CN114598581A - Two-stage detection model training method, identification method and device for probabilistic shaped signal - Google Patents

Abstract

The invention provides a two-stage detection model training method, identification method and device for probabilistic shaping signals. The electrical domain signal obtained by processing the DSP module at the signal receiving end is based on the amplitude histogram processed by the blind equalization algorithm. The mapping of the amplitude histogram to the optical signal-to-noise ratio and the modulation format is constructed in a manner to realize the identification of the optical signal-to-noise ratio and the modulation format of the probability shaped signal. At the same time, since the phase noise of the signal laser and the local oscillator laser under the Lorentzian line type is a micro-nano process, and the variance of the phase noise increment is linearly related to the laser linewidth, different symbol intervals are constructed by machine learning. The root mean square root and average value of the lower phase noise increments are mapped to the laser linewidth to realize the identification of the laser linewidth and effectively improve the generalization ability of the identification.

Description

Two-stage detection model training method, identification method and device for probabilistic shaped signal

technical field

The invention relates to the technical field of optical communication, in particular to a training method, identification method and device for a two-stage detection model of a probability shaping signal.

Background technique

In recent years, new technologies such as cloud devices and smart devices have developed rapidly, and data traffic has continued to grow. With the development of optical communication, the capacity of single-mode fiber has gradually approached the Shannon limit. In order to cope with the rapidly increasing data traffic demand, the resources of the physical layer need to be used more efficiently. The elastic optical network can adaptively adjust the modulation format and data rate according to channel conditions and capacity requirements, and can better allocate physical resources. To enable elastic optical networks to flexibly control optical transmission systems, optical performance monitoring and modulation format identification have been extensively studied.

The parameters of optical signal-to-noise ratio (OSNR) and laser linewidth are extremely important for the design of transmission systems in elastic optical networks, and monitoring of laser linewidth is also helpful in diagnosing sudden anomalies in lasers. In addition, modulation format identification (MFI) is particularly important for the selection of receiver algorithms in elastic optical networks. With the development of digital signal processing (DSP) technology, optical performance monitoring has shifted from the optical domain to the electrical domain, which further reduces the monitoring cost. To achieve high-precision performance monitoring, machine learning techniques are widely used.

In coherent optical communication systems, OSNR and laser linewidth are two key indicators for measuring amplitude noise and phase noise. However, there is no detection method for probability shaped waves in the prior art, and laser linewidth cannot be identified at the same time.

SUMMARY OF THE INVENTION

The embodiments of the present invention provide a two-stage detection model training method, identification method and device for probabilistic shaping signals, so as to eliminate or improve one or more defects existing in the prior art, and solve the problem that the prior art cannot effectively identify probabilistic shaping signals wave and the inability to identify the laser linewidth.

The technical scheme of the present invention is as follows:

In one aspect, the present invention provides a two-stage detection model training method for probabilistic shaped signals, including:

Acquire a training sample set, the training sample set includes multiple samples, and each sample includes an amplitude histogram obtained by processing a DSP module at the receiving end of the elastic optical network through a blind equalization algorithm under various signal modulation formats; each sample also includes a phase Noise; mark the optical signal-to-noise ratio, modulation format and laser line width corresponding to each sample as a label; wherein, the elastic optical network adopts a Lorentzian line-type signal laser and a local oscillator laser;

Obtain a first initial model and a second initial model, where the first initial model and the second initial model are ANN (Artificial Neural Network, artificial neural network) networks; wherein, the first initial model includes an input layer, A shared hidden layer, a first specific hidden layer and a second specific hidden layer respectively connecting the shared hidden layer, the first specific hidden layer is used to identify the modulation format and connect the first output layer, the second specific hidden layer Used to identify the optical signal-to-noise ratio and connect the second output layer;

Taking the amplitude histogram of each sample as the input, taking the optical signal-to-noise ratio and the modulation format as the output, using the training sample set to train the first initial model to obtain the optical signal-to-noise ratio and modulation format recognition model;

Taking the sequence formed by the root mean square value and the average value of the phase noise increments of each sample at different symbol intervals as input, and taking the laser linewidth as output, the second initial model is trained by using the training sample set, and the result is obtained: Laser linewidth identification model.

In some embodiments, the samples in the training sample set are generated under five modulation formats of QPSK, 16QAM, PS-16QAM, 64QAM and PS-64QAM.

In some embodiments, the training sample set is in the range of the laser line width of 50KHz to 500KHz, and the amplitude histogram is sampled according to the step size of 50KHz. At the same time, for the elastic optical network using the QPSK modulation format, the optical signal-to-noise ratio is Amplitude histogram sampling for each parameter in the range of 10 to 25dB; amplitude histogram sampling for each parameter in the range of optical signal-to-noise ratio 15 to 30dB for elastic optical networks using 16QAM and PS-16QAM modulation formats ; For elastic optical networks using 64QAM and PS-64QAM modulation formats, amplitude histogram sampling is performed for each parameter in the range of optical signal-to-noise ratio 20 to 35 dB; for each laser linewidth, modulation format and optical signal-to-noise ratio The combination collects the first set number of samples respectively.

In some embodiments, the amplitude histogram includes an amplitude histogram of a dual polarization state signal.

In some embodiments, the training sample set is in the range of the laser line width of 50KHz to 500KHz, and the phase noise is collected according to the step size of 50KHz. At the same time, for the elastic optical network using the QPSK, 16QAM and PS-16QAM modulation formats, The phase noise is collected separately for each parameter in the range of optical signal-to-noise ratio of 20 to 30dB; for elastic optical networks using PS-64QAM and 64QAM modulation formats, each parameter is collected in the range of optical signal-to-noise ratio of 30 to 40dB. Phase noise collection; separately collect a second set number of samples for each combination of laser linewidth, modulation format, and optical signal-to-noise ratio.

In some embodiments, the input is a sequence formed by the root mean square value and the average value of the phase noise increments of each sample at different symbol intervals, including:

For signal lasers and local oscillator lasers using the Lorentzian line type, the phase noise is a Wiener process, expressed as:

的表达式为：Among them, w(n) is a sequence that obeys a Gaussian distribution, the mean of w(n) is 0, and the variance is

The expression is:

where Δv is the sum of the full width at half maximum of the signal and the LO, and τ _s is the time interval between adjacent signals;

的表达式为：Then the phase noise increments

The expression is:

Among them, abs( ) is the absolute value function, N represents the number of symbol intervals, and n represents the number of time series;

Calculate the phase noise increments corresponding to multiple set symbol intervals, and obtain the root mean square value and average value corresponding to each symbol interval;

The root mean square value and the average value of the phase noise increment corresponding to each symbol interval number form a sequence, which is used as the input of the second initial model.

In some embodiments, in calculating the phase noise increment corresponding to the plurality of preset numbers of symbol intervals, the multiple preset numbers of symbol intervals include 2, 5, and 10.

On the other hand, the present invention also provides a two-stage detection method for a probability shaped signal, comprising:

Obtaining a histogram of the amplitude to be detected obtained by processing the blind equalization algorithm of the signal to be detected, and inputting the histogram of the amplitude to be detected into the optical signal-to-noise ratio and modulation format recognition model in the above-mentioned two-stage detection model training method for probabilistic shaped signals, and output the optical signal-to-noise ratio and modulation format identification result of the signal to be detected;

Obtain the root mean square value and average value of the phase noise increment of the signal to be detected at different symbol intervals, and form a sequence to be detected, and input the sequence to be detected into the laser line in the above-mentioned two-stage detection model training method for probability-shaping signals Wide recognition model and output laser line type recognition results.

In another aspect, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, where the processor implements the steps of the above method when the processor executes the program.

In another aspect, the present invention also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the steps of the above method.

The beneficial effects of the present invention are:

In the two-stage detection model training method, identification method and device of the probability shaping signal of the present invention, the electrical domain signal obtained by processing the DSP module at the signal receiving end is based on the amplitude histogram processed by the blind equalization algorithm, and is processed by machine learning. The mapping of the amplitude histogram to the optical signal-to-noise ratio and the modulation format is constructed in a manner to realize the identification of the optical signal-to-noise ratio and the modulation format of the probability shaped signal. At the same time, since the phase noise of the signal laser and the local oscillator laser under the Lorentzian line type is a micro-nano process, and the variance of the phase noise increment is linearly related to the laser linewidth, the phase noise increment is constructed by machine learning. The rms and average values of the quantities are mapped to the laser linewidth to enable identification of the laser linewidth.

Further, by introducing the root mean square value and average value of phase noise increments at different symbol intervals to form a sequence, and constructing the mapping relationship between the sequence and the laser linewidth by means of machine learning, the generalization ability of recognition can be effectively improved.

Additional advantages, objects, and features of the present invention will be set forth in part in the description that follows, and in part will become apparent to those of ordinary skill in the art upon study of the following, or may be learned from practice of the invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

Those skilled in the art will appreciate that the objects and advantages that can be achieved with the present invention are not limited to those specifically described above, and that the above and other objects that can be achieved by the present invention will be more clearly understood from the following detailed description.

Description of drawings

The accompanying drawings described herein are used to provide a further understanding of the present invention, and constitute a part of the present application, and do not constitute a limitation to the present invention. In the attached image:

1 is a schematic flowchart of a method for training a two-stage detection model for probabilistic shaped signals according to an embodiment of the present invention;

FIG. 2 is a logical schematic diagram of a method flow for training a two-stage detection model for probabilistic shaped signals according to an embodiment of the present invention;

3 is a schematic structural diagram of a first initial model in a method for training a two-stage detection model for probabilistic shaped signals according to an embodiment of the present invention;

4 is a schematic structural diagram of a second initial model in the two-stage detection model training method for probabilistic shaped signals according to an embodiment of the present invention;

Fig. 5 is a histogram of amplitude distribution of signals under different modulations with different optical signal-to-noise ratios;

Fig. 6 is the phase noise curve graph under different laser linewidths;

7 is an estimation result of the optical signal-to-noise ratio of the two-stage optical signal-to-noise ratio of the probability-shaping signal and the laser linewidth monitoring scheme according to an embodiment of the present invention;

FIG. 8 is an estimation result of the laser linewidth by the two-stage optical signal-to-noise ratio of the probability-shaping signal and the laser linewidth monitoring scheme according to an embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but not to limit the present invention.

Here, it should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only the structures and/or processing steps closely related to the solution according to the present invention are shown in the drawings, and the related structures and/or processing steps are omitted. Other details not relevant to the invention.

It should be emphasized that the term "comprising/comprising" when used herein refers to the presence of a feature, element, step or component, but does not exclude the presence or addition of one or more other features, elements, steps or components.

Here, it should also be noted that, if there is no special description, the term "connection" herein may not only refer to direct connection, but also to indicate indirect connection with intermediates.

In order to efficiently detect the modulation format, optical signal-to-noise ratio and laser linewidth of the probability shaping signal, the optical performance detection is shifted from the optical domain to the electrical domain through the digital signal processing module (DSP module), which reduces the detection cost and improves the performance.

Specifically, the present invention provides a two-stage detection model training method for probabilistic shaped signals, as shown in FIG. 1 , including steps S101 to S104:

Step S101: Acquire a training sample set, the training sample set includes multiple samples, and each sample includes an amplitude histogram obtained by processing a DSP module at the receiving end of the elastic optical network through a blind equalization algorithm under a variety of signal modulation formats; each sample also includes Phase noise; mark the optical signal-to-noise ratio, modulation format and laser line width corresponding to each sample as a label; among them, the elastic optical network adopts Lorentzian line type signal laser and local oscillator laser;

Step S102: Obtain a first initial model and a second initial model, where the first initial model and the second initial model are ANN networks; wherein, the first initial model includes an input layer, a shared hidden layer, and a first initial model that is respectively connected to the shared hidden layer. The specific hidden layer and the second specific hidden layer, the first specific hidden layer is used to identify the modulation format and connect the first output layer, and the second specific hidden layer is used to identify the optical signal-to-noise ratio and connect the second output layer.

Step S103: Take the amplitude histogram of each sample as input, and take the OSR and modulation format as output, use the training sample set to train the first initial model, and obtain the optical signal-to-noise ratio and modulation format recognition model.

Step S104: Take the sequence formed by the root mean square value and the average value of the phase noise increments of each sample at different symbol intervals as input, take the laser line width as output, use the training sample set to train the second initial model, and obtain the laser Line width recognition model.

In this embodiment, the optical signal-to-noise ratio, modulation format and laser linewidth are identified by using the electrical signal processed by the DSP module in the optical network, and machine learning is introduced to construct a mapping relationship to achieve efficient identification. Specifically, in this embodiment, a two-stage identification model is constructed, and the optical signal-to-noise ratio, modulation format, and laser linewidth are identified by constructing two sets of mapping relationships.

In step S101, the optical signal obtained by the probability shaping at the transmitting end is transmitted through an optical fiber, and amplified by an optical fiber amplifier. digital signal processing. The amplitude histogram obtained through blind equalization algorithm processing is used as a parameter for identifying optical signal-to-noise ratio and modulation format. Specifically, the amplitude histogram of the signal distribution after processing by the constant modulus algorithm can be used. At the same time, the phase noise calculated by the blind phase search algorithm is used as the parameter for identifying the laser linewidth.

Filter the existing data or build an experimental platform to collect sample data and form a training sample set. Exemplarily, an experimental platform can be built, which includes lasers, double-biased IQ modulators, and single-mode fibers of multiple lengths at the transmitting end; After the optical signal is received, it is processed through an analog-to-digital converter, through digital signal processing, through upsampling, dispersion compensation, IQ equalization, and constant modulus algorithm to obtain an amplitude histogram. The phase noise is further obtained by calculating the blind phase search algorithm.

In the process of constructing the sample data set, in order to improve the generalization ability and the recognition accuracy, data collection is performed in various modulation modes. In some embodiments, the samples in the training sample set are QPSK, 16QAM, PS-16QAM, 64QAM and PS-64QAM under five modulation formats. Various lengths of single-mode fiber can also be set, such as 80KM, 160KM and 240KM.

In some embodiments, the training sample set is in the range of the laser line width of 50KHz to 500KHz, and the amplitude histogram is sampled according to the step size of 50KHz. Meanwhile, for the elastic optical network using the QPSK modulation format, the optical signal-to-noise ratio is 10 to Amplitude histogram sampling is performed for each parameter within the range of 25dB; for elastic optical networks using 16QAM and PS-16QAM modulation formats, amplitude histogram sampling is performed for each parameter within the optical signal-to-noise ratio range of 15 to 30dB; Using elastic optical networks with 64QAM and PS-64QAM modulation formats, amplitude histogram sampling is performed for each parameter in the range of optical signal-to-noise ratio 20 to 35 dB; for each combination of laser linewidth, modulation format and optical signal-to-noise ratio A first set number of samples are collected. Further, the amplitude histogram is an amplitude histogram including signals of dual polarization states. In this way, there are at least 8000 (5×16×10×5×2) magnitude histogram samples in the training sample set. The 8000 samples (5 × 16 × 10 × 5 × 2) in this dataset were collected under 5 modulation formats, 16 signal-to-noise ratios, 10 line widths, and 2 permutations and combinations of polarization states, and each 5 sets of data were collected.

In some embodiments, the training sample set may be further divided into a training set and a test set according to a set ratio, the training set is used for training and updating model parameters, and the test set is used for testing and optimizing model parameters. Exemplarily, 90% of the amplitude histogram samples are randomly selected as the training set, and the other 10% of the amplitude histogram samples are selected as the test set.

Further, for the phase noise used for laser linewidth identification, the training sample set is in the range of the laser linewidth from 50KHz to 500KHz, and the phase noise is collected according to the step size of 50KHz. For the elastic optical network in the optical signal-to-noise ratio range of 20 to 30dB, the phase noise of each parameter is collected separately; for the elastic optical network with PS-64QAM and 64QAM modulation formats, the optical signal-to-noise ratio is in the range of 30 to 40dB. Phase noise collection is performed separately for each parameter; for each combination of laser linewidth, modulation format and optical signal-to-noise ratio, a second set number of samples are collected respectively. In this way, at least 2200 (4×5×11×10) phase noises are collected as data sets, and in some embodiments, 70% of the data are randomly selected as training sets and the rest as test sets.

Finally, for each sample, the corresponding modulation format, optical signal-to-noise ratio and laser linewidth are marked as labels.

In step S102, the first initial model and the second initial model are defined as an ANN network. The ANN network is an extensive parallel interconnected network composed of simple adaptive units, and its organization can simulate the biological nervous system to real-world objects. interactive response. In this embodiment, the structure of the first initial model based on multi-task learning is specifically limited. Since the first initial model needs to use the amplitude histogram to simultaneously identify the optical signal-to-noise ratio and the modulation format, the structure of the first initial model is as follows: After sharing the hidden layers, specific hidden layers and output layers are constructed for the two recognition tasks, respectively. In addition to the input layer and the output layer, the second initial model can contain multiple hidden layers.

In step S103, for the training of the first initial model, network update may be performed based on error backpropagation, and specifically, the method of loss function and stochastic gradient descent is used for processing.

是一个维纳过程，由于维纳过程是独立增量过程，相位噪声增量的方差与激光器线宽是线性关系，因此监测激光器线宽的关键是监测相位噪声增量的方差。In step S104, when the signal laser and the local oscillator laser are of the Lorentzian line type, the phase noise

is a Wiener process. Since the Wiener process is an independent incremental process, the variance of the phase noise increment has a linear relationship with the laser linewidth. Therefore, the key to monitoring the laser linewidth is to monitor the variance of the phase noise increment.

In some embodiments, the input is a sequence formed by the root mean square value and the average value of the phase noise increments of each sample at different symbol intervals, including:

For signal lasers and local oscillator lasers using the Lorentzian line type, the phase noise is a Wiener process, expressed as:

的表达式为：Among them, w(n) is a sequence that obeys a Gaussian distribution, the mean of w(n) is 0, and the variance is

The expression is:

where Δv is the sum of the full width at half maximum of the signal and the LO, and _τs is the time interval between adjacent signals.

的表达式为：Then the phase noise increment

The expression is:

Among them, abs( ) is the absolute value function, N represents the number of symbol intervals, and n represents the number of time series.

Calculate the phase noise increments corresponding to a number of set symbol intervals, and obtain the root mean square value and average value corresponding to each symbol interval number; The averages make up the series and serve as input to the second initial model.

的绝对值。

即为N个w(n)的叠加，w(n)是服从高斯分布的序列，w(n)的均值为0，方差为

在绝对值函数的作用下，得到了方差叠加后的参数。Specifically, in this example, the phase noise in the integrated coherent receiver can be estimated by the result of carrier phase recovery (CPR). Specifically, it can be processed by blind phase search. After the phase noise is obtained, it can be calculated by formula (3)

the absolute value of .

That is, the superposition of N w(n), w(n) is a sequence obeying a Gaussian distribution, the mean of w(n) is 0, and the variance is

Under the action of the absolute value function, the parameters after the variance is superimposed are obtained.

Due to the different laser linewidths, the N values required to accurately predict the linewidth are different. In order to enable the scheme to accurately predict the linewidth under multiple linewidths, this embodiment needs to take multiple N values, and use the phase noise increment data at different symbol intervals as the input of the ANN network to improve the robustness of the scheme sex. After parameter optimization, the values of N are 2, 5, and 10, that is, in the calculation of phase noise increments corresponding to multiple preset numbers of symbol intervals, the multiple preset numbers of symbol intervals include 2, 5, and 10. Since the increment is a Gaussian distribution with zero mean, only the absolute value information needs to be obtained when calculating the increment.

During the training process of the second initial model, the network can be updated based on error backpropagation, and specifically, the method of loss function and stochastic gradient descent is used for processing.

On the other hand, the present invention also provides a two-stage detection method for a probability shaped signal, including steps S201-S202:

Step S201: Obtain a histogram of the amplitude to be measured obtained by processing the blind equalization algorithm of the signal to be detected, and input the histogram of the amplitude to be measured into the optical signal noise in the two-stage detection model training method for probabilistic shaped signals described in steps S101 to S104 above. Compare with the modulation format recognition model, and output the optical signal-to-noise ratio and modulation format recognition results of the signal to be detected.

Step S202: Obtain the root mean square value and the average value of the phase noise increments of the signal to be detected at different symbol intervals, and form a sequence to be detected, and input the sequence to be detected into the double signal of the probability shaping signal in the above steps S101-S104. The laser line width recognition model in the model training method is detected in stages, and the laser line shape recognition result is output.

In this embodiment, after the signal to be detected is received by the integrated coherent receiver at the receiving end, DSP processing is performed, and the histogram of the amplitude to be detected obtained by processing the blind equalization algorithm is input to the optical signal obtained by pre-training in steps S101 to S104. Noise ratio and modulation format recognition model to detect optical signal-to-noise ratio and modulation format. Further, the input parameter format and requirements of the pre-trained laser linewidth identification model according to steps S101 to S104.

In another aspect, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, where the processor implements the steps of the above method when the processor executes the program.

In another aspect, the present invention also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the steps of the above method.

Below in conjunction with a specific embodiment, the present invention will be described:

This embodiment provides a two-stage optical signal-to-noise ratio and laser linewidth monitoring solution for probability shaped signals.

The two-stage optical signal-to-noise ratio and laser linewidth monitoring scheme is shown in Figure 2. In the first stage, optical signal-to-noise ratio estimation and modulation format identification are realized by using ANN network based on multi-task learning. The first stage is located after the blind equalization algorithm, and the amplitude histogram of the equalized signal is input into the neural network based on multi-task learning. The structure of the neural network is shown in Figure 3, where the shared layer can simultaneously learn the modulation format (MFI) and the optical signal-to-noise ratio (OSNR) to estimate the commonality of the two tasks, thus improving the accuracy of the results and reducing the complexity of the network structure . The second stage is located after carrier phase recovery (CPR). A simple ANN network is used to estimate the laser linewidth. The structure of the ANN network is shown in Figure 4.

Specifically, a multi-task learning ANN network is used in the first stage to map the amplitude histogram to the optical signal-to-noise ratio and modulation format. An ANN network is used in the second stage to map the rms and average values of the phase noise increments to the laser linewidth.

During the training process, a training sample set is first established. In this embodiment, screening and establishment may be performed according to the existing data in the existing database, or an experimental platform may be established to perform simulated operation and collect data. Exemplarily, the establishment of the experimental platform includes: a laser (1550nm) and a double-biased IQ modulator at the transmitting end, a standard single-mode fiber of 80 km, an Erbium-Doped Fiber Amplifier (EDFA) at the receiving end, and introduced amplifier spontaneous emission noise (ASE noise) , obtain the optical signal through the integrated coherent receiver, perform analog-to-digital conversion, and perform digital processing after up-sampling, including at least dispersion compensation, IQ equalization and constant modulus algorithm processing, after the constant modulus algorithm processing, the amplitude of the dual polarization state signal is obtained Histogram. The phase noise is then obtained by blind phase search.

For the amplitude histogram, the training sample set is in the range of the laser line width of 50KHz to 500KHz, and the amplitude histogram is sampled according to the step size of 50KHz. At the same time, for the elastic optical network using the QPSK modulation format, the optical signal to noise ratio is 10 to 25dB. The amplitude histogram is sampled separately for each parameter within the range; for the elastic optical network using 16QAM and PS-16QAM modulation formats, the amplitude histogram is sampled for each parameter within the optical signal-to-noise ratio range of 15 to 30dB; For elastic optical networks in 64QAM and PS-64QAM modulation formats, amplitude histogram sampling is performed for each parameter in the range of optical signal-to-noise ratio of 20 to 35 dB; separately for each combination of laser linewidth, modulation format and optical signal-to-noise ratio The first set number of samples. Further, the amplitude histogram is an amplitude histogram including signals of dual polarization states. In this way, there are at least 8000 (5×16×10×5×2) magnitude histogram samples in the training sample set. 90% of the amplitude histogram samples are randomly selected as the training set, and the other 10% of the amplitude histogram samples are selected as the test set. Add the corresponding optical signal-to-noise ratio and modulation format as labels for the amplitude histogram of each sample.

For the phase noise used for laser linewidth identification, the training sample set is in the range of the laser linewidth from 50KHz to 500KHz, and the phase noise is collected according to the step size of 50KHz. For optical networks, phase noise collection is performed for each parameter in the range of optical signal-to-noise ratio of 20 to 30dB; for elastic optical networks using PS-64QAM and 64QAM modulation formats, each of the The parameters are separately collected for phase noise; for each combination of laser linewidth, modulation format and optical signal-to-noise ratio, a second set number of samples are collected respectively. In this way, at least 2200 (4×5×11×10) phase noises are collected as data sets, and in some embodiments, 70% of the data are randomly selected as training sets and the rest as test sets. In this embodiment, the laser and the local oscillator laser used are Lorentzian line type, so the phase noise is a Wiener process. Referring to the description of the above equations 1 to 3, calculate the mean square of the phase noise increment corresponding to different symbol intervals The root value and the mean value form a sequence, which is used as the input of the second-stage ANN network, and the corresponding laser linewidth is used as the output.

The ANN network in the first stage and the ANN network in the second stage are respectively trained using the training sample set to obtain the optical signal-to-noise ratio and modulation format recognition model, and the laser linewidth recognition model.

Further, in order to study the performance of the two-stage optical signal-to-noise ratio and laser linewidth monitoring scheme for probability-shaping signals proposed in this embodiment, an 80km transmission coherent optical communication system simulation is built, and QPSK, 16QAM, probability shaping (PS)-16QAM, 64QAM, PS-64QAM five kinds of signals, collect the amplitude distribution histogram and the root mean square value and mean value of the phase noise increment, and input the pre-trained optical signal-to-noise ratio and modulation format identification according to the corresponding format Model and Laser Linewidth Identification Model.

As shown in Figure 5, the shape of the amplitude histogram of the signal is related to both the modulation format and the signal-to-noise ratio. As shown in Figure 6, the change in phase noise is related to the size of the line width. The simulation results show that the accuracy of modulation format identification is 100%, and the root mean square error (RMSE) of OSNR and linewidth monitoring errors are 0.2dB and 18KHz, respectively. Among them, the curves of predicted OSNR and actual OSNR are shown in FIG. 7 , and the value of predicted linewidth and the value of actual linewidth are shown in FIG. 8 .

In the two-stage detection model training method, identification method and device of the probability shaping signal of the present invention, the electrical domain signal obtained by processing the DSP module at the signal receiving end is based on the amplitude histogram processed by the blind equalization algorithm, and is processed by machine learning. The mapping of the amplitude histogram to the optical signal-to-noise ratio and modulation format is constructed in a manner to realize the identification of the optical signal-to-noise ratio and the modulation format of the probability shaped signal. At the same time, since the phase noise of the signal laser and the local oscillator laser under the Lorentzian line type is a micro-nano process, and the variance of the phase noise increment is linearly related to the laser linewidth, the phase noise increment is constructed by machine learning. The rms and average values of the quantities are mapped to the laser linewidth to enable identification of the laser linewidth.

Further, by introducing the root mean square value and average value of phase noise increments at different symbol intervals to form a sequence, and constructing the mapping relationship between the sequence and the laser linewidth by means of machine learning, the generalization ability of recognition can be effectively improved.

It should be understood by those of ordinary skill in the art that the various exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is implemented in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may implement the described functionality using different methods for each particular application, but such implementations should not be considered beyond the scope of the present invention. When implemented in hardware, it may be, for example, an electronic circuit, an application specific integrated circuit (ASIC), suitable firmware, a plug-in, a function card, or the like. When implemented in software, elements of the invention are programs or code segments used to perform the required tasks. The program or code segments may be stored in a machine-readable medium or transmitted over a transmission medium or communication link by a data signal carried in a carrier wave. A "machine-readable medium" may include any medium that can store or transmit information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, and the like. The code segments may be downloaded via a computer network such as the Internet, an intranet, or the like.

It should also be noted that the exemplary embodiments mentioned in the present invention describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above steps, that is, the steps may be performed in the order mentioned in the embodiments, or may be different from the order in the embodiments, or several steps may be performed simultaneously.

In the present invention, features described and/or illustrated with respect to one embodiment may be used in the same or similar manner in one or more other embodiments, and/or in combination with or in place of features of other embodiments Features of the implementation.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, various modifications and changes may be made to the embodiments of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims (10)

1. a two-stage detection model training method of probability shaping signal, is characterized in that, comprises: Obtain a training sample set, the training sample set includes multiple samples, and each sample includes an amplitude histogram obtained by processing a DSP module at the receiving end of the elastic optical network through a blind equalization algorithm under various signal modulation formats; each sample also includes a phase Noise; mark the optical signal-to-noise ratio, modulation format and laser line width corresponding to each sample as a label; wherein, the elastic optical network adopts a Lorentzian line-type signal laser and a local oscillator laser; Obtain a first initial model and a second initial model, where the first initial model and the second initial model are ANN networks; wherein, the first initial model includes an input layer, a shared hidden layer, and the shared The first specific hidden layer and the second specific hidden layer of the hidden layer, the first specific hidden layer is used to identify the modulation format and connect the first output layer, and the second specific hidden layer is used to identify the optical signal-to-noise ratio and connect The second output layer; Taking the amplitude histogram of each sample as the input, taking the optical signal-to-noise ratio and the modulation format as the output, using the training sample set to train the first initial model to obtain the optical signal-to-noise ratio and modulation format recognition model; Taking the sequence formed by the root mean square value and the average value of the phase noise increments of each sample at different symbol intervals as input, and taking the laser linewidth as output, the second initial model is trained by using the training sample set, and the result is obtained: Laser linewidth identification model. 2. the two-stage detection model training method of probability shaping signal according to claim 1, is characterized in that, the sample in described training sample set is in QPSK, 16QAM, PS-16QAM, 64QAM and PS-64QAM five kinds of modulation formats produced below. 3. The two-stage detection model training method of probability shaping signal according to claim 2, is characterized in that, described training sample set is in the scope of laser line width 50KHz to 500KHz, according to 50KHz is step size to carry out amplitude histogram sampling , at the same time, for the elastic optical network using the QPSK modulation format, the amplitude histogram is sampled for each parameter in the range of the optical signal-to-noise ratio of 10 to 25dB; for the elastic optical network using the 16QAM and PS-16QAM modulation formats, in the optical Amplitude histogram sampling is performed for each parameter in the range of SNR 15 to 30dB; for elastic optical networks using 64QAM and PS-64QAM modulation formats, the amplitude of each parameter in the range of OSR 20 to 35dB Histogram sampling; respectively collect a first set number of samples for each combination of laser line width, modulation format and optical signal-to-noise ratio. 4 . The method for training a dual-stage detection model of a probability-shaping signal according to claim 3 , wherein the amplitude histogram comprises an amplitude histogram of a dual-polarization state signal. 5 . 5. The two-stage detection model training method of probability shaping signal according to claim 2, is characterized in that, described training sample set is in the scope of laser line width 50KHz to 500KHz, according to 50KHz is step size to carry out phase noise collection, At the same time, for the elastic optical network using QPSK, 16QAM and PS-16QAM modulation formats, the phase noise of each parameter is collected separately in the optical signal-to-noise ratio range of 20 to 30 dB; for the elastic optical network using PS-64QAM and 64QAM modulation formats Network, phase noise collection is performed for each parameter within the range of optical signal-to-noise ratio of 30 to 40 dB; a second set number of samples are collected for each combination of laser linewidth, modulation format and optical signal-to-noise ratio. 6. the two-stage detection model training method of probability shaping signal according to claim 1 is characterized in that, with the sequence formed by the root mean square value of phase noise increment and the average value of each sample under different symbol intervals as input, include: For signal lasers and local oscillator lasers using the Lorentzian line type, the phase noise is a Wiener process, expressed as:

Among them, w(n) is a sequence that obeys a Gaussian distribution, the mean of w(n) is 0, and the variance is

The expression is:

where Δv is the sum of the full width at half maximum of the signal and the LO, and τ _s is the time interval between adjacent signals; Then the phase noise increments

The expression is:

Among them, abs( ) is the absolute value function, N represents the number of symbol intervals, and n represents the number of time series; Calculate the phase noise increments corresponding to multiple set symbol intervals, and obtain the root mean square value and average value corresponding to each symbol interval; The root mean square value and the average value of the phase noise increment corresponding to each symbol interval number form a sequence, which is used as the input of the second initial model. 7. The two-stage detection model training method of probability shaped signal according to claim 6, is characterized in that, in calculating the phase noise increment corresponding to the symbol interval number of a plurality of settings, the symbol interval number of a plurality of settings comprises 2, 5 and 10. 8. A two-stage detection method for a probability shaped signal, characterized in that, comprising: Obtain a histogram of the amplitude to be detected obtained by processing the blind equalization algorithm of the signal to be detected, and input the histogram of the amplitude to be detected into the light in the training method of the two-stage detection model of the probability-shaping signal according to any one of claims 1 to 7. Signal-to-noise ratio and modulation format identification model, and output the optical signal-to-noise ratio and modulation format identification results of the signal to be detected; Obtain the root mean square value and the average value of the phase noise increment of the signal to be detected under different symbol intervals, and form a sequence to be detected, and input the sequence to be detected into the double signal of the probability shaping signal of any one of claims 1 to 7. The laser line width recognition model in the model training method is detected in stages, and the laser line shape recognition result is output. 9. An electronic device, comprising a memory, a processor and a computer program stored on the memory and running on the processor, wherein the processor implements any one of claims 1 to 8 when the processor executes the program the steps of the method described in item. 10. A computer-readable storage medium on which a computer program is stored, characterized in that, when the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images