Open Access
Add to BibSonomyAbstract
An Auto Bias Control (ABC) technique for the IQ-modulator of a flexible optical QAM transmitter is described. This technique can support various optical QAM signal formats with Nyquist filtering and electronic dispersion pre-compensation. 16, 32 and 64-QAM signals (21 Gbaud) are successfully generated, and all bias voltages are held to their optimum value even when signal format is changed.
© 2014 Optical Society of America
1. Introduction
To handle the ever-increasing data traffic, high order optical QAM schemes utilizing Nyquist filtering are being widely investigated [1,2]. Because QAM signal order is strongly restricted by transmission length, transmitter flexibility is an important goal for future elastic optical networks. When the transmission length is changed due to dynamic routing, the transmitter must change the order of QAM or baud rate [3], while operating the IQ-modulator in the best condition. For long time operation in commercial use, bias condition monitoring and ABC technique are essential to cancel the DC bias drift of the optical modulator. We have already proposed an ABC technique for the IQ-modulator in high order QAM transmitters, based on asymmetric bias dithering [4–7]. Several other approaches have since been proposed; analyzing the statistics of output optical signal [8] or using differential phasor monitor [9]. To realize a next generation flexible QAM transmitter, the ABC circuit must handle IQ-modulators driven with analog-like drive signals, for example, high order QAM with Nyquist filter and/or electronic pre-distortion. Moreover, the ABC circuit must offer stable operation even when the signal format and/or baud rate are changed. To the best of our knowledge, however, no report on ABC circuit performance has addressed these requirements. This paper shows that asymmetric bias dithering can be applied to high order QAM signals with Nyquist filtering and electronic dispersion pre-compensation. We also show that our ABC circuit can hold bias voltages to their optimum value even while changing the optical modulation format. Experiments show our ABC works correctly for 16, 32, and 64-QAM signals (21 Gbaud) with Nyquist filtering (roll off factor of 0.1). Measured Q-penalty induced by our ABC is 0.3 dB.
2. Construction of transmitter
Figure 1 shows the construction of the proposed flexible optical Dual Polarization (DP) 2n-QAM transmitter with ABC circuit. CW light is divided and launched into two IQ-modulators. Each IQ-modulator is driven by two 2[n/2]-level drive signals (denoted DATAI and DATAQ), corresponding to In-phase data and Quadrature-phase data, where [x] denotes the ceiling of x. The Phase Shifter (PS) and Child Mach-Zehnder modulator (CM) inside the IQ-modulator are biased at the θ = ± π/2 and null points, respectively. In Fig. 1, bias voltages are denoted BIASI, BIASQ and BIASPH. DATAI and DATAQ are generated by a Digital Signal Processor (DSP) and Digital Analogue Converter (DAC), using n bit streams denoted D1~Dn. To generate high order QAM signals, nonlinearity of the driver amplifiers and IQ-modulator must be suppressed. In previous work, we used pre-emphasis to suppress this nonlinearity [6]. In this study, we utilize the linear region of these devices, so the swing voltages of DATAI and DATAQ are set to less than 0.4x2Vπ, where Vπ is the half wave voltage at the RF port of the IQ-modulator. Optical electric fields EI and EQ generated by two CMs are coupled after the PS, yielding a 2n-QAM signal with EI ± jEQ. The sign of ± jEQ corresponds to the sign of θ = ± π/2. Two 2n-QAM signals are coupled using Polarization Beam Coupler (PBC), which generates a DP-2n-QAM signal.
3. ABC for flexible QAM transmitter based on asymmetric bias dithering
Figure 1 also shows the construction of the ABC circuit based on asymmetric bias dithering [4–7]. The oscillator in Fig. 1 generates three sine waves with angular frequency ωd. Here, ωd is very small compared to the baud rate. Two of them (DitherI and DitherQ) are used to dither BIASI and BIASQ. The phases of DitherI and DitherQ are orthogonal, and they are described as DitherI = cos (ωdt), DitherQ = sin (ωdt). The third sine wave is used as the reference clock of the synchronous detector. The power of the optical signal is monitored by a Photo Detector (PD) located in the modulator. Detected optical power is input to the synchronous detector.
3.1 Detection of BIASk drift (k = I or Q) for QAM signal with Nyquist filtering
Figure 2(a) shows the simulated waveform of DATAk, for 16-QAM with Nyquist filtering. Four signal levels of DATAk are denoted as ± V0 and ± V1. Between the time slots, DATAk has overshoot induced by Nyquist filtering. The voltages of these overshoots are denoted as ± Vtop. Figure 2(b) shows |Ek|2 as a function of the voltage of DATAk. To simplify, the effect of Ditherk is not shown in Fig. 2(b). When BIASk is optimum, |Ek|2 induced by four signal levels (black circle) and overshoot (black triangle) are symmetrical. When BIASk drifts (green arrow in Fig. 2), |Ek|2 at these voltages are changed (white symbol). Because the optical power at CM output is proportional to the summation of |Ek|2, the drift of BIASk can be monitored by the PD. As mentioned before, swing voltage 2Vtop is set to less than 0.4x2Vπ. With this swing voltage, PD output is minimized when BIASI and BIASQ are optimum. Note that, if the swing voltage is large, PD output for non-Nyquist 16-QAM or QPSK is maximized with optimum BIASk. A detailed description is given in Chapter 3 of [6]. Because BIASk is dithered by Ditherk, the drift of BIASI (or BIASQ) induces intensity modulation at the output of PD with angular frequency ωd, and its phase is the same as that of DitherI (or DitherQ) [4–7].
Fig. 2 DATAk and Ek with and without bias drift (k = I or Q). (a) Simulated waveform of DATAk for 16-QAM with Nyquist filtering (b) |Ek|2 of each of the four signal levels (circle) and overshoot (triangle). Black: BIASk is optimum, White: BIASk has drifted
3.2 Detection of the BIASPH drift for QAM signal with Nyquist filtering
Figure 3 schematically shows the four constellations of 16-QAM, assuming that BIASk is optimum though BIASPH has drifted. θ is ± π /2 ± δ, where 0 < δ < π/2. Without dithering (black symbol), constellations are symmetrical with respect to the origin, though θ is not rectangular. Overshoot induced by Nyquist filtering does not break this symmetry, because + Vtop and -Vtop are symmetrical (see Fig. 2(a)). However, when asymmetric bias dithering is added, the symmetry is slightly broken, and optical power is modulated [4]. If -π/2 < θ < π/2, maximum shift appears when ωdt = π/4 (see red symbol at ‘I’) and ωdt = −3π/4 (blue symbol at ‘III’). If θ < -π/2 or π/2 < θ, maximum shift appears when ωdt = 3π/4 (green symbol at ‘II’) and ωdt = -π/4 (brown symbol at ‘IV’). It means that when BIASPH drifts, optical power is modulated with angular frequency 2ωd, even if Nyquist filtering is used. For m2-QAM, all of the constellations in Fig. 3 contain m2 symbols, but the outward form of the constellation (diamond shape) is not altered by m2. Maximum shift appears at the same phase and position, so optical power is modulated at the same angular frequency of 2ωd, independent of order m2 or baud rate.
Fig. 3 Constellations with drifted BIASPH. θ = ± π /2 ± δ, where 0 < δ < π/2. Black: without dithering, Red: ωdt = + π/4, Green: ωdt = + 3π/4, Brown: ωdt = -π/4, Blue: ωdt = −3π/4
3.3 Error signal for feedback loop
Because all bias drifts induce intensity modulation, error signals for each bias drift can be generated by a synchronous detector [4–7]. Figure 4 shows the error signal for BIASPH as a function of phase shift θ. Solid and dashed lines show error signal with and without Nyquist filtering, respectively. Order of QAM is 64 and 16; roll off factor α of Nyquist filter is 0.1. All lines have a zero cross point at the optimum bias point, θ = ± π /2, which means this ABC circuit worked correctly with and without Nyquist filtering. In Fig. 4, all lines have basically the same slope. This means that the sensitivity of error signals to BIASPH is almost identical with or without Nyquist filtering. The bias controller feeds back the error signals for each bias voltage.
Fig. 4 Simulated Error Signal for BIASPH. Solid line:with Nyquist filtering (α = 0.1),Dashed line:w/o Nyquist filtering
3.4 Polarity selector switch for θ
We have already show that our ABC can work correctly with electronic pre-compensation to counter the chromatic dispersion in transmission line, using the DATAk generated by an offline computer [7]. To achieve pre-compensation in real time operation, (EI ± j EQ) T−1 is calculated by a DSP, where T−1 is the inverse transfer function of the transmission line. The sign of ± j EQ and sign of θ ( ± π/2) are decided by the program running on the DSP. Figure 5 schematically shows the symbol rotation (white to black) induced by pre-compensation. Here, we assumed that sign of j EQ is plus and correct θ is + π/2. When θ is set to the correct value, the optical phase is decreased; conversely, if θ is incorrect, - π/2, optical phase is increased. In real time operation with electronic pre-compensation, therefore, the ABC circuit must select the appropriate sign of θ. This study introduces the polarity selector switch (see Fig. 1) for θ. Figure 4 shows that slope of the error signal for BIASPH is reversed at θ = + π/2 and θ = −π/2. If the bias controller is programed to increase (decrease) θ when the error signal is positive (negative), θ will converge on −π/2. If the logic of the feedback signal is reversed, θ will converge on + π/2. This logic is selected by the polarity selector switch mentioned before. Because this switch is a software switch, it can be operated manually, or controlled by the DSP.
3.5 Bias voltage hold function
To realize elastic networks, the transmitter is required to change the signal format flexibly. While changing the format, DATAI and DATAQ are unstable. If these are just white noise, the output of the IQ-modulator can be treated as a non-Nyquist m2-QAM signal with infinite symbol rate, where m is infinite. This is because white noise is an analog signal with broad spectrum, so Vn-Vn-1 in Fig. 2 is zero, and the symbol rate can be treated as infinite. As mentioned before, this ABC technique is not affected by m or symbol rate, so the ABC circuit can work correctly. If DATAI and DATAQ drop to GND level, BIASI and BIASQ keep to the null point, because the ABC circuit in this setup minimizes the optical power. This means EI and EQ are almost extinguished, and all error signals become 0. Thus, while changing the signal format, this ABC circuit can hold the bias voltages, even though DATAI and DATAQ are unstable.
4. Experimental setup
We measured the performance of our ABC circuit for 16, 32 and 64-QAM signals (21 Gbaud). Because of the restriction of the experimental setup, DATAI and DATAQ were generated by offline processing and arbitrary waveform generator (AWG), instead of a DSP and DAC. Data sequences for DATAI and DATAQ were generated using multiple PRBS patterns. After Nyquist filtering with roll off factor of 0.1, generated data sequences were uploaded to the AWG, and drive signals DATAI and DATAQ for X and Y polarization were generated. Because no optical signal was transmitted in this experiment, electronic dispersion pre-compensation was not used. Vπ of the IQ-modulator was 3.6 V at the RF port, and 9.6~10.0V at the bias ports. Swing voltages of DitherI and DitherQ were controlled by the ABC circuit. The maximum swing voltages were 0.1 Vpp, with frequency of 530 Hz. Linewidths of the signal light and local light used for self-homodyne detection were less than 100 KHz. QAM signal and local light were launched into a coherent receiver based on a dual polarization optical hybrid and balanced photodetectors. The detected signals were stored in a digital storage oscilloscope, and the Q-factor was calculated by offline processing.
5. Results
Though both X-polarization and Y-polarization were controlled by the ABC circuit in our experiment, we show the result for single polarization only, because both polarizations showed almost the same performance. Figure 6 shows the convergence of BIASI, BIASQ and BIASPH, as functions of time t. Though BIASI, BIASQ were dithered at 530 Hz, the data logging system used in this experiment showed its mean value. ABC started at t = 0. After t = 3 min, all bias voltages had converged; the slight changes are due to DC bias drift. Hatched periods in Fig. 6 are periods for data uploading to AWG. In these periods, new data patterns (DATAI and DATAQ for X and Y) were sequentially uploaded, and each channel of AWG output random noise in the free running state. Optical 16, 64, and 32 QAM signals were generated in the periods without hatching. From t = 0 to t = 26 min., ABC circuit worked continuously, without reset or restart. Figure 7 shows the measured optical power spectrum and constellation of 64-QAM (t = 11min.), 32-QAM (t = 17min.) and 16-QAM (t = 23 min.). Side lobes of the optical spectra were well suppressed by Nyquist filtering, and clear constellations were achieved. We also measured the Q factor of the 64-QAM signal. Measured Q factor with manual bias control was 7.0 dB, and Q factor with ABC was 6.7 dB. It means that the Q-penalty induced by ABC was 0.3 dB.
Fig. 6 Bias Voltages vs. time t. Only results for Y polarization shown. ABC started at t = 0. Shaded areas shows data upload periods.
Fig. 7 Received optical power spectra and constellations (polarization y only). (a): 64-QAM (t = 11 min), (b): 32-QAM (t = 17min), (c): 16-QAM (t = 23min).
6. Conclusions
An ABC technique with Nyquist filtering that realizes flexible transmitters for m-QAM signals was described. 16, 32 and 64-QAM signals (21 Gbaud) were successfully generated, and all bias voltages were held to optimum values even when signal format was changed. Measured Q-penalty induced by the proposal was just 0.3 dB.
Acknowledgments
This work was supported in part by the Lambda Reach Project of the National Institute of Information and Communications Technology (NICT) of Japan.
References and links
1. T. Kobayashi, A. Sano, A. Matsuura, M. Yoshida, T. Sakano, H. Kubota, Y. Miyamoto, K. Ishihara, M. Mizoguchi, and M. Nagatani, “45.2Tb/s C-band WDM transmission over 240km using 538Gb/s PDM-64QAM single carrier FDM signal with digital pilot tone,” in Proceedings of ECOC2011, Th.13.C.6, (2011).
2. K. Roberts and C. Laperle, “Flexible transceivers,” in Proceedings of ECOC 2012, We.3.A.3, (2012).
3. M. Jinno, B. Kozicki, H. Takara, A. Watanabe, Y. Sone, T. Tanaka, and A. Hirano, “Distance-adaptive spectrum resource allocation in spectrum-sliced elastic optical path network,” IEEE Commun. Mag. 48(8), 138–145 (2010). [CrossRef]
4. H. Kawakami, E. Yoshida, and Y. Miyamoto, “Asymmetric dithering technique for bias condition monitoring in optical QPSK modulator,” Electron. Lett. 46(6), 430–431 (2010). [CrossRef]
5. H. Kawakami, E. Yoshida, and Y. Miyamoto, “Bias control technique based on asymmetric bias dithering for optical QPSK modulation,” J. Lightwave Technol. 30(7), 962–968 (2012). [CrossRef]
6. 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]
7. T. Kobayashi, Y. Miyamoto, K. Yonenaga, and H. Kawakami, “Pulse-carver-free RZ-64 QAM transmitter with electronic CD pre-compensation and auto bias control,” in Proceedings of ECOC 2013, We.4.C.4 (2013). [CrossRef]
8. P. S. Cho and M. Nazarathy, “Bias control for optical OFDM transmitters,” IEEE Photon. Technol. Lett. 22(14), 1030–1032 (2010). [CrossRef]
9. H. G. Choi, Y. Takushima, H. Y. Choi, J. H. Chang, and Y. C. Chung, “Modulation-format-free bias control technique for MZ modulator based on differential phasor monitor,” in Proceedings of OFC/NFOEC2011, JWA33. [CrossRef]
