1. Introduction

Spatial-division multiplexing (SDM) systems are a promising approach to support future high-capacity optical networks [1]. These systems rely on the use of spatially diverse parallel optical channels to multiply the transmission throughput of optical links. A possible transmission media for SDM networks are weakly coupled multi-core fibers (MCFs). These have been intensely researched during the last decade and are characterized by the use of multiple cores within a common cladding, which act as spatial channels. The homogeneous variants of MCFs have similar propagation characteristics on all cores, which enable time-aligned transmission of spatial channels [2]. This supports the use of spatial super channels (SSCs) [3], where a high capacity signal is multiplexed and transmitted over multiple spatial channels sharing a common wavelength band. This approach has also been extended to high-throughput short reach optical interconnects using intensity-modulated direct detection (IM-DD) signals, where the use of SSCs transported in MCFs is analogous to existing multi-fiber lane standard systems [4]. In these systems, the skew is expected to result mainly from the transceivers electronics and optical propagation and will be managed with electronic buffering techniques, which introduce electronic complexity and latency, with potential impact on cost [5]. In this sense, the use of MCFs promises to reduce propagation skew by at least an order of magnitude [6,7], which may relax the requirements for electronic buffering mechanisms and extend the reach.

Nevertheless, these applications rely on achieving small differences between the propagation delays of different cores, referred to here as inter-core skew (ICS) [8]. ICS in homogeneous MCFs depends on the fabrication of the fiber as well as its deployment, being particularly sensitive to bends and twists [9] as well as the arbitrarily varying environmental conditions the fiber is subject to [6,7]. For these reasons, ICS may be difficult to predict and dynamically compensate for. With reported ICS values up to 0.5 ns/km [7], it can be easily extrapolated that even short reach optical interconnects up to 10 km in length may require some form of ICS compensation for the transmission of SSCs.

This work proposes a simple approach for the compensation of ICS, which, to a certain extent, is independently of the transmission distance. The proposed approach is suitable for systems where the definition of SSC can be relaxed to accommodate distinct wavelengths on different spatial channels. By appropriately tuning the wavelength used by different spatial channels, it is possible to adjust their relative group delays, referred to here as walkoff, to compensate the ICS along the fiber, as shown in Fig. 1. This approach is particularly suitable for MCF-based short-reach interconnects using a single wavelength per spatial channel, such as the systems used in intra data-center networks. In these systems, the operation band of the receiver is usually wide enough to consider it agnostic to the signal wavelength. Our approach would require the use of tunable transmitter lasers with each spatial channel. We note that exploiting the interplay between wavelength-dependent group delay in MCFs and ICS has been extensively proposed in works such as [10,11], among others. However, those works focus on the design of heterogeneous MCFs, where the ICS and dispersion properties of each core are carefully engineered to enable simultaneous high-frequency radio-over-fiber signal processing and distribution. In this work, we focus on the compensation of ICS on homogeneous MCFs that were designed for transmission of time-aligned spatial channels but carry some residual ICS or have ICS induced from environmental conditions, bending or twisting.

 

Fig. 1. Principle of ICS compensation using wavelength dependent group delay. Top - uncompensated case with spatial channels transmitted at a common wavelength. Bottom - the wavelength of the signal transmitted in core 2 is chosen to have the group delay counter the ICS.

Download Full Size | PDF

This paper is structured as follows. Section 2 presents an experimental demonstration of the proposed method using a 111 Gb/s spatial super channel composed by 4 spatial channels, each with a 27.8 Gb/s on-off keying (OOK) signal, and transmitted through 4 cores of a 10.1 km 19-core homogeneous MCF. A discussion on the advantages and disadvantages of this technique as well as alternative applications is included in section 3 and final conclusions are outlined in section 4.

2. Experimental demonstration

Figure 2(a) shows the experimental setup. The lightwaves from 4 tunable external cavity lasers (ECLs) were modulated using Mach-Zehnder modulators (MZMs) driven by a 4-channel pulse pattern generator (PPG). The transmitted signal was a 2¹⁶-1 symbol pseudo-random binary sequence time demultiplexed onto the 4 output channels of the PPG, each operating at 27.8 Gb/s for a combined throughput of 111.2 Gb/s. The optical signals were spatially multiplexed by a 3-D waveguide device onto 4 cores of a 10.1 km MCF. The latter was fully described in [12] and had a 200 µm cladding with 19 trench-assisted cores. The cores were placed in an hexagonal arrangement, as shown in Fig. 2 with an average pitch of 35 µm. The average crosstalk between adjacent cores, including the multiplexers, was −32 dB at a wavelength of 1550 nm, which, for the purpose of this work, had a negligible impact.

 

Fig. 2. Experimental setup (a) and ICS per unit distance of each core of the 19-core fiber with respect to core 1 (b).

Download Full Size | PDF

The cores selected for this experiment are highlighted and numbered 10, 11, 12, and 13 in Fig. 2(a), Fig. 2(b) shows the ICS of each core with respect to the center core 1. It is shown that the selected cores have a relatively high maximum ICS of 0.29 ns/km, despite being adjacent. The maximum ICS between any pair of cores in this fiber was 0.38 ns/km. At the fiber output, a 3-D waveguide device was used to spatially demultiplex the transmitted signals, which were then detected by photodetectors (PDs). The resulting electrical signals were sent to a 4-channel bit-error-rate tester (BERT) for performance analysis. Alternatively, the electrical signals were sent to a real-time digital sampling oscilloscope (RT-DSO) operating at 80 GS/s. This was used to measure the skew between channels with a cross-correlation method similar to the method presented in [6].

Figure 3 shows the wavelength dependence of the ICS for the 4 considered cores with respect to core 10 at 1528 nm. It is shown that the delays between channels remain approximately unchanged across the C-band but increase linearly with respect to a fixed reference, mainly as a consequence of GVD. In this case, the reference was the propagation delay of channel 10 at a wavelength of 1528 nm. The ICS between channels 10 and 11 ranged between −233 ps/km and −240 ps/km at the wavelengths of 1528 nm and 1565 nm, respectively. Similarly, the ICS between channels 10 and 12 ranged between 50 ps/km and 55 ps/km. Finally, the ICS between channels 10 and 13 ranged between −116 ps/km and −121 ps/km. The maximum ICS between any pair of signals with similar nominal wavelengths in this system was approximately 296 ps/km and the average dispersion parameters of the cores ranged between 19.3 ps/nm/km and 19.5 ps/nm/km, measured using the slopes of the curves in Fig. 3. As such, the maximum wavelength spacing required to fully align these signals was 15.3 nm. However, it should be stressed that the actual required wavelength spacing varies slightly with the wavelength selected for transmission due to the core-dependent dispersion parameters.

 

Fig. 3. Wavelength dependence of the ICS of cores 10, 11, 12, and 13 with respect to the reference core 10 at a 1528 nm. Scenarios A and B correspond to the transmission of all channels at the same wavelength, 1550 nm, or with the same delay, 200 ps/km with respect to the reference core, respectively.

Download Full Size | PDF

We considered two transmission scenarios A and B, as shown in Fig. 3. In scenario A, the spatial channels were aligned at a nominal wavelength of 1550 nm. In scenario B, the spatial channels were aligned at the similar propagation delay. For this we positioned channels 10, 11, 12, and 13 at the wavelengths of 1538.8 nm, 1551.1 nm, 1536.1 nm, and 1545.0 nm, respectively. We note that scenario A corresponds to the typical use case for spatial super channels, with similar wavelengths for all spatial channels, whereas scenario B corresponds to the proposed approach. Figure 4(a) and (b) show short segments of the detected traces in scenarios A and B, respectively. For this figure, the PPG was programmed to introduce long sequences of zeros, in order to facilitate the visualization of the delays between channels. Figure 4(a) illustrates the delay between channels typical of transmission in MCFs. In contrast, Fig. 4(b) shows the delays fully compensated without the use of delay lines. The alignment was achieved solely by tuning the wavelengths of the transmitter lasers.

 

Fig. 4. Traces of the detected signals from channels 10, 11, 12, and 13 when transmitted at the same nominal wavelength of 1550 nm - Scenario A (a) and when transmitted with the same delay - Scenario B (b).

Download Full Size | PDF

For performance evaluation, we implemented scenario B, with appropriately aligned spatial channels and measured the performance of the combined 111 Gb/s signal. Figure 5 shows the BER estimated in real-time by the BERT as a function of the frequency offset of channel 10 with respect to the remaining channels. The sampling phase of the BERT for each channel was set initially by tuning all the channels at zero frequency offset. Afterwards, if remained fixed during the measurement. It is shown that a frequency offset range between −2 GHz and 2.8 GHz is required to achieve a BER below 10⁻⁹. This corresponds to a delay spread of of 6.6 ps, which is little more than 18% of the symbol period. This may be considered a very narrow range. However, it is worth stressing that all alignment mechanisms that would support clock phase tracking have been disabled. These would allow extending the allowed delay spread significantly. Nevertheless, it demonstrates the effectiveness of the proposed method to align spatial channels with significant precision without the use of optical or electrical delays.

 

Fig. 5. Dependence of the BER of the 111 Gb/s SSC on the optical frequency offset of channel 10 with respect to the optimum alignment.

Download Full Size | PDF

3. Discussion

The previous section demonstrated the feasibility of an ICS compensation technique based on tuning the wavelengths of the transmitted spatial channels in order to have the differences in group delay between cores compensate for ICS. The proposed technique has the advantage of being independent of transmission distance and does not require mechanical systems, such as motorized variable optical delay lines. It was shown to allow sub-symbol duration precision thanks to the sub-gigahertz tuning precision of the transmitter lasers. This technique may also be suitable for more complex network architectures and account not only for ICS but also for delays within network elements, such as SDM reconfigurable optical add and drop multiplexers, or electrical connections within the transmitter and receiver sub-systems.

However, despite its ease of use, there are significant limitations that must be taken into account at this point. We begin with the bandwidth requirements of the proposed approach. In the example shown in the previous section, the required wavelength range for ICS compensation was approximately 15.3 nm. On a C-band system with a transmission bandwidth of 35 nm, the need to reserve more than 15 nm for ICS compensation alone may be overwhelming unless single wavelength systems are used. In fact, if we assume that the entire C-band is available for ICS compensation, our approach could accommodate an ICS parameter of less than 0.68 ns/km. In cases where the ICS compensation range is insufficient, additional compensation techniques would be required. Another important limitation is the need for transmission at wavelengths with non-zero dispersion, which may degrade the signal performance. Also, this excludes the O-band, where fiber dispersion is typically very low and is commonly used for short reach interconnects. Finally, an important outcome of Fig. 5 is the fact that the sensitivity of the ICS alignment to laser detuning may be significant. This implies the need for relatively high-end temperature-stabilized lasers. Low-cost uncooled lasers may have frequency wander that is sufficiently high to prevent a stable ICS compensation. Nevertheless, and despite its shortcomings, the proposed technique remains as a valid approach for a simple method of ICS compensation, which may be used in conjunction with other techniques, such as electronic buffers. It may also be used as a means to perform corrections to static ICS compensation methods, such as optical delay lines, to account for dynamic fluctuations of ICS [6,7].

4. Conclusion

This work presented a method for the compensation of residual inter-core skew in homogeneous multi-core fiber transmission systems. The proposed method is based on the transmission of the different spatial channels composing a spatial super channel at different wavelengths. The selected wavelengths are such that the differences in group delay between spatial channels compensate the inter-core skew. The proposed method is experimentally demonstrated using a 111 Gb/s spatial super channel transmitted through 4 cores of a 19-core 10.1 km homogeneous multi-core fiber. The system performance is evaluated, showing that the proposed method allows inter-core skew compensation with sub-symbol duration precision, dispensing any further alignment assistance.