Joint monitoring method for modulation format and optical signal-to-noise ratio based on amplitude analysis of complex plane

Abstract

The application discloses a modulation format and optical signal-to-noise ratio joint monitoring method based on an amplitude analytic complex plane, and relates to the field of communication, and comprises the following steps: performing constant modulus equalization processing on a signal to be identified to obtain a preprocessed digital signal; generating an amplitude analytic complex plane based on the preprocessed digital signal; and identifying the modulation format and the optical signal-to-noise ratio of the amplitude analytic complex plane through a trained multi-target neural network.The application can simultaneously perform modulation format identification and optical signal-to-noise ratio monitoring through the multi-target neural network, share bottom layer feature extraction, avoid redundant calculation, reduce the complexity of the scheme, and improve the joint identification efficiency.Through verification, the method can achieve 100% correct identification rate of five kinds of modulation formats when the optical signal-to-noise ratio is not lower than the threshold corresponding to 20% FEC, and the maximum error of optical signal-to-noise ratio monitoring of the five kinds of modulation formats is controlled within 1 dB.

Description

Technical Field

[0001] This invention relates to the field of communications, and more specifically to a modulation format based on amplitude analysis complex plane and a method for joint monitoring of optical signal-to-noise ratio. Background Technology

[0002] Elastic Optical Networks (EONs) can better address the problem of insufficient spectrum resource utilization in traditional wavelength division multiplexing (WDM) systems, and further realize the digitization and dynamism of the network with their flexible spectrum allocation and parameter adjustment capabilities. Meanwhile, nodes and receivers in EONs need to implement Optical Performance Monitoring (OPM) to obtain information such as signal dispersion, modulation format, and optical signal-to-noise ratio (OSNR) to ensure the efficient operation of the EON.

[0003] In recent years, researchers have broadly categorized OPM research in EON into two types: single-parameter monitoring and joint multi-parameter monitoring. In single-parameter monitoring schemes, improved principal component analysis algorithms, trajectory information, and particle swarm optimization can achieve modulation format (MFI) identification; training sequences and delay-tap-based sampling techniques are used for dispersion monitoring; and random forests and support vector machines are used for optical signal-to-noise ratio (SNR) estimation. However, these schemes can only achieve single-parameter monitoring. When monitoring other parameters, new features need to be selected or the data retrained, leading to higher computational complexity and time delays. To address this issue, joint multi-parameter monitoring has received increasing attention. Deep neural networks (DNNs) are widely used in EON OPM in joint multi-parameter monitoring schemes because they offer high monitoring accuracy and can autonomously learn features in dynamic, high-volume, and complex data environments. Some techniques propose using features such as constellation diagrams, spectra, intensity and differential phase images, and amplitude histograms as inputs to multi-task DNNs to achieve modulation format identification and SNR estimation. However, with the continuous development of EON, the requirements for convergence speed and computational complexity are becoming increasingly stringent. Therefore, it is necessary to study joint multi-parameter OPM schemes suitable for high-order modulation formats (such as 128QAM) in the next generation of EON. Summary of the Invention

[0004] To address the aforementioned shortcomings in existing technologies, the modulation format based on amplitude analysis complex plane and the optical signal-to-noise ratio joint monitoring method provided by this invention solves the problems of existing methods being unable to simultaneously recognize 128QAM signals and having low recognition speed.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0006] A method for jointly monitoring modulation format and optical signal-to-noise ratio based on amplitude analytical complex plane is provided, which includes the following steps:

[0007] The signal to be identified is subjected to CMA equalization to obtain a preprocessed digital signal.

[0008] Amplitude analytical complex plane is generated based on the preprocessed digital signal;

[0009] The initial convolution module performs preliminary feature extraction and normalization on the amplitude analytical complex plane;

[0010] The spatial dimension of the features output by the initial convolutional module is reduced by using max pooling layers;

[0011] The feature extraction module performs multi-level feature reuse and fusion on the features output by the max pooling layer.

[0012] The transition layer compresses the dimensionality of the features output by the feature extraction module and reduces the spatial resolution of the features output by the feature extraction module.

[0013] The spatial dimension of the feature map output by the transition layer is reduced to 1×1 by using a global average pooling layer, while keeping the channel dimension unchanged.

[0014] The modulation format and optical signal-to-noise ratio are jointly monitored by the output layer based on the features output by the global average pooling layer, and the detection results are output.

[0015] Furthermore, the specific method for generating an amplitude analytical complex plane based on the preprocessed digital signal includes the following steps:

[0016] The power of the preprocessed digital signal is normalized to extract amplitude information;

[0017] The amplitude information is analyzed to obtain a two-dimensional planar diagram;

[0018] The obtained two-dimensional planar diagram is plotted as a heat map to obtain the amplitude analytical complex plane.

[0019] Furthermore, the expression for power normalization is:

[0020]

[0021] in This represents the m-th level amplitude value of the preprocessed digital signal; This represents the correlation probability of the m-th level of the preprocessed digital signal; M represents the total number of levels in the modulation format.

[0022] Furthermore, the specific method for analyzing the amplitude information to obtain a two-dimensional planar diagram includes the following steps:

[0023] Based on the amplitude information, a discrete Fourier transform is performed on the preprocessed digital signal to obtain the frequency domain signal, whose expression is:

[0024]

[0025] in Represents a frequency domain signal; This represents the preprocessed digital signal; is the natural constant; N is the sign number; k is the frequency domain index; j is the imaginary unit; Pi; n is the time-domain index;

[0026] For frequency domain signals For frequency domain processing, the expression is:

[0027]

[0028] in The signal after frequency domain processing;

[0029] The signal after frequency domain processing The inverse Fourier transform is expressed as follows:

[0030]

[0031] in The time-domain signal is obtained by the inverse Fourier transform;

[0032] The time-domain signal obtained by inverse Fourier transform Together with the amplitude information, a two-dimensional graph is formed, resulting in a two-dimensional planar graph.

[0033] Furthermore, the initial convolutional module comprises a Conv2D layer, a BatchNorm layer, and a ReLU activation layer connected in sequence.

[0034] Furthermore, the feature extraction module employs a fully connected structure built using a multi-level gated aggregation mechanism, comprising a first branch, a second branch, and a feature fusion module, wherein:

[0035] The first branch includes the following connected components:

[0036] The first convolutional submodule is used to adjust the channel dimensions through 1×1 convolution and output the channel-adjusted features.

[0037] The first activation submodule is used to perform SiLU activation on the output of the first convolution submodule and output the activated features.

[0038] The second branch includes the following connected routes:

[0039] The multi-level convolution submodule contains parallel depthwise separable convolution groups that extract features at different scales with different dilation rates and kernel sizes, and then concatenate the features at different scales to output multi-scale features.

[0040] The second convolutional submodule is used to adjust the channel dimensions through 1×1 convolution and output the channel-adjusted features.

[0041] The second activation submodule is used to perform SiLU activation on the output of the second convolution submodule and output the activated features.

[0042] The feature fusion module is used to add and fuse the outputs of the first branch and the second branch, and then perform a 1×1 convolution on the fusion result to adjust the channel dimension and output the final feature.

[0043] The inputs to the first convolutional submodule and the multi-level convolutional submodule are both inputs to the feature extraction module, and the output of the feature fusion module is the output of the feature extraction module.

[0044] Furthermore, the transition layer includes batch normalization units, ReLU activation units, 1×1 convolution units, and average pooling operation units connected in sequence.

[0045] Furthermore, in the output layer, the joint monitoring of modulation format and optical signal-to-noise ratio includes modulation format identification and optical signal-to-noise ratio identification. Modulation format identification is a classification identification performed as a classification task, while optical signal-to-noise ratio identification is a regression identification performed as a regression task.

[0046] Furthermore, the signals to be identified include QPSK signals, 16QAM signals, 32QAM signals, 64QAM signals, and 128QAM signals.

[0047] Furthermore, the initial convolutional module, max pooling layer, feature extraction module, transition layer, global average pooling layer, and output layer constitute a multi-objective neural network. The training process of the multi-objective neural network includes the following steps:

[0048] The amplitude analytical complex planes corresponding to QPSK, 16QAM, 32QAM, 64QAM, and 128QAM signals under known optical signal-to-noise ratios (SNRs) were obtained respectively, and used as training samples. The SNR ranges for QPSK signals (7–26 dB), 16QAM signals (15–33 dB), 32QAM signals (18–37 dB), 64QAM signals (19–38 dB), and 128QAM signals (23–40 dB).

[0049] The actual optical signal-to-noise ratio and modulation format corresponding to the training samples are used as labels. The multi-target neural network is trained using the training samples, and the parameters of the multi-target neural network are updated by backpropagation until the recognition accuracy of the multi-target neural network reaches the set value.

[0050] The beneficial effects of this invention are as follows:

[0051] 1. This method can simultaneously achieve accurate identification of modulation format and accurate estimation of optical signal-to-noise ratio based on the characteristic differences of different modulation formats and different optical signal-to-noise ratio values ​​on the amplitude resolution complex plane. It does not require multiple neural networks to perform modulation format identification and optical signal-to-noise ratio monitoring separately, which reduces the complexity of the scheme and the time required, and improves the efficiency of joint identification of modulation format and optical signal-to-noise ratio.

[0052] 2. This method uses the amplitude analysis complex plane as input features, which enhances the signal discrimination and facilitates the subsequent training and recognition of modulation format and optical signal-to-noise ratio information by the multi-objective neural network. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the method.

[0054] Figure 2 This is a simulation schematic diagram of the coherent optical communication system in the embodiment;

[0055] Figure 3 for Figure 2 Architecture diagram of the digital signal processing module;

[0056] Figure 4 For the amplitude analysis complex plane of the QPSK signal;

[0057] Figure 5 The complex plane for amplitude analysis of a 16QAM signal;

[0058] Figure 6 The complex plane for amplitude analysis of a 32QAM signal;

[0059] Figure 7 The complex plane for amplitude analysis of a 64QAM signal;

[0060] Figure 8 The complex plane for amplitude analysis of a 128QAM signal;

[0061] Figure 9 Here is a diagram of a multi-objective neural network architecture; where (a) is the multi-objective neural network architecture and (b) is the network architecture of the feature extraction layer.

[0062] Figure 10 The graph shows the change in modulation format recognition accuracy as a function of optical signal-to-noise ratio.

[0063] Figure 11The diagrams illustrate the relationship between the actual and estimated optical signal-to-noise ratios (OSRs) of five signals in the embodiments; where (a) is the relationship between the actual and estimated OSRs of a QPSK signal; (b) is the relationship between the actual and estimated OSRs of a 16QAM signal; (c) is the relationship between the actual and estimated OSRs of a 32QAM signal; (d) is the relationship between the actual and estimated OSRs of a 64QAM signal; and (e) is the relationship between the actual and estimated OSRs of a 128QAM signal.

[0064] Figure 12 This is a performance comparison diagram. Detailed Implementation

[0065] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0066] like Figure 1 and Figure 9 As shown, the modulation format and optical signal-to-noise ratio joint monitoring method based on amplitude analysis complex plane includes the following steps:

[0067] S1. Perform CMA equalization on the signal to be identified to obtain the preprocessed digital signal;

[0068] S2. Generate an amplitude analytical complex plane with dimensions of 28×28×3 based on the preprocessed digital signal;

[0069] S3. The initial convolution module performs preliminary feature extraction and normalization on the amplitude analytical complex plane; the output feature size of the initial convolution module is 14×14×64.

[0070] S4. Reduce the spatial dimension of the features output by the initial convolutional module by using a max pooling layer; the output feature size of the max pooling layer is 7×7×64.

[0071] S5. The feature extraction module (Feature block 1) performs multi-level feature reuse and fusion on the features output by the max pooling layer; the output feature size of the feature extraction module is 7×7×128.

[0072] S6. The dimensionality of the features output by the feature extraction module is compressed and the spatial resolution of the features output by the feature extraction module is reduced through the transition layer (Translayer1); the output feature size of the transition layer is 3×3×64.

[0073] S7. The spatial dimension of the feature map output by the transition layer is reduced to 1×1 by a global average pooling layer (Avg Pool Layer) while keeping the channel dimension unchanged; the output feature size of the global average pooling layer is 1×1×64.

[0074] S8. The modulation format and optical signal-to-noise ratio are jointly monitored by the output layer based on the features output by the global average pooling layer, and the detection results are output.

[0075] The initial convolutional module, max pooling layer, feature extraction module, transition layer, global average pooling layer, and output layer constitute a multi-objective neural network.

[0076] In this embodiment, the specific method for generating the amplitude analytical complex plane based on the preprocessed digital signal in step S2 includes the following steps:

[0077] S2-1. Normalize the power of the preprocessed digital signal and extract the amplitude information;

[0078] S2-2. Analyze the amplitude information to obtain a two-dimensional planar diagram;

[0079] S2-3. Draw the obtained two-dimensional planar diagram into a heat map to obtain the amplitude analytical complex plane.

[0080] The expression for power normalization is:

[0081]

[0082] in This represents the m-th level amplitude value of the preprocessed digital signal; This represents the correlation probability of the m-th level of the preprocessed digital signal; M represents the total number of levels in the modulation format.

[0083] In this embodiment, the specific method for parsing the amplitude information to obtain the two-dimensional planar diagram in step S2-2 includes the following steps:

[0084] S2-2-1. Perform a Discrete Fourier Transform on the preprocessed digital signal based on the amplitude information to obtain the frequency domain signal, the expression of which is:

[0085]

[0086] in Represents a frequency domain signal; This represents the preprocessed digital signal; is the natural constant; N is the sign number; k is the frequency domain index; j is the imaginary unit; π is the mathematical constant pi; n is the time-domain index; although k and n have different meanings, since the Discrete Fourier Transform maps a time-domain signal of length N to a frequency-domain signal of length N, their values ​​are in the same range, from 0 to 1. ;

[0087] S2-2-2, For frequency domain signals For frequency domain processing, the expression is:

[0088]

[0089] in The signal after frequency domain processing;

[0090] S2-2-3, The signal after frequency domain processing The inverse Fourier transform is expressed as follows:

[0091]

[0092] in The time-domain signal is obtained by the inverse Fourier transform;

[0093] S2-2-4. The time-domain signal obtained by inverse Fourier transform Together with the amplitude information, a two-dimensional graph is formed, resulting in a two-dimensional planar graph.

[0094] The above operations normalize the signal amplitude information to a specific range, dividing the image into 80×80 squares. Each square contains information about the parsed signal amplitude and the number of symbols within the square. However, when the number of symbols is fixed, the number of symbols within an interval varies depending on the modulation format and the optical signal-to-noise ratio (SNR) of the signal. QPSK signal amplitudes exhibit a single-level distribution, with the total number of symbols concentrated in the squares corresponding to the single-level distribution; 16QAM signal amplitudes exhibit a three-level distribution, with the total number of symbols concentrated in the squares corresponding to the three-level distribution; 32QAM signal amplitudes exhibit a five-level distribution, with the total number of symbols concentrated in the squares corresponding to the five-level distribution; 64QAM signal amplitudes exhibit a nine-level distribution, with the total number of symbols concentrated in the squares corresponding to the nine-level distribution; and 128QAM signals exhibit a thirteen-level distribution, with the total number of symbols concentrated in the squares corresponding to the thirteen-level distribution. Since the total number of symbols is fixed, but the amplitude distribution levels of each modulation format are different, the number of squares corresponding to each distribution level is also different. Therefore, the number of symbols allocated to each square is also different, thus laying the foundation for subsequent recognition.

[0095] In this embodiment, in order to perform preliminary feature extraction and normalization on the amplitude analytical complex plane, the initial convolutional module includes a Conv2D layer (Conv Layer), a BatchNorm layer, and a ReLU activation layer connected in sequence.

[0096] In this embodiment, the feature extraction module adopts a fully connected structure constructed using a multi-level gated aggregation mechanism, including a first branch, a second branch, and a feature fusion module, wherein:

[0097] The first branch includes the following connected components:

[0098] The first convolutional submodule (Conv 1*1) is used to adjust the channel dimensions through 1×1 convolution and output the channel-adjusted features;

[0099] The first activation submodule (SiLU) is used to activate the output of the first convolutional submodule using SiLU, and output the activated features.

[0100] The second branch includes the following connected routes:

[0101] The multi-level convolutional submodule (DWConv Block) contains parallel depthwise separable convolutional groups that extract features at different scales with different dilation rates and kernel sizes, and then concatenate the features at different scales to output multi-scale features.

[0102] The second convolutional submodule (Conv 1*1) is used to adjust the channel dimensions through 1×1 convolution and output the channel-adjusted features;

[0103] The second activation submodule (SiLU) is used to activate the output of the second convolutional submodule using SiLU, and outputs the activated features.

[0104] The feature fusion module is used to add and fuse the outputs of the first branch and the second branch, and then perform a 1×1 convolution on the fusion result to adjust the channel dimension and output the final feature.

[0105] The inputs to the first convolutional submodule and the multi-level convolutional submodule are both inputs to the feature extraction module, and the output of the feature fusion module is the output of the feature extraction module.

[0106] In this embodiment, the transition layer includes a batch normalization unit, a ReLU activation unit, a 1×1 convolution unit, and an average pooling operation unit connected in sequence.

[0107] In this embodiment, the joint monitoring of modulation format and optical signal-to-noise ratio (OSN) includes modulation format identification and OSN identification. Modulation format identification is a classification task, while OSN identification is a regression task. Specifically, the output layer includes a classifier (FC1) and a regressor (FC2) connected in parallel: the classifier performs modulation format identification (classification task), and the regressor performs OSN identification (regression task). The signals to be identified include QPSK, 16QAM, 32QAM, 64QAM, and 128QAM signals.

[0108] In this embodiment, the training process of the multi-objective neural network includes the following steps:

[0109] A1. Obtain the amplitude analytical complex planes corresponding to QPSK, 16QAM, 32QAM, 64QAM, and 128QAM signals under known optical signal-to-noise ratios (SNRs), and use them as training samples; where the SNR range of QPSK signals is 7~26dB; the SNR range of 16QAM signals is 15~33dB; the SNR range of 32QAM signals is 18~37dB; the SNR range of 64QAM signals is 19~38dB; and the SNR range of 128QAM signals is 23~40dB.

[0110] A2. Use the actual optical signal-to-noise ratio and modulation format corresponding to the training samples as labels, train the multi-target neural network using the training samples, update the parameters of the multi-target neural network in the backpropagation manner, until the recognition accuracy of the multi-target neural network reaches the set value.

[0111] In the specific implementation process, in order to control the number of training samples, the optical signal-to-noise ratio of each signal is selected at 1dB intervals. That is, the five signals contain 20, 19, 20, 20 and 18 optical signal-to-noise ratio values ​​respectively. The total number of symbols corresponding to each feature map is 8000. Each optical signal-to-noise ratio value of each modulation format contains 100 sets of feature maps as samples, so a total of 9700 sets of samples are obtained.

[0112] The 9700 samples were divided into training and validation samples in an 8:2 ratio. The classification task labels for the five modulation format signals were set to '0', '1', '2', '3' and '4', respectively, and the regression task labels were set to the true optical signal-to-noise ratio value corresponding to each modulation format signal. The multi-target neural network was trained and its parameters were updated by backpropagation until the recognition accuracy of the multi-target neural network reached the set value.

[0113] To verify the feasibility of this method, numerical simulations were performed on the transmission of 28GBaud polarization-multiplexed QPSK / 16QAM / 32QAM / 64QAM / 128QAM signals. The simulation block diagram of the 28GBaud coherent optical communication system is shown below. Figure 2 As shown. At the transmitting end, the laser wavelength is 1550nm, the linewidth is 100kHz, and I and Q are the in-phase and quadrature components, respectively. The electrical signal drives the IQ modulator to generate QPSK / 16QAM / 32QAM / 64QAM / 128QAM signals with a transmission rate of 28GBaud, which are then converted into polarization multiplexed signals by a polarization beam combiner and enter the transmission channel. The optical signal-to-noise ratio setting module can adjust the optical signal-to-noise ratio value of the signal. At the receiving end, the out-of-band noise of the optical signal is first filtered by an optical bandpass filter, and then the optical signal is split into two paths by a polarization beam splitter. These two paths are then mixed 90° with the two local oscillator optical signals split from the local oscillator laser. The mixed signals are converted into electrical signals by a balanced detector, filtered by a low-pass filter, and finally digitized by an analog-to-digital converter for subsequent digital signal processing. Figure 3 As shown, Ex and Ey are the x-polarized and y-polarized signals orthogonal to each other on the same wavelength received by the coherent optical receiver. The entire digital signal processing part includes three types of algorithms: modulation format uncorrelated algorithm, the present method, and modulation format correlated algorithm. The modulation format uncorrelated algorithm and the modulation format correlated algorithm are existing algorithms and are not the inventive features of this application.

[0114] In this embodiment, the amplitude resolution complex planes of QPSK / 16QAM / 32QAM / 64QAM / 128QAM signals are respectively as follows: Figure 4 , Figure 5 , Figure 6 , Figure 7 and Figure 8 As shown.

[0115] The modulation format recognition accuracy of this method under different optical signal-to-noise ratios is as follows: Figure 10 As shown in the figure, the vertical axis represents the correct recognition rate (%), and the horizontal axis represents the optical signal-to-noise ratio (dB). The optical signal-to-noise ratio ranges for QPSK, 16QAM, 32QAM, 64QAM, and 128QAM signals are 7~26dB, 15~33dB, 18~37dB, 19~38dB, and 23~40dB, respectively, with 1dB intervals. For 28GBaud polarization-multiplexed QPSK / 16QAM / 32QAM / 64QAM / 128QAM signals, a 100% correct recognition rate was achieved across the entire optical signal-to-noise ratio range. The lowest optical signal-to-noise ratios of 7dB, 15dB, 18dB, 19dB, and 23dB are all below the optical signal-to-noise ratio threshold corresponding to 20% forward error correction (FEC) for each modulation format (corresponding to...). Figure 10 The vertical dashed lines of the same color in the image have a bit error rate (BER) of 2.4 × 10⁻⁶. -2 ).

[0116] The relationship between the actual optical signal-to-noise ratio (OSR) and the estimated OSR for the five signals proposed in this method is as follows: Figure 11 As shown, the reference line for the true optical signal-to-noise ratio (SNR) is depicted by a black diagonal line. The horizontal axis represents the true SNR value, and the vertical axis represents the estimated SNR value. It can be seen that the estimation error is slightly larger at lower or higher SNR values ​​compared to intermediate SNR values, and the estimation error for higher-order modulation schemes is larger than that for lower-order modulation schemes. This is because at low SNR, the significant fluctuations in features caused by noise increase the difficulty of SNR estimation. At high SNR, the influence of noise is greatly reduced, and the distinguishability of features between adjacent SNR values ​​is significantly reduced, further increasing the difficulty of SNR estimation.

[0117] To further verify the complexity of our proposed solution, we compared it with four typical neural network models (DenseNet, ResNet, MobileNet, and VGG-like), and the results are shown in Table 1. To ensure fairness in the comparison, all models maintained the same input image size (28×28) and number of symbols (8000). As can be seen from Table 1, our proposed solution contains only 0.04M parameters and 112.41M FLOPs. In contrast, DenseNet and ResNet have as many as 38.39M and 23.52M parameters, respectively, and their computational cost is significantly higher. Even the lightweight MobileNet and VGG-like models have far more parameters and computational cost than our proposed solution. Figure 12 The paper further demonstrates the performance of five schemes in modulation format recognition accuracy and optical signal-to-noise ratio (SNR) monitoring. It can be observed that although all models achieved high recognition rates (close to or equal to 100%) in the modulation format recognition task, the proposed method exhibits lower mean absolute error (MAE) across various modulation formats in the SNR monitoring task, especially under the more complex 128QAM, where its performance advantage is even more significant. Therefore, this method maintains high recognition accuracy and superior monitoring performance while significantly reducing model complexity, making it particularly suitable for optical network scenarios with limited computational resources, demonstrating good practical potential.

[0118] Table 1: Performance Comparison

[0119]

[0120] In summary, this invention, through a specially designed multi-objective neural network, can simultaneously and accurately identify the modulation format and precisely estimate the optical signal-to-noise ratio based on the characteristic differences of different modulation formats and different optical signal-to-noise ratio values ​​on the amplitude analytical complex plane. This eliminates the need for multiple neural networks to perform modulation format identification and optical signal-to-noise ratio monitoring separately, reducing the complexity and time required for the solution and improving the efficiency of joint identification of modulation format and optical signal-to-noise ratio.

Claims

1. A method for jointly monitoring modulation format and optical signal-to-noise ratio based on amplitude analytical complex plane, characterized in that, Includes the following steps: The signal to be identified is subjected to CMA equalization to obtain a preprocessed digital signal. Amplitude analytical complex plane is generated based on the preprocessed digital signal; The initial convolution module performs preliminary feature extraction and normalization on the amplitude analytical complex plane; The spatial dimension of the features output by the initial convolutional module is reduced by using max pooling layers; The feature extraction module performs multi-level feature reuse and fusion on the features output by the max pooling layer. The transition layer compresses the dimensionality of the features output by the feature extraction module and reduces the spatial resolution of the features output by the feature extraction module. The spatial dimension of the feature map output by the transition layer is reduced to 1×1 by using a global average pooling layer, while keeping the channel dimension unchanged. The modulation format and optical signal-to-noise ratio are jointly monitored by the output layer based on the features output by the global average pooling layer, and the detection results are output. The specific method for generating an amplitude analytical complex plane based on a preprocessed digital signal includes the following steps: The power of the preprocessed digital signal is normalized to extract amplitude information; The amplitude information is analyzed to obtain a two-dimensional planar diagram; The obtained two-dimensional planar diagram is plotted as a heat map to obtain the amplitude analytical complex plane; The specific method for parsing amplitude information to obtain a two-dimensional planar graph includes the following steps: Based on the amplitude information, a discrete Fourier transform is performed on the preprocessed digital signal to obtain the frequency domain signal, whose expression is: in Represents a frequency domain signal; This represents the preprocessed digital signal; is the natural constant; N is the sign number; k is the frequency domain index; j is the imaginary unit; Pi; n is the time-domain index; For frequency domain signals For frequency domain processing, the expression is: in The signal after frequency domain processing; The signal after frequency domain processing The inverse Fourier transform is expressed as follows: in The time-domain signal is obtained by the inverse Fourier transform; The time-domain signal obtained by inverse Fourier transform Together with the amplitude information, a two-dimensional graph is formed, resulting in a two-dimensional planar graph; The feature extraction module employs a fully connected structure built using a multi-level gated aggregation mechanism, comprising a first branch, a second branch, and a feature fusion module, wherein: The first branch includes the following connected components: The first convolutional submodule is used to adjust the channel dimensions through 1×1 convolution and output the channel-adjusted features. The first activation submodule is used to perform SiLU activation on the output of the first convolution submodule and output the activated features. The second branch includes the following connected routes: The multi-level convolution submodule contains parallel depthwise separable convolution groups that extract features at different scales with different dilation rates and kernel sizes, and then concatenate the features at different scales to output multi-scale features. The second convolutional submodule is used to adjust the channel dimensions through 1×1 convolution and output the channel-adjusted features. The second activation submodule is used to perform SiLU activation on the output of the second convolution submodule and output the activated features. The feature fusion module is used to add and fuse the outputs of the first branch and the second branch, and then perform a 1×1 convolution on the fusion result to adjust the channel dimension and output the final feature. The inputs to the first convolutional submodule and the multi-level convolutional submodule are both inputs to the feature extraction module, and the output of the feature fusion module is the output of the feature extraction module.

2. The method according to claim 1, characterized in that, The expression for power normalization is: in This represents the m-th level amplitude value of the preprocessed digital signal; This represents the correlation probability of the m-th level of the preprocessed digital signal; M represents the total number of levels in the modulation format.

3. The method according to claim 1, characterized in that, The initial convolutional module consists of a Conv2D layer, a BatchNorm layer, and a ReLU activation layer connected in sequence.

4. The method according to claim 1, characterized in that, The transition layer consists of batch normalization units, ReLU activation units, 1×1 convolution units, and average pooling operation units connected in sequence.

5. The method according to claim 1, characterized in that, In the output layer, the joint monitoring of modulation format and optical signal-to-noise ratio includes modulation format identification and optical signal-to-noise ratio identification. Modulation format identification is a classification identification performed as a classification task, while optical signal-to-noise ratio identification is a regression identification performed as a regression task.

6. The method according to claim 1, characterized in that, The signals to be identified include QPSK signals, 16QAM signals, 32QAM signals, 64QAM signals, and 128QAM signals.

7. The method according to claim 1, characterized in that, The initial convolutional module, max pooling layer, feature extraction module, transition layer, global average pooling layer, and output layer constitute a multi-objective neural network. The training process of a multi-objective neural network includes the following steps: The amplitude analytical complex planes corresponding to QPSK, 16QAM, 32QAM, 64QAM, and 128QAM signals under known optical signal-to-noise ratios (SNRs) were obtained respectively, and used as training samples. The SNR ranges for QPSK signals (7–26 dB), 16QAM signals (15–33 dB), 32QAM signals (18–37 dB), 64QAM signals (19–38 dB), and 128QAM signals (23–40 dB). The actual optical signal-to-noise ratio and modulation format corresponding to the training samples are used as labels. The multi-target neural network is trained using the training samples, and the parameters of the multi-target neural network are updated by backpropagation until the recognition accuracy of the multi-target neural network reaches the set value.

Images