An optical fiber nonlinear equalization method, system, electronic device and medium

Abstract

The application discloses an optical fiber nonlinear equalization method, system, electronic equipment and medium, and relates to the technical field of optical fiber communication, and the method comprises the following steps: acquiring a target symbol sequence; the target symbol sequence is a symbol sequence of a received optical signal to be tested received by a receiving end; a perturbation triple product eigenvalue of each symbol in the target symbol sequence is constructed by using a perturbation theory to obtain a target eigenvalue set; the target eigenvalue set is input into an equalizer model to obtain a data category of the target symbol sequence after equalization; wherein the equalizer model is obtained by training a complex full-connection neural network by using training data. The application can obtain an equalizer model closer to a real value, and improve the performance of optical fiber signal nonlinear equalization.

Description

Technical Field

[0001] This invention relates to the field of optical fiber communication technology, and in particular to an optical fiber nonlinear equalization method, system, electronic device, and medium. Background Technology

[0002] With the rapid development of optical fiber communication technology in recent years, high-capacity, long-distance, and high-speed optical fiber communication systems have gained increasing attention. However, due to various linear and nonlinear impairments affecting optical signals during transmission, the transmission channel capacity faces increasing pressure. Digital signal processing (DSP) algorithms, such as maximum likelihood sequence equalization (MLSE), can effectively address most linear impairments, including chromatic dispersion (CD) and polarization mode dispersion (PMD). However, when transmitting high-order modulation formats, the nonlinear effects of the optical fiber intensify as the transmitted optical power increases, severely impacting system performance. Therefore, researching DSP algorithms that effectively overcome optical fiber nonlinear effects has become a hot topic in the field of optical fiber communication. Classic nonlinear equalization algorithms include digital backpropagation (DBP), Volterra series transfer function (VSTF), and equalization algorithms based on perturbation theory (PB). Among them, the PB algorithm treats nonlinear impairment as a nonlinear perturbation term and solves the perturbation term according to the nonlinear Schrödinger equation (NLSE) for nonlinear compensation. However, this scheme requires prior knowledge of the accurate parameter information of the optical fiber channel and requires a lot of computational resources, which is not conducive to practical applications.

[0003] Currently, the rise of artificial intelligence has led to machine learning being increasingly recognized as a promising tool for addressing various challenges in optical communication. In particular, neural network (NN)-based algorithms have demonstrated their significant potential in mitigating nonlinear transmission impairments in optical communication links. The main idea behind neural network-based optical signal equalization is to continuously learn the complex mapping relationship between the transmitting and receiving signals, and adjust the weight matrices of each layer of the network according to certain rules (learning algorithms). Once the weights of each layer converge to a certain value, a well-fitted model is established, and finally, this model is used to equalize other distorted received signals. For example, fully connected neural networks (FNN), convolutional neural networks (CNN), recurrent neural networks (RNN), long short-term memory neural networks (LSTM), and gated recurrent neural networks (GRU) have all been proven effective for nonlinear equalization of optical signals.

[0004] However, in signal processing within fiber optic communication systems, digital signals are typically represented using complex numbers. Since most current neural network frameworks use real numbers instead of complex numbers, it's difficult to represent the correlation between amplitude and phase when processing optical signals. Given the mathematical rationale behind complex number representation, researchers have further considered the potential of artificial neural networks using complex numbers to represent parameters such as inputs, outputs, and weights in these areas. Therefore, complex valued neural networks (CVNNs), which use fully complex parameters and variables to process information, have gradually come into the researchers' view.

[0005] In summary, although significant progress has been made in research on nonlinear equalization using neural networks, these networks typically require a large amount of feature data to drive training and obtain accurate models. In practical applications, the benefits gained by neural networks from directly learning and training from the symbol sequences of received signals are limited. Therefore, obtaining more meaningful feature data is the primary consideration when applying neural networks as nonlinear equalizers. Summary of the Invention

[0006] Based on this, embodiments of the present invention provide a fiber optic nonlinear equalization method, system, electronic device, and medium. The method uses perturbation theory to analyze the input signal to promote the initialization of the complex-valued neural network, thereby obtaining an equalizer model that is closer to the true value and improving the performance of fiber optic signal nonlinear equalization.

[0007] To achieve the above objectives, the present invention provides the following solution:

[0008] A fiber nonlinear equalization method includes:

[0009] Obtain the target symbol sequence; the target symbol sequence is the symbol sequence of the received optical signal to be tested received by the receiver; the symbol sequence includes multiple symbols;

[0010] The perturbation triple product eigenvalues ​​of each symbol in the target symbol sequence are constructed using perturbation theory to obtain the target eigenvalue set;

[0011] The target feature value set is input into the equalizer model to obtain the data category after the target symbol sequence is equalized;

[0012] The equalizer model is obtained by training a complex-valued fully connected neural network using training data.

[0013] The training data includes: a training feature set and data categories of symbol sequences of transmitted optical signals from the transmitter; the training feature set includes perturbation triple product feature values ​​of each symbol in the symbol sequence of the received training optical signals constructed using perturbation theory.

[0014] Optionally, the perturbation triple product eigenvalues ​​of each symbol in the target symbol sequence are constructed using perturbation theory to obtain the target eigenvalue set, specifically including:

[0015] For the k-th symbol in the target symbol sequence, determine the real and imaginary component data of the k-th symbol and its adjacent symbols, wherein the adjacent symbols include symbols within a set range adjacent to the k-th symbol;

[0016] The nonlinear Schrödinger equation that must be satisfied by the transmission characteristics of optical signals when they are transmitted in single-mode optical fiber;

[0017] Based on the aforementioned nonlinear Schrödinger equation, the perturbation triple product eigenvalue of the k-th symbol in the target symbol sequence is constructed using perturbation theory.

[0018] A target feature value set is constructed based on the perturbation triple product feature values ​​of all symbols in the target symbol sequence.

[0019] Optionally, the expression for the perturbation triple product eigenvalue of the k-th symbol is:

[0020]

[0021] Among them, T x / y (z=L,t=k) represents the perturbation triple product eigenvalue of the k-th symbol; z represents the transmission distance; t represents the symbol number; L represents the length of the single-mode fiber; A represents the first part of the perturbation triple product eigenvalue of the k-th symbol; U represents the second part of the perturbation triple product eigenvalue of the k-th symbol; x (z=L,t=k+n) represents the optical field of the (k+n)th symbol of the optical signal in the x-polarization state; U represents the conjugate transpose of the optical field of the (k+m+n)th symbol of the optical signal in the x-polarization state; x (z=L,t=k+m) represents the optical field of the (k+m)th symbol of the optical signal in the x-polarization state; U y (z=L,t=k+n) represents the optical field of the (k+n)th symbol of the optical signal in the y polarization state; U represents the conjugate transpose of the optical field of the (k+m+n)th symbol in the y-polarization state of an optical signal; y (z=L,t=k+m) represents the light field of the (k+m)th symbol of the optical signal in the y-polarization state; the x-polarization state and the y-polarization state are perpendicular; m and n determine the size of the set range.

[0022] Optionally, m and n satisfy the constraint condition; the constraint condition is:

[0023]

[0024] Where P is a set constant.

[0025] Optionally, the method for determining the equalizer model is as follows:

[0026] The data categories of the symbol sequence of the training received optical signal received by the receiver and the symbol sequence of the transmitted optical signal at the transmitter are obtained.

[0027] The perturbation triple product eigenvalues ​​of each symbol in the symbol sequence of the received optical signal are constructed using perturbation theory to obtain the training feature set.

[0028] The training data is determined based on the training feature set and the data category of the symbol sequence of the transmitted optical signal from the transmitter.

[0029] A complex-valued fully connected neural network is constructed; the complex-valued fully connected neural network includes: an input layer, a hidden layer and an output layer connected in sequence; the hidden layer includes two fully connected layers; the neurons in the fully connected layers are all complex-valued neurons;

[0030] The training data is input into the complex-valued fully connected neural network. The Adam optimizer is used to optimize the network parameters of the complex-valued fully connected neural network with the goal of minimizing the error loss function, resulting in a trained complex-valued fully connected neural network. The error loss function is determined based on the data category of the symbol sequence of the transmitted optical signal at the transmitter and the category of the predicted data output by the complex-valued fully connected neural network.

[0031] The trained complex-valued fully connected neural network was determined as the equalizer model.

[0032] The present invention also provides an optical fiber nonlinear equalization system, comprising:

[0033] The signal acquisition module is used to acquire a target symbol sequence; the target symbol sequence is the symbol sequence of the received optical signal to be tested received by the receiver; the symbol sequence includes multiple symbols.

[0034] The feature value construction module is used to construct the perturbation triple product feature value of each symbol in the target symbol sequence using perturbation theory, so as to obtain the target feature value set;

[0035] The data equalization module is used to input the target feature value set into the equalizer model to obtain the data category after the target symbol sequence is equalized;

[0036] The equalizer model is obtained by training a complex-valued fully connected neural network using training data.

[0037] The training data includes: a training feature set and data categories of symbol sequences of transmitted optical signals from the transmitter; the training feature set includes perturbation triple product feature values ​​of each symbol in the symbol sequence of the received training optical signals constructed using perturbation theory.

[0038] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the above-described fiber nonlinear equalization method.

[0039] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described fiber nonlinear equalization method.

[0040] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0041] This invention proposes a fiber optic nonlinear equalization method, system, electronic device, and medium. It employs perturbation theory to construct the perturbation triple product eigenvalues ​​of each symbol in the target symbol sequence, obtaining a target eigenvalue set. This target eigenvalue set is then input into an equalizer model to obtain the data category of the equalized target symbol sequence. The equalizer model is obtained by training a complex-valued fully connected neural network using training data. This invention uses perturbation theory to analyze the input signal to facilitate the initialization of the complex-valued neural network, thereby obtaining an equalizer model that more closely approximates the true values ​​and improving the performance of fiber optic signal nonlinear equalization. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 A flowchart of the fiber nonlinear equalization method provided in an embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram of the complex-valued neuron structure provided in an embodiment of the present invention;

[0045] Figure 3 This is a block diagram of the CFNN structure provided in an embodiment of the present invention;

[0046] Figure 4 This is a schematic diagram illustrating the selection of m and n values ​​provided in an embodiment of the present invention;

[0047] Figure 5This is a schematic diagram comparing the performance of the equalization algorithm provided in the embodiments of the present invention;

[0048] Figure 6 This is a structural diagram of an optical fiber nonlinear equalization system provided in an embodiment of the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] The purpose of this invention is to provide a fiber optic nonlinear equalization method, system, electronic device, and medium based on a perturbation theory-based complex-valued fully connected neural network (PB-CFNN). The method employs perturbation theory to analyze the input signal to facilitate the initialization of the CFNN. This physically meaningful initialization can better guide the network training towards a model that more closely approximates the true values, thereby significantly reducing the training workload. Furthermore, because the complex characteristics of the signal's features perfectly match the fully complex structure of the CFNN, it can achieve better performance than real-valued neural networks.

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] Example 1

[0053] See Figure 1 The fiber nonlinear equalization method in this embodiment includes:

[0054] Step 101: Obtain the target symbol sequence; the target symbol sequence is the symbol sequence of the received optical signal to be tested received by the receiver; the symbol sequence includes multiple symbols.

[0055] Step 102: Construct the perturbation triple product eigenvalues ​​of each symbol in the target symbol sequence using perturbation theory to obtain the target eigenvalue set.

[0056] Step 103: Input the target feature value set into the equalizer model to obtain the data category after the target symbol sequence is equalized.

[0057] The equalizer model is obtained by training a complex-valued fully connected neural network using training data. The training data includes: a training feature set and data categories of the symbol sequences of the transmitted optical signals from the transmitter; the training feature set includes the perturbation triple product feature values ​​of each symbol in the symbol sequence of the received training optical signals, constructed using perturbation theory.

[0058] In one instance, step 102 specifically includes:

[0059] 1) For the k-th symbol in the target symbol sequence, determine the real and imaginary component data of the k-th symbol and its adjacent symbols. The adjacent symbols include symbols within a defined range adjacent to the k-th symbol. In this embodiment, the adjacent symbols include the first m, first n, first m+n, last m, last n, and last m+n symbols adjacent to the k-th symbol.

[0060] 2) Construct the nonlinear Schrödinger equation that the transmission characteristics of optical signals in single-mode optical fiber must satisfy.

[0061] 3) Based on the aforementioned nonlinear Schrödinger equation, the perturbation triple product eigenvalue of the k-th symbol in the target symbol sequence is constructed using perturbation theory.

[0062] 4) Construct a target feature value set based on the perturbation triple product feature values ​​of all symbols in the target symbol sequence. The expression for the perturbation triple product feature value of the k-th symbol is:

[0063]

[0064] Among them, T x / y (z=L,t=k) represents the perturbation triple product eigenvalue of the k-th symbol; z represents the transmission distance; t represents the symbol number; L represents the length of the single-mode fiber; A represents the first part of the perturbation triple product eigenvalue of the k-th symbol; U represents the second part of the perturbation triple product eigenvalue of the k-th symbol; x (z=L,t=k+n) represents the optical field of the (k+n)th symbol of the optical signal in the x-polarization state; U represents the conjugate transpose of the optical field of the (k+m+n)th symbol of the optical signal in the x-polarization state; x (z=L,t=k+m) represents the optical field of the (k+m)th symbol of the optical signal in the x-polarization state; U y (z=L,t=k+n) represents the optical field of the (k+n)th symbol of the optical signal in the y polarization state; U represents the conjugate transpose of the optical field of the (k+m+n)th symbol in the y-polarization state of an optical signal; y(z = L, t = k + m) represents the optical field of the (k + m)th symbol of the optical signal in the y-polarization state; the x-polarization state and the y-polarization state are perpendicular; m and n determine the size of the set range. m and n satisfy the constraint condition; the constraint condition is...

[0065]

[0066] Where P is a set constant.

[0067] In one example, the method for determining the equalizer model in step 103 is as follows:

[0068] 1) Obtain the data categories of the symbol sequence of the training received optical signal received by the receiver and the symbol sequence of the transmitted optical signal from the transmitter.

[0069] 2) The perturbation triple product eigenvalues ​​of each symbol in the symbol sequence of the received optical signal are constructed using perturbation theory to obtain the training feature set. This step is similar to the process of constructing the perturbation triple product eigenvalues ​​in step 102 above, and will not be repeated here.

[0070] 3) Determine the training data based on the training feature set and the data category of the symbol sequence of the transmitted optical signal at the transmitter.

[0071] 4) Construct a Complex Value Fully Connected Neural Network (CFNN). The CFNN includes: an input layer, a hidden layer, and an output layer connected in sequence; the hidden layer includes two fully connected layers; and the neurons in the fully connected layers are all complex value neurons.

[0072] 5) Input the training data into the complex-valued fully connected neural network, and use the Adam optimizer to optimize the network parameters of the complex-valued fully connected neural network with the goal of minimizing the error loss function, so as to obtain the trained complex-valued fully connected neural network; the error loss function is determined according to the data category of the symbol sequence of the transmitted optical signal at the transmitter and the predicted data category output by the complex-valued fully connected neural network.

[0073] 6) The trained complex-valued fully connected neural network is determined as the equalizer model.

[0074] The fiber nonlinear equalization method described above will be further explained below.

[0075] Perturbation theory is used to analyze symbol samples at the optical signal receiver, constructing physically meaningful perturbation features, which are then used as the training set input to the neural network. The original training dataset is used to train the CFNN, and the training model that best fits the true value equalizer model is obtained. Input features are constructed for the test signal and used as the test set input to the trained model for equalization.

[0076] The input features for constructing symbol samples include:

[0077] When compensating the kth symbol, the I and Q data of that symbol are obtained, as well as the I and Q component data of the m, n, and m+n adjacent symbols before and after that symbol. When processing information, I and Q can be regarded as the real and imaginary parts of the signal, respectively.

[0078] The transmission characteristics of polarization-multiplexed optical pulses in ordinary single-mode optical fibers satisfy the nonlinear Schrödinger equation:

[0079]

[0080] Where U(z,t) represents the optical field, the subscripts x / y represent the two polarization states of the optical pulse, α represents the fiber loss, β² represents the second-order dispersion, and γ represents the nonlinear parameter related to the effective mode area of ​​the fiber's nonlinear refractive index. According to perturbation theory, the nonlinear term jγ|U(z,t)| 2 U(z,t) can be considered as a perturbation term. At the fiber optic transmitter, i.e., at z=0, the transmitted optical signal is assumed to be a Gaussian pulse sequence:

[0081]

[0082] Where A k Let represent the amplitude of the k-th symbol, T represent the symbol duration, and τ represent the width of the Gaussian pulse. Based on the above two formulas, the nonlinear perturbation triple product term T experienced by the k-th symbol at a transmission distance z = L is obtained. x / y (z = L, t = k) is represented as follows:

[0083]

[0084] This allows us to construct the complex-valued input features of the symbol. After reprocessing all the symbol sample sets sequentially, we obtain the original training dataset.

[0085] Training the CFNN using the original training dataset and fitting the equalizer model involves two stages: a forward propagation stage and a back propagation stage, specifically:

[0086] Forward propagation phase: The feature triple product term of the current input signal is passed to the CFNN, and data classification is performed through the input layer, fully connected layer, and output layer. A neuron is the smallest unit that makes up a neural network. Figure 2 Let y be a complex-valued neuron in a CFNN, where the input x, bias b, weight ω, nonlinear activation function f, and output y are all in complex-valued form, and i refers to the i-th neuron. i The complex-valued input of the i-th neuron, ω i This refers to the complex weight of the i-th neuron. The output expression is:

[0087]

[0088] Figure 2 The output of a complex-valued neuron is the weighted sum of all its inputs. The weights of different inputs are the neuron parameters, and the optimization process of a neural network is the process of optimizing the values ​​of these neuron parameters. Figure 3 The basic structure of CFNN is shown, where the hidden layers consist of two fully connected layers. The obtained N-dimensional perturbation triple product eigenvalues ​​are fed into the input layer, where N is determined by the values ​​of m and n. Figure 3 The arrows in the diagram illustrate its forward propagation process within the fully connected layer. Figure 3 It can be seen that each node in CNFF has an operational relationship with all nodes in the next layer, that is, the number of parameters input to the next layer = the number of input nodes in the previous layer × the total number of nodes in this layer + the number of biases. The input data is activated by the complex-valued nonlinear activation function f and then passed to the next layer, until it is passed to the output layer for classification and output. The complex-valued nonlinear activation function f is the complex form of the ReLU function.

[0089] CReLU=max(0,Re(x))+jmax(0,Im(x)).

[0090] Furthermore, the output activation function of the output layer is also in complex form, expressed as:

[0091]

[0092] Backpropagation stage: Based on the final classification of the signal obtained from the forward propagation algorithm, the error between the predicted result and the correct answer is compared with the correct classification of the original training data. Based on this difference, the backpropagation algorithm optimizes the loss function that measures the error to find its minimum extreme value. This invention uses the classic Adam optimizer for optimization. Backpropagation continuously updates a series of weights ω and biases b, making the values ​​of the neural network parameters closer to the true answer, until the expected result is fitted.

[0093] The test signal is used as the test set input to the trained model for equalization: all the test signals in the test set are used to construct triple product eigenvalues ​​through perturbation theory, and then they are sequentially input into the fitted neural network model to carry out the complete propagation process and obtain the output classification. The output classification result is compared with the original data classification to obtain the accuracy. From the test accuracy, the bit error rate, Q factor and other key indicators of the optical fiber communication system can be calculated.

[0094] In practical applications, a specific implementation process of the above-mentioned fiber nonlinear equalization method is as follows:

[0095] This specific example uses a dual-polarization 64th-order quadrature amplitude modulation (PD-64QAM) signal commonly used in high-capacity fiber optic communication. At the coherent receiver, the received signal is converted into a digital signal by an analog-to-digital converter (ADC), and then signal equalization is performed by an offline DSP. The offline DSP mainly includes dispersion compensation (CDC), clock recovery (CR), polarization mode dispersion (PMD) compensation, frequency offset estimation (FOE), carrier phase recovery (CPR), and PB-CFNN equalization methods. The specific steps of the equalization are as follows:

[0096] Step S1: When compensating the kth symbol, the I and Q data of the symbol are obtained respectively, as well as the I and Q component data of the m, n, and m+n adjacent symbols before and after the symbol. When processing information, I and Q can be regarded as the real part and imaginary part of the signal respectively.

[0097] S11. Based on the above relevant formulas, the nonlinear perturbation triple product term T of the current symbol caused by in-channel four-wave mixing and in-channel cross-phase modulation is obtained. x / y (z=L,t=k), thus constructing the complex-valued input features of the symbol.

[0098] S12, m, and n are the key factors determining the triplet. Considering the complexity of actual calculations, the smallest region that has a greater impact on the k-th sign should be selected. Therefore, this specific example uses the following inequality to constrain the values ​​of m and n:

[0099]

[0100] like Figure 4 As shown, this represents the range of values ​​for m and n.

[0101] S13. Construct the complex-valued input features of the symbol using steps S11 and S12. Repeat this process on all symbol sample sets to obtain the original training dataset.

[0102] Step S2: Training the CFNN using the original training dataset and fitting the equalizer model requires two processes.

[0103] S21. Perform the forward propagation process, passing the feature triple product term of the current input signal to the CFNN. Data classification is then performed through the input layer, fully connected layer, and output layer. In this specific example, the PD-64QAM signal has 64 neurons in the output layer used for classification output. Figure 3 The arrows in the diagram illustrate its forward propagation process in the fully connected layer.

[0104] S22. Perform the backpropagation process. Based on the signal classification results from step S21, use the error loss function to calculate the error between the predicted result and the correct answer. Based on this gap, use the classic Adam optimizer to optimize the loss function, continuously updating a series of neural network-related parameters such as weights ω and biases b, so that the values ​​of the neural network parameters are closer to the true answer, until the expected result is fitted.

[0105] S23. Repeat the above process iteratively until the loss function of the neural network no longer changes and approaches convergence. At this point, the network model has completed the fitting.

[0106] Step S3: Construct input features for the signal under test and input them into the CFNN equalizer for classification. The input feature construction method is the same as in step S1. Construct triple product feature values ​​for all the signals under test in the test set using perturbation theory, and then input them sequentially into the fitted neural network model to complete the propagation process and obtain the output classification. Compare the output classification result with the original data classification to obtain the accuracy. From the test accuracy, key indicators such as the Q factor of the optical fiber communication system can be calculated. Figure 5 This example demonstrates the compensation effect of different equalization algorithms on the received signal under different transmit optical power (LOP). The Q factor of PB-CFNN is significantly higher than that of the real-valued fully connected neural network (PB-RFNN) equalizer based on perturbation theory when the LOP is between -2 and 5 dBm, and it is also much higher than the algorithm without nonlinear compensation (w / oNLC).

[0107] Example 2

[0108] In order to implement the method corresponding to Embodiment 1 above and achieve the corresponding functions and technical effects, an optical fiber nonlinear equalization system is provided below.

[0109] See Figure 6 The system includes:

[0110] The signal acquisition module 601 is used to acquire a target symbol sequence; the target symbol sequence is the symbol sequence of the received optical signal to be tested received by the receiver; the symbol sequence includes multiple symbols.

[0111] The feature value construction module 602 is used to construct the perturbation triple product feature value of each symbol in the target symbol sequence using perturbation theory, so as to obtain the target feature value set.

[0112] The data equalization module 603 is used to input the target feature value set into the equalizer model to obtain the data category after the target symbol sequence is equalized.

[0113] The equalizer model is obtained by training a complex-valued fully connected neural network using training data. The training data includes a training feature set and data categories of the symbol sequence of the transmitted optical signal at the transmitter. The training feature set includes the perturbation triple product feature values ​​of each symbol in the symbol sequence of the received training optical signal constructed using perturbation theory.

[0114] Example 3

[0115] This embodiment provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor runs the computer program to enable the electronic device to perform the fiber nonlinear equalization method of Embodiment 1.

[0116] Alternatively, the aforementioned electronic device may be a server.

[0117] In addition, embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the fiber nonlinear equalization method of Embodiment 1.

[0118] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0119] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A fiber optic nonlinear equalization method, characterized in that, include: Obtain the target symbol sequence; the target symbol sequence is the symbol sequence of the received optical signal to be tested received by the receiver. The code element sequence includes multiple symbols; The perturbation triple product eigenvalues ​​of each symbol in the target symbol sequence are constructed using perturbation theory to obtain the target eigenvalue set; The target feature value set is input into the equalizer model to obtain the data category after the target symbol sequence is equalized; The equalizer model is obtained by training a complex-valued fully connected neural network using training data. The training data includes: a training feature set and data categories of symbol sequences of transmitted optical signals from the transmitter; the training feature set includes perturbation triple product feature values ​​of each symbol in the symbol sequence of the received training optical signals constructed using perturbation theory. The perturbation triple product eigenvalues ​​of each symbol in the target symbol sequence are constructed using perturbation theory to obtain the target eigenvalue set, which specifically includes: For the first in the target symbol sequence k The symbol determines the first... k The real and imaginary data of the first symbol and its adjacent symbols, wherein the adjacent symbols include those related to the first symbol. k Symbols within a defined range adjacent to each other; The nonlinear Schrödinger equation that must be satisfied by the transmission characteristics of optical signals when they are transmitted in single-mode optical fiber; Based on the aforementioned nonlinear Schrödinger equation, perturbation theory is used to construct the first... k The perturbation triple product eigenvalues ​​of each symbol; A target feature value set is constructed based on the perturbation triple product feature values ​​of all symbols in the target symbol sequence.

2. The fiber nonlinear equalization method according to claim 1, characterized in that, No. k The expression for the eigenvalues ​​of the perturbation triple product of denoted symbols is: ； in, Indicates the first k The perturbation triple product eigenvalue of each symbol; z represents the transmission distance; t represents the symbol number; L Indicates the length of a single-mode fiber; the first... k The first part of the perturbation triple product eigenvalue of the nth symbol; A represents the first part of the perturbation triple product eigenvalue of the nth symbol. k The second part of the perturbation triple product eigenvalues ​​of each symbol; Indicates the light signal at x The first polarization state k+n The light field of a symbol; Indicates the light signal at x The first polarization state k+m+n The conjugate transpose of the light field of each symbol; Indicates the light signal at x The first polarization state k+m The light field of each symbol; Indicates the light signal at y The first polarization state k+n The light field of a symbol; Indicates the light signal at y The first polarization state k+m+n The conjugate transpose of the light field of each symbol; Indicates the light signal at y The first polarization state k+m The light field of each symbol; x polarization state and y Polarization state is perpendicular; m and n Determine the size of the set range.

3. The fiber nonlinear equalization method according to claim 2, characterized in that, m and n The constraints are satisfied; the constraints are: ； in, P This is a set constant.

4. The fiber nonlinear equalization method according to claim 1, characterized in that, The method for determining the equalizer model is as follows: The data categories of the symbol sequence of the training received optical signal received by the receiver and the symbol sequence of the transmitted optical signal at the transmitter are obtained. The perturbation triple product eigenvalues ​​of each symbol in the symbol sequence of the received optical signal are constructed using perturbation theory to obtain the training feature set. The training data is determined based on the training feature set and the data category of the symbol sequence of the transmitted optical signal from the transmitter. Construct a complex-valued fully connected neural network; The complex-valued fully connected neural network includes: an input layer, a hidden layer, and an output layer connected in sequence; the hidden layer includes two fully connected layers; and the neurons in the fully connected layers are all complex-valued neurons. The training data is input into the complex-valued fully connected neural network. The Adam optimizer is used to optimize the network parameters of the complex-valued fully connected neural network with the goal of minimizing the error loss function, resulting in a trained complex-valued fully connected neural network. The error loss function is determined based on the data category of the symbol sequence of the transmitted optical signal at the transmitter and the category of the predicted data output by the complex-valued fully connected neural network. The trained complex-valued fully connected neural network was determined as the equalizer model.

5. A fiber optic nonlinear equalization system, characterized in that, include: The signal acquisition module is used to acquire the target symbol sequence; the target symbol sequence is the symbol sequence of the received optical signal to be tested received by the receiver. The code element sequence includes multiple symbols; The feature value construction module is used to construct the perturbation triple product feature value of each symbol in the target symbol sequence using perturbation theory, so as to obtain the target feature value set; The data equalization module is used to input the target feature value set into the equalizer model to obtain the data category after the target symbol sequence is equalized; The equalizer model is obtained by training a complex-valued fully connected neural network using training data. The training data includes: a training feature set and data categories of symbol sequences of transmitted optical signals from the transmitter; the training feature set includes perturbation triple product feature values ​​of each symbol in the symbol sequence of the received training optical signals constructed using perturbation theory. The perturbation triple product eigenvalues ​​of each symbol in the target symbol sequence are constructed using perturbation theory to obtain the target eigenvalue set, which specifically includes: For the first in the target symbol sequence k The symbol determines the first... k The real and imaginary data of the first symbol and its adjacent symbols, wherein the adjacent symbols include those related to the first symbol. k Symbols within a defined range adjacent to each other; The nonlinear Schrödinger equation that must be satisfied by the transmission characteristics of optical signals when they are transmitted in single-mode optical fiber; Based on the aforementioned nonlinear Schrödinger equation, perturbation theory is used to construct the first... k The perturbation triple product eigenvalues ​​of each symbol; A target feature value set is constructed based on the perturbation triple product feature values ​​of all symbols in the target symbol sequence.

6. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the fiber nonlinear equalization method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the fiber nonlinear equalization method as described in any one of claims 1 to 4.

Images