Neural network adaptive modulation method and system for optical communication device
The neural network adaptive modulation method enhances the responsiveness of optical communication systems by using a hybrid neural network model to dynamically adjust modulation parameters in response to environmental changes, improving communication quality and stability.
Patent Information
- Application Number
- CN202411118359.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-08-15
AI Technical Summary
Optical communication systems face challenges in adapting to changing environments due to fixed modulation methods that lack rapid response capabilities, leading to fluctuations in communication quality and poor stability.
A neural network adaptive modulation method using a hybrid model comprising a deep convolutional neural network and long short-term memory network to process multimodal sensing data, extracting time series pattern features and generating adaptive modulation parameters for modulation mode, depth, and coding rate.
Enables dynamic adjustment of modulation parameters in response to environmental changes, enhancing anti-interference ability and communication quality stability, improving resource utilization efficiency, and reducing operation and maintenance costs.
Smart Images

Figure CN118643871B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of optical communication equipment management, and in particular to a neural network adaptive modulation method and system for an optical communication device. Background Art
[0002] Since its inception, optical communication technology has rapidly become the backbone of modern communication networks with its advantages of high bandwidth, low attenuation and anti-electromagnetic interference. However, with the continuous growth of communication needs and the increasingly complex communication environment, the challenges faced by optical communication systems are becoming increasingly severe. In this context, the selection and adaptability of modulation methods are particularly important.
[0003] At present, most optical communication systems use fixed modulation methods, such as Phase Shift Keying (PSK), Amplitude Shift Keying (ASK) and Quadrature Amplitude Modulation (QAM). These modulation methods are usually designed for specific communication environments and requirements, and are difficult to cope with changes in the environment during actual operation. For example, in an environment with less interference, a fixed modulation method may perform well, but in the case of a sudden increase in interference, the communication quality will drop significantly.
[0004] In addition, the operating environment and equipment status of the optical communication system may change at any time, such as changes in ambient temperature, channel differentiation status, differences in environmental interference noise of various optical communication devices or channels, etc. Existing modulation methods lack rapid response capabilities and it is difficult to adjust modulation parameters in time to adapt to these changes, resulting in large fluctuations in communication quality and poor stability.
[0005] In response to the above problems, the industry has not yet proposed a better technical solution. Summary of the invention
[0006] The present application provides a neural network adaptive modulation method, system, storage medium, computer program product and electronic device for an optical communication device, which are used to at least solve the problems of weak anti-interference ability and poor stability of fixed modulation methods in current related technologies.
[0007] In a first aspect, an embodiment of the present application provides a neural network adaptive modulation method for an optical communication device, comprising: acquiring multimodal sensing timing data of a target optical fiber channel in the optical communication device; the multimodal sensing timing data comprises multiple historical time steps and corresponding multidimensional sensing data, and the multidimensional sensing data comprises signal-to-noise ratio, bit error rate, polarization mode dispersion, ambient temperature and ambient interference noise; processing the multimodal sensing timing data based on an automatic encoder to determine a corresponding reconstructed timing feature matrix; inputting the reconstructed timing feature matrix into an adaptive modulation model to determine corresponding target modulation data; the parameter types of the target modulation data comprise modulation mode, modulation depth and coding rate; The method comprises the following steps: the adaptive modulation model adopts a hybrid neural network model, which includes a deep convolutional neural network, a long short-term memory network and a modulation parameter generation module; the deep convolutional neural network is used to extract the timing pattern features corresponding to each time step from the reconstructed timing feature matrix; the timing pattern features define the association patterns between different features that change over time; the long short-term memory network is used to process the reconstructed timing feature matrix and the timing pattern features to obtain a comprehensive feature representation; the modulation parameter generation module is used to process the comprehensive feature representation through a fully connected layer to obtain target modulation data.
[0008] In a second aspect, an embodiment of the present application provides a neural network adaptive modulation system for an optical communication device, comprising: a channel data acquisition unit, used to acquire multimodal sensing time series data of a target optical fiber channel in the optical communication device; the multimodal sensing time series data comprises multiple historical time steps and corresponding multidimensional sensing data, and the multidimensional sensing data comprises signal-to-noise ratio, bit error rate, polarization mode dispersion, ambient temperature and ambient interference noise; a feature extraction unit, used to process the multimodal sensing time series data based on an automatic encoder to determine a corresponding reconstructed time series feature matrix; a modulation data determination unit, used to input the reconstructed time series feature matrix into an adaptive modulation model to determine the corresponding target modulation data; the parameter type of the target modulation data comprises a modulation mode, a modulation depth and an encoding. bit rate; a channel modulation setting unit, used to update the modulation parameter settings for the target optical fiber channel in the optical communication device according to the target modulation data; wherein the adaptive modulation model adopts a hybrid neural network model, which includes a deep convolutional neural network, a long short-term memory network and a modulation parameter generation module; the deep convolutional neural network is used to extract the timing pattern features corresponding to each time step from the reconstructed timing feature matrix; the timing pattern features define the association pattern between different features that changes over time; the long short-term memory network is used to process the reconstructed timing feature matrix and the timing pattern features to obtain a comprehensive feature representation; the modulation parameter generation module is used to process the comprehensive feature representation through a fully connected layer to obtain the target modulation data.
[0009] According to a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the neural network adaptive modulation method of the optical communication device of any embodiment of the present application.
[0010] In a fourth aspect, an embodiment of the present application provides a storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the neural network adaptive modulation method of the optical communication device of any embodiment of the present application are implemented.
[0011] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the neural network adaptive modulation method of the optical communication device of any embodiment of the present application.
[0012] The neural network adaptive modulation method and system of an optical communication device provided by the present application can produce at least the following technical effects:
[0013] (1) By acquiring the multimodal sensing time series data of the optical fiber channel, it is possible to monitor and identify in real time the changes in various environmental factors and equipment status in the optical fiber channel, including signal-to-noise ratio (SNR), bit error rate (BER), polarization mode dispersion (PMD), ambient temperature and environmental interference noise, etc., and then dynamically adjust the modulation mode, modulation depth and coding rate through the adaptive modulation model, thereby realizing the adaptive modulation of the optical fiber channel to different environments and states, enhancing the anti-interference ability and communication quality stability.
[0014] (2) The adaptive modulation model of the hybrid neural network architecture, including a deep convolutional neural network and a long short-term memory network, is used to perform deep learning processing on multimodal sensor time series data. The deep convolutional neural network extracts the time series pattern features, the long short-term memory network processes these features and generates a comprehensive feature representation, and then the modulation parameter generation module determines the target modulation data. In this way, the time-varying correlation patterns between different features in the multidimensional features and the temporal dependencies of each feature can be learned and captured, enabling the system to more accurately identify potential abnormal modes of the channel and adjust the corresponding channel modulation parameters, thereby maintaining high-quality signal transmission in different communication environments and reducing fluctuations in communication quality.
[0015] (3) By integrating the autoencoder and hybrid neural network, the feature analysis and extraction function of the autoencoder is used to efficiently extract and enhance the effective features, and the prediction results of the hybrid neural network are used. The system can adaptively select and adjust the modulation mode, modulation depth and coding rate. It can dynamically select the modulation mode that best suits the current environment based on real-time monitoring data, ensuring that efficient modulation modes can be used to improve data transmission efficiency in low-interference environments, and can switch to more robust modulation modes in high-interference environments to ensure communication reliability.
[0016] (4) For optical communication devices with multiple optical fiber channels, this technical solution can also detect the real-time status of different optical fiber channels and perform adaptive modulation respectively, thereby realizing personalized modulation of different optical fiber channels by the optical communication device.
[0017] Through this technical solution, by introducing a neural network adaptive modulation method, a dynamic response to the changeable channel environment in the optical communication system is achieved, the stability of communication quality and resource utilization efficiency are improved, and operation and maintenance costs are effectively reduced. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0019] Figure 1 A flowchart showing an example of a neural network adaptive modulation method for an optical communication device according to an embodiment of the present application;
[0020] Figure 2 A schematic diagram showing a structural connection of an example of an adaptive modulation model according to an embodiment of the present application is shown;
[0021] Figure 3 An operation flow chart of an example of acquiring multi-modal sensing time series data of a target optical fiber channel according to an embodiment of the present application is shown;
[0022] Figure 4 An operation flow chart of an example of determining a reconstructed time series feature matrix based on an autoencoder according to an embodiment of the present application is shown;
[0023] Figure 5 A schematic diagram of the structural connection of an example of a deep convolutional neural network according to an embodiment of the present application is shown;
[0024] Figure 6 A schematic diagram showing a structural connection of an example of a long short-term memory network based on an attention mechanism according to an embodiment of the present application is shown;
[0025] Figure 7 A schematic diagram showing a structural connection of an example of a modulation parameter generation module according to an embodiment of the present application is shown;
[0026] Figure 8 A structural block diagram showing an example of a neural network adaptive modulation system for an optical communication device according to an embodiment of the present application;
[0027] Fig. 9 It is a schematic structural diagram of an embodiment of an electronic device of the present application. DETAILED DESCRIPTION
[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0029] In the technical solution of this application, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved shall comply with the provisions of relevant laws and regulations and shall not violate public order and good morals.
[0030] Figure 1 A flowchart showing an example of a neural network adaptive modulation method for an optical communication device according to an embodiment of the present application is shown.
[0031] Regarding the executor of the method of the embodiment of the present application, it can be any controller or processor with computing or processing capabilities. By introducing sampled multimodal data and using advanced deep learning technologies such as autoencoders, deep convolutional neural networks, and long short-term memory networks, the system can complete the analysis and decision-making process of sensor data at a faster speed, so that when the communication environment changes, the modulation parameters can be adjusted quickly to improve the system's response speed to environmental changes.
[0032] In some examples, it can be a channel adaptive modulation platform for an optical communication device, and can be integrated and configured in an electronic device, terminal or server by means of software, hardware or a combination of software and hardware, and the types of terminals, electronic devices or servers can be diverse, such as mobile phones, tablet computers or desktop computers, etc.
[0033] like Figure 1 As shown, in step S110, multi-modal sensing timing data of a target optical fiber channel in an optical communication device is acquired.
[0034] Here, the multimodal sensing time series data includes multiple historical time steps and corresponding multidimensional sensing data, and the multidimensional sensing data includes signal-to-noise ratio, bit error rate, polarization mode dispersion, ambient temperature, and ambient interference noise. In some embodiments, multiple sensors are arranged on the optical fiber channel of the optical communication device to monitor and record the multidimensional sensing data in real time, and combined with the working parameters regularly collected by the data acquisition module, the multimodal sensing time series data is formed. By using multimodal sensing time series data, the system can effectively capture the dynamic changes of the working state of the optical fiber channel and the external environmental parameters, providing higher adaptability than the traditional fixed modulation method.
[0035] In step S120, the multimodal sensing time series data is processed based on the autoencoder to determine the corresponding reconstructed time series feature matrix.
[0036] Here, the collected multimodal sensor time series data is reduced in dimension and features are extracted through an autoencoder. Specifically, the autoencoder consists of two parts: an encoder and a decoder. The encoder part maps the high-dimensional input data to a low-dimensional latent space to remove redundant information and extract the main features, and the decoder part reconstructs the original data from the low-dimensional latent space. Here, the autoencoder can be trained with a preset data sample set, so that the system can learn how to effectively represent the input data and generate a reconstructed time series feature matrix.
[0037] Through this embodiment, the automatic encoder can effectively reduce the dimension of multimodal data, remove redundant information, retain the most critical features, and reconstruct the time series feature matrix containing the core information of the multimodal data, thereby reducing the computational complexity of subsequent models and facilitating accurate analysis and prediction of the state of the optical fiber channel.
[0038] In step S130, the reconstructed timing characteristic matrix is input into an adaptive modulation model to determine corresponding target modulation data.
[0039] It should be noted that the adaptive modulation model can adopt various machine learning models, so that it can adaptively predict the corresponding modulation parameter recommendations based on the channel state, which include modulation mode, modulation depth and coding rate. In some examples of the embodiments of the present application, the adaptive modulation model adopts a hybrid neural network model, and more details will be expanded in conjunction with other examples below.
[0040] In step S140, the modulation parameter settings for the target optical fiber channel in the optical communication device are updated according to the target modulation data.
[0041] In some embodiments, the modulation parameter settings of the corresponding optical fiber channel are automatically updated according to the target modulation data generated by the adaptive modulation model. Specifically, the modulation mode for this optical fiber channel is switched, such as switching from PSK to QAM; or the modulation depth is adjusted, such as changing the size or shape of the modulation constellation; or the coding rate is adjusted to optimize the balance between the transmission rate and the bit error rate. More preferably, before the system updates the modulation parameters, a one-time verification or simulation test can also be performed to ensure the effectiveness and stability of the new parameter settings in the current environment.
[0042] Figure 2 A schematic diagram of the structural connection of an example of an adaptive modulation model according to an embodiment of the present application is shown.
[0043] like Figure 2 As shown, the adaptive modulation model 200 includes a deep convolutional neural network 210, a long short-term memory network 220 and a modulation parameter generation module 230.
[0044] The deep convolutional neural network 210 is used to extract the time series pattern features corresponding to each time step from the reconstructed time series feature matrix, and the time series pattern features define the correlation pattern between different features that changes over time.
[0045] Specifically, the deep convolutional neural network 210 may include multiple convolutional layers and pooling layers to gradually extract more abstract and advanced temporal pattern features, and the output feature map of each layer is used as the input of the next layer. Exemplarily, the local features in the input reconstructed temporal feature matrix are extracted by the convolutional layer, and the local pattern of the data can be detected by using the sliding window operation. The spatial dimension of the feature map is reduced by the pooling layer while retaining important information, such as MaxPooling or AveragePooling.
[0046] It should be noted that deep convolutional neural networks can efficiently extract local time series features in input data, and are particularly suitable for processing input data with spatial locality and time dependence. In this embodiment, the relationship between different features is used to define spatial features. Specifically, the input data does not have a traditional spatial structure, but the relationship between different features (such as the correlation between signal-to-noise ratio and bit error rate) and the dependency between time steps can be regarded as a certain "structure" or "pattern". In this way, by capturing the complex relationships and patterns between these features through deep convolutional neural networks, significant patterns and anomalies can be identified in the feature data, such as short-term burst noise, high bit error rate events, etc., or long-term trend changes, such as continuous channel attenuation, etc.
[0047] The long short-term memory network 220 is used to process and reconstruct the time series feature matrix and the time series pattern features to obtain a comprehensive feature representation.
[0048] Here, the Long Short-Term Memory (LSTM) network 220 may include multiple layers of LSTM units. Specifically, the LSTM unit includes an input gate, a forget gate, an output gate, and a memory unit, and can effectively process and store long-range dependencies in time series data. In addition, the input gate controls the input of new information, the forget gate determines which old information needs to be discarded, and the output gate determines the output of the current time step. By adopting a multi-layer LSTM structure, such as a multi-layer LSTM stacked structure, that is, the output of each layer of LSTM is used as the input of the next layer, the expression ability of the model can be effectively enhanced.
[0049] Through this embodiment, based on the characteristics of the long short-term memory network that is good at capturing long-range dependencies in time series data, the model can remember and utilize key information within a long time span, thereby improving the prediction accuracy of the modulation parameters. In addition, by comprehensively processing the reconstructed time series feature matrix and the time series pattern features extracted by the deep convolutional neural network, the long short-term memory network can integrate the time series information and generate a more comprehensive feature representation.
[0050] The modulation parameter generation module 230 is used to process the comprehensive feature representation through a fully connected layer to obtain target modulation data.
[0051] In some embodiments, the modulation parameter generation module may include multiple fully connected layers to predict different types of modulation parameters, such as modulation mode, modulation depth, and coding rate, and may use a multi-head output design so that each output node corresponds to a different modulation parameter. Here, in response to the needs of multi-task learning, a weighted loss function may be used to jointly train different output tasks. For example, for the modulation mode, a classification cross entropy loss function may be used, and for the modulation depth and coding rate, a mean square error loss function may be used.
[0052] Through this embodiment, the modulation parameter generation module can simultaneously predict multiple modulation parameters to meet different communication requirements and channel states. Combining the feature representation capabilities of deep convolutional neural networks and long short-term memory networks, the modulation parameter generation module can accurately generate modulation parameters that adapt to the current channel state and optimize the performance of the communication link.
[0053] Figure 3 An operational flow chart of an example of acquiring multi-modal sensing time series data of a target optical fiber channel according to an embodiment of the present application is shown.
[0054] like Figure 3 As shown, in step S310, each optical fiber channel in the optical communication device is polled according to a preset polling cycle to determine a target optical fiber channel.
[0055] In some embodiments, the optical communication device periodically polls each optical fiber channel to trigger sampling of the timing status of each optical fiber channel, and the polling frequency can be set or adjusted according to business needs or channel conditions. Specifically, a reasonable polling interval can be set according to the specific application scenario and the dynamic change characteristics of the channel to balance real-time performance and system load. Furthermore, the optical communication device can sample the multimodal sensing timing data of each optical fiber channel in turn, and obtain a targeted modulation scheme for the target optical fiber channel by calling (for example, remotely calling) the autoencoder and the adaptive modulation model, thereby realizing personalized modulation management for different channels in the optical communication device.
[0056] In step S320, first sensing time series data of the target optical fiber channel is collected according to the first sampling frequency, and the parameter types of the first sensing time series data include signal-to-noise ratio, bit error rate and polarization mode dispersion.
[0057] In step S330, second sensing target optical fiber data of the target optical fiber channel is sampled according to the second sampling frequency, the parameter types of the second sensing time series data include ambient temperature and ambient interference noise, and the first sampling frequency is greater than the second sampling frequency.
[0058] In step S340, multimodal sensing time series data is determined according to the first sensing time series data and the second sensing time series data.
[0059] In some embodiments, for parameters with fast dynamic changes such as signal-to-noise ratio, bit error rate and polarization mode dispersion, high-frequency acquisition is used to accurately capture the instantaneous changes in channel status, such as several times per second, to avoid missing critical burst signals. For relatively stable parameters such as ambient temperature and ambient noise, low-frequency acquisition can be used, such as once per minute, to avoid data redundancy and system load caused by unnecessary frequent acquisition. In addition, in order to deal with the imbalance of time series data, the information of the vacant time steps of ambient temperature and ambient noise can be filled and improved by interpolation to achieve data information alignment, and then multi-modal sensing time series data can be obtained through data connection.
[0060] Through this embodiment, different sampling frequencies are set according to the characteristics of the sampled data, so that the system can use resources more effectively, and while ensuring the accuracy of data collection, the data processing efficiency and system response speed are significantly improved.
[0061] Figure 4 An operational flow chart of an example of determining a reconstructed time series feature matrix based on an autoencoder according to an embodiment of the present application is shown.
[0062] like Figure 4 As shown, in step S410, the signal-to-noise ratio, bit error rate and environmental interference noise in the multimodal sensing time series data are processed by Min-Max normalization to obtain corresponding Min-Max normalized time series features.
[0063] Here, the features with a large numerical range, such as the signal-to-noise ratio, bit error rate, and environmental interference noise, are processed by the Min-Max normalization method. All data are scaled to the range of [0, 1], which helps to eliminate the dimensional differences between different features.
[0064] In step S420, the polarization mode dispersion and the ambient temperature in the multimodal sensing time series data are processed by Z-score standardization to obtain corresponding Z-score standard time series features.
[0065] Here, the features with significant differences in distribution, such as polarization mode dispersion and ambient temperature, are processed by Z-score normalization. In this way, after standardization, the data will have a standard normal distribution with a mean of 0 and a standard deviation of 1, which helps the neural network converge faster and reduces the impact of numerical differences between features.
[0066] In step S430, a fused standard time series feature is determined according to the Min-Max normalized time series feature and the Z-score standard time series feature.
[0067] Here, the information in the Min-Max normalized time series features and the Z-score standard time series features can be aligned and connected according to the time step to obtain the corresponding fused standard time series features.
[0068] Through this embodiment, the corresponding types of data are processed respectively by using Min-Max normalization and Z-score standardization, so that the system can eliminate the dimensional differences between different features, so that the data can be processed and analyzed on the same scale, and the comparability between features is improved. In addition, corresponding standard preprocessing methods are used for different types of data. The features such as signal-to-noise ratio and bit error rate processed by Min-Max normalization reflect the changes in signal quality and noise environment, while the features such as polarization mode dispersion and ambient temperature processed by Z-score standardization reflect the changes in the physical environment.
[0069] In step S440, the encoder based on the autoencoder compresses the fused standard time series features into a low-dimensional space, and the decoder based on the autoencoder reconstructs the original data from the low-dimensional space to determine the corresponding reconstructed time series feature matrix.
[0070] Through this embodiment, an automatic encoder is used for feature extraction. The high-dimensional input data is mapped to a low-dimensional space through the encoder, and important information of the original data is retained as much as possible. This not only achieves data dimensionality reduction and reduces the redundancy of data dimensions, but also captures the nonlinear relationship in the data to extract the most useful features for adaptive modulation decisions.
[0071] In some examples of the embodiments of the present application, the loss function of the autoencoder is for:
[0072] , formula (1);
[0073] , formula (2);
[0074] , formula (3);
[0075] , formula (4);
[0076] In the formula, They represent the original reconstruction error loss term, the temporal smoothness constraint loss term, and the feature importance constraint loss term, respectively. Respectively represent the weight coefficients of each corresponding loss term; Respectively represent Min-Max normalized data of samples and their reconstructed data, Respectively represent The Z-score standardized data and its reconstructed data of samples, is the total number of samples in the data sample set of the autoencoder; Respectively represent and Min-Max normalized reconstruction data of time steps, Respectively represent and The Z-score normalized reconstructed data of time steps, Indicates the total number of time steps corresponding to the sample; Represents a set of key feature indexes, where the key features include signal-to-noise ratio and bit error rate; Indicates The weight coefficient of each key feature; Indicates The sample The Min-Max normalized value of the key features, Indicates The sample Min-Max normalized reconstruction value of key features, Indicates The sample The Z-score standardized value of the key features, Indicates The sample The Z-score standardized reconstruction value of the key features.
[0077] In an embodiment of the present application, the autoencoder adopts a comprehensive loss function based on the original reconstruction error loss term (mean square error), the temporal smoothness constraint loss term and the feature importance constraint loss term. The difference between the original data and the reconstructed data is measured through the original reconstruction error loss term, which serves as part of the basic loss function of the autoencoder. Through the temporal smoothness constraint loss term, it is ensured that the data of adjacent time steps maintain a certain smoothness after reconstruction to avoid abnormal mutations caused by reconstruction errors. Through the feature importance constraint loss term, by giving higher weights to specific key features, the importance of these features in the reconstruction process is enhanced. By combining the above-mentioned loss terms in a weighted summation manner, the model can not only effectively reconstruct the input data, but also pay special attention to the reconstruction quality of key features, ensuring accuracy and robustness in multimodal data processing.
[0078] Constrain the loss term by feature importance , the autoencoder gives higher reconstruction accuracy to key features (such as signal-to-noise ratio and bit error rate). These features, as core parameters for evaluating the quality of communication channels, have a direct impact on the optimization of modulation parameters. Therefore, by improving the reconstruction accuracy of these features, it is possible to ensure that the model accurately grasps the state of the communication channel, thereby making more accurate modulation decisions, which helps to improve the overall quality and reliability of communication. , ensuring the smooth transition of reconstructed data between adjacent time steps, avoiding sudden or abnormal changes, so that the system can better track the dynamic changes of channel states, adjust modulation parameters in time, and ensure the stability and response speed of the system in a changing environment. Therefore, by focusing on the accurate reconstruction of key features and the smoothness of time series data, the loss function helps the system enhance its resistance to noise and abnormal data, enabling the system to more accurately identify and respond to changes in the channel and improve its adaptability to a changing environment.
[0079] In some examples of the embodiments of the present application, the deep convolutional neural network in the adaptive modulation model adopts a multi-scale deep convolutional neural network. Figure 5 A schematic diagram of the structural connection of an example of a deep convolutional neural network according to an embodiment of the present application is shown.
[0080] like Figure 5 As shown, the multi-scale deep convolutional neural network 500 includes a scale fusion layer 510 and multiple scale convolution blocks (521, 522...52M). Each scale convolution block includes a cascaded convolution layer (5211, 5221...52M1) and a pooling layer (5212, 5222...52M2), and each scale convolution block has a convolution kernel of a corresponding scale.
[0081] Specifically, in each scale convolution block, the convolution layer is used to extract the feature correlation information of the time period corresponding to the convolution kernel scale, and obtain the corresponding scale time series feature pattern through pooling operation. The first scale convolution block is used to capture short-term time series feature patterns, and the second scale convolution block is used to capture long-term time series feature patterns. The convolution kernel length of the first scale convolution block is smaller than the convolution kernel length of the second scale convolution block. In other words, through a small convolution kernel, the interaction of different features in a short time can be captured, such as the impact of burst interference on the signal-to-noise ratio and bit error rate; through a large convolution kernel, the overall trend and changes in a longer time period can be captured, such as the impact of long-term changes in environmental noise levels on the system.
[0082] More specifically, each scale convolutional block is used to perform the following operations:
[0083] , formula (5);
[0084] , formula (6);
[0085] The scale fusion layer is used to fuse the scale time series feature patterns output by each scale convolution block to determine a multi-scale time series feature pattern:
[0086] , formula (7);
[0087] In the formula, Indicates the reconstructed time series feature matrix corresponding to the time step arrive The information fragments between Indicates The convolution kernel is for The output characteristics of and Respectively represent The weights and biases of the convolution kernels, Indicates The length of the convolution kernel, represents the convolution operation, represents the RELU activation function, represents the maximum pooling function, Express Features after maximum pooling processing; Indicated in The multi-scale time series feature pattern of time steps, represents the feature concatenation operation, Represents the total number of convolution kernels.
[0088] Through this embodiment, using convolution kernels of different lengths, the multi-scale deep convolutional neural network can extract multi-time scale features from multi-modal time series data, short convolution kernels can capture short-term features, and long convolution kernels can identify long-term features. Thus, through multi-scale feature extraction, the system can fully understand the channel state and environmental changes, ensure that short-term and long-term influencing factors can be considered when making modulation parameter decisions, and improve the accuracy and adaptability of modulation. In addition, through the multi-level convolution structure of the multi-scale deep convolutional neural network, complex patterns in multi-modal data can be effectively identified and separated, including the relationship and dependence between multiple features. These patterns generally reflect information in multiple aspects such as signal quality, interference type and environmental changes. Thus, by identifying these complex patterns, the system can more accurately evaluate the characteristics and potential risks of the current communication environment, thereby optimizing the modulation strategy. For example, when a specific interference pattern is detected, the system can automatically adjust the modulation mode to reduce the bit error rate.
[0089] In some examples of the embodiments of the present application, the long short-term memory network in the adaptive modulation model can adopt a long short-term memory network based on the attention mechanism. Figure 6 A schematic diagram of the structural connection of an example of a long short-term memory network based on the attention mechanism according to an embodiment of the present application is shown.
[0090] like Figure 6 As shown, the long short-term memory network 600 based on the attention mechanism includes a cascaded LSTM layer 610 and an attention mechanism layer 620.
[0091] The time-dependent features of the reconstructed time-series feature matrix and the multi-scale time-series feature pattern of each time step are extracted through the LSTM layer 610 to update the hidden state of each time step. The attention weight of each time step is calculated through the attention mechanism layer 620, and the hidden states of all time steps are weighted to generate a weighted comprehensive feature representation.
[0092] More specifically, it calculates the weighted comprehensive feature representation in the following way:
[0093] , formula (8);
[0094] , formula (9);
[0095] , formula (10);
[0096] , formula (11);
[0097] In the formula, Respectively represent the LSTM layer in hidden state and cell state at time steps, Respectively represent the LSTM layer in Hidden state and cell state at time steps; Indicates The attention score of time steps, Represent the weight matrix and bias term of the attention mechanism layer respectively, Indicates The attention weights for time steps, represents the total number of time steps, Indicates The attention score of time steps, represents the tanh activation function, represents the sum of the attention scores of all time steps after exponential operation, represents weighted comprehensive feature representation.
[0098] In optical communication systems, changes in channel states and external environments often have temporal continuity and dependency. The LSTM layer can effectively process and memorize relevant information in time series, which is particularly suitable for processing various sensor parameters with timing dependency. Through the attention mechanism layer, a weight is assigned to each time step. According to the needs of the current task, the model's attention to different time steps is dynamically adjusted, so that the network can dynamically adjust the attention weight of the input features. The weighted sum highlights the most important time step features for the current task, which can ensure that the model gives priority to the information that is most critical to the modulation decision when dealing with complex and changeable optical communication environments, thereby improving the model's discrimination ability and decision quality. In addition, by observing the distribution of attention weights, users can understand which time steps and features the model focuses on when making decisions, achieve the interpretability of feature reasoning, and help operation and maintenance personnel discover potential problems and improvement points in the system, and improve their understanding and trust in system behavior.
[0099] In some examples of the embodiments of the present application, the modulation parameter generation module adopts a modulation parameter generation module based on adaptive weight adjustment. Figure 7 A schematic diagram of the structural connection of an example of a modulation parameter generation module according to an embodiment of the present application is shown.
[0100] like Figure 7 As shown, the modulation parameter generation module 700 adopts a modulation parameter generation module based on adaptive weight adjustment, which includes an adaptive weight update layer 710, and a classification fully connected layer 720, a first regression fully connected layer 730 and a second regression fully connected layer 740 respectively connected to the output of the adaptive weight update layer 710.
[0101] The adaptive weight updating layer 710 is used to determine the adaptive weights corresponding to various modulation parameters:
[0102] , formula (12);
[0103] In the formula, Indicates modulation parameters Adaptive weights of Representation and Modulation Parameters The associated weight adjustment coefficient vectors, different types of modulation parameters have unique corresponding weight adjustment coefficient vectors; Represents a collection of all types of modulation parameters, including modulation mode, modulation depth and coding rate; is the sum of the exponentially weighted scores of all types of modulation parameters, which is represented in the weighted comprehensive feature representation The total weight of all modulation parameters.
[0104] It should be noted that Determines the importance of the current specific modulation parameters (such as modulation mode, modulation depth or coding rate) in the final output. The higher the weight, the greater the influence of the corresponding type of modulation parameters in the system modulation process. In addition, is a feature representation The vector of the same dimension is used to calculate the weight of the modulation parameter, which can be optimized during the model training process. For example, during the model training process, the weight adjustment coefficient vector As part of the model parameters, it is automatically learned through the back propagation algorithm and gradient descent method. When the model is trained, it will represent the input data according to its characteristics. and its corresponding target modulation parameters, and continuously adjust to minimize the loss function.
[0105] In optical communication systems, the impact of different modulation parameters on system performance may vary due to environmental changes. Here, through the adaptive weight adjustment mechanism, the system can flexibly adjust the priority and weight of modulation parameters in different situations to ensure the optimization of communication quality.
[0106] The classification fully connected layer 720 is used to output the modulation mode in the target modulation data:
[0107] Formula (13);
[0108] In the formula, Indicates the adaptive weight corresponding to the modulation mode, Indicates the modulation mode in the output target modulation data; and Represent the weight matrix and bias vector of the classification fully connected layer respectively; Represents the softmax activation function.
[0109] The first regression fully connected layer 730 is used to output the modulation depth in the target modulation data:
[0110] Formula (14);
[0111] In the formula, represents the adaptive weight corresponding to the modulation depth, Indicates the modulation depth in the output target modulation data; and Represent the weight matrix and bias vector of the first regression fully connected layer respectively.
[0112] The second recurrent fully connected layer 740 is used to output the coding rate in the target modulation data:
[0113] Formula (15);
[0114] In the formula, represents the adaptive weight corresponding to the coding rate, Indicates the coding rate in the output target modulation data; and They represent the weight matrix and bias term of the second regression fully connected layer respectively.
[0115] The adaptive weight adjustment mechanism in the embodiment of the present application allows the model to represent the current comprehensive features. Dynamically adjust the weight of each modulation parameter (such as modulation mode, modulation depth, coding rate) so that the model can adapt to different channel states and external environment changes. As a result, the system can automatically adjust the modulation strategy according to the real-time channel conditions. For example, in a high-noise environment, the system can automatically increase the weight of the modulation depth to enhance the signal strength; while under stable channel conditions, it may be more inclined to increase the coding rate to increase data throughput. Therefore, through the selection and modulation of adaptive modulation parameter types, the system can always maintain the best communication performance in a complex and changing environment.
[0116] Through this embodiment, the modulation parameter generation module based on the adaptive weight adjustment mechanism can dynamically weight the influence of different features, thereby comprehensively considering the weights of multiple factors in the prediction of the modulation parameters, making the generated modulation parameters more accurate and targeted, and able to better match the requirements of the current channel state, achieve the reliability of the modulation parameters, and ensure the high quality and stability of channel communication.
[0117] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of actions combined, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.
[0118] Figure 8 A structural block diagram of an example of a neural network adaptive modulation system for an optical communication device according to an embodiment of the present application is shown.
[0119] like Figure 8 As shown, the neural network adaptive modulation system 800 of the optical communication device includes a channel data acquisition unit 810, a feature extraction unit 820, a modulation data determination unit 830 and a channel modulation setting unit 840.
[0120] The channel data acquisition unit 810 is used to acquire multimodal sensing timing data of a target optical fiber channel in an optical communication device; the multimodal sensing timing data includes multiple historical time steps and corresponding multidimensional sensing data, and the multidimensional sensing data includes signal-to-noise ratio, bit error rate, polarization mode dispersion, ambient temperature and ambient interference noise.
[0121] The feature extraction unit 820 is used to process the multimodal sensing time series data based on an autoencoder to determine a corresponding reconstructed time series feature matrix.
[0122] The modulation data determination unit 830 is used to input the reconstructed timing feature matrix into the adaptive modulation model to determine the corresponding target modulation data; the parameter types of the target modulation data include modulation mode, modulation depth and coding rate.
[0123] The channel modulation setting unit 840 is used to update the modulation parameter setting for the target optical fiber channel in the optical communication device according to the target modulation data.
[0124] The adaptive modulation model adopts a hybrid neural network model, which includes a deep convolutional neural network, a long short-term memory network and a modulation parameter generation module;
[0125] The deep convolutional neural network is used to extract the time series pattern features corresponding to each time step from the reconstructed time series feature matrix; the time series pattern features define the association pattern between different features that changes over time;
[0126] The long short-term memory network is used to process the reconstructed time series feature matrix and the time series pattern features to obtain a comprehensive feature representation;
[0127] The modulation parameter generation module is used to process the comprehensive feature representation through a fully connected layer to obtain target modulation data.
[0128] In some embodiments, an embodiment of the present application provides a non-volatile computer-readable storage medium, in which one or more programs including execution instructions are stored, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute the steps of the neural network adaptive modulation method of any of the above-mentioned optical communication devices of the present application.
[0129] In some embodiments, the embodiments of the present application also provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the steps of any of the above-mentioned neural network adaptive modulation methods for optical communication devices.
[0130] In some embodiments, the embodiments of the present application also provide an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the neural network adaptive modulation method of the optical communication device.
[0131] Fig. 9 is a schematic diagram of the hardware structure of an electronic device for executing a neural network adaptive modulation method for an optical communication device provided by another embodiment of the present application, such as Fig. 9 As shown, the device includes:
[0132] One or more processors 910 and memory 920, Fig. 9 A processor 910 is taken as an example.
[0133] The device for executing the neural network adaptive modulation method of the optical communication device may further include: an input device 930 and an output device 940 .
[0134] The processor 910, the memory 920, the input device 930 and the output device 940 may be connected via a bus or other means. Fig. 9 The example of connecting through bus is taken in the following.
[0135] The memory 920, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the neural network adaptive modulation method of the optical communication device in the embodiment of the present application. The processor 910 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 920, that is, the neural network adaptive modulation method of the optical communication device in the above method embodiment is implemented.
[0136] The memory 920 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 920 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 920 may optionally include a memory remotely arranged relative to the processor 910, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0137] The input device 930 may receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 940 may include a display device such as a display screen.
[0138] The one or more modules are stored in the memory 920, and when executed by the one or more processors 910, the neural network adaptive modulation method of the optical communication device in any of the above method embodiments is executed.
[0139] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.
[0140] The electronic devices of the embodiments of the present application exist in various forms, including but not limited to:
[0141] (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.
[0142] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, etc.
[0143] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.
[0144] (4) Other onboard electronic devices with data interaction functions, such as on-board devices installed in vehicles.
[0145] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0146] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the relevant technology, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiment.
[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A neural network adaptive modulation method for an optical communication device, comprising: Acquiring multi-modal sensing time series data of a target optical fiber channel in an optical communication device; The multimodal sensing time series data includes multiple historical time steps and corresponding multidimensional sensing data, and the multidimensional sensing data includes signal-to-noise ratio, bit error rate, polarization mode dispersion, ambient temperature and ambient interference noise; Processing the multimodal sensor time series data based on an autoencoder to determine a corresponding reconstructed time series feature matrix; Inputting the reconstructed timing characteristic matrix into an adaptive modulation model to determine corresponding target modulation data; The parameter types of the target modulation data include modulation mode, modulation depth and coding rate; updating, according to the target modulation data, a modulation parameter setting for the target optical fiber channel in the optical communication device; The adaptive modulation model adopts a hybrid neural network model, which includes a deep convolutional neural network, a long short-term memory network and a modulation parameter generation module; The deep convolutional neural network is used to extract the time series pattern features corresponding to each time step from the reconstructed time series feature matrix; the time series pattern features define the association pattern between different features that changes over time; The long short-term memory network is used to process the reconstructed time series feature matrix and the time series pattern features to obtain a comprehensive feature representation; The modulation parameter generation module is used to process the comprehensive feature representation through a fully connected layer to obtain target modulation data; Wherein, the processing of the multimodal sensing time series data based on the autoencoder to determine the corresponding reconstructed time series feature matrix includes: The signal-to-noise ratio, bit error rate and environmental interference noise in the multimodal sensing time series data are processed by Min-Max normalization to obtain corresponding Min-Max normalized time series features; The polarization mode dispersion and the ambient temperature in the multimodal sensing time series data are processed by Z-score standardization to obtain corresponding Z-score standard time series features; Determine a fusion standard time series feature according to the Min-Max normalized time series feature and the Z-score standard time series feature; An encoder based on an autoencoder compresses the fused standard time series features into a low-dimensional space, and a decoder based on the autoencoder reconstructs the original data from the low-dimensional space to determine a corresponding reconstructed time series feature matrix; The loss function of the autoencoder is: In the formula, They represent the original reconstruction error loss term, the temporal smoothness constraint loss term, and the feature importance constraint loss term, respectively. Respectively represent the weight coefficients of each corresponding loss term; Respectively represent Min-Max normalized data of samples and their reconstructed data, Respectively represent The Z-score standardized data and its reconstructed data of samples, is the total number of samples in the data sample set of the autoencoder; Respectively represent and Min-Max normalized reconstruction data of time steps, Respectively represent and The Z-score normalized reconstructed data of time steps, Indicates the total number of time steps corresponding to the sample; Represents a set of key feature indexes, where the key features include signal-to-noise ratio and bit error rate; Indicates The weight coefficient of each key feature; Indicates The sample The Min-Max normalized value of the key features, Indicates The sample Min-Max normalized reconstruction value of key features, Indicates The sample The Z-score standardized value of the key features, Indicates The sample The Z-score standardized reconstruction value of the key features.
2. The method according to claim 1, wherein: The method of acquiring multi-modal sensing time series data of a target optical fiber channel in an optical communication device comprises: Polling each optical fiber channel in the optical communication device according to a preset polling cycle to determine a target optical fiber channel; The multi-modal sensing time series data of the target optical fiber channel is sampled.
3. The method according to claim 2, wherein: The sampling of the multi-modal sensing time series data of the target optical fiber channel includes: According to a first sampling frequency, first sensing time series data of the target optical fiber channel is collected; parameter types of the first sensing time series data include signal-to-noise ratio, bit error rate and polarization mode dispersion; According to a second sampling frequency, sampling second sensing time series data of the target optical fiber channel; the parameter types of the second sensing time series data include ambient temperature and ambient interference noise; the first sampling frequency is greater than the second sampling frequency; Multimodal sensing time series data is determined according to the first sensing time series data and the second sensing time series data.
4. The method according to any one of claims 1 to 3, wherein: The deep convolutional neural network in the adaptive modulation model adopts a multi-scale deep convolutional neural network; The multi-scale deep convolutional neural network includes a scale fusion layer and multiple scale convolution blocks, each of the scale convolution blocks includes a cascaded convolution layer and a pooling layer, and each of the scale convolution blocks has a convolution kernel of a corresponding scale; in each scale convolution block, the convolution layer is used to extract feature association information of a time period corresponding to the convolution kernel scale, and obtain a corresponding scale time series feature pattern through a pooling operation; the first scale convolution block is used to capture a short-term time series feature pattern, and the second scale convolution block is used to capture a long-term time series feature pattern, and the convolution kernel length of the first scale convolution block is less than the convolution kernel length of the second scale convolution block: The scale fusion layer is used to fuse the scale time series feature patterns output by each scale convolution block to determine a multi-scale time series feature pattern: In the formula, Indicates the reconstructed time series feature matrix corresponding to the time step arrive The information fragments between Indicates The convolution kernel is for The output characteristics of and Respectively represent The weights and biases of the convolution kernels, Indicates The length of the convolution kernel, represents the convolution operation, represents the RELU activation function, represents the maximum pooling function, Express Features after maximum pooling processing; Indicated in The multi-scale time series feature pattern of time steps, represents the feature concatenation operation, Represents the total number of convolution kernels.
5. The method according to any one of claims 1 to 3, wherein: The long short-term memory network adopts a long short-term memory network based on an attention mechanism, which includes an LSTM layer and an attention mechanism layer; The LSTM layer is used to extract and reconstruct the time series feature matrix and the time series dependency features of the multi-scale time series feature pattern of each time step to update the hidden state of each time step; the attention weight of each time step is calculated through the attention mechanism layer, and the hidden states of all time steps are weighted to generate a weighted comprehensive feature representation: In the formula, Respectively represent the LSTM layer in hidden state and cell state at time steps, Respectively represent the LSTM layer in Hidden state and cell state at time steps; Indicates The attention score of time steps, Represent the weight matrix and bias term of the attention mechanism layer respectively, Indicates The attention weights for time steps, represents the total number of time steps, Indicates The attention score of time steps, represents the tanh activation function, represents the sum of the attention scores of all time steps after exponential operation, represents weighted comprehensive feature representation.
6. The method according to claim 5, wherein: The modulation parameter generation module adopts a modulation parameter generation module based on adaptive weight adjustment, which includes an adaptive weight update layer, a classification fully connected layer, a first regression fully connected layer, and a second regression fully connected layer; The adaptive weight update layer is used to determine the adaptive weights corresponding to various modulation parameters: In the formula, Indicates modulation parameters Adaptive weights of Representation and Modulation Parameters The associated weight adjustment coefficient vectors, different types of modulation parameters have unique corresponding weight adjustment coefficient vectors; Represents a collection of all types of modulation parameters, including modulation mode, modulation depth and coding rate; is the sum of the exponentially weighted scores of all types of modulation parameters, which is represented in the weighted comprehensive feature representation The total weight of all modulation parameters under The classification fully connected layer is used to output the modulation mode in the target modulation data: In the formula, Indicates the adaptive weight corresponding to the modulation mode, Indicates the modulation mode in the output target modulation data; and Represent the weight matrix and bias vector of the classification fully connected layer respectively; represents the softmax activation function; The first regression fully connected layer is used to output the modulation depth in the target modulation data: In the formula, represents the adaptive weight corresponding to the modulation depth, Indicates the modulation depth in the output target modulation data; and Represent the weight matrix and bias vector of the first regression fully connected layer respectively; The second regressive fully connected layer is used to output the coding rate in the target modulation data: In the formula, represents the adaptive weight corresponding to the coding rate, Indicates the coding rate in the output target modulation data; and They represent the weight matrix and bias term of the second regression fully connected layer respectively.
7. A neural network adaptive modulation system for an optical communication device, comprising: A channel data acquisition unit, used to acquire multi-modal sensing time series data of a target optical fiber channel in an optical communication device; The multimodal sensing time series data includes multiple historical time steps and corresponding multidimensional sensing data, and the multidimensional sensing data includes signal-to-noise ratio, bit error rate, polarization mode dispersion, ambient temperature and ambient interference noise; A feature extraction unit, configured to process the multimodal sensing time series data based on an autoencoder to determine a corresponding reconstructed time series feature matrix; A modulation data determination unit, used for inputting the reconstructed timing characteristic matrix into an adaptive modulation model to determine corresponding target modulation data; The parameter types of the target modulation data include modulation mode, modulation depth and coding rate; a channel modulation setting unit, configured to update the modulation parameter setting for the target optical fiber channel in the optical communication device according to the target modulation data; The adaptive modulation model adopts a hybrid neural network model, which includes a deep convolutional neural network, a long short-term memory network and a modulation parameter generation module; The deep convolutional neural network is used to extract the time series pattern features corresponding to each time step from the reconstructed time series feature matrix; the time series pattern features define the association pattern between different features that changes over time; The long short-term memory network is used to process the reconstructed time series feature matrix and the time series pattern features to obtain a comprehensive feature representation; The modulation parameter generation module is used to process the comprehensive feature representation through a fully connected layer to obtain target modulation data; Wherein, the processing of the multimodal sensing time series data based on the autoencoder to determine the corresponding reconstructed time series feature matrix includes: The signal-to-noise ratio, bit error rate and environmental interference noise in the multimodal sensing time series data are processed by Min-Max normalization to obtain corresponding Min-Max normalized time series features; The polarization mode dispersion and the ambient temperature in the multimodal sensing time series data are processed by Z-score standardization to obtain corresponding Z-score standard time series features; Determine a fusion standard time series feature according to the Min-Max normalized time series feature and the Z-score standard time series feature; An encoder based on an autoencoder compresses the fused standard time series features into a low-dimensional space, and a decoder based on the autoencoder reconstructs the original data from the low-dimensional space to determine a corresponding reconstructed time series feature matrix; The loss function of the autoencoder is for: In the formula, They represent the original reconstruction error loss term, the temporal smoothness constraint loss term, and the feature importance constraint loss term, respectively. Respectively represent the weight coefficients of each corresponding loss term; Respectively represent Min-Max normalized data of samples and their reconstructed data, Respectively represent The Z-score standardized data and its reconstructed data of samples, is the total number of samples in the data sample set of the autoencoder; Respectively represent and Min-Max normalized reconstruction data of time steps, Respectively represent and The Z-score normalized reconstructed data of time steps, Indicates the total number of time steps corresponding to the sample; Represents a set of key feature indexes, where the key features include signal-to-noise ratio and bit error rate; Indicates The weight coefficient of each key feature; Indicates The sample The Min-Max normalized value of the key features, Indicates The sample Min-Max normalized reconstruction value of key features, Indicates The sample The Z-score standardized value of the key features, Indicates The sample The Z-score standardized reconstruction value of the key features.
