0 Bewertungen0% fanden dieses Dokument nützlich (0 Abstimmungen)
19 Ansichten4 Seiten
Blind method for chromatic dispersion estimation in coherent optical receivers. Method is independent on modulation formats and signal spectra. Excellent performance is demonstrated even in extremely bandlimited systems.
Blind method for chromatic dispersion estimation in coherent optical receivers. Method is independent on modulation formats and signal spectra. Excellent performance is demonstrated even in extremely bandlimited systems.
Blind method for chromatic dispersion estimation in coherent optical receivers. Method is independent on modulation formats and signal spectra. Excellent performance is demonstrated even in extremely bandlimited systems.
Abstract In this paper, we present an accurate blind
method for chromatic dispersion estimation in coherent
optical receivers. The method is independent on modulation formats and signal spectra. Additionally, the estimation method is proven to be robust against sampling phase, polarization mode dispersion, carrier phase and frequency offset, and clock frequency offset. Excellent performance is demonstrated even in extremely bandlimited systems, so- called faster than Nyquist systems. Keywords Chromatic dispersion, polarization-mode dispersion, coherent detection, digital receiver, performance monitoring. I. INTRODUCTION NHANCED digital signal processing (DSP) has significantly improved the performance of coherent optical transmission systems. Due to IC technology advances, longer transmission distances can be achieved due to the digital compensation of chromatic dispersion (CD), polarization mode dispersion (PMD), laser noise, nonlinear effects, hardware imperfections, etc. Higher order modulation formats and efficient spectral shaping are currently one of the main research topics. The modulation format and baud rate are supposed to be flexible in the next generation (elastic) optical networks. In such networks, the network management will be able to select the best/optimum transmission scenario depending on the link parameters. Some link parameters, such as the amount of CD, must be initially efficiently estimated, and later compensated in the digital domain. Estimation techniques can be divided into two categories: i) blind non-data added methods without training symbols (additional redundancy), and ii) training based methods utilizing specially designed sequences. Techniques in both groups have their own advantages and disadvantages related to complexity, accuracy, tracking capability, required redundancy, sensitivity to sampling clock and laser frequency offset, and various related aspects. In CD uncompensated links, after coherent detection and correct signal amplification and quantization, the first most important operation in digital coherent receivers is CD estimation. Accumulated CD is efficiently compensated using frequency domain equalization.
The authors are with Huawei Technologies Duesseldorf GmbH, Riesstrasse 25, 80992, Munich, Germany. (e-mail: nebojsa.stojanovic@huawei.com).
Specially designed training sequences can be used to estimate channel imperfections such as CD and PMD [1 2]. These methods introduce significant redundancy for significant CD estimation and require careful handling of the carrier and clock phase/frequency offset. However, when fast start-up is required their complex hardware realizations may prohibit the use of these methods. On the other hand, blind methods based on the received signal (or signal modifications) autocorrelation functions [35] (directly related to the clock tone quality) completely fail in Nyquist systems with a roll-off factor (ROF) equal to 0. Additionally, none of these algorithms can work in faster than Nyquist systems (FTN) [6] that has become very attractive in strongly bandlimited systems applications. In case of fast start-up, hardware complexity becomes a critical issue for algorithms presented in [45] which also do not distinguish the sign of CD. A simple blind algorithm based on the peak-to-average-power ration (PAPR) is presented in [7]. This algorithm is extremely robust in any transmission scenario and can be realized with small implementation effort. In this paper, we present a novel CD estimation algorithm that is also applicable in any scenario, while at the same time is more accurate than the PAPR method. The new method is based on monitoring small signal values (MSSV) that fall below the predefined threshold. Some algorithm modifications are discussed as well. II. AUTOMATIC GAIN CONTROL IN COHERENT OPTICAL RECEIVERS The block diagram of a polarization division multiplexed (PDM) coherent receiver is presented in Fig. 1. After a polarization-beam splitter (PBS), the two orthogonal polarizations are mixed with a local oscillator (LO) laser in two 90 22 optical hybrids and sampled by an analog-to-digital converter (ADC). The automatic gain control (AGC) follows the power signal variations, which are mainly caused by switching channels in wavelength- division multiplexing (WDM) systems, and adjusts the signal swing to the ADC input in order to effectively exploit ADC resolution and to avoid serious ADC clipping. DSP is realized in a parallel structure with clocks between 100 and 500 MHz in 32-GBaud systems, and performs the operation by block processing. Some slow varying processes such as tuning laser frequency, AGC parameters adjustment, and ADC levels optimization, can be efficiently estimated and controlled by the use of a micro-controller (C). The phase detectors (PD) output is filtered by a low-pass filter (LPF) and controls a voltage Efficient and Low-Complexity Chromatic Dispersion Estimation in Coherent Optical Systems Nebojsa Stojanovic, Member, IEEE, Bangning Mao, and Fotini Karinou E 21st Telecommunications forum TELFOR 2013 Serbia, Belgrade, November 26-28, 2013. 978-1-4799-1420-3/13/$31.00 2013 IEEE 153
controlled oscillator (VCO) that provides the reference clock to the 4 ADCs. In practical implementations, each ADC block consists of subunits clocked at much lower frequencies. The AGC control unit should prevent signal clipping in both AGC and ADC block. Often, AGC is realized as a two-stage block. The first block uses feedback loop in order to track fast power variations without signal clipping. The second one is used for fine signal tuning to avoid signal clipping in the ADC block. The DSP provides the control signal to the second block by monitoring quantized values in outer bins. Usually, signal from both quadratures of the same polarization are monitored and one control signal is used for the AGC adjustment of the two AGC responsible for two real signals amplification in a single polarization. In the paper, an 8-bit ADC is used to obtain all simulation results. Absolute values of outer ADC limits are set to one. The C control signal adjusts the second amplifier so that outer ADC bins contain only 1/256 percent of quantized data. III. ESTIMATION METHOD DESCRIPTION AND SIMULATION RESULTS In coherent receivers, CD is efficiently compensated in the frequency domain. Polarization demultiplexing of two polarizations and PMD compensation are done by a 22 MIMO filter. Before CD compensation the residual CD value must be known. Therefore, we need accurate algorithms for estimating this value. In Fig. 2, we have plotted the histograms of the real part of a single polarization of a 32-GBaud PDM-QPSK signal with the following parameters: OSNR=12 dB, laser frequency offset of 300 MHz, laser linewidth (LW) of 100 kHz, and ROF=0.2, for four CD values. The size of the histogram bin was 1/254, and two samples per symbol have been used. Values on y-axis represent the number of observations of signal amplitudes falling in the considered bin. The x-axis represents signal amplitudes before quantization (after AGC, but before ADC). It is easy to observe that the portion of small amplitudes is minimal for CD=0. The amount of small amplitudes increases with CD. One can conclude that monitoring small values can be used to indicate the amount of CD. However, when LW and LO were set to 0 the signal statistics for CD=0 has changed the shape as presented in the last histogram (QPSK has two levels per quadrature). In practical systems, LOs linewidth is equivalent to at least several MHz. Using only 32000 symbols in 32-GBaud systems with CD=0 and LO=1 MHz one can obtain the histogram similar to the first histogram in Fig. 2. Later we will demonstrate that frequency offset does not influence the accuracy of the estimation method. Motivated by results presented in Fig. 2 we have developed the new estimation method that scans CD over the predefined range and monitor small quantized amplitudes. CD corresponding to the smallest number of counted quantized data falling below the specified monitoring threshold MT is selected. Let us denote by R I x x jx = + the complex signal from an arbitrary polarization. Several MSSV variants are investigated which count the following events: a) ( ) R abs x MT < , -2 0 2 0 1 2 3 4 x 10 4 data before ADC h i s t o g r a m CD=0 -2 0 2 0 1 2 3 4 5 x 10 4 data before ADC h i s t o g r a m CD=500ps/nm -2 0 2 0 1 2 3 4 5 6 x 10 4 data before ADC h i s t o g r a m CD=1000ps/nm -2 0 2 0 2 4 6 8 x 10 4 data before ADC h i s t o g r a m CD=10000ps/nm -2 0 2 0 500 1000 1500 2000 data before ADC h i s t o g r a m CD=0 LO=0 LW=0
Fig. 3. Histograms of data before ADC for different CD values
-2 0 2 0 1 2 3 4 x 10 4 data before ADC h i s t o g r a m CD=0 -2 0 2 0 1 2 3 4 5 x 10 4 data before ADC h i s t o g r a m CD=500ps/nm -2 0 2 0 1 2 3 4 5 6 x 10 4 data before ADC h i s t o g r a m CD=1000ps/nm -2 0 2 0 2 4 6 8 x 10 4 data before ADC h i s t o g r a m CD=10000ps/nm -2 0 2 0 500 1000 1500 2000 data before ADC h i s t o g r a m CD=0 LO=0 LW=0
Fig. 2. Histograms of data before ADC for different CD values.
Fig. 1. Phase- and polarization- diversity optical coherent receiver. 154
b) 2 R x MT < , c) ( ) ( ) R I abs x abs x MT + < , and d) 2 2 R I x x MT + < . The simulation results of 32-GBaud PDM-QPSK signal with OSNR=10 dB, CD=10000 ps/nm, LO=300 MHz, LW=100 kHz, ROF=0.1, and MT=0.5 using 2 15 symbols are presented in Fig. 3a. A cost function CF represents the normalized number of counted events for the specific CD value (CF normalization was different in figures 5, 6, and 7 that does not influences final conclusions). All four methods provide identical estimation results with zero estimation error. In the same figure we present the PAPR estimation results. We can observe that this method produces an estimation error of 300 ps/nm (marked in the figure). In Fig. 3b and 3c we present the estimation results with MT=0.2 and MT=0.4, respectively. Different methods requires different MT values in order to maximize the ratio max(CF)/min(CF), and to decrease the estimation error. For example, the first two methods should use MT=0.1, the third one MT=0.4, and the last estimation method performs well with MT=0.2. We have investigated the estimation method in 32- Gbaud PDM-QPSK (OSNR=12 dB) and 16QAM (OSNR=17 dB) systems with CD=10000 ps/nm and ROF of 0 and 0.5. Also, estimation for both modulation formats is carried out in FTN systems with 25 GHz 3-dB Gaussian optical spectrum. In all cases we have applied the method a) with MT=0.1 over 2 15 symbols. The results presented in Fig. 4 confirm the excellent performance of the new estimation method. In all six investigated cases the estimation error was 0. Unlike methods using autocorrelation [35], the new method performs well in case of ROF=0 and in FTN systems. The accuracy of estimation depending on sampling phase and PMD has been investigated. First, we have simulated 32 sampling phases within one symbol interval for the QPSK with OSNR=14dB and ROF=0. The simulation results presented in Fig. 5 indicate that the absolute maximum estimation error caused by sampling phase variations can reach 120 ps/nm. In the second type of simulations, we have introduced the first-order PMD -1000 -500 0 500 1000 0.18 0.2 0.22 0.24 0.26 0.28 0.3 CD-10000 (ps/nm) C F
QPSK FTN 16QAM FTN a)
-1000 -500 0 500 1000 0.2 0.25 0.3 CD-10000 (ps/nm) C F
QPSK ROF=0 16QAM ROF=0 b)
-1000 -500 0 500 1000 0.1 0.15 0.2 0.25 0.3 0.35 CD-10000 (ps/nm) C F
QPSK ROF=0.5 16QAM ROF=0.5 c)
Fig. 4. Estimation results in different systems: a) FTN, b) ROF=0, c) ROF=0.5. -600 -400 -200 0 200 400 600 1 1.5 2 2.5 CD-10000(ps/nm) C F 120ps/nm
X: -300 Y: 0.2078 CD-10000 (ps/nm) C F a b c d PAPR a)
-5000 0 5000 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 CD-10000 (ps/nm) C F
a b c d b)
-5000 0 5000 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 CD-10000 (ps/nm) C F
a b c d c)
Fig. 3. Estimation results of four proposed methods: a) MT=0.1, b) MT=0.2, c) MT=0.4. 155
defined by a polarization rotation angle and a differential group delay of half a symbol interval. The polarization angle was varied from 0 to 345 in step of 15. In this scenario, the absolute estimation error was 210 ps/nm as one can observe in Fig. 6. All these problems are overcome by using several short blocks of data (e.g. 1024 symbols), scanning CD over each block, and summing up all estimation results. Using this method, the results presented in Fig. 3 and 4 will not influenced by sampling phase and PMD variations. The estimation method behavior of two modulation formats, QPSK and 16QAM, has been investigated in the presence of frequency offset (ROF=0). Frequency offset was varied from 0 to 2 GHz in step of 100 MHz. The clock offset of 200 ppm was also added. The estimation results presented in Fig. 7 prove the robustness of the new method against the frequency and clock offset. The estimation accuracy was not affected at all. IV. IMPLEMENTATION All commercial coherent receivers include the CD compensation block. CD scanning can be easily done per data block (e.g. block of 1024 samples) that can be frozen in the input buffer of the CD compensation block. For each data block, CF values can be collected in counters corresponding to CD values. After several data blocks the final CD estimation can be done (selecting CD value corresponding to the smallest CF value). Hardware requirements for CD estimation are only CF counters and comparators for selection of samples with values lower than MT (e.g. the first proposed method). As all operations can be done in hardware at CMOS speed, the proposed method can also provide very fast start-up. After summarizing all simulation results, including minimum hardware requirements, and comparing the method with other algorithms already discussed in Introduction, the novel method most probably stands out as the best candidate for CD estimation in coherent optical receivers. V. CONCLUSION Fast start-up in coherent receivers is dominated by channel parameters estimation. We have proposed the novel method for fast and low-complexity CD estimation that can be easily implemented in the CMOS ASIC part of the digital receiver. The method is robust against sampling phase, PMD, and laser frequency offset. The accuracy of the proposed method can be additionally improved by optimizing the monitoring threshold. REFERENCES [1] E. Ip and J. M. Kahn, Compensation of dispersion and nonlinear impairments using digital backpropagation, J. Lightw. Technol., vol. 26, pp. 34163425, Oct. 2008. [2] C. R. S. Fludger, T. Duthel, D. van den Borne, C. Schulien, E.-D. Schmidt, T. Wuth, J. Geyer, E. De Man, G.-D. Khoe, and H. de Waardt, Coherent equalization and POLMUX-RZ-DQPSK for robust 100-GE transmission, J. Lightw. Technol., vol. 26, pp. 64 72, Jan. 2008. [3] F.N. Hauske, Z. Zhang, C. Li, C. Xie, and Q. Xiong, Precise, robust and least complexity CD estimation, in Proc. OFC, Los Angeles, USA, Mar. 2011, Paper JWA32. [4] F. Pittala, F. N. Hauske, Y. Ye, N. G. Gonzalez, and I. T. Monroy, Combined CD and DGD monitoring based on data-aided channel estimation, in Proc. SPPCom, Toronto, Jun. 2011, paper SPTuC3. [5] Q. Sui, A. P. T. Lau, and C. Lu, Fast and robust chromatic dispersion estimation using auto-correlation of signal power waveform for DSP based-coherent systems, in Proc. OFC, Los Angeles, USA, Mar. 2011, Paper OW4G.3. [6] C. Malouin, P. Thomas, B. Zhang, J. O'Neil, and T. Schmidt, Natural expression of the best-match search Godard clock-tone algorithm for blind chromatic dispersion estimation in digital coherent receivers, in Proc. SPPCom, Colorado Springs, USA, Jun. 2012, Paper SpTh2B.4. [7] C. Xie, Chromatic dispersion estimation for single-carrier coherent optical communications, IEEE Photon. Technol. Lett., vol. 10, pp. 992995, May 2013.
-600 -400 -200 0 200 400 600 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 CD-10000(ps/nm) C F 210ps/nm
Fig. 6. Estimation results in the presence of first-order PMD with 24 different power coupling factors. -1000 -500 0 500 1000 1 1.5 2 2.5 CD-10000(ps/nm) C F QPSK 16QAM
Fig. 7. Estimation results in the presence of frequency offset. -600 -400 -200 0 200 400 600 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 CD-10000(ps/nm) C F 210ps/nm
Fig. 6. Estimation results in the presence of first-order PMD with 24 different power coupling factors. 156