The 2012 International Conference on Advanced Technologies for Communications (ATC 2012) Electro-Optic Modulator Drift Compensation Based on DSP System Pham Toan Thang1, Vu Van Yem2, Bach Gia Duong3 and Bernard Journet1, Member IEEE SATIE, Institut d'Alembert, ENS Cachan, UniverSud, CNRS 61 Avenue du Président Wilson, 94230 Cachan, France School of Electronic and Telecommunications Hanoi University of Science and Technology Dai Co Viet Road, Hanoi, Vietnam University of Engineering and Technology Vietnam National University in Hanoi 144 Xuan Thuy, Quân Câu Giây, Hanoi, Vietnam bernard.journet@ens-cachan.fr The optical response of our modulator has been determined by applying a very low frequency signal (in this case Hz) triangle signal to the DC electrode of the modulator The frequency is chosen low enough to consider the transfer function as static The resulting curve shown in Fig presents the output voltage of a photodetector placed at the output of the modulator, as a function of the slowly varying voltage applied to the DC electrode Abstract—This paper presents a study concerning the improvement of an electro-optic modulator operation by controlling its optical bias point The evolution of the modulator characteristics can be followed through its non-linear behavior by detecting the second harmonic of a low-frequency modulating signal The control system has been implemented in a kit based on a DSP board and a data acquisition card; it is designed in order to keep the bias point at the optimal value for minimizing the second order non-linear effects So it is possible to compensate completely the effects of the modulator transfer function drift for more than hours 2.5 I INTRODUCTION Detector output (V) Optical telecommunication systems at high data rate (higher than some tens of GHz) require external modulation Electrooptic modulators (EOM) are therefore the fundamental devices used for modulating the intensity of a laser beam The EOM is characterized by its transfer function, corresponding to the optical output power Pout as a function of the modulating signal vm (t ) applied to the electrode In case of a MZ modulator, with an input power Pin , the relation between the output power and the applied voltage is a sine function where α is an attenuation coefficient and η is the visibility factor [1]; Vπ is called the half-wave voltage ⎡ ⎞⎤ ⎛π Pout (t ) = αPin ⎢1 + η cos⎜⎜ vm (t ) + φ ⎟⎟⎥ ⎠⎦⎥ ⎝ Vπ ⎣⎢ A B 1.5 1.0 -6 -4 -2 Biasing voltage (V) Figure Transfer function of the electrooptic modulator The average value of the signal is indicated by the horizontal line The transfer function displays a sinusoidal behavior and the equation corresponding to the experimental results in Fig is given by: (1) Vout (Vb ) = VoM sin (K Vb + φ0 ) + V0 The EOM used for this study is a Mach-Zehnder (MZ) intensity modulator It is a commercial device made of lithium niobate material (LiNbO3, ref MX-LN-10 from Photline Company) with a 12 GHz modulation bandwidth It has aready been studied in a previous work [2] The EOM has two electrodes, one for the biasing signal (for DC or quasi DC signals) and one for the modulating signal corresponding to the data to be transmitted (AC high frequency signals) A laser diode (Fujitsu FLD5F6CX) at 1535 nm wavelength provides the constant laser beam intensity at the optical input of the EOM 978-1-4673-4352-7/12/$31.00 ©2012 IEEE 2.0 (2) where VoM = (0.899 ± 0.003) V , V0 = (1.748 ± 0.004) V , K = (0.457 ± 0.002) V -1 and φ0 = (0.960 ± 0.005) rd From this equation we can calculate the half-wave voltage such as K = π / Vπ and we find Vπ = (6.880 ± 0.023) V [2] A good transmission is obtained in case of a wide open eye pattern and the modulator must be biased for a symmetric operation The best behavior of the modulator is obtained at its 367 half-transparency bias points, such as A or B in Fig In this case, the symmetry of the transfer function with respect to the average value is optimal These points are called quadrature bias points [2] If the EOM is biased at the quadrature position then the transfer function is symmetric around this bias point, and hop (t ) contains only odd harmonics In case of a drift of the transfer function, the bias point is no more at the quadrature position and the curve is no more symmetric, even harmonics will appears in hop (t ) The synchronous detection process is built in order to Unfortunately electro-optic modulators are not perfectly stable with time and the optimum bias point changes during the operating time In fact the phase shift φ in (1) is a function of time leading to the transfer function drift When φ (t ) changes the transfer function of the EOM is shifted accordingly, to the right or to the left, during the operating time The drift of the EOM transfer function can be explained by different effects such as changes of working temperature, optical coupling efficiency and photorefractive effects [3, 4, 5, 6] detect the second harmonic of the modulating signal vd (t ) The optical signal at the output of the EOM is detected by a photodetector, then filtered around f d ; then there is the multiplication process by a reference signal at f d and a low pass filtering (see Fig 2) If the bias voltage is constant, and if the transfer function is horizontally shifted then the bias point is no more in the optimal quadrature position Modulators made of lithium niobate material are much more stable than electro-optic polymer based devices [7, 8] Nevertheless a drift can be observed even at a medium term time scale The purpose of this paper is to show how to compensate the drift of the transfer function by using a system based on a DSP kit II Figure Principle of the modulator non-linearity estimation method The output of the second harmonic detection scheme indicates how much second order non-linearity there is in the modulated signal and so it is an indication about the position of bias point on the transfer function This quantity is called here the non-linear indicator and it is noted NLI DETERMINATION OF THE MODULATOR DRIFT EFFECTS A The Measurement method The technique which has been developed is based on a synchronous detection scheme The method for estimating the drift of the transfer function is based on an evaluation of the non linear behavior of the modulator [8] Testing the linearity of the modulator is performed by the way of a dither signal which is a sine signal, applied to the biasing DC electrode for modulating the optical beam vd (t ) = VD sin( 2πf d t ) B Experimental setup The complete system is organized around a DSP board eZdspTM F28335 from Spectrum Digital Inc (based on a Texas Instrument TMS320F28335 DSP at 150 MHz) which is associated to a data acquisition daughter card F28335 from LinkResearch Company The system can be designed in MatlabSimulink for an easy graphical programming, and then the working sheet is exported into Code Composer Studio software for building the final application which has to be loaded to the DSP board In this system we use the digital-to-analog and analog-to-digital converters of the Link-Research F28335 card, which are working on 13 bits and ±10 V range (3) As the modulating signal vm (t ) should not perturb the high frequency modulation dedicated to the transmitted data, we have chosen a rather low frequency f d = 500 Hz For detecting a non linear effect with a good sensitivity the dither signal is a large signal with an amplitude VD = 850 mV The modulating signal vd (t ) is obtained by a VCO block in Matlab/Simulink with a sampling frequency f s = 40 kHz leading to 80 points in one period The signal at f d = kHz is build with the same method Taking into account a DC bias voltage Vb and the modulating signal vd (t ) applied simultaneously to the DC electrode vm (t ) = Vb + vd (t ) The signal at the output of the photodetector is directly acquired by the Link-Research board The sampling frequency is always 40 kHz (4) The bandpass filter is an important part of the system It has been designed with the Filter Design & Analysis toolbox of Matlab In this study an IIR has been chosen, it is a second order Butterworth For designing the filter the center frequency is kHz, the sampling frequency 40 kHz and the two cut-off frequency are 998 Hz and 1002 Hz The frequency response of the filter (shown in Fig 3) has been obtained from a function generator and an oscilloscope Then the transfer function in the time domain of the EOM is obtained from (1) and (4) ⎞⎤ ⎛π P (t ) α ⎡ hop (t ) = out = ⎢1 + η cos⎜⎜ [Vb0 + vd (t )] + φ (t ) ⎟⎟⎥ ⎢⎣ Pin (t ) ⎠⎥⎦ ⎝ Vπ (5) 368 1.0 100 50 NLI (mV) Vout / Vin 0.8 0.6 -50 0.4 -100 0.2 10 15 20 25x10 Time (s) 980 990 1000 1010 1020 Figure Long term evolution of the non linear indicator caused by the EOM transfer function drift Frequency (Hz) Figure Frequency response of the bandpass filter III Experimentally the bandwidth is found to be Δf = 4.2 Hz The corresponding quality factor is therefore equal to f d / Δf and so Q = 238 EXPERIMENTAL RESULTS WITH CONTROL PROCESS A Compensation of the drift From the knowledge of NLI it is possible to build a control system for compensating the changes of φ (t ) by adding a control signal to the biasing voltage Vb in order to keep NLI ≈ : the bias point is controlled at the quadrature position The control loop is designed with a PID controller [9, 10] The experimental system for controlling the bias point of the EOM at a quadrature position is presented as it has been developped with Matlab Simulink in Fig C Determination of the nonlinear indicator The behavior of the modulator can be determined by applying a scanning bias voltage Vb from −8.4 V to +8.5 V The modulating signal has an amplitude of 0.85 V and a frequency 500 Hz This experiment is equivalent to an imposed drift of the transfer function in a short time in order to avoid changes of the modulator characteristics It is why the value of NLI can be rather high The variations of NLI as a function of the biasing voltage are presented in Fig The behavior of NLI (Vb ) can be well modeled by a sine function as shown by the dashed lines in Fig The signal vm (t ) applied to the DC electrode has three components, tuned from the eZdsp-DSP-F28335 board: the DC biasing voltage Vb0 , the dither signal vd (t ) at f d = 500 Hz and the control signal Vbc obtained from the digital PID A second study is performed by registering the natural drift for a long time experiment The evolution of the NLI parameter as a function of time has been recorded and the corresponding results for an experimental time of hours are shown in Fig The peak-to-peak variation of NLI is 248 mV and the standard deviation is 56 mV 2000 NLI (mV) 1000 -1000 -2000 -8000 -6000 -4000 -2000 2000 4000 6000 8000 Vb0 (mV) Figure Evolution of the Non Linear Indicator (NLI), when the DC bias voltage changes from -8.4 V to +8.5 V Figure Experimental system for controlling the EOM bias point at quadrature position as designed with Matlab Simulink 369 vm (t ) = Vb0 + Vbc + vd (t ) These results are preliminary results with the system based on the DSP kit As it is well known from other experiments that the temperature is also a key parameter for drift effect [11], it is important to remind that in this experiment there is no temperature control for the EOM With temperature control the drift would be reduced and the variations of the biasing voltage would be smaller The future works concern the integration of two control process (bias point and temperature) on the same DSP board (6) The generation of dither signal and of reference signal at kHz is achieved by two VCO blocks The input signal at ADC1 comes from the photodetector and the output signal vm (t ) is placed at DAC1 for applying to the EOM electrode The other outputs are Vbias = Vb + Vbc at DAC3 and NLI at DAC5 The system is also linked to the PC computer by an USB/RS232 data link for data acquisition IV In this paper, it has been shown that the effects of the drift of an EOM can be measured by determining the nonlinear behavior of the modulator By controlling, with a digital PID, the second harmonic of a low frequency modulating signal, it is possible to optimize the bias point of the EOM for keeping the bias point in quadrature position Finally a reduction of the non linear indicator by a coefficient of at least 18 has been obtained thanks to the bias point control process That demonstrates the feasibility of the method and the efficiency of the designed system B Variations of the NLI with the control process The effects of the EOM drift are presented by the way of the nonlinear indicator in Fig The duration of the experiment is 30000 s or 8.3 hours 40 NLI (mV) 20 REFERENCES [1] B.E.A Saleh and M.C Teich, Fundamental of photonics, WileyInterscience, 2nd Edition, 1200 pages, 2007 [2] Dang Thanh Bui, Chi Thanh Nguyên, Isabelle Ledoux-Rak, Joseph Zyss, Bernard Journet, “Instrumentation system for determination and compensation of electro-optic modulator transfer function drift”, IOP, Meas Sci Technol Vol 22, 125105, 2011 [3] M Aillerie, N Théofanous, H L Saadon, “Thermo-optic effect in an electro-optic modulation system,” Proceedings of the New Achievements in Materials and Environmental Sciences (NAMES), 3rd France-Russia Seminar, pp 87-91, Metz, France, 2007 [4] Weirong Mo, Nanguang Chen, “Source stabilization for high quality timedomain diffuse optical tomography,” Proceeding of SPIE, vol 7170, 2009 [5] Jeffrey Snoddy, Yun Li, Fabien Ravet, Xiaoyi Bao, “Stabilization of electro-optic modulator bias voltage drift using a lock-in amplifier and a proportional-intergral-derivative controller in a distributed Brillouin sensor system,” Applied Optics vol 46, no 9, 20 Mar 2007 [6] G L Li, R B Welstand, W X Chen, J T Zhu, S A Pappert, C K Sun, Y Z Liu, and P K L Yu Wang, “Novel Bias Control of Electroabsorption Waveguide Modulator,” IEEE Photonics Technology Letters, vol 10, no 5, May 1998 [7] Suntak Park, Jung Jin Ju, Jung Yun Do, Seung Koo Park, and MyungHyun Lee, "Thermal stability enhancement of electrooptic polymer modulator,” IEEE Photonics Technology letters, Vol 16 (1), Jan 2004 [8] Heuk Park and Wol-Yon Hwang, "Origin of direct current drift in electrooptic polymer modulator,” Applied Physics Letters, vol 70 (21), May 1997 [9] Gene F Franklin, J David Powell, Michael L Workman “Digital Control of Dynamic Systems,” 2nd ed., Addison-Wesley Publishing Company, pp 222-229, 1990 [10] Takahashi, Y., Rabins, M J., and Auslander, D M., “Control and Dynamic Systems,” Addison-Wesley Reading, Massachusetts, 1972 [11] Dang Thanh Bui, Lam Duy Nguyên, Bernard Journet, “Improving the behavior of an electro-optic modulator by controlling its temperature”, Journal of Electronics and Communications - REV, Vol 1, N°1, pp 75-79, Jan.-March 2011 -20 -40 10 15 20 25 30x10 Time (s) Figure Evolution of the non linear indicator during the control process The average value of NLI is −1.314 μV , the standard deviation is 3.074 mV and the peak-to-peak variation is 21 mV Comparing the peak to peak variations with the case without control (Fig 5) there is a reduction by a coefficient 11.8 for the NLI variations; comparing the standard deviation the reduction of the variations corresponds to a factor 18.2 thanks to the bias point control process For compensating the drift there is a change of the biasing voltage The variations of Vbias = Vb + Vbc as a function of time are shown in Fig 5850 Biasing voltage (mV) CONCLUSION 5800 5750 5700 5650 10 15 20 25 30x10 Time (s) Figure Evolution of the EOM biasing voltage during the control process 370 ... function by using a system based on a DSP kit II Figure Principle of the modulator non-linearity estimation method The output of the second harmonic detection scheme indicates how much second order... variations with the case without control (Fig 5) there is a reduction by a coefficient 11.8 for the NLI variations; comparing the standard deviation the reduction of the variations corresponds... DETERMINATION OF THE MODULATOR DRIFT EFFECTS A The Measurement method The technique which has been developed is based on a synchronous detection scheme The method for estimating the drift of