A method and system for mapping gradient-weighted svm dispersion monitoring

Abstract

This invention relates to the field of optical fiber communication technology, specifically to a mapping gradient weighted SVM dispersion monitoring method and system. The key technical points include the following steps: constructing asynchronous histogram statistical features of the signal; analyzing and calculating the mapping gradient between the histogram features and dispersion parameters; designing a sample weighting strategy based on the mapping gradient curve; embedding the gradient weights into the SVM loss function to construct a mapping gradient weighted SVM regression model; inputting the feature vector of the signal to be monitored into the trained model to solve and output accurate dispersion parameter values. This invention analyzes the non-uniform nonlinear mapping relationship between statistical features and dispersion parameters, introduces a mapping gradient to weight the samples, and enables the model to obtain stronger constraints in the smooth mapping range, thereby improving the accuracy and stability of dispersion monitoring across the entire range. It effectively solves the problem of error accumulation in traditional equal-weighted regression in the large dispersion range and is suitable for online dispersion monitoring in high-speed coherent optical communication systems.

Description

Technical Field

[0001] This invention relates to the field of optical fiber communication technology, and specifically to a method and system for SVM dispersion monitoring using mapped gradient weighting. Background Technology

[0002] With the continuous evolution of information technology, the demand for cloud computing, video services, and large-scale data interaction is growing rapidly, leading to a continuous upward trend in network traffic. As a key infrastructure supporting high-capacity information transmission, optical fiber communication systems are constantly evolving towards higher transmission rates and larger capacities. Coherent optical communication technology, with its high receiver sensitivity and spectral efficiency, has become the mainstream solution for current high-speed optical transmission systems. During long-distance optical fiber transmission, signals are inevitably affected by various physical effects. Among them, dispersion effects cause optical pulse broadening and inter-symbol interference, thereby reducing system performance. With the increase of modulation order and symbol rate, the impact of dispersion on signal quality becomes more significant, becoming one of the important factors restricting system performance. Therefore, accurate monitoring of dispersion parameters in optical signals is an important prerequisite for realizing intelligent operation and maintenance and dynamic optimization of optical networks.

[0003] Various methods have been proposed for addressing the dispersion monitoring problem. Pilot-based or training sequence-based methods achieve dispersion monitoring by introducing known signals; digital signal processing-based methods rely on precise equalization and synchronization processes, placing high demands on system complexity and computational resources; in recent years, machine learning-based dispersion monitoring methods have gained increasing attention, estimating dispersion parameters by constructing a mapping relationship between features and dispersion parameters, including methods such as neural networks and support vector machines. These methods have improved monitoring accuracy to some extent. However, under actual transmission conditions, the mapping relationship between statistical features and dispersion parameters often exhibits non-uniform variation characteristics, meaning that the rate of change varies significantly across different dispersion intervals. Traditional methods assign equal weights to each sample during modeling, making it difficult to balance modeling accuracy across different intervals, especially in intervals where the mapping change slows down, easily leading to large estimation errors and affecting overall monitoring performance. To address these issues, we propose a mapping gradient-weighted SVM dispersion monitoring method and system. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a mapping gradient weighted SVM dispersion monitoring method and system, solving the problems mentioned in the background art.

[0005] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0006] On the one hand, this invention provides a method and system for detecting dispersion in SVM using mapped gradient weighting, comprising the following steps:

[0007] S1. Construct asynchronous histogram statistical features for the signal;

[0008] S2. Analyze and calculate the mapping gradient between histogram features and dispersion parameters;

[0009] S3. Design a sample weighting strategy based on the mapping gradient curve;

[0010] S4. Embed the gradient weights into the SVM loss function to construct a mapped gradient-weighted SVM regression model;

[0011] S5. Input the asynchronous histogram features into the SVM regression model to output accurate dispersion parameter values.

[0012] Preferably, in step S1, the received signal is received asynchronously via sampling.

[0013] Preferably, asynchronous histogram statistical features are constructed for the signal, assuming the received signal is a complex signal sequence:

[0014]

[0015] in, This is the in-phase component corresponding to this sampling point. Let n be the orthogonal component corresponding to the sampling point, where n represents the discrete sampling number and N represents the total number of sampling points.

[0016] Its amplitude is:

[0017]

[0018] Normalize the amplitude:

[0019]

[0020] Preferably, in step S1, the amplitude is divided into M intervals. The asynchronous histogram feature is defined as follows:

[0021]

[0022] in:

[0023]

[0024] The final feature vector is:

[0025]

[0026] Preferably, in step S2, considering that the full width at half maximum (FWHM) can effectively characterize the broadening of the main peak of the histogram and exhibits a relatively stable monotonic response characteristic with the change of dispersion, FWHM is selected as a one-dimensional statistical descriptor to construct the mapping relationship between histogram features and dispersion.

[0027] First, FWHM features are extracted from histogram samples under different dispersion conditions. Then, multiple samples are statistically averaged at each discrete dispersion point to obtain a discrete sequence of FWHM as a function of dispersion point D_k:

[0028]

[0029] in, This represents the FWHM corresponding to the nth sample at the kth dispersion point. This represents the number of samples at this dispersion point.

[0030] Preferably, in step S2, the gradient of the mapping relationship is estimated using a discrete difference method:

[0031]

[0032] in, To prevent the denominator from approaching zero, the above weights are adjusted. It can reflect the mapping change characteristics of different dispersion ranges.

[0033] Preferably, in step S3, to avoid the influence of the gradient sign on the weight distribution, a value is taken as follows: Amplitude is used to construct the sample weight function:

[0034]

[0035] in, Indicates the dispersion value The local gradient obtained at that point.

[0036] Preferably, in the SVM dispersion detection method with mapped gradient weighting, the standard optimization problem of the SVM algorithm in step S4 is:

[0037]

[0038] Preferably, in step S4, after introducing the mapped gradient weights, the standard optimization problem is transformed into:

[0039]

[0040] in, For the determined sample weights, this formula is equivalent to extending the uniform penalty parameter C into a sample correlation form:

[0041]

[0042] Preferably, in step S4, the kernel function chosen for the SVM algorithm is the RBF kernel, which can perform nonlinear mapping on the input features, and its expression is:

[0043]

[0044] Preferably, the method is applicable to multi-modulation format scenarios.

[0045] On the other hand, the present invention also provides a method for detecting SVM dispersion using a mapping gradient weighted approach, comprising:

[0046] The signal acquisition module is used to acquire the optical communication signal from the receiving end;

[0047] The feature extraction module is used to perform asynchronous histogram statistics on the received signal and construct an amplitude distribution feature vector;

[0048] The mapping analysis module is used to establish the mapping relationship between the feature vector and the dispersion parameter, and to calculate the mapping gradient;

[0049] The weight calculation module is used to generate corresponding sample weights based on the mapping gradient.

[0050] The dispersion monitoring module is used to construct a gradient-weighted support vector regression model and output dispersion parameter monitoring values.

[0051] In summary, the present invention has the following main beneficial effects:

[0052] This invention relates to a signal dispersion monitoring method and system for high-speed coherent optical communication systems. By analyzing the non-uniform nonlinear mapping relationship between statistical characteristics and dispersion parameters, a mapping gradient is introduced to weight the samples, enabling the model to obtain stronger constraints in the smooth mapping range. This improves the accuracy and stability of dispersion monitoring across the entire range and effectively solves the problem of error accumulation in traditional equal-weight regression over large dispersion ranges. It is suitable for online dispersion monitoring in high-speed coherent optical communication systems. Attached Figure Description

[0053] Figure 1 This is a flowchart of the method of the present invention;

[0054] Figure 2 This is a system flowchart view of the present invention;

[0055] Figure 3 This is a flowchart of the mapping gradient weighted SVM algorithm of the present invention;

[0056] Figure 4 This is a graph showing the relationship between FWHM and dispersion in Example 1;

[0057] Figure 5 The figures show the dispersion and dispersion error curves, actual OSNR value, and MAE curve of the QAM signal after dispersion monitoring using the SVM algorithm with mapped gradient weighting in Example 1. Detailed Implementation

[0058] To make the technical objectives, technical solutions, and beneficial effects of the present invention clearer, the embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be noted that the described embodiments are only a part of the embodiments of the present invention and are not intended to limit all implementations. Other implementations obtained by those skilled in the art based on the technical concept of the present invention without creative effort should all fall within the protection scope of the present invention.

[0059] The following embodiments are used to illustrate the present invention, but do not constitute a limitation on the scope of protection. The parameters or conditions involved in the implementation process can be appropriately adjusted according to the specific application scenario. Equivalent substitutions or conventional optimizations made to the method without departing from the core idea of ​​the present invention are all within the scope of protection of the present invention.

[0060] Example 1

[0061] With the increasing application of high-speed coherent optical communication systems in high-capacity and long-distance transmission, they offer significant advantages in improving spectrum utilization efficiency and system performance. However, signals are affected by factors such as device noise, dispersion effects, and Kerr nonlinearity during transmission, leading to waveform distortion and performance degradation. Among these, dispersion effects cause optical pulse broadening and exacerbate inter-symbol interference, becoming a crucial factor affecting system transmission quality. Therefore, accurate monitoring of dispersion parameters is of great significance for achieving link state awareness and adaptive compensation.

[0062] refer to Figures 1-5 This invention provides a method for detecting dispersion in SVM using a mapped gradient weighted approach, comprising the following steps:

[0063] S1. Construct asynchronous histogram statistical features for the signal;

[0064] S2. Analyze and calculate the mapping gradient between histogram features and dispersion parameters;

[0065] S3. Design a sample weighting strategy based on the mapping gradient curve;

[0066] S4. Embed the gradient weights into the SVM loss function to construct a mapped gradient-weighted SVM regression model;

[0067] S5. Input the asynchronous histogram features into the SVM regression model to output accurate dispersion parameter values.

[0068] On the other hand, the present invention provides a mapping gradient weighted SVM dispersion monitoring system, comprising:

[0069] The signal acquisition module is used to acquire the optical communication signal from the receiving end;

[0070] The feature extraction module is used to perform asynchronous histogram statistics on the received signal and construct an amplitude distribution feature vector;

[0071] The mapping analysis module is used to establish the mapping relationship between the feature vector and the dispersion parameter, and to calculate the mapping gradient;

[0072] The weight calculation module is used to generate corresponding sample weights based on the mapping gradient.

[0073] The dispersion monitoring module is used to construct a gradient-weighted support vector regression model and output dispersion parameter monitoring values.

[0074] Figure 1 In the dispersion monitoring system shown, the QAM signal is asynchronously received by the signal acquisition module and then output to the feature extraction module to extract the amplitude histogram features of the asynchronous signal.

[0075] Input into the mapping analysis module to establish the mapping relationship between histogram features and dispersion parameters;

[0076] The input is fed into the weight calculation module, which generates the corresponding sample weights based on the mapping relationship.

[0077] The dispersion monitoring module is used to construct a gradient-weighted support vector regression model and output the dispersion parameter monitoring values.

[0078] The specific algorithm flow is as follows:

[0079] 1. Construct asynchronous histogram statistical features for the signal.

[0080] (1) Assume the received signal is a complex signal sequence:

[0081]

[0082] Its amplitude is:

[0083]

[0084] in, This is the in-phase component corresponding to this sampling point. Let n be the orthogonal component corresponding to the sampling point, where n represents the discrete sampling number and N represents the total number of sampling points.

[0085] (2) Normalize the amplitude:

[0086]

[0087] (3) Divide the amplitude into M intervals Here, M is set to 64, then the asynchronous histogram feature is defined as:

[0088]

[0089] in:

[0090]

[0091] The resulting feature vector is:

[0092]

[0093] 2. Analyze and calculate the mapping gradient between histogram features and dispersion parameters.

[0094] (1) Considering that the full width at half maximum (FWHM) can effectively characterize the broadening of the main peak of the histogram and exhibits a relatively stable monotonic response characteristic with the change of dispersion, FWHM is selected as a one-dimensional statistical descriptor to construct the mapping relationship between histogram features and dispersion.

[0095] (2) Extract FWHM features from histogram samples under different dispersion conditions, and perform statistical averaging on multiple sets of samples at each discrete dispersion point to obtain a discrete sequence of FWHM as a function of dispersion point Dk:

[0096]

[0097] in, This represents the FWHM corresponding to the nth sample at the kth dispersion point. This represents the number of samples at this dispersion point.

[0098] (2) Gradient estimation of the mapping relationship is performed using the discrete difference method:

[0099]

[0100] in, An adjustment factor to prevent the denominator from approaching zero. The above weights. It can reflect the mapping change characteristics of different dispersion ranges.

[0101] 3. Design a sample weighting strategy based on the mapping gradient curve.

[0102] (1) To avoid the influence of the gradient sign on the weight distribution, take Amplitude is used to construct the sample weight function:

[0103]

[0104] in, Indicates the dispersion value The local gradient obtained is not directly calculated from the input features, but is determined by its mapping change characteristics in the dispersion dimension, thereby realizing the mapping from the mapping structure to the sample weights.

[0105] 4. Embed gradient weights into the SVM loss function to construct a gradient-weighted SVM regression model.

[0106] (1) This method uses the SVM algorithm to establish a nonlinear mapping relationship between histogram features and dispersion parameters.

[0107] (2) Since the histogram features and dispersion parameters are nonlinearly mapped, the standard optimization problem of the SVM algorithm is:

[0108]

[0109] (3) After introducing the mapped gradient weights, the optimization objective can be expanded to:

[0110]

[0111] in, The sample weights are defined. This form is equivalent to extending the uniform penalty parameter C to the sample correlation form:

[0112]

[0113] (4) The kernel function of the SVM algorithm uses the RBF kernel to perform nonlinear mapping on the input features, and its expression is:

[0114]

[0115] 5. Input the asynchronous histogram features into the SVM regression model to output accurate dispersion parameter values.

[0116] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that, unless otherwise defined, the technical or scientific terms used in this invention should be understood in their ordinary sense as understood by one of ordinary skill in the art to which this invention pertains. Terms such as "comprising" or "including" as used in this invention mean that the element or object preceding the word encompasses the element or object listed after the word and its equivalents. Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A gradient-weighted SVM dispersion monitoring method, characterized in that, Includes the following steps: S1. Construct asynchronous histogram statistical features for the signal; S2. Analyze and calculate the mapping gradient between histogram features and dispersion parameters; S3. Design a sample weighting strategy based on the mapping gradient curve; S4. Embed the gradient weights into the SVM loss function to construct a mapped gradient-weighted SVM regression model; S5. Input the asynchronous histogram features into the SVM regression model to output accurate dispersion parameter values.

2. The SVM dispersion monitoring method based on mapped gradient weighting according to claim 1, characterized in that, In step S1, asynchronous histogram statistical features are constructed for the signal, assuming the received signal is a complex signal sequence: in, This is the in-phase component corresponding to this sampling point. Let n be the orthogonal component corresponding to the sampling point, where n represents the discrete sampling number and N represents the total number of sampling points. Its amplitude is: Normalize the amplitude: Divide the amplitude into M intervals. The asynchronous histogram feature is defined as follows: in: The final feature vector is: 。 3. The SVM dispersion monitoring method based on mapped gradient weighting according to claim 1, characterized in that, In step S2, considering that the full width at half maximum (FWHM) can effectively characterize the broadening of the main peak of the histogram and exhibits a relatively stable monotonic response characteristic with the change of dispersion, FWHM is selected as a one-dimensional statistical descriptor to construct the mapping relationship between histogram features and dispersion. First, FWHM features are extracted from histogram samples under different dispersion conditions. Then, multiple samples are statistically averaged at each discrete dispersion point to obtain a discrete sequence of FWHM as a function of dispersion point D_k. in, This represents the FWHM corresponding to the nth sample at the kth dispersion point. This represents the number of samples at that dispersion point; Under different dispersion conditions, the FWHM sequences under the corresponding dispersion are extracted, and the gradient of the mapping relationship is estimated by discrete difference: in, To prevent the denominator from approaching zero, the above weights are adjusted. It can reflect the mapping change characteristics of different dispersion ranges.

4. The SVM dispersion monitoring method based on mapped gradient weighting according to claim 1, characterized in that, In step S3, to avoid the influence of the gradient sign on the weight distribution, we take... Amplitude is used to construct the sample weight function: in, Indicates the dispersion value The local gradient obtained at that point.

5. The SVM dispersion monitoring method based on mapped gradient weighting according to claim 1, characterized in that, In step S4, the standard optimization problem of the SVM algorithm is: 。 6. The SVM dispersion monitoring method based on mapped gradient weighting according to claim 1, characterized in that, In step S4, after introducing the mapped gradient weights, the optimization objective can be expanded to: in, For the determined sample weights, this form is equivalent to extending the uniform penalty parameter C to the sample correlation form: 。 7. The SVM dispersion monitoring method based on mapped gradient weighting according to claim 5, characterized in that, In step S4, the RBF kernel is chosen to perform nonlinear mapping on the input features, and its expression is: 。 8. A gradient-weighted SVM dispersion monitoring system, characterized in that, include: The signal acquisition module is used to acquire the optical communication signal from the receiving end; The feature extraction module is used to perform asynchronous histogram statistics on the received signal and construct an amplitude distribution feature vector; The mapping analysis module is used to establish the mapping relationship between the feature vector and the dispersion parameter, and to calculate the mapping gradient; The weight calculation module is used to generate corresponding sample weights based on the mapping gradient. The dispersion monitoring module is used to construct a gradient-weighted support vector regression model and output dispersion parameter monitoring values.

Images