Not so far in the future, coherent technologies are expected to penetrate into access and intra-data-center networks, where the large capacity and high receiver sensitivity yielded by intradyne detection with a local oscillator (LO) will be maximally utilized. Laser diodes (LDs) need to have narrow linewidths and high stability. In addition, when applications require LDs with fast wavelength tunability [1,2], the attendant increase in transceiver cost can be hinder the introduction of these applications in cost-sensitive areas. One solution to this problem is loopback modulation, which is being eagerly investigated as a means to reduce operational and maintenance costs in passive optical networks (PONs) [3,4]. This approach provides a cost-effective means of eliminating the need for an LD at the transmitter, as an optical carrier generated in a central office (CO) is remotely intensity modulated at the user site and the resulting signal is launched to the CO.

Loopback modulation is effective not only in PONs but also when creating optical switching networks. High-port-count optical switches are expected to innovate intra-data-center networks, which are now confronted with problems stemming from Moore’s law saturation [5–7]. A large-port-count optical switch can be effectively created by the combined use of wavelength routing and space switches [7], and various architectures have been presented [8–10]. Wavelength routing can be implemented by placing a wavelength-tunable laser diode (TLD) at each switch input port (transmitter); the target switch output port is selected by setting the TLD wavelength. While data center networks need fast wavelength tunability O (µs), current fast TLDs pose several barriers to their application in data center, including cost, availability, reliability, and testability. We propose herein a new configuration, loopback modulation of LO light to remove the need for TLDs at the input ports. When a sender sets the TLD light at ${\lambda _i}$, it is delivered to the targeted output port i associated with ${\lambda _i}$. Inversely, the ${\lambda _i}$ light launched from output port i can be sent along the same route to reach the sender’s input port. This is because reciprocity is retained in an optical switch that relies on single-mode operation. In loopback coherent modulation, the LO light (${\lambda _i}$) supplied at output port i can be utilized as a carrier to be modulated in an external modulator set at the input port.

Polarization division multiplexing (PDM) enhances spectral efficiency in coherent systems where signals are mapped onto two orthogonal polarizations. In loopback modulation, the polarization of the light must be linear and aligned at the input port with the polarization axis of a dual-polarization IQ modulator (DP-IQM). However, the polarization state fluctuates randomly along the loopback path due to the birefringence of the transmission fibers and devices traversed. The use of polarization-maintaining fibers (PMFs) and devices removes the fluctuation, but PDM becomes unavailable. Polarization stabilization using passive optical components (e.g., a Faraday rotator) in front of modulators [4] alleviates the polarization fluctuation problem, but it halves the spectral efficiency due to single-polarization transmission. The challenge is to realize stable and cost-effective remote modulation for coherent PDM signals over standard single-mode fiber.

In this Letter, we propose a new polarization-alignment circuit (PAC) for loopback dual-polarization transmission that is applicable to optical switching systems. The proposed technique utilizes passive optical components but can ensure that the polarization state of seed LO light always remains aligned with the polarization direction of the DP-IQM. The LO light, the wavelength of which is dedicated to each output port, is supplied by a fixed-wavelength LD. Stable loopback modulation without power fading is cost-effectively achieved by employing a PAC and a silicon-photonic tunable filter (TF). Both theoretical and experimental analyses verify that the power fluctuation is reduced from 25 dB (no PAC) to 7.2 dB with the proposed technique. No significant degradation in Q-factor ($< $ 0.8 dB) is observed during a 100-min experiment, even though random polarization fluctuations at a rate of 5 rad/ms are applied. We demonstrate a 1,856 $\times$ 1,856 optical switch experiment using the proposed PAC for 256-Gb/s dual-polarization quadrature phase-shift-keying (DP-QPSK) loopback modulation. The measured bit error rates (BERs) are below the forward error correction limit (3.8 $\times$ 10⁻³). To the best of our knowledge, this is the first demonstration of loopback dual-polarization coherent transmission using a commercial DP-IQM.

Figure 1(a) depicts the proposed MN $\times$ MN optical switch architecture for coherent detection. The basic configuration closely follows our previously reported TLD-based architecture [8] and consists of M sets of N wavelength-tunable transmitters ${\mathrm{\lambda }_1}{-}{\mathrm{\lambda }_N}$, N M $\times $ M delivery-and-coupling switches (DCSs), and M N $\times $ N wavelength-routing (WR) switches. The transmitter generates an optical signal whose carrier wavelength is tuned to select the target output port. The N transmitted signals are delivered by the DCSs to each WR switch. The multiple signals are aggregated by an S $\times$ 1 optical coupler and then amplified by an erbium-doped fiber amplifier (EDFA) for loss compensation. Although the EDFA is a relatively expensive device, the per-port cost is reduced, as S input ports share one EDFA. After the amplified signals have been further combined by an (N/S) $\times$ 1 optical coupler, a 1 ${\times} $ N demultiplexer routes the N wavelength-multiplexed signals according to wavelength. Finally, the wavelength-demultiplexed signal is coherently detected by using an LO whose wavelength is dedicated to each port.

 

Fig. 1. (a) Proposed MN $\times$ MN optical switch architecture for coherent detection. (b) Loopback transmitter incorporating a passive polarization-alignment circuit (PAC). SEL: selector, DAC: digital-to-analog converter.

Download Full Size | PDF

By setting the wavelength and the switch state in a DCS, a path is established between arbitrary input and output ports without any blocking. Based on this configuration, a new optical switch architecture is developed that dispenses with the need for fast-tunable TLDs at transmitters. Continuous-wave (CW) light is generated by an LO at each receiver and looped back to an input port for modulation. Figure 1(b) illustrates the loopback transmitter formed by a silicon-photonic TF, PAC, and conventional TE-input DP-IQM. The input CW light is fed to the lower path by an optical circulator. One specific channel associated with the target output port is extracted from the loopbacked multiplexed N-wavelength light by a silicon-photonic TF. The PAC rotates the state of polarization (SOP) of the extracted LD light so that the incident x/y-linearly polarized light is transformed away from a single polarized state. The optical signal modulated by the DP-IQM is launched toward the receiver. To compensate for system loss, the output at the TF and modulator is amplified by compact and low-cost EDFAs for single-channel application that need no thermo-electric cooler. This yields a cost-effective and high-port-count optical switch that supports coherent detection without using TLDs. In the case of wideband WDM transmission, another TF is necessary after the DP-IQM to reject out-band amplified spontaneous emission (ASE) noise. The filter’s 3-dB bandwidth should be larger than the modulated signal bandwidth. We define the three ports of the 1 × 2 splitter/coupler as ports 1, 2, and 3. An optical signal input from port 1 is evenly split between ports 2 and 3. Conversely, acting as a coupler, signals input to ports 2 and 3 are combined and output from port 1 with a 3-dB loss.

The performance of the loopback transmitter is largely dependent on the polarization rotation caused by the optical switch. If the input polarization state is orthogonal with respect to the polarization axis of the IQMs, e.g., y-linearly polarized light is injected into the TE-input IQMs, we cannot obtain optical power. The PAC divides the y-linearly polarized light into two arms; one is converted into a circularly polarized state by a ${45^\circ }$ phase rotator (PR) while the other retains its y-polarization. Both lights are combined at a polarization beam combiner (PBC) followed by a $- {45^\circ }$ PR. Finally, the PAC outputs ${45^\circ }$ linear polarization light since the PBC passes the y-component of the circular polarization (the upper branch before the PBC) and rejects the y-linearly polarized light (the lower branch before the PBC). Therefore, stable modulation is realized, as the PAC mitigates the polarization mismatch and reduces the power variation due to the polarization fluctuation caused by fiber and device traversal by the light. Although this explanation assumes the injection of y-linearly polarized light against the TE-input IQMs (i.e., the worst condition without the PAC), the general behavior of the PAC is formulated below for arbitrary input polarization states. The loopback transmitter can be cost-effectively fabricated by integrating all the components together with semiconductor optical amplifiers on a single chip [11].

To assess the polarization dependency of the PAC, we derive an analytical expression for the output power. Even without high polarization-dependent loss, the incident (x/y-linearly polarized) light from an LO suffers random polarization rotation in reaching the loopback transmitter. At the PAC, the input electric field vector ${{\mathbf E}_{\textrm{in}}}$ is written, using the polarization rotation angle $\theta $ and the retardation $\varphi ,$ as ${{\mathbf E}_{\textrm{in}}} = \cos \theta {{\mathbf e}_\textrm{x}} + {\textrm{e}^{ - j\varphi }}\sin \theta {{\mathbf e}_{\textrm{y}}}$, where ${{\mathbf e}_\textrm{x}}$ and ${{\mathbf e}_\textrm{y}}$ are the unit vectors of the orthogonal polarization components. After the randomly polarized light passes through the ${45^\circ }$ PR, the electric field vector ${{\mathbf E}_{\textrm{out}}}$ at the output of the PBC is given by ${{\mathbf E}_{\textrm{out}}} = {E_{x,\,\textrm{out}}}{{\mathbf e}_\textrm{x}} + {E_{y,\textrm{out}}}{{\mathbf e}_{\textrm{y}}},$ where ${E_{x,\,\textrm{out}}} ={-} j\cos \theta /\sqrt 2 $ and ${E_{y,\textrm{out}}} = \; ({\cos \theta + {\textrm{e}^{ - j\varphi }}\sin \theta } )/2\; $ represent the electric field components of the x- and y-polarization components, respectively. The output power ${I_{\textrm{out}}}\; $ for the 45° angle used in modulation is expressed as

Validation experiments are carried out by launching light with arbitrary polarization into the loopback transmitter. The SOP of the input CW light is strictly controlled by a polarization synthesizer. At the DP-IQM output, we evaluate the power ratio normalized to the maximum measured power. A polarization rotation retardation of ${0^\circ }$ ($\varphi = {0^\circ }$) is set to minimize the power ratio (i.e., the power fluctuation is maximized in this case). Figure 2 shows the measured power ratio as a function of the polarization rotation angle $\theta $ with and without the PAC. As coplotted in Fig. 2, the Q-factor is also tested by modulating a 128-Gb/s DP-QPSK signal. Without the PAC, the power ratio decreases to $- $ 25 dB at around $\theta = {90^\circ }$ and the Q-factor varies from 24 dB to 14 dB. With the PAC, the power difference is suppressed to $- $ 7.5 dB [Fig. 2(b)]. Thanks to the PAC, the Q-variation becomes small ($< $ 0.72 dB), regardless of the polarization state. The measured power ratio agrees well with the calculated one.

 

Fig. 2. Measured power ratio and Q-factor versus the polarization rotation angle (a) without and (b) with the PAC at $\varphi = {0^\circ }$.

Download Full Size | PDF

The long-term stability of the loopback transmitter is examined by changing the input polarization scrambling at 5 rad/ms. This rapid SOP change assumes severe conditions comparable with a ∼1,000-km terrestrial link [12,13]. Figure 3 plots the measured power ratios with and without PAC recorded every minute for a 100-min period. Without the PAC, the output power fluctuated and faded to $- $ 25dB, while the PAC mitigated the power fluctuation to less than 9.8 dB. A slight discrepancy can be seen between the minimum power ratios of the static [${\approx}{-} $ 7.5 dB in Fig. 2(b)] and dynamic [${\approx}{-} $ 9.8 dB in Fig. 3] characteristics with PAC use. This can be attributed to the fiber-based optical components of the PAC, which are susceptible to external disturbances such as environmental temperature changes and stress applied to the fibers. The performance will be improved by integrating all PAC components. The transmission performance of a 128-Gb/s DP-QPSK signal is shown in Fig. 4, where the Q-factor is measured with the PAC under the same conditions as Fig. 3. The constellations displayed in the insets clearly show that we achieved polarization-insensitive operation, as the Q-variation was held to within 0.8 dB.

 

Fig. 3. Time evolution of the power ratio without and with the PAC.

Download Full Size | PDF

 

Fig. 4. Measured Q-factor transition with the PAC. Inset: constellations obtained in the best and worst cases.

Download Full Size | PDF

We conducted a proof-of-concept experiment to verify the effectiveness of the proposed optical switch and PAC. Figure 5 shows the experimental setup used to demonstrate 1,856 $\times$ 1,856 optical switching. A recently fabricated silicon-photonic TF [14] was utilized; the module and chip are shown in Figs. 5(a) and 5(b), respectively. Two identical filter modules were assembled in an off-chip polarization-diversity configuration [15], where each chip consisted of eight Mach–Zehnder interferometers connected in series. The measured 3-dB bandwidth and fiber-to-fiber insertion loss of the filter were 17 GHz and 8.9 dB, respectively. Loopback modulation was performed by using CW light [2 $\times$ 58 (N) channels] delivered from receivers [see Fig. 5(c)]. At the transmitter, the silicon-photonic TF selected one of the CW light carriers that matched the target LO wavelength (1547.12 nm) [see Fig. 5(d)]. After loss compensation by an EDFA, the extracted CW light carrier was driven by a DAC to produce a 32-Gbaud DP-QPSK signal. To simulate the worst case (the modulated output is minimized), the optical power was adjusted by the polarization controller (PC) placed in front of the PAC. After amplification, the transmitted signal was delivered to a 32 (M) ${\times} $ 32 (M) DC switch. The switched signal was coupled with spectrally shaped ASE (SS-ASE) light using a wavelength-selective switch (WSS). The resulting signal emulated 75-GHz-spaced 58-channel dual-carrier 256-Gb/s DP-QPSK signals covering the wavelength range from 1,530 nm to 1,565 nm [see Fig. 5(e)]. The WDM signal was sent to an S $\times$ 1 coupler and amplified by a two-way EDFA consisting of two EDFAs and couplers. Here, the saturation power P_S was defined at the output of each EDFA, and the same value was set for both directions. The amplified signal was further passed through a (58/S) ${\times} $ 1 coupler and demultiplexed by a 75-GHz-grid arrayed waveguide grating (AWG). The received signal was coherently detected by mixing it with LO light and then demodulated through offline digital signal processing (DSP). At the receiver, the LD light at 1547.12 nm was divided by a 3-dB splitter; one of the splitter outputs was used as an LO and the other was launched into the loopback transmitter. The output CW light was multiplexed with SS-ASE light [as shown in Fig. 5(c)] and propagated through the optical switch in the opposite direction to signal transmission.

 

Fig. 5. Experimental setup of the 1,856 $\times$ 1,856 optical switch ($N = $ 58, $M = $ 32). Photographs of the fabricated silicon-photonic TF (a) module and (b) chip. Measured optical spectra at (c) the loopback transmitter input, (d) the silicon-photonic TF output, and (e) the AWG input. (f) Measured BERs as a function of the two-way EDFA saturation power (P_S) for different sharing numbers (S).

Download Full Size | PDF

Figure 5(f) plots the measured BERs versus the two-way EDFA saturation power (P_S) for different EDFA sharing numbers (S). We achieved BERs of below 1 $\times$ 10⁻³ at EDFA saturation powers (P_S) of 19.8, 23.2, and 26.0 dBm for S= 1, 2, and 4, respectively. Analytically calculated curves are coplotted in the figure. When the saturation power is less than 24 dBm, economical optical amplification (S = 2) is possible. Thus, we confirmed that a cost-effective 1,856 $\times$ 1,856 optical switch can be created without the use of TLDs.

Tuning the LD wavelength is often considered a viable means of realizing wavelength-routing systems; however, fast-tunable LDs are not mature enough for practical application. The loopback configuration of fixed-wavelength LD lights can avoid the use of TLDs, provided polarization-related issues can be resolved. Our proposed PAC effectively realizes this, as verified by 1,856 $\times$ 1,856 optical switch experiments. The developed PAC is expected to be applied to not only the optical switch tested but also to various wavelength-routing systems to avoid the use of TLDs.