Design and Performance of 1×8 Core Selective Switch Supporting 15 Cores Per Port Using Bundle of Three 5-Core Fibers

A core-selective switch (CSS) is a key building block for a port modular spatial cross-connect where an optical signal launched into any core in the input multi-core fiber (MCF) can be switched to a core that has the same core identifier of any output MCF. One way to increase the core count in a CSS is to use several MCFs as a bundle and collimate/demultiplex beams emitted from it all at once using a single microlens. In this paper, we report a 1×8 CSS prototype based on a bundle of three 5-core fibers (5-CFs), which supports 15 cores per port, demonstrating the feasibility of a high core count CSS through MCF bundling. The CSS prototype exhibits insertion loss of less than 4.8 dB, and polarization dependent loss of less than 0.5 dB over an ultrawide wavelength range of 1480 nm to 1630 nm. The inter-core crosstalk (XT) characteristics as a function of the wavelength are thoroughly investigated. We show that the intra 5-CF XT imparted to the center core from the other four outer cores in the same 5-CF is less than −52 dB and the inter 5-CF XT from cores in the other 5-CFs in the same bundle is less than -64 dB, which are sufficiently low values for practical networks. No OSNR penalty in the 100-Gb/s spatial channel was observed when ten 100-Gb/s DP-QPSK WDM signals in the C-band were simultaneously routed by the bundled 5-CF 1×8 CSS prototype.

Color being developed to accommodate the ever-increasing data traffic demand [1], [2]. Among such spatial division multiplexing (SDM) fibers, a nominally uncoupled four or five core fiber (4-CF or 5-CF) with a standard 125-μm cladding will probably be the earliest SDM fiber to market due to its higher degrees of reliability and compatibility with the current single-mode fiber (SMF) based optical transmission technology [3]. Novel optical node technologies that support SDM fibers including joint switching of a spatial superchannel [4], [5], [6], [7], [8], [9] and a subsystem-modular wavelength cross-connect (WXC) [10], [11], [12] have also been investigated. However, in the optical node technologies proposed thus far, all traffic entering a node is still processed in the fine-granular spectrum domain, which may suffer from scaling and cost increase issues. On the other hand, the idea of a multigranular optical network with the coarsest granular fiber switching was already proposed at the end of the '90s [13]. As SDM technology began to gather much attention, some experimental demonstrations on multigranular (time, wavelength, and space) optical networks based on a multi-core fiber (MCF) link and a high port count SMF switch were reported [14], [15], [16]. The spatial channel network (SCN) architecture [17], [18], [19], [20] was recently proposed as another approach to take full advantage of the spatial dimension. In an SCN, the current optical layer evolves into hierarchical wavelength division multiplexing (WDM) and SDM layers, and an optical node is decoupled into a spatial cross-connect (SXC) and a conventional WXC to form a hierarchical optical cross-connect [18]. Here, a spatial channel is an optical channel in the SDM layer that is constructed by connecting cores in each SDM link (a MCF or parallel SMFs) on a route using SXCs. The SCN architecture will yield two major benefits: a reduction in the total node cost in the network and an extension to the optical reach for optical signals that spatially bypass the overlying WDM layer [17], [20]. As a key building block for a port modular SXC, a novel spatial optical switch referred to as a core selective switch (CSS) was recently proposed [21], [22], [23], [24], [25], [26]. A CSS is a one-input MCF and N-output MCF device where an optical signal launched into any core in the input MCF can be switched to a core that has the same core identifier of any output MCF. A 1×8 CSS supporting a 5-CF per port was prototyped using a This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ micro-electromechanical systems (MEMS) mirror array, which exhibits low insertion loss (IL) and low polarization dependent loss (PDL) over a 1500-nm to 1630-nm wavelength range [24]. Considering that wavelength channels from 4 to 96 are used in the current WDM network, it will most likely be required that several tens or more spatial channels (cores) be used in an SCN as well. A straightforward way to increase the core count in a CSS is to employ a high core count MCF, for example, a 19-core fiber [26]. Another way is to use several MCFs as a bundle and collimate/demultiplex beams emitted from it all at once using a single microlens [25]. A further increase in the number of cores is expected by using both schemes together.
In this paper, we report a 1×8 CSS prototype based on a bundle of three 5-CFs, which supports 15 cores per port, demonstrating the feasibility of a high core count CSS through MCF bundling. This paper is an expanded version of [25] with detailed descriptions of the bundled 5-CF collimator array design and the ray-tracing simulation of the 1×8 CSS prototype. We added experimental results including intra-and inter-MCF crosstalk (XT) and bit error rate (BER) measurements of routed SDM optical signals. The rest of this paper is organized as follows. Section II gives details on the MEMS-based CSS prototype design including the optimization of ray tracing simulations and the alignment accuracy estimation of the MCF collimator array. Section III describes performance evaluation results for the CSS prototype including wavelength dependence of the IL, PDL, and XT, and BER measurements of routed SDM optical signals. Section IV presents our conclusions.

A. Design and Function of Bundled 5-CF 1×8 CSS
The bundled 5-CF 1×8 CSS prototype comprises a 3×3 array of bundles of three 5-CFs, a 3×3 microlens array, a condenser lens, and a 3×5 MEMS mirror array each aligned in a 4-f system as shown in Fig. 1. Fig. 2 shows arrangements for nine 5-CF bundles labeled B 0 to B 8 , three 5-CFs in a bundle labeled F 0 to F 2 , and five cores in a 5-CF labeled C 0 (center core) and C 1 to C 4 (outer cores). The bundle of three 5-CFs in the center of the 3×3 array (B 4 ) is used as an input port of the CSS and the others are output ports. As illustrated using ray tracing in Figs. 1, 15 beams emitted from the input 5-CF bundle are collimated and spatially demultiplexed by the center microlens and imaged on each corresponding mirror of the 3×5 MEMS mirror array by the condenser lens. They are steered toward their respective output ports using the MEMS mirrors, and spatially re-multiplexed into the output 5-CFs. Reflected ray traces are not shown for better visibility. Fig. 3 shows a high-level view of the bundled 5-CF 1×8 CSS. The CSS has an input port (bundle B 4 ) that comprises three 5-CFs labeled F 0 to F 2 , each supporting five cores labeled C 0 to C 4 and eight output ports (B 1 to B 3 and B 5 to B 8 ). If we number the 15 cores in each bundle from C 0 to C 14 , the function of the CSS is described such that an optical signal input to any core of an input port can be output to a core having the same core number in any output port.     5-CFs are closely placed next to each other on the vertices of an equilateral triangle with fiber-to-fiber spacing g of 12.5 μm. The core spacing, c, between the center and outer cores of the 5-CF, cladding diameter d, mode field diameter, and cut-off wavelength of the 5-CF are 31.6 μm, 125 μm, 9 μm, and <1290 nm, respectively [3]. The attenuation and XT at the wavelength of 1550 nm are 0.22 dB and less than −60 dB after 1-km transmission, respectively.
The 5-CF bundle and microlens array, as shown in Fig. 4(b), are assembled to form a 3×3 bundled 5-CF collimator array as shown in Fig. 4(d). Fig. 5 shows a side view of the collimator array. The microlens array comprises nine plano-convex microlenses with the focal length of 810 μm and the thickness of 1 mm, each orthogonally arranged with a 750-μm lens pitch. In order to set the facet of the 5-CF bundle array at the front focus  point of the microlens, a silica spacer block with the thickness of 350 μm is inserted. The 5-CF bundle array and the microlens array with anti-reflection (AR) coating are precisely adjusted so that each center of the three center cores of the 5-CF bundles is aligned with the optical axis of each microlens. Then, they are fixed via the silica spacer block with refractive index matching adhesive.
The 4-f system based on the microlens with focal length f 1 and the condenser lens with focal length f 2 forms a point-symmetric enlarged image of the 15 beams exiting the input 5-CF bundle on each MEMS mirror with magnification factor M = f 2 /f 1 . We first designed the 5-MEMS mirror array that has five 1-mm diameter mirrors placed at a 2.1-mm pitch in the same arrangement as the core placement in the 5-CF as shown in Fig. 5. The required magnification factor, M , is determined from the core and mirror pitches as M = 2.1/0.0316 = 66.5. The available microlens array has a 0.81-mm focal length, which requires the condenser lens focal length of 53.8 mm. Since 5-CFs in the bundle are arranged at a 137.5-μm spacing, three 5-MEMS mirror arrays are placed at a spacing of 9.14 mm ( = 137.5 × M ) as shown in Fig. 6. Parameters for the components used in the CSS prototype are summarized in Table I.

C. Beam Clipping Estimation
The required aperture of the microlens, D, to prevent significant beam clipping for beams emitted from the outermost cores is given by Here, l, w col , and α are the length from the center of the 5-CF bundle to an outermost core, the spot size on the microlens given by where w 0 is the beam waist of a beam at the exit facet of each core and λ is the wavelength, and clipping factor, respectively. If we need the beam clipping for beams emitted from the outermost cores to be less than 1%, α should be greater than 1.5. The l for the bundle of three 5-CFs is given by The solid line in Fig. 7 shows the required aperture of the microlens, 2(αw col + l 3 ), for the bundle of three 5-CFs, which becomes 485 μm when w 0 , λ, α, and f 1 are 4.55 μm, 1.55 μm, 1.6, and 810 μm, respectively. Since this is much less than the effective aperture of the employed microlens (750 μm), we can expect no beam clipping to occur in the CSS prototype employing bundles of three 5-CFs. If seven 5-CFs are arranged in a hexagonal closely packed structure, l is given by and 2(αw col + l) is 601 μm for the same parameters as shown by the dotted line in Fig. 7. We conclude that a CSS supporting seven 5-CFs per port, which yield a high core count per port of 35, is possible using this CSS design.

D. Design and Optimization Using Ray Tracing Software
We optimized the design of the bundled 5-CF 1×8 CSS shown in Fig. 1 by using a commercially available optical design software based on ray tracing. Fig. 8 shows simulation results of ILs for 15 optical beams at a wavelength of 1550 nm emitted from the input 5-CF bundle (B 4 ) and coupled to the output bundle (B 1 ). The figure indicates that if the optical alignment is perfect, an IL of approximately 0.15 dB can be achieved. Fig. 9 shows ILs as a function of wavelength ranging from 1480 nm to 1630 nm. Reflecting that the design optimization was conducted to minimize the IL for three wavelengths (1530 nm, 1545 nm, and 1560 nm) in the C-band, the IL is less than 0.2 dB in the C band and gradually increases as the input signal wavelength moves away from the C-band to the S-band or Lband. However, the highest IL still remains less than 0.5 dB even at the wavelength of 1480 nm or 1630 nm.
Taking the actual power transmittance of the microlens and condenser lens and the power reflectance of the MEMS mirror of approximately 95%, we estimate an additional loss of approximately 0.7 dB, resulting in the total IL of approximately 0.85 dB at 1550 nm.

E. Collimator Array Accuracy Evaluation
In order to know how accurately the collimator array was fabricated, we launched a 1550-nm light wave into each core of the 3×3 bundled 5-CF array in turn and observed using a spatial beam profiler placed at the position of a MEMS mirror array. Fig. 10 shows the overlaid beam positions.
In the 4-f system of a CSS, if the collimator array is fabricated ideally, beams from cores with the same fiber ID and core ID should be focused on the same position. If a core position is deviated by d from the ideal position due to (i) a misalignment between the bundled 5-CF collimator array and the micro-lens array and (ii) a rotational misalignment of a 5-CF in the holed Si substrate, the core deviation is observed as the shift of the beam position by Md. From the distribution of vertical and horizontal beam positions shown in Fig. 10, we estimated the standard deviation of the core position to be 1.4 μm. The coupling efficiency of a Gaussian beam η in the presence of misalignment d perpendicular to the optical axis is given by η = exp(−d 2 /w 2 ) , where w is a beam waist radius. Using 9.1 μm, which is the mode field diameter of the 5-CF used in the CSS, we estimated the CSS insertion loss attributed to the core position deviation in the bundled 5-CF collimator to be 0.5 dB. Fig. 11 shows the IL of the 15 cores for the connections from B 4 to all other bundles measured using amplified spontaneous emission (ASE) in the C-band, which excludes the IL of the fan-in fan-out (FIFO) device used for launching the ASE. Here, the FIFO device used for the experiments comprises four thincladding SMFs with a square lattice structure [27], which can access the four outer cores in a 5-CF. A standard SMF is used to access the center core in a 5-CF. ILs for cores in F 2 in B 3 are not available due to a fiber disconnection that occurred during the fabrication process of the collimator array. Although the ILs for 115 outputs for this first prototype vary from core to core (the Fig. 11. IL of the 15 cores for the connections from B4 to all other bundles measured using ASE in the C-band, which excludes the IL of the FIFO device used for launching the ASE. averaged and highest ILs are 2.5 dB and 5.3 dB, respectively), a fairly low IL at a level of approximately 1 dB is achieved for cores in particular 5-CFs (F 2 in B 1 , B 5 , B 6 , and B 8 ), which is close to the ray trace simulation results considering the lens transmittance and mirror reflectance. This indicates that if the alignment of optical components including rotational alignment of the 5-CFs is improved, we can expect that the ILs in the C-band for cores in all output bundles will decrease to the same level. Fig. 12 shows the IL and PDL characteristics for the connection from B 4 to B 1 as a function of the wavelength in the range from 1480 nm to 1630 nm. The IL and PDL include those of FIFO devices used for launching light from a wavelengthtunable laser diode to each core. An IL of less than 4.8 dB and a PDL of less than 0.5 dB are achieved for all cores across a very wide-wavelength range of 150 nm.

B. Intra-Bundle Crosstalk Characteristics
A MEMS mirror-based CSS has many advantages compared to liquid crystal on silicon (LCoS)-based conventional Fig. 13. Inter-core XT in the same output 5-CF bundle.
wavelength selective switches (WSSs): low insertion loss, fast switching speeds, no polarization dependent loss, and a wide wavelength range. It has another important advantage of no higher-order diffraction. Combined with the high output MCF pitch design enabled by the large tilt angle of a MEMS mirror, this feature yields a negligibly small amount of inter-port XT (large port separation) in a CSS. This is a significant difference from the case of a conventional flexible-grid WSS, which suffers from non-negligible inter-port XT due to the coupling of higher order diffractions originating from the imperfection of the phase grating written on the LCoS spatial light modulator. On the other hand, a major type of XT in a CSS is the inter-core XT in the same output MCF (same output 5-CF bundle in this case), where a portion of the optical signal power directed to the target core couples with other cores in the same 5-CF or in different 5-CFs in the same bundle as shown in Fig. 13(a) and (b), respectively. Fig. 14 shows the XT imparted to center core C 0 in F 2 in B 1 as a function of the wavelength, which includes the XT that occurs in FIFO devices arranged at the input and output sides of the CSS under test. We observe very low intra 5-CF XT imparted to C 0 in F 2 from the other four outer cores, C 1 to C 4 , in the same 5-CF, F 2 , at less than −52 dB as shown in Fig. 14(a). The inter 5-CF XT in bundle B 1 from cores C 0 to C 4 in different 5-CFs, F 0 and F 1, in the same B 1 bundle exhibits even lower XT of less than −64 dB as shown in Fig. 14(b). These XT values are much lower than those observed for the 5-CF CSS (−32 dB and −54 dB, respectively) in our previous reports [22], [24]. We attribute this significant improvement in XT characteristics to the introduction of the new collimator array design shown in Fig. 5 where a microlens array with AR coating and a 5-CF bundle array are fixed via a silica spacer block with refractive index matching adhesive, which may decrease stray lights in a CSS. We also measured the XT from cores in different bundles and found that inter-core XT from different cores is less than −68 dB.
The aggregated total XT from all the cores in the same bundle calculated using the results in Fig. 14(a) and (b) is −50 dB in the C-band and L-band and less than −48 dB in the S-band as shown in Fig. 14(c). Let suppose that (i) a 0.5 dB penalty for a quadrature phase shift keying (QPSK) optical signal caused by the total in-band XT of −20 dB [28] is acceptable and (ii) the same amount of −23 dB is assigned to both the cumulative XT generated by MCF links and the cumulative XT generated by SXCs. The experimental results indicate that we can transport QPSK optical signals through more than 150 SXCs with the route-and-select configuration, which is a sufficiently large number of nodes for practical networks.

C. BER Measurement of Switched WDM Signals
In order to confirm that there are no unknown deteriorating factors in the CSS prototype, we tested it by launching WDM signals into cores of the input bundle and routing them to the  corresponding core of the output bundles. Fig. 15 shows the experimental configuration for BER performance measurement. Due to the limitation of the available number of 5-CF 1×2 couplers, which are used to launch a WDM signal into the center core in a 5-CF while FIFO devices are used to launch four WDM signals into the four outer cores in a 5-CF, 10 cores among the 15 cores in the bundle of three 5-CFs are filled at once in this experiment. The WDM signal comprises a 100-Gb/s dual-polarization (DP) QPSK optical signal and dummy ASE spectrum to emulate a fully loaded spatial channel in the C-band.
The BER for the 100-Gb/s DP-QPSK signal exiting from the CSS prototype is measured using a digital coherent receiver while changing the optical signal-to-noise ratio (OSNR) of the received signal by controlling the power of the ASE loading. Fig. 16 shows the impact of XT on the BER performance of the employed digital coherent transceiver. Circles represent the OSNR penalties for the transceiver in the back-to-back configuration at the BER of 10 −3 , which are caused by the intentionally added XT signal. The dotted curve represents the theoretical curve given by , where erfc[·] is a complementary error function and XT is the XT power normalized by the signal power [29], [30]. The figure shows that the OSNR penalty increases with XT and the XT of −20 dB causes an OSNR penalty of 0.5 dB, which agrees with the results in [28]. Fig. 17 shows the BER for the 100-Gb/s DP-QPSK signal exiting from cores in B 4 of the CSS prototype as a function of the OSNR of the received signal. The BER curve for the 100-Gb/s DP-QPSK transceiver back-to-back configuration is also shown for comparison. Fig. 17(a) and (b) show the BER vs. OSNR curves for five cores in F 0 and F 1 in B 1 , respectively, when 10-WDM signals are simultaneously launched into cores in F 0 and F 1 in B 4 . Fig. 17(c) shows the BER vs. OSNR curves for five cores in F 2 in B 1 when 10-WDM signals are simultaneously launched into cores in F 1 and F 2 in B 4 . The figures show that the bundled 5-CF 1×8 CSS prototype can route at least ten 100-Gb/s DP-QPSK WDM signals in the C-band at once without any OSNR penalty. These observations confirmed that in the CSS prototype there are effectively no deteriorating factors including inter core XT and inter-wavelength channel XT.

IV. CONCLUSION
We showed the feasibility of a high core count CSS using the MCF bundling scheme by demonstrating a bundled 5-CF 1×8 CSS having 15 cores per port. The CSS prototype exhibited low IL of less than 4.8 dB, and a low PDL of less than 0.5 dB over an ultrawide wavelength range of 1480 nm to 1630 nm. The inter-core XT characteristics as a function of the wavelength showed that the intra 5-CF XT imparted to the center core from the other four outer cores in the same 5-CF is less than −52 dB and the inter 5-CF XT from cores in different 5-CFs in the same bundle is even lower at less than −64 dB, which are sufficiently low values for practical networks. No OSNR penalty in the 100-Gb/s spatial channel was observed when ten 100-Gb/s DP-QPSK WDM signals in the C-band were simultaneously routed by the bundled 5-CF 1×8 CSS prototype.