A joint recognition method of modulation format and optical signal-to-noise ratio based on color polar coordinate diagram
Through the multi-task learning neural network method based on color polar graphs, the problem that the prior art is difficult to identify 128QAM signals is solved, and the accurate identification of modulation format and optical signal-to-noise ratio is realized, which reduces complexity and time and improves efficiency.
Patent Information
- Application Number
- CN202411180764.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-08-27
AI Technical Summary
The prior art is difficult to take into account the modulation format identification and optical signal-to-noise ratio monitoring of 128QAM signals, especially in the larger OSNR range, and the traditional methods are highly complex and inefficient.
Using a multi-task learning neural network method based on color polar graphs, the color polar graphs are modulated by CMA equalization and multi-task learning neural network to identify the color polar graphs in the modulation format and optical signal-to-noise ratio, and using a multi-task learning neural network (MTL-Reskanet) Achieve joint identification of modulation format and optical signal-to-noise ratio.
Accurate recognition of different modulation formats and OSNR values is achieved, which reduces the recognition complexity and time, improves efficiency, and enhances the discrimination of recognition through color polar graph feature mapping.
Smart Images

Figure CN118713966B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of communications, and in particular to a method for jointly identifying a modulation format and an optical signal-to-noise ratio based on a color polar coordinate diagram. Background Art
[0002] Elastic optical networks (EONs) have received much attention in recent years. Their notable feature is that they can adaptively adjust different transmission parameters according to channel conditions and traffic requirements. At the same time, as the structure of optical networks becomes more complex and link conditions become more dynamic, monitoring is needed to ensure the reliability and stability of optical networks. Therefore, as a key technology to ensure the reliable and stable operation of optical fiber networks, optical performance monitoring (OPM) technology plays an increasingly important role. In particular, optical signal-to-noise ratio (OSNR) and modulation format are important parameters that need to be identified during the implementation of OPM.
[0003] As optical networks become more complex and dynamic, traditional OPM technology is no longer suitable for current optical networks. Although higher-order modulation formats can further improve the spectral efficiency of the system, the difficulty of modulation format identification and OSNR monitoring is also significantly increased accordingly. Therefore, most of the current modulation format and OSNR joint identification technologies are mainly aimed at modulation formats of 64QAM (QAM: quadrature amplitude modulation) and below. Therefore, a modulation format and OSNR joint identification technology is needed that can simultaneously realize modulation format identification and OSNR monitoring of multiple signals (especially 128QAM) within a larger OSNR range. Summary of the invention
[0004] In view of the above-mentioned deficiencies in the prior art, the modulation format and optical signal-to-noise ratio joint identification method based on color polar coordinate diagram provided by the present invention solves the problem that the prior art is difficult to identify 128QAM signals.
[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:
[0006] A method for jointly identifying a modulation format and an optical signal-to-noise ratio based on a color polar coordinate diagram is provided, which comprises the following steps:
[0007] S1, performing CMA equalization processing on the signal to be identified to obtain a pre-processed digital signal;
[0008] S2, generating a color polar coordinate map based on the preprocessed digital signal;
[0009] S3, using the trained multi-task learning neural network to identify the modulation format and optical signal-to-noise ratio of the color polar coordinate image;
[0010] The signals to be identified include QPSK signals, 8QAM signals, 16QAM signals, 32QAM signals, 64QAM signals and 128QAM signals.
[0011] Furthermore, the specific method for generating a color polar coordinate map based on the preprocessed digital signal includes the following sub-steps:
[0012] S2-1, converting the preprocessed digital signal into a polar coordinate system, and obtaining the amplitude and phase of the preprocessed digital signal according to the converted real and imaginary part information;
[0013] S2-2, normalize the amplitude of the preprocessed digital signal to [0, 2.5], and normalize the phase of the preprocessed digital signal to [-3.2, 3.2];
[0014] S2-3, divide the current amplitude into 50 intervals in the range of [0,2.5], and divide the current phase into 80 intervals in the range of [-3.2,3.2], forming 50×80 squares; each square contains the amplitude and phase of the preprocessed digital signal;
[0015] S2-4. Obtain the number of symbols in the squares, map the number of symbols in each square into color information, and obtain a color polar coordinate diagram.
[0016] Furthermore, the expression for obtaining the amplitude and phase of the preprocessed digital signal according to the converted real and imaginary information is:
[0017]
[0018]
[0019] in is the amplitude of the preprocessed digital signal; is the phase of the preprocessed digital signal; is the real part after conversion; is the imaginary part after conversion; Represents the inverse tangent function.
[0020] Furthermore, when the number of symbols in the square is 0, the number of symbols is mapped to black, that is, [0 0 0].
[0021] Furthermore, when the number of symbols in the square is greater than 78, the number of symbols is mapped to white, that is, [1 1 1].
[0022] Furthermore, when the number of symbols in the grid is greater than or equal to 1 and less than or equal to 78, it is mapped to 39 different colors at intervals of 2, which are:
[0023] [1 0 0], [0 1 0], [0 0 1], [1 1 0], [1 0 1], [0 1 1], [0.5 0 0], [0 0.5 0], [0 0 0.5], [0.5 0.5 0], [0 0.5 0.5], [1 0.5 0], [0 1 0.5], [0.5 0 1], [1 0.75 0.75], [0.75 1 0.75], [0.75 0.75 1], [0.75 0.375 0], [0 0.75 0.375], [0.375 0 0.75], [1 0.25 0.25]、[0.25 1 0.25]、[0.25 0.25 1]、[0.75 0.75 0]、[0.750 0.75]、[0 0.75 0.75]、[1 0.625 0.478]、[0.478 1 0.625]、[0.625 0.478 1]、[0.8750.4375 0]、[0 0.875 0.4375]、[0.4375 0 0.875]、[1 0.375 0.75]、[0.75 1 0.375]、[0.375 0.75 1]、[0.625 0.3125 0]、[0 0.625 0.3125] and [0.3125 0 0.625].
[0024] Further, the multi-task learning neural network includes a first Conv2Dlayer module, a first feature extraction layer, a second feature extraction layer, a third feature extraction layer, an adaptive average pooling layer, a flattening layer and an output layer connected in sequence;
[0025] The first Conv2Dlayer module is used to perform preliminary feature extraction on the image input into the multi-task learning neural network;
[0026] The first feature extraction layer is used to further extract features from the output of the first Conv2Dlayer module;
[0027] A second feature extraction layer, used for further feature extraction of the output of the first feature extraction layer;
[0028] A third feature extraction layer, used for further feature extraction of the output of the second feature extraction layer;
[0029] An adaptive average pooling layer to reduce the spatial dimension of the output of the third feature extraction layer;
[0030] The flattening layer is used to flatten the output of the adaptive average pooling layer into a one-dimensional vector;
[0031] The output layer is used for modulation format recognition and optical signal-to-noise ratio recognition; modulation format recognition is a classification task; optical signal-to-noise ratio recognition is a regression task.
[0032] Furthermore, the first Conv2Dlayer module includes a traditional convolution submodule, a Fast KAN convolution submodule and an addition submodule; the traditional convolution submodule and the Fast KAN convolution submodule are in parallel, and the input of the traditional convolution submodule and the input of the FastKAN convolution submodule are jointly used as the input of the first Conv2Dlayer module; the addition submodule is used to add the output of the traditional convolution submodule and the output of the Fast KAN convolution submodule; the output of the addition submodule is the output of the first Conv2Dlayer module; wherein the traditional convolution submodule is activated by the GELU activation function.
[0033] Furthermore, the structures of the three feature extraction layers are the same, and all include a second Conv2Dlayer module, a third Conv2Dlayer module, a fourth Conv2Dlayer module, a fifth Conv2Dlayer module, and a sixth Conv2Dlayer module; in the same feature extraction layer:
[0034] The input of the second Conv2Dlayer module and the input of the third Conv2Dlayer module are the input of the feature extraction layer to which they are located; the output of the second Conv2Dlayer module is the input of the fourth Conv2Dlayer module; the output of the fourth Conv2Dlayer module and the output of the third Conv2Dlayer module are connected through residual connection as the input of the fifth Conv2Dlayer module; the output of the fifth Conv2Dlayer module is the input of the sixth Conv2Dlayer module; the input of the fifth Conv2Dlayer module and the output of the sixth Conv2Dlayer module are connected through residual connection as the output of the feature extraction layer to which they are located.
[0035] Furthermore, the training process of the multi-task learning neural network includes the following sub-steps:
[0036] S3-1, respectively obtain the color polar coordinate diagrams corresponding to the QPSK signal, 8QAM signal, 16QAM signal, 32QAM signal, 64QAM signal, and 128QAM signal under the condition of known optical signal-to-noise ratio, and use them as training samples;
[0037] The optical signal-to-noise ratio range of QPSK signal is 7~26dB; the optical signal-to-noise ratio range of 8QAM signal is 9~31dB; the optical signal-to-noise ratio range of 16QAM signal is 14~33dB; the optical signal-to-noise ratio range of 32QAM signal is 18~37dB; the optical signal-to-noise ratio range of 64QAM signal is 19~38dB; the optical signal-to-noise ratio range of 128QAM signal is 23~40dB;
[0038] S3-2. Use the actual optical signal-to-noise ratio value corresponding to each modulation format signal as a label, train the multi-task learning neural network through training samples, and update the parameters of the multi-task learning neural network in a back-propagation manner until the recognition accuracy of the multi-task learning neural network reaches a set value, or the number of training times of the multi-task learning neural network reaches a set value, to obtain a trained multi-task learning neural network.
[0039] The beneficial effects of the present invention are:
[0040] 1. This method can simultaneously realize accurate recognition of modulation formats and precise estimation of OSNR based on the slight feature differences of different modulation formats and different OSNR values on the color polar coordinate diagram through a multi-task learning neural network (MTL-Reskanet). It does not require multiple neural networks to realize modulation format recognition and OSNR monitoring respectively, which reduces the complexity and time required for the solution and improves the efficiency of joint recognition of modulation format and OSNR.
[0041] 2. This method maps the symbol number distribution of the preprocessed digital signal in the polar coordinate system into color information (RGB). The dimension of the generated color polar coordinate map features is further improved, which greatly enhances the feature discrimination and facilitates the subsequent multi-task learning neural network to train and identify the modulation format and OSNR information. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 is a schematic diagram of the process of this method;
[0043] Figure 2 A diagram showing a simulation setting of a coherent optical communication system in an embodiment of the present invention;
[0044] Figure 3 for Figure 2 The architecture diagram of the digital signal processing module in the figure;
[0045] Figure 4 It is the top view corresponding to the color polar coordinate diagram of the QPSK signal;
[0046] Figure 5 It is the top view corresponding to the color polar coordinate diagram of 8QAM signal;
[0047] Figure 6It is the top view corresponding to the color polar coordinate diagram of 16QAM signal;
[0048] Figure 7 It is the top view corresponding to the color polar coordinate diagram of 32QAM signal;
[0049] Figure 8 It is the top view corresponding to the color polar coordinate diagram of 64QAM signal;
[0050] Fig. 9 It is the top view corresponding to the color polar coordinate diagram of 128QAM signal;
[0051] Fig.10 Diagram of neural network architecture for multi-task learning;
[0052] Fig.11 The performance of multi-task learning neural networks during training and the relationship between the number of training rounds;
[0053] Fig.12 The performance of multi-task learning neural networks is related to the number of training rounds during verification;
[0054] Fig.13 This is a graph showing the modulation format recognition accuracy changing with the optical signal-to-noise ratio;
[0055] Fig.14 It is the relationship diagram between the real optical signal-to-noise ratio value and the estimated optical signal-to-noise ratio value of the QPSK signal;
[0056] Fig.15 It is the relationship diagram between the real optical signal-to-noise ratio value and the estimated optical signal-to-noise ratio value of the 8QAM signal;
[0057] Fig.16 It is the relationship diagram between the real optical signal-to-noise ratio value and the estimated optical signal-to-noise ratio value of the 16QAM signal;
[0058] Fig.17 It is the relationship diagram between the real optical signal-to-noise ratio value and the estimated optical signal-to-noise ratio value of the 32QAM signal;
[0059] Fig.18 It is the relationship diagram between the real optical signal-to-noise ratio value and the estimated optical signal-to-noise ratio value of the 64QAM signal;
[0060] Fig.19 It is the relationship diagram between the real optical signal-to-noise ratio value and the estimated optical signal-to-noise ratio value of the 128QAM signal;
[0061] Fig. 20 is the average absolute error of the optical signal-to-noise ratio estimation of the six signals under different optical signal-to-noise ratio conditions. DETAILED DESCRIPTION
[0062] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.
[0063] like Figure 1 As shown, the modulation format and optical signal-to-noise ratio joint identification method based on the color polar coordinate diagram includes the following steps:
[0064] S1, performing CMA (constant modulus algorithm) equalization processing on the signal to be identified to obtain a pre-processed digital signal;
[0065] S2, generating a color polar coordinate map based on the preprocessed digital signal;
[0066] S3, using the trained multi-task learning neural network to identify the modulation format and optical signal-to-noise ratio of the color polar coordinate image;
[0067] The signals to be identified include QPSK (Quadrature Phase Shift Keying) signals, 8QAM signals, 16QAM signals, 32QAM signals, 64QAM signals and 128QAM signals.
[0068] The specific method for generating a color polar coordinate map based on the preprocessed digital signal includes the following sub-steps:
[0069] S2-1, converting the preprocessed digital signal into a polar coordinate system, and obtaining the amplitude and phase of the preprocessed digital signal according to the converted real and imaginary part information;
[0070] S2-2, normalize the amplitude of the preprocessed digital signal to [0, 2.5], and normalize the phase of the preprocessed digital signal to [-3.2, 3.2];
[0071] S2-3, divide the current amplitude into 50 intervals in the range of [0,2.5], and divide the current phase into 80 intervals in the range of [-3.2,3.2], forming 50×80 squares; each square contains the amplitude and phase of the preprocessed digital signal;
[0072] S2-4. Obtain the number of symbols in the squares, map the number of symbols in each square into color information, and obtain a color polar coordinate diagram.
[0073] The color polar coordinate diagram maps the symbol number information into color information (RGB) in the form of polar coordinates within a grid composed of different phases and amplitudes, which increases the dimension of the information, greatly reduces the difficulty of subsequent neural network recognition, and improves recognition accuracy.
[0074] The digital signal expression in the polar coordinate system is , A preprocessed complex digital signal can be represented by real and imaginary parts. The expression for obtaining the amplitude and phase of the preprocessed digital signal based on the converted real and imaginary information is:
[0075]
[0076]
[0077] in is the amplitude of the preprocessed digital signal; is the phase of the preprocessed digital signal; is the real part after conversion; is the imaginary part after conversion; Represents the inverse tangent function.
[0078] The above operation normalizes the amplitude and phase information of the signal to a specific range, and divides the image into 50×80 squares according to the amplitude and phase information. These squares are formed by dividing the amplitude and phase information of the signal at equal intervals. Therefore, the position of the amplitude and phase information of each interval in the two-dimensional space is determined. Each square contains the amplitude, phase and number of symbols in the square of the signal. However, when the number of symbols is determined, the number of symbols in the interval will vary with the type of modulation format and the OSNR value of the signal. Therefore, this embodiment sets color information in the square to represent the number of symbols, which can indirectly reflect the amplitude and phase information of the signal. The number of symbols can be used to infer which squares the signal appears in. Each square represents specific amplitude and phase information. The amplitude of the QPSK signal presents a single-level distribution, and the total number of symbols will be concentrated in the square corresponding to the single-level distribution; the amplitude of the 8QAM signal presents a two-level distribution, and the total number of symbols will be concentrated in the square corresponding to the two-level distribution; the amplitude of the 16QAM signal presents a three-level distribution, and the total number of symbols will be concentrated in the square corresponding to the three-level distribution; the amplitude of the 32QAM signal presents a five-level distribution, and the total number of symbols will be concentrated in the square corresponding to the five-level distribution; the amplitude of the 64QAM signal presents a nine-level distribution, and the total number of symbols will be concentrated in the square corresponding to the nine-level distribution; the 128QAM signal presents an 11-level distribution, and the total number of symbols will be concentrated in the square corresponding to the 11-level distribution. Since the total number of symbols is fixed, and the distribution level of the signal amplitude of each modulation format is different, the number of squares corresponding to the distribution level is also different. Therefore, the number of symbols allocated to each square is also different. In this way, information that originally belonged to the three-dimensional space can be displayed on a two-dimensional plane: the amplitude and phase of the signal and the number of symbols in the corresponding square.
[0079] In this embodiment, when the number of symbols in the square is 0, the symbol number is mapped to black, that is, [0 0 0]. When the number of symbols in the square is greater than 78, the symbol number is mapped to white, that is, [1 1 1]. When the number of symbols in the square is greater than or equal to 1 and less than or equal to 78, it is mapped to 39 different colors (RGB) at intervals of 2, namely:
[0080] [1 0 0], [0 1 0], [0 0 1], [1 1 0], [1 0 1], [0 1 1], [0.5 0 0], [0 0.5 0], [0 0 0.5], [0.5 0.5 0], [0 0.5 0.5], [1 0.5 0], [0 1 0.5], [0.5 0 1], [1 0.75 0.75], [0.75 1 0.75], [0.75 0.75 1], [0.75 0.375 0], [0 0.75 0.375], [0.375 0 0.75], [1 0.25 0.25]、[0.25 1 0.25]、[0.25 0.25 1]、[0.75 0.75 0]、[0.750 0.75]、[0 0.75 0.75]、[1 0.625 0.478]、[0.478 1 0.625]、[0.625 0.478 1]、[0.8750.4375 0]、[0 0.875 0.4375]、[0.4375 0 0.875]、[1 0.375 0.75]、[0.75 1 0.375]、[0.375 0.75 1]、[0.625 0.3125 0]、[0 0.625 0.3125] and [0.3125 0 0.625].
[0081] It should be noted that if the order of the above arrays changes, it will basically not affect the final recognition accuracy, etc., because these arrays represent color information, and the color information represents the number of symbols in the squares. The number of symbols in the squares is the fundamental feature, so other distinguishable colors can be used to represent the number of symbols in different squares.
[0082] like Fig.10 As shown, the multi-task learning neural network includes a first Conv2Dlayer module, a first feature extraction layer, a second feature extraction layer, a third feature extraction layer, an adaptive average pooling layer, a flattening layer and an output layer connected in sequence;
[0083] The first Conv2Dlayer module is used to perform preliminary feature extraction on the image input into the multi-task learning neural network, and the output feature size is 8×224×224;
[0084] The first feature extraction layer (corresponding to Fig.10 Layer1 in ), used to further extract features from the output of the first Conv2Dlayer module, with an output feature size of 16×112×112;
[0085] The second feature extraction layer (corresponding to Fig.10 Layer2 in ), used to further extract features from the output of the first feature extraction layer, with an output feature size of 32×56×56;
[0086] The third feature extraction layer (corresponding to Fig.10 Layer3 in ), used to further extract features from the output of the second feature extraction layer, with an output feature size of 64×28×28;
[0087] Adaptive average pooling layer (corresponding to Fig.10 AdaptivaAvgpool in ), which is used to reduce the spatial dimension of the output of the third feature extraction layer and improve the generalization ability of the model. The output feature size is 64×1×1;
[0088] Flattening layer (corresponding to Fig.10 The Flatten Layer in the adaptive average pooling layer is used to flatten the output of the adaptive average pooling layer into a one-dimensional vector with a length of 64;
[0089] The output layer (corresponding to Fig.10 OSNR and MF in the image are used for modulation format recognition and optical signal-to-noise ratio recognition; modulation format recognition is a classification task; optical signal-to-noise ratio recognition is a regression task.
[0090] The first Conv2Dlayer module includes a traditional convolution submodule, a Fast KAN convolution submodule and an addition submodule; the traditional convolution submodule and the Fast KAN convolution submodule are in parallel, and the input of the traditional convolution submodule and the input of the Fast KAN convolution submodule are used as the input of the first Conv2Dlayer module; the addition submodule is used to add the output of the traditional convolution submodule and the output of the Fast KAN convolution submodule; the output of the addition submodule is the output of the first Conv2Dlayer module; wherein the traditional convolution submodule is activated by the GELU activation function.
[0091] Fast KAN convolution is very similar to traditional convolution, but instead of applying a dot product between the kernel and the corresponding pixel in the image, a learnable nonlinear function (radial basis function) is applied to each element and then they are summed.
[0092] The structures of the three feature extraction layers are the same, including the second Conv2Dlayer module, the third Conv2Dlayer module, the fourth Conv2Dlayer module, the fifth Conv2Dlayer module and the sixth Conv2Dlayer module; in the same feature extraction layer:
[0093] The input of the second Conv2Dlayer module and the input of the third Conv2Dlayer module are the input of the feature extraction layer to which they are located; the output of the second Conv2Dlayer module is the input of the fourth Conv2Dlayer module; the output of the fourth Conv2Dlayer module and the output of the third Conv2Dlayer module are connected through residual connection as the input of the fifth Conv2Dlayer module; the output of the fifth Conv2Dlayer module is the input of the sixth Conv2Dlayer module; the input of the fifth Conv2Dlayer module and the output of the sixth Conv2Dlayer module are connected through residual connection as the output of the feature extraction layer to which they are located.
[0094] In this embodiment, the training process of the multi-task learning neural network is:
[0095] Obtain the color polar coordinate diagrams corresponding to the QPSK signal, 8QAM signal, 16QAM signal, 32QAM signal, 64QAM signal, and 128QAM signal under the condition of known optical signal-to-noise ratio, and use them as training samples;
[0096] The optical signal-to-noise ratio range of the QPSK signal is 7-26dB; the optical signal-to-noise ratio range of the 8QAM signal is 9-31dB; the optical signal-to-noise ratio range of the 16QAM signal is 14-33dB; the optical signal-to-noise ratio range of the 32QAM signal is 18-37dB; the optical signal-to-noise ratio range of the 64QAM signal is 19-38dB; the optical signal-to-noise ratio range of the 128QAM signal is 23-40dB; in order to control the number of training samples, the optical signal-to-noise ratio added to each signal is selected in this embodiment at an interval of 1dB, that is, the six signals contain 20, 23, 20, 20, 20, and 18 OSNR values respectively; the total number of symbols corresponding to each color polar coordinate diagram is 5000, and a certain OSNR value of each modulation format contains 100 sets of color polar coordinate diagrams as samples, so a total of 12100 sets of samples are obtained;
[0097] The 12100 groups of samples were divided into training samples and verification samples in a ratio of 8:2. The classification task labels of the six modulation format signals were set to '0', '1', '2', '3', '4' and '5', respectively. The regression task labels were set to the real optical signal-to-noise ratio values corresponding to each modulation format signal. The multi-task learning neural network was trained, and the parameters of the multi-task learning neural network were updated by back propagation until the recognition accuracy of the multi-task learning neural network reached the set value, or the number of training times of the multi-task learning neural network reached the set value, thus obtaining the trained multi-task learning neural network.
[0098] In one embodiment of the present invention, in order to verify the feasibility of the method, a numerical simulation of 28GBaud polarization multiplexing QPSK / 8QAM / 16QAM / 32QAM / 64QAM / 128QAM signal transmission is performed. The simulation block diagram of the 28GBaud coherent optical communication system is shown in FIG. Figure 2 As shown, Figure 2 In it (I and Q are the in-phase component and the orthogonal component respectively), it includes the transmitting end and the receiving end. At the transmitting end, the wavelength of the laser is 1550nm and the line width is 100kHz. The electrical signal drives the IQ modulator to generate QPSK / 8QAM / 16QAM / 32QAM / 64QAM / 128QAM signals with a transmission rate of 28GBaud, and is converted into polarization multiplexing signals through the polarization combiner to enter the transmission channel. The optical signal-to-noise ratio setting module can adjust the OSNR value of the signal. At the receiving end, the out-of-band noise of the optical signal is first filtered out by an optical bandpass filter, and then the optical signal is divided into two paths by a polarization beam splitter, and the two local oscillator optical signals separated by the local oscillator laser are mixed at 90° respectively. The mixed signal is converted into an electrical signal by a balanced detector, and the electrical signal is filtered based on a low-pass filter. Finally, the electrical signal is digitized by an analog-to-digital converter for subsequent digital signal processing. Figure 3 As shown, E x and E y They are mutually orthogonal signals on the same wavelength received by the coherent optical receiver. x polarization and y polarization signals, the entire digital signal processing part includes three types of algorithms: modulation format non-related algorithm, this method and modulation format related algorithm, among which the modulation format non-related algorithm and the modulation format related algorithm are both existing algorithms and are not the creativity of this application.
[0099] In this embodiment, the color polar coordinate diagrams of QPSK / 8QAM / 16QAM / 32QAM / 64QAM / 128QAM signals are respectively as follows: Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 and Fig. 9 shown.
[0100] Since the performance of the training set and the validation set can be used as a criterion for judging whether the multi-task learning neural network is overfitting or underfitting, this embodiment analyzes the performance of the training set and the validation set in detail. Fig.11 As shown in the figure, for the training curve, the accuracy of modulation format recognition reached 100% after 277 training rounds, and the training curve of the mean absolute error of optical signal-to-noise ratio estimation stabilized at 0.58 dB after 569 training rounds. Fig.12As shown in the figure, for the validation set, the accuracy of modulation format recognition reached 100% after 181 training rounds, and the training curve of the estimated optical signal-to-noise ratio mean absolute error stabilized at 0.315 dB after 547 training rounds. Therefore, the performance difference between the training set and the validation set is not large, indicating that the multi-task learning neural network has good generalization ability.
[0101] The modulation format recognition accuracy of this method under different OSNR conditions is as follows: Fig.13 As shown in the figure, the ordinate is the correct recognition rate (%), the abscissa is the optical signal-to-noise ratio (dB), the OSNR ranges of QPSK, 8QAM, 16QAM, 32QAM, 64QAM, and 128QAM signals are 7~26dB, 9~31dB, 14~33dB, 18~37dB, 19~38dB, and 23~40dB, respectively, and the interval of OSNR is 1dB. For 28GBaud polarization multiplexed QPSK / 8QAM / 16QAM / 32QAM / 64QAM / 128QAM signals, 100% correct recognition rate is achieved in the entire OSNR range, and the lowest optical signal-to-noise ratio corresponding to 7dB, 9dB, 14dB, 18dB, 19dB, and 23dB are all lower than the optical signal-to-noise ratio threshold corresponding to 20% forward error correction (FEC) of each modulation format (corresponding to Fig.13 The vertical dashed line in the figure shows the bit error rate (BER) = 2.4 × 10 -2 ). Fig.13 FEC_X indicates 20% forward error correction for X modulation format.
[0102] The relationships between the real OSNR values and the estimated OSNR values of the six signals proposed by this method are as follows: Fig.14 , Fig.15 , Fig.16 , Fig.17 , Fig.18 and Fig.19 As shown, the reference line of the real OSNR value is described by a black diagonal line, the horizontal axis represents the real OSNR value, and the vertical axis represents the estimated OSNR value. It can be seen that when the OSNR value is low or high, the estimation error is slightly larger than the intermediate OSNR value, and the estimation error of the high-order modulation format is larger than that of the low-order modulation format. The reason is that at low OSNR, the large fluctuation of the characteristics caused by noise increases the difficulty of OSNR estimation. At high OSNR, the influence of noise is greatly reduced, and the distinguishability of the characteristics of adjacent OSNR values is significantly reduced, which increases the difficulty of OSNR estimation.
[0103] In addition, the average absolute error of the optical signal-to-noise ratio estimation of the six signals under different optical signal-to-noise ratios is as follows: Fig. 20As shown in the figure, at low and high OSNR values, the mean absolute error of each modulation format is higher than that at the intermediate OSNR value, that is, the estimation error at low and high OSNR values is larger than that at the intermediate OSNR value. However, the method still shows excellent performance, and the average mean absolute error of the OSNR estimation for each modulation format is less than 1dB, as shown in the figure. Fig. 20 As shown by the dotted lines in , the average mean absolute errors of the six modulation formats are 0.23dB, 0.26dB, 0.23dB, 0.26dB, 0.39dB and 0.56dB respectively.
[0104] In summary, the present invention can synchronously realize accurate recognition of modulation formats and precise estimation of OSNR according to the slight feature differences of different modulation formats and different OSNR values on the color polar coordinate diagram through multi-task learning neural network. There is no need for multiple neural networks to realize modulation format recognition and OSNR monitoring respectively, which reduces the complexity and time required for the solution and improves the efficiency of joint recognition of modulation format and OSNR.
Claims
1. A method for joint identification of modulation format and optical signal-to-noise ratio based on color polar coordinate diagram, characterized in that: The following steps are involved: S1, performing CMA equalization processing on the signal to be identified to obtain a pre-processed digital signal; S2, generating a color polar coordinate map based on the preprocessed digital signal; S3, using the trained multi-task learning neural network to identify the modulation format and optical signal-to-noise ratio of the color polar coordinate image; The signals to be identified include QPSK signals, 8QAM signals, 16QAM signals, 32QAM signals, 64QAM signals and 128QAM signals; The specific method for generating a color polar coordinate map based on the preprocessed digital signal includes the following sub-steps: S2-1, converting the preprocessed digital signal into a polar coordinate system, and obtaining the amplitude and phase of the preprocessed digital signal according to the converted real and imaginary part information; S2-2, normalize the amplitude of the preprocessed digital signal to [0, 2.5], and normalize the phase of the preprocessed digital signal to [-3.2, 3.2]; S2-3, divide the current amplitude into 50 intervals in the range of [0,2.5], and divide the current phase into 80 intervals in the range of [-3.2,3.2], forming 50×80 squares; each square contains the amplitude and phase of the preprocessed digital signal; S2-4, obtaining the number of symbols in the squares, mapping the number of symbols in each square into color information, and obtaining a color polar coordinate diagram; The multi-task learning neural network includes a first Conv2Dlayer module, a first feature extraction layer, a second feature extraction layer, a third feature extraction layer, an adaptive average pooling layer, a flattening layer and an output layer connected in sequence; The first Conv2Dlayer module is used to perform preliminary feature extraction on the image input into the multi-task learning neural network; The first feature extraction layer is used to further extract features from the output of the first Conv2Dlayer module; A second feature extraction layer, used for further feature extraction of the output of the first feature extraction layer; A third feature extraction layer, used for further feature extraction of the output of the second feature extraction layer; An adaptive average pooling layer to reduce the spatial dimension of the output of the third feature extraction layer; The flattening layer is used to flatten the output of the adaptive average pooling layer into a one-dimensional vector; Output layer, used for modulation format recognition and optical signal-to-noise ratio recognition; Among them, modulation format recognition is a classification task; optical signal-to-noise ratio recognition is a regression task.
2. The method for joint identification of modulation format and optical signal-to-noise ratio based on color polar coordinate diagram according to claim 1, characterized in that: The expression for obtaining the amplitude and phase of the preprocessed digital signal based on the converted real and imaginary information is: in is the amplitude of the preprocessed digital signal; is the phase of the preprocessed digital signal; is the real part after conversion; is the imaginary part after conversion; Represents the inverse tangent function.
3. The method for joint identification of modulation format and optical signal-to-noise ratio based on color polar coordinate diagram according to claim 1, characterized in that: When the number of symbols in the square is 0, the symbol number is mapped to black, that is, [0 0 0].
4. The method for joint identification of modulation format and optical signal-to-noise ratio based on color polar coordinate diagram according to claim 1, characterized in that: When the number of symbols in the square is greater than 78, the number of symbols is mapped to white, that is, [1 1 1].
5. The method for joint identification of modulation format and optical signal-to-noise ratio based on color polar coordinate diagram according to claim 1, characterized in that: When the number of symbols in the grid is greater than or equal to 1 and less than or equal to 78, it is mapped to 39 different colors at intervals of 2, which are: [1 0 0], [0 1 0], [0 0 1], [1 1 0], [1 0 1], [0 1 1], [0.5 0 0], [0 0.5 0], [0 00.5], [0.5 0.5 0], [0 0.5 0.5], [1 0.5 0], [0 1 0.5], [0.5 0 1], [10.75 0.75], [0.75 1 0.75], [0.75 0.75 1], [0.75 0.375 0], [0 0.75 0.375], [0.375 00.75], [1 0.25 0.25], [0.25 1 0.25], [0.25 0.25 1], [0.75 0.75 0], [0.75 0 0.75], [0 0.75 0.75], [1 0.625 0.478], [0.478 1 0.625], [0.625 0.478 1], [0.875 0.43750], [0 0.875 0.4375], [0.4375 0 0.875], [1 0.375 0.75], [0.75 1 0.375], [0.3750.75 1], [0.625 0.3125 0], [0 0.625 0.3125] and [0.3125 0 0.625].
6. The method for joint identification of modulation format and optical signal-to-noise ratio based on color polar coordinate diagram according to claim 1, characterized in that: The first Conv2Dlayer module includes a traditional convolution submodule, a Fast KAN convolution submodule and an addition submodule; the traditional convolution submodule and the Fast KAN convolution submodule are in parallel, and the input of the traditional convolution submodule and the input of the Fast KAN convolution submodule are used together as the input of the first Conv2Dlayer module; The addition submodule is used to add the output of the traditional convolution submodule and the output of the Fast KAN convolution submodule; the output of the addition submodule is the output of the first Conv2Dlayer module; wherein the traditional convolution submodule is activated by the GELU activation function.
7. The method for joint identification of modulation format and optical signal-to-noise ratio based on color polar coordinate diagram according to claim 1, characterized in that: The structures of the three feature extraction layers are the same, including the second Conv2Dlayer module, the third Conv2Dlayer module, the fourth Conv2Dlayer module, the fifth Conv2Dlayer module and the sixth Conv2Dlayer module; in the same feature extraction layer: The input of the second Conv2Dlayer module and the input of the third Conv2Dlayer module are the input of the feature extraction layer to which they are located; the output of the second Conv2Dlayer module is the input of the fourth Conv2Dlayer module; the output of the fourth Conv2Dlayer module and the output of the third Conv2Dlayer module are connected through residual connection as the input of the fifth Conv2Dlayer module; the output of the fifth Conv2Dlayer module is the input of the sixth Conv2Dlayer module; the input of the fifth Conv2Dlayer module and the output of the sixth Conv2Dlayer module are connected through residual connection as the output of the feature extraction layer to which they are located.
8. The method for joint identification of modulation format and optical signal-to-noise ratio based on color polar coordinate diagram according to claim 1, characterized in that: The training process of a multi-task learning neural network consists of the following sub-steps: S3-1, respectively obtain the color polar coordinate diagrams corresponding to the QPSK signal, 8QAM signal, 16QAM signal, 32QAM signal, 64QAM signal, and 128QAM signal under the condition of known optical signal-to-noise ratio, and use them as training samples; The optical signal-to-noise ratio range of QPSK signal is 7~26dB; the optical signal-to-noise ratio range of 8QAM signal is 9~31dB; the optical signal-to-noise ratio range of 16QAM signal is 14~33dB; the optical signal-to-noise ratio range of 32QAM signal is 18~37dB; the optical signal-to-noise ratio range of 64QAM signal is 19~38dB; the optical signal-to-noise ratio range of 128QAM signal is 23~40dB; S3-2. Use the actual optical signal-to-noise ratio value corresponding to each modulation format signal as a label, train the multi-task learning neural network through training samples, and update the parameters of the multi-task learning neural network in a back-propagation manner until the recognition accuracy of the multi-task learning neural network reaches a set value, or the number of training times of the multi-task learning neural network reaches a set value, to obtain a trained multi-task learning neural network.
Citation Information
Patent Citations
Road scene recognition method and system based on multi-task learning neural network
CN112183395A
New signal modulation format recognition method based on neural network
CN112953871A
Neural network-based modulation signal identification method and application thereof
CN113361433A
Key parameter joint monitoring method in elastic optical network
CN113452439A
Cascade modulation format identification and optical signal-to-noise ratio monitoring method
CN115733566A
