Reconfigurable polarization processor based on coherent four-port micro-ring resonator

3.1 Device characteristics

The experimental setup is shown in Figure 1(d). Light from a continuous-wave laser was split into two paths and used as the optical source for two transmission lanes formed by two polarization channels (CH1 and CH2). On each path, a polarization controller (PC) was used to adjust the polarization before combining the two channels using a fiber-optic packaged polarization beam splitter (PBS). In each path, we used a variable optical attenuator (VOA) to vary the power and measure the scattering matrix of the system. In this way, two independent data channels were coupled to the orthogonal modes in the single mode fiber. After the PBS, another PC enabled the setting of different test input polarizations orientations before coupling into the device under test (DUT). A simplified schematic of DUT is shown in Figure 1(c). Before applying the control algorithm on the two PS, the performance of the MRR and PS was characterized independently. Figure 2(a) and (b) shows the transmission spectra of the fabricated MRR, the resonance of the MRR was fitted with the Lorentz curve, and the loaded quality (Q) factor of the MRR was measured to be 1185. The tunability of the MRR is shown in Figure 2(d) and (e). The power consumption to tune the MRR over one free-spectral range was 59.5 mW. Figure 2(f) shows transmission spectra from port “IN” to port “OUT1.” With electrical power of 49.6 mW, the transmission spectra repeated, i.e., the input waveguide PS was tuned from 0 to 2π with 49.6 mW power.

Figure 2:

Device characteristics. (a) Transmission spectra of the through port of MRR. (b) Transmission spectra of the drop port of MRR. (c) Lorentz fit of the resonance of MRR, the quality factor was extracted to be 1185. (d) Transmission spectra of the through port of MRR and (e) transmission spectra of the drop port of MRR when applied with voltage from 0 V to 3 V. (f) Transmission spectra from port “IN” to port “OUT1” in Figure 1(c) when the PS was applied with a voltage range from 0 V to 3 V.

3.2 Polarization descrambler and switching

We performed two experiments to demonstrate the polarization switching and the polarization mode unscrambling using the whole system, respectively. To facilitate the testing of polarization switching, the input fiber polarization controller was manually adjusted to maximize the coupling from TE mode into port “OUT1” (S₁₁) and from TM mode into port “OUT2” (S₂₂) as shown in Figure 3(c). To achieve polarization switching, we used particle swarm optimization to optimize the applied voltages on the PS and MRR in order to completely switch the output to the other output port “OUT2.” During the optimization process, we set the FOM = S₁₂ + S₂₁ − S₁₁ − S₂₂. The value of FOM evolved with optimization epochs are shown in Figure 3(a). S _ij represents power transmission from port “CHi” to port “OUTj”. Scattering matrix S _ij evolved with epochs was shown in Figure 3(b). FOM value was optimized from −40 dB to over 40 dB. This shows that the system can be configured as a polarization switch with polarization extinction ratio of 16 dB. The optimized scattering matrix is plotted in Figure 3(d) with crosstalk level < −16 dB. To achieve mode unscrambling, we used particle swarm optimization to optimize the applied voltages on the PSs. During the optimization process, we set the FOM = min((S₁₂ − S₁₁), (S₂₁ − S₂₂)). The value of FOM evolved with optimization epochs was shown in Figure 3(e). Scattering matrix S _ij evolved with epochs was shown in Figure 3(f). FOM value was optimized from −7 dB to 21 dB. The scattering matrix after optimization is plotted in Figure 3(d) with crosstalk level < −21 dB. This shows that two arbitrary input orthogonal modes in the fiber, irrespective of external disturbances and rotation in the fiber transmission, can always be unscrambled and restored to two different spatial outputs with crosstalk level lower than −21 dB in the proposed system. The same algorithm based on pilot tones can be used to adapt dynamically to changes in the state of polarization during the period of high-speed data transmission of a data packet. Continuous tracking of changes in the polarization state is limited by the maximum phase shift of about 2π in the integrated thermo-optic phase shifters, as there is a need to reset to zero phase shift beyond 2π, but the reset can be carried out during the initial setup for data packet transmission. The polarization processor can be suitable for data packet transmission such as the 100 G Ethernet standards used in data centers. The connection setup to establish the two lanes of data transmission can be established during the data link initiation protocol using the algorithms we mentioned above. The results also show that arbitrary polarization can be coupled to an optical fiber from the 2D GC if we operate the device in reverse and send light from port “OUT1” or port “OUT2” back to the 2DGC, because we can always tune the ratio and phase difference of the x- and y-polarization launched into the fiber.

Figure 3:

The training process of mode switching. (a) FOM value evolved with epoch number. (b) Scattering matrix evolved with epoch number. (c) Scattering matrix before switching and (d) scattering matrix after switching. (e–h) The training process of polarization unscrambling. (e) FOM value and (f) scattering matrix evolved with epoch number. Scattering matrix (g) before and (h) after polarization unscrambling.

3.3 High speed data transmission

One question we wish to address experimentally is to assess the limitation on signal bandwidth introduced by the use of the MRR in the polarization processor, and whether the polarization processor can be used to unscramble the pol-mux data lanes in a high-speed optical communication system. The important parameter in this application scenario is the optical bandwidth of the polarization processor. The optical bandwidth is determined by the optical path length difference within the system and the Q of the MRR (which had a loaded Q of 1185). Intersymbol crosstalk will be introduced if the delay time is too large. The system should have large bandwidth (small path length difference) if it was aimed at supporting high-speed data communication. The bandwidth of MRR Δf can be estimated by the photon life time τ within the MRR by Δf = 1/τ. Photon lifetime τ of an MRR can be derived from the Q of the MRR by τ = n_effQλ/πc, where n_eff represents the effective index of waveguides, λ is the resonance wavelength. In this scenario, the system’s bandwidth was estimated to be around 200 GHz. To further investigate the bandwidth and the ability of the system to support high-speed communications, we carried out the experiment as shown in Figure 4(c). In Figure 4(c), we generated 40 GHz none-return-to-zero pseudo-random binary sequence signals and fed the two channels of pol-mux 40 GHz data into the optical fiber transmission link. On the receiver side, the two output channels were monitored by the power meters and the high-speed real-time oscilloscopes. Figure 4(a) and (b) shows the spectra of S _ij before and after being unscrambled by the MRR-based coherent network. Figure 4(d) shows the eye diagram generated by the Mach–Zehnder Modulator (MZM). Figure 4(e) and (f) shows the eye diagram acquired before being processed by the network, while Figure 4(g)–(h) shows the eye diagrams after the unscrambling of the data lanes by the integrated polarization processor. The unprocessed eye diagram is completely closed because the arbitrary choice of orthogonal polarization basis at the receiver does not match the original transmitter polarization basis and severe crosstalk is present in the two unprocessed output channels. However, after processing by the integrated 4-port MRR coherent network, the eye diagrams become opened by the reduction of crosstalk. Besides, we can initially observe that the shape of the eye diagram was not changed after transmitting through the optical system, which shows that the 4-port MRR processor can support high-speed optical communication at speed of at least 40 GHz. Furthermore, we applied this setup to showcase an all-optical transceiver system to implement PAM4 and polarization multiplexing in a direct-detection transceiver operating at 400 Gb/s [45]. The experimental results are consistent with system being transparent to both data rates and modulation formats for signals that occupy a frequency band under 200 GHz.

Figure 4:

Transmission spectra of S _ij (a) before and (b) after being unscrambled by the coherent network. (c) Experimental setup of high-speed optical communication system using coherent network to do polarization unscrambling. TLD, MZM, PRBS, PM, EDFA, and OSC represent tunable laser diode, Mach–Zehnder modulator, pseudo-random binary sequence generation pattern, power monitor, erbium-doped fiber application amplifier, and high-speed oscilloscope. (d) Eye diagram generated by the MZM without going through the PIC system. Eye diagram before being scrambled by the optical system acquired by (e) OSC1 and (f) OSC2. Eye diagram unscrambled by the optical system acquired by (g) OSC1 and (h) OSC2.

3.4 Polarization analyzer

3.4.1 Division-of-space polarization analyzer

In this section, we show that the 4-port coherent MRR system can be configured as a DOS polarization analyzer. The schematic of the system is shown in Figure 5(a) and (b) while the experimental setup is shown in Figure 5(c). It can be proved that there exists a linear relationship between the Stokes parameters vector S and the four measured spatial outputs I (from O1 to O4) related by the matrix equation S = TI. The transmission matrix T can always be calibrated by four known polarization states. Then, any polarization state can be retrieved by measuring the spatial intensity vector I. The normalized Stokes parameter can be retrieved by S = TI and S ⃗ = ( S 1 , S 2 , S 3 ) / S 0 . Figure 5(d) illustrates the Poincare sphere containing the recovered and reference points. The reference parameters are those defined by the commercial polarization generator/analyzer. Figure 5(e) depicts the recovered Stokes parameters, and Figure 5(f) shows the corresponding deviations from the reference parameters. One can see that the systems work well with only small measurement errors: the deviation of the orientation angle (2ψ) measured by the polarization analyzer varied from −0.5° to 1.3° and the deviation of ellipticity angle (2χ) varied from −0.98° to 7.27°.

Figure 5:

Division-of-space polarization analyzer. (a) Micrograph image and (b) schematic of the system when it was configured as a DOS polarization analyzer. (c) Experimental setup of the polarization analyzer. (d) The Poincare sphere shows the recovered polarization state. (e) The recovered Stokes parameters and (f) deviations of polarization parameters.

3.4.2 Division-of-time polarization analyzer

In this section, we show that the 4-port MRR system can be configured as a DOT PA. The schematic of the system is shown in Figure 6(a) and (b) while the experimental setup is shown in Figure 6(c). There exists a linear relationship between the Stokes parameters vector S and the four measured temporal output I (from O1) related by S = TI. The transmission matrix T can also be calibrated by four known polarization states. Then, any polarization state can be retrieved by measuring the temporal intensity vector I. The normalized Stokes parameter can be retrieved by S = TI, and Figure 6(d) illustrates the Poincare sphere containing the recovered and reference points. The reference parameters are those defined by a commercial polarization generator/analyzer. Figure 6(e) depicts the recovered Stokes parameters, and Figure 6(f) shows the corresponding deviations from the reference parameters. One can see that the measurement error of orientation angle relative to the reference is slightly worse than the DOS measurement: the DOT measurements had deviation of the orientation angle (2ψ) varied from −2.93° to 3.49°. However, the DOT measurement had slightly better performance than the DOS measurement for the ellipticity angle (2χ): the DOT measurement had deviation of ellipticity angle (2χ) that varied from −3.5° to 3.05°.

Figure 6:

Division-of-time polarization analyzer. (a) Micrograph image and (b) schematic of the system when it was configured as a DOT polarization analyzer. (c) Experimental setup of the polarization analyzer. (d) The Poincare sphere shows the recovered polarization state. (e) The recovered Stokes parameters and (f) deviations of polarization parameters.

4.1 Discussion

Table 1 summarized and compared the state-of-the-art techniques that were utilized for polarization processing. Comparing with the traditional method using coherent receiver with DSP, the advantages of using integrated processor for polarization processor is lower latency and lower power consumption for the high data rate signals. Compared with the MZI-based integrated photonic processor, the MRR-based photonic processor has smaller footprint with rings of radius as small as 3 μm and possible of high-speed configuration without further sacrificing footprint. The other advantage is due to the smaller finesse (F) of MRR, the thermal power required to tune from 0 to 1 in MRR is finesse (F) times lower than that required for MZI. F is defined as the ratio of the FSR (Free Spectral Range) and the resonance full width at half maximum (FWHM), i.e., F = FSR/FWHM. It should be noted that normally F should be larger for MRRs compared to MZI and increases with the resonance Q-factor. The disadvantage of MRR is that the resonances are sensitive to fabrication errors. This is a well-known and well-studied problem. To address fabrication errors, there are three primary solutions commonly employed. One approach involves phase trimming [46] by laser annealing of germanium-implanted waveguides. Additionally, employing lithography with a higher resolution (<45 nm) can mitigate fabrication errors and enhance the consistency of MRR characteristics [47]. While these methods can reduce the spread of resonances of MRRs, thermal heaters are needed anyway to realize reconfigurable circuits. Notably, efforts have been undertaken to employ feedback loops for adaptive tuning of MRRs, aimed at mitigating the impact of environmental fluctuations [48].

4.2 Conclusions

In conclusion, we have proposed and experimentally validated a novel high-performance polarization processor by using coherent 4-port MRR integrated with a 2DGC. The 4-port MRR can be used to tune the splitting ratio and phase difference between two polarization components and, therefore, can be configured to control and monitor the polarization states. The polarization processor can be used as polarization descrambler, polarization switch, polarizers, and polarization analyzer. This paper reports the first experimental demonstration of the coherent 4-port MRR for polarization mode switching and polarization mode unscrambling. We demonstrated how two pol-mux data channels can be recovered after transmission in nonpolarization maintaining optical fiber (introducing arbitrary polarization rotation during transmission) and that the integrated receiver can reduce the crosstalk level between the two data lanes to below −21 dB. These results show that pol-mux communications can be used for direct-detection optical fiber interconnects in data centers, without the need for using polarization maintaining optical fibers nor high-speed DSP to recover the transmitted polarization channels. The device demonstrates the successful use of the 4-port coherent network for matrix multiplications to process 40 GHz polarization-multiplexed communications signals. For the polarization analyzer, the DOS polarization analyzer achieved good accuracy in the measurement of the orientation angle (2ψ), with measurement errors varying from −0.5° to 1.3°, and reasonable accuracy in measurement of the ellipticity angle (2χ), with errors varying from −0.98° to 7.27°. The DOT polarization analyzer achieved the deviation of the orientation angle (2ψ) is from −2.93° to 3.49° and deviation of ellipticity angle (2χ) is from −3.5° to 3.05°.