1. Introduction

Point to point links without active inline elements, often referred to as unrepeatered links, are an essential component of submarine communications to support connections to remote locations, such as isolated islands and festoon networks [1]. By eliminating active inline components, such as amplifiers or repeaters, these systems are substantially less costly and less complex than links with undersea amplification. Significant research has been recently devoted towards increasing the capacity and reach of unrepeatered systems [2–6]. This includes wideband amplification using Raman amplifiers [2] or semiconductor optical amplifiers [3], compensation of fiber nonlinearities, [4,5] and highly spectral efficient modulation formats [6]. Recent demonstrations of some of these techniques in real-time have also cemented their feasibility [7].

More recently, unrepeatered links have also been proposed to support highly cost-sensitive terrestrial inter-datacenter links with reach up to regional distances [1]. These links would avoid the need for potentially costly repeater sites whilst making use of terrestrial grade fibers, which typically present higher loss than ultra low-loss high effective area submarine-grade fibers. High capacity can be achieved using C+L band transmission supported by wideband amplification schemes, such as semiconductor optical amplifiers [3]. The latter have enabled the current record capacity$\times$distance product of 20.6 Pb/s·s for unrepeatered systems on a 257.5 km link.

Here, we propose the use of bidirectional distributed Raman amplification to achieve a similar capacity$\times$distance product whilst using 234.8 km of standard single mode fiber (SSMF). For this purpose, we transmit 424 wavelength division multiplexed (WDM) channels at 24.5 GBaud with polarization-division-multiplexed (PDM) 16-ary, 32-ary, and 64-ary-quadrature amplitude modulated (QAM) signals across a bandwidth exceeding 87.2 nm, encompassing the C and L bands. Bidirectional Raman amplification is implemented using 12 backwards propagating and 11 forwards propagating pumps spread over a wavelength range between 1410.8 nm and 1502.7 nm for wideband amplification. The total pump powers were 1.4 W and 1.16 W for backwards and forwards propagating pumps, respectively. The total launch power is optimized to balance the impact of stimulated Raman scattering (SRS) to reach a combined throughput above 90 Tb/s and a capacity$\times$distance product above 21 Pb/s$\cdot$km after LDPC decoding with PDM-64QAM. This paper is structured as follows. Section 2 describes the experimental setup used for this demonstration with the corresponding results and discussion in section 3. Final conclusions can be found in section 4.

2. Experimental setup

Figure 1 shows a simplified diagram of the experimental setup used in this demonstration. A high quality 3-channel sliding test band was produced by modulating three lightwaves from <100 kHz linewidth tunable external cavity lasers (ECLs) and spaced by 25 GHz. The center lightwave was independently modulated by a dual-polarization IQ modulator (DP-IQ) to produce the channel under analysis. The outer lightwaves were combined at a polarization-maintaining coupler, prior to modulation at a separate DP-IQ modulator. The modulators were driven by 4 arbitrary waveform generators (AWGs) with electrical bandwidths around 12 GHz and operating at 49 GS/s to produce pre-equalized 24.5 GBaud PDM-16QAM, PDM-32QAM, or PDM-64QAM signals with 2¹⁶ symbols. The AWGs had approximately 5 effective bits at a modulation frequency of 25 GHz. Outer and center channels were combined at a 2$\times$2 coupler to form the sliding test band with the outer channels optically decorrelated by a 50 ns optical delay. C-band or L-band erbium-doped fiber amplifiers (EDFAs) were used to amplify the test band, depending on its wavelength and a VOA was used to adjust the power to match a wideband low-OSNR dummy band. The dummy band was produced by modulating the signal at the output of a comb generator using a single-polarization IQ (SP-IQ) modulator. The comb produced 25 GHz spaced carriers across a wavelength range >100 nm, encompassing the C and L bands. C and L band optical processors (OP) positioned in-between corresponding C and L band EDFAs were used to define the power profile of the launch spectrum and set a pre-determined power tilt. The bandwidths of the EDFAs and the WDM couplers conditioned the minimum C band and maximum L band wavelengths, as well as the gap between the C and L bands. The inset diagram in Fig. 1 shows the structure of the transmitted spectrum with 421 dummy channels, a 75 GHz notch to accommodate the sliding test band and a 25 GHz gap between the C and L bands. Both test and dummy bands were combined in a 1:10 coupler and amplified by a C+L band amplifier followed by a variable optical attenuator (VOA) for power control. Figure 2-a) shows the spectrum of the transmitted signal, measured with a resolution bandwidth of 1 nm. Note that, due to the limited resolution, the C/L gap at 1567.952 nm is not distinguishable. We used a power tilt of 16 dB between the shortest C-band wavelength of 1527.994 nm and the longest L-band wavelength of 1615.261 nm, as shown in Fig. 2-a). This tilt was selected to balance the power and OSNR spectra after transmission, as described in the following.

 

Fig. 1. Experimental setup for unrepeatered transmission of 424 WDM channels consisting of a 3-channel sliding test band and a 421-channel dummy band across C and L bands.

Download Full Size | PDF

 

Fig. 2. Spectra measured with a resolution bandwidth of 1 nm of a) the transmitted WDM signal, b) span loss, and c) received WDM signal. The latter was measured for launch powers between 10 dBm and 18 dBm.

Download Full Size | PDF

The transmission line was composed by a 234.8 km standard single-mode fiber with WDM couplers at the terminal points to couple the signal with co- and counter-propagating Raman pumps. Fig. 2-b) shows the measured span loss of the fiber including splices. The minimum loss was 26.5 dB at a wavelength of 1569.6 nm. The dispersion parameter and effective area were 18 ps/nm/km and 90 µm² at 1550 nm, respectively. For each propagation direction, we used up to 12 Raman pumps with wavelengths between 1410.8 nm and 1502.7 nm. To optimize the launch power, launch power tilt and the Raman pump powers, we began by setting a given combination of launch power and launch power tilt. In general, the power of the short wavelength C-band channels will be severely degraded without Raman amplification. We then adjusted the Raman pumps manually to achieve an OSNR of approximately 15 dB at the short end of the C-band. This was done by directly observing the received signal spectrum as shown in Fig. 2-c). Afterwards, we optimized the Raman pumps using a simplex algorithm. The algorithm used a target function based on an estimation of the OSNR spectrum, which was obtained out by carving 5 gaps of 250 GHz at different points of the transmitted signal spectrum and measuring the corresponding noise power levels. This process was repeated for multiple launch power and launch power tilt combinations until reaching a maximum performance for 16 dB tilt.

Figure 3 shows the OSNR after the optimization for each channel. In this case, the OSNR was estimated by turning off the test band for each wavelength and comparing the noise floor with the signal power. This method had limited accuracy at wavelengths near the seed wavelength of the comb source, around 1568 nm, due to the varying powers of the comb lines at that region. As shown in Fig. 2-c) and Fig. 3, the launch power affects significantly the received power and the OSNR, in particular for the long L-band wavelength region. The impact of SRS translates into a tilt of the OSNR that favors the L-band when increasing the launch power. This suggests that the optimum launch power will correspond to a balance between the achievable throughput of the C and L bands. The final Raman pump powers are listed in Table 1, which also includes the corresponding wavelengths and fiber attenuation parameters.

 

Fig. 3. OSNR estimated at a resolution bandwidth of 0.1 nm for launch powers between 10 dBm and 18 dBm.

Download Full Size | PDF

Table 1. Forward and backward propagating Raman pump wavelengths and powers.

View Table

We note that the optimization process used in this work may be sub-optimal. Advanced methods, such as the one proposed in [8] would be more suitable given the large number of Raman pumps as well to allow different options for the launch power spectrum. Nevertheless, we found that our procedure was sufficient to achieve an acceptable performance. The development of more complex methods of optimization are outside the scope of this work.

At the receiver, the WDM signal was first amplified by a C+L-band EDFA. A tunable band-pass filter (TBF) was used to select the test channel, which was then amplified by C or L band EDFAs, depending on the operation wavelength. After a VOA for power control, the signal was mixed with a local oscillator in a polarization-diverse coherent receiver. The local oscillator was a 100 kHz ECL. The electrical output of the receiver was digitized by a real-time digital sampling oscilloscope operating at 80 GS/s with an electrical bandwidth of approximately 33 GHz. For performance evaluation, we used 10 µs traces of the received signal, which were stored for offline processing. The digital signal processing consisted in stages for resampling to 2 samples per symbol, orthonormalization and frequency-domain dispersion compensation followed by timing recovery. Afterwards, a 2$\times$2 MIMO was used to undo polarization multiplexing and the remaining channel response. The MIMO used 17-tap equalizers, which were updated using a least-mean squares algorithm. The taps were initially estimated using a data-aided approach, which was switched to decision-directed after convergence. Carrier recovery was performed within the equalizer loop, using the method proposed in [9]. We note that we used a fixed window length for PDM-16QAM and PDM-64QAM signals of 256 symbols for carrier phase recovery. However, for PDM-32QAM signals, convergence was significantly harder to achieve due to cycle slips and required a varying window length of up to 2048 symbols. The post-FEC throughput was computed by emulating the decoding of the received signals using LDPC codes from the DVBS2 standard [10], as previously described in [11]. We implemented code puncturing on the original codes to achieve a rate-granularity of 0.01. Each code rate was evaluated by generating at least 100 random code words using symbols of the received signals. When reaching a target post-LDPC BER of 4.5$\times$10^-5 minus a 10% margin, we considered the code to be sufficiently acceptable for applying a 1% overhead hard decision outer code described in [12] to recover the remaining errors.

3. Performance

Figure 4(a) to (c) show the throughput for each WDM channel after LDPC decoding for PDM-16QAM, PDM-32QAM, and PDM-64QAM signals and launch powers varying between 10 dBm and 18 dBm. In the case of PDM-16QAM, it is clear that for most of the transmission spectrum, the throughput per channel is nearly independent of the wavelength as the performance is nearly the theoretical maximum. For both PDM-32QAM and PDM-64QAM, the L-band performance tends to increase steadily with the launch power. For the C-band, the performance decreases with the launch power for powers above 12 dBm, as a result of SRS.

 

Fig. 4. Throughput of the 424 WDM channels estimated after LDPC decoding for PDM-16QAM, 32QAM and 64QAM with launch powers between 10 dBm and 18 dBm.

Download Full Size | PDF

 

Fig. 5. Dependence of the total throughput after LDPC decoding on the total launch power for PDM-16QAM, PDM-32QAM, PDM-64QAM and selecting the best format for each WDM channel (a). For the latter case, (b) to (f) show the constellation cardinality of the selected format for each of the WDM channels for launch powers from 10 dBm to 18 dBm.

Download Full Size | PDF

Figure 5-a) shows the total throughput dependence on the launch power for the considered modulation formats. The highest throughput of 90 Tb/s is reached with PDM-64QAM and a launch power of 16 dBm. However, it should be noted that for launch powers below 14 dBm we observed higher throughput using PDM-32QAM. The maximum throughput corresponds to a capacity$\times$distance product of 21.132 Pb/s$\cdot$km. Figure 5-a) also shows the throughput that would be achieved if one was to select the best of the 3 considered modulation formats for each wavelength. In this case, the maximum throughput was 91.59 Tb/s for a capacity$\times$distance product of 21.507 Pb/s$\cdot$km, 1.7% higher than the use of PDM-64QAM only. It is interesting to note the distribution of the best performing modulation format across the transmitted spectrum, shown in Fig. 5-b) to -f) for increasing launch power. For 10 dBm launch power, the spectrum would be dominated by PDM-32QAM signals, using PDM-16QAM only for the short C-band wavelengths, which are degraded by SRS. As the launch power is increased, the use of PDM-64QAM signals becomes advantageous, especially to exploit the high performance of the L-band wavelengths boosted by SRS. Nevertheless, it is important to note that the gain achieved with the complexity of using multiple modulation formats may be unjustifiable considering the possible added cost. In fact, the use of PDM-32QAM alone would enable 98% of the maximum throughput achievable using PDM-64QAM. Another possibility is the use of modulation formats with highly granular throughput, such as probabilistic shaping, as in [6]. In this case, the entropy of each wavelength could be finely adjusted to exploit the channel capacity and improve the overall throughput. This topic will be subject of future research.

4. Conclusion

This work presented a demonstration of an unrepeatered link with a maximum capacity$\times$distance product of 21.132 Pb/s$\cdot$km after 234.8 km transmission using SSMF and bidirectional Raman amplification. For this purpose, the transmitted WDM signal had a bandwidth exceeding 87 nm across C and L bands and carried 424 wavelengths with 24.5 GBaud PDM-64QAM signals. Transmission line amplification was performed using 12 forward propagating and 11 backwards propagating pumps spread over a wavelength range above 90 nm with total pump powers of 1.45 W and 1.16 W, respectively. We evaluated the impact of varying launch power to balance the degradation imposed by stimulated Raman scattering on the C and L bands. We also evaluated the benefits of using multiple modulation formats to reach a relatively small throughput gain around 1.7%. This may be insufficient to justify the complexity of multiple modulation formats on a real link. Nevertheless, to the authors’ knowledge, this work reported the highest capacity$\times$distance product for an unrepeatered link with more than 200 km and without the use of remote optically pumped amplifiers.