Open Access
Add to BibSonomyAbstract
We experimentally investigate various methods for reducing cross-phase modulation in hybrid networks with mixed 100G and 10G traffic. The experimental results over standard single-mode and non-zero dispersion-shifted fiber types demonstrate the effectiveness of several different XPM reduction techniques as well as the interplay between them. Nonlinear transmission performance is quantified using the Nonlinear Threshold metric as a function of key system features, including DCM type, dispersion map, spectral guard bands, and carrier phase estimation window size. Fiber Bragg grating-based DCMs are shown to offer a distinct advantage over fiber-based DCMs under certain conditions, particularly in dispersion-managed systems with very strong XPM. The average walk-off per span is introduced as a simple yet effective metric to compare different methods of XPM mitigation.
© 2013 Optical Society of America
1. Introduction
Polarization-multiplexed quadrature phase-shift keying (PM-QPSK) with coherent detection has gained wide acceptance as the optimum modulation format for increasing capacity in regional and long-haul networks employing 100Gb/s channel data rates on the 50GHz spectral grid. However, many existing networks are being upgraded gradually and thus need to support a “hybrid” scenario with existing 10G OOK-modulated channels co-propagating with 100G PM-QPSK channels. In this scenario performance can be significantly limited by fiber nonlinearities, and therefore simple, inexpensive methods are needed to reduce the impact of inter-channel nonlinearity, particularly cross-phase modulation (XPM).
The nature and impact of XPM in 100G systems have been studied previously, and several system design approaches have been proposed to reduce XPM. These include varying the dispersion map [1–4], increasing the spectral guard band between 100G and 10G channels [1,2,5], using alternative types of DCMs [4,6], and adding extra differential delay at intermediate nodes [7,8]. However, the experimental studies reported to date have not compared the efficiency of XPM suppression among the various available methods, or the performance impact when multiple techniques are combined in the same system. Of particular interest is the performance over a challenging but realistic network scenario: co-propagating 100G and 10G channels transported over widely-deployed low-dispersion fiber, commonly known as non-zero dispersion-shifted fiber (NZ-DSF), which presents greater sensitivity to nonlinear effects than standard single-mode fiber (SSMF), due to its smaller fiber dispersion and core effective area.
This paper focuses on practical and low-cost methods for reducing XPM via system design in hybrid 100G-10G networks, and demonstrates experimentally the impact of several system design options. We compare the use of recently-developed 50GHz fiber Bragg grating (FBG) based dispersion-compensation modules (DCMs) [9] with traditional dispersion compensating fiber (DCF) in a mixed 100G-10G system, and show that FBGs can provide significant benefits under certain conditions. We also investigate the nonlinear performance when using DCFs and FBGs as a function of several other system parameters, including residual dispersion per span (RDPS), local fiber dispersion, frequency of DCM placement, addition of spectral guard band, and carrier phase estimation window size in the receiver DSP.
2. Theoretical XPM analysis
Cross-phase modulation induces two distinct types of signal distortions: phase noise and polarization crosstalk. The latter is the most significant nonlinearity in polarization-multiplexed systems [1,6]. The worst-case XPM for a phase-modulated signal occurs if the neighboring channels are intensity-modulated, as in the hybrid 100G-10G systems which we explore in this work. Although this is the most challenging scenario for 100G deployment, the same general concepts for XPM reduction are also valid for other types of systems, including homogeneous 100G QPSK. In this paper we focus specifically on hybrid 100G-10G systems since these represent the cases which are most severely limited by XPM.
A useful analytical model for considering the impact of XPM is the “XPM filter,” H(ω), as described in [1]. For a multi-span, multi-channel system, the XPM filter is given as
where m is the channel index, n is the span index, IFT denotes the inverse Fourier transform, and the terms γ, α, L, and Δβ′ represent respectively the fiber nonlinear coefficient, fiber loss, span length, and difference in group velocity between interacting channels. Typically, , which means H(ω) is a low-pass filter with cutoff frequency fc = αn / Δβ′m,n, where the walk-off Δβ′m,n can also be represented as DnΔλm,n. A smaller fc reduces the total amount of XPM that passes the filter, which is desirable for improving system performance.The two types of XPM distortions can be expressed mathematically with the terms ϕx/y, which represents the XPM-induced phase noise for each polarization, and wyx/xy, which represents the XPM-induced polarization crosstalk. These two terms are given as
where u(z,t) is the dual-polarization signal, τm,n is the accumulated differential delay between channels, and denotes convolution [1]. The differential delay τm,n is a key parameter; a larger differential delay causes the XPM in each span to be summed up in random phase, producing a smaller total XPM impact.Using Eqs. (1)–(3), we can identify the XPM impact of several key system parameters. XPM can be reduced by increasing the accumulated differential delay, τm,n, or the walk-off, Δβ′m,n. In practice, the differential delay can be increased by eliminating or reducing the amount of in-line dispersion compensation, using DCMs which preserve the differential delay between channels, or introducing differential delay through additional devices or diverse paths at intermediate nodes. The walk-off may be increased by using fiber with larger local dispersion, using wider channel spacing, or adding a guard band.
In addition to these factors related to the optical system design, the receiver DSP algorithms can also play an important role in the impact of XPM. In this study we consider options which can be implemented in a real-time receiver and are available for our experimental study. In particular, the carrier phase estimation (CPE) averaging window size can improve tolerance to XPM-induced phase noise if it is optimized appropriately for a given system [10]. Smaller window sizes are more tolerant to the distortions induced by strong XPM, and are thus desirable in most hybrid systems with strong XPM. In contrast, larger window sizes are better in systems with weak nonlinearity and/or large cumulative dispersion.
3. Experimental configuration
A commercially-available MSA module is used to generate the 120 Gb/s NRZ-PM-QPSK test signal [11], which is tuned to 1543.73 nm, with framed PRBS 231-1 data. The five nearest neighboring channels on the 50 GHz grid on each side of the test channel are generated by independently modulated commercial 10G transponders, as shown in Fig. 1. The rest of the C-band is filled with DFB lasers modulated at 10.7 Gb/s, with the odd and even channels de-correlated. The launch power of all channels is uniform. The tests are performed in a conventional recirculating loop, by applying a gating signal to the 100G MSA module on an evaluation board. By disabling the blind CD estimation and applying the CD compensation values manually, the convergence time for the remaining DSP algorithms fits within the loop gating time. This allows for a sufficient time window to calculate the BER in real time for each block of gated data coming from the recirculating loop into the coherent receiver.
Fig. 1 Experimental configuration of 12 span recirculating loop system and illustrative example of two different TWRS dispersion maps used in the study.
The recirculating loop system, with 4 spans of fiber per loop and all-EDFA amplification is illustrated in Fig. 1. The fiber type is either SSMF (G.652) or TrueWave-RS, a type of NZ-DSF (G.655). Pre-compensation DCF with −360 ps/nm is used, as well as under-compensation every 1, 2, or 4 spans, via DCF or FBG. Input power to the DCF was kept sufficiently low to avoid nonlinearity in the DCF. The fiber Bragg gratings operate on the 50 GHz ITU-T grid with −0.5 dB bandwidth of 40 GHz [9]. We note a very small penalty on the 100G signal due to the FBGs, which is below 0.4 dB for all test cases (up to 12 devices), and may be due to passband narrowing and/or residual phase ripple originating from the channelized FBG. This small penalty is factored out of the analysis of nonlinear performance reported in the following sections. By varying the span lengths and frequency of DCM placement, a number of different CD maps are created, with average RDPS varying from ~10-80 ps/nm. We use RDPS according to the conventional definition, as the difference between the total fiber CD for each span and the CD compensation applied after each span. For cases with DCMs placed every span, this yields a small positive RDPS each span, whereas for cases with less frequent DCM placement this produces alternation between large positive and negative RDPS, which produces a small positive value for the average RDPS over the whole link. Two different CD maps are shown as illustrative examples in Fig. 1. All dispersion maps with optical compensation were chosen to allow for sufficiently low pre-FEC BER (≤10−3) to achieve post-FEC error-free operation (BER<10−15) on the 10G channels, assuming use of hard-decision FEC common in long-haul 10G DWDM systems. At the end of the link, a noise-loading stage is employed to vary the received OSNR. The same 100G transceiver module is used for coherent detection, digitization, and demodulation. The carrier phase estimation has variable averaging window size, which for the purposes of this study is always restricted to either 8 or 32. While larger or smaller window sizes may yield slightly better or worse performance in some cases, these two values are used for study since they present a good contrast between optimal performance in the linear and nonlinear regimes.
The 100G signal’s tolerance to nonlinearity is quantified using the Nonlinear Threshold (NLT) metric, defined here as the product of the number of spans and the launch power corresponding to a 1.0 dB OSNR penalty at a reference BER. The NLT parameter is a useful metric for quantifying nonlinear penalties independent of the link amplifiers and span losses. The 1 dB penalty point approximately corresponds to the maximum system margin, and thus provides a good balance between signal degradation due to ASE noise and nonlinear distortions. We choose reference BER = 10−2, which is close to the error correction threshold for latest generation Soft-Decision FEC used in commercial DWDM systems with 100G PM-QPSK transceivers. It should be noted that a lower reference BER level (e.g., 10−3, which corresponds to the threshold for common hard-decision FEC) will increase the OSNR penalties and the NLT differences measured and described below.
4. Results and discussion
4.1 TWRS results
Figure 2 shows the 100G Nonlinear Threshold for transmission over 12 spans of TWRS fiber, with in-line compensation via DCF or FBG placed every 2 spans, with varying average RDPS. An additional case with no in-line DCMs is also shown (dashed lines). In practice, this type of uncompensated system is not viable for networks supporting 10G traffic, without the added cost and complexity of DSP for the 10G channels. However, it is included here as a reference for “best-case” XPM performance.
Fig. 2 100G Nonlinear Threshold for 12 spans TWRS system, (a) with in-line DCMs every 2 spans, and (b) with added 50 GHz guard band. CPE window size = 8.
For the cases with DCF compensation, the XPM penalty can be reduced by ~1 dB by adding a 50 GHz guard band (i.e., leaving empty the adjacent 50 GHz channel slots on each side of the test channel). This can be noted by comparing Figs. 2(a) and 2(b). This benefit can be increased with a larger guard band, though eventually, with very large guard band, the incremental improvement from further increasing the guard band is severely diminished, approaching zero [5]. Increasing the walk-off via insertion of guard bands can be an effective means of reducing XPM, but is not an optimal solution since it directly reduces the overall system capacity and adds complexity to the wavelength management across the network.
Another important observation for the cases with DCF compensation in Fig. 2 is that the NLT increases with RDPS, asymptotically approaching the value for the uncompensated case. It is clear that avoiding a very low RDPS is beneficial for reducing XPM in a mixed 100G-10G system; however, changing the CD map is not always an option. It is also significant that for larger average RDPS values, there is almost no difference in nonlinear performance when compared with the uncompensated case (<0.5 dB difference in NLT). With no guard band, this condition is observed for average RDPS >~70 ps/nm, and with additional 50 GHz guard band this occurs for RDPS >~40 ps/nm. This demonstrates that a dispersion map with enough accumulated residual CD (or less frequent in-line compensation) sufficiently reduces XPM such that DCMs can be used without additional penalty compared to the uncompensated case. This enables a system that maintains good performance for the 10G channels with less detrimental impact on the 100G channel.
Next, the in-line DCFs were substituted with 50 GHz channelized fiber Bragg grating DCMs. FBGs introduce phase ripple, which causes a penalty for 10G signals, but this penalty is small enough for the number of FBGs employed in this study to allow error free operation on 10G channels with adequate operating margin. For coherent 100G signals, phase ripple penalty can be mitigated if the linear equalizer in the DSP has a sufficient number of taps [12]. Additionally, it has been shown previously that the use of a device similar to a 50 GHz channelized FBG can help reduce XPM in a direct-detection 10G/40G network [13].
Figure 2 shows that the use of FBGs instead of DCFs can produce an NLT improvement as large as 2 dB. An improvement is expected by theory and simulation [1,6], since the FBG device, unlike DCF, preserves the differential delay between channels, τm,n, while compensating the dispersion. The nonlinear performance with FBGs is almost constant across RDPS, and nearly equivalent to the uncompensated case. This means that the use of FBG-based DCMs can produce nearly the same 100G performance as an uncompensated system, with little to no dependence on the dispersion map, while at the same time providing a suitable environment for direct-detection 10G operation. This advantage is particularly significant in systems with close to 100% in-line compensation. Lack of performance dependence on RDPS also makes network deployment and maintenance much simpler in deployed systems where the exact RDPS values may not be known or readily measured, or in systems where changes in the average RDPS may occur over time.
Less frequent CD compensation is another system design option which can reduce XPM, since it allows the differential delay between channels to accumulate for a larger number of spans between compensation points. The impact of this technique is shown in Fig. 3. By comparing Fig. 3 with Fig. 2 we note that the NLT can be improved ~1 dB by compensating every 4 spans compared with compensating every 2 spans. Again, the XPM reduction benefit from compensating less frequently is also reduced when FBGs are used instead of DCF, or if the RDPS is large. Conversely, compensating more frequently (e.g., every span) will worsen XPM and the nonlinear performance will more strongly depend on other system parameters.
Fig. 3 100G Nonlinear Threshold for 12 spans TWRS system, (a) with in-line DCMs every 4 spans, and (b) with added 50 GHz guard band. CPE window size = 8.
The carrier phase estimation window size was initially set to 8 for all the results in Figs. 2 and 3. This smaller value is beneficial for hybrid systems with strong XPM. However, for a purely linear system, a larger value such as 32 yields higher performance. The impact of the CPE window size relative to other system parameters is quantified by measuring the NLT for window size of 32, for the same cases shown in Fig. 2 (DCMs placed every 2 spans). The results, plotted in Fig. 4, show that a window size of 32 instead of 8 can decrease the NLT by 1 dB or more in the worst case (very low RDPS). This difference is reduced in cases with larger RDPS and when FBGs are used instead of DCF (i.e., cases with weaker XPM).
Fig. 4 100G Nonlinear Threshold for 12 spans TWRS system, (a) with in-line DCMs every 2 spans, and (b) with added 50 GHz guard band. CPE window size = 32.
In systems with strong XPM (such as hybrid systems with low RDPS), replacing DCFs with FBG-based compensation offers a clear advantage; however, this benefit is reduced if other system conditions are present which reduce XPM, such as additional guard bands or large RDPS. This is clearly demonstrated in Fig. 2, where the benefit of FBGs over DCF is less prominent if a guard band is added, and in the cases with the largest RDPS, where the XPM reduction due to the CD map effectively eliminates the benefit of FBGs. Figures 2 and 3 show that large RDPS and/or less frequent in-line compensation and/or additional guard band(s) produce performance that is roughly equal, independent of the type of compensation: DCF, FBGs, or no in-line DCMs. Thus these various system design factors cannot be examined totally independently when analyzing nonlinearity or attempting to reduce XPM.
4.2 SSMF results
Next, SSMF fiber was studied under similar conditions, to determine the impact of the same system parameters on another common fiber type with different dispersion and nonlinearity profile. Figure 5 shows the NLT results for SSMF with DCMs placed every span (representing a typical CD map for a 10G system over SSMF), with CPE window sizes of 8 and 32. The higher local dispersion and larger core effective area of SSMF make it more resilient to XPM effects than TWRS. This advantage of SSMF is reflected by the higher maximum NLT for the uncompensated case, which is ~4 dB higher than for TWRS.
Fig. 5 100G Nonlinear Threshold for 12 spans SSMF system, (a) with in-line DCMs every span and CPE windows size = 8, (b) with added 50 GHz guard band, (c) with in-line DCMs every span and CPE window size = 32, (d) with added 50 GHz guard band.
However, in the case with SSMF we note a larger difference between the NLT with DCF and with no in-line DCMs; this is due to more frequent DCM placement. In contrast to TWRS, sparser placement of DCMs is not practical with SSMF, due to its larger amount of accumulated CD, which would require DCMs with very high negative CD, resulting in high DCM loss. For SSMF, the same NLT dependence on RDPS is seen, but the performance with DCF does not reach the same NLT as the uncompensated case, for RDPS up to 80 ps/nm. In this scenario with DCMs placed every span, in order to approximate the performance without DCMs the RDPS must be very large—possibly beyond the range of normal operation for direct-detect 10G channels. The benefit of FBGs remains the same, though it should be noted that in this case with DCMs every span, even the performance with FBGs is not quite equal to the uncompensated case when the RDPS is very low. This equivalence in performance between FBGs and no DCMs is achieved for RDPS greater than ~25 ps/nm. The larger CPE window size of 32 also shows the same characteristics noted previously: worse overall performance, and larger XPM penalty, particularly for the worst cases with low RDPS.
4.3 Discussion
Finally, we attempt to compare performance between multiple cases by calculating the walk-off due to the accumulated net dispersion at the input to each span, and averaging the walk-off across the number of spans in the link This average walk-off per span is defined here as DinΔλ, where Din is the average input dispersion at each span in ps/nm, and Δλ is the wavelength distance to the nearest neighbor—either 0.4 nm, corresponding to 50 GHz channel spacing, or 0.8 nm for the case with added 100 GHz guard band. Figure 6 shows the NLT, as an indicator of the overall nonlinear performance, plotted against the average walk-off per span, as described above. The average walk-off per span depends on two features of the dispersion map—average RDPS and frequency of DCM placement—as well as the spacing of the nearest neighbor channels. Larger walk-off improves the nonlinear performance up to the limit determined by a given fiber type, which is represented by the uncompensated case (dashed lines), as noted earlier. For a given fiber type, nonlinear effects can be reduced to the same best-case level using different mitigation techniques which yield the same average walk-off per span. For both SSMF and TWRS, the NLT with DCMs is roughly equivalent to the NLT without DCMs (or with FBGs) when the average walk-off per span is >~400 ps. This type of generalized walk-off assessment offers the flexibility to implement the simplest or lowest-cost solution when multiple means for XPM reduction exist, in order to achieve the same performance benefit. The results in Fig. 6 also indicate that the combined effectiveness of multiple methods for XPM reduction can be quantified when assessing and comparing nonlinear performance across multiple system configurations.
It should be noted that this walk-off analysis is approximate because it does not account for all the XPM interactions with more distant 10G channels, but it provides a good estimate of the total XPM impact in most cases studied here. The exception occurs for very low walk-off values (<100 ps), noted particularly for the SSMF CD map with lowest RDPS and DCMs placed every span (average walk-off per span = ~10-20 ps). In this case the impact of more distant neighbors is strong due to the tighter CD map, but this is not accounted for in the walk-off calculation, causing some deviation in the NLT trend from the rest of the cases with larger walk-off. Further experimental results and expansion of the total walk-off calculation are needed to verify the application of this analysis to other possible system configurations, including addition of differential delay at intermediate nodes as in [7,8].
5. Conclusions
We have shown experimentally that the use of fiber Bragg grating DCMs can significantly reduce XPM in hybrid 100G-10G networks with in-line optical dispersion management. However, the performance difference between DCF and FBG is diminished or eliminated if other conditions exist which reduce the total system XPM. Nonlinear performance of 100G in hybrid systems also depends on the dispersion map, guard band size, and carrier phase estimation window size. The various methods for reducing XPM are not necessarily independent and additive; thus, a comprehensive understanding of 10G and 100G performance as well as all system parameters is needed to determine whether a certain XPM reduction technique can be effective. The average walk-off per span is shown to be a useful tool to compare the efficacy of different optical methods of XPM reduction, and possibly to predict the XPM impact in cases where multiple mitigation techniques are simultaneously employed. While this paper focused on 100G-10G systems as the most challenging scenario for 100G deployment, the same general concepts for XPM reduction can also be applied to other systems with mixed modulation formats, or even 100G-only transmission, though further study is needed to determine the effectiveness of each technique for a given type of system.
References and links
1. Z. Tao, W. Yan, L. Liu, L. Li, S. Oda, T. Hoshida, and J. C. Rasmussen, “Simple fiber model for determination of XPM effects,” J. Lightwave Technol. 29(7), 974–986 (2011). [CrossRef]
2. X. Liu and S. Chandrasekhar, “Suppression of XPM penalty on 40-Gb/s DQPSK resulting from 10-Gb/s OOK channels by dispersion management,” in Proceedings of Optical Fiber Communications Conference (San Diego, Calif., 2008), paper OMQ6. [CrossRef]
3. C. Xia, J. F. Pina, A. G. Striegler, and D. van den Borne, “On the nonlinear threshold of polarization-multiplexed QPSK transmission with different dispersion maps,” in Proceedings of European Conference on Optical Communications (Geneva, Switzerland, 2011), paper We.10.P1.72. [CrossRef]
4. S. Searcy and S. Tibuleac, “System design tradeoffs for XPM mitigation in hybrid 100G-10G networks,” in Proceedings of Optical Fiber Communications Conference (Anaheim, Calif., 2013), paper NW4E.3. [CrossRef]
5. O. Bertran-Pardo, J. Renaudier, G. Charlet, H. Mardoyan, P. Tran, M. Salsi, and S. Bigo, “Overlaying 10 Gb/s legacy optical networks with 40 and 100 Gb/s coherent terminals,” J. Lightwave Technol. 30(14), 2367–2375 (2012). [CrossRef]
6. C. Xie, “Suppression of interchannel nonlinearities in WDM coherent PDM-QPSK systems using periodic-group-delay dispersion compensators,” in Proceedings of European Conference on Optical Communications (Vienna, Austria, 2009), paper P4.08.
7. D. Sperti, P. Serena, and A. Bononi, “Optical solutions to improve PDM-QPSK resilience against cross-channel nonlinearities: a comparison,” IEEE Photonics Technol. Lett. 23(11), 667–669 (2011). [CrossRef]
8. O. Vassilieva, T. Hoshida, K. Croussore, I. Kim, and T. Naito, “Suppression of XPM penalty in dispersion managed hybrid 10G/40G/100G DWDM networks using group delay management,” in Proceedings of European Conference on Optical Communications (Vienna, Austria, 2009), paper P4.04.
9. G. Brochu, M. Morin, F. Trépanier, B. Maheux-Lacroix, C. Paquet, A. Patel, M. Filer, C. Meyer, and S. Tibuleac, “Channelized fiber Bragg gratings for inline chromatic dispersion compensation of 40 Gb/s links on ITU-50 grid,” in Proceedings of European Conference on Optical Communications (Geneva, Switzerland, 2011), paper Th.12.B.3. [CrossRef]
10. M. S. Alfiad, D. van den Borne, T. Wuth, M. Kuschnerov, B. Lankl, C. Weiske, E. de Man, A. Napoli, and H. de Waardt, “111-Gb/s POLMUX-RZ-DQPSK transmission over 1140 km of SSMF with 10.7-Gb/s NRZ-OOK neighbours,” in Proceedings of European Conference on Optical Communications (Brussels, Belgium, 2008), paper Mo.4.E.2. [CrossRef]
11. L. E. Nelson, G. Zhang, M. Birk, C. Skolnick, R. Isaac, Y. Pan, C. Rasmussen, G. Pendock, and B. Mikkelsen, “A robust real-time 100G transceiver with soft-decision forward error correction,” J. Opt. Commun. Networking 4(11), B131–B141 (2012). [CrossRef]
12. E. Tipsuwannakul, J. Li, T. A. Eriksson, F. Sjöström, J. Pejnefors, P. A. Andrekson, and M. Karlsson, “Mitigation of fiber Bragg grating-induced group-delay ripple in 112 Gbit/s DP-QPSK coherent systems,” in Proceedings of Optical Fiber Communications Conference (Los Angeles, Calif., 2013), paper JW2A.69.
13. X. Liu, X. Wei, J. Ying, and D. A. Fishman, “Scalable dispersion management for hybrid 10-Gb/s and 40-Gb/s DWDM transmissions with high nonlinear tolerance,” IEEE Photonics Technol. Lett. 17(9), 1980–1982 (2005). [CrossRef]
