Method, device and communication system for reducing optical transmission impairments

Abstract

A method and device is provided for reducing optical transmission impairments, particularly nonlinear effects, of at least one link Said method comprising the following steps: extracting a phase information () from an optical signal (120) received via that at least one link, determining a nonlinear coefficient (), associated with the at least one link, based on the phase information (), applying a control mechanism (202) using the nonlinear coefficient (). Furthermore, a communication system is suggested comprising said device.

Claims

1. A method for reducing optical transmission impairments, particularly nonlinear effects, of at least one link, comprising the following steps: a) receiving an optical signal via the at least one link; b) converting the optical signal into a digital electrical representation; c) extracting a phase information of the optical signal from the digital electrical representation, d) determining with a processor a nonlinear coefficient, associated with the at least one link, based on the phase information, e) applying with the processor a control mechanism to the digital electrical representation using the nonlinear coefficient to compensate for impairments of the optical signal, wherein the phase information is extracted after a carrier recovery of the optical signal, wherein a cost function is derived based on the extracted phase information, and an optimization algorithm is applied in connection with the cost function to determine the nonlinear coefficient, and wherein the optimization algorithm is based on a steepest descent algorithm defined as
(i+1)=(i)+(i), wherein i is an index of a discrete time (i+1) is representing a value of the nonlinear coefficient at iteration (i+1) (i) is representing a value of the nonlinear coefficient at the preceding iteration step (i) is representing a convergence factor, comprising an effective fiber length L.sub.eff and a channel power P, and (i)=CF ()/, wherein CF denotes the cost function.

2. The method according to claim 1, repeating the steps c) to e) until a value of the nonlinear coefficient reaches or exceeds a given value.

3. The method according to claim 1, wherein the extracted phase information comprises a spreading of received symbols being part of the optical signal, and the determination of the nonlinear coefficient is such that a reduced spreading of the received symbols is achieved.

4. The method according to claim 3, wherein a respective spreading of the received symbols comprises respective phase differences between the received symbols and respective transmitted symbols which are derived either with or without a training sequence.

5. The method according to claim 4, wherein the optical signal is a coherent signal based on a 16 QAM modulation, 4.sup.th power is applied to the optical signal, the respective phase differences are derived from a 4.sup.th power signal, a cost function (CF), based on those derived phase differences, is defined as
CF=[.sub.1+.sub.3]*(1/R.sub.2)+.sub.2*(1/R.sub.1+1/R.sub.3) wherein .sub.1, .sub.2, and .sub.3 are representing a standard deviation for each of the respective phase differences, and R1, R2 and R3 are representing radii of a 16 QAM constellation.

6. The method according to claim 1, wherein the optimization algorithm starts by calculating two values of the cost function corresponding to two different values of the nonlinear coefficient, and a first starting value of the nonlinear coefficient is represented by a selected initial value.

7. The method according to claim 1, wherein the nonlinear coefficient is refined as a n-dimensional nonlinear coefficient, representing n single links, and the n-dimensional nonlinear coefficient is determined by a n-dimensional calculation.

8. A device comprising a converter for converting an optical signal received via at least one link into a digital electrical representation; and a processor unit configured for: a) extracting phase information of the optical signal from the digital electrical representation, b) determining a nonlinear coefficient, associated with the at least one link, based on the phase information, c) applying a control mechanism to the digital electrical representation using the nonlinear coefficient to reduce optical transmission impairments of the optical signal, including non-linear effects of the at least one link, wherein the phase information is extracted after a carrier recovery of the optical signal, wherein a cost function is derived based on the extracted phase information, and an optimization algorithm is applied in connection with the cost function to determine the nonlinear coefficient, and wherein the optimization algorithm is based on a steepest descent algorithm defined as
(i+1)=(i)+(i), wherein i is an index of a discrete time (i+1) is representing a value of the nonlinear coefficient at iteration (i+1) (i) is representing a value of the nonlinear coefficient at the preceding iteration step (i) is representing a convergence factor, comprising an effective fiber length L.sub.eff and a channel power P, and (i)=CF ()/, wherein CF denotes the cost function.

9. The device according to claim 8, wherein the processor unit is arranged so as to repeat the steps a) to c) until a value of the nonlinear coefficient reaches or exceeds a value.

10. The device according to claim 8, wherein said device is a communication device that is or is associated with a receiver for optical signals.

11. A communication system comprising the device according to claim 8.

Description

(1) Embodiments of the invention are shown and illustrated in the following figures:

(2) FIG. 1 shows a block diagram of an optical coherent receiver based on digital signal processing (DSP);

(3) FIG. 2a shows a block diagram of a DSP-based coherent receiver according to the proposed solution;

(4) FIG. 2b shows an exemplary flow chart of the proposed solution;

(5) FIG. 3 shows a histogram of the derived phase difference of a 16 QAM modulated signal;

(6) FIG. 4 shows a histogram of the derived phase difference of the 4.sup.th power of a 16 QAM modulated signal;

(7) FIG. 5 shows a 16 QAM constellation diagram to the 4.sup.th-power;

(8) FIG. 6 exemplarily shows a derived cost function CF for determining an optimized value of the nonlinear coefficient based on a 16 QAM modulated signal over a standard single-mode fiber;

(9) FIG. 7 shows an example of a derived cost function based on a 16 QAM modulated signal over a large-effective area pure silica core fiber;

(10) FIGS. 8 and 9 each shows an example concerning the quality performance of the proposed method;

(11) FIG. 10 to 13 each shows a further example of a signal constellation diagram applied to the 4.sup.th power, based on a further exemplary modulation format.

(12) With reference to FIG. 1, a common used DSP-based coherent receiver is exemplarily depicted according to prior art. In a first step a received signal 120 is digitally converted by a block of four analog-to-digital converters 101. In a following step bulk chromatic dispersion and nonlinear effects are compensated by a Digital Back-Propagation (DBP) algorithm implemented by a DBP Module 102. After a time synchronization provided by a clock recovery module 103, a signal polarization de-multiplexing is performed by a time domain equalizer 104, which can also be implemented in a carrier recovery module 105. The succeeding steps process received coherent signals via the modules carrier recovery 105, decision making on received symbols 106 and estimation of a bit error rate 107.

(13) The coherent receiver can be refined as a data-aided receiver (i.e. using training sequences (TS)). Nevertheless, the proposed method can also be realized by utilizing a receiver, which operates without training sequences (also referred to as blind receiver).

(14) The DBP algorithm or DBP module 102 requires a description of the link, which is used for back-propagation purposes. It is one of the advantages of the proposed solution that DBP can be used even by applying an arbitrary or incorrect link description. An incorrect link description can result (among others) the following, statistically independent, sources of errors: the fiber length (which is possibly incorrect), the fiber type (which is possibly wrong) or the power levels (which can not be measured accurately)

(15) Generally, it can be distinguished between homogeneous and inhomogeneous links. Homogeneous links comprise equal fibers for all spans (an optical fiber/cable terminated at both ends which may include devices that add, subtract, or attenuate optical signals) which is the usual scenario for point-to-point connections. Inhomogeneous links are usually found in meshed optical networks, where links comprising the same type of fiber can hardly be found.

(16) In a homogeneous scenario, the error on estimating the length of a single link or span does not really cause a problem as, after compensation of linear and nonlinear effects, this error will be averaged outprovided that the error is confined to a reasonable range. Errors of up to 20% on the length specification do not induce any significant impairment, in case DBP is used.

(17) On the other hand, in a meshed network or even on a single link, an error concerning the type of fiber may not be averaged out, resulting in a system outage after DBP is applied.

(18) Finally, in case of wrong measured power levels along the link, the same disadvantage is valid as for estimating the wrong length of the fiber: if the error is uniformly distributed, DBP provides an improvement, otherwise the system performance deteriorates.

(19) The solution presented herein solves the problem mentioned above: Exemplary results for a single homogenous link with a wrong estimation concerning the type of fiber but with exact knowledge of the CD value will be presented.

(20) Examples for different types of fibers are: Large-Effective Area Pure Silica Core Fiber (LA-PSCF) Standard Single-Mode optical Fiber (SSMF)

(21) The coherent receiver 200 shown in FIG. 2a is based on the receiver according to FIG. 1. In addition to FIG. 1, a feedback connection 220 is provided between the carrier recovery module 105 and the DBP module which is now an adaptive DBP (A-DBP) module 202. Further, an adaptive estimation module 210 is part of the feedback connection 220. A signal 230 which is the resulting outcome of the carrier recovery 105 is passed on to the Estimation Module 210, where the nonlinear coefficient parameter (Gamma) is estimated or calculated respectively by processing the internal signal 230 forwarded from the carrier recovery module 105 as will be described further below.

(22) According to an embodiment of the proposed solution a wrong description with regard to the type of fiber can initially be provided to the adaptive DBP module 202. Additionally, a correct description of a dispersion parameter is provided to the adaptive DBP module 202. Apart from this, further knowledge being available concerning the link will be the number of spans and the individual length of the spans.

(23) The adaptive algorithm according to the proposed solution, implemented in the adaptive estimation module 210, is explained in more detail, wherein a flow chart of the proposed solution is shown in FIG. 2b.

(24) If it is started with a wrong description of the type of fiber, the initial value of the nonlinear coefficient , i.e. (0), has to be estimated. Based on the fact that the nonlinear fiber coefficient , for commonly installed fibers, varies from 0.6 1/(W*km) (in case of Large-Effective Area Pure Silica Core Fiber (LA-PSCF)), to about 2 1/(W*km) (for the case of non-zero-dispersion shift fiber), the proposed initial value for (0) can be set (e.g., per default) at 1.3 1/(W*km), which may correspond to an average value for commercially available fibers. Accordingly, a different value is selected for (1), wherein (1) indicates the next iteration after (0). The actual selected value of the nonlinear coefficient (represented by signal 231 in FIG. 2) is forwarded to the Adaptive DBP Module 202.

(25) After frame recovery of the incoming signal 120 processed by the carrier recovery module 105, training sequences (TS) being part of the received signal 120 are extracted and used to derive the residual nonlinear phase difference between received and (originally) transmitted symbols or sequences of symbolssee step 250 in FIG. 2b. It is noted that this is also possible without any training sequence by deriving the residual nonlinear phase difference between received symbols and respective symbols after decision, (which is also called blind method or blind receiver).

(26) The phase difference between the two sequences is defined as
(t)=(t).sub.RX(t)

(27) wherein (t) either represents a sequence of training symbols, ((t)=.sub.TS(t)) or a sequence of already decided symbols ((t)=.sub.DEC(t)) and .sub.RX(t) represents the received symbols.

(28) Both, the use of a blind receiver (using decided symbols (t)=.sub.DEC(t)) and the use of a data-aided method (using training symbols (t)=.sub.TS(t)) is valid.

(29) According to a further embodiment the phase difference between two symbols (or sequences of symbols) can also be determined as follows:
(t)=|(t).sub.RX(t)|

(30) where | . . . | represents the absolute value of the phase difference (t).

(31) For this example a 16 QAM modulation format is applied for the received signal 120. FIG. 3 presents the respective histogram 300 of the derived phase difference (t) of such kind of 16 QAM modulated signal, comprising 12 peaks corresponding to the 12 phases of a 16 QAM modulated signal, whereby the value of the phase difference (t) at the very left and right side of the histogram 300 represent the same angle. This phase information can be obtained by mathematically manipulating the phase information being part of the received symbols (e.g., elimination of phase ambiguity).

(32) Following the proposed method, the 4.sup.th power is applied to the incoming signal 120 before deriving the residual nonlinear phase difference at the adaptive Estimation Module 210. The respective histogram 400 of the 4.sup.th power signal is shown in FIG. 4 where only three phases of a single quadrant can be identified accordingly. This information (spreading of received symbols), presented in FIG. 4, is the basis for calculating the nonlinear coefficient by deriving and evaluating a cost function as suggestedstep 251 in FIG. 2b.

(33) As it can be seen by the peak in the middle of the histogram (=0) of FIG. 4, some phases of the received symbols are identified more frequently, because two of the symbols of the 4.sup.th-power of a 16 QAM constellation as shown in FIG. 5 are corresponding to the same phaserepresented by symbol 502 and 503 in FIG. 5.

(34) Based on the information available in FIG. 4, i.e. based on the identified phase difference of the 4.sup.th-power of the incoming 16 QAM signal, the following cost function can be determined:
CF=[.sub.1+.sub.3]*(1/R.sub.2)+.sub.2*(1/R.sub.1+1/R.sub.3)

(35) wherein .sub.1, .sub.2, and .sub.3 are representing the standard deviation for each of the respective phase differences as shown in FIG. 4, and R1, R2 and R3 are representing the radii of the 16 QAM constellation as shown in FIG. 5.

(36) The cost function CF is the basis for estimating a variation of gamma, which is now explained in more detail:

(37) An optimized (in the purpose of improved) value of the nonlinear coefficient can be calculated by minimizing the cost function CF mentioned above. The algorithm for optimizing the nonlinear coefficient value (e.g., according to the steepest descent algorithm, which is a known optimization algorithm, http://en.wikipedia.org/wiki/Gradient_descent) is iteratively applied and, e.g., implemented in the adaptive Estimation Module 210 as follows:
(i+1)=(i)+(i)

(38) wherein: i is the index of the discrete time; (i+1) represents the value of the nonlinear coefficient at the iteration (i+1); (i) represents the value of the nonlinear coefficient at the preceding iteration (i); represents a convergence factor, comprising an effective fiber length L.sub.eff and a channel power P.

(39) The effective fiber length may be derived according to the following exemplary relation:

(40) $L_{\mathrm{eff}}=\frac{1-\exp \left(-L\right)}{}$

(41) wherein

(42) is a fiber attenuation defined in [Np/km].

(43) The algorithm can also be applied by considering only the algebraic sign of the gradient.

(44) Each iteration (i) can be derived according to the following equation:
(i)=CF()/

(45) wherein CF()/ is a gradient of the cost function over the nonlinear coefficient.

(46) By substituting (i) in the iterative optimization algorithm, the new value of the nonlinear coefficient can be determined according to:

(47) $\left(i+1\right)=\left(i\right)+\frac{\mathrm{CF}\left(\right)}{}$

(48) The new value of the nonlinear coefficient (i+1) is forwarded to the adaptive DBP module 202, wherein the received signal 120 is processed by applying the new value (i+1)step 253 in FIG. 2b.

(49) After each iteration, the gradient of the cost function CF is evaluated with respect to the previous iteration, whereas a change of the sign of the gradient indicates an end of the iteration loop, i.e. a minimum of the cost function has been reached. At this stage the iterative optimization algorithm can be stoppedstep 252 in FIG. 2b.

(50) FIG. 6 shows an example (calculated by simulated data) of the derived cost function CF as a function of the nonlinear coefficient based on a SSMF fiber with 16-QAM.

(51) The algorithm for determining the optimized value of the nonlinear coefficient starts by calculating two results of the cost function CF (corresponding to two different initial values of ).

(52) In addition, the convergence factor has to be optimized as well to achieve a reduction of computational time without losing quality in estimation accuracy.

(53) In a further example shown in FIG. 7 the respective cost function CF of an 882 km SSMF fiber was investigated based on experimental data, considering a launch power of 3 dBm.

(54) According to FIGS. 6 and 7 several important aspects of the proposed solution can be identified: The information being available after carrier recovery of the received signal is sufficient for determination of the optimized value of the nonlinear coefficient , i.e. FEC (forward error correction) based on a BER calculation can be avoided. Advantageously, the convergence factor for estimating the optimum value of the nonlinear coefficient can be significantly accelerated. The cost function CF can be derived analytically wherein verification of the results can be achieved by post-processing simulated and experimental data. The robustness of the proposed approach has been verified under extreme conditions, showing that an appropriate determination of the nonlinear coefficient is always successful.

(55) FIG. 8 and FIG. 9 show examples concerning the quality performance of the proposed method based on a Log 10(BER) versus power (dBm) performance, wherein Log 10(BER) is correlated with the quality of the received signal 120 after BER calculation.

(56) FIG. 8 is showing the Log 10(BER) versus power (dBm) performance for simulated data propagated over a 882 km LA-PSCF. The first curve (FDE) is showing the alignment of the signal-quality dependent from the power injected into the fiber by compensating only linear impairments using a Frequency Domain Equalizer (FDE). The second curve (0.6=.sub.BP) is showing the respective quality alignment by applying a Digital Back Propagation based on a fixed nonlinear coefficient =0.6 1/(W*km) which is assumed to be the correct value for the fiber. The third curve (1=.sub.BP) is showing the respective quality alignment by applying a Digital Back Propagation based on a wrong nonlinear coefficient value =1 1/(W*km). The forth curve (A_BP with initial =1) is showing the respective quality alignment by applying an adaptive Digital Back Propagation according to the proposed solution by starting with a (wrong) initial value of the nonlinear coefficient =1. As there is only a small difference between the alignment of the second and forth curve it can been verified, that the proposed method is working correctly, i.e. the derived optimized value of the nonlinear coefficient after termination of the optimization algorithm according to the proposed method is exactly the same or nearly the same value like the nonlinear coefficient value of the real fiber.

(57) FIG. 9 is showing the Log 10(BER) versus power (dBm) performance for experimental data propagated over 882 km of SSMF. Again, the first curve (FDE) is showing the alignment of the quality dependent from the power injected into the fiber by only compensating linear impairments using a Frequency Domain Equalizer (FDE). The second curve (0.6=.sub.BP) is showing the respective quality alignment by applying a Digital Back Propagation based on a fixed nonlinear coefficient =0.6 1/(W*km) which is the correct value according to information of the supplier of the fiber. The third curve (1.3=.sub.BP) is showing the respective quality alignment by applying a Digital Back Propagation based on a wrong nonlinear coefficient value =1.3 1/(W*km). The forth curve (A_BP with initial =1.3) is again showing the respective quality alignment by applying an adaptive Digital Back Propagation according to the proposed solution by starting with a (wrong) initial value of the nonlinear coefficient =1.3. Again only small differences can be identified between the second and forth curve, which means that the proposed adaptive Back Propagation algorithm is working correctly even by selecting a wrong initial value of the nonlinear coefficient .

(58) It should be noted, that the aforementioned cost function CF, determined exemplarily for processing a 16 QAM modulated signal, is one possible embodiment applying the proposed solution. The proposed solution can be applied for all kinds of modulation formats.

(59) The aforementioned cost function can be generalized as follows:

(60) ${\mathrm{CF}}_{\mathrm{gen}}=\underset{kK}{.Math.}\left[\left({}_{\mathrm{upper},k}+{}_{\mathrm{lower},k}\right).Math.\frac{1}{R_k}\right]+{}_{\mathrm{center}}.Math.\underset{iI}{.Math.}\frac{1}{R_i}$$\mathrm{with}kK;iI$

(61) wherein CF.sub.gen is a general cost function .sub.upper,k represents a standard deviation for each of the phase differences .sub.upper higher than a central phase .sub.center per radius R.sub.k .sub.lower,k represents a standard deviation for each of the phase differences .sub.lower lower than the central phase .sub.central central per radius R.sub.k .sub.center represents a central phase. I represents a set of distinct radii of the signal constellation K represents a set of distinct phase angles of the histogram of the signal constellation after a M-th power operation

(62) Adapting the general cost function CF.sub.gen for receiving of the 16 QAM modulated signal (as already being part of the exemplary description of the proposed solution), may result to the following cost function CF.sub.16:

(63) ${\mathrm{CF}}_{16}=\stackrel{K_{\max }}{\underset{kK}{.Math.}}\left({}_{\mathrm{upper}}+{}_{\mathrm{lower}}\right).Math.\frac{1}{R_2}+{}_{\mathrm{center}}.Math.\left(\frac{1}{R_1}+\frac{1}{R_3}\right)$$\mathrm{with}k\left[2\right];i\left[1,3\right]$

(64) FIG. 10 shows the corresponding 16 QAM constellation diagram applied to the 4.sup.th power.

(65) Hereinafter, examples are provided, adapting the general cost function CF.sub.gen for processing different modulation formats of the received signal.

(66) Adapting the general cost function CF.sub.gen for receiving of a 32 QAM modulated signal:

(67) ${\mathrm{CF}}_{32}=\underset{kK}{.Math.}\left[\left({}_{\mathrm{upper},k}+{}_{\mathrm{lower},k}\right).Math.\frac{1}{R_k}\right]+{}_{\mathrm{center}}.Math.\underset{iI}{.Math.}\frac{1}{R_i}$$\mathrm{with}k\left[2,4,5\right];i\left[1,3\right]$

(68) FIG. 11 shows the corresponding 32 QAM constellation diagram applied to the 4.sup.th power.

(69) Adapting the general cost function CF.sub.gen for receiving of a 64 QAM modulated signal:

(70) ${\mathrm{CF}}_{64}=\underset{kK}{.Math.}\left[\left({}_{\mathrm{upper},k}+{}_{\mathrm{lower},k}\right).Math.\frac{1}{R_k}\right]+{}_{\mathrm{center}}.Math.\underset{iI}{.Math.}\frac{1}{R_i}$$\mathrm{with}k\left[2,4,5,6,7,8\right];i\left[1,3,6,9\right]$

(71) FIG. 12 shows the corresponding 64 QAM constellation diagram to the 4.sup.th power.

(72) For receipt of a M-PSK modulated signal, the following cost function can be determined:

(73) ${\mathrm{CF}}_{M\text{-}\mathrm{PSK}}=\frac{}{R}$$\mathrm{with}k=0;i=1$

(74) FIG. 13 shows the corresponding M-PSK constellation diagram to the 4.sup.th power.

(75) In real DWDM systems details on the transmission link are not or only partially available. Even, if details were available, with the upcoming automatically switched optical networks (ASON/GMPLS) an exact knowledge of the link description would not be available any more, particularly after protection switching or even active traffic routing. The proposed solution for adaptive Digital Back Propagation is capable of a suitable set of parameters for a DBP implementation after a very short initialization cycle.

(76) As a further advantage, no significant changes in the optical receiver are necessary for implementing the proposed solution. The coding of the optimization algorithm can be implemented in a DSP (Digital Signal Processor).

(77) The proposed approach can be implemented in various optical transmission systems using coherent detection, including single carrier and multi carrier, single mode and multi mode.

LIST OF ABBREVIATIONS

(78) DBP Digital Back-Propagation DEC Decision DSP Digital Signal Processor DWDM Dense Wavelength Division Multiplex BER Bit Error Rate CD Chromatic Dispersion CF Cost Function CPR Carrier Phase Recovery CR Clock Recovery DSP Digital Signal Processing DM Dispersion Managed FDE Frequency-Domain Equalizer NDM Non-Dispersion Managed NLPN Nonlinear Phase Noise PMD Polarization Mode Dispersion RX Receive SPM Self Phase Modulation TDE Time-Domain Equalizer TS Training Sequence XPM Cross Phase Modulation