JThA51.pdf
Receiver-side a1814_1.pdf adaptive opto-electronic chromatic dispersion compensation
Arup Polley and Stephen E. Ralph
School of Electrical and Computer Engineering Georgia Institute of Technology 777 Atlantic Drive, Atlanta, Georgia 30332-0269 [e-mail]
Abstract: We demonstrate, via comprehensive simulation, a receiver-side adaptive chromatic dispersion compensation technique which reconstructs the received optical signal as an electrical baseband signal and subsequently applies linear electrical filtering to remove chromatic dispersion.
2006 Optical Society of America
OCIS codes: 070.6020 (Signal Processing); 060.2330 (Fiber Optics Communications)
1. Introduction Electronic dispersion compensation (EDC) and equalization of single mode fiber (SMF) has recently attracted considerable interest as a potentially low-cost, adaptive solution to chromatic dispersion (CD) [1,2]. However, the performance improvement via an electronic only method is limited due to the nonlinear effects of square-law photodetection and the loss of phase information in the process. Although signal processing on the full optical signal has been performed using predistortion techniques [3] and coherent detection, a receive-side solution has advantages. Recently, a receiver-side solution has been proposed [4] which uses passive asymmetric Mach Zehnder interferometer (AMZI) to extract the instantaneous frequency information of the received optical signal. This information has been used to construct a replica of the optical signal in the radio frequency (RF) range and subsequently a dispersive medium was used to compensate the CD. Here, we propose baseband signal processing after the frequency information has been extracted using an AMZI. The method relies on analog signal processing to reconstruct the optical signal as a baseband electrical signal with phase information and subsequently employ a modified feedforward equalizer (FFE) to compensate the linear channel distortion due to CD. We thus avoid the requirement for a broadband frequency modulator and RF dispersive medium and enable adaptive tuning. Furthermore, the adaptive equalizer can mitigate other channel distortion along with the chromatic dispersion. The entire electrical signal processing block can be implemented using analog integrated circuits. We evaluate the performance using a full channel model including nonlinear effects as well as optical noise. 2. System description Figure 1 shows the optical channel, extraction of instantaneous frequency of the modulating signal using the AMZI and reconstruction of the baseband signal using the subsequent signal processing blocks. The basic principle is to integrate the frequency to obtain the phase and pass it through cosine and sine functions and then multiply these signals with the square root of the intensity signal to obtain the in-phase (I-phase) and quadrature-phase (Q-phase) of the baseband signal.
10 Transmitter SMF VOA EDFA 0-60km 12 dB PD1 AMZI PD2 TIA Irb1 |Erb|
+
Irb2
f
Integrator + bandpass filter
Cos Sin
I-phase Q-phase
Fig. 1 Optical channel and receiver reconstructing the baseband signal A realistic evaluation of the above scheme requires design considerations and proper consideration of the presence of noise. The received optical signal in baseband representation (Ebr(t)) is the output of an essentially linear channel. Considering the transfer function of the AMZI with -3dB transmission of both outputs tuned at the carrier frequency, the photodetector (PD) and the trans-impedance amplifier (TIA) the two outputs are given by
OSA 1-55752-830-6
a1814_1.pdf JThA51.pdf
I rb1 ( t ) = C
1 2
Erb ( t ) 1 + sin
2
f (t ) + n1 ( t ) F
(1)
1 f (t ) 2 I rb 2 ( t ) = C Erb ( t ) 1 sin + n2 ( t ) F 2 where, the period of frequency response F = 1/, and is the delay between two arms of the interferometer. The instantaneous frequency f(t) of the baseband signal can be extracted with small nonlinearity for f ( t ) F 1 .
The noise of the two signals n1(t) and n2(t) is correlated. C is the conversion gain of the detector-TIA combination. The method of Fig. 1 provides estimates of Erb(t) and f(t) (denoted by Erb ( t ) and f ( t ) respectively) are given by
2 2 Erb ( t ) = I rb1 ( t ) + I rb 2 ( t ) = C Erb ( t ) + n1+ 2 ( t )
rb 2
(2)
(3) + n1+ 2 ( t ) + Vb / C Erb ( t ) + n1+ 2 ( t ) + Vb / C rb Where, n1+2(t) and n1-2(t) is the noise obtained by adding and subtracting n1(t) and n2(t) respectively. From Eq. (3), it is apparent that the estimate of f(t) becomes noisy when signal strength | Erb(t)|2 becomes comparable to the noise power n1+2(t). To suppress this undesired noise enhancement an offset Vb is applied during the estimation of f(t). The value of Vb is chosen such that for large signal strength, distortion in f ( t ) is negligible. However for small
rb1 2
( I ( t ) I ( t ) ) = f t / F f ( t ) / F = ( () ) ( I (t ) + I (t )) + V E (t )
rb1 rb 2 b
Erb ( t )
2
+
n1 2 ( t )
2
signal strength estimation of f(t) is saturated and the noise is also suppressed. Though, the optimum value of Vb/C depends on the signal to noise ratio (SNR), we find a value of 0.2 times the average signal strength provides good results. We also note that the choice of F involves a tradeoff between the nonlinearity and SNR of the extracted estimate of f(t). The SNR of f ( t ) degrades for f ( t ) F 1 , as the first term in Eq. (3) becomes smaller. We find that a F=50GHz, corresponding to of 20 ps, provides good SNR while keeping the nonlinearity sufficiently low. The intensity and frequency estimates are used to generate the I-phase and the Q-phase components of the received optical signal (Fig. 1). The components used to create the functional blocks are realizable using analog circuits. Specifically, the square-root circuit the and analog divider circuit may be realized using MOS translinear circuits [5,6] with some bandwidth limitations. In the simulation we have assumed a bandwidth of 7.5 GHz for all the functional blocks. The integrator is coupled with a bandpass filter to provide a phase output which eliminates the slow phase drift but retains the necessary fast phase change information. We find that a bandwidth of 0.1GHz to 7.5GHz limits the phase excursion well within . This also helps in the realization of cosine and sine output with piece-wise linear function generation. The adder and multiplier circuits with necessary bandwidths are realizable. It is noted that the reconstructed signal may have a relatively slowly changing phase offset along with the noise and nonlinearity induced impairments. However, it contains the essential phase information necessary to mitigate the channel impairments. The linearized and reconstructed baseband optical signal is fed to a modified dual phase feedforward equalizer (FFE) as shown in Fig. 2a. The magnitude square of the output of the complex FFE makes the output independent of the phase drift extending over several FFE span. This feature makes the modified FFE insensitive to relatively fast phase variations which are difficult to track with the adaptation algorithm. This also allows for modest source drift and linewidth. The filter coefficients are updated using a least mean square error (LMSE)-like algorithm. It is noted that the FFE span required to effectively compensate the chromatic dispersion increases with the length of the fiber and is proportional to the pulse spread due to chromatic dispersion. Figure 2b shows the adapted FFE taps compensating 400km SMF.
I-phase Q-phase
Dual phase FFE
Magnitude square
Output
Fig. 2a Modified feed forward equalizer
Fig. 2b Adapted I phase (D)and Q phase (o) FFE taps for 400km SMF
OSA 1-55752-830-6
a1814_1.pdf JThA51.pdf
3. Performance evaluation The optical channel is simulated using the Optisystem software. The transmitter is a 10Gbps amplitude modulator with extinction ratio of 10dB modulating a laser with 10MHz linewidth and center frequency at 193.1THz. Each span consists of variable length of SMF from 0 to 60 km, a variable optical attenuator (VOA) and an Erbium doped fiber amplifier (EDFA) of 12 dB gain and 4 dB noise figure. Simulation is performed for multiple lengths of fiber. The SMF model includes chromatic dispersion of 16.75ps/nm-km, dispersion slope of 0.075ps/nm2-km, attenuation coefficient of 0.2dB/km and nonlinear coefficient n2 of 2.610-20m2/W. The transmitted power is swept from 0dBm to -15dBm to vary the received OSNR. The signal is preamplified before the receiver to maintain constant received optical power. A 42GHz, 1st order Gaussian bandpass filter is used to attenuate the out of band ASE noise. A balanced PIN photodetector with 8GHz bandwidth is used to detect the signals after AMZI. The net noise after the photodetection incorporates signal-ASE beat noise, ASE-ASE, thermal noise, and shot noise. However the received power is adjusted so that the dominant contribution comes from signal-ASE beat noise. Noise figure of the TIA is 6 dB. The 7.5GHz Bessel-Thomson lowpass filter is used to limit the receiver bandwidth. Figure 3 shows the performance of the receiver for a fixed received OSNR of 23 dB. The compensator can extend the reach beyond 600km with forward error correction (FEC). Figure 4 illustrates the Q vs. received OSNR for different lengths of fiber. The receiver OSNR penalty is within 10 dB for 600km at the FEC limit.. As mentioned before, the accuracy of the reconstructed signal is limited by the noise. Furthermore, square-law detector and square-rooting circuit generate higher frequency harmonics which due to bandwidth limitation are lost in the signal reconstruction process. This produces a nonlinear distortion in the recovered baseband signal and limits the performance.
Fig. 3 Q after FFE vs. fiber length for 23 dB of Received OSNR
Fig. 4 Q vs. OSNR for different lengths of fiber
4. Conclusion We show that the frequency information of the received optical signal extracted using a passive optical component can be used to image the optical signal in baseband and make the end to end fiber channel linear. Linear electrical filter can then compensate for the chromatic dispersion impairment in the reconstructed signal. A modified FFE is demonstrated which can extend the SMF link lengths over 600km at 10Gb/s for low cost transmitters. The entire signal processing can be implemented using analog integrated circuits and therefore, can potentially reduce the cost of EDC implementation for metro area networks. 5. References
1. N. Alic, G. C. Papen, R. E. Sperstein, R. Jiang, C. Marki, Y. Fainman, S. Radic, P.A. Andrekson, Experimental demonstration of 10Gb/s NRZ extended dispersion-limited reach over 600km-SMF link without optical dispersion compensation, Proc OFC 2006, paper OWB7, (2006) 2. P. M. Watts, V. Mikhailov, M. Glick. P. Bayvel, R. I. Killey, Performance of optical single sideband transmission systems using adaptive electronic dispersion compensators, Proc ECOC 2005, paper Tu 4.2.4 (2005) 3. D. McGhan, C. Laperie, A Savchenko, C. Li, G. Mak, M. OSullivan, 5120 km RZ-DPSK transmission over 652 fibre at 10 Gbit/s with no optical dispersion compensation, Proc OFC 2005, paper PDP-27, (2005) 4. A. D. Ellis, M. E. McCarthy, Receiver-side electronic dispersion compensation using passive optical field detection for low cost 10Gbit/s 600 km-reach applications, Proc OFC 2006, paper OTuE4, (2006) 5. W. Gai, H. Chen, E. Seevinck, Quadratic-translinear CMOS multiplier-divider circuit, Elec. Lett. 33, 860-861 (1997) 6. V. Riewruja, K. Anuntahirunrat, W. Surakampontorn, A class AB CMOS square-rooting circuit, Int J. Electronics 85, 55-60 (1998)
OSA 1-55752-830-6