the R&S node network is 25.4% if Ns = 2. In this way, the traffic distribution has virtually no impact on spectral efficiency in the large-traffic situations.
Fig. 7 Spectral efficiency calculated as a function of the average number of wavelength path demands between each node pair, D. Scheme A, Scheme B, and Scheme C are defined in Table 2.
5. Transmission experiments
We conducted transmission experiments to confirm the feasibility of the proposed network architecture. The experimental configuration is depicted in Fig. 8. The 4.8-THz bandwidth between 191.32500 THz and 196.12500 THz was fully utilized. At the transmitter side, six continuous waves (CWs) consecutively aligned in the frequency domain were generated by six independent tunable lasers (TLs). The frequencies were set at 193.54167-193.70833 THz with 33.3-GHz spacing. To create three 32-Gbaud/400-Gbps dual-carrier DP-16QAM signals, the CWs were modulated by a lithium-niobate IQ modulator (IQM) driven by a two-channel arbitrary-waveform generator (AWG) with eight-bit resolution; the electrical-signal spectrum was formed by a root-raised cosine filter with a roll-off factor of 0.01. Then, polarization-division multiplexing (PDM) was emulated by a polarization-beam splitter (PBS), 10-ns delay fiber, and polarization-beam combiner (PBC). On the other hand, 48 CWs aligned with 100-GHz spacing were generated from 48 TLs and combined in an optical coupler. CW intensity was modulated by a lithium-niobate intensity modulator (IM) driven by a 33.3-GHz cosine wave generated from a synthesizer; the IM bias was set to yield the carrier-residual condition. This yielded 144 CWs aligned with 33.3-GHz spacing. Next, 72-channel 400-Gbps dual-carrier DP-16QAM signals were created by an IQM and a PDM emulator. The target bundle consisting of three 400-Gbps dual-carrier DP-16QAM signals was added to the network by a WSS and multiplexed with the other wavelength signals, where the non-target signals having the same frequency with the added bundle were terminated by the WSS. The resulting 72-channel 400-Gbps dual-carrier DP-16QAM signals were aligned with 66.6-GHz spacing over the 4.8-THz bandwidth. After passing through an erbium-doped fiber amplifier (EDFA), the signals entered the link consisting of a 100-km standard single-mode fiber (SMF). The loss coefficient, dispersion parameter, and nonlinearity coefficient of the SMF were 0.18 dB/km, 16.5 ps/nm/km, and 1.5 /W/km, respectively. The noise figure of the EDFA was around 5 dB. The signals were amplified by an EDFA and delivered to the next node. Each node comprised a 1 × 9 splitter and 9 × 1 WSS in the B&S arrangement or a 1 × 9 WSS and 9 × 1 WSS in the R&S arrangement. The loss of each node was around 18 dB for the B&S architecture and 17 dB for the R&S architecture. The bandwidth resolution of the WSS was 12.5 GHz. After passing through two nodes, the signals were inserted into a recirculating fiber loop that comprised two synchronous loop-controlling switches (SW), a 2 × 2 splitter, a 100-km SMF, an EDFA, and a node. Here, the excess loss of 3.5 dB due to the loop-controlling switch and the 2 × 2 splitter was pre-compensated by the preceding EDFA so as to avoid excess SNR degradation. After multiple loops, the target signal was dropped and detected by a digital coherent receiver; here, the signal on 193.70833 THz was evaluated because both cross-phase modulation and spectrum narrowing were worst at this frequency. The detected signal was sampled at 50 Gsample/s by a four-channel analog-to-digital converter with eight-bit resolution. The digitized data was processed by a digital-signal-processing (DSP) circuit. The DSP circuit performs chromatic dispersion compensation, polarization division demultiplexing, phase and frequency estimation, and symbol decoding [20]. The clock recovery was performed based on adaptive filtering [21]. In our scheme, spectrum narrowing is induced at only one side of each signal spectrum because of path bundling and hence the clock can be extracted even if the single-side filtering is too tight. The target bit-error ratio (BER) was 2.7 × 10−2 as we assumed the use of forward error correction (FEC) [22]. The maximum number of spectrum-narrowing events triggered by adjacent paths, Ns, was controlled as follows. If the given WSS imposes spectrum narrowing on the target signal, we set the passband of WSS port #1 to 200 GHz and that of WSS port #2 to residual 4,600 GHz. Port #1 was used for the target bundle while port #2 was used for the non-target bundles. The 200-GHz passband causes spectrum narrowing at the edge channels of the target bundle. Here, the symbol timing skew between port #1 and port #2 was shifted with each WSS traversal. If the given WSS does not impose spectrum narrowing on the target signal, we set the passband of WSS port #1 to 4,800 GHz and no passband is assigned to the other WSS ports. Considering the worst case, spectrum narrowing was caused at the node/s closest to the source node when Ns was limited.
Fig. 8 Experimental setup for evaluating the performance of quasi-Nyquist WDM networks, where 72-channel 400-Gbps dual-carrier DP-16QAM signals were aligned with 66.6-GHz spacing over the entire C-band.
Figure 9 plots the BER characteristics as a function of transmission distance measured for each node architecture. The maximum transmittable distances are summarized in Table 3. Note that spectrum narrowing could occur at each traversed node, i.e. the possible Ns is equal to H-1 for the B&S nodes and 2(H-1) for the R&S nodes, where H is the hop count. In addition, B&S node networks suffer from spectrum narrowing at the source node whereas R&S node networks suffer at the source and destination nodes. Without limiting Ns, the transmission distance is severely bounded since the signals are impaired with each WSS traversal. The transmission distances with B&S nodes and R&S nodes were 400 km and 200 km, respectively. The use of small Ns suppresses the signal degradation induced by the spectrum narrowing and extends the maximum transmission distance. When Ns is 2/1/0, the transmission distances in B&S node networks are 600/900/900 km whereas those in R&S node networks are 500/900/1000 km. In both cases, Ns needs to be less than or equal to 1 for the longest path for the network examined in Section 4. Note that in general the required Ns depends on transmission characteristics such as transmission distance and hop count of each path. Here, the simulation results show that the improvements in spectral efficiency reach 30.8% for the B&S node network and 13.4% for the R&S node network when Ns = 1. Thus, the network using R&S nodes cannot enjoy the benefits of quasi-Nyquist WDM. The B&S node networks in conjunction with our restriction-aware wavelength-assignment algorithm can realize truly efficient quasi-Nyquist WDM networks without replacing currently deployed WSSs. The obtained spectral-efficiency improvement, 30.8%, is quite close to the theoretical limit of 31.3% elucidated in Subsection 2.1.
Fig. 9 BER characteristics for 400-Gbps dual-carrier DP-16QAM signals. Ns denotes the maximum number of spectrum-narrowing events triggered by adjacent paths.
Table 3. Transmittable distance for 400-Gbps signals
It should be noted that the allowable number of spectrum-narrowing events can be increased and the number of channels to be bundled would be changed if WSSs with finer frequency resolution and better filtering characteristic become cost-effective in the future. These attributes also facilitate the realization of quasi-Nyquist WDM networks.
6. Conclusion
We demonstrated a highly efficient quasi-Nyquist WDM network architecture that allows us to utilize widely deployed WSSs. Our wavelength-assignment algorithm can suppress the impact of restrictions imposed by the limited bandwidth granularity of WSS passbands and the significant spectrum narrowing occasioned by WSS traversal. Intensive network analyses based on computer simulations showed that our network architecture improves the spectral efficiency of 32-Gbaud/400-Gbps DP-16QAM networks by 30.8%. To confirm the feasibility of the proposed architecture, we conducted transmission experiments on 72-channel 400-Gbps dual-carrier DP-16QAM signals aligned with 66.6-GHz spacing in the full C-band. The fiber capacity reached 28.8 Tbps even after 900-km transmission and 9 node hops. Note that our proposed scheme can be applied to core mesh networks by using DP-QPSK signals, as they are much more tolerant of transmission impairments.
Funding
National Institute of Information and Communications Technology; Japan Society for the Promotion of Science KAKENHI (18K13756).
References
1. V. N. I. Cisco, “Forecast and methodology, 2017–2022” (2019).
2. T. Zami, I. Fernandez de Jauregui Ruiz, B. Lavigne, and A. Ghazisaeidi, “Growing impact of optical filtering in future WDM networks,” in Optical Fiber Communication Conference (Optical Society of America, 2019), paper M1A.6. [CrossRef]
3. 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, (Optical Society of America, 2014), paper Th1I.2. [CrossRef]
4. C. Pulikkaseril, L. A. Stewart, M. A. F. Roelens, G. W. Baxter, S. Poole, and S. Frisken, “Spectral modeling of channel band shapes in wavelength selective switches,” Opt. Express 19(9), 8458–8470 (2011). [CrossRef] [PubMed]
5. K. Kayano, S. Yamaoka, Y. Mori, H. Hasegawa, and K. Sato, “Highly dense elastic optical networks enabled by grouped routing with distance-adaptive modulation,” Photonic Tech. L. 31(4), 295–298 (2019). [CrossRef]
6. Y. Terada, Y. Mori, H. Hasegawa, and K. Sato, “Highly spectral efficient networks based on grouped optical path routing,” Opt. Express 24(6), 6213–6228 (2016). [CrossRef] [PubMed]
7. S. Kilmurray, T. Fehenberger, P. Bayvel, and R. Killey, “Comparison of the nonlinear transmission performance of quasi-Nyquist WDM and reduced guard interval OFDM,” Opt. Express 20(4), 4198–4205 (2012).
8. G. Bosco, V. Curri, A. Carena, P. Poggiolini, and F. Forghieri, “On the performance of Nyquist-WDM terabit superchannels based on PM-BPSK, PM-QPSK, PM-8QAM or PM-16QAM subcarriers,” J. Lit. Technol. 29(1), 53–61 (2011). [CrossRef]
9. D. Sinefeld, S. Ben-Ezra, and D. M. Marom, “Nyquist-WDM filter shaping with a high-resolution colorless photonic spectral processor,” Opt. Lett. 38(17), 3268–3271 (2013). [CrossRef] [PubMed]
10. 249025-ICT OASE Project, D4.2.1, “Technical assessment and comparison of next-generation optical access system concepts,” (2011).
11. R. Shiraki, Y. Mori, H. Hasegawa, and K. Sato, “Demonstration of quasi-Nyquist WDM networks using widely deployed wavelength-selective switches in Optical Fiber Communication Conference (Optical Society of America, 2019), paper M1A.7. [CrossRef]
12. Optical Internetworking Forum, “Technology options for 400G implementation,” OIF-Tech-Options-400G–01.0 (2015).
13. Optical Internetworking Forum, “Integrable tunable laser assembly multi source agreement,” OIF-ITLA-MSA-01.3 (2015).
14. ITU-T, “Spectral grids for WDM applications: DWDM frequency grid,” ITU-T Recommendation G.694.1 (2012).
15. J. Doucette, D. He, W. D. Grover, and O. Yang, “Algorithmic approaches for efficient enumeration of candidate p-cycles and capacitated p-cycle network design,” in Design of Reliable Communication Networks (IEEE, 2003), paper 212–220.
16. K. Kikuchi, “Fundamentals of coherent optical fiber communications,” J. Lit. Technol. 34(1), 157–179 (2016). [CrossRef]
17. S. L. Woodward, M. D. Feuer, and P. Palacharla, Optical Fiber Telecommunications 6th ed. (Academic, 2013), Chap. 15.
18. “N-degree ROADM architecture comparison: broadcast-and-select vs route-and-select in 120 Gb/s DP-QPSK transmission systems,” https://www.slideshare.net/ADVAOpticalNetworking/n-degree-roadm-architecture-comparison, accessed 10 May 2019.
19. R. Shiraki, S. Yamaoka, Y. Mori, H. Hasegawa, and K. Sato, “Novel network architecture enabling quasi-Nyquist wavelength-division multiplexing,” in Photonics West (Optical Society of America, 2019), paper 10946–11.
20. Y. Mori, C. Zhang, and K. Kikuchi, “Novel configuration of finite-impulse-response filters tolerant to carrier-phase fluctuations in digital coherent optical receivers for higher-order quadrature amplitude modulation signals,” Opt. Express 20(24), 26236–26251 (2012). [CrossRef] [PubMed]
21. K. Kikuchi, “Clock recovering characteristics of adaptive finite-impulse-response filters in digital coherent optical receivers,” Opt. Express 19(6), 5611–5619 (2011). [CrossRef] [PubMed]
22. D. Chang, F. Yu, Z. Xiao, N. Stojanovic, F. Hauske, Y. Cai, C. Xie, L. Li, X. Xu, and Q. Xiong, “LDPC convolutional codes using layered decoding algorithm for high speed coherent optical transmission,” in Optical Fiber Communication Conference (Optical Society of America, 2012), paper OW1H.4. [CrossRef]
