A coherent optical communication auxiliary optical fiber channel modeling method, system and electronic equipment

Abstract

The present invention relates to a method, system, and electronic device for modeling coherent optical communication-assisted optical fiber channels. The modeling method comprises: obtaining a multi-span transmission signal sequence to construct an initial neural network model; using the multi-span transmission signal sequence to train the initial neural network model to obtain a tuned neural network model; importing the tuned neural network model and cascading a linear physical model and an amplifier to obtain a single-span optical fiber link channel model; and iteratively transmitting signals based on the single-span optical fiber link channel model according to a preset transmission scenario to obtain multi-span transmission results. The present invention can simultaneously achieve low-complexity and high-prediction accuracy optical communication transmission simulation modeling.

Description

Technical Field

[0001] The present invention relates to the field of optical communication technology, and in particular to a method, system and electronic equipment for coherent optical communication auxiliary optical fiber channel modeling. Background Art

[0002] As the main artery of information and communication networks, optical communications are the solid foundation of the contemporary digital economy, deeply empowering emerging industries such as 5G networks, cloud computing, and large-scale computing networks. In the development of optical fiber communication technology, increasing system capacity and optimizing system design are crucial to address potential capacity shortages. Establishing optical fiber channel models for high-speed, ultra-wideband, and long-distance transmission plays a vital role in signal transmission evolution analysis, communication quality assessment, and overall system optimization design.

[0003] The goal of fiber channel modeling is to identify a method that can quickly convert input signals into output signals, and can characterize the channel transmission characteristics with high accuracy and low computational complexity. The transmission response of signals in optical fibers is characterized by the nonlinear Schrödinger equation. For solving this complex two-dimensional partial differential equation, there is no mathematically achievable analytical solution, and it can only be approximated through numerical calculations. Although the traditional distributed Fourier method can provide reliable simulation results, its modeling accuracy is affected by the degree of discreteness of the fiber link grid. With the increase in sampling rate and the introduction of complex nonlinear characteristics, the corresponding modeling complexity and simulation time will increase dramatically, inevitably leading to large storage requirements and long time consumption, and its application in multi-span wavelength division multiplexing long-distance transmission simulation is limited.

[0004] Based on this, there is an urgent need for a simulation modeling method with high modeling accuracy and low complexity to solve the problems existing in the above-mentioned existing technologies. Summary of the Invention

[0005] In order to solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a coherent optical communication auxiliary optical fiber channel modeling method, system and electronic equipment, which can simultaneously have low complexity and high prediction accuracy optical communication transmission simulation modeling.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A coherent optical communication-assisted optical fiber channel modeling method, comprising:

[0008] Obtain multi-segment transmission signal sequences and construct an initial neural network model;

[0009] Training the initial neural network model using a multi-segment transmission signal sequence to obtain an optimized neural network model;

[0010] Importing the tuned neural network model, and cascading the linear physical model and the amplifier to obtain a single-span optical fiber link channel model;

[0011] A transmission scenario is preset, and according to the transmission scenario, a signal is iteratively transmitted based on the single-span optical fiber link channel model to obtain a multi-span transmission result.

[0012] Optionally, a polarization multiplexing-wavelength division multiplexing coherent optical communication system is used to obtain the multi-span transmission signal sequence.

[0013] Optionally, the polarization multiplexing-wavelength division multiplexing coherent optical communication system includes: a laser, which generates an optical pulse, which is polarized and split by a polarization beam splitter, and outputs X-polarized and Y-polarized optical carriers. The X-polarized and Y-polarized optical carriers are respectively and simultaneously acted on an IQ modulator with an electrical signal based on a preset modulation format for electro-optical conversion to complete the modulation of the electrical signal and output an optical signal. After the polarization combiner, wavelength division multiplexer and amplifier perform wave combining and fiber input power control, the input signal is obtained, the input signal is input into a channel model, and the output signal is obtained. After that, the output signal is received by wavelength division multiplexing and a coherent receiver, and then undergoes subsequent digital signal processing and demodulation judgment.

[0014] Optionally, the channel model includes: a plurality of optical fiber channels and amplifiers;

[0015] The transmission evolution process of the input signal in the optical fiber channel is expressed as follows:

[0016]

[0017] Where z is the transmission distance, t is the time, α, β2, and γ are the attenuation coefficient, group velocity dispersion coefficient, and nonlinear coefficient, respectively. x (z, t) is the X-polarized optical carrier, E y (z, t) is the Y-polarized optical carrier, is the partial derivative operation, and j is the imaginary unit.

[0018] Optionally, obtaining the multi-span transmission signal sequence includes:

[0019] Processing the multi-span transmission signal sequence:

[0020] The linear effect of the output signal in the multi-span transmission signal sequence is extracted and compensated once, the input signal at different spans and the compensated output signal are power normalized respectively, the real and imaginary part information of the signal are extracted respectively, and one-dimensional splicing and cutting are performed.

[0021] Optionally, extracting and compensating for linear effects of output signals in the multi-span transmission signal sequence includes:

[0022] Set the received signal transmission length to have the same linear characteristics as the actual link, and set α and β2 to the opposite of their original values:

[0023]

[0024] Where C is the compensation operator, is the frequency domain representation of the electric field at z+L, is the frequency domain representation of the electric field at z, α is the attenuation, β2 is the group velocity dispersion, ω is the angular frequency, j is the imaginary unit, and L is the transmission distance.

[0025] Optionally, the initial neural network model adopts an MLP-Mixer model, wherein the MLP-Mixer model mainly extracts complex features based on a channel feature fusion module and a spatial feature fusion module, and introduces layer normalization operations and residual connections into the module to further improve the feature extraction and characterization capabilities of the model.

[0026] Optionally, obtaining the tuned neural network model includes:

[0027] Using the stochastic gradient descent method, the processed multi-span transmission signal is input into the MLP-Mixer model to perform model parameter training to obtain a tuned neural network model;

[0028] During the training process, the mean square error loss function is used to calculate the loss value and the loss value is used as the judgment standard when it drops to the minimum during iterative training until it stabilizes, so as to end the training and save the tuned model parameters; the mean square error is the mean square error between the neural network prediction value and the distribution Fourier method calculation result.

[0029] To achieve the above object, the present invention further provides a coherent optical communication auxiliary optical fiber channel modeling system, comprising:

[0030] A sequence acquisition module is used to acquire a multi-segment transmission signal sequence;

[0031] Model building module, used to build the initial neural network model;

[0032] A model training module is used to train the initial neural network model using a multi-segment transmission signal sequence to obtain an optimized neural network model;

[0033] The multi-span transmission module is used to import the tuned neural network model and cascade it with the linear physical model and amplifier to obtain a single-span optical fiber link channel model, preset a transmission scenario, and according to the transmission scenario, iteratively transmit the signal based on the single-span optical fiber link channel model to obtain a multi-span transmission result.

[0034] To achieve the above-mentioned purpose, the present invention also provides an electronic device for coherent optical communication-assisted optical fiber channel modeling, comprising a memory, a processor, a communication interface and a communication device, wherein the processor, the communication interface and the memory communicate with each other through a communication bus, and a program is stored on the memory and applied to the channel modeling system, and the method described is implemented when the program is executed by the processor.

[0035] The beneficial effects of the present invention are:

[0036] (1) The present invention selects the MLP-Mixer model as the neural network channel model. Its modeling accuracy is better than the existing deep learning modeling method. Compared with the traditional distributed Fourier method, the simulation time is shortened by dozens of times, and it has the advantages of high accuracy and low complexity.

[0037] (2) The present invention can integrate data-driven neural network models and physical models. The neural network can effectively extract the nonlinear characteristics of the optical fiber transmission channel and cascade the linear physical model and amplifier to effectively model the real optical fiber transmission link, avoiding the traditional modeling method that uses complex numerical calculations and discrete iterative calculations, and significantly improving the simulation operation speed.

[0038] (3) The technical solution provided by the present invention, including the optical communication-assisted fiber channel modeling method, system, and electronic device, can have good generalization performance while ensuring accuracy, and does not require repeated training for different transmission scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 This is a flow chart of a coherent optical communication auxiliary optical fiber channel modeling method according to an embodiment of the present invention;

[0041] Figure 2 This is a block diagram of the coherent optical communication simulation system architecture used in an embodiment of the present invention; wherein, Figure 2 (a) is the block diagram of the coherent optical communication simulation system used. Figure 2 (b) is the digital signal processing flow chart of the transmitting end of the coherent optical communication simulation system. Figure 2 (c) is a flow chart of digital signal processing at the receiving end of the coherent optical communication simulation system;

[0042] Figure 3 A schematic diagram of a data set preparation process used in an embodiment of the present invention;

[0043] Figure 4 A schematic diagram of the neural network model structure used in an embodiment of the present invention;

[0044] Figure 5 Schematic diagram of the comparison results between the modeling method and the distributed Fourier method under the same conditions in an embodiment of the present invention; wherein, Figure 5 (a) is the time domain waveform comparison result of the modeling method and the distributed Fourier method under the same conditions. Figure 5 (b) is the spectrum distribution comparison result of the modeling method and the distributed Fourier method under the same conditions. Figure 5 (c) Comparison results of constellation diagrams between the modeling method and the distributed Fourier method under the same conditions;

[0045] Figure 6 Schematic diagram of error accumulation results and time consumption in link transmission of the modeling method and the distributed Fourier method under the same conditions in an embodiment of the present invention;

[0046] Figure 7 Schematic diagram of a coherent optical communication auxiliary optical fiber channel modeling system according to an embodiment of the present invention;

[0047] Figure 8 Schematic diagram of an electronic device for modeling a coherent optical communication auxiliary optical fiber channel according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] like Figure 1 As shown, this embodiment discloses a coherent optical communication-assisted optical fiber channel modeling method, including: obtaining a multi-span transmission signal sequence to construct an initial neural network model; using the multi-span transmission signal sequence to train the initial neural network model to obtain a tuned neural network model; importing the tuned neural network model, and cascading the linear physical model and the amplifier to obtain a single-span optical fiber link channel model; presetting a transmission scenario, and according to the transmission scenario, iteratively transmitting the signal based on the single-span optical fiber link channel model to obtain a multi-span transmission result.

[0051] Specifically: This embodiment discloses a coherent optical communication auxiliary optical fiber channel modeling method, including:

[0052] Step 1: Acquire a multi-span transmission signal sequence based on a preset polarization multiplexing-wavelength division multiplexing coherent optical communication system.

[0053] Step 2: Preprocess the input and output signals and create a data set.

[0054] Step 3: Build a neural network model and initialize model parameters.

[0055] Step 4: Complete the training and testing of the neural network based on the dataset.

[0056] Step 5: Store the tuned neural network model parameters as a single-span channel model.

[0057] Step 6: For the set long-distance transmission scenario, a neural network model is imported and multi-segment transmission modeling is completed through loop iteration.

[0058] The long-distance transmission scenario here involves transmitting optical signals over a distance of 1200 km in an optical fiber. During this process, the optical signal will pass through 15 spans of optical fiber, each span being 80 km long. Between adjacent spans, the optical signal is relayed and amplified by an erbium-doped fiber amplifier (EDFA) to compensate for the effects of channel attenuation on the optical signal power.

[0059] Furthermore, a polarization multiplexing-wavelength division multiplexing coherent optical communication system is used to obtain a multi-span transmission signal sequence.

[0060] Furthermore, the polarization multiplexing-wavelength division multiplexing coherent optical communication system includes: a laser, which generates an optical pulse, which is polarized and split by a polarization beam splitter, and outputs an X-polarized and a Y-polarized optical carrier. The X-polarized and the Y-polarized optical carriers are respectively acted on the IQ modulator simultaneously with the electrical signal based on the preset modulation format to complete the electrical signal modulation and output the optical signal. After the polarization combiner, the wavelength division multiplexer and the amplifier are combined and the fiber input power is controlled, the fiber input optical signal is obtained, the optical signal is input into the channel model, and the output signal is obtained. After that, it is received by the wavelength division multiplexing and coherent receiver, and then undergoes subsequent digital signal processing and demodulation judgment, such as Figure 2 As shown in (a).

[0061] Specifically, in this embodiment, a polarization multiplexing-wavelength division multiplexing coherent optical communication system is constructed to obtain a multi-span transmission signal sequence. The system framework includes: a transmitter, a channel, and a receiver, such as Figure 2(a) shows the transmitter parameter settings. The transmitter parameter settings include the signal modulation format, transmission rate, sampling rate, number of symbols, number of channels, channel spacing, modulator bias voltage, and the roll-off factor and filter required for DSP processing. Channel parameters include the attenuation coefficient, group velocity dispersion coefficient, and self-phase modulation coefficient. The receiver parameters include the corresponding parameter settings required for DSP processing. The communication system settings are completed according to the preset parameters, and the Quadrature Amplitude Modulation (QAM) electrical signal is generated based on the transmitter digital signal processing (DSP) process.

[0062] For example, the number of transmission channels is 5, the frequency interval is 50 GHz, and a single-channel signal is generated by 5 independent transmitters, where 5 lasers are centered at a wavelength of 1550 nm and the frequency interval between adjacent ones is 50 GHz. In each channel, the signal is modulated in QAM format, the modulation order is 16, the upsampling rate is 16, and the pulse shaping roll-off factor is 0.1, so that a 16QAM random bit stream can be obtained. The laser generates and outputs an optical pulse, which is polarized and split by a polarization beam splitter. The X-polarized / Y-polarized optical carrier acts on the IQ modulator simultaneously with the electrical signal to perform electro-optical conversion on the electrical signal, and then is combined by a polarization combiner. After that, the optical signals of different wavelengths are combined by a wavelength division multiplexer, and then input into the optical fiber channel model after the fiber input power is controlled by the amplifier. The evolution process of the optical signal during transmission in the optical fiber is characterized by the Manakov equation, as shown in the following formula:

[0063]

[0064] Wherein, signal E=[E x ,E y ] T The signal containing X and Y polarization can be expressed as: z represents the transmission distance, t represents the time, C represents the number of channels, Δf k is the frequency interval, α, β2, and γ are the attenuation coefficient, group velocity dispersion coefficient, and nonlinear coefficient, respectively. The distributed Fourier method is applied to obtain the input and output signal sequences of each span.

[0065] The distributed Fourier transform method discretizes a fiber link into subspans using a symmetrical partitioning scheme. Based on the following equations, nonlinear effects are calculated in the time domain, or dispersion effects are calculated in the frequency domain. Ultimately, an iterative operation is used to construct a fiber channel model. The distributed Fourier transform solution is used as a reference. The channel output data is obtained after 80 km of fiber. Both input and output data exhibit all the characteristics of simulated optical signals.

[0066]

[0067] Where L and N represent the linear operator and nonlinear operator respectively, E represents the electric field distribution, z represents the transmission distance, ω represents the angular frequency, T represents time, and j is the imaginary unit.

[0068] like Figure 2 As shown in (b), the DSP process at the transmitter end of the coherent optical communication simulation system includes: generating a random bit stream, mapping it based on a preset QAM modulation format to generate an electrical signal, upsampling the electrical signal, and pulse shaping the electrical signal based on a root cosine filter and a preset roll-off factor.

[0069] At the receiving end, the optical signals of each channel are extracted after passing through the wavelength division multiplexer, and the electrical signals received by the coherent receiver are based on Figure 2 (c) Perform DSP processing. The DSP process includes low-pass filtering, resampling to 2x the sampling rate, dispersion compensation / digital backpropagation, matched filtering using a root-spurious cosine filter, carrier phase recovery algorithm to estimate the signal phase, demapping based on the preset QAM modulation format to restore the random bit stream, and bit error rate calculation.

[0070] Furthermore, obtaining the multi-segment transmission signal sequence includes:

[0071] Processing of multi-span transmission signal sequences: Extraction and compensation of linear effects on the output signals in the multi-span transmission signal sequences, power normalization of the input signals at different spans and the compensated output signals, extraction of the real and imaginary part information of the signals, and one-dimensional splicing and cutting.

[0072] Specifically, in step 2, the acquired multi-span optical fiber transmission signal is preprocessed, mainly by extracting and compensating the linear effect of the single-span output signal at one time. The specific implementation can be based on the digital back propagation algorithm, setting the optical fiber characteristic parameters α and β2 to the opposite of the original values ​​to simulate the received signal transmission with the same length having opposite linear characteristics to the actual link. The expression can be simplified as:

[0073]

[0074] like Figure 3 As shown in the figure, the power normalization is performed on the preprocessed output signals and the original input signals of each span. Considering that the neural network processes real numbers, while the QAM signal is complex, the real and imaginary parts of the signal are extracted and stacked and spliced ​​separately, so that different parts of the same signal are represented at the same location (at the same time). A sliding window is then set to segment and arrange the sequence.

[0075] For example, a window size of 16 symbols is selected. Taking into account the inter-symbol interference introduced by dispersion on the symbols, in order to better extract features, there is an overlap of 8 symbols between the segmented signals. The dimension of the signal length is defined as the spatial dimension, and the dimension of the single real-valued part of the corresponding signal at the same time is defined as the channel dimension to complete the construction of the dataset.

[0076] Furthermore, in step 3, the neural network is constructed, and the initial neural network model adopts the MLP-Mixer model, including: input layer, linear layer, MLP-Mixer module layer, linear layer and output layer;

[0077] Input layer, used to receive input sequence information;

[0078] Linear layer, used to extract feature information of the input sequence and map it to a high-dimensional space;

[0079] The MLP-Mixer module layer is constructed by stacking several feature fusion modules, each of which performs channel feature fusion operations and spatial feature fusion operations in sequence. The feature fusion operation is performed through two linear layers and a nonlinear activation function. Layer normalization is performed before the fusion operation. Residual connections are introduced in the fusion process. The corresponding features are fused according to the channel dimension and spatial dimension respectively, so that the neural network can capture and extract key features and has strong representation capabilities.

[0080] The linear layer is used to map the features extracted by the MLP-Mixer module layer to the original space where the signal is located;

[0081] The output layer is used to obtain the output prediction sequence results based on the mapped signal in the original space.

[0082] Specifically, the MLP-Mixer structure is as follows Figure 4 As shown, its notable feature is that the entire model is completely built based on the Multi-Layer Perception (MLP) layer. Its advantage is that data processing and feature extraction can be performed only through basic matrix multiplication. The nonlinear activation function selects GELU, and introduces advanced technologies such as layer normalization and residual links to help the model learn better to extract features. The core of the structure is composed of a stack of feature fusion modules. The feature fusion module contains two steps. First, the channel dimension feature fusion is performed, and then the position dimension feature fusion is performed. Each feature fusion is mainly completed by two layers of linear layers combined with a nonlinear activation function operation. Before fusion, the data needs to be layer normalized. After the model is built, the neural network is initialized; the initialization adopts the Xavier initialization method. For example, the feature fusion module here is 4, and the MLP-Mixer module layer is expressed by the following formula:

[0083] Z′ l =W2σ(W1LN(Zl-1))+Z l-1

[0084] Z l =W4σ(W3LN(Z′ l ))+Z′ l ;

[0085] Among them, W i (=1,2,3,4) are weight parameters, is the l-1th hidden layer output, d c is the channel dimension, d s is the position dimension, LN represents the layer normalization operation, and σ is the nonlinear activation function.

[0086] Furthermore, obtaining the tuned neural network model includes: using the stochastic gradient descent method to input the processed multi-segment transmission signal into the MLP-Mixer model to train the model parameters to obtain the tuned neural network model; during the training process, using the mean square error loss function to calculate the loss value and using the value dropped to the lowest level in iterative training until it stabilizes as a judgment criterion to end the training and save the tuned model parameters; the mean square error is the mean square error between the neural network prediction value and the distribution Fourier method calculation result.

[0087] Specifically, in step 4, the training of the neural network is to calculate the error based on the set loss function and update the weights and biases of the neural network through the back-propagation mechanism. The parameters are adjusted with an appropriate learning rate in multiple iterative learning to minimize the loss function. For example, the loss function is defined as the mean square difference between the output signal predicted by the neural network and the simulation result based on the distributed Fourier method. The optimizer is Adam, and the learning rate is adjusted using warm-up cosine annealing. During the training process, the training mode is full-batch, that is, a batch size of data is randomly captured from the data set for training in one iteration epoch, rather than traversing the entire data set. The iteration cycle is 50,000 times.

[0088] After training is completed, the model parameters are saved and the accuracy and generalization of the model are tested based on the test set data.

[0089] Specifically, step 5 stores the tuned neural network model parameters, mainly saving the model structure, the tuned weights and biases carried by each neuron.

[0090] Specifically, step 6 imports the neural network model for the specified transmission scenario and completes multi-span transmission modeling through a loop of iterations. For example, for a long-haul transmission scenario with 15 spans, each span spanning 80 km of fiber, the input optical signal will undergo 15 iterative transmissions. In each iteration, the output of the neural network channel model, the linear characteristic module, and the amplifier module is used as the output of the single-span fiber channel. Figure 5 (a)-5(c) respectively give the distribution of the signal in the time domain, frequency domain and the constellation diagram restored by DSP after reception. It can be seen that the proposed channel model has good consistency with the results calculated by the traditional distributed Fourier method.

[0091] exist Figure 6 The normalized mean square error and the cumulative computation time for each span of 131,072 symbols in iterative transmission are given in

[15] . Under the same conditions, the neural network channel model (11 seconds) is much faster than the distributed Fourier method (259 seconds). The cumulative modeling error after 1,200 km transmission is 5.5×10 -3 , which verifies the effectiveness of the proposed channel modeling scheme.

[0092] like Figure 7 As shown, this embodiment also provides a coherent optical communication auxiliary optical fiber channel modeling system, including:

[0093] A sequence acquisition module is used to acquire a multi-segment transmission signal sequence;

[0094] Model building module, used to build the initial neural network model;

[0095] A model training module is used to train an initial neural network model using a multi-segment transmission signal sequence to obtain an optimized neural network model;

[0096] The multi-span transmission module is used to import the tuned neural network model and cascade it with the linear physical model and amplifier to obtain a single-span fiber link channel model. Based on the preset transmission scenario, the input optical signal of the starting span is fed into the single-span fiber link channel model to iteratively transmit the signal and obtain the multi-span transmission result based on the preset integer multiple transmission distance.

[0097] Among them, the linear physical model includes attenuation and group velocity dispersion effects.

[0098] This embodiment also discloses an electronic device for coherent optical communication-assisted optical fiber channel modeling. Figure 8The specific structure of the device is demonstrated, including a memory, a processor, a communication interface, a communication device, and a communication bus. The processor, the communication interface, and the memory communicate with each other via the communication bus, while a computer program for the optical fiber channel model is stored in the memory. When the computer program is executed on the processor, a coherent optical communication-assisted optical fiber channel modeling method can be implemented.

[0099] The memory is used to store computer programs, that is, all scripts and codes that implement the Fibre Channel model. The memory can be a storage device such as a USB flash drive, a mobile hard drive, or a random access memory (RAM).

[0100] The processor is used to execute the program stored in the memory to implement the steps of the fiber channel model provided in the above embodiment. The processor can be a general-purpose processor, or a digital signal processor, field programmable gate array, or other device;

[0101] Communication interfaces are used for communication between the above-mentioned electronic devices and other devices, including input / output (I / O) interfaces, physical interfaces, and logical interfaces that realize the interconnection between internal devices of the device and between the device and external devices;

[0102] The term "communication bus" is represented by a double-arrowed line, but this does not necessarily mean there is only one bus or type of bus. Communication buses can be standard peripheral component interconnect buses or extended industry standard architecture buses, and can be categorized as data buses, control buses, address buses, and so on.

[0103] Those skilled in the art will understand that Figure 8 The hardware structure shown in the figure does not constitute a limitation to the present application and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0104] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A coherent optical communication-assisted fiber channel modeling method, characterized in that: include: Obtain multi-segment transmission signal sequences and construct an initial neural network model; Training the initial neural network model using a multi-segment transmission signal sequence to obtain an optimized neural network model; After obtaining the multi-span transmission signal sequence, the method includes: Processing the multi-span transmission signal sequence: Extracting and compensating the linear effect of the output signal in the multi-span transmission signal sequence, performing power normalization processing on the input signal at different spans and the compensated output signal, extracting the real and imaginary part information of the signal, and performing one-dimensional splicing and cutting; Extracting and compensating for linear effects of output signals in the multi-span transmission signal sequence includes: Set the received signal transmission length to have the same linear characteristics as the actual link, and set α and β2 to the opposite of their original values: Where C is the compensation operator, is the frequency domain representation of the electric field at z+L, is the frequency domain representation of the electric field at z, α is the attenuation, β2 is the group velocity dispersion, ω is the angular frequency, j is the imaginary unit, and L is the transmission distance; The initial neural network model adopts an MLP-Mixer model, wherein the MLP-Mixer model mainly extracts complex features based on a channel feature fusion module and a spatial feature fusion module, and introduces layer normalization operations and residual connections in the module to further improve the feature extraction and representation capabilities of the model; Using the stochastic gradient descent method, the processed multi-span transmission signal is input into the MLP-Mixer model to perform model parameter training to obtain a tuned neural network model; During the training process, the loss value is calculated using the mean square error loss function and is judged by the minimum value dropped to a stable value during iterative training to end the training and save the tuned model parameters; the mean square error is the mean square error between the neural network prediction value and the distribution Fourier method calculation result; Importing the tuned neural network model, and cascading the linear physical model and the amplifier to obtain a single-span optical fiber link channel model; A transmission scenario is preset, and according to the transmission scenario, a signal is iteratively transmitted based on the single-span optical fiber link channel model to obtain a multi-span transmission result.

2. The coherent optical communication assisted fiber channel modeling method according to claim 1, characterized in that: A polarization multiplexing-wavelength division multiplexing coherent optical communication system is used to obtain the multi-span transmission signal sequence.

3. The coherent optical communication assisted fiber channel modeling method according to claim 2, characterized in that: The polarization multiplexing-wavelength division multiplexing coherent optical communication system includes: a laser, which generates an optical pulse, and after polarization control and beam splitting are performed by a polarization beam splitter, the laser outputs an X-polarized and a Y-polarized optical carrier, and the X-polarized and the Y-polarized optical carriers are respectively acted on an IQ modulator simultaneously with an electrical signal based on a preset modulation format to complete electrical-optical conversion to complete electrical signal modulation and output an optical signal, and then after wavelength division multiplexing and fiber input power control are performed by a polarization combiner, a wavelength division multiplexer and an amplifier, the input signal is obtained, and the input signal is input into a channel model to obtain an output signal, which is then received by wavelength division multiplexing and a coherent receiver, undergoes subsequent digital signal processing, and is demodulated and judged.

4. The coherent optical communication-assisted optical fiber channel modeling method according to claim 3, characterized in that: The channel model includes: a plurality of optical fiber channels and amplifiers; The transmission evolution process of the input signal in the optical fiber channel is expressed as follows: Where z is the transmission distance, t is the time, α, β2, and γ are the attenuation coefficient, group velocity dispersion coefficient, and nonlinear coefficient, respectively. x (z, t) is the X-polarized optical carrier, E y (z, t) is the Y-polarized optical carrier, is the partial derivative operation, and j is the imaginary unit.

5. A coherent optical communication assisted fiber channel modeling system implemented according to the method according to any one of claims 1 to 4, characterized in that: include: A sequence acquisition module is used to acquire a multi-segment transmission signal sequence; Model building module, used to build the initial neural network model; A model training module is used to train the initial neural network model using a multi-segment transmission signal sequence to obtain an optimized neural network model; The multi-span transmission module is used to import the tuned neural network model and cascade it with the linear physical model and amplifier to obtain a single-span optical fiber link channel model, preset a transmission scenario, and according to the transmission scenario, iteratively transmit the signal based on the single-span optical fiber link channel model to obtain a multi-span transmission result.

6. An electronic device for coherent optical communication-assisted optical fiber channel modeling, comprising a memory, a processor, a communication interface, and a communication device, wherein the processor, the communication interface, and the memory communicate with each other via a communication bus, and a program is stored on the memory and applied to a channel modeling system. When the program is executed by the processor, the method described in any one of claims 1 to 4 is implemented.

Images