CNN-LSTM-based mode division multiplexing optical fiber communication phase calibration method

Abstract

The invention discloses a mode division multiplexing optical fiber communication phase calibration method based on CNN-LSTM, and belongs to the technical field of optical fiber communication. By constructing a double-input CNN-LSTM hybrid neural network architecture, MIMO input signals and original target signals of mode division multiplexing optical fiber communication are obtained, data preprocessing is carried out, and the spatial feature extraction capability of a convolutional neural network and the time sequence modeling capability of a long-short-term memory network are combined, so that the MIMO input signals and the original target signals of mode division multiplexing optical fiber communication are obtained. Effective compensation of phase noise in long-distance strong-coupling mode division multiplexing optical fiber communication is realized, and a double-branch network comprising an image input layer, a convolutional layer, an LSTM layer, a deep connection layer and a user-defined phase calibration layer is constructed by separating complex MIMO signals into real and imaginary parts and generating a sliding window training sample. The problems that a traditional algorithm is limited in phase noise compensation capacity and poor in convergence performance under the strong coupling condition are effectively solved, and the balance performance and the transmission quality of a mode division multiplexing optical fiber communication system are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical fiber communications and relates to a CNN-LSTM-based mode division multiplexing optical fiber communication phase calibration method. Background Art

[0002] With the rapid development of applications such as 5G, the Internet of Things, and cloud computing, the demand for data transmission is exploding. Mode division multiplexing (MDM) has become a key technology for increasing the capacity of optical fiber communications. However, over long-distance transmission, factors such as mode coupling, phase noise, and dispersion severely impact signal quality. Traditional linear equalization algorithms have limitations when dealing with strong coupling and nonlinear effects.

[0003] In the prior art, commonly used equalization algorithms include:

[0004] 1. Least Mean Square Algorithm: This is an adaptive filtering algorithm based on gradient descent. It has slow convergence, limited performance in strong coupling situations, and difficulty handling rapidly changing channel conditions.

[0005] 2. Normal mode algorithm: a blind equalization algorithm that does not require a training sequence, but is sensitive to phase noise and prone to mode aliasing in multi-mode systems;

[0006] 3. Neural network equalization: It can handle nonlinear effects, but lacks effective extraction of time series features and has high training complexity.

[0007] The shortcomings of the existing technology are:

[0008] Limited phase noise compensation capability: Traditional algorithms mainly focus on amplitude equalization, and lack effective compensation mechanisms for laser phase noise and nonlinear phase noise in optical fibers; Insufficient utilization of spatiotemporal correlation: Failure to fully utilize the correlation of mode-division multiplexing signals in the time and space dimensions leads to limited equalization performance; Poor convergence performance under strong coupling conditions: Under the strong mode coupling conditions of long-distance transmission, traditional algorithms are prone to fall into local optimality, with slow convergence and poor stability; Lack of adaptive phase calibration mechanism: In high-speed transmission systems, existing methods find it difficult to track and compensate for rapidly changing phase noise in real time; Difficulty in balancing computational complexity and performance: High-performance algorithms are often accompanied by high computational complexity, which makes it difficult to meet real-time processing requirements. Summary of the Invention

[0009] In response to the existing technical problems, the present invention provides a CNN-LSTM-based mode division multiplexing optical fiber communication phase calibration method.

[0010] A CNN-LSTM-based phase calibration method for mode-division multiplexing (MDM) fiber communications is proposed. By constructing a dual-input CNN-LSTM hybrid neural network architecture, the MIMO input signal and original target signal of the MDM fiber communications are obtained, and data preprocessing is performed. The spatial feature extraction capability of the convolutional neural network and the temporal modeling capability of the long short-term memory network are combined to effectively compensate for phase noise in long-distance, strongly coupled MDM fiber communications. By separating the complex MIMO signal into real and imaginary parts and generating sliding window training samples, a dual-branch network consisting of an image input layer, a convolutional layer, an LSTM layer, a deep connection layer, and a custom phase calibration layer is constructed. Adaptive estimation and compensation of phase noise are achieved through iterative training and a phase feedback mechanism, and decision feedback is used to optimize test results.

[0011] The beneficial effects of the present invention include:

[0012] 1. Strong Coupling Adaptability: The CNN-LSTM architecture can effectively extract the spatiotemporal characteristics of signals and adapt to long-distance, strongly coupled environments. The 3×3 kernel function of the convolutional layer and the 256 hidden units of the LSTM can handle complex inter-mode coupling relationships, significantly improving adaptability to strongly coupled environments compared to traditional linear equalizers.

[0013] 2. Phase noise compensation capability: Real-time estimation and compensation of phase noise are achieved through a phase calibration layer and iterative feedback mechanism. The block mean-based phase estimation algorithm can effectively suppress phase jitter caused by laser phase noise and fiber nonlinear effects, effectively improving phase noise suppression capabilities.

[0014] 3. Adaptive convergence performance: Decision feedback and multiple rounds of iterative training (configurable number of iterations) are used to improve the system's adaptability and convergence. The Adam optimizer and gradient clipping technology ensure training stability, and the convergence speed is faster than the traditional recursive least squares algorithm.

[0015] 4. Multi-mode collaborative processing: Fully utilize the correlation information between multiple modes through deep connection layers; support 6-mode MIMO processing, and can simultaneously process multiple spatially multiplexed light field modes.

[0016] 5. Real-time processing capability: The LSTM's "last" output mode ensures real-time processing capability; the single sample processing latency is less than 10μs, meeting the real-time requirements of high-speed optical communication systems and supporting data transmission rates exceeding 100Gbps.

[0017] 6. Signal quality improvement: Through the sliding window mechanism and timing modeling, signal quality is significantly improved; the bit error rate (BER) is reduced by 1-2 orders of magnitude compared to traditional DSP algorithms, and the convergence of the signal constellation diagram is significantly improved.

[0018] 7. System robustness: A dual-input architecture and phase feedback mechanism enhance the system's robustness to channel variations. Even when channel conditions deteriorate, it maintains stable demodulation performance, offering greater adaptability than traditional fixed-parameter algorithms.

[0019] 8. Computational efficiency optimization: Improve computing efficiency through network structure optimization and batch processing technology; compared with traditional deep learning methods, the number of parameters is reduced by 40% and the training time is shortened by 50%, which facilitates hardware implementation and actual deployment.

[0020] 9. Engineering Practicality: The algorithm has good engineering practicality and scalability. It supports configurations with different numbers of modes and channel conditions, can flexibly adapt to the needs of different optical fiber communication systems, and provides an effective technical solution for the industrial application of mode division multiplexing technology.

[0021] 10. Innovative Technology Fusion: This innovative technology combines CNN spatial feature extraction, LSTM timing modeling, and a custom phase calibration layer for the first time, creating a complete end-to-end optimization solution. Compared to single technology approaches, this solution significantly improves overall performance and opens up a new technical path for intelligent signal processing in optical fiber communications. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] 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 the embodiments or the description of the prior art. 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 work. As shown in the figure:

[0023] Figure 1 This is a schematic diagram of the overall structure of the CNN-LSTM network architecture of the present invention, showing the complete process of dual input branches, feature extraction, fusion processing and output.

[0024] Figure 2 This is a schematic diagram of sliding window data preprocessing, which illustrates the generation method of time series windows and the boundary processing strategy.

[0025] Figure 3 This is a detailed structural diagram of the phase calibration layer, showing the specific implementation of phase noise estimation and compensation.

[0026] Figure 4 This is the iterative training flow chart, which illustrates the iterative process of phase noise adaptive estimation and network parameter update.

[0027] Figure 5 This is the convergence diagram of the method of the present invention, which reflects the effects of the algorithm convergence speed, convergence error, etc.

[0028] Figure 6 The constellation diagrams of the 6-mode MDM system are compared, showing the equalization effect of the method of the present invention in different modes. DETAILED DESCRIPTION

[0029] 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.

[0030] Example 1: Figure 1 、 Figure 2 、 Figure 3 、 Figure 4 、 Figure 5 and Figure 6 As shown, a phase calibration method for mode division multiplexing optical fiber communication based on CNN-LSTM, a phase calibration method for mode division multiplexing optical fiber communication based on convolutional neural network and long short-term memory network, is suitable for signal equalization and phase noise compensation in long-distance strongly coupled mode division multiplexing optical fiber communication system. It is a new method that can effectively deal with phase noise in long-distance strongly coupled mode division multiplexing optical fiber communication, which can not only ensure high-performance signal equalization, but also meet the requirements of real-time processing.

[0031] A CNN-LSTM-based phase calibration method for mode-division multiplexing optical fiber communications combines the feature extraction capabilities of convolutional neural networks with the timing modeling capabilities of long-short-term memory networks to effectively compensate for phase noise in long-distance strongly coupled mode-division multiplexing optical fiber communications, thereby improving the system's equalization performance and transmission quality.

[0032] A CNN-LSTM-based mode division multiplexing optical fiber communication phase calibration method comprises the following steps:

[0033] Step 1: Data Preprocessing

[0034] Obtain the MIMO input signal and the original signal of the mode division multiplexing optical fiber communication, and separate the complex signal into real and imaginary parts;

[0035] Set the sliding window parameters, including tap length and sample block size for phase calculation;

[0036] The sliding window technique is used to generate training samples, and the window size is 2×window length+1;

[0037] Step 2: Network architecture construction

[0038] Construct a dual-input CNN-LSTM hybrid neural network, including:

[0039] First input branch: The image input layer receives data of dimension [number of patterns × 2, window size, 1], performs feature extraction through the convolution layer, and then outputs it through the flattening layer;

[0040] LSTM feature extraction layer: an LSTM layer with 256 hidden units, using the "last" output mode and an output dimension of 12;

[0041] The second input branch receives a phase noise compensation signal with a dimension of [12, 1, 1].

[0042] Deep connection layer: concatenates the outputs of the two branches;

[0043] Phase calibration layer: Customize the phase calibration layer to achieve real-time compensation of phase noise;

[0044] Fully connected output layer: outputs a 12-dimensional signal, corresponding to the real and imaginary parts of the 6 modes;

[0045] Step 3: Iterative training and phase estimation

[0046] Initialize the phase noise compensation vector PN to [1,1,1,1,1,1,0,0,0,0,0,0];

[0047] Conduct the first network training to obtain initial prediction results;

[0048] Calculate phase noise based on the predicted complex signal and the training target signal;

[0049] Update the phase noise compensation vector;

[0050] Repeat the training process to achieve adaptive estimation and compensation of phase noise;

[0051] Step 4: Testing and decision feedback

[0052] Predict the test data and obtain the initial equilibrium result;

[0053] Applying phase calibration and computing the calibrated complex signal;

[0054] Obtain the decision signal through hard decision and update the training target;

[0055] Perform multiple rounds of iterative optimization to improve balancing accuracy;

[0056] Step 5: Output Processing

[0057] The 12-dimensional real vector output by the network is converted into 6-mode complex signal output.

[0058] like Figure 1Figure 2 shows the neural network structure with two input branches. The feature extraction path includes input layer 1 (12×(tap×2+1)×1 dimensions), a 3×3 convolutional layer (12 filters), a flattening layer, an LSTM layer (256 hidden units, last output mode), and a fully connected layer 2 (12 neurons). The phase compensation path includes input layer 2 (12×1×1 dimensions) and a flattening layer. The two branches are merged through a deep connection layer (Concat), then pass through a custom phase calibration layer (phaseCalibrationLayer3), and finally pass through a fully connected layer 3 (12 neurons) and an output layer to produce a 12-dimensional output.

[0059] like Figure 2 The top diagram shows the sliding window mechanism for the input data sequence, constructing a window centered around the current sample and encompassing samples from all preceding and following moments. The bottom diagram illustrates how to handle three edge cases: zero padding at the beginning of the sequence, a full window under normal circumstances, and zero padding at the end of the sequence, ensuring that all samples can construct input windows of uniform size.

[0060] like Figure 3 As shown in Figure 2, the internal working mechanism of the phase calibration layer is shown. lstm ) and the phase input (P_ n ) as input, the phase angle and phase calibration factor are calculated by the phase estimation module, and finally the phase calibration is performed in the phase compensation module, and the calibrated signal F_ is output. out .

[0061] like Figure 4 The complete training and testing process is shown in Figure 1. It starts with initializing the PN (phase noise vector) and then enters a loop: training the network, predicting the output, calculating the phase noise, updating the PN vector, and determining whether the maximum number of iterations has been reached. If not, the loop continues, and if so, the final result is output.

[0062] like Figure 5 The figure below shows the trend of mean squared error (MSE) as the number of iterations increases. The horizontal axis represents the number of iterations, and the vertical axis represents the MSE. The curve shows that the error decreases rapidly from an initial value of approximately 0.67, decreasing sharply over the first five iterations. It stabilizes around the tenth iteration and eventually converges to near zero, demonstrating the algorithm's rapid convergence.

[0063] like Figure 6Figure 1 shows a comparison of the constellation plots for the six patterns on the training and test sets. The left side shows the training set results, and the right side shows the test set results. Each subplot shows the distribution of the predicted and true values on the complex plane. The training set results show that the predicted points are highly concentrated and overlap well with the true values. While the test set results are slightly more dispersed, they still maintain a good clustering effect, verifying the effectiveness and generalization ability of the algorithm.

[0064] Example 2: Figure 1 、 Figure 2 、 Figure 3 、 Figure 4 、 Figure 5 and Figure 6 As shown in the figure, a CNN-LSTM-based mode division multiplexing optical fiber communication phase calibration method is described in detail using a 6-mode mode division multiplexing optical fiber communication system as an example:

[0065] Network parameter settings:

[0066] Input dimension: 12×(2×window size + 1).

[0067] Number of LSTM hidden units: 256

[0068] Convolution kernel size: 3×3, number of output channels: 12.

[0069] Learning rate: 1e-4.

[0070] Batch size: 128.

[0071] Phase estimation block size: 20.

[0072] Data preprocessing:

[0073] Convert complex signals into real and imaginary part formats.

[0074] Generate a sliding window.

[0075] Network training: The first phase uses an initial phase estimate for training, followed by adaptive optimization using phase feedback. Phase noise estimates are updated after each epoch to achieve gradual convergence.

[0076] Phase calibration implementation:

[0077] 1. Receive feature and phase inputs from the LSTM layer.

[0078] 2. Calculate the phase compensation factor.

[0079] 3. Perform phase calibration on the signal.

[0080] 4. Output the calibrated signal characteristics.

[0081] Example 3: Figure 1 、 Figure 2 、 Figure 3 、 Figure 4 、 Figure 5 and Figure 6 As shown in the figure, a CNN-LSTM-based phase calibration method for mode division multiplexing optical fiber communication can be adapted to MDM systems with different numbers of modes:

[0082] 4-MODE SYSTEM:

[0083] The input dimension is resized to 8×(2×window size+1).

[0084] The output dimension is adjusted to 8.

[0085] Other parameters remain unchanged.

[0086] 12-Mode System:

[0087] The input dimension is resized to 24×(2×window size+1).

[0088] The output dimension is adjusted to 24.

[0089] The number of LSTM hidden units can be increased to 512 to improve processing power.

[0090] Example 4: Figure 1 、 Figure 2 、 Figure 3 、 Figure 4 、 Figure 5 and Figure 6 As shown, a CNN-LSTM-based mode division multiplexing optical fiber communication phase calibration method includes the following steps:

[0091] Step 1: Data preprocessing,

[0092] Get the MIMO input signal and original target signal of mode division multiplexing optical fiber communication,

[0093] Separate the complex signal into real and imaginary parts, forming a data format in which the real and imaginary parts are interleaved.

[0094] Set the sliding window parameters, including the tap length and sample block size for phase calculation,

[0095] The sliding window technique is used to generate training samples. The window size is 2×window length+1, and the boundaries are padded with zeros.

[0096] Step 2: Build a dual-input CNN-LSTM hybrid neural network architecture,

[0097] First input branch: Set the image input layer to receive sliding window data of dimension [number of patterns × 2, window size, 1], perform spatial feature extraction through the convolution layer, and then output a one-dimensional feature vector through the flattening layer;

[0098] LSTM feature extraction layer: An LSTM layer containing 256 hidden units, using the "last" output mode to model temporal features.

[0099] The second input branch receives a phase noise compensation signal of dimension [12, 1, 1].

[0100] Deep connection layer: deeply concatenate the LSTM output features and phase compensation signals.

[0101] Phase calibration layer: Customize the phase calibration layer to achieve real-time compensation of phase noise.

[0102] Fully connected output layer: outputs 12-dimensional signals, corresponding to the real and imaginary parts of 6 modes,

[0103] Step 3: Iterative training and adaptive phase estimation,

[0104] Initialize the phase noise compensation vector PN to [1,1,1,1,1,1,0,0,0,0,0,0],

[0105] Conduct the first round of network training to obtain initial prediction results.

[0106] Calculate the phase noise based on the predicted complex signal and the target signal,

[0107] Update the phase noise compensation vector,

[0108] Repeat the training process to achieve adaptive estimation and compensation of phase noise,

[0109] Step 4: Testing and decision feedback optimization,

[0110] Make initial predictions on the test data to get balanced results,

[0111] Apply phase calibration and compute the calibrated complex signal,

[0112] Obtain the decision signal through hard decision and update the training target.

[0113] Perform multiple rounds of iterative optimization to improve the balancing accuracy.

[0114] Step 5: Output complex signal,

[0115] The 12-dimensional real vector output by the network is converted into 6-mode complex signal output.

[0116] During the sliding window generation process, zero padding is performed on the boundary samples: when the window length is insufficient, zero padding is performed at the beginning of the sequence or at the end of the sequence to ensure that the window size of each sample is consistent.

[0117] In the CNN-LSTM network architecture, the convolutional layer uses a 3×3 convolution kernel, 12 output channels, and the same padding method. The LSTM layer has 256 hidden units, and the output mode is set to "last" to ensure real-time processing capabilities.

[0118] The phase noise estimation adopts a block processing method, and estimates the phase noise by calculating the phase difference between the predicted signal and the target signal within a sample block. The sample block size is a configurable parameter.

[0119] The iterative training process uses the Adam optimizer, with a learning rate set to 1e-4, a batch size of 128, and a gradient clipping threshold of 1. Multiple rounds of iterations are used to achieve joint optimization of network parameters and phase estimation.

[0120] In the decision feedback process, 4-QAM hard decision is used to obtain a decision signal, and the decision result is used as a new training target to achieve adaptive optimization in the test phase.

[0121] The method is applicable to MDM systems with different numbers of modes.

[0122] For a 4-mode system, the input dimension is adjusted to 8×(2×window size+1) and the output dimension is 8.

[0123] For the 12-mode system, the input dimension is adjusted to 24×(2×window size+1), the output dimension is 24, and the number of LSTM hidden units is increased to 512.

[0124] The phase calibration layer receives the feature vector and phase compensation signal from the LSTM, performs real-time phase calibration on the signal through the phase compensation factor, and outputs the calibrated signal feature.

[0125] The network training adopts a data storage combination method to combine the sliding window data, the phase compensation signal and the target signal, and supports batch parallel processing.

[0126] Example 5: Figure 1 、 Figure 2 、 Figure 3 、 Figure 4 、 Figure 5 and Figure 6As shown, a CNN-LSTM-based mode division multiplexing optical fiber communication phase calibration method is shown. By constructing a dual-input CNN-LSTM hybrid neural network architecture, the MIMO input signal and the original target signal of the mode division multiplexing optical fiber communication are obtained, and data preprocessing is performed. The spatial feature extraction capability of the convolutional neural network and the temporal modeling capability of the long short-term memory network are combined to achieve effective compensation of the phase noise in long-distance strongly coupled mode division multiplexing optical fiber communication. By separating the complex MIMO signal into real and imaginary parts and generating sliding window training samples, a dual-branch network including an image input layer, a convolution layer, an LSTM layer, a deep connection layer and a custom phase calibration layer is constructed. The adaptive estimation and compensation of the phase noise are achieved through iterative training and a phase feedback mechanism, and the test results are optimized using decision feedback.

[0127] The data preprocessing process includes: separating a complex signal into a real part and an imaginary part to form a data format in which the real and imaginary parts are interleaved, setting sliding window parameters including tap length and phase calculation sample block size, generating training samples using a sliding window technique, wherein the window size is 2×window length+1, and zero padding is performed on the boundaries to ensure that the window size of each sample is consistent.

[0128] The dual-input CNN-LSTM hybrid neural network architecture includes: the first input branch sets the image input layer to receive sliding window data of the dimension [number of patterns × 2, window size, 1], performs spatial feature extraction through a convolution layer using a 3×3 convolution kernel, 12 output channels and the same padding method, and then outputs a one-dimensional feature vector through a flattening layer; an LSTM feature extraction layer containing 256 hidden units and using the "last" output mode models the temporal features; the second input branch receives a phase noise compensation signal of the dimension [12, 1, 1]; a deep connection layer deeply splices the LSTM output features and the phase compensation signal; a custom phase calibration layer realizes real-time compensation of phase noise; and a fully connected output layer outputs a 12-dimensional signal corresponding to the real and imaginary parts of the six patterns.

[0129] The iterative training and phase feedback mechanism includes: initializing the phase noise compensation vector PN to [1,1,1,1,1,1,0,0,0,0,0,0], performing a first round of network training to obtain an initial prediction result, calculating the phase noise based on the predicted complex signal and the target signal, updating the phase noise compensation vector, and repeating the training process to achieve adaptive estimation and compensation of the phase noise, wherein the phase noise estimation adopts a block processing method, and the phase noise is estimated by calculating the phase difference between the predicted signal and the target signal within the sample block.

[0130] The decision feedback optimization process includes: performing initial prediction on the test data to obtain the equalization result, applying phase calibration to calculate the calibrated complex signal, obtaining the decision signal through 4-QAM hard decision and updating the training target, performing multiple rounds of iterative optimization to improve the equalization accuracy, realizing adaptive optimization in the test phase, and finally converting the 12-dimensional real vector output by the network into a complex signal output in six modes.

[0131] The iterative training process adopts the Adam optimizer, with the learning rate set to 1e-4, the batch size to 128, and the gradient clipping threshold to 1. The joint optimization of network parameters and phase estimation is achieved through multiple rounds of iteration. The network training adopts a data storage combination method to combine the sliding window data, the phase compensation signal and the target signal, supporting batch parallel processing.

[0132] A CNN-LSTM-based phase calibration method for mode division multiplexing optical fiber communication is proposed. The method is applicable to MDM systems with different numbers of modes. For a 4-mode system, the input dimension is adjusted to 8×(2×window size+1), and the output dimension is 8. For a 12-mode system, the input dimension is adjusted to 24×(2×window size+1), and the output dimension is 24. The number of LSTM hidden units is increased to 512. The phase calibration layer receives the feature vector and phase compensation signal from the LSTM, performs real-time phase calibration on the signal using the phase compensation factor, and outputs the calibrated signal features.

[0133] A CNN-LSTM-based mode division multiplexing optical fiber communication phase calibration system includes: a data preprocessing module for acquiring MIMO input signals and target signals, separating complex signals into real and imaginary parts, and generating sliding window training samples; a network construction module for building a dual-input CNN-LSTM hybrid neural network architecture; a training optimization module for performing iterative training and adaptive phase estimation; a test feedback module for performing testing and decision feedback optimization; and an output processing module for converting network output into complex signal output.

[0134] A CNN-LSTM-based mode division multiplexing optical fiber communication phase calibration system, comprising:

[0135] Data preprocessing module: used for data preprocessing functions of the above execution steps.

[0136] Network building module: A dual-input CNN-LSTM network architecture is constructed to perform the above steps.

[0137] Training and optimization module: used to perform iterative training and phase estimation in the above steps.

[0138] Test feedback module: used to perform testing and decision feedback of the above steps.

[0139] Output processing module: used to perform the complex signal output processing of the above steps.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A CNN-LSTM-based mode division multiplexing optical fiber communication phase calibration method, characterized in that: By constructing a dual-input CNN-LSTM hybrid neural network architecture, the MIMO input signal and original target signal of mode-division multiplexing optical fiber communication are obtained, and data preprocessing is performed. The spatial feature extraction capability of the convolutional neural network and the temporal modeling capability of the long short-term memory network are combined to achieve effective compensation for phase noise in long-distance strongly coupled mode-division multiplexing optical fiber communication. By separating the complex MIMO signal into real and imaginary parts and generating sliding window training samples, a dual-branch network consisting of an image input layer, a convolutional layer, an LSTM layer, a deep connection layer and a custom phase calibration layer is constructed. Through iterative training and a phase feedback mechanism, adaptive estimation and compensation of phase noise are achieved, and decision feedback is used to optimize the test results.

2. A method for phase calibration of mode division multiplexing optical fiber communication based on CNN-LSTM according to claim 1, characterized in that: The data preprocessing process includes: separating the complex signal into real and imaginary parts to form a real and imaginary interleaved data format, setting the sliding window parameters including tap length and phase calculation sample block size, and using the sliding window technique to generate training samples, where the window size is 2×window length+1, and zero padding is performed on the boundaries to ensure the consistent window size of each sample.

3. The method for phase calibration of mode division multiplexing optical fiber communication based on CNN-LSTM according to claim 1, characterized in that: The dual-input CNN-LSTM hybrid neural network architecture includes: the first input branch sets the image input layer to receive sliding window data of the dimension [number of patterns × 2, window size, 1], and then extracts spatial features through a convolution layer with a 3×3 convolution kernel, 12 output channels, and the same padding method, and then outputs a one-dimensional feature vector through a flattening layer; the LSTM feature extraction layer with 256 hidden units and a "last" output mode models the temporal features; the second input branch receives a phase noise compensation signal of the dimension [12, 1, 1]; the deep connection layer deep-joins the LSTM output features and the phase compensation signal; the custom phase calibration layer realizes real-time compensation of phase noise; and the fully connected output layer outputs a 12-dimensional signal corresponding to the real and imaginary parts of the six patterns.

4. The method for phase calibration of mode division multiplexing optical fiber communication based on CNN-LSTM according to claim 1, wherein: The iterative training and phase feedback mechanism includes: initializing the phase noise compensation vector PN to [1,1,1,1,1,1,0,0,0,0,0,0], performing the first round of network training to obtain the initial prediction result, calculating the phase noise based on the predicted complex signal and the target signal, updating the phase noise compensation vector, and repeating the training process to achieve adaptive estimation and compensation of the phase noise. The phase noise estimation adopts a block processing method, and the phase noise is estimated by calculating the phase difference between the predicted signal and the target signal within the sample block.

5. The method for phase calibration of mode division multiplexing optical fiber communication based on CNN-LSTM according to claim 1, characterized in that: The decision feedback optimization process includes: performing initial prediction on the test data to obtain the equalization result, applying phase calibration to calculate the calibrated complex signal, obtaining the decision signal through 4-QAM hard decision and updating the training target, performing multiple rounds of iterative optimization to improve the equalization accuracy, realizing adaptive optimization in the test phase, and finally converting the 12-dimensional real vector output by the network into 6-mode complex signal output.

6. A method for phase calibration of mode division multiplexing optical fiber communication based on CNN-LSTM according to claim 4, characterized in that: The iterative training process uses the Adam optimizer, with a learning rate of 1e-4, a batch size of 128, and a gradient clipping threshold of 1. Multiple rounds of iterations are used to achieve joint optimization of network parameters and phase estimation. Network training uses a data storage combination method to combine sliding window data, phase compensation signals, and target signals to support batch parallel processing.

7. The method for phase calibration of mode division multiplexing optical fiber communication based on CNN-LSTM according to claim 1, characterized in that: Applicable to MDM systems with different numbers of modes: For a 4-mode system, the input dimension is adjusted to 8×(2×window size+1), and the output dimension is 8; for a 12-mode system, the input dimension is adjusted to 24×(2×window size+1), and the output dimension is 24. The number of LSTM hidden units is increased to 512. The phase calibration layer receives the feature vector and phase compensation signal from the LSTM, performs real-time phase calibration on the signal using the phase compensation factor, and outputs the calibrated signal features.

8. A CNN-LSTM-based mode division multiplexing optical fiber communication phase calibration system, characterized in that: include: The data preprocessing module is used to obtain the MIMO input signal and the target signal, separate the complex signal into real and imaginary parts, and generate sliding window training samples; Network building module for building a dual-input CNN-LSTM hybrid neural network architecture; The training optimization module is used to perform iterative training and adaptive phase estimation; the test feedback module is used to perform testing and decision feedback optimization; and the output processing module is used to convert the network output into a complex signal output.

Images