1. Introduction

Driven by the ever-growing demand for high-capacity services such as social media, video streaming, and cloud computing, high-speed short-reach optical communication solutions based on direct detection are being extensively studied owing to their relatively low complexity and cost compared to coherent detection systems. However, the expensive discrete devices employed in commercial communication systems are not suitable for short-reach applications. To overcome the high cost of discrete devices, silicon photonics has emerged as a candidate solution owing to its small footprint, high integration density, and low cost.

Silicon photonic modulators have received ever-increasing attention from the research and development community. Among various kinds of silicon modulators, the depletion-type of Mach-Zehnder modulator (MZMs) [1] is the most common for practical use due to higher modulation speed compared to the carrier injection type [2], better tolerance to fabrication errors as well as environmental variations compared to resonant modulators [3], and stability compared to silicon-organic hybrid modulators [4,5]. Various transmission experiments have been performed in recent years employing different direct detection modulation schemes. In 2015, 112 Gb/s four level pulse amplitude modulation (PAM-4) and eight-level PAM (PAM-8) transmission over 20 km based on a silicon intensity MZM was shown in [6] exhibiting bit error rates (BERs) below 3.8 × 10⁻³. Recently, a monolithically integrated silicon photonic Stokes vector receiver was used along with an integrated silicon IQ modulator in order to demonstrate 128 Gb/s transmission over 100 km standard single mode fiber (SSMF) with a BER below 3.8 × 10⁻³ [7].

Chromatic dispersion (CD) introduces power fading distortion due to the square-law photo-detection, which is a major limitation to the transmission distance and the bit-rate of conventional double-sideband (DSB) direct detection systems [8]. To overcome the CD induced spectrally selective fading in DSB systems, single sideband (SSB) modulation can be applied to direct detection optical systems, as has been widely demonstrated with Lithium Niobate modulators [9–11]. It has been shown that a SSB signal can be generated with either an IQ-MZM or a dual-drive intensity MZM by employing a Hilbert transform within the digital signal processing (DSP) at the transmitter [11]. For SSB signal detection using a single-ended photodiode, signal-signal beat interference (SSBI) should be mitigated [10]. Recently, 320 km transmission employing 56 Gb/s single-sideband discrete multi-tone (DMT) format based on a silicon IQ-MZM was also achieved [12]. To further reduce the cost, dual-drive silicon intensity MZMs are highly desired in SSB optical transmission systems. The primary objective of this paper is to validate silicon DD-MZM in metro-scale optical SSB direct detection transmission and make a comparison with a silicon IQ-MZM.

In this paper, the electrically packaged silicon IQ-MZM and silicon DD-MZM are presented and used for a Nyquist-shaped SSB modulation scheme. The mentioned silicon DD-MZM investigated in this paper is in fact a single child MZM of the silicon IQ-MZM. Detailed comparisons are demonstrated between silicon IQ-MZM and silicon DD-MZM in generating 40 Gb/s (10Gbaud) 16-ary quadrature amplitude modulation (16-QAM) Nyquist-shaped SSB signals along with a series of transmission experiments. In order to exploit the potential of the silicon IQ-MZM, 50 Gb/s (12.5Gbaud) 16-QAM Nyquist-shaped SSB transmission over different distances is also successfully demonstrated without CD compensation.

2. Device design and characterization

Figure 1 shows the silicon modulator chip and the mounting Printed Circuit Board (PCB). The silicon in-phase-quadrature (IQ) Modulator consists of two nested Mach-Zehnder Modulators. As highlighted in Fig. 1(b), each MZM has phase modulators of 1.8 mm length in each arm. Low-loss 1x2 and 2x2 MMI structures are used to split and recombine the light both to and within the two MZM. The phase modulators operate via the plasma dispersion effect, where the depletion of free carriers from a reverse biased PN junction which is positioned in the center of the waveguide causes a shift in the phase of the propagating light. Coplanar waveguide electrodes allow the electrical drive signals to co-propagate along the phase modulators at a similar velocity to the light in the waveguide. Grating couplers are positioned at either end of the device to allow the coupling of light to and from the device. Localized heating elements are positioned in each Mach-Zehnder arm and in both I and Q paths to allow fine phase adjustment. The MZM chip is surface mounted on a purposely built radio frequency (RF) PCB board (ROGER4350) as shown in Fig. 1(a). The input differential electrical signals are connected onto the PCB via four K-connectors and microstrip lines route these signals to the input side of MZM. On the output side of MZM, wired bonds are used to connect the differential signals to the PCB, on which the on board 50-ohm termination is built. Several direct current (DC) connections are made to the PCB, which are used for tuning the heating elements on the MZM chip.

 

Fig. 1 Structure of (a) the PCB board and (b) the silicon chip with IQ modulators.

Download Full Size | PDF

Although the purpose-built PCB makes the testing easier, the unavoidable RF loss within the microstrip line on the PCB and impedance discontinuity introduced by bonding wires could significantly reduce the system bandwidth. To assess the bandwidth reduction introduced by the packaging, several dedicated PCBs were fabricated to identify the bandwidth bottleneck within whole system. The electrical S21 results under different packaging conditions (namely wire-bonded and RF probe) were evaluated by using an Agilent 50 GHz network analyzer system (Agilent N5225A).

Firstly, the bandwidth of the electrodes on the silicon MZM chip was evaluated by applying an RF probe onto the pads of the chip. As shown in Fig. 2, by changing the reverse bias voltage from 0 V to 8 V, the 6 dB bandwidth of the MZM chip gradually increased from 7.5 GHz to 22.5 GHz. In contrast, when the MZM chip was wire-bonded onto the PCB, the 6-dB bandwidth dropped to 4 GHz. Changing the reverse bias can still enhance the bandwidth, but only within a narrow range. Furthermore, a 40-mm long 50-ohm microstrip line printed on PCB was measured separately and showed a 6-dB bandwidth up to 10.6 GHz. Therefore, it can be concluded that bonding wires between the PCB and MZM chip are the major reason that the system bandwidth deteriorates. Indeed, flip-chip bonding was considered during the initial experimental stage, but numerous RF input-output (IO) pads and several additional DC control nodes within the IQ-MZM prevented the feasibility of flip-chip bonding.

 

Fig. 2 Electrical S21 parameter measured for different packaging conditions under different reverse bias voltages.

Download Full Size | PDF

The modulator linearity and its power handling capability were assessed by measuring the third order inter-modulation Spurious Free Dynamic Range (SFDR) [13]. The measured value was SFDR = 106 dB/Hz^2/3 at a bias voltage of −8V, which demonstrates that our modulator shows comparable performance to the LiNbO₃ device in linearity. More detailed measurement experiments and results are reported in [14].

3. Principle of optical SSB signal generation by IQ-MZMs and DD-MZMs

In this section, we simply present theoretical studies on the generation of optical SSB signals via IQ-MZM and DD-MZM. An electrical SSB signal can be easily generated by DSP-based Hilbert transform as shown in our previous work [10]. Therefore, an optical SSB signal can be obtained if the real and imaginary components of an electrical SSB signal are used to drive an IQ-MZM, which has been detailed described in [15]. Based on an IQ-MZM, the bias voltage of each child MZM should be finely adjusted to control the carrier-to-signal power ratio (CSPR) of the signal, which plays an important role on system performance [16].

The DD-MZM comprised two parallel phase modulators (PMs) with independent RF and bias ports, as shown in Fig. 3, and is also capable of linearly converting electrical complex signals to the optical domain. A theoretical description of SSB generation using DD-MZM has already been presented in [11], according to which the transfer function of a DD-MZM can be described as Eq. (1) using small signal approximation.

 

Fig. 3 Schematic diagram of DD-MZM. RF: radio frequency; PM: phase modulator.

Download Full Size | PDF

Here $V_I$ and $V_Q$ are independent RF signals for the two PMs, respectively, and the bias difference between the two PMs is set to $V_{\pi }/2$. From Eq. (1), it is observed that the electrical complex signal is linearly mapped to the optical domain with an extra carrier term $1-j$. Compared to the IQ-MZM, however, the CSPR of a DD-MZM cannot be changed through the bias voltages, which have already been used to keep a $\pi /2$ phase difference between two PMs. Therefore, the only way to change the CSPR of a DD-MZM is to adjust the amplitude of the input electrical signal.

4. Experimental setup and DSP

Figure 4(a) and 4(b) illustrate the experimental setup of the Nyquist pulse-shaped SSB transmission systems based on the IQ-MZM and the DD-MZM, respectively. In Fig. 4(a), an external cavity laser (ECL) exhibiting a line-width of ~100 kHz is employed at the transmitter side as the optical source. After a polarization controller (PC), the input light is coupled to the silicon IQ-MZM through a vertical surface grating coupler. The silicon IQ-MZM is fed by an arbitrary waveform generator (Keysight M8195A) operating at 60 GSa/s to generate 10 GBaud 16-QAM Nyquist-shaped SSB signals. Each branch of the silicon IQ-MZM is driven by differential electrical signals in a push-pull mode. Four different bias voltages are applied to the IQ-MZM. The IQ-MZM is biased above the null point by controlling $V_I$and $V_Q$ to generate an optical carrier for the SSB signal. These two biases are finely adjusted to achieve the desired CSPR. A global reverse bias voltage $V_{\mathrm{Re}v}$is applied to all PN junctions to control the most suitable operation point and we set it to −8V for the largest bandwidth consideration with respect to the result in Fig. 2. $V_{\pi /2}$ is applied to the $\pi /2$phase shift heater. After coupling back to fiber, an erbium-doped optical fiber amplifier (EDFA) is used to control the launch power into the fiber link. The transmission link consists of $N$ spans of 80 km SSMF and EDFAs to compensate for fiber attenuation after each span. No inline optical CD compensation is used. At the receiver, a tunable optical filter (Yenista Optics XTM-50) with an edge roll-off factor of 500 dB/nm is added to filter out out-of-band noise. The signal is detected by a single-ended photodiode (PD) and subsequently amplified by an electrical amplifier (EA) both with a bandwidth around 50 GHz. The converted electrical signal is sampled by a real-time digital storage oscilloscope (Keysight DSA-X 96204Q) operating at 80 GSa/s to perform off-line DSP.

 

Fig. 4 Experimental setup for (a) IQ-MZM and (b) DD-MZM. ECL: external cavity laser; AWG: arbitrary waveform generator; EDFA: erbium-doped optical fiber amplifier; SSMF: standard single-mode fiber; OBPF: optical band-pass filter; PD: photodiode; EA: electrical amplifier; DSO: digital storage oscilloscope; PC: polarization controller. (a) $V_{\pi /2}$: bias voltage to control $\pi /2$phase shift heater;$V_I$or $V_Q$: bias voltage to change the phase difference between two arms of a single MZM; $V_{\mathrm{Re}v}$: reverse bias voltage to control the operation point of all PN junctions; $I$and $\overline{I}$: RF signal and its opposite of I branch; $Q$ and $\overline{Q}$: RF signal and its opposite of Q branch. (b) $V_{\pi /2}$: bias voltage to set the phase difference between two arms of a DD-MZM as $\pi /2$; $I$and $Q$: independent RF signals to drive two arms of a DD-MZM.

Download Full Size | PDF

In Fig. 4(b), one branch of the IQ-MZM is used as a DD-MZM to minimize the experimental difference in device fabrication. The two PM arms of the DD-MZM are driven by independent electrical I and Q signals, respectively. The function of $V_{\pi /2}$as well as $V_{\mathrm{Re}v}$is the same as that in the IQ-MZM.

The diagram of the DSP is shown in Fig. 5(a)-5(b). The structure of the transmitted signal frame is shown in Fig. 5(c). At the transmitter, the bit streams are mapped to 16-QAM first. The preamble includes two 64-symbol synchronization sequences and four 128-symbol training sequences. Following the preamble, 25600 data symbols are transmitted. Pre-equalization with a tap number of 97 is performed by convoluting the mapped symbols with the inverse system response in the time domain. As the AWG in our experiment has a sampling rate range of 53.76~65GSa/s, 6 times up-sampling is chosen to generate a 10Gbaud signal. The signal is digitally shaped using a root raised cosine filter with a roll-off factor of 0.01 and a tap number of 1537. The signal is up-converted with a subcarrier frequency of 0.51 of the symbol rate to realize a nearly half-cycle subcarrier modulation (only a 1% guard band between the optical carrier and the SSB signal). Finally, after Hilbert transform, the SSB signal is generated.

 

Fig. 5 Generation and detection of the SSB signal. The insets are the corresponding spectra of the signals at each stage of the DSP. The blue part is the signal, and the red part is the interference. The symbol rate is F_S and the subcarrier frequency is F_C. (a) Transmitter side DSP. RRC: root raise cosine. (b) Receiver side DSP. SSBI: signal-signal beat interference. (c) Frame structure of baseband signal.

Download Full Size | PDF

At the receiver, as the SSB signal will be converted to double-sideband (DSB) signal after photodiode detection, 4-time resampling is required to avoid aliasing. After SSBI compensation with the iterative method in [10], the signal is down-converted by the subcarrier of $0.51F_s$ and matched-filtered with a tap number of 1025. After synchronization, two-point resampling is performed. Then the signal is equalized with a $T_s/2$ spaced training sequence based time domain equalization (where $T_s$ is the symbol time period). A 97-tap finite impulse response (FIR) filter whose taps are updated from the training sequence is used to equalize the data. Finally, after down sampling, the BER is calculated by error counting with ~5 × 10⁵ bits.

5. Experimental results and discussion

Note that the CSPR is defined as the power ratio between the carrier and the sideband signal. A system with large CSPR suffers degradation in the optical signal-to-noise ratio (OSNR) while small CSPR suffers relatively severe SSBI. In general, the CSPR and the total launch power should be optimized for the specific transmission distance. However, as we have mentioned above, the CSPR of a DD-MZM is only determined by the amplitude of input RF signals, which were optimized to provide a suitable driving voltage. Therefore, there is little space to control the CSPR of a DD-MZM. In our experiment, the CSPR generated by the DD-MZM was about 16.4 dB. For a fair comparison, the CSPR of the IQ-MZM was also set as 16.4 dB by finely adjusting its bias voltages. The RF driving voltage for the IQ-MZM was also optimized. As shown in Fig. 6, we tested the BER performance versus the optical launch power in both IQ-MZM and DD-MZM systems. To achieve the best average BER performance, the launch powers of 10.0 dBm and 6.0 dBm were chosen for DD-MZM and IQ-MZM, respectively.

 

Fig. 6 Measured BERs as a function of the launch power for DD-MZM and IQ-MZM after 160 km SSMF transmission with the CSPR fixed at 16.4 dB. The baud rate of the signal is 10 Gbaud.

Download Full Size | PDF

Figure 7(a) and 7(b) shows the received optical spectra of the SSB signal at different transmission distances for DD-MZM and IQ-MZM, respectively. The back-to-back (BTB), 80 km and 160 km curves were measured before the optical band-pass filter (OBPF). Note that the EDFA was also included in the BTB scenario for a fair comparison with different transmission distance. The filtered curves were measured after the OBPF, which demonstrates that most of the residual sideband has been removed. The lower amplitude of filtered spectrum is due to the insertion loss of the OBPF.

 

Fig. 7 Received optical spectra at different transmission distances with 0.01 nm resolution for (a) DD-MZM and (b) IQ-MZM. (c) Comparison of transmitted optical spectrums between the DD-MZM and the IQ-MZM with 0.01 nm resolution. The baud rate of the signal is 10 Gbaud.

Download Full Size | PDF

Figure 7(c) compares the output optical spectra from the two kinds of modulators. It is obvious that the optical sideband suppression ratio (OSSR) for IQ-MZM is about 10 dB larger than that for DD-MZM. We first attributed this effect to the difference in extinction ratios induced by non-ideal optical splitting ratios of MZMs as described in [16]. However, this assumption seems unreasonable after we found that both extinction ratios of the two kinds of modulators were around 30 dB, demonstrating well matched optical splitting ratios. Excluding the fault of optical chip, it is most likely that the unequal RF loss within the four microstrip lines on the PCB and the different impedance discontinuity introduced by bonding wires contribute together to the large OSSR difference. The lower OSSR of the DD-MZM will degrade the BER performance of the system.

Figure 8 gives the received electrical spectra after down-converting in the BTB scenario with and without pre-equalization for the silicon IQ-MZM. The 6-dB bandwidth of the whole system can be calculated from the spectrum without pre-equalization, which is around 4 GHz, similar to the result measured by the network analyzer.

 

Fig. 8 Received electrical spectra after down-converting in BTB scenario with and without pre-equalization for silicon IQ-MZM.

Download Full Size | PDF

The BER performance versus the OSNR of the DD-MZM and the IQ-MZM systems at different transmission distances is shown in Fig. 9(a), and the corresponding constellation maps are also shown in Fig. 9(b)-9(e). In our experiment, we choose a BER threshold of 4.5 × 10⁻³ for 7% HD-FEC, which can be constructed from continuously interleaved Bose-Chaudhuri-Hocquenghem (BCH)(1020,988) codewords [17]. For the back-to-back (BTB) scenarios, compared with the theoretical curve, the implementation OSNR penalties at the BER of 4.5 × 10⁻³ for the DD-MZM and the IQ-MZM systems are about 6.4 dB and 1.4 dB, respectively. The performance degradation of the DD-MZM system is mainly due to the low OSSR value in the SSB signal generation. After 160 km SSMF transmission, the BERs of the DD-MZM and the IQ-MZM systems are 5.4 × 10⁻⁴ and 9.0 × 10⁻⁵, respectively. The OSNR penalties are about 1.4 dB and 1.5 dB, respectively, compared with the BTB results.

 

Fig. 9 (a) BER performance versus OSNR at different transmission distances with different modulators. The baud rate of the signal is 10 Gbaud. (b) (c) Constellation maps for IQ-MZM and DD-MZM in BTB scenario, respectively. (d) (e) Constellation maps for IQ-MZM and DD-MZM after 160 km transmission, respectively.

Download Full Size | PDF

In the above experiment, the CSPRs of the IQ-MZM and the DD-MZM systems were chosen to have the same value for the sake of a fair comparison. To fully explore the capability of the silicon IQ-MZM in SSB direct detection transmission, we optimized the CSPR of the IQ-MZM system at a higher baud rate of 12.5 Gbaud (50 Gbit/s) and a longer transmission distance of 320 km. As shown in Fig. 10, the optical launch power is optimized for different CSPR values. To achieve the best BER performance, a CSPR of 14.4 dB and an optical launch power of 5.0 dBm were chosen.

 

Fig. 10 Measured BERs as a function of the launch power after 320 km SSMF transmission with different CSPRs.

Download Full Size | PDF

Figure 11(a) shows the OSNR degradation as a function of transmission distance. Figure 11(b) shows the BER dependence on transmission distance with the CSPR of 14.4 dB and the optimum launch power of 5.0 dBm. The BERs below the 7% HD-FEC threshold of 4.5 × 10⁻³ were measured for all the SSMF transmission distances studied, up to 320km.

 

Fig. 11 (a) OSNR as a function of transmission distance measured at 0.1 nm resolution. (b) Measured BER as a function of transmission distance with the CSPR of 14.4 dB and launch power of 5.0 dBm.

Download Full Size | PDF

The BER performance versus OSNR at different transmission distances of the 12.5 Gbaud IQ-MZM system is shown in Fig. 12. The OSNR penalties at the BER of 4.5 × 10⁻³ compared to the BTB scenario vary between 0.25 dB and 2.0 dB, for transmission distances in the range 80 km to 320 km, which demonstrates the robustness of the system in a CD channel.

 

Fig. 12 BER performance versus OSNR at different transmission distances with the silicon IQ-MZM. The baud rate of the signal is 12.5 Gbaud.

Download Full Size | PDF

6. Conclusion

The comparative performance of silicon DD-MZM and silicon IQ-MZM was investigated in direct-detection SSB systems. Both kinds of silicon modulators were employed to generate 10 Gbaud 16-QAM Nyquist-shaped SSB signals with the same CSPR. The measured OSSR of the DD-MZM was 10 dB lower than that of the IQ-MZM. Transmission experiments over 80 km and 160 km SSMF have demonstrated that at a BER of 4.5 × 10⁻³, the required OSNR for the DD-MZM is about 5 dB higher than for the IQ-MZM. Our work is the first experimental comparison between silicon IQ-MZMs and DD-MZMs in generating SSB signals. After optimizing the CSPR, this silicon IQ-MZM can support 50 Gb/s 16-QAM Nyquist-shaped SSB transmission up to 320 km with a BER of 2.7 × 10⁻³.

We can conclude that both silicon IQ-MZMs and DD-MZMs have great potential for low cost metro-scale optical transmission and data center interconnection. Although not as good as silicon IQ-MZM in generating SSB signals, silicon DD-MZM is still attractive due to its lower cost and higher energy efficiency. Note that the silicon modulator in our experiment only has a limited bandwidth of 4 GHz after wire-bonding based packaging. To enhance the system bandwidth, more advanced packaging techniques (such as flip-chip bonding) can be employed. Due to their smaller footprint and fewer IO pads, the silicon DD-MZMs will be favored in advanced compact packaging approaches.