CN118764097A - WDM system nonlinear compensation method and device requiring only target channel information - Google Patents

Abstract

The present invention provides a WDM system nonlinear compensation method and device that only requires target channel information, belonging to the field of optical fiber communication technology. The method integrates a single-channel DBP and a JNL neural network that combines NPCC and LMS functions, adaptively updates weights, and constructs a hidden layer with a known working principle. On the basis of single-channel DBP compensating for the SPM effect, the XPM effect is compensated based on the target channel information. The present invention only compensates for the signal distortion caused by the nonlinear effect based on the target channel information, does not require the assistance of other WDM channels, and can effectively equalize the signal nonlinear distortion while controlling the calculation complexity and reducing the implementation cost, thereby improving the equalization performance.

Description

WDM system nonlinear compensation method and device requiring only target channel information

Technical Field

The present invention belongs to the technical field of optical fiber communication, and in particular relates to a WDM system nonlinear compensation method and device which only requires target channel information.

Background Art

In recent decades, due to the continuous expansion of current emerging digital applications and services such as big data, cloud computing, and 5G communication technology, the capacity demand of communication networks has shown a steady growth trend. Optical fiber communication networks are facing unprecedented challenges and need to further expand transmission capacity and transmission distance. It is well known that the distortion caused by linear impairments including chromatic dispersion (CD) and polarization mode dispersion (PMD) can be equalized by advanced digital signal processing (DSP) techniques to achieve satisfactory performance improvement. However, fiber Kerr nonlinearity has long been a bottleneck for long-distance optical communications. Although the propagation dynamics of signals in nonlinear fibers are well known and governed by the nonlinear Schrödinger equation (NLSE), the interaction between fiber nonlinearity-induced self-phase modulation (SPM), cross-phase modulation (XPM), and four-wave mixing (FWM) in wavelength division multiplexing (WDM) systems has proven to be difficult to statistically characterize and compensate. So far, in most cases, a single target channel cannot directly obtain information about other adjacent channels. Therefore, in the absence of other channel information, effectively mitigating or compensating for inter-channel nonlinear impairments based only on the target channel information is crucial to improving the performance of high-speed, long-distance, and high-capacity fiber-optic communication systems.

In order to alleviate the distortion caused by the nonlinear effects of optical fibers, researchers have proposed a variety of effective nonlinear compensation techniques. The digital back propagation (DBP) algorithm and its improved method solve the fiber back propagation equation by the split-step Fourier method (SSFM) to achieve alternating compensation for dispersion and nonlinearity. However, the iteration of DBP requires multiple Fourier transform pairs, and the performance improves with the increase of the number of steps per span, which means that superior performance requires higher computational complexity. In addition, the algorithm is a method of compensating nonlinear distortion in theory, which requires transparent optical fiber link parameters, and faces huge challenges in direct application in practice. In addition, the optical phase conjugation (OPC) technology for nonlinear compensation in the optical domain, the nonlinear equalization method based on Volterra series, and the nonlinear compensation algorithm based on perturbation theory have also been proven to be effective. However, OPC is very expensive in practical applications and has very low conversion efficiency, which limits its performance. Since the Volterra series requires a Fourier transform module, it faces the dilemma of increasing complexity as the dispersion accumulates. Nonlinear compensation based on perturbation theory requires a higher computational complexity to achieve the desired quantization accuracy.

With the rapid development of machine learning in recent years, the powerful learning ability of neural networks (NN) has attracted widespread attention. It does not require too much prior link information of the system to complete the operation. Therefore, artificial neural networks (ANN) and convolutional neural networks (CNN) have been introduced into the field of fiber nonlinear compensation to further improve system performance. The correlation between adjacent symbols of triples has made memory neural networks a research hotspot. Taking the long short-term memory (LSTM) network and its variants as an example, they effectively realize the nonlinear compensation of coherent optical communication systems by memorizing the correlation between adjacent symbols. However, in order to well compensate for the XPM damage caused by other channels in the WDM system to the target channel, in addition to the information of the target channel, the information of other channels is also needed, which brings great inconvenience and high computational complexity to the nonlinear compensation.

From the patent search, it can be seen that the schemes for nonlinear compensation in optical communication systems mainly include: In existing research, the signal is first compensated for dispersion and nonlinearity, and then the compensated signal is regressed to determine the result. In existing research, a triplet is used to make the neural network learn the nonlinear damage value, and then the nonlinear damage value is subtracted from the received signal to complete the nonlinear compensation. In existing research, an optical phase conjugation algorithm is used to compensate for nonlinear damage. The above-mentioned research has high computational complexity in nonlinear compensation of optical communication systems and requires clear information about other channels, and cannot independently process different channels in the WDM system.

Summary of the invention

In view of the above-mentioned deficiencies in the prior art, the present invention provides a WDM system nonlinear compensation method and device that only requires target channel information, which can achieve performance improvement with lower computational complexity based only on the target channel information, and the present invention can greatly extend the effective transmission distance.

In order to achieve the above object, the technical solution adopted by the present invention is: a WDM system nonlinear compensation method that only requires target channel information, comprising the following steps:

S1. Using a coherent receiver to separately receive the discrete signal of each channel in the WDM system, and resample the discrete signal to 2 samples/symbol;

S2, using the resampled signal from a certain target channel to compensate for the signal damage caused by dispersion and SPM using single-channel DBP;

S3, sequentially performing polarization multiplexing processing, downsampling to 1 symbol/sample processing, frequency offset estimation processing, and carrier phase recovery processing on the signal after single-channel DBP compensation;

S4, input the signal after carrier phase recovery into the JNL neural network, and combine the LMS algorithm and the NPCC algorithm to compensate for the signal damage caused by XPM, wherein in the JNL neural network, the nonlinear phase noise NPN, a component of XPM, is compensated by the iterative process of the linear digital filter of the LMS algorithm; the nonlinear polarization crosstalk NPC, a component of XPM, is compensated by the NPCC layer based on the NPCC principle;

S5. Calculate the bit error rate of the damaged signal after JNL neural network compensation to complete the joint compensation of nonlinear damage within and between channels of the WDM system.

The beneficial effect of the present invention is that the present invention integrates the single-channel DBP and the neural network of the joint NPCC and LMS algorithms. The key to this solution is the JNL neural network of the joint NPCC and LMS. The JNL neural network can adaptively update the weights, construct a hidden layer with a known working principle, and complete the compensation of the XPM effect based on the target channel information on the basis of the single-channel DBP compensation of the SPM effect. The present invention compensates for the signal distortion caused by the nonlinear effect based on the target channel information, does not require the assistance of other channels, and can effectively equalize the nonlinear distortion of the signal while controlling the calculation complexity and reducing the implementation cost, thereby improving the equalization performance.

Furthermore, the step S2 is specifically as follows:

The resampled signal from a certain target channel is compensated for the signal damage caused by chromatic dispersion and SPM alternately by the dispersion compensation layer and the nonlinear compensation layer in the single-channel DBP.

Furthermore, the expression of the JNL neural network is as follows:

in, and Both represent samples output by the JNL neural network after learning.

and They represent the samples at time k output after the JNL neural network is learned, H _x and _Hy represent the time-varying ISI matrices, X _kl and Y _kl represent l different adjacent samples centered on the sample at time k on the x polarization and y polarization of the input JNL neural network, respectively, m represents the number of adjacent samples, W _yx represents the crosstalk factor from x polarization to y polarization, and W _xy represents the crosstalk factor from y polarization to x polarization.

The beneficial effect of the above further scheme is that the JNL neural network of the present invention only needs the target channel information on the basis of the SPM effect in the single-channel DBP compensation channel to accurately compensate for the nonlinear damage caused by the XPM effect, and the computational complexity is low.

Furthermore, the JNL neural network includes:

The input layer is used to receive the signal after the carrier phase is recovered with XPM damage;

Hidden layer, used to combine the LMS algorithm and the NPCC algorithm to compensate for the signal damage caused by XPM;

The output layer is used to output the damaged signal after compensation by the adaptive filter.

Further, the hidden layer includes an LMS layer and an NPCC layer;

The LMS layer is used to compensate for XPM damage through an iterative process of a linear digital filter of an LMS algorithm;

The NPCC layer is used to compensate for the NPC component of the XPM.

Furthermore, the step S5 is specifically as follows:

The bit error rate of the damaged signal after JNL neural network compensation is calculated to achieve offline processing of the target channel signal and complete the joint compensation of nonlinear damage within and between channels of the WDM system.

The beneficial effect of the above further solution is that the present invention can more clearly evaluate the recovery of the distorted signal through the bit error rate BER processing of the digital signal processing module.

The present invention provides a WDM system nonlinear compensation device that only requires target channel information, including:

A first processing module is used to receive the signal of each channel in the WDM system separately by using a coherent receiver, and resample the received signal to 2 samples/symbol;

The second processing module is used to compensate the signal damage caused by dispersion and SPM using the single-channel DBP for the resampled signal from a certain target channel;

The third processing module is used to sequentially perform polarization multiplexing processing, down-sampling to 1 symbol/sample processing, frequency offset estimation processing and carrier phase recovery processing on the signal after single-channel DBP compensation;

The fourth processing module is used to input the signal after carrier phase recovery into the JNL neural network, and combine the LMS algorithm and the NPCC algorithm to compensate the signal damage caused by the XPM, wherein in the JNL neural network, the NPN component of the XPM is compensated by the iterative process of the linear digital filter of the LMS algorithm; the NPC component of the XPM is compensated by the NPCC layer based on the NPCC principle;

The fifth processing module is used to calculate the bit error rate of the damaged signal after JNL neural network compensation, and complete the joint compensation of nonlinear damage within and between channels of the WDM system.

The present invention provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement any step of the WDM system nonlinear compensation method requiring only target channel information.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a flow chart of the method of the present invention.

FIG. 2 is a block diagram of an 11-channel WDM simulation system and a complete off-line DSP flow chart of the present invention.

FIG3 is a schematic diagram of the JNL neural network architecture in the present invention.

FIG. 4 is a Q factor performance curve diagram of DBP and single-channel DBP+JNL in the 11-channel WDM simulation system when 36GBaud PDM-16QAM is transmitted over 1600km in this embodiment.

FIG. 5 is a Q factor performance curve diagram of DBP and single-channel DBP+JNL in an 11-channel WDM simulation system with 64GBaud PDM-64QAM transmission over 400km in this embodiment.

FIG6 is a schematic diagram showing the relationship between the Q factor performance and the transmission distance of DBP and single-channel DBP+JNL in the PDM-16QAM modulation format of the 11-channel WDM simulation system of this embodiment.

FIG7 is a schematic diagram showing the relationship between the Q factor performance and the transmission distance of DBP and single-channel DBP+JNL in the 11-channel WDM simulation system PDM-64QAM modulation format according to this embodiment.

FIG8 is a Q factor performance curve diagram of DBP+NPCC and DBP+JNL in the 11-channel WDM simulation system when 36GBaud PDM-16QAM is transmitted over 1600km in this embodiment.

FIG. 9 is a Q factor performance curve diagram of DBP+NPCC and DBP+JNL in the embodiment when 64GBaud PDM-64QAM is transmitted over 400km in an 11-channel WDM simulation system.

FIG10 is a schematic diagram showing the relationship between the Q factor performance and the transmission distance of DBP+NPCC and DBP+JNL in the PDM-16QAM modulation format of the 11-channel WDM simulation system of this embodiment.

FIG11 is a schematic diagram showing the relationship between the Q factor performance and the transmission distance of DBP+NPCC and DBP+JNL in the 11-channel WDM simulation system PDM-64QAM modulation format according to this embodiment.

FIG. 12 is a schematic diagram of the device structure of the present invention.

DETAILED DESCRIPTION

The specific implementation modes of the present invention are described below to facilitate those skilled in the art to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

Example 1

As shown in FIG1 , the present invention provides a WDM system nonlinear compensation method that only requires target channel information, and the implementation method thereof is as follows:

S1. A coherent receiver is used to separately receive the discrete signal of each channel in the WDM system, and the discrete signal is resampled to 2 samples/symbol. The implementation method is as follows:

S101, in the WDW system, using a coherent receiver to individually receive discrete signals of each channel;

S102 , resample the discrete signal to 2 samples/symbol.

S2. The resampled signal from a target channel is compensated for the signal damage caused by dispersion and SPM using single-channel DBP. Specifically:

The resampled signal from a certain target channel is compensated for the signal damage caused by chromatic dispersion and SPM alternately by the dispersion compensation layer and the nonlinear compensation layer in the single-channel DBP.

S3, sequentially performing polarization multiplexing processing, downsampling to 1 symbol/sample processing, frequency offset estimation processing, and carrier phase recovery processing on the signal after single-channel DBP compensation;

In this embodiment, the signal after DBP compensation of a single channel is downsampled to 1 symbol/sample after passing through a polarization demultiplexing algorithm, and then processed through a frequency offset estimation algorithm and a carrier phase recovery algorithm.

S4. Input the signal after carrier phase recovery into the JNL neural network, and combine the LMS algorithm and the NPCC algorithm to compensate for the signal damage caused by XPM. In the JNL neural network, the nonlinear phase noise NPN, a component of XPM, is compensated by the iterative process of the linear digital filter of the LMS algorithm; the nonlinear polarization crosstalk NPC, a component of XPM, is compensated by the NPCC layer based on the NPCC principle.

In this embodiment, the signal after carrier phase recovery enters the neural network (Joint NPCC and LMS, JNL) of nonlinear polarization crosstalk canceller (NPCC) and least mean square (LMS) to compensate for the signal damage caused by XPM. In the JNL neural network, the nonlinear phase noise (NPN) of XPM can be compensated by the iterative process of the linear digital filter of the LMS algorithm; the nonlinear polarization crosstalk (NPC) of XPM is compensated by the NPCC layer based on the NPCC principle.

In this embodiment, the signal after the carrier phase recovery algorithm compensation enters the JNL neural network to compensate for the damage caused by XPM. The JNL neural network can adaptively update the weights, construct a hidden layer with a known working principle, and complete the compensation of the cross-phase modulation XPM effect based on the target channel information on the basis of the single-channel DBP compensation of the fiber nonlinear induced self-phase modulation SPM effect. The present invention compensates for the signal distortion caused by the nonlinear effect based on the target channel information, does not require the assistance of other channels, and can effectively equalize the signal nonlinear distortion while controlling the calculation complexity and reducing the implementation cost, thereby improving the equalization performance.

JNL neural network includes:

An input layer, used to receive an input signal with XPM impairment;

The hidden layer is used to compensate for the signal damage caused by XPM, and is composed of an LMS layer and an NPCC layer. In the JNL neural network, the NPN component of XPM can be compensated by the iterative process of the linear digital filter of the LMS algorithm; the NPC component of XPM is compensated by the NPCC layer based on the NPCC principle. The LMS layer is used to compensate for the damage with XPM through the iterative process of the linear digital filter of the LMS algorithm; the NPCC layer is used to compensate the NPC component of XPM;

The output layer is used to output the signal compensated by the adaptive filter.

In this embodiment, for the NPN caused by XPM, according to the first-order perturbation theory analysis, the nonlinear interaction of XPM can be modeled as time-varying inter-symbol interference (ISI), so the kth sample of the signal received in the target channel is expressed as:

Where _Rk represents the received signal, _Tk represents the transmitted signal, i represents the imaginary part, m,l represent the number of adjacent samples before and after the XPM effect of the current sample, and the number used to traverse m, respectively, and _Tkl represents l different adjacent signals centered on the signal at time k at the transmitter. _Nk represents a Gaussian process, and [-m,m] represents the range of adjacent samples related to the XPM effect of the current sample, which is 2m+1 in total. represents the time-varying ISI matrix, whose values are determined by the transmitted signal on the interference channel. The above formula shows that NPN is modeled as at least partially correlated time-varying ISI, so according to the characteristic that it can be represented by a finite memory term, NPN can be compensated by the iterative process of the linear digital filter of the LMS algorithm.

In this embodiment, according to the principle of NPCC, if NPN is not considered and only the NPC caused by XPM is considered, since |W _xy ² and |W _yx ² are much less than 1, the time-varying XPM matrix can be written as follows:

In the above formula, x and y represent x and y polarization signals respectively, _Rx represents the received signal corresponding to x polarization, _Ry represents the received signal corresponding to y polarization, _Tx represents the ideal signal corresponding to x polarization, _Ty represents the ideal signal corresponding to y polarization, _Wyx represents the crosstalk factor from x polarization to y polarization, and _Wxy represents the crosstalk factor from y polarization to x polarization. In the formula, _Tx = _Rx - _WxyTy , _{Ty = Ry} - _WyxTx , _{and NPC} can be compensated by subtracting the estimated nonlinear crosstalk _WxyxTy , _WyxTx . However, the accurate ideal signal of the receiving end cannot be obtained at the receiving end, so in the calculation of weights, that is, _Wxy = ( _Rx - _Tx )/ _Ty and _Wyx = ( _{Ry - Ty} )/ _Tx , Tx _/ y is the result of direct judgment of Rx _/ y. Taking into account that the weight parameters will be optimized through the error cycle, the initial weight value can be sent to the neural network for adaptive optimization by the optimizer, which can avoid direct judgment of the output value in each cycle, improve the compensation accuracy and reduce the computational complexity.

In this embodiment, the nonlinear distortion caused by XPM has correlation between adjacent samples. Therefore, considering these two aspects, the LMS algorithm and the NPCC algorithm can be combined to obtain a neural network compensation model that can improve the efficiency of XPM mitigation:

Wherein, R _x and R _y both represent the received symbol sequences corresponding to the two polarizations, x and y both represent polarization signals, W _yx represents the crosstalk factor from x polarization to y polarization, W _xy represents the crosstalk factor from y polarization to x polarization, T _x and _Ty both represent the ideal signals sent, X and Y represent the input signal samples, and All represent the signal samples output after JNL neural network learning.

and They represent the samples at time k output by the JNL neural network after learning. _Hx and _Hy both represent the time-varying ISI matrix, which includes the influence of adjacent samples. _Wyx represents the crosstalk factor from x polarization to y polarization, _Wxy represents the crosstalk factor from y polarization to x polarization, k represents the current sample, m and l are the adjacent sample number representation methods, kl represents different adjacent samples, [-m,m] is the adjacent sample range related to the XPM effect of the current sample, which is 2m+1 in total, and also represents the number of taps of LMS and NPCC, that is, the number of weights that the neural network needs to optimize in each algorithm.

S5. Calculate the bit error rate of the damaged signal after JNL neural network compensation to complete the joint compensation of nonlinear damage within and between channels of the WDM system, which is specifically as follows:

The bit error rate of the damaged signal after JNL neural network compensation is calculated to achieve offline processing of the target channel signal and complete the joint compensation of nonlinear damage within and between channels of the WDM system.

In this embodiment, steps S2-S3 and step S5 complete signal compensation in an offline DSP algorithm, as shown in FIG2 ; and step S4 is completed in a JNL neural network, as shown in FIG3 .

In this embodiment, as shown in Figure 2, the JNL neural network is located after the carrier phase recovery algorithm. The order in which the received signal passes through the processing algorithm in the DSP in the steps is: resampling, dispersion compensation (Chromatic dispersion compensation, CDC) or DBP, polarization demultiplexing, downsampling, frequency offset estimation algorithm, carrier phase recovery algorithm, JNL neural network, and finally the bit error rate calculation is performed on the output of the JNL neural network.

This embodiment builds an 11-channel WDM system of 36GBaud PDM-16QAM 1600km and 64GBaud PDM-64QAM 400km based on VPIDesign Suite 11.1 and Matlab joint simulation, as shown in Figure 5. In order to prevent the neural network from learning the data generation law, the random function of Matlab is used to randomly generate the bit sequence, and the symbol length of each polarization is 65536. The frequency offset and laser linewidth are set to 100MHz and 100KHz respectively. Each transmission span of the system consists of 80km of standard single mode fiber (SSMF) and an erbium-doped fiber amplifier (EDFA), and the waveform of the modulated signal is formed by a root raised cosine (RRC) filter with a roll-off factor of 0.1. The wavelength of the central channel of the WDM system is 1550nm, and the channel spacing of PDM-16QAM and PDM-64QAM is 50GHz and 75GHz respectively. The loss, dispersion coefficient, nonlinear coefficient, and polarization mode dispersion coefficient of the optical fiber are set to 0.2 dB/km, 16 ps/nm/km, 1.3 W ^-1 /km, The dispersion slope is set to 0.08ps/nm ² /km. The EDFA works in gain control mode, and the noise figures are 5dB (PDM-16QAM) and 4dB (PDM-64QAM) respectively. At the receiving end, this embodiment uses a coherent receiver to receive signals, and then performs offline processing.

In this embodiment, for the received discrete data, whether it is PDM-16QAM or PDM-64QAM, the data length of each channel is 131072 samples, the first 50% of 65536 samples are used for training, and the remaining 65536 samples are used for testing. In this embodiment, the number of hidden layers of the neural network used for training is 1.

In this embodiment, in order to train the JNL neural network, the Adam optimizer is selected, which has a relatively fast convergence speed, is simple to implement, and is suitable for large-scale data and parameter scenarios. The learning rate of the optimizer is set to 0.001 to ensure the stability of the neural network training performance. The mean squared error (MSE) is selected as the loss function to measure the closeness between the data obtained by the neural network training and the label.

In this embodiment, in order to better compensate for the nonlinear damage between channels, before the JNL neural network is trained, the present invention optimizes the number of filter taps of the LMS algorithm and the NPCC, and the number of taps of the two algorithms in the neural network compensation model is consistent, both of which are 2m+1. For PDM-16QAM and PDM-64QAM, the present invention selects the number of taps equal to 17 and 15 as the optimal parameters after optimization, that is, m is 8 and 7 respectively.

In this embodiment, FIG4 shows the Q factor performance curve of DBP and single-channel DBP+JNL when the WDM simulation system 36GBaud PDM-16QAM is transmitted for 1600km. As shown in FIG4, for the PDM-16QAM signal, the optimal transmission power of the present invention is 0dBm, which is 1dBm higher than the linear compensation scheme. Compared with the linear compensation scheme, the signal-to-noise ratio at the optimal transmission power of the present invention can be improved by about 3.2dB, and the Q factor is improved by about 1.12dB. For the PDM-16QAM signal, the performance of the present invention is higher than that of DBP-5StPS with 5 steps per span, and the Q factor is improved by about 0.54dB at the optimal transmission power. Compared with DBP-2StPS with the same number of steps, the Q factor is improved by about 0.87dB.

In this embodiment, Figure 5 shows the Q factor performance curves of DBP and single-channel DBP+JNL when the WDM simulation system 64GBaud PDM-64QAM is transmitted for 400km. For the 64GBaud PDM-64QAM transmission scenario, the optimal transmission power of the present invention is 2dBm, which is 1dBm higher than the linear compensation scheme. Compared with the linear compensation scheme, the signal-to-noise ratio at the optimal transmission power of the present invention can be improved by about 4.1dB, and the Q factor is improved by about 1.93dB. The performance of the PDM-64QAM signal under the present invention is higher than that of DBP-10StPS and the Q factor is improved by about 1.66dB at the optimal transmission power. Compared with DBP-5StPS with the same number of steps, the Q factor is improved by about 1.84dB.

In this embodiment, based on the above analysis of Figures 4 and 5, the comparison results with DBP reflect the effective compensation effect of the present invention on nonlinear damage between channels, and the comparison results with the linear compensation scheme can illustrate the effectiveness of the present invention in jointly compensating for nonlinear damage within and between channels.

In this embodiment, FIG6 shows a schematic diagram of the relationship between the Q factor performance and transmission distance of DBP and single-channel DBP+JNL under the PDM-16QAM modulation format of the WDM simulation system. For PDM-16QAM, the linear compensation scheme can transmit about 1680km under the condition of meeting the 7% FEC threshold line, DBP-2StPS can transmit about 1850km, DBP-5StPS can transmit about 2080km, and DBP+JNL, which performs joint compensation for intra-channel SPM and inter-channel XPM, can transmit about 2400km with an accuracy of 2 steps per span. Compared with the linear compensation scheme, the proposed scheme extends the transmission distance by 720km, and extends the transmission distance by 550km compared to DBP with the same number of steps. In addition, compared with DBP-5StPS with more steps, DBP(2StPS)+JNL can transmit about 320km more.

In this embodiment, FIG7 shows a schematic diagram of the relationship between the Q factor performance and transmission distance of DBP and single-channel DBP+JNL under the PDM-16QAM modulation format of the WDM simulation system. For PDM-64QAM, the linear compensation scheme can transmit about 240km under the condition of meeting the 7% FEC threshold line, DBP-5StPS can transmit about 300km, DBP-10StPS can transmit about 340km, and DBP(5StPS)+JNL with the same number of steps as DBP-5StPS can transmit about 560km. Compared with the linear compensation scheme, the proposed scheme can transmit 320km more, compared with the DBP-5StPS scheme, it can transmit 260km more, and compared with DBP-10StPS, DBP(5StPS)+JNL can extend the transmission distance by about 220km.

In this embodiment, based on the above analysis of FIG. 6 and FIG. 7 , for signals of two modulation formats, the proposed solution can extend the transmission distance of the system under the condition of meeting the 7% FEC threshold line.

In this embodiment, Figure 8 shows the Q factor performance curves of DBP+NPCC and DBP+JNL when the WDM simulation system 36GBaud PDM-16QAM is transmitted for 1600km. We set the frequency deviation and linewidth of the transmission system to 0, without the influence of carrier phase noise. As can be seen from the figure, for PDM-16QAM signals, the optimal transmission power of DBP(2StPS)+JNL is 1dBm, which is 2dBm higher than the linear compensation scheme. Compared with the linear compensation scheme, the signal-to-noise ratio of the present invention can be improved by about 3.5dB at the optimal transmission power, and the Q factor is improved by about 1.27dB. For PDM-16QAM signals, the performance of the proposed scheme is higher than that of DBP-2StPS with the same number of steps, and the Q factor is improved by about 0.7dB at the optimal transmission power. Compared with DBP(2StPS)+NPCC, the Q factor is improved by about 0.34dB.

In this embodiment, FIG9 shows the Q factor performance curves of DBP+NPCC and DBP+JNL when the WDM simulation system transmits 64GBaud PDM-64QAM for 400km. For the 64GBaud PDM-64QAM transmission scenario, the optimal transmission power of DBP(5StPS)+JNL is 2dBm, which is 1dBm higher than that of the linear compensation scheme. Compared with the linear compensation scheme, the signal-to-noise ratio of the present invention can be improved by about 2.8dB at the optimal transmission power, and the Q factor is improved by about 0.76dB. The performance of the proposed scheme for PDM-64QAM signals is higher than that of DBP-5StPS, and the Q factor is improved by about 0.53dB at the optimal transmission power. Compared with DBP(5StPS)+NPCC, the Q factor is improved by about 0.33dB.

In this embodiment, based on the above analysis of FIG. 8 and FIG. 9 , the comparison results with the traditional cascaded solution DBP (5StPS) + NPCC demonstrate the advantage of the proposed solution in applying JNL to jointly compensate for intra-channel and inter-channel nonlinear impairments.

In this embodiment, FIG10 shows a schematic diagram of the relationship between the Q factor performance and transmission distance of DBP+NPCC and DBP+JNL under the PDM-16QAM modulation format of the WDM simulation system. As shown in the figure, under the condition of meeting the 7% FEC threshold line, for PDM-16QAM, the linear compensation scheme can transmit about 1840km, DBP-2StPS can transmit about 2080km, and DBP+NPCC and DBP+JNL can transmit about 2240km and 2350km respectively at an accuracy of 2 steps per span. Therefore, compared with the linear compensation scheme, the transmission distance of the proposed scheme is extended by 510km, compared with the DBP-2StPS transmission distance is extended by 270km, and compared with DBP+NPCC, the transmission distance is extended by 110km.

In this embodiment, FIG11 shows a schematic diagram of the relationship between the Q factor performance and the transmission distance of DBP+NPCC and DBP+JNL under the PDM-64QAM modulation format of the WDM simulation system. For the 64GBaud PDM-64QAM transmission scenario, the transmission distance of DBP(5StPS)+JNL increases by approximately 120km, 80km, and 50km, respectively, compared with the linear compensation scheme, DBP-5StPS, and DBP+NPCC with 5 steps.

In this embodiment, the total number of real number multiplications required for the neural network training process is used as a measure of complexity for complexity analysis. For 16/64QAM signals, compared with the traditional algorithm of DBP and NPCC cascade on the basis of the same step number DBP, the proposed scheme significantly reduces the complexity while improving the performance. For PDM-16QAM signals, DBP(2StPS)+NN is only 41% of DBP-5StPS under the premise of improving the performance by 0.54dB. For PDM-64QAM signals, the proposed scheme improves the performance by 1.66dB compared with DBP-10StPS at an accuracy of 5 steps per step, and the complexity is only 51.78% of it.

Example 2

As shown in FIG. 12 , the present invention provides a WDM system nonlinear compensation device that only requires target channel information to perform any one of Embodiment 1, including:

A first processing module is used to receive the signal of each channel in the WDM system separately by using a coherent receiver, and resample the received signal to 2 samples/symbol;

The second processing module is used to compensate the signal damage caused by dispersion and SPM using the single-channel DBP for the resampled signal from a certain target channel;

The third processing module is used to sequentially perform polarization multiplexing processing, down-sampling to 1 symbol/sample processing, frequency offset estimation processing and carrier phase recovery processing on the signal after single-channel DBP compensation;

The fourth processing module is used to input the signal after carrier phase recovery into the JNL neural network, and combine the LMS algorithm and the NPCC algorithm to compensate the signal damage caused by the XPM, wherein in the JNL neural network, the NPN component of the XPM is compensated by the iterative process of the linear digital filter of the LMS algorithm; the NPC component of the XPM is compensated by the NPCC layer based on the NPCC principle;

The fifth processing module is used to calculate the bit error rate of the damaged signal after JNL neural network compensation, and complete the joint compensation of nonlinear damage within and between channels of the WDM system.

In this embodiment, the main structure of the JNL neural network is selected to train the neural network with a relatively fast convergence speed, simple implementation, and suitable for large-scale data and parameter scenarios. The learning rate of the optimizer is set to 0.001 to ensure the stability of the neural network training performance. The mean square error (MSE) is selected as the loss function to measure the closeness between the data obtained by the neural network training and the label.

The WDM system nonlinear compensation device that only requires target channel information provided by the embodiment shown in Figure 12 can execute the technical solution shown in the WDM system nonlinear compensation method that only requires target channel information in the above method embodiment. Its implementation principle and beneficial effects are similar and will not be repeated here.

In this embodiment, the present application can divide the functional units according to the WDM system nonlinear compensation method that only requires target channel information. For example, each function can be divided into each functional unit, or two or more functions can be integrated into one processing unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional units. It should be noted that the division of units in the present invention is schematic and is only a logical division. There may be other division methods in actual implementation.

In this embodiment, the WDM system intra-channel and inter-channel nonlinear impairment joint compensation system based on the target channel includes hardware structures and/or software modules for executing various functions in order to realize the principles and beneficial effects of the method. Those skilled in the art should easily realize that, in combination with the schematic units and algorithm steps described in the embodiments disclosed in the present invention, the present invention can be implemented in the form of hardware and/or a combination of hardware and computer software. Whether a function is executed in a hardware or computer software driven manner depends on the specific application and design constraints of the technical solution. Different methods can be used for each specific application to implement the described function, but such implementation should not be considered to exceed the scope of this application.

In this embodiment, the present invention makes full use of the modeling principle of XPM, and can effectively equalize the nonlinear distortion of the signal while controlling the calculation complexity and reducing the implementation cost, thereby improving the equalization performance.

Example 3

The present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of a WDM system nonlinear compensation method requiring only target channel information as described in any one of Embodiment 1.

In this embodiment, the electronic device may include: a processor, a memory, a bus and a communication interface. The processor, the communication interface and the memory are connected via a bus. The memory stores a computer program that can be run on the processor. When the processor runs the computer program, it executes part or all of the steps of the WDM system nonlinear compensation method that only requires target channel information provided in the aforementioned embodiment 1 of the present application.

Claims (8)

1. A method for compensating for nonlinearity of a WDM system requiring only target channel information, comprising the steps of:

s1, independently receiving discrete signals of each channel in a WDM system by using a coherent receiver, and resampling the discrete signals to 2 samples/symbol;

S2, utilizing a single channel DBP to compensate signal damage caused by chromatic dispersion and SPM to resample signals from a certain target channel;

S3, carrying out polarization multiplexing processing, downsampling to 1 symbol/sample processing, frequency offset estimation processing and carrier phase recovery processing on the signal after single-channel DBP compensation in sequence;

S4, inputting the signal with the recovered carrier phase into a JNL neural network, and combining an LMS algorithm and an NPCC algorithm to compensate signal damage caused by XPM, wherein in the JNL neural network, nonlinear phase noise NPN of a component part of the XPM is compensated through an iterative process of a linear digital filter of the LMS algorithm; NPCC layer based on NPCC principle compensates nonlinear polarization crosstalk NPC of XPM component part;

S5, calculating bit error rate of the damaged signal compensated by the JNL neural network to finish nonlinear damage joint compensation in and among channels of the WDM system.

2. The method for compensating the nonlinearity of the WDM system requiring only the target channel information according to claim 1, wherein said step S2 is specifically:

The resampled signal from a certain target channel is alternately compensated for chromatic dispersion and signal impairments by SPM by a dispersion compensation layer and a nonlinear compensation layer in a single channel DBP.

3. The method for compensating for nonlinearity of a WDM system requiring only target channel information according to claim 1, wherein the JNL neural network is expressed as follows:

Wherein, AndAll represent samples output after JNL neural network learning,AndRespectively representing samples of k moments outputted after the JNL neural network is learned, wherein both H _x and H _y represent time-varying ISI matrices, X _k-l and Y _k-l are respectively on X-polarization and Y-polarization representing the input JNL neural network, l different adjacent samples centered on the samples of k moments, m represents the number of adjacent samples, W _yx represents the crosstalk factor from X-polarization to Y-polarization, and W _xy represents the crosstalk factor from Y-polarization to X-polarization.

4. The method for compensating for nonlinearity of a WDM system requiring only target channel information according to claim 1, wherein the JNL neural network comprises:

An input layer for receiving the carrier phase recovered signal with XPM impairments;

The hidden layer is used for combining an LMS algorithm and an NPCC algorithm and compensating signal damage caused by XPM;

And the output layer is used for outputting the damaged signal compensated by the adaptive filter.

5. The method of compensating for nonlinearity of a WDM system in which only target channel information is required as claimed in claim 4, wherein the hidden layer comprises an LMS layer and an NPCC layer;

The LMS layer is used for compensating XPM damage through the iterative process of a linear digital filter of an LMS algorithm;

The NPCC layer is used for compensating NPC which is a component part of XPM.

6. The method for compensating the nonlinearity of the WDM system requiring only the target channel information according to claim 1, wherein said step S5 is specifically:

And (3) calculating bit error rate of the damaged signal compensated by the JNL neural network, and realizing off-line processing of the target channel signal to finish nonlinear damage joint compensation in and among channels of the WDM system.

7. A WDM system nonlinearity compensation device that performs only target channel information as recited in any one of claims 1-6, comprising:

a first processing module for individually receiving a signal of each channel in the WDM system using the coherent receiver and resampling the received signal to 2 samples/symbol;

The second processing module is used for utilizing a single channel DBP to compensate signal damage caused by chromatic dispersion and SPM to resample a signal from a certain target channel;

the third processing module is used for sequentially carrying out polarization multiplexing processing, downsampling to 1 symbol/sample processing, frequency offset estimation processing and carrier phase recovery processing on the signal after single-channel DBP compensation;

The fourth processing module is used for inputting the signal after carrier phase recovery into the JNL neural network, and combining an LMS algorithm and an NPCC algorithm to compensate signal damage caused by XPM, wherein in the JNL neural network, an NPN component part of the XPM is compensated through an iterative process of a linear digital filter of the LMS algorithm; NPCC layer based on NPCC principle compensates NPC of XPM component part;

And the fifth processing module is used for calculating the bit error rate of the damaged signal compensated by the JNL neural network to finish nonlinear damage joint compensation in and among channels of the WDM system.

8. An electronic device comprising a memory, a processor and a computer program stored on the memory and running on the processor, the processor executing the program to implement the steps of the WDM system nonlinearity compensation method that requires only target channel information as claimed in any one of claims 1-6.