1. Introduction

Optical fiber communications have employed on-off-keying intensity modulation (IM) for about 40 years. Over the last ten years, coherent optical systems have enabled higher-capacity information to be encoded in four dimensions, namely amplitudes and phases in two polarizations [1]. This enables ~1 Tb/s channel rate [2]. On the other hand, IM with direct detection (IMDD) with complex modulation formats also receives significant intention, pushing the channel rate to 100-200G by employing multi-level pulse amplitude modulation (PAM) [3,4], discrete multi-tone (DMT) modulation [5–10] or carrierless amplitude phase (CAP) modulation [11]. Further increase of the channel rate can be realized by multi-dimensional IMDD. Recently, there is increasing interest in utilizing Stokes vector (SV) receiver, which is a DD technique with the capability to digitally track the polarization changes in fibers and decode information in two to four dimensions [12–19]. Therefore, it can serve as an intermediate bridging technology between conventional intensity and coherent optical systems. The DD technology is particularly attractive for short-reach interconnects operating at 100 Gb/s and beyond for metro and data center applications [20,21], since these applications are very sensitive to cost [22].

The SV receiver requires several complex and precise optical components, far more complicated than a single photodetector (PD) used in other DD techniques. This receiver complexity poses challenges for its practical implementation. In this paper, we report a highly compact and monolithically integrated silicon photonic (SiPh) SV receiver. SiPh is an emerging disruptive optical device technology with the promise to provide compact, low-cost, and low-power optical transceivers [23]. Its principal technical merits include high levels of optical integration and simple implementation of polarization diversity, making it an ideal platform to implement a SV receiver.

Together with a silicon in-phase/quadrature (I/Q) modulator fabricated on the same wafer, we demonstrate the generation and detection of a 32-Gbaud 16-ary quadrature amplitude modulation (16-QAM) signal, and its transmission over 100-km fiber with a bit error ratio (BER) less than the threshold for the hard-decision forward error correction (HD-FEC) with a 7% overhead. The salient features of this demonstration are: (a) we implement the first monolithically integrated SV receiver, significantly promoting SV-DD to be a practical technique; (b) the 128-Gb/s line rate for 100-km transmission sets the record for SiPh devices with DD (monolithic SiPh modulators and detectors) to the best of our knowledge; (c) both the modulator and receiver chips were fabricated on the same wafer, allowing for future single-chip ultra-compact and low-cost transceivers.

2. Silicon photonic integrated circuits (PICs)

We first explain the working principle of the SiPh SV receiver. The SV is defined as [S₁, S₂, S₃]^T = [|E_x|²-|E_y|², 2Re(E_xE_y*), 2Im(E_xE_y*)]^T, where E_x and E_y are electrical fields in two orthogonal polarizations. Hence, the SV is determined by the amplitudes in two polarizations and the inter-polarization phase. The first parameter S₁ can be measured by detecting the optical powers in two polarizations, while the second and third parameters require mixing the electrical fields in a 90-degree mixer. The circuit used is shown in Fig. 1(a). The optical signal is first split into TE/TM by a polarization beam splitter (PBS), then each polarization is power-divided into two portions. One of TE branches is detected by a PD and the other enters into a 90-degree mixer. For the TM lights, they are rotated to TE first by polarization rotators (PRs), then one branch is detected by a PD with the other entering the mixer. The four outputs of the mixer produce the S₂ and S₃ with balanced detections. Figure 1(b) presents a photograph of the SiPh chip, with a size of 2 mm x 3 mm. This chip integrates all the components required in Fig. 1(a). The PBS is implemented based on a silicon waveguide directional coupler where the PR utilizes the adiabatic mode conversion to convert TM₀ mode to TE₁ mode and then coupling to TE₀ mode [24,25]. We use a 4x4 multimode interference coupler (MMI) to provide a 90-degree hybrid function. There are six single-ended germanium PDs. More detailed device information can be found in [26]. The SiPh receiver chip is packaged with a cleaved fiber and printed circuit boards for RF access, shown in Fig. 1(c).

 

Fig. 1 Silicon photonic integrated circuits. (a) and (b) Optical circuit and photograph of SiPh SV receiver. (c) Photograph of a packaged SV receiver with PCBs and fiber. PBS: polarization beam splitter, PR: polarization rotator, PD: photodetector, MMI: multimode interference coupler, PCB: printed circuit board. (d) and (e) Optical circuit and photograph of a SiPh I/Q modulator with a power-tunable carrier in the orthogonal polarization. (f) Photograph of the packaged I/Q modulator chip with a PCB. TC: tunable coupler.

Download Full Size | PDF

On the transmitter side, the information is encoded in one polarization, while the other is a continuous-wave (CW) carrier [12]. With the CW carrier, the polarization changes in fibers can be tracked by the SV detection. Although it is possible to encode information in both polarizations [15, 17], this transmitter scheme allows digital dispersion compensation. The silicon PIC to realize this configuration consists of a tunable coupler (TC), a PR, an I/Q modulator and a polarization beam combiner (PBC), shown in Fig. 1(d). It is to be noted that this PIC is much more complicated than just a single silicon modulator used in [15, 17]. After TE light is launched into the chip, the TC can be adjusted to split the power between the I/Q modulator and the CW path. In the CW path, the PR is used to rotate the TE mode to the TM mode, which is then combined with the modulator output using the PBC. The TC is realized by using a Mach-Zehnder interferometer (MZI) with tunable phase elements. The silicon MZMs are formed by using two waveguide phase shifters with embedded reverse-biased pn junctions in a single-drive push-pull configuration [26]. The waveguides in the MZMs have a cross section of 0.5 μm x 0.22 μm and a slab thickness of 90 nm. The phase shifter lengths in the MZMs are 5.5 mm and the modulation efficiency is ~2.4 V-cm, resulting in a V_π of ~4.5 V. Figure 1(e) and 1(f) exhibit the photographs of the PIC and the packaged device, with a PIC size of 6 mm x 2 mm.

Figures 2 and 3 show measured characteristics of both the receiver and modulator PICs. The fiber-to-PD responsivity spectra are shown in Fig. 2(a) for six PDs and for two polarizations. The polarization extinction ratios are more than 20 dB, shown from the responsivity differences between TE and TM for PD1 and PD2. The total responsivity is ~0.23 A/W for the TE mode and 0.18 A/W for the TM mode at 1540 nm, including both the coupling and on-chip losses [see Fig. 2(b)]. The responsivity drop for longer wavelength is mainly attributed to absorption decrease of germanium. The additional loss for the TM mode comes from the excess loss of the PBS and PR. For the modulator PIC, the total insertion loss is about 10 dB when the light goes through the CW path (including both the fiber coupling and on-chip losses), while it is about 17 dB if all the input light is tuned to the I/Q modulator path, shown in Fig. 3(a). The transmission ripple in the modulator path is attributed to the length difference between I and Q. Figure 3(b) shows the MZM’s frequency response with a 3-dB bandwidth of ~18 GHz at a reverse bias of 2 V.

 

Fig. 2 Fiber-to-PD responsivities. (a) Responsivity spectra of six PDs in the SV receiver, blue and red lines represent TE and TM, respectively. (b) Total responsivities for TE and TM inputs.

Download Full Size | PDF

 

Fig. 3 (a) Transmission spectra of the SiPh modulator PIC when the input light is adjusted to CW path (blue) and modulator path (green). (b) Small-signal electro-optic (EO) frequency response of the silicon MZM under different reverse biases.

Download Full Size | PDF

3. 32-Gbaud 16-QAM modulation, detection and transmission

Figure 4(a) illustrates the experimental setup to generate and detect 32-Gbaud 16-QAM signals. On the transmitter side, a CW laser at 1550.12 nm with a power of + 14 dBm is launched into the SiPh modulator PIC. The two MZMs are driven by Nyquist-shaped RF signals, each generated by a 8-bit digital-analog-converter (DAC) with a sampling rate of 88 GSa/s and amplified to a peak-to-peak voltage swing of ~3.7 V. We adjust the TC so that the carrier to signal power ratio is about 1:1. The output of the transmitter PIC is amplified by an erbium-doped fiber amplifier (EDFA), and then launched into a standard single mode fiber (SSMF) with a power of ~ + 6 dBm (The launch power consists of power for both the 16-QAM signal and the carrier. The + 6 dBm is the optimal based on our measurement.). At the receiver, the signal is first amplified by another EDFA and then filtered by an optical band-pass filter. The received signal with a power of about + 14 dBm is launched into the SiPh SV receiver. The output from the SV receiver are collected by six analog-to-digital converters (ADCs) embedded in real-time oscilloscopes with a sampling rate of 40 GSa/s and an analog bandwidth of ~17 GHz. Balanced detection is accomplished with the signal subtraction between the two sampled channels. Offline digital signal processing (DSP) is carried out, as explained below.

 

Fig. 4 (a) Optical setup to generate, transmit and detect a 32-Gbaud 16-QAM optical signal. (b) Flow chart of offline digital signal processing (DSP).

Download Full Size | PDF

The DSP procedures follow those in [12]. In the Stokes space, the received SV is related to the transmitted SV $T$ by the equation $\Re =H\cdot T$_, where $H$ is a 3x3 real-valued matrix determined by the polarization rotation in fiber. The rotation matrix $H$can be calculated in Stokes Space with the assistance of training sequences [12]. After $H$is obtained, the transmitted $T$ can be recovered from that is measured by the SiPh receiver. Then the full-field information of the transmitted signal can be obtained from $T$. Digital dispersion compensation and a least mean squares (LMS) equalizer are applied to the 16-QAM Nyquist-shaped signals. A digital phase locked loop is implemented in the LMS equalizer to perform clock recovery. No carrier or phase recovery is needed [12]. Figure 4(b) presents the flow chart of the DSP procedure.

The measured BER as a function of optical signal-to-noise ratio (OSNR) is shown in Fig. 5(a) for a back-to-back link and through a 100-km fiber span (the span loss is ~20 dB). The required OSNRs for BERs of 2.4 x 10⁻² and 3.8 x 10⁻³ (20% and 7% FEC threshold, respectively) are 22.4 dB and 26.6 dB, respectively, and increase to 23.9 dB and 29.0 dB after transmission. The transmission penalty at a BER of 2.4 x 10⁻² is therefore 1.5 dB. The recovered constellations for the back-to-back case and after 100 km transmission are presented in Figs. 5(b) and 5(c) (each with an OSNR of 33.3 dB and 31.2 dB). The BER and SNR for the back-to-back link are 5 x 10⁻⁵ and 19 dB. After 100-km transmission, they become 1.4 x 10⁻³ and 16.4 dB. After 100-km, the BER is still below 3.8 x 10⁻³, the threshold of 7% overhead FEC.

 

Fig. 5 Experimental results of 32-Gbaud 16-QAM modulation, detection and transmission. (a) BER versus OSNR for back-to-back (b2b) and 100 km. (b) and (c) Constellations for b2b and 100 km.

Download Full Size | PDF

4. Discussions and conclusion

In conclusion, we have demonstrated a monolithically integrated SiPh SV receiver along with an integrated silicon I/Q modulator and use the pair to demonstrate 128-Gb/s transmission over 100-km SSMF fiber. Since there is still margin for 20% FEC threshold, a longer transmission distance is possible. If transimpedance amplifiers (TIAs) are used, the EDFAs can be replaced by low-cost semiconductor optical amplifiers (SOAs). This SV receiver can also be used to detect dual-polarization signals with information encoded in three dimensions, allowing reception of >300 Gb/s as demonstrated in [17]. Additionally, it can be used to characterize fast polarization changes in fibers.