1. Introduction

Due to the wide spreading of broadband communication systems, such as mobile front/back haul and cloud services, demands on the capacity, reach and splitting ratio of optical access network are growing continuously. One technology that simultaneously provides good power budget, high capacity/speed, and being compatible with current fiber plant is coherent detection [1–3]. The coherent dense or ultra-dense wavelength-division multiplexing PON (DWDM/UDWDM-PON) is capable of providing ≥1-Gb/s access speed to individual subscribers, with Terabit-class aggregated capacity. Besides, there is no need for optical filtering or amplifying at the optical network unit (ONU), achieving both high flexibility and sensitivity. For instance, N. Cvijetic et al. have demonstrated 1.92-Tb/s coherent DWDM PON using the orthogonal frequency-division multiplexing (OFDM) technology without high-speed ONU-side electronics [4]. And, A. Shahpari et al. have demonstrated Terabit-class coherent UDWDM-PON using Nyquist shaping technology over 12.8-nm band [5].

Although available coherent technology for the core networks offers sophisticated and high performance transceivers [6–8], other alternatives should be explored to trade-off performance with simplicity and low-cost criterions in the access network. One possible solution is to use the low-cost tunable laser, such as the distribution feedback (DFB) laser. The main challenge is that the DFB laser usually takes unacceptable high linewidth and instable frequency. To solve such problem, M. Presi et al. proposed a novel 3x3 optical coupler based coherent scheme for 1.25-Gb/s on-off keying (OOK) signal [9]. We have demonstrated a computationally-efficient bi-directional 2.5-Gb/s/ch coherent UDWDM-PON with 4-level pulse amplitude modulation (PAM-4) and heterodyne detection [10]. Both methods employ the intensity modulation and coherent envelop detection for the amplitude-shift keying (ASK) signal. Error-free detection is achieved without any carrier frequency or phase estimation. Another approach to reduce the overall cost is to simplify the transmitter structure of ONU. The direct-modulation scheme is cost-effective, such as using the directly modulated DFB laser or vertical cavity surface-emitting laser (VCSEL) [11–13]. However, one additional laser source is required in ONU to act as the local oscillator (LO) for downlink coherent detection.

On the other hand, the external modulators using silicon photonics, such as silicon based Mach-Zehnder modulator (MZM) may be promising due to its advantages of low-cost, high reliability, and high density integration by CMOS fabrication processes [14]. In this paper, we propose a novel coherent PON scheme by employing the linear silicon MZM for cost reduction. Intensity modulation of DMT signal and coherent detection is used to achieve both high SE and receiver sensitivity. We also propose a digital carrier regeneration (DCR) method to further reduce the CSPR (carrier-to-signal power ratio) of intensity modulated signal. Compared to conventional DMT intensity modulation scheme, the receiver sensitivity is increased by about 16 dB with proposed DCR method due to its low CSPR. No carrier frequency offset and phase estimations are needed in the receiver, which greatly reduces the complexity for both laser and DSP. Finally, a 10-Gb/s/ch uplink transmission is experimentally demonstrated with silicon MZM over 20-km SSMF, achieving about −44dBm sensitivity with 7% FEC overhead.

2. Silicon Mach-Zehnder modulator

The external modulator based on silicon photonics is a key aspect for the proposed coherent PON, which enables both high density integration and low cost of ONU. Besides, a common laser can be used as both the optical carrier and LO, simultaneously. In this paper, a 2-mm silicon MZM is designed based on the balanced Mach-Zehnder interferometer (MZI) structure, which is embedded with reverse biased PN junction. Figure 1(a) illustrates the cross-section diagram of the single push-pull PN junction phase shifter with key dimensions annotated. The single push-pull operation setup is adopted to reduce the modulation-induced chirp [15]. The phase shifter uses rib waveguide with a slab thickness of 90 nm and waveguide height of 220 nm. The junction location is shifted by 50 nm from the center of the waveguide to reduce the loss and driving voltage by maximizing the interaction between the carriers and the optical field. The doping concentrations are 10¹⁸/cm³ for p-dopants and n-dopants in the PN junctions, and 10²⁰/cm³ for both P + and N + contact regions. The ${\text{V}}_{\pi }\cdot {\text{L}}_{\pi }$ of phase shifter is 1.7 V·cm.

 

Fig. 1 (a) Cross-section diagram of the single push-pull reverse biased PN junction phase shifter; (b) effective index (Neff) and characteristic impedance (Zc) results from finite element method simulation.

Download Full Size | PDF

Figure 1(b) shows the RF effective index (Neff) and characteristic impedance (Zc) of the travelling electrode. 3D electromagnetism simulation is used to extract the loaded RF transmission line scattering parameters. The RF effective index is about 3.7, which is very close to the optical group index of 3.8 and the characteristic impedance is 50 Ω. The effective index and impedance match well at the same time. The device uses grating couplers to couple light into and off the chip. The whole on-chip loss is 3.6 dB.

3. Digital carrier regeneration

Typically, directly modulated optical signal is not power efficient, since strong optical carrier exists in the generated signal. To achieve higher aggregated capacity in the power splitter based UDWDM-PON, such modulation scheme will reduce the effective signal power, thus limiting the OSNR and maximum optical distribution network (ODN) loss budget. To solve such problem, a carrier suppressed DMT signal is generated by external modulation based on the designed silicon MZM. The silicon MZM is biased around the null point to generate a bi-polar intensity signal. With coherent detection, such signal can be recovered in digital domain using the proposed DCR method.

Figure 2(a) shows a schematic diagram of the generated optical DMT signal in our DCR scheme. Note that a guard band is reserved at the inner subcarriers of DMT signal to accommodate the residual carrier. Figure 2(b) shows the flow chart of proposed DCR method. Suppose ${\overline{r}}_n$ represents the received signal after analog-to-digital conversion. A digital low pass filter (LPF) is first applied to divide ${\overline{r}}_n$ into two parts. The first part is the residual carrier, which is then amplified by the digital amplifier. The amplified digital carrier is represented as

 

Fig. 2 (a) Schematic diagram of the optical DMT signal in proposed scheme; (b) flow chart of the proposed DCR method.

Download Full Size | PDF

The DCR method may look similar with the well-known RF-pilot tone methods [16–18], as shown in Fig. 3(a). The basic ideas behind them are different. The RF-pilot tone method aims to compensate the frequency offset and phase noise directly from the phase of filtered pilot tone. But, the DCR method aims to restore the intensity of DMT signal, while the frequency offset and phase noise is removed by envelop detection (see Eq. (3). Since we focuse on how to reduce the cost, we have to assume the use of low-cost lasers with large linewidth (typically > 1 MHz) in the ONU. Besides, the DCR method should be robust to the noise by digital filtering. Under the above concerns, a comparison is done by simulation to show the advantages of DCR method over the RF-pilot tone method, as shown in Fig. 3(b). The filter bandwidth is 200 MHz, which is large enough to capture the RF-pilot or residual carrier. The size of guard band is 500 MHz. The above parameters are the same for both methods. The Q-factor versus CSPR curves are measured, under the linewidth of 10 kHz, 100 kHz, and 1 MHz, respectively. For simplicity, the same effective OSNR of 12 dB (excluding the residual carrier) and 16-QAM bit loading is assumed in all the measurements. From Fig. 3(b), we can see that the DCR method outperforms the RF-pilot method under the same linewidth and CSPR (when CSPR > −16 dB). Linewidth is more critical to the RF-pilot method. The Q-factor degrades dramatically, when the linewidth of RF-pilot method becomes larger than 1 MHz. Increasing the CSPR does not improve the Q-factor of RF-pilot method effectively. However, the DCR method shows satisfying robustness to the linewidth and unfiltered ASE noise, when CSPR is high enough.

 

Fig. 3 (a) Flow chart of the RF-pilot tone based method; (b) comparison between DCR and RF-pilot tone method under different linewidth and CSPR.

Download Full Size | PDF

4. Experimental setup

Figure 4 shows the experimental setup for the 10-Gb/s/ch uplink transmission of DMT signal over 20-km SSMF using silicon MZM and proposed DCR method. Firstly, a continuous wave (CW) operating around 1552 nm with 16-dBm output power is used as the optical carrier, which is then modulated by the silicon MZM. The CW linewidth is 100 KHz. The silicon MZM is biased around the null point to produce a carrier suppressed DMT signal. For comparison, we also generate a DMT signal in the conventional way, where the silicon MZM is biased in the linear modulation region. Figures 4(a) and 4(b) shows the optical spectrums of the carrier suppressed DMT signal and conventional intensity modulated DMT signal, respectively. The driving voltages for these two cases are 4V and 2V, accordingly. The arbitrary waveform generator (AWG) works at 12 GSa/s. The FFT size is 512, in which 116 are loaded with 16-QAM signal. The first 6 subcarriers are reserved as the guard band. Hermitian symmetry is used to produce the real-valued DMT signal. The length of cyclic prefix (CP) is 1/128 of the FFT size. There are 6 training symbols, which are followed by 96 payload symbols. The FEC overhead is 7%, corresponding to a BER threshold of 3.8x10⁻³. The net data rate is about 10 Gb/s, after excluding all the overheads. The modulated optical signal is amplified by an Erbium-doped-fiber-amplifier (EDFA), and then fed into the fiber.

 

Fig. 4 Experimental setup for the 10-Gb/s/ch transmission over 20-km SSMF using silicon MZM and proposed DCR method. (a) Optical spectrum of the generated carrier suppressed DMT signal, (b) optical spectrum of conventional intensity modulated DMT signal. TOF: tunable optical filter.

Download Full Size | PDF

The optical distribution network is composed of 20-km SSMF and a variable optical attenuator (VOA), which is used to emulate the passive power splitter and dynamically adjust the power budget during the measurement. Without any optical amplification and filtering, the received signal is coherently detected in the ONU by a commercial integrated coherent receiver (ICR), with an analog bandwidth of 25 GHz. The optical power of LO is 16 dBm. The four outputs of the ICR are sampled by an oscilloscope of 50 GSa/s, and then processed offline in a Matlab program. The DSP process includes: digital LPF to filter the residual carrier, the DCR process for DMT signal recovering, symbol synchronization, fast Fourier transform (FFT), channel estimation (CE), and decision. To show the advantages of proposed DCR method, we also tested the performance using conventional IM/DD scheme. In the IM/DD scheme, the intensity modulated DMT signal, as show in Fig. 4(b), is first optically amplified and filtered in the ONU. Then, the filtered signal is directly detected by a photo detector (PD) of 40-GHz bandwidth. The output of PD is sampled by an oscilloscope of 50 GSa/s, and then processed offline. The offline processing includes: synchronization, FFT, CE, and decision.

4. Results and discussion

Figure 5 shows the receiver sensitivities for both coherent detection (without EDFA) and direct detection (with EDFA) at back-to-back measurement. In the direct detection scheme, the receiver sensitivity is measured before the optical amplifier in ONU. Through EDFA and a tunable optical filter (TOF), the signal power is kept at 0 dBm before photo detection. The maximum sensitivity is about −28 dBm for the FEC limit of about 8.5-dB Q-factor, corresponding to a BER threshold of 3.8x10⁻³. In the coherent scheme based on DCR method, we can achieve about −44dBm sensitivity without any optical amplifying or filtering, which shows great advantage over the direct detection scheme for both complexity and sensitivity. Thanks to the low CSPR, the sensitivity of DCR method increases by about 16 dB than the unsuppressed optical carrier scheme in direct detection. Thus, the same Q-factor can be achieved even at lower received optical power, due to CSPR improvement. And this is especially important for system requiring large ODN loss and high-order modulation.

 

Fig. 5 Receiver sensitivities for both coherent detection and direct detection schemes at back-to-back measurement.

Download Full Size | PDF

We further measured the receiver sensitivity for the coherent detection scheme using DCR method over 20-km SSMF, as shown in Fig. 6. Almost the same performance is achieved as the back-to-back measurement. The insets of Fig. 6 also show the 16QAM constellations at sensitivities of −36 dBm and −45 dBm, respectively. When 0-dBm launch power is supposed, the proposed method can provide 20-km reach, 1024 splitting ratio, still with 10-dB power margin for the 10-Gb/s DMT-16QAM signal, which shows great potential for both high SE and power budget.

 

Fig. 6 Receiver sensitivities for the coherent detection scheme using DCR method over 20-km SSMF.

Download Full Size | PDF

5. Conclusion

In this paper, we proposed a cost-effective coherent PON scheme using the linear silicon MZM. Intensity modulation of DMT signal and coherent detection is used to achieve both high SE and high sensitivity. We also propose a DCR method to further improve the power efficiency of the modulated DMT signal. Significant increase in power budget is achieved with the proposed DCR method. No carrier frequency offset and phase estimations are needed in the DSP, which greatly reduces the requirements on both laser and DSP. Finally, a 10-Gb/s/ch uplink transmission is experimentally demonstrated over 20-km SSMF, achieving about −44dBm sensitivity with 7% FEC overhead.