A novel method to simultaneously detect power imbalance, modulation strength, and bias drift of coherent IQ transmitter during the initial power-up is presented. This is achieved by sweeping gain scaling factor of finite impulse filter in a digital domain and monitoring the combined output power. Furthermore, by dithering gain scaling factor of finite impulse filter, the power imbalance is measured with live traffic. Those impairments can be compensated accordingly. For example, the power imbalance is compensated through adjustment of gain setting of a radio frequency amplifier. This novel method works for multiple channels over C band, and the built-in photodiode of coherent transmitter provides sufficient accuracy.
© 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
To meet the ever-growing requirement to carry data traffic, the high-order quadrature amplitude modulation (QAM) running at high baud-rate has been widely used in the long-haul transmission system. The data is modulated in both the in-phase (I) and the quadrature (Q) domain, as well as in the two orthogonal polarizations (X and Y) [1, 2]. Furthermore, Nyquist pulse shaping is ubiquitously used to improve the spectral efficiency [3, 4]. Since the initial inception of probability shaping constellation (PSC) , the advantage and flexibility of PSC 64-QAM has made this technique as the essential de-factor modulation format for the next generation coherent DWDM system . All those advanced systems utilize the coherent IQ transmitter, which is composed of four tributary channels XI, XQ, YI, and YQ. The power imbalance among the tributaries need to be minimized to achieve the best performance.
Some solutions have been demonstrated to compensate the IQ or XY power imbalance. In , a dithering signal is applied to the bias of MZM. The strength of the second order harmonics of the dithering signal can be used to determine the power imbalance. Simultaneously, the drift of the bias point for MZM can be compensated. However, this method requires the external circuit to apply and detect the dithering. Because the dithering signal is small, the analysis needs to be performed in the frequency domain after fast Fourier transform (FFT). The complexity of the system is relatively high.
In , the Gram-Schmidt orthogonalization procedure (GSOP) is used to compensate the quadrature imbalance introduced by the coherent receiver and transmitter. In , a sixteen-real value adaptive equalizer (AEQ) replaces the traditional 2x2 complex AEQ, which allows the compensation of the power imbalance in the coherent receiver and transmitter. In , the power imbalance is compensated by a 2x2 complex adaptive equalizer (AEQ) which is implemented after the carrier phase estimation (CPE). The skew from the transmitter can be compensated simultaneously. Here the challenges for those methods [8–10] are two-fold: First, the power imbalance limits the transmission distance of optical signal; the receiver may not be able to fully compensate the power imbalance from the coherent transmitter. Second, the multi-in multi-out (MIMO) equalization is done in the receiver. The impairment introduced during the transmission through fiber (for example, nonlinear distortion) and the impairment introduced by phase noises (for example, equalizer enhanced phase noise (EEPN)) will limit the compensation capabilities for the transmitter’s power imbalance.
In , during the initial power-up, the identical data pattern can be loaded into two tributaries. By adjusting phase difference between two tributaries, a destructive interference can be achieved. The power level of destructive interference can be used to determine the power imbalance between tributaries. Meanwhile, the time skew can be compensated. In , through an innovative cooperative coevolution genetic algorithm with modified clock tone amplitude being the fitness function, the time skew among tributaries, the bias voltage for MZM, the phase imbalance between I tributary and Q tributary, and the amplitude imbalance among tributaries can be all compensated. However, the method in [11, 12] are not applicable to the live traffic.
In addition to the imbalance caused by tributary power difference, the imbalance can be caused by the other factors. Ideally, the peak-to-peak phase swing (modulation strength) applied to the Mach-Zehnder modulator (MZM) should be the same across four tributaries; the bias points of the MZM should be close to its null point (90 degree) for all tributaries. The deviation from this ideal case also cause the imbalance among the tributaries.
To overcome the certain limitations of previous methods, we propose a novel method to detect and compensate power imbalance, modulation strength, and bias drift during the initial power-up. Furthermore, power imbalance can be monitored and compensated with the live traffic. Our method utilizes the existing hardware required to implement the coherent IQ transmitter. In addition, our method relies on the adjustment of the finite impulse filter (FIR) in the digital domain. As a result, our method is highly accurate and robust against noise.
The paper is organized as following: in section 2, we establish the principle to detect power imbalance, modulation strength and bias point for MZM; in section 3, we discuss the experimental results for four different scenarios; in section 4, we draw the conclusion.
Figure 1 shows the block diagram of coherent IQ transmitter, which is composed by the analog coherent optics (ACO)  and the digital signal processing (DSP) application specific integrated circuit (ASIC). The client’s data first passes through the layer of forward error correction (FEC) where FEC overhead is added for error correction. Then the data passes through the FIR filter in the tap-and-delay structure. The FIR filter is Ts/2 spaced, where Ts is the symbol period. The output from FIR filter is converted from digital domain to analog domain through a high-speed digital-to-analog (DAC) converter. The analog electrical signal first goes through the traces on the radio-frequency (RF) print circuit board (PCB) and pluggable interface (if applicable). The electrical signal is then boosted by the linear RF amplifiers and finally applied to the MZM of a particular tributary.
In the coherent IQ transmitter, the output from a continuous wave (CW) laser diode (LD) is split into four tributaries. The tunable optical coupler (TOC), like the thermo-optical switch, can be integrated into the coherent IQ transmitter to balance the power among the tributaries. One can also integrate the variable optical attenuator (VOA) or semiconductor optical amplifier (SOA) to regulate the power among the tributaries . After the MZM, the optical powers from the four tributaries are combined. The phase shifter (PS) is used to introduce 90-degree difference between I tributary and Q tributary. The polarization rotator (Pol-Rot) is used to generate the orthogonal polarization and the polarization beam combiner (PBC) is used to combine the outputs from two polarizations.
An important block within the DSP ASIC on the transmitter side is a finite impulse response (FIR) filter. There are multiple functionalities of the FIR filter: to overcome the bandwidth limitation imposed by the components on the data path; to perform the Nyquist pulse shaping which improves the spectral efficiency; to compensate the impairment like the time skew and the power imbalance. The output from the FIR filter is a convolution between the input signal and the FIR’s impulse response, which can be expressed as
One feature of the tap-delay FIR filter is as following: if we multiply a scaling factor Scale with all the tap coefficients, the spectral response does not change. But the output from FIR is changed by Scale. To demonstrate this, we choose a typical FIR filter. The tap coefficients of the FIR are set to provide certain amount of gain at Nyquist frequency to compensate the RF insertion loss introduced by the PCB traces and pluggable connector (if applicable) . This compensates the inter-symbol interference (ISI) due to the limited bandwidth. We also apply multiple scaling factors to this typical FIR filter to create a set of new FIR filters. In our implementation, the tap coefficients of FIR filter h(j) is represented by an 8-bit register with the first bit being the sign bit, thus h(j) is within the range of [-128, 127]. When we apply the scaling factor, we round h(j) to the nearest integer as shown in the equation below.
We plot the frequency response of all those FIR filters in Fig. 2. As shown, the spectral shapes of all those FIR filters are very similar. Some small difference is caused by the rounding error. This demonstrates that we can apply the scaling factor to the FIR filter to simultaneously adjust the output of the FIR filter and maintain the spectral shape roughly the same.
Next, we discuss how the gain scaling factor influences the combined output power of the coherent IQ transmitter. From Eq. (1), one can see that the maximum output value from the FIR filter is reached when the input signal has the same sign as the tap coefficients of the FIR filter as shown in the equation belowEq. (1) are in phase and added constructively, leading to the maximum output power. We define this particular data stream as χ(j).
The output of FIR is converted to analog signal through high speed DAC. Then, the maximum output of the FIR filter (ymax) from the particular data patter χ(j) corresponds to the peak-to-peak voltage swing from the DAC. The output from DAC further passes through pluggable connector (if applicable), RF trace, RF amplifier to MZM. Then the peak-to-peak voltage swing (Vswing) applied to MZM to create phase shift can be expressed as following:
The total output from a polarization-multiplexed IQ transmitter is shown below
Next, we define the modulation strength factor α and the bias drift factor β. We combine the Eqs. (2), (4) and (5), and express the combined output power from the MZM as following:
Initially, we set the gain scaling factor for all tributaries as 1 and measure the combined output power in normal operation Poutini. Then we sweep the scaling factor of the ith tributary, measure the output power Pouti(Scalei). Meanwhile, we keep the scaling factors of the other tributaries fixed at 1. Since the power from the other three tributaries will remain unchanged during the process, the difference between Pouti(Scalei) and Poutini will contain only the terms from the ith tributary whose scaling factor is swung. This difference is shown below:Eq. (7) for ith tributary, and M is the total measurement points. A typical procedure like sequential quadrature programming (SQP) can be used to perform the curve fitting .
Once those underlying fitting parameters for four tributaries are decided, we can calculate the power from each tributary. Then the imbalance (IMB, unit in dB) between the different tributaries can be determined by the following equations:
One important advantage of the proposed method is that gain scaling factor can be adjusted completely independent of the shape of the FIR filter which is determined by the relative ratio between the tap coefficients. This allows the power imbalance, the modulation depth, and the bias drift be determined through the swing of the gain scaling factor. Meanwhile, the adjustment of the shape of the FIR filter can compensate the bandwidth characteristics of particular tributary and perform the Nyquist pulse shaping accordingly.
Another impairment for the coherent IQ transmitter is the modulation nonlinearity from various components. By sweeping the frequency and the amplitude of the sinusoidal stimulus during the initial power-up, the bandwidth limitation and the modulation nonlinearity are detected for coherent IQ transmitter. The bandwidth limitation can be compensated by the shape of the FIR filter. The modulation nonlinearity can be compensated by a memoryless Volterra filter, independent from the FIR filter .
In this section, we first show the experimental results during the initial calibration. Then, we present the method to detect power imbalance with live traffic. Next, we show that this novel method works over 100 channels within C band. We also demonstrate that the accuracy of built-in photo diode of the coherent transmitter is sufficient. Finally, we demonstrate that the power imbalance can be compensated through adjustment of gain setting of radio-frequency (RF) amplifier.
3.1 Power-up calibration with swinging in gain scaling factor
Figure 3(a) shows our experimental setup utilizing a Juniper PTX 3000 chassis. In the PTX 3000 chassis, there are all necessary line-cards like routing engine (RE), routing control board (RCB), switch interface board (SIB), physical interface card (PIC) and flexible PIC concentrator (FPC). The line-side 5х100G DWDM (dense wavelength division multiplexing) coherent PIC (Juniper P2-PTX-5-100G-DWDM) hosts the DSP ASIC and the CFP2 form-factor analog coherent optics (CFP2-ACO) pluggable module. The output of the coherent transmitter is split into two paths by a 3dB coupler. One path is connected to an external power meter (PM), and the other path is looped back to the coherent receiver. An optical network tester (ONT, from Viavi) is used to generate 100 gigabit Ethernet (100GE) traffic. A pair of QSFP28 (quad small form-factor pluggable) modules are used to transport the 100GE traffic from the ONT to a client-side PIC hosting 10 QSFP28 modules (Juniper P3-10-U-QSFP28). The 100GE traffic is forwarded from the client-side PIC to the line-side PIC through a packet forwarding engine (PFE). One concern is whether there is any Ethernet packet loss during the adjustment of the gain scaling factor. So, we continuously monitor the ONT during the experiment.
The modulation format for the optical signal is polarization multiplexed QPSK (PM-QPSK). Nyquist pulse shaping is not used for this experiment and the baud rate is approximately 30GBd. The 3dB bandwidth of the whole Tx chain is approximately 16GHz . The effective number of bits (ENoB) is similar to the DSP chip in . The optical signal to noise ratio (OSNR) is approximately 42dB at the transmitter’s output. The required OSNR (ROSNR) at the receiver side is approximately 13dB. The performance of BER vs. OSNR for the coherent IQ transponder can be found in .
Figure 3(b) shows the block diagram of coherent DSP ASIC and optical front-end. We have discussed the functionality of each component in the transmitter side in section 2, so we only describe the functionality of each component in the receiver side. The optical front-end of coherent IQ receiver is composed of 90-degree optical hybrid, balanced photo diodes, linear trans impedance amplifier, and integrated tunable laser assembly (ITLA) serving as local oscillator (LO). The DSP ASIC on the receiver side is composed of the following building block : static equalization to compensate the static impairment and chromatic dispersion (CD); adaptive equalization to perform polarization de-multiplexing and compensation of polarization mode dispersion (PMD); time recovery using Gardner’s method; frequency offset estimation; carrier phase estimation (CPE); symbol estimation and recovery; and forward error correction to remove the error.
We choose the initial FIR filter so that ymax is 212. This is based on the optimal bit error rate (BER) from the coherent transponder, as discussed in section 3.2. Since we use 8-bit DAC, the maximum value is 256. We intentionally only use 212 / 256 = 83% of the available range of the DAC to avoid the output being saturated and clipped. With ymax of the initial FIR being 212, the gain scaling factor can be changed from 0 to 1.2. This is the range we swing the gain scaling factor during the initial power-up calibration.
To adjust the Vswing by changing the gain scaling factor (Scale), the RF linear amplifier within the pluggable module should work in the manual gain control (MGC) mode. Figure 4 shows the output of the peak detector (PKD) which is proportional to Vswing at the output of the RF amplifier. As seen, Vswing changes linearly with Scale. Since we adjust the scaling factor in the digital domain, the accuracy of the adjustment of Vswing is very high, only limited by the rounding error in the digital domain. It is also robust against the analog noise. On the other hand, it is also possible to adjust Vswing by adjusting the gain setting of the RF amplifier. However, it is usually within a limit range, the linearity between the gain setting and the output swing is guaranteed by the RF amplifier’s specification. Thus, our method offers an important advantage.
Figure 5 shows the measurement results and the curve fitting results for three typical tributaries. The curve fitting parameters are shown in the inset of each sub-figure. As seen, there is a very good agreement between the measurement and the curve fitting results. Table 1 further summarizes the fitting parameter for three channels within C band. The frequency of each channel is 191.1THz + chan*0.05THz, where chan is the channel number from 1 to 100. Based on the fitting parameters [p, α, β], we calculate the power imbalance between the tributaries and show them in Table 1.
Since the swinging of the gain scaling factor is implemented in the digital domain with high accuracy, we can generate multiple values of Scale with a fine step and measure the corresponding ΔP. The influence of noise on the measurement accuracy can be greatly reduced by measuring multiple data points and applying the curve fitting as shown in Eq. (7). The improvement in accuracy is proportional to the square root of M, which is the total measurement points. In this way, we can improve the accuracy of measured power imbalance, modulation depth and bias point. For example, in the high baud-rate system, the ENoB is smaller, leading to the larger quantization noise, which will limit the setting accuracy of the gain scaling factor. The proposed method can effectively ‘average’ the quantization noise.
The estimation error Errrms is also shown in Table 1. As seen, the estimation error is very small, indicating a good agreement between the theoretical analysis and the experimental results. It is also noticeable that when the bias of IQ modulator drifts, Errrms increases. Thus, the accuracy of the estimation is dependent on the bias point. However, the amount of the change in Errrms is small even when the bias is drifted from β = 1 (0.5π) to a large value of β = 1.25 (0.625π). This demonstrates that our method is robust against the drift of bias point.
3.2 In-flight measurement with dithering in gain scaling factor
The method above can be used to detect and compensate the power imbalance, modulation strength and bias drift in the calibration phase and the module reconfiguration phase like switching to a new channel. However, a full-scale swing of DSP scaling factor is not possible with live traffic. With DSP scaling factor being 0, the particular tributary will be turned off, leading to loss of client’s data.
Figure 6 shows the bit error rate (BER) / Q2 factor of the coherent IQ transmitter vs. the gain scaling factor. As seen, if we limit the gain scaling factor within the range of 0.9 to 1.1, the impact on BER / Q2 is small. However, if we only use the measurement data within this small range to perform curve fitting, we could get large error. So, a different methodology is needed.
By performing Taylor expansion of Eq. (7) near Scale = 1, we can see ΔPi is roughly linear with Scalei. We can further define a slope coefficient, as shown in the equation below:Table 1, the largest contribution of the power imbalance is from the difference in the steady state power (p). The difference in α and β is relatively small except the XI tributary of the channel 87. Thus, the imbalance between the tributaries can be roughly estimated as following
Inevitably, there is noise associated with each measurement. If we measure two points with Scale = 0.9 and Scale = 1.1 and calculate the Slope with two points, the result is susceptible to the measurement noise. Since the dithering of the gain scaling factor is implemented in the digital domain with high accuracy, we can generate multiple values of Scale with a fine step and measure the corresponding ΔP. The improvement in accuracy is proportional to the square root of M, which is the total measurement points. In this way, we can improve the accuracy of measured power imbalance.
We validate through the ONT that on the client-side, there is no packet loss when the gain scaling factor is changed from 0.9 to 1.1. We present the measurement result with the live traffic in Table 1, in comparison with the measurement result with the full swing of the gain scaling factor. As seen, the difference between two methods is small most of time. One noticeable difference (0.43dB) between two methods is for IMBIQ,X of channel 87. The root cause is that the bias point of XI tributary (as shown in Fig. 5(c), β = 1.24) is significant away from the optimal null point. This leads to a large error in the estimation of power imbalance using Slope. The large bias drift should be detected and corrected during the initial power-up to reduce this error. Even with this large drift of bias point presented, the estimation error is still <0.5dB, demonstrating the feasibility of measurement and compensation of the power imbalance with the live traffic.
One concern is that the full range of DAC output is not used. In theory, the larger the peak-to-peak swing of the modulated signal is, the higher the output power and the OSNR are. However, due to the nonlinearity of the DAC, the RF amplifier and the IQ modulator, the signal modulated with the full peak-to-peak swing (256 for 8-bit DAC, corresponding to Scale = 1.2) suffers from the degradation of BER / Q2 factor, as shown in Fig. 6. The optimal peak-to-peak swing is ~212 for 8-bit DAC, which corresponds to 83% of the full swing and Scale = 1. In addition, the clipping effect of the DAC can cause the degradation in the performance when the full range of DAC output is used. As seen, the penalty of Q2 factor is approximately 0.5dB at the full range of DAC output.
The impact of the gain scaling factor is more noticeable at the high OSNR region. In this work, we focus on the influence of power imbalance, so we use the back-to-back configuration without OSNR loading. When the OSNR is low, the influence from the noise is more dominant.
During the normal operation of the coherent IQ transmitter, it is essential to keep the bias point from drifting due to the environmental change. The automatic bias control (ABC) is implemented to maintain the optimal bias point. Many approaches have been demonstrated by first dithering the bias voltage at a particular frequency, then detecting the various parameters at that particular frequency to compensate the bias drifting. In , one detects the change in optical power using the first order lock-in amplifier. In , one monitors the power of the first and the second harmonics of the dithering signal. In , one measures the RF power at the dithering frequency and the RF power of the transmitted signal simultaneously. In , one calculates the dithering correlation function. In general, those techniques can maintain the bias voltage near the null point when the ABC is initially engaged with a ‘good’ starting point close to the optimal value. Our proposed technique can be used to determine the initial ‘good’ starting point for ABC, and the ABC holds the bias voltage to the null point during the normal operation.
3.3 Experimental results over C band using the internal photo diode
We further demonstrate our novel method works over different channel across C band. Here we use an improved CFP2-ACO modules with a better bias control circuit. We also apply our curve-fitting method using the measured result using the internal built-in photo diode of CFP2-ACO module. The results are shown in Fig. 7. As seen, there is a noticeable variation between different channels. For QPSK modulation format, this variation of the imbalance does not impose a severe penalty. However, for high order modulation format like 16-QAM or 64-QAM, the penalty due to the imbalance is much higher. So, it is necessary to measure the imbalance during the initial power-up and compensate it accordingly.
It is also noticeable that the result from the internal photo-diode closely tracks with the result from the external instrument-grade power meter. This demonstrates that the internal PD offers sufficient accuracy for this measurement. Thus, this calibration process can be implemented without any external hardware. In general, the accuracy of the internal PD is not high. It suffers from the thermal noise and drifts over the temperature. In this case, the accuracy is +/− 0.5dB from + 3dBm to −10dBm. However, in our method, we measure the difference between the output power at certain gain scaling factor and the initial output power (ΔPi), as shown in Eq. (7). Thus, the noise and the drift of internal PD cancels during this ‘differential’ measurement. A more relevant specification for the photo diode in this application is the linearity. It is +/− 0.2dB from + 3dBm to −10dBm. The relaxed specification on the photo diode is another advantage for our method.
3.4 Compensation of power imbalance through gain setting of RF amplifier
During the initial calibration, we can measure three underlying root causes contributing to the imbalance in coherent IQ transmitter. They are optical power (p), modulation strength (α) and bias point (β). Table 2 summarizes the compensation methods for the three underlying root causes. As shown in Fig. 1, it is possible to integrate some optical devices, like TOC and VOA / SOA, within the coherent IQ transmitter. Then, one can use those devices to compensate imbalance in optical power. The difference in modulation strength can be compensated by adjusting the gain setting of RF amplifier or the gain scale of the FIR filter. The drift of bias point can be adjusted by the bias voltage.
However, not all coherent IQ transmitter integrate TOC / VOA / SOA. Particularly, the coherent IQ transmitter we used in our experiment does not integrate those devices. Our experimental results show that the difference in optical power (p) contributes most to the imbalance. To compensate the imbalance in this case, we adjust the gain setting of RF amplifier to adjust the modulation strength (α), so that the output power after the modulation Ptrib are the same across four tributaries.
Figure 8 shows the experimental results of compensating imbalance through this method. We choose two channels with large IQ imbalance for demonstration based on the measured result in Fig. 7 (indicated by the green arrow). The 74th channel has a large IQ imbalance in X polarization, where the steady-state optical power (p) in the XI tributary is higher than that in the XQ tributary. Then, when we increase the gain setting of the RF amplifier in the MGC mode for the XQ tributary, we increase the peak-to-peak swing (Vswing) applied to the MZM, leading to a linear increment of the voltage of the PKD for the RF amplifier and the modulation strength (α) in the XQ tributary. In turn, the output power after the MZM in the XQ tributary increases, leading to that the imbalance between XI and XQ tributaries approaches zero (the target setting).
The 60th channel has a large IQ imbalance in Y polarization, where the steady-state optical power (p) in the YI tributary is higher than that in the YQ tributary. Then, when we decrease the gain setting of the RF amplifier in the MGC mode for the YI tributary, we decrease the peak-to-peak swing (Vswing) applied to the MZM, leading to a linear decrement of the voltage of the PKD for the RF amplifier and the modulation strength (α) in the YI tributary. In turn, the output power after the MZM in the YI tributary decreases, leading to that the imbalance between YI and YQ tributaries approaches zero (the target setting).
In this paper, we demonstrate a novel method to detect and compensate the power imbalance and the modulation imperfection in coherent IQ transmitter. This method can be applied during the initial power-up by sweeping the gain scaling factor of the FIR filter and monitor the combined output from the coherent IQ transmitter. Then a curve fitting procedure to minimize the least mean square error can be used to determine three underlying parameters: the optical power, the modulation strength, and the bias point. Based on those parameters, the imbalance among the tributaries and the imperfection in modulation can be determined and compensated accordingly.
This method can be applied with the live traffic by dithering the gain scaling factor of FIR filter. The slope of the change in the combined output power can be used to estimate the power imbalance. Once estimated, the power imbalance can be compensated accordingly. For example, the gain setting of the RF amplifier can be adjusted to compensate the power imbalance.
The method is applicable over different channels as well. The variation of the imbalance over different channel is not negligible. So, it is necessary to compensate the imbalance for different channels. The accuracy of the internal photo diode built in the coherent IQ transmitter is sufficient. The noise and drift of internal PD is cancelled by the ‘differential’ process in the measurement.
The novel method relies on the adjustment of the gain scaling factor of the FIR filter within the DSP, which is external to the analog optical front-end. Thus, this method can be universally applied to the various form-factor of the coherent IQ transmitter. Also, the implementation of this method is within the digital domain, thus it is highly accurate and robust against noise. Our method is also independent from the modulation format, so it can be applied to the next-generation coherent IQ transmitter running at high baud-rate and advanced modulation format.
The gain scaling factor can be adjusted completely independent of the shape of the FIR filter. This allows the power imbalance, the modulation depth, and the bias drift be determined through the swing of the gain scaling factor. Meanwhile, the adjustment of the shape of the FIR filter can compensate the bandwidth characteristics of one particular tributary. In addition, the modulation nonlinearity can be compensated by a memoryless Volterra filter independent from the FIR filter.
The authors gratefully acknowledge vigorous encouragement and sturdy support on innovation from Dr. Domenico Di Mola at Juniper Networks.
References and links
1. K. Roberts, Q. Zhuge, I. Monga, S. Gareau, and C. Laperle, “Beyond 100 Gb/s: capacity, flexibility, and network optimization,” J. Opt. Commun. Netw. 9(4), C12–C24 (2017). [CrossRef]
2. J. Wei, Q. Cheng, R. Penty, I. White, and D. Cunningham, “400 Gigabit Ethernet using advanced modulation formats: performance, complexity, and power dissipation,” IEEE Commun. Mag. 53(2), 182–189 (2015). [CrossRef]
3. B. Châtelain, C. Laperle, K. Roberts, M. Chagnon, X. Xu, A. Borowiec, F. Gagnon, and D. V. Plant, “A family of Nyquist pulses for coherent optical communications,” Opt. Express 20(8), 8397–8416 (2012). [CrossRef] [PubMed]
4. J. Wang and Z. Pan, “Generate Nyquist-WDM Signal Using a DAC With Zero-Order Holding at the Symbol Rate,” J. Lightwave Technol. 32(24), 4085–4091 (2014).
5. F. Buchali, F. Steiner, G. Bocherer, L. Schmalen, P. Schulte, and W. Idler, “Rate Adaptation and Reach Increase by Probabilistically Shaped 64-QAM: An Experimental Demonstration,” J. Lightwave Technol. 34(7), 1599–1609 (2016). [CrossRef]
6. J. Cho, X. Chen, S. Chandrasekhar, G. Raybon, R. Dar, L. Schmalen, E. Burrows, A. Adamiecki, S. Corteselli, Y. Pan, D. Correa, B. McKay, S. Zsigmond, P. Winzer, and S. Grubb, “Trans-atlantic field trial using high spectral efficiency probabilistically shaped 64-QAM and single-carrier real-time 250-Gb/s 16-QAM,” J. Lightwave Technol. 36(1), 103–113 (2018). [CrossRef]
7. M. Sotoodeh, Y. Beaulieu, J. Harley, and D. McGhan, “Modulator Bias and Optical Power Control of Optical Complex E-Field Modulators,” J. Lightwave Technol. 29(15), 2235–2248 (2011). [CrossRef]
8. I. Fatadin, S. J. Savory, and D. Ives, “Compensation of quadrature imbalance in an optical QPSK coherent receiver,” IEEE Photonics Technol. Lett. 20(20), 1733–1735 (2008). [CrossRef]
9. M. Faruk and K. Kikuchi, “Compensation for In-Phase/Quadrature Imbalance in Coherent-Receiver Front End for Optical Quadrature Amplitude Modulation,” IEEE Photonics J. 5(2), 7800110 (2013). [CrossRef]
10. C. Fludger and T. Kupfer, “Transmitter Impairment Mitigation and Monitoring for High Baud-Rate, High Order Modulation Systems,” in European Conference on Optical Communication (2016), paper Tu.2.A.2.
11. Y. Yue, Q. Wang, B. Zhang, A. Vovan, and J. Anderson, “Detection and compensation of power imbalances for DP-QAM transmitter using reconfigurable interference,” Proc. SPIE 10130, 101300M (2017).
12. J. Diniz, F. Da Ros, E. Da Silva, R. Jones, and D. Zibar, “Optimization of DP-MQAM Transmitter Using Cooperative Coevolutionary Genetic Algorithm,” J. Lightwave Technol. 36(12), 2450–2462 (2018). [CrossRef]
13. H. Chaouch and M. Filer, “Analog Coherent Optics for Long Haul Datacenter Regional Networks,” J. Lightwave Technol. 36(2), 372–376 (2018). [CrossRef]
14. C. Doerr, “Silicon photonic integration in telecommunications,” Front. Phys. 3, 1–16 (2015). [CrossRef]
15. F. Lu, B. Zhang, Y. Yue, J. Anderson, and G. Chang, “Investigation of Pre-Equalization Technique for Pluggable CFP2-ACO Transceivers in Beyond 100 Gb/s Transmissions,” J. Lightwave Technol. 35(2), 230–237 (2017). [CrossRef]
16. “Sequential quadratic programming,” https://en.wikipedia.org/wiki/Sequential_quadratic_programming
17. Q. Wang, Y. Yue, and J. Anderson, “Detection and Compensation of Bandwidth Limitation and Modulation Nonlinearity in Coherent IQ Transmitter”, accepted by European Conference on Optical Communication (2018), paper Th.2.25.
18. Z. Zhang, C. Li, J. Chen, T. Ding, Y. Wang, H. Xiang, Z. Xiao, L. Li, M. Si, and X. Cui, “Coherent transceiver operating at 61-Gbaud/s,” Opt. Express 23(15), 18988–18995 (2015). [CrossRef] [PubMed]
19. Q. Wang, Y. Yue, X. He, A. Vovan, and J. Anderson, “Accurate model to predict performance of coherent optical transponder for high baud rate and advanced modulation format,” Opt. Express 26(10), 12970–12984 (2018). [CrossRef] [PubMed]
20. M. Faruk and S. Savory, “Digital signal processing for coherent transceivers employing multilevel formats,” J. Lightwave Technol. 35(5), 1125–1141 (2017). [CrossRef]
21. “Coefficient of determination,” https://en.wikipedia.org/wiki/Coefficient_of_determination
22. H. Kawakami, T. Kobayashi, E. Yoshida, and Y. Miyamoto, “Auto Bias Control technique for optical 16-QAM transmitter with asymmetric bias dithering,” Opt. Express 19(26), B308–B312 (2011). [CrossRef] [PubMed]
23. T. Gui, C. Li, Q. Yang, X. Xiao, L. Meng, C. Li, X. Yi, C. Jin, and Z. Li, “Auto bias control technique for optical OFDM transmitter with bias dithering,” Opt. Express 21(5), 5833–5841 (2013). [CrossRef] [PubMed]
24. H. S. Chung, S. H. Chang, J. H. Lee, and K. Kim, “Field trial of automatic bias control scheme for optical IQ modulator and demodulator with directly detected 112 Gb/s DQPSK Signal,” Opt. Express 21(21), 24962–24968 (2013). [CrossRef] [PubMed]
25. X. Li, L. Deng, X. Chen, M. Cheng, S. Fu, M. Tang, and D. Liu, “Modulation-format-free and automatic bias control for optical IQ modulators based on dither-correlation detection,” Opt. Express 25(8), 9333–9345 (2017). [CrossRef] [PubMed]