Disclosure of Invention
The embodiment of the invention provides a method and equipment for regulating and controlling the dynamic gain of an optical fiber Raman amplifier, which are used for eliminating or improving one or more defects in the prior art and solving the problems of uneven Raman amplification gain spectrum caused by the traditional multi-pump light source and uneven gain spectrum caused by the dynamic change of signal light to be amplified.
The technical scheme of the invention is as follows:
in one aspect, the present invention provides a method for adjusting and controlling a dynamic gain of an optical fiber raman amplifier, including:
receiving signal light to be amplified, and acquiring state parameters of the signal light to be amplified, wherein the state parameters comprise: the light intensity, the frequency and the central wavelength of the signal light to be amplified;
obtaining target parameters, wherein the target parameters at least comprise: flatness and bandwidth of the ideal gain spectrum; acquiring equipment parameters, wherein the equipment parameters at least comprise: the working wavelength range and the working temperature range of the optical fiber Raman amplifier;
acquiring a preset neural network model, wherein the preset neural network model is used for mapping the state parameters, the target parameters and the equipment parameters to the pumping number, the pumping power and the pumping wavelength required for reaching the ideal gain spectrum;
inputting the state parameters, the target parameters and the equipment parameters into the preset neural network to output the predicted pumping number, the predicted pumping power and the predicted pumping wavelength for realizing the ideal gain spectrum; adjusting the optical fiber Raman amplifier to amplify the signal light to be amplified according to the predicted pumping number, the predicted pumping power and the predicted pumping wavelength, and detecting an actual gain spectrum;
calculating the mean square error of an actual gain spectrum and an ideal gain spectrum, if the mean square error is larger than a set threshold value, adjusting the number of predicted pumps, the predicted pump power and the predicted pump wavelength by using the preset neural network model in combination with a gradient descent method, detecting based on the adjusted number of predicted pumps, the predicted pump power and the predicted pump wavelength to obtain an adjusted actual gain spectrum, and calculating the adjusted mean square error; until the adjusted mean square error is smaller than the set threshold value, and outputting the adjusted predicted pumping number, predicted pumping power and predicted pumping wavelength.
In some embodiments, the preset neural network model is obtained by training the convolutional neural network initial network or the fully-connected neural network initial network with a training sample set.
In some embodiments, before obtaining the preset neural network model, the method further includes:
acquiring a fully-connected neural network initial model;
acquiring a training sample set, wherein the training sample set comprises a set number of samples, and each sample comprises the light intensity of sample signal light to be amplified, the frequency of the sample signal light to be amplified, the central wavelength of the sample signal light to be amplified, the number of pumps, the pumping power, the pumping wavelength, the flatness of an actual pumping gain spectrum and the bandwidth of the actual pumping gain spectrum in the primary optical fiber Raman amplification process;
and training the fully-connected neural network initial model by using the training sample set to obtain the preset neural network model, wherein the input is the light intensity of the sample signal light to be amplified, the frequency of the sample signal light to be amplified, the central wavelength of the sample signal light to be amplified, the flatness of the actual pumping gain spectrum and the bandwidth of the actual pumping gain spectrum, and the output is the pumping number, the pumping power and the pumping wavelength.
In some embodiments, the pump wavelength is also tuned by a flexible grid.
In some embodiments, before obtaining the preset neural network model, the method further includes:
acquiring the initial network model of the convolutional neural network; the convolutional neural network initial network comprises a feature extraction layer and a feature mapping layer, wherein the feature extraction layer comprises a convolutional layer, an activation function layer and a pooling layer, the input of each neuron in the convolutional layer is connected with a local receiving domain of the previous layer and used for extracting the local features and determining the position relation, and the feature mapping layer is used as a classifier and used for identifying and classifying the results;
acquiring a training sample set, wherein the training sample set comprises a set number of samples, and each sample comprises the light intensity of sample signal light to be amplified, the frequency of the sample signal light to be amplified, the central wavelength of the sample signal light to be amplified, the number of pumps, the pumping power, the pumping wavelength, the flatness of an actual pumping gain spectrum and the bandwidth of the actual pumping gain spectrum in the primary optical fiber Raman amplification process;
and training the convolutional neural network initial network model by using the training sample set to obtain the preset neural network model, wherein the input is the light intensity of the sample signal light to be amplified, the frequency of the sample signal light to be amplified, the central wavelength of the sample signal light to be amplified, the flatness of the actual pumping gain spectrum and the bandwidth of the actual pumping gain spectrum, and the output is the pumping number, the pumping power and the pumping wavelength.
In some embodiments, adjusting the predicted number of pumps, the predicted pump power, and the predicted pump wavelength by using the preset neural network model in combination with a gradient descent method includes:
obtaining the current predicted pumping number, the predicted pumping power and the predicted pumping wavelength, and recording as pumping parameter vectors;
obtaining a difference value between the current actual gain spectrum and the ideal gain spectrum, solving a gradient based on the parameters of the preset neural network model, and updating the predicted pumping number, the predicted pumping power and the predicted pumping wavelength, wherein the calculation formula is as follows:
wherein the content of the first and second substances,
for the updated pump parameter vector,
for the current pump parameter vector, i denotes the number of iterations,
it is indicated that the learning rate is,
which means that the gradient is determined,
representing the difference between the current actual gain spectrum and the ideal gain spectrum.
In some embodiments, the method further extends the training sample set using perturbation theory, including:
acquiring an initial training sample set only containing existing data, and acquiring a differential equation of pump light power of pump light with each wavelength relative to the distribution condition of an optical fiber with unit length under the condition of different state parameters of signal light to be amplified, equipment parameters of a Raman amplifier and pump wavelength combinations so as to express a transmission state;
introducing a pumping light power variable in the differential equation through a perturbation theory, constructing a linear homogeneous differential equation set, solving the linear homogeneous differential equation set by adopting a forward Euler method to obtain a pumping light power variable in each unit length of optical fiber in the transmission process of each wavelength of pumping light in the optical fiber, and calculating an integral variable of the pumping light power in the optical fiber;
calculating the pumping power integral after perturbation according to the integral variable of the pumping light power in the optical fiber, calculating the gain spectrum of the optical fiber Raman amplifier obtained after perturbation, performing perturbation on a plurality of preset pumping wavelength combinations and constructing a correlation matrix;
and expanding the training sample set according to the correlation matrix, the state parameters of the signal light to be amplified corresponding to each element in the correlation matrix, the equipment parameters of the Raman amplifier and the combination of the pumping wavelengths.
In one aspect, the present invention provides a fiber raman amplifier based on a flexible grid network, including:
the pumping laser is used for generating pumping light according to the set pumping wavelength, the set pumping power and the set pumping number;
the wavelength division multiplexer is used for guiding signal light to be amplified and pump light into the first end of the optical fiber, and the wavelength division multiplexer is provided with a flexible grid for dynamically adjusting the pump wavelength;
the optical splitter is arranged at the second end of the optical fiber and used for leading out the signal light to be amplified after the signal light is subjected to optical fiber Raman amplification and splitting a sub-beam from the signal light to be amplified after the signal light is subjected to optical fiber Raman amplification;
the output monitor is used for measuring the sub-beams to obtain an actual gain spectrum of the signal light to be amplified after the signal light is subjected to fiber Raman amplification;
and the signal processing unit is used for acquiring the actual gain spectrum, calculating the predicted pumping number, the predicted pumping power and the predicted pumping wavelength by adopting the optical fiber Raman amplifier dynamic gain regulation and control method, and adjusting the pumping laser according to the predicted pumping number, the predicted pumping power and the predicted pumping wavelength.
In some embodiments, the flexible grid has 6.25GHz, 12.5GHz, 25GHz, 50GHz and/or 100GHz as the grid spacing.
In one aspect, the present invention provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method as described above.
The invention has the beneficial effects that:
the method comprises the steps of mapping state parameters of signal light to be amplified, target parameters of an ideal gain spectrum and equipment parameters to obtain pumping parameters comprising the number of pumps, pumping power and pumping wavelength through a preset neural network model obtained through training, automatically generating the pumping parameters comprising the predicted number of pumps, the predicted pumping power and the predicted pumping wavelength to control the light Raman amplifier to work, adjusting the pumping parameters by using a gradient descent method through a preset neural network after calculating the mean square error of an actual gain spectrum and the ideal gain spectrum, optimizing until the mean square error of the adjusted actual gain spectrum and the ideal gain spectrum is smaller than a set threshold value, and quickly and automatically adjusting the pumping parameters to achieve a flat ideal gain spectrum.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
It will be appreciated by those skilled in the art that the objects and advantages that can be achieved with the present invention are not limited to the specific details set forth above, and that these and other objects that can be achieved with the present invention will be more clearly understood from the detailed description that follows.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the following embodiments and accompanying drawings. The exemplary embodiments and descriptions of the present invention are provided to explain the present invention, but not to limit the present invention.
It should be noted that, in order to avoid obscuring the present invention with unnecessary details, only the structures and/or processing steps closely related to the scheme according to the present invention are shown in the drawings, and other details not so relevant to the present invention are omitted.
It should be emphasized that the term "comprises/comprising" when used herein, is taken to specify the presence of stated features, elements, steps or components, but does not preclude the presence or addition of one or more other features, elements, steps or components.
It is also noted herein that the term "coupled," if not specifically stated, may refer herein to not only a direct connection, but also an indirect connection in which an intermediate is present.
The working principle of the optical fiber Raman amplifier is the stimulated Raman scattering effect in the quartz optical fiber, which is expressed in the form that weak signals and strong pump light within the Raman gain bandwidth of the pump light are transmitted in the optical fiber, so that the weak signal light is amplified. The gain wavelength of the optical fiber Raman amplifier is determined by the wavelength of the pump light, signals with any wavelength in an optical fiber window can be amplified theoretically, and the gain spectrum is wide. In practical applications, several raman pumps are integrated into a module to amplify signals in order to enlarge the gain range of the raman amplifier. When amplifying, due to the requirement of receiver sensitivity, the amplified signals need to have the same power at different wavelengths, i.e. there is a high requirement for gain flatness, so it is a necessary problem to realize the gain flatness of the raman amplifier.
In order to realize flat gain on a plurality of wavelengths, the invention provides a method for regulating and controlling the dynamic gain of an optical fiber Raman amplifier, as shown in FIG. 1, comprising the following steps of S101-S105:
it should be clear that, the steps S101 to S105 described in this embodiment are not limited to the order of the steps, and it should be understood that the order of the steps may be changed or the steps may be parallel in a specific application scenario. Further, the method for adjusting and controlling the dynamic gain of the fiber raman amplifier according to the present embodiment is used for operating on a multi-pump light fiber raman amplifier. For adjusting the pumping parameters to achieve a flattening of the gain spectrum of the signal to be amplified.
Step S101: receiving signal light to be amplified, and acquiring state parameters of the signal light to be amplified, wherein the state parameters at least comprise: the light intensity, frequency and center wavelength of the signal light to be amplified.
Step S102: obtaining target parameters, wherein the target parameters at least comprise: flatness and bandwidth of the ideal gain spectrum; acquiring equipment parameters, wherein the equipment parameters at least comprise: the working wavelength range and the working temperature range of the fiber Raman amplifier.
Step S103: and acquiring a preset neural network model, wherein the preset neural network model is used for mapping the state parameters, the target parameters and the equipment parameters to the pumping number, the pumping power and the pumping wavelength required for achieving the ideal gain spectrum.
Step S104: inputting the state parameters, the target parameters and the equipment parameters into a preset neural network to output the predicted pumping number, the predicted pumping power and the predicted pumping wavelength for realizing the ideal gain spectrum; and adjusting the fiber Raman amplifier to amplify the signal light to be amplified according to the predicted pumping number, the predicted pumping power and the predicted pumping wavelength, and detecting an actual gain spectrum.
Step S105: calculating the mean square error of the actual gain spectrum and the ideal gain spectrum, if the mean square error is larger than a set threshold value, adjusting the number of predicted pumps, the predicted pump power and the predicted pump wavelength by using a preset neural network model in combination with a gradient descent method, detecting based on the adjusted number of predicted pumps, the predicted pump power and the predicted pump wavelength to obtain an adjusted actual gain spectrum, and calculating the adjusted mean square error; until the adjusted mean square error is smaller than the set threshold value, and outputting the adjusted predicted pumping number, predicted pumping power and predicted pumping wavelength.
In step S101 of this embodiment, the signal light to be amplified is input through the transmission fiber, and in the practical application process, the generator of the signal light to be amplified may obtain corresponding light intensity, frequency and central wavelength information, or may separately set a detection device for detection and acquisition.
In step S102, an expected signal amplification effect is determined according to the requirements of the actual application scenario, and a corresponding target parameter is obtained, that is, an ideal gain spectrum of the signal light to be amplified is determined according to the characteristics of actual fiber loss, an expected raman gain coefficient, pump light interaction, pump failure, amplified spontaneous emission noise, rayleigh scattering, and the like, and a corresponding flatness and bandwidth are obtained, where the ideal gain spectrum is a theoretical value. To be more practical, a certain degree of gain jitter may be set for the ideal gain piece. Wherein the pump wavelength is also dynamically adjusted according to the flexible grid of channels.
In step S103, the preset neural network model is obtained through pre-training and is used to establish a mapping relationship between the state parameters, the target parameters, the device parameters, and the pumping parameters required for achieving the ideal gain spectrum, where the pumping parameters at least include the number of pumps, the pumping power, and the pumping wavelength. That is, the preset neural network model in this embodiment is used to reversely deduce the pumping parameters based on the known state parameters, device parameters, and target parameters of the signal light to be amplified, so as to guide the control and adjustment.
In some embodiments, parameters such as the bandwidth and the flatness of an ideal gain spectrum and the wavelength and the power of a signal to be amplified are input into a trained preset neural network model, and parameters such as the working wavelength (1528-1605 nm), the working temperature (-20-65 ℃), the input/output pigtail type (common single-mode fiber SMF-28) and the like of a Raman amplifier are input into the model in the form of condition parameters, so that parameters such as the pumping number, the pumping power and the pumping wavelength corresponding to the gain spectrum are obtained.
In some embodiments, the predetermined neural network model is obtained by training the convolutional neural network initial network or the fully-connected neural network initial network with a training sample set. The training sample set is established based on actual pumping parameters, state parameters of signal light to be amplified, equipment parameters of the Raman amplifier and finally obtained actual gain spectrums in the previous optical amplification process, and in the training sample set establishing process, the actual gain spectrums in the existing data, the state parameters of the signal light to be amplified and the equipment parameters of the Raman amplifier are used as output labels to form samples, so that the training sample set is formed.
In some embodiments, correlation matrixes of different pump powers and ideal gain spectrums under the combined conditions of different state parameters of signal light to be amplified, equipment parameters of the raman amplifier and pump wavelengths can be obtained through calculation based on a perturbation theory, and a training sample set with a larger data volume is constructed based on data in the correlation matrixes to obtain a better fitting effect.
Specifically, under the combined conditions of different state parameters of signal light to be amplified, equipment parameters of the raman amplifier and pumping wavelengths, the construction mode of correlation matrixes of different pumping powers and ideal gain spectrums comprises the following steps of S201-S203:
step S201: and under the combined conditions of different state parameters of the signal light to be amplified, equipment parameters of the Raman amplifier and pumping wavelengths, acquiring a differential equation of the pumping light power of the pumping light with each wavelength relative to the distribution condition of the optical fiber with unit length to express the transmission state.
Step S202: and introducing a pumping light power variable in a differential equation through a perturbation theory, constructing a linear homogeneous differential equation set, solving the linear homogeneous differential equation set by adopting a forward Euler method to obtain the pumping light power variable in each unit length of optical fiber in the transmission process of each wavelength of pumping light in the optical fiber, and calculating an integral variable of the pumping light power in the optical fiber.
Step S203: and calculating the pumping power integral after perturbation according to the integral variable of the pumping light power in the optical fiber, calculating the gain spectrum of the optical fiber Raman amplifier obtained after perturbation, performing perturbation on a plurality of preset pumping wavelength combinations and constructing a correlation matrix.
In some embodiments, in step S201, for the fiber raman amplifier with a preset pump wavelength combination, obtaining a differential equation of pump light power of each wavelength of pump light with respect to a unit length of fiber distribution includes:
for a fiber Raman amplifier with N pump wavelengths, a differential equation of each pump light power with respect to a unit length of the fiber is established, as shown in the following formula 1:
wherein the content of the first and second substances,
is an N x 1 vector representing the pump light power;
is an N x 1 vector representing pump light loss;
is an NxN matrix representing the Raman gain coefficient among the wavelengths of the pump light; z represents the fiber length.
In this embodiment, the transmission law of the pump light is expressed by equation 1, and since the fiber raman amplifier usually operates in a small-signal or near-small-signal state, the distribution of the pump light power along the fiber depends mainly on the interaction between the pump lights. Further, defining an Nx 1 vector representing the pump light power along the fiber integral (pump integral for short); wherein, the expression of the pump integral I is as follows 2:
wherein the content of the first and second substances,
to represent the N × 1 vector of the pump light power, dz is the differential of the fiber length, and L is the fiber length.
In step S202, a pump light power variable is introduced in a differential equation through a perturbation theory, and a linear homogeneous differential equation set is constructed, including:
introducing variable to input pump light power
,
Become into
Neglecting to
The linear homogeneous partial differential equation system is constructed according to the second order term of (3):
wherein the content of the first and second substances,
is an N x N matrix representing the effect of fiber characteristics on the distribution of pump light power along the fiber. Here matrix
Is obtained by extracting the formula of the formula
And (4) obtaining the product.
Solving the linear homogeneous differential equation set by adopting a forward Euler method to obtain the pump light power variable in each unit length optical fiber in the transmission process of each wavelength pump light in the optical fiber, and the method comprises the following steps:
adopting a forward Euler method to numerically solve the linear homogeneous partial differential equation set (formula 3), and taking
For step size, then:
……
……
then, in step S202, the integral variable of the pump light power in the optical fiber
I.e., the variable of the pump integral, can be calculated by the following equation 7:
wherein, H is an NxN matrix and represents the linear relation between the power change of input pump light and the integral variable of the power of the pump light in the optical fiber; k is the step size of the k-th segment,
l is the pump length of the fiber, step size.
The gain of the fiber raman amplifier is determined by the pump integral I, and for the case of M input channels, the fiber raman amplifier gain spectrum obtained after calculating the perturbation is as follows 8:
wherein the content of the first and second substances,
the M multiplied by 1 vector is used for representing the gain spectrum of each channel of the fiber Raman amplifier after perturbation;
is the loss of each channel and is,
is an M × N matrix representing the Raman gain coefficient between the signal light wavelength and the pump light wavelength;
is an
mx 1 vector representing the power spectrum tilt of the channel due to inter-channel SRS effects;
for the pump integral, i.e. the integral of the pump light power in the current state in the optical fiber, the calculation formula is as follows:
wherein the content of the first and second substances,
is the pump integral for the k-1 th state,
is the pump integral for the k-th state,
is a variable of the pump integral calculated in equation 7.
Therefore, correlation matrixes of different pump powers and ideal gain spectrums under the conditions of different state parameters of signal light to be amplified, equipment parameters of the Raman amplifier and pump wavelength combination are obtained through theoretical calculation, and therefore the training sample set is expanded under the condition that the existing data quantity is limited. The accuracy of training the preset neural network model is improved.
In some embodiments, before obtaining the preset neural network model, the method further includes steps S301 to S303:
step S301: and acquiring a fully-connected neural network initial model.
Step S302: the method comprises the steps of obtaining a training sample set, wherein the training sample set comprises a set number of samples, and each sample comprises the light intensity of sample signal light to be amplified, the frequency of the sample signal light to be amplified, the central wavelength of the sample signal light to be amplified, the number of pumps, the pumping power, the pumping wavelength, the flatness of an actual pumping gain spectrum and the bandwidth of the actual pumping gain spectrum in the primary optical fiber Raman amplification process.
Step S303: and training the fully-connected neural network initial model by using a training sample set to obtain a preset neural network model, wherein the input is the light intensity of the sample signal light to be amplified, the frequency of the sample signal light to be amplified, the central wavelength of the sample signal light to be amplified, the flatness of an actual pumping gain spectrum and the bandwidth of the actual pumping gain spectrum, and the output is the pumping number, the pumping power and the pumping wavelength.
In this embodiment, the fully-connected neural network includes an input layer, an output layer, and one or more hidden layers. The structure of a fully-connected neural network initial model (DNN) is not fixed, a general neural network comprises an input layer, a hidden layer and an output layer, one DNN structure only comprises one input layer and one output layer, and the hidden layer is arranged between the input layer and the output layer. Each layer of neural network is provided with a plurality of neurons, the neurons in the layers are mutually connected, the neurons in the layers are not mutually connected, and the neurons in the next layer are connected with all the neurons in the upper layer. Since DNN can fit almost any function, the non-linear fit capability of DNN is very strong. Often deeper and narrower networks will be more resource efficient. However, DNN is not easy to train, and a large amount of data is required to train a deep network, so that under the condition that the existing data is limited, correlation matrices of different pump powers and ideal gain spectrums under the conditions of different state parameters of signal light to be amplified, equipment parameters of a raman amplifier and pump wavelength combinations can be obtained based on the steps in steps S201 to S203, and a training sample set with a larger data amount is constructed based on the data in the correlation matrices.
In some embodiments, the samples in the training sample set with the first set proportion are used for training the fully-connected neural network initial model to obtain the preset neural network model, and the rest samples are used for testing the preset neural network model.
In some embodiments, before obtaining the preset neural network model, the method further includes steps S301 to S303:
step S301: acquiring a convolutional neural network initial network model; the convolutional neural network initial network comprises a feature extraction layer and a feature mapping layer, wherein the feature extraction layer comprises a convolutional layer, an activation function layer and a pooling layer, the input of each neuron in the convolutional layer is connected with a local receiving domain of the previous layer and used for extracting the features of the local part and determining the position relation, and the feature mapping layer is used as a classifier and used for identifying and classifying the results.
Step S302: and acquiring a training sample set, wherein the training sample set comprises a set number of samples, and each sample comprises the light intensity of the signal light of the sample to be amplified, the frequency of the signal light of the sample to be amplified, the central wavelength of the signal light of the sample to be amplified, the number of pumps, the pumping power, the pumping wavelength, the flatness of an actual pumping gain spectrum and the bandwidth of the actual pumping gain spectrum in the primary optical fiber Raman amplification process.
Step S303: and training the convolutional neural network initial network model by using a training sample set to obtain a preset neural network model, wherein the input is the light intensity of the sample signal light to be amplified, the frequency of the sample signal light to be amplified, the central wavelength of the sample signal light to be amplified, the flatness of an actual pumping gain spectrum and the bandwidth of the actual pumping gain spectrum, and the output is the pumping number, the pumping power and the pumping wavelength.
Similarly, in steps S301 to S303, under the condition that the existing data is limited, the correlation matrices of different pump powers and ideal gain spectrums under the conditions of different state parameters of the signal light to be amplified, device parameters of the raman amplifier, and pump wavelength combinations can be obtained based on the steps in steps S201 to S203, and a training sample set with a larger data size is constructed based on the data in the correlation matrices.
In step S104, a predicted pumping parameter is calculated according to the current state parameter, the target parameter and the device parameter through the trained preset neural network model. And adjusting the light Raman amplifier to amplify based on the predicted pumping parameters, and obtaining an actual gain spectrum. Since the actual gain spectrum obtained by the first prediction has a difference from the ideal gain spectrum, in order to obtain a better effect, the flatness of the actual gain spectrum needs to be judged and optimized.
In step S105, when the mean square error between the actual gain spectrum and the ideal gain spectrum is greater than the threshold, the trained preset neural network model is combined to adjust and update the predicted number of pumps, the predicted pump power, and the predicted pump wavelength by using a gradient descent method. And adjusting the fiber Raman amplifier by utilizing the updated predicted pumping number, predicted pumping power and predicted pumping wavelength to obtain an updated actual gain spectrum, repeatedly calculating the mean square error of the updated actual gain spectrum and the ideal gain spectrum, and repeatedly circulating until the adjusted mean square error is smaller than a set threshold value and outputting the adjusted predicted pumping number, predicted pumping power and predicted pumping wavelength.
In some embodiments, in step S105, adjusting the predicted number of pumps, the predicted pump power, and the predicted pump wavelength by using a preset neural network model in combination with a gradient descent method includes steps S401 to S402:
step S401: and acquiring the current predicted pumping number, the predicted pumping power and the predicted pumping wavelength, and recording as pumping parameter vectors.
Step S402: obtaining the difference value between the current actual gain spectrum and the ideal gain spectrum, solving the gradient based on the parameters of the preset neural network model, and updating the predicted pumping number, the predicted pumping power and the predicted pumping wavelength, wherein the calculation formula is as follows:
wherein the content of the first and second substances,
for the updated pump parameter vector,
for the current pump parameter vector, i denotes the number of iterations,
it is indicated that the learning rate is,
which means that the gradient is determined,
representing the difference between the current actual gain spectrum and the ideal gain spectrum.
In one aspect, the present invention provides a fiber raman amplifier based on a flexible grid network, as shown in fig. 2, including: the device comprises a pump laser, a wavelength division multiplexer, an optical splitter, an output monitor and a signal processing unit.
The pumping laser is used for generating pumping light according to the set pumping wavelength, the set pumping power and the set pumping number;
the wavelength division multiplexer is used for guiding signal light to be amplified and pump light into the first end of the optical fiber and is provided with a flexible grid;
the optical splitter is arranged at the second end of the optical fiber and used for leading out the signal light to be amplified after the signal light is subjected to optical fiber Raman amplification and splitting a sub-beam from the signal light to be amplified after the signal light is subjected to optical fiber Raman amplification;
the output monitor is used for measuring the quantum beam to obtain an actual gain spectrum of the signal light to be amplified after the signal light is subjected to fiber Raman amplification;
and the signal processing unit is used for acquiring an actual gain spectrum, calculating the predicted pumping number, the predicted pumping power and the predicted pumping wavelength by adopting the dynamic gain control method of the optical fiber Raman amplifier in the steps S101 to S105, and adjusting the pumping laser according to the predicted pumping number, the predicted pumping power and the predicted pumping wavelength.
Based on the working principle of the optical fiber Raman amplifier, the wavelength interval between the pump light and the signal light to be amplified needs to reach the Stokes frequency. Therefore, in the optical fiber raman amplification process, if the signal light to be amplified has a slight variation, the gain spectrum flatness is necessarily changed, the optical fiber raman amplifier in the prior art can only process the signal to be amplified with a fixed wavelength and power, and if the signal with different wavelengths needs to be processed, the hardware structure needs to be changed. However, in the practical application process, the signal light to be amplified usually generates disturbance due to the change of the service adjustment, and in order to cope with the dynamic change, and enable the fiber raman amplifier based on the flexible grid network according to this embodiment to dynamically regulate and control the changed signal light to be amplified, the pump light is regulated through the flexible grid to realize the dynamic regulation and control, so as to ensure the stability and the flatness of the gain spectrum.
In some embodiments, the flexible grid has 6.25GHz, 12.5GHz, 25GHz, 50GHz and/or 100GHz as the grid spacing.
In one aspect, the present invention provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method as described above.
In one aspect, the invention provides a computer-readable storage medium, on which a computer program is stored, characterized in that the program, when executed by a processor, implements the steps of the above-described method.
The invention is illustrated below with reference to a specific example:
the present embodiment provides a raman amplifier capable of implementing flexible grid dynamic pump gain control, which includes a transmission fiber, a raman pump source, an output detector, and a signal processing unit. The Raman pump source is used for providing pump light for the transmission signal to generate gain, and the device can emit a flexible pump signal light source; the output detector is used for detecting the output signal and converting the output signal into an electric signal; the signal processing unit comprises a gain spectrum detection module and a neural network optimization control module, wherein the gain spectrum detection module is used for calculating an output signal gain spectrum according to the electric signal, calculating average gain, gain flatness, flat bandwidth and the like, and optimizing multiple parameters of the pump light through the neural network optimization control module, so that gain flatness is realized. The amplifier can automatically realize the optimal distribution of the grid signals, reduces the gain fluctuation and has high flexibility.
In the embodiment, the central wavelength of the signal to be amplified is 193.1THz, the number of pumps is 2, the pump power is 1W, and the pump wavelength is 179.9THz to 206.3 THz. And constructing a training sample set with 3000-scale training data and 500-scale testing data, and training the initial neural network model. And mapping the state parameters, the target parameters and the equipment parameters to the number of pumps, the pump power and the pump wavelength required for achieving the ideal gain spectrum by using a trained neural network model. Adjusting the light Raman amplifier by using the predicted pumping number, pumping power and pumping wavelength to obtain an actual gain spectrum, calculating the mean square error of the actual gain spectrum and an ideal gain spectrum, if the mean square error is larger than a set threshold value, adjusting the predicted pumping number, the predicted pumping power and the predicted pumping wavelength by using the preset neural network model in combination with a gradient descent method, detecting based on the adjusted predicted pumping number, the predicted pumping power and the predicted pumping wavelength to obtain an adjusted actual gain spectrum, and calculating the adjusted mean square error; until the adjusted mean square error is smaller than the set threshold value, and outputting the adjusted predicted pumping number, predicted pumping power and predicted pumping wavelength to optimize the parameters.
Referring to fig. 3 and 4, the optimization process includes:
1. constructing an initial neural network model; the initial neural network model adopts a Convolutional Neural Network (CNN) structure, and the basic structure of the CNN comprises a feature extraction layer and a feature mapping layer. The characteristic extraction layer comprises a convolution layer, an activation function layer and a pooling layer, the input of each neuron is connected with a local receiving domain of the previous layer, and the characteristic of the local part can be extracted and the position relation can be determined; the feature mapping layer plays a role of a classifier, namely, the result is identified and classified. And training the initial neural network model by using a training sample set to obtain a preset neural network model, wherein the input is the light intensity of the sample signal light to be amplified, the frequency of the sample signal light to be amplified, the central wavelength of the sample signal light to be amplified, the flatness of an actual pumping gain spectrum and the bandwidth of the actual pumping gain spectrum, and the output is the pumping number, the pumping power and the pumping wavelength.
2. Receiving the signal light to be amplified, and detecting the wavelength and power of the signal light.
3. And inputting parameters such as the bandwidth and the flatness of the ideal gain spectrum and the wavelength and the power of the signal to be amplified into the trained preset neural network model to obtain the pumping parameters corresponding to the gain spectrum.
4. And adjusting the pump wave parameters by adjusting the pump control unit, applying the predicted pump parameters to an actual system, and measuring to obtain an actual gain spectrum.
Comparing the MSE and ideal precision of the ideal gain spectrum and the actual gain spectrum and feeding back the MSE and ideal precision to a module of a pumping signal, wherein the specific circulation steps comprise:
s1: a threshold value δ is set.
S2: the MSE of the ideal gain spectrum and the actual gain spectrum is obtained.
S3: and comparing the MSE of the ideal gain spectrum and the actual gain spectrum with a target threshold value delta, and detecting whether the target precision is achieved.
S4: if MSE is less than or equal to delta, ending the circulation and outputting the number of pumps, the pump power and the pump wavelength used for predicting the time.
S5: if MSE is more than delta, parameters such as the number of pumps, pump power, pump wavelength and the like are updated by adopting a gradient descent method, and the pump wave parameters output by the neural network in the previous step are recorded as
Corresponding actual gain spectrum
And the ideal gain spectrum
Average error MSE is noted as
The pump wave parameters are updated according to the following formula:
wherein the content of the first and second substances,
for the updated pump parameter vector,
for the current pump parameter vector, i denotes the number of iterations,
it is indicated that the learning rate is,
which means that the gradient is determined,
representing the difference between the current actual gain spectrum and the ideal gain spectrum. Inputting the updated parameters into the neural network again to obtain a new measured gain spectrum, and calculating the MSE of the new measured gain spectrum and the MSE of the ideal gain spectrum again to be compared with a threshold value until the required precision is reached;
the schematic diagram of the signal gain spectrum before and after optimization by the scheme is shown in fig. 4, and it can be seen that the scheme can achieve the effect of gain spectrum flattening.
In summary, according to the method and the device for adjusting and controlling the dynamic gain of the optical fiber raman amplifier, the state parameter of the signal light to be amplified, the target parameter of the ideal gain spectrum and the device parameter are mapped to obtain the pumping parameter including the pumping number, the pumping power and the pumping wavelength through the preset neural network model obtained through training, the pumping parameter including the predicted pumping number, the predicted pumping power and the predicted pumping wavelength is automatically generated to control the operation of the optical fiber raman amplifier, after the mean square error of the actual gain spectrum and the ideal gain spectrum is calculated, the preset neural network is used for adjusting the pumping parameter by a gradient descent method, optimization is carried out until the mean square error of the adjusted actual gain spectrum and the ideal gain spectrum is smaller than a set threshold value, the pumping parameter can be adjusted rapidly and automatically, and a flat ideal gain spectrum is achieved.
Those of ordinary skill in the art will appreciate that the various illustrative components, systems, and methods described in connection with the embodiments disclosed herein may be implemented as hardware, software, or combinations of both. Whether this is done in hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from 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, plug-in, function card, or the like. When implemented in software, the elements of the invention are the 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 by a data signal carried in a carrier wave over a transmission medium or a communication link. A "machine-readable medium" may include any medium that can store or transfer information. Examples of a machine-readable medium 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 so forth. The code segments may be downloaded via computer networks such as the internet, intranet, etc.
It should also be noted that the exemplary embodiments mentioned in this patent 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-described steps, that is, the steps may be performed in the order mentioned in the embodiments, may be performed in an order different from the order in the embodiments, or may be performed simultaneously.
Features that are described and/or illustrated with respect to one embodiment may be used in the same way or in a similar way in one or more other embodiments and/or in combination with or instead of the features of the other embodiments in the present invention.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes may be made to the embodiment of the present invention by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.