Open Access
Add to BibSonomy
Abstract
We address the potential application of G.654.C optical fiber for O-band transmission in the wavelength range of 1270 nm to 1330 nm. Fiber samples at the extreme upper end of the cable cutoff manufacturing distribution are chosen for modeling and experimentation. Modeling of multipath interference (MPI) generation in bend conditions representative of cable deployment suggests minimal to negligible penalty and transmission experiments at 100 Gb/s and 400 Gb/s with commercial IMDD transceivers demonstrate longer transmission with increased power margin compared to standard G.652 fiber due to lower O-band attenuation and no adverse impacts from MPI.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Optical links inside hyperscale data centers (DCs) are still largely implemented with intensity-modulated direct detection (IMDD) transceivers operating in the O-band in the 1300 nm region even as data rates climb to 400 Gb/s and higher [1]. Such intra-DC links are typically on the order of 2 km or less. However, longer reach versions of O-band IMDD transceivers are also used for DC interconnections (DCI) and metro transmission for links ranging from 10 km (e.g., 100GBASE-LR4) up to 80 km (e.g., 100GBASE-ZR4). These transceivers are now available for data rates of 100 Gb/s and 400 Gb/s. To date, the optical fiber deployed for O-band transmission in both intra-DC and DCI/metro applications is primarily G.652-compliant standard single-mode fiber having cable cutoff (CC) specified as less than 1260 nm [2]. Cable cutoff below 1260 nm essentially ensures that O-band transmission will be purely single-mode, given that O-band transceivers generally have laser wavelengths in the 1270-1330 nm range. The channel spacing between wavelengths depends on the transceiver configuration and may be up to 20 nm for coarse wavelength division multiplexing (CWDM) but may be less than that for other schemes, e.g., medium WDM (MWDM) with more wavelengths [3].
G.652-compliant optical fibers are commonly manufactured with Germanium-doped cores and may have attenuation levels at 1310 nm and 1550 nm of about 0.35 dB/km and 0.20 dB/km. However, silica-core fibers with nominally the same 1550 nm effective area (∼80 µm2) and zero-dispersion wavelength (∼1310 nm) but much lower attenuation (<0.16 dB/km at 1550 nm) are now widely deployed in terrestrial long-haul systems for C-band transmission. The ultra-low loss of silica-core fibers can offer significantly longer reach at 1550 of more than 30% [4]. Furthermore, the significant attenuation advantage of silica-core fibers compared to Ge-doped fibers also applies to the O-band and can similarly enable longer system reach or greater power margin because of lower O-band attenuation. One example of a silica-core fiber with many comparable characteristics to standard single-mode fibers is G.654.C-compliant Corning SMF-28 ULL optical fiber. However, besides having much lower attenuation, another important distinction of SMF-28 ULL fiber is implied by the G.654.C standard designation, namely that the G.654.C standard provides an upper CC limit of 1530 nm [5]. As a result of this higher CC allowance, ultra-low loss G.654.C fibers have not been historically used in the O-band where the possibility of MPI impairments might be perceived to outweigh potential transmission benefits from significantly lower attenuation like those in the C-band.
Here, we investigate the application of O-band transmission over this G.654.C fiber by evaluating the potential for MPI generation and its impact on transmission system performance. This work is an expansion of [6] including greater details and new experimental transmission data. We demonstrate the feasibility of O-band transmission over the highest cutoff SMF-28 ULL fiber available when deployed in bend conditions similar to terrestrial outside plant cables. We begin by modeling the LP11 higher order mode attenuation for several bend diameters including those corresponding to maximum diameter conditions in loose-tube cables. We then calculate expected levels of MPI for transmission in DCI/metro link lengths accounting for splice losses, splice lengths, fiber LP11 mode attenuation in the cabled fiber and in splice trays, and estimated mode coupling strength. The calculations are based on an established approach for evaluating MPI from propagation in few-mode fibers [7–11]. The modeling predicts negligible MPI penalty for IMDD signals for transmission in expected cable conditions, even for links comprised solely from fibers with the highest cutoff. We obtain further evidence of the suitability of O-band transmission over this fiber from transmission experiments conducted with commercial O-band transceivers operating at 100 Gb/s (100 G) and 400 Gb/s (400 G), for which no penalty from MPI is observed and successful error-free transmission is accomplished with fiber lengths at least 25% longer than the transceiver specifications.
2. Fiber selection and higher order mode attenuation behavior
For this study, we selected several spools of SMF-28 ULL fiber with measured CC values at the extreme upper end of the manufacturing distribution to try and assess fiber samples potentially having the greatest impact on MPI generation and transmission performance. While the G.654.C standard allows CC up to 1530 nm, the manufacturer’s specified maximum CC is 1520 nm. Five spools were chosen, each with measured CC greater than 1450 nm and in the 99th percentile of the CC distribution or higher. For context, the average CC in the manufacturing distribution is generally in the range 1300-1320 nm. A histogram of the CC distribution for one recent 3-month manufacturing period is shown in Fig. 1.
To facilitate the modeling of potential MPI generation for O-band transmission in the high cutoff G.654.C fiber, we first measured the refractive index profiles of the fiber samples. The index profiles were modeled with a beam propagation method [12] to calculate the LP11 higher order mode attenuation as a function of wavelength over a wide range encompassing the O-band. The impact of fiber bending is modeled by transforming a curved fiber into a straight one using a refractive index profile conformal mapping approach [13]. The higher order mode attenuation is a key parameter for estimating MPI growth for transmission in a few-mode regime below CC. The LP11 attenuation was calculated for several bend conditions including straight line, and bend diameters of 80, 165, and 190 mm. The 80 mm diameter corresponds to the approximate loop diameter experienced in splice trays on either side of splices, and the 165- and 190-mm bend diameters correspond approximately to maximum fiber bend conditions expected in different loose-tube cable configurations. The LP11 attenuation results are given in Fig. 2. These values represent the differential mode attenuation (DMA) of the higher order mode relative to the fundamental mode. The attenuation results for the cable deployment bend diameters are given in units of dB/km and shown for a range 1250-1350 nm, encompassing the O-band transmission range. The LP11 attenuation for the 80 mm diameter is expressed in units of dB/loop (full circle) as is relevant to the configuration in a splice tray. We also showed this attenuation over a much wider range up to the maximum wavelength modeled of 1500 nm because this illustrates the modeled prediction of the CC values where the attenuation data cross the dashed line representing the attenuation of a single 80 mm loop in a CC measurement [14]. The results in Fig. 2(a) show that the LP11 mode attenuation is at least 20 dB/km at 1270 nm for any of the fiber samples and increasing with wavelength, while the results in Fig. 2(b) show that the model predictions for the CC values are all greater than 1450 nm, consistent with measured CC values. The predicted CC for two of the samples was above 1500 nm. The LP11 mode attenuation in the O-band range is on the order of 0.5 dB/loop and below.
Fig. 2. Differential higher order mode LP11 attenuation as a function of wavelength for high-cutoff G.654.C fiber samples. (a) 165 mm and 190 mm bend diameters, (b) 80 mm diameter.
3. MPI generation modeling for a single span link and the system impact
To model potential MPI levels generated, and thus potential DCI/metro system impacts for O-band transmission in the high CC samples identified, we considered the single span system illustrated schematically in Fig. 3. Typical link lengths for this system type are up to 80 km, and terrestrial system deployments usually have splice points in the cable on the order of every 4-8 km. We chose a 5 km splice length for this study. The splices are assumed to be situated in splice trays with several 80 mm diameter loops of fiber on each side of the splice. For this study, we assumed 4 loops are used on each side, although the results were not found to be sensitive to this number. The splice points are one source of MPI as coupling between the fundamental and higher order modes can occur here. Taking a conservative approach, we assumed that any splice loss Lsplice (dB) of the fundamental mode results in the full transfer of the lost power to the higher order mode with coupling coefficient $\varepsilon = 1 - {10^{( - {L_{splice}}(dB)/10)}}$ and that any power already propagating in the LP11 mode couples back to the fundamental mode with the same coefficient [8–10]. We assumed average splice losses of 0.05 dB at each intra-span splice point as well as for splices of the G.654.C fiber to G.652 fiber at the transmitter and receiver ends of the link, with a small random variation of the loss around the average. In general, it is the coupling of light from the fundamental mode out to the higher order mode and then later coupling back to the fundamental mode after a relative propagation delay that creates MPI. As mentioned earlier, the average fiber bend diameter in the cable was modeled as 165 mm (82.5 mm radius) or 190 mm (95 mm radius).
Besides mode coupling at the splice points, distributed mode coupling can occur during propagation in the transmission fiber. To account for this MPI source, we use an established model of MPI generation from transmission below CC in a few-mode fiber [7–11]. The key parameters governing MPI are the differential mode attenuation (DMA) expressed here as the LP11 attenuation and the power coupling coefficient expected during propagation. In the regime where the DMA is high according to the relationship
where Δα is the DMA in linear units km-1 and κ is the power coupling coefficient in the same units, then the MPI generated during transmission over fiber length L can be estimated as and MPI is defined here as the ratio of the total power of the interfering terms (expressed as a summation here) to the average signal power in the fundamental mode at the span end, orAs Eq. (2) shows, the two main fiber parameters needed to estimate the level of MPI generated from distributed coupling during propagation are the DMA and the coupling coefficient. Since we can model the DMA as a function of wavelength from the index profiles as in Fig. 2, we can find an estimate for κ by making MPI measurements over known span lengths for fiber in a given bend diameter and then using Eq. (2). To that end, we experimentally measured MPI for two different span lengths of 63 km and 118 km constructed from several of the fiber samples selected by analyzing the power fluctuations of a CW laser transmitted through the spans over a range of wavelengths [10,15,16]. The measurements were made with the G.654.C fibers wound on reels with inner diameter of about 165 mm. The results of the MPI measurements are shown in Fig. 4(a) including back-to-back measurement data that sets a floor for the measurement system. The floor is due to several factors including laser intensity noise and polarization-dependent loss (PDL) of a polarization scrambler and other components. The laser linewidth was 100 kHz, the polarization rotation speed was 8,640 degrees/sec, and the MPI was calculated over 100,000 power measurements. While there is clearly scatter in the data, the measured MPI levels are generally higher for the longer span as expected. Since the outer diameter of fiber on the reels was greater than 165 mm, we used an average of the DMA values calculated for 165- and 190-mm diameters to estimate the coupling coefficient from Eq. (2). The results from those estimates are given in Fig. 4(b). The upper and lower boundaries of the shaded region in Fig. 4(b) are the average coupling coefficient values obtained for the two span lengths, with the higher one being about 4 × 10−3 km-1.
Fig. 4. (a) Measured MPI as a function of wavelength for back-to-back, 63 km span, and 118 km span. (b) Estimated power coupling coefficient for the two tested span lengths.
Using a distributed coupling coefficient of 4 × 10−3 km-1, we then calculated the expected MPI for different conditions and configurations. MPI was calculated including both discrete coupling events at splice points and distributed coupling during propagation according to the model as described in [7,8,11]. The first results given in Fig. 5(a) correspond to a modeled span of length 80.7 km comprised of three of the fiber samples. The average CC of the span was 1472 nm. The results for four cases are shown in the figure. The first case corresponds to a hypothetical straight-line deployment with splices but without 80 mm diameter fiber loops surrounding the splices. This is by far the worst case and unrealistic in nature. The second case is straight-line but including the effects of 4 fiber loops on each side of the splices to determine their effect. The third and fourth cases are for the realistic configurations with average bend diameters of 165 mm and 190 mm, respectively, experienced by the fiber as deployed in loose tube cables, and including the fiber loops in splice trays. The results show that MPI is greatly reduced with the average bend diameters of cable deployment, and the maximum MPI for the 190 mm bend diameter at 1270 nm is about -34 dB for the 80.7 km span. For comparison, we also modeled an 80 km span comprised solely of a fiber with one of the highest CC values in the sample set of about 1500 nm. The results for that span are given in Fig. 5(b) for the same cases and show only very minor differences from the span comprised of 3 fibers in Fig. 5(a). For this span, the maximum MPI at 1270 nm is -33 dB for the 190 mm bend diameter.
Fig. 5. MPI per span as a function of wavelength. (a) 80.7 km span comprised of 3 spools of fiber with average CC 1472 nm. (b) 80 km span comprised solely of fiber with CC ∼1500 nm.
The system impact of MPI for IMDD systems can be estimated in terms of power penalty and depends on the modulation format and the target bit error ratio (BER). The two most common modulation formats used for O-band IMDD transmission systems today are non-return to zero (NRZ) and 4-level pulse amplitude modulation (PAM-4). Power penalty results for those two formats [17–20] with a target BER of 1 × 10−6 (Q = 13.5 dB) assuming random polarizations of the interfering terms and optimized decision thresholds are shown in Fig. 6 according to [18, Eq. (10), 20, Eq. (5)] for NRZ, and [20, Eq. (7)] for PAM-4. Given that the highest MPI level from Fig. 5(b) was about -33 dB for an 80 km span, that would result in maximum power penalties of less than 0.2 dB and 0.03 dB for PAM-4 and NRZ signals, respectively.
Fig. 6. Predicted power penalties for PAM-4 and NRZ signals as a function of total MPI level. Target BER = 1 × 10−6 and random polarizations are assumed for interfering terms.
4. Transmission experiments and results
The modeling results of the previous section suggest that even if a DCI/metro link is constructed completely of the SMF-28 ULL G.654.C fiber with CC at the upper end of the manufacturing distribution, minimal or negligible power penalty is expected due to the high attenuation of the LP11 mode in the bend conditions of a cable deployment. Here we perform transmission experiments with commercial O-band transceivers over links made from these high cutoff fiber samples to investigate the performance. The spans constructed were deployed on shipping reels with inner (minimum) winding diameter of about 165 mm, so the bend condition is on the same order as expected in loose tube cables in the range of 165-190 mm. The general test set-up is illustrated schematically in Fig. 7. Pluggable QSFP28 or QSFP-DD transceivers operating at either 100 Gb/s or 400 Gb/s were plugged into an optical network tester (ONT) to transmit and detect internally generated signals. All digital signal processing (DSP) and forward error correction (FEC) were performed in the transceiver modules. G.652-compliant optical jumpers were connected to the transceiver with LC connectors. The fiber under test was either the G.654.C fiber with high cable cutoff or standard G.652 fiber, and individual reels of fiber were spliced to each other and to optical jumpers at each end. The output of the fiber under test was directed to a variable optical attenuator (VOA) so that the received power at the receiver could be varied and controlled. FC/APC connectors were used for the connections into and out of the VOA.
Fig. 7. Test set-up for experimental transmission tests through the G.654.C fiber with either 100 G or 400 G O-band transceivers.
The first transceiver tested was a 100 G transceiver rated for 40 km transmission, type QSFP28-ER4 manufactured by QSFPTEK. Four wavelengths modulated with the NRZ format at 25.78 Gb/s per wavelength were transmitted in the wavelength range 1296-1310 from the transceiver. Raw BER data (pre-FEC) was obtained in the Viavi ONT (model ONT-603) as a function of received power for back-to-back transmission, over 80 km of the G.654.C fiber, and over 69.5 km of standard G.652 fiber. The results are given in Fig. 8 showing that transmission over both fibers demonstrated better performance than in back-to-back, perhaps due to favorable interaction of fiber dispersion in the O-band with laser chirp [21] or to optimization of receiver equalization digital signal processing (DSP) that is designed to work best for fiber transmission. The signals were error-free post-FEC for all data points in Fig. 8. The results also demonstrate no evidence of any adverse effect or penalty from MPI from the G.654.C fiber transmission, and 80 km transmission was achieved with more than 6 dB power margin above the FEC threshold. For comparison, the G.652 fiber span had about 1 dB higher loss while being 10.5 km shorter than the G.654.C SMF-28 ULL fiber span.
The next transmission experiments conducted using the same ONT were with another 100 G O-band transciever, this one rated for 80 km link length. This transceiver was a QSFPTEK QSFP28-ZR4 pluggable module. This unit also employed four 25.78 Gb/s NRZ signals transmitted over four wavelengths ranging from 1296-1310 nm. The transceiver was tested over link lengths of 99 km of the G.654.C fiber and 85 km of G.652 fiber and these pre-FEC BER results are shown in Fig. 9 along with those in the back-to-back condition. The G.654.C link was comprised of four reels of fiber involving 5 splices, with no fiber loops around the splices. We again find no penalty or impact from MPI in the comparison to the G.652 fiber transmission, and both fibers again displayed better performance than back-to-back transmission. The G.654.C SMF-28 ULL link had a total loss of about 28.2 dB, and still had 7.2 dB of power margin at this distance.
The third O-band transceiver tested for transmission over the high cutoff G.654.C fiber was a 400 G QSFP-DD-ER8 pluggable module. The ONT used for these measurements was a Viavi ONT-804. This transceiver has 8 wavelengths, each modulated with PAM-4 signals at about 26.56 Gbaud for a total data rate of 400 Gb/s. The transceiver is rated for 40 km transmission. The transmitted optical spectrum in Fig. 10 shows the 8-wavelength channel plan, ranging from about 1273 nm to 1309 nm.
We transmitted this 400 G signal over link lengths of 50 km and 40 km for the G.654.C and G.652 fibers, respectively. The 50 km G.654.C span was comprised of two fiber reels and three splices with no fiber loops. The results for overall pre-FEC BER as a function of received power are given in Fig. 11 along with back-to-back results. These overall BER results represent the average pre-FEC BER over the 8 wavelength channels. All signals were error-free after FEC (KP4, BER threshold of ∼2.2 × 10−4). We observe that transmission over both fiber types again exhibits a small negative power penalty at least for lower received power levels. The G.654.C SMF-28 ULL fiber transmission shows no MPI impact again, even though PAM-4 signals are more sensitive to this impairment as conveyed in Fig. 6. This strongly suggests that no significant MPI was generated during transmission over the G.654.C fiber. Relative to the minimum received power value with error-free detection after FEC, there was still 7.6 dB of power margin over this fiber span.
We also explored the transmission performance over the 50 km G.654.C fiber span in more detail by measuring the pre-FEC BER of each individual wavelength channel in the ONT. This was meant to assess if there were any significant differences in performance between channels that might suggest MPI effects. The results of this study are provided in Fig. 12 in terms of the maximum, average, and minimum BER values over the set of 8 wavelengths as a function of received power for the fiber transmission and in back-to-back (B-to-B). While there was a significant spread in performance over the channels for both fiber transmission and B-to-B, the results are very consistent in that the 50 km span showed slightly better performance (negative power penalty) for both the maximum BER channel and minimum BER channel, as well as the average over most of the received power range. As a last measure of transmission performance over the 50 km span of G.654.C fiber, we continuously monitored the real-time data in the ONT for over 100 hours at the maximum received power level. The received signal was confirmed to be error-free for the full period.
Fig. 12. Experimental transmission results for the QT-QSFP-DD-ER8 transceiver with results for maximum, minimum, and average BER over the 8 wavelength set.
Finally, we measured the PAM-4 eye diagrams for each wavelength as a qualitative performance measure. This was done for transmission over both fiber types and in B-to-B. We used a real-time oscilloscope (Tektronix DPO73304D) to acquire the transmitted signals as illustrated in Fig. 13. A Viavi optical network tester (ONT-804) was used to drive the 400 G ER8 transceiver and feed the transmitter signals into the fiber under test (FUT). The optical receiver used was a Thorlabs RXM42AF single mode ultrafast receiver with 42 GHz bandwidth. The real-time scope sampled the data at the rate of 100 GS/s. The optical data baud rate is around 26.5625 Gb/s corresponding to an optical data rate around 56.125 Gb/s. We were able to turn on one wavelength at a time from the ONT to acquire the detected output data. We set a 21 GHz bandwidth limited filter in the real-time scope, which optimized the displayed eye diagram. Using tools in the communications toolbox of MATLAB, we constructed the eye diagrams from the acquired data. The benefit of using a real-time scope in the configuration shown in Fig. 13 is that it avoids splitting a portion of the light out to drive a clock recovery unit. This was especially important for obtaining eye diagrams after 40-50 km of transmission since the optical power reaching the optical receiver is low at those distances at around -12.5 dBm per wavelength. In Fig. 14, we show the eye diagrams obtained from the B-to-B condition for channel 1 associated with the lowest wavelength and channel 5. The PAM4 eyes are wide open indicating good transmission quality. The eye diagrams from other channels are very similar in the B-to-B configuration. In Figs. 15 and 16, we show the eye diagrams of every other channel, i.e., Channel 1, 3, 5 and 7, to illustrate the transmission quality after transmitting over 40 km of the G.652 fiber and 50 km of the SMF-28 ULL fiber, respectively. The eye diagrams still show reasonable quality after transmitting over these long distances. The eye diagrams from the G.654.C SMF-28 ULL fiber after 50 km transmission appear similar or slightly better than those after 40 km of the G.652 fiber. The fact that the received eye diagrams after 40-50 km appear somewhat degraded from the B-to-B condition may be partly due to dispersion effects as well as low received power. However, as observed in Fig. 11 the BER after transmission over both fibers was generally lower than, or comparable to, B-to-B and this might be due to equalization DSP applied in the receiver that is optimized for fiber transmission.
Fig. 13. The experimental setup using a real-time scope to acquire received signals to reconstruct eye diagrams.
Fig. 15. Eye diagrams for transmission over 40 km of G.652 fiber. (a) Channel 1. (b) Channel 3. (c) Channel 5. (d) Channel 7.
Fig. 16. Eye diagrams for transmission over 50 km of G.654.C SMF-28 ULL fiber. (a) Channel 1. (b) Channel 3. (c) Channel 5. (d) Channel 7.
5. Conclusions
We have investigated the feasibility of O-band DCI transmission over ultra-low loss G.654.C optical fiber. We identified and obtained fiber samples of SMF-28 ULL fiber at the extreme upper end of the manufacturing distribution for cable cutoff. We calculated the higher order mode attenuation for wavelengths in the O-band in various bend conditions and used this data to model the expected MPI generation in DCI link lengths for bend diameters representative of deployment in loose tube cables. The modeled MPI generation even for a full link comprised of only the highest cutoff fiber was sufficiently low to predict negligible power penalty. Of course, real links in the field would be comprised of many fiber sections with a range of different cable cutoff values so real deployments should have much better characteristics, i.e., lower MPI. We then performed experimental transmission measurements of 100 G and 400 G O-band commercial transceivers over the high cutoff G.654.C fiber and found no trace of any MPI impairment. However, the ultra-low attenuation of the SMF-28 ULL fiber allowed longer span lengths with increased power margin compared to standard G.652 fiber. These results support the use case of such ultra-low attenuation G.654.C fiber for DCI applications with O-band transmission beyond the usual application space of C- and L-band transmission systems.
Acknowledgments
The authors thank Mark Gray, Duane Robbins, and Gabe Sudduth for helpful suggestions and assistance.
Corning and SMF-28 are registered trademarks of Corning, Incorporated, Corning, NY.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. H. Isono, “Latest standardization trend for next-gen high speed optical transceivers,” Proc. SPIE 12027, 26 (2022). [CrossRef]
2. Recommendation ITU-T G.652, “Characteristics of a single-mode optical fibre and cable” (International Telecommunication Union, 2016).
3. X. Chen, K. Li, R. A. McCool, H. Chen, H. Dong, S. M. Patel, J. E. Hurley, J. S. Stone, S. Bickham, and M.-J. Li, “Feasibility of 25Gb/s MWDM Transmission Over a 15-km G652. D Compliant Fiber for 5 G Fronthaul Networks,” in Opto-Electronics and Communications Conference, 1–3 (2021).
4. J. Downie, J. Hurley, D. Pikula, S. Ten, and C. Towery, “Study of EDFA and Raman system transmission reach with 256 Gb/s PM-16QAM signals over three optical fibers with 100 km spans,” Opt. Express 21(14), 17372–17378 (2013). [CrossRef]
5. Recommendation ITU-T G.654, “Characteristics of a cut-off shifted single-mode optical fibre and cable” (International Telecommunication Union, 2020).
6. J. D. Downie, P. Sterlingov, J. Hurley, H. D. Pedro, X. Chen, and D. Seddon, “O-band Transmission over High-Cutoff G.654.C Fiber,” Optical Fiber Communications Conference and Exhibition (2023), paper Th3F.7.
7. M. Mlejnek, I. Roudas, J. D. Downie, N. Kaliteevskiy, and K. Koreshkov, “Coupled-Mode Theory of Multipath Interference in Quasi-Single Mode Fibers,” IEEE Photonics J. 7(1), 1–16 (2015). [CrossRef]
8. J. D. Downie, M. Mlejnek, I. Roudas, W. A. Wood, A. Zakharian, J. E. Hurley, S. Mishra, F. Yaman, S. Zhang, E. Ip, and Y.-K. Huang, “Quasi-Single-Mode Fiber Transmission for Optical Communications,” IEEE J. Select. Topics Quantum Electron. 23(3), 31–42 (2017). [CrossRef]
9. J. D. Downie, J. Hurley, H. DePedro, S. Garner, J. Blaker, A. Zakharian, S. Ten, and G. Mills, “Investigation of Potential MPI Effects on Supervisory Channel Transmission Below Cable Cut-off in G.654 Fibres,” 42nd European Conference on Optical Communication, 2016, pp. 1–3 (2016).
10. J. D. Downie, J. Hurley, H. DePedro, S. Garner, J. Blaker, A. Zakharian, S. Ten, and G. Mills, “Measurements and modeling of multipath interference at wavelengths below cable cut-off in a G.654 optical fiber span,” Opt. Express 25(8), 9305–9311 (2017). [CrossRef]
11. Q. Sui, H. Zhang, J. Downie, W. Wood, J. Hurley, S. Mishra, A. Lau, C. Lu, H. Tam, and P. Wai, “Long-haul quasi-single-mode transmissions using few-mode fiber in presence of multi-path interference,” Opt. Express 23(3), 3156–3169 (2015). [CrossRef]
12. R. Scarmozzino, A. Gopinath, R. Pregla, and S. Helfert, “Numerical techniques for modeling guided-wave photonic devices,” IEEE J. Select. Topics Quantum Electron. 6(1), 150–162 (2000). [CrossRef]
13. D. Marcuse, “Influence of curvature on the losses of doubly clad fibers,” Appl. Opt. 21(23), 4208–4213 (1982). [CrossRef]
14. Recommendation ITU-T G.650.1, “Definitions and test methods for linear, deterministic attributes of single-mode fibre and cable” (International Telecommunication Union, 2020).
15. J. D. Downie, J. Hurley, I. Roudas, K. Koreshkov, and M. Mlejnek, “MPI Measurements of quasi-single-mode fibers,” Photonics Conference (IEEE, 2015), 273-274, paper MG3.4.
16. J. D. Downie, W. A. Wood, J. Hurley, M. Mlejnek, I. Roudas, A. Zakharian, S. Mishra, F. Yaman, S. Zhang, E. Ip, and Y.-K. Huang, “Quasi-single-mode fiber transmission for submarine systems,” in SubOptic (2016), paper TH1A.3.
17. E. L. Goldstein, L. Eskildsen, and A. F. Elrefaie, “Performance implications of component crosstalk in transparent lightwave networks,” IEEE Photon. Technol. Lett. 6(5), 657–660 (1994). [CrossRef]
18. H. Takahashi, K. Oda, and H. Toba, “Impact of Crosstalk in an Arrayed-Waveguide Multiplexer on N x N Optical Interconnection,” J. Lightwave Technol. 14(6), 1097–1105 (1996). [CrossRef]
19. W. Zheng, H. P. Sardesai, M. G. Taylor, D. L. Craig, J. Fowlkes, and J. R. Simpson, “Measurement and system impact of multipath interference from dispersion compensating fiber modules,” IEEE Trans. Instrum. Meas. 53(1), 15–23 (2004). [CrossRef]
20. C. H. Kim and Optics Letters, “In-band crosstalk-induced power penalties in various signal modulation formats,” Opt. Lett. 44(24), 5965–5968 (2019). [CrossRef]
21. I. Tomkos, B. Hallock, I. Roudas, R. Hesse, A. Boskovic, J. Nakano, and R. Vodhanel, “10-Gb/s transmission of 1.55-µm directly modulated signal over 100 km of negative dispersion fiber,” IEEE Photon. Technol. Lett. 13(7), 735–737 (2001). [CrossRef]
