Four-channel F-DFBS array parallel photon reserve pool optical fiber compensation method

Abstract

The invention discloses a four-channel F-DFBS array parallel photon reserve pool optical fiber compensation method comprising the following steps: inputting a pre-processed damaged optical fiber signal into a trained optical fiber compensation model to obtain a compensation signal; wherein the optical fiber compensation model comprises a convolution preprocessing network and a four-channel F-DFBS array parallel photon time delay reserve pool network; the convolution preprocessing network is used for performing preliminary compensation on the normalized damaged optical fiber signal to obtain a preliminarily compensated damaged optical fiber signal; the four-channel F-DFBS array parallel photon time delay reserve pool network is used for carrying out secondary compensation to obtain a compensation signal; and obtaining an evaluation result of the compensation signal according to the mean square error and the error rate. According to the invention, optical fiber damage equalization can be carried out on four paths of signals in parallel, the signal recovery efficiency is obviously improved, a complex photon integrated structure is avoided, the hardware complexity and cost are greatly reduced, and a channel equalization solution which is efficient, practical and easy to deploy is provided.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of signal processing, and particularly relates to a four-channel F-DFBS array parallel photon reservoir pool optical fiber compensation method. BACKGROUND

[0002] Reservoir Computing (RC) as an important paradigm in the field of machine learning has gained sustained attention due to its unique architecture design and efficient training mechanism. The core advantage of the model lies in its structural simplicity, efficient training characteristics and memory ability for dynamic systems, which has important application value in channel equalization, optical fiber compensation and other scenes.

[0003] Optical fiber communication provides a solid foundation for modern communication systems due to its high capacity, low loss and excellent anti-interference ability. However, channel impairments during signal transmission not only cause signal distortion and attenuation, but also limit the transmission distance, thereby affecting the overall performance of the communication system. Therefore, effective signal compensation has become a key research focus. SUMMARY

[0004] In order to solve the above problems existing in the prior art, the application provides a four-channel F-DFBS array parallel photon reservoir pool optical fiber compensation method.

[0005] The technical problem to be solved by the application is solved by the following technical scheme: In a first aspect, the application provides a four-channel F-DFBS array parallel photon reservoir pool optical fiber compensation method, which comprises: normalizing a damaged optical fiber signal to obtain a normalized damaged optical fiber signal; inputting the normalized damaged optical fiber signal into a trained optical fiber compensation model to obtain a compensation signal; wherein the optical fiber compensation model comprises a pre-constructed convolutional preprocessing network and a pre-constructed four-channel F-DFBS array parallel photon time delay reservoir pool network; the convolutional preprocessing network is used to preliminarily compensate the pre-processed damaged optical fiber signal to obtain a preliminarily compensated damaged optical fiber signal; the four-channel F-DFBS array parallel photon time delay reservoir pool network is used to secondarily compensate the preliminarily compensated damaged optical fiber signal to obtain the compensation signal; obtaining an evaluation result of the compensation signal according to the mean square error and the bit error rate.

[0006] Optionally, the training process of the convolutional preprocessing network comprises: constructing a training data set; updating the network parameters of the convolutional preprocessing network through back propagation according to the training data set and the mean square error.

[0007] Optionally, the training process of the four-channel F-DFBS array parallel photonic delay reservoir network includes: The state of the reservoir of the four-channel F-DFBS array parallel photonic delay reservoir network is obtained from the preliminary compensation signal output by the completed convolutional preprocessing network. The signal with secondary compensation is obtained based on the state of the reservoir and the output layer weight matrix of the four-channel F-DFBS array parallel photonic time delay reservoir network. The output layer weight matrix of the four-channel F-DFBS array parallel photonic delay reservoir network is updated based on the signal and ridge regression algorithm after the secondary compensation.

[0008] Optionally, the signal for the secondary compensation is represented as follows: ; in, Indicates the first The state of the reservoir at discrete time. The sampled signal is obtained based on the random mask and the signal obtained from the preliminary compensation. This is the weight matrix between the internal nodes of the reservoir in the parallel photonic delay reservoir network of the four-channel F-DFBS array. For time delay interval, For time intervals, It is a non-linear activation function.

[0009] Optionally, the time interval is represented as follows: ; in, It is a virtual node.

[0010] Alternatively, each virtual node can be represented according to the Langevin-Kobayashi rate equation as follows: ; ; in, This represents the first [channel] in the parallel photonic delay reservoir network of the four-channel F-DFBS array. The distributed feedback laser in the first The complex amplitude of the slowly varying electric field corresponding to a continuous time interval. This represents the first [channel] in the parallel photonic delay reservoir network of the four-channel F-DFBS array. The distributed feedback laser in the first The average carrier number density corresponding to each consecutive time period, where α is the linewidth enhancement factor. Represents the differential gain coefficient It is the gain compression factor denotes the transparent carrier density, is the photon lifetime, is the carrier lifetime is the feedback strength, denotes the laser angular frequency, denotes the injection strength, denotes the first distributed feedback laser in the first continuous-time received sampled signal.

[0011] Optionally, the evaluation result includes a mean square error value and a bit error rate; the mean square error value is obtained according to the following formula: ; wherein, denotes the mean square error value, denotes the number of signal sample values, denotes the first signal sample value of the original signal corresponding to the damaged optical fiber signal, denotes the first signal sample value of the compensation signal.

[0012] Optionally, the bit error rate is obtained according to the following formula: ; wherein, denotes the bit error rate, denotes the number of bits that are incorrectly received or detected as errors in the signal transmission process, denotes the total number of bits in the signal transmission process.

[0013] The technical scheme provided by the embodiment of the application can include the following beneficial effects: In the above technical scheme, the application proposes an optical fiber compensation model, wherein the four-channel F-DFBS array parallel photon time delay reservoir network is realized based on four distributed feedback lasers (F-DFBs) array, which is suitable for optical platform hardware implementation, can perform optical fiber damage equalization on four parallel signals, and significantly improves the signal recovery efficiency. The model uses a single nonlinear node of a distributed feedback laser, combines time division multiplexing to construct a virtual node, avoids complex photon integrated structure, greatly reduces hardware complexity and cost, and provides an efficient, practical and easy-to-deploy channel equalization solution.

[0014] The application will be further described in detail below in combination with the drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a flowchart of a four-channel F-DFBS array parallel photon reserve pool optical fiber compensation method provided by an embodiment of the application; Figure 2 is a model running flowchart provided by an embodiment of the application; Figure 3 is a test result diagram provided by an embodiment of the application; Figure 4 is a hardware experimental device diagram provided by an embodiment of the application; Figure 5 is a diagram of a constellation diagram corresponding to the compensated signal. DETAILED DESCRIPTION

[0016] The application will be further described in detail below in combination with specific embodiments, but the embodiments of the application are not limited thereto.

[0017] Figure 1 is a flowchart of a four-channel F-DFBS array parallel photon reserve pool optical fiber compensation method provided by an embodiment of the application, as shown in Figure 1 , the method can include the following steps: S101, normalizing the damaged optical fiber signal to obtain a normalized damaged optical fiber signal.

[0018] Specifically, the damaged optical fiber signal is an 8 / 16 / 32 / 64QAM four-channel signal, and the process of normalizing the damaged optical fiber signal includes: wherein, is the damaged optical fiber signal, and are the minimum value and the maximum value in , and is the normalized damaged optical fiber signal.

[0019] S102, inputting the normalized damaged optical fiber signal into a trained optical fiber compensation model to obtain a compensation signal; wherein the optical fiber compensation model includes a pre-constructed convolutional preprocessing network and a pre-constructed four-channel F-DFBS array parallel photon time delay reserve pool network; the convolutional preprocessing network is used to preliminarily compensate the normalized damaged optical fiber signal to obtain a preliminarily compensated damaged optical fiber signal; the four-channel F-DFBS array parallel photon time delay reserve pool network is used to secondarily compensate according to the preliminarily compensated damaged optical fiber signal to obtain the compensation signal.

[0020] ​Understandably, convolutional neural networks use 1D convolutional kernels to perform nonlinear transformations on the input features at each time step, multiple filters to extract local features, and the ReLU activation function. The data extracted in the preceding steps is mapped to a fully connected layer, which uses two linear mapping layers to continuously reduce the number of neurons. Finally, a preliminary compensated signal is obtained through a 2D output layer.

[0021] The input fiber compensation model includes a trained convolutional preprocessing network and a trained four-channel F-DFBS array parallel photonic delay reservoir network.

[0022] Optionally, the training process of the convolutional preprocessing network includes: Build the training dataset; The network parameters of the convolutional preprocessing network are updated via backpropagation based on the training dataset and mean squared error.

[0023] Specifically, for each damaged fiber signal, the complex damaged fiber signal is first split into two real channels: a real part and an imaginary part. Then, local features are extracted from these two real channels using a sliding window to form the input feature matrix. Simultaneously, based on the target signal dataset, the original signal dataset is obtained from the undamaged original signals corresponding to the damaged fiber signals. The damaged fiber signal corresponding to the center position of each sliding window is taken as the label of the window's features, constructing a label matrix. Each sample in the training dataset includes a feature matrix and a corresponding label matrix. The initial learning rate of the convolutional preprocessing network and a callback function for adjusting the learning rate are set; the mean squared error (MSE) is used as the loss function, and the weights and biases of the convolutional preprocessing network are updated through backpropagation. The trained weights are read, and the test dataset is used for testing. The construction method of the test dataset can refer to the training dataset.

[0024] For a four-channel F-DFBS array parallel photonic delay reservoir network, a random mask signal is generated during initialization. Weight matrix connecting internal reservoir nodes The update rule for the state of neurons inside the reservoir of the four-channel F-DFBS array parallel photonic delay reservoir network is as follows: ; in, Indicates the first To ensure that the input signal can effectively access the entire dimension of the reservoir and to keep the system transient, the state of the reservoir at discrete time is considered. The sampled signal is obtained based on the random mask and the signal with preliminary compensation. This is the internal connection matrix of the parallel photonic delay reservoir network of a four-channel F-DFBS array. is a time delay interval, is a time interval, is a nonlinear activation function, typically , provides storage capability for a four-channel F-DFBS array parallel photonic time-delay reservoir network. In the current continuous-time reservoir, the state of a neuron node depends on its own state within a continuous-time interval . Define time points as virtual nodes, the time distance between them keeps consistent, separated by . This

[0025] The final output layer is represented by a linear combination of the neuron node states in the reservoir:

[0026] where, is the signal of the quadratic compensation, is the weight matrix of the output layer.

[0027] Distributed feedback lasers are used as virtual nodes of the nonlinearity, and four channels are implemented through a distributed feedback laser array. Each virtual node is represented by the Langmuir-kinoshita rate equation as follows: ; ; where, represents the complex amplitude of the slowly-varying electric field of the th distributed feedback laser in the four-channel F-DFBS array parallel photonic time-delay reservoir network at the th continuous time, represents the average carrier number density of the th distributed feedback laser in the four-channel F-DFBS array parallel photonic time-delay reservoir network at the th continuous time, and α is the linewidth enhancement factor represents the differential gain coefficient is the gain compression factor represents the transparent carrier density, is the photonic lifetime, is the carrier lifetime is the feedback strength, represents the laser angular frequency, represents the injection strength, represents the The distributed feedback laser in the first One of the sampled signals received in a continuous time period This represents the injected current density.

[0028] It can be described by the following equation.

[0029] ; in, Indicates the amplitude of the injected field. This represents the bias voltage of the Mach-Zehnder modulator. yes With the The detuning frequency between the distributed feedback lasers. Furthermore, It is the first Feature values ​​of the path corresponding to each distributed feedback laser. It is the first The mask matrix of the path corresponding to each distributed feedback laser. The equations are solved numerically using the second-order Runge-Kutta method with a step size of 2 picoseconds.

[0030] The pre-compensated signals (8 / 16 / 32 / 64QAM four-channel signals) obtained through the convolution preprocessing module are converted into one-dimensional signals with alternating real and imaginary parts, and used as inputs to the parallel photonic delay reservoir network of the four-channel F-DFBS array for secondary signal compensation.

[0031] S103. Obtain the evaluation results of the compensation signal based on the mean square error and the bit error rate.

[0032] Optionally, the evaluation results include the mean squared error (MSE) and the bit error rate (BER); the MSE is obtained according to the following formula: ; in, This represents the mean square error value. Indicates the number of signal sample values. This represents the first undamaged original signal corresponding to the damaged fiber optic signal. Each signal sample value, The first compensation signal Each signal sample value.

[0033] The bit error rate is obtained using the following formula: ; in, This represents the bit error rate; the lower the bit error rate, the better the compensation effect. This indicates the number of bits that were incorrectly received or detected during signal transmission. The total number of bits in the signal transmission process is represented.

[0034] In an embodiment, the flow of the present application can be divided into two parts: software and hardware, steps 1 to 7 will explain the software flow, and steps 8 to 9 will explain the hardware flow.

[0035] Figure 2 is a model running flow diagram provided by an embodiment of the present application, as Figure 1 shown, step 1: first determine whether the CPU device of the host is normal; store the 8QAM, 16QAM, 32QAM and 64QAM damaged optical fiber signal and target signal data set file in the current working directory. Save the normalized damaged optical fiber signal data set; split the complex signal into two real channels to form a two-dimensional feature matrix. A sliding window with a length of 49 is used to extract features. In this process, the signal is mirror filled to avoid losing edge information. The input feature matrix with a dimension of (N, 49, 2) is obtained, where N is the number of samples, 49 is the window length, and 2 is the number of real / imaginary channels. The real and imaginary parts of the target signal at the center position of the window form a two-dimensional label matrix ; divide the data set, the first 70% is the training set, and the last 30% is the test set.

[0036] Step 2: preliminary feature extraction is performed on the damaged optical fiber signal data. Convolutional neural network parameter setting: the convolutional neural network adopts a three-layer 1D convolution structure (Conv1, Conv2, Conv3), each layer is configured with a convolution kernel with a length of 15 and uses same padding, the first layer uses 16 filters, and the second and third layers are 32 filters to extract features layer by layer. Finally, it is activated through the ReLU function. The Flatten flattening layer converts the three-dimensional feature map into a 1568-dimensional vector, which is further compressed by two 16-dimensional fully connected layers, and finally the 2-dimensional output layer is used to regress the real and imaginary parts of the QAM symbol.

[0037] Step 4: use mean square error (MSE) as the loss function, set the epoch number and batch size to 80 and 32, and update the weights according to the backpropagation before the end of the epoch. The model is based on the Adam optimizer, and the weights are updated according to the gradient calculated by the backpropagation after each batch training. The initial learning rate is set to 0.001. If the validation set loss does not decrease within 10 training rounds, the learning rate will be multiplied by 0.1 to reduce it. After training is completed, the model directly processes the test set data, combines the real and imaginary parts obtained by processing into complex form, and obtains the preliminary recovered QAM signal. The recovered signal is saved in text file and NPZ format to the current directory.

[0038] The length of the signal entering the four-channel F-DFBS array in parallel photonic time-delay reservoir pool network is 120000, and the idle point is 100. The number of virtual nodes is 90. The mask matrix is a matrix with a size of 192xNum of virtual nodes, and the elements are randomly selected as -1, 1 or 3. In addition, the virtual node interval is set to 1 ns, and the time delay interval is 1 ns. The parameter values of the four distributed feedback laser reservoir pool computing systems in the numerical simulation are set as follows: Table 1

[0039] Step 6: The training set of four pieces of convolution-preprocessed QAM signals is simultaneously input into the four distributed feedback laser (F-DFB) reservoir pool computing (RC) systems with set parameters. The length of the training set is 100000. The weight matrix in the output layer is trained by using ridge regression.

[0040] Step 7: The test set sample data is input, and the trained weight is imported to compensate the damaged optical fiber signal of the test set. The constellation diagram corresponding to the recovered signal is clear in profile and good in constellation point gathering degree, Figure 3 is a test result schematic diagram provided by an embodiment of the present application.

[0041] The index data is shown in Table 2: Table 2

[0042] Step 8: After the simulation training is completed, the model can be deployed on a hardware experimental platform. Figure 4 is a hardware experimental device diagram provided by an embodiment of the present application, as shown in Figure 4As shown, the system is driven by tunable lasers (TL1, TL2, TL3, TL4) as driving lasers to provide optical carriers. The results of four modulation format impaired fiber signals and mask signals after processing are respectively encoded into arbitrary waveform generators (AWG1, AWG2, AWG3, AWG4) to generate at a speed of 1GSa / s. Then the generated input signals are sent to the radio frequency port of electro-optical modulators (MZM1, MZM2, MZM3, MZM4). The polarization state of light is adjusted by polarization controllers (PC1, PC2, PC3 and PC4) through MZM loading on the optical carrier. The injection intensity of the input signal is regulated by variable optical attenuators (VOA1, VOA2, VOA3, VOA4) and erbium-doped fiber amplifiers (EDFA1, EDFA2, EDFA3, EDFA4). Then, each input signal is injected into the distributed feedback laser in the array through the optical circulator. Subsequently, the output light of the distributed feedback laser is converted into an electrical signal by a photoelectric converter PD, and finally the time domain characteristics are observed by an oscilloscope OSC and the output is recorded. Due to the limitation of too much data in the experiment, the sampling rate of the OSC is set to 2GSa / s.

[0043] Step 9: Post-processing of oscilloscope collected data. Figure 5 is a schematic diagram of the constellation diagram corresponding to the compensated signal, as Figure 5 shown, The index data is as follows: Table 3

[0044] The application can simultaneously equalize four-way signals. Based on the optical transmission simulation platform, the transmission process of 8QAM, 16QAM, 32QAM and 64QAM signals in a single-mode fiber is simulated with high precision to obtain a distorted signal data set. Then, the convolutional neural network is used to extract features and preliminarily denoise the impaired optical signal. Finally, a time delay reserve pool structure containing a four-way distributed feedback laser array, i.e. a four-channel F-DFBS array parallel photonic time delay reserve pool network, is constructed to realize the parallel recovery of the four modulation format signals. The four-way parallel signal integrated network architecture improves the efficiency limitation of traditional single-channel signal processing. The method uses a distributed feedback laser as a nonlinear node and constructs a virtual node through time division multiplexing, avoiding the complexity of physical structure and interconnection of multi-node integrated photonic chips, significantly reducing the hardware complexity and system cost, and providing an innovative solution with hardware parallel advantages for practical channel equalization of high-speed optical communication systems.

[0045] It is to be understood that the terms "first", "second", and the like, used in the description and in the claims, are used to describe different objects, and do not imply a particular order or sequence. The use of the terms "first" and "second" are not meant to denote a particular order or sequence, but are used to distinguish one element from another. It is to be understood that the data used herein can be interchanged, where appropriate, so that the embodiments described herein can be carried out in sequences other than those illustrated or described herein. The implementations described in the following example embodiments are not meant to represent all implementations consistent with the present application. Rather, they are simply examples of apparatus and methods consistent with some aspects of the present application.

[0046] In the description of the description, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the description, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in the description.

[0047] Although the present application is described herein in conjunction with various embodiments, those skilled in the art, by viewing the drawings and the disclosure, can understand and implement other variations of the disclosed embodiments in the implementation of the claimed application. In the description of the present application, the word "comprising" does not exclude other components or steps, "one" or "one" does not exclude multiple, "multiple" means two or more, unless otherwise explicitly specified. In addition, some measures are described in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0048] The above is a further detailed description of the present application in conjunction with specific preferred embodiments, and cannot be considered as limiting the specific implementation of the present application to these descriptions. For those skilled in the art, without departing from the concept of the present application, a number of simple deductions or substitutions can be made, which should be considered as falling within the scope of protection of the present application.

Claims

1. A fiber compensation method for a four-channel F-DFBS array parallel photonic reservoir, characterized in that, The method includes: The damaged fiber optic signal is normalized to obtain the normalized damaged fiber optic signal. The normalized damaged fiber signal is input into the trained fiber compensation model to obtain a compensation signal. The fiber compensation model includes a pre-built convolutional preprocessing network and a pre-built four-channel F-DFBS array parallel photonic delay reservoir network. The convolutional preprocessing network performs preliminary compensation on the pre-processed damaged fiber signal to obtain a pre-compensated damaged fiber signal. The four-channel F-DFBS array parallel photonic delay reservoir network performs secondary compensation based on the pre-compensated damaged fiber signal to obtain the compensation signal. The evaluation result of the compensated signal is obtained based on the mean square error and the bit error rate.

2. The fiber compensation method for a four-channel F-DFBS array parallel photonic reservoir according to claim 1, characterized in that, The training process of the convolutional preprocessing network includes: Build the training dataset; Based on the training dataset and mean squared error, the network parameters of the convolutional preprocessing network are updated via backpropagation.

3. The fiber compensation method for a four-channel F-DFBS array parallel photonic reservoir according to claim 1, characterized in that, The training process of the four-channel F-DFBS array parallel photonic delay reservoir network includes: The state of the reservoir of the four-channel F-DFBS array parallel photonic delay reservoir network is obtained from the preliminary compensation signal output by the completed convolutional preprocessing network. The signal with secondary compensation is obtained based on the state of the reservoir and the output layer weight matrix of the four-channel F-DFBS array parallel photonic time delay reservoir network. The output layer weight matrix of the four-channel F-DFBS array parallel photonic delay reservoir network is updated based on the signal and ridge regression algorithm after the secondary compensation.

4. The fiber compensation method for a four-channel F-DFBS array parallel photonic reservoir according to claim 3, characterized in that, The update rule for the state of neurons inside the reservoir of the four-channel F-DFBS array parallel photonic delay reservoir network is as follows: ； in, Indicates the first The state of the reservoir at discrete time. The sampled signal is obtained based on the random mask and the signal obtained from the preliminary compensation. This is the weight matrix between the internal nodes of the reservoir in the parallel photonic delay reservoir network of the four-channel F-DFBS array. For time delay interval, For time intervals, It is a non-linear activation function.

5. The fiber compensation method for a four-channel F-DFBS array parallel photonic reservoir according to claim 4, characterized in that, The time interval is represented as follows: ； in, It is a virtual node.

6. The fiber compensation method for a four-channel F-DFBS array parallel photonic reservoir according to claim 5, characterized in that, Each virtual node is represented according to the Langevin-Kobayashi rate equation as follows: ； ； in, This represents the first [channel] in the parallel photonic delay reservoir network of the four-channel F-DFBS array. The distributed feedback laser in the first The complex amplitude of the slowly varying electric field corresponding to a continuous time interval. This represents the first [channel] in the parallel photonic delay reservoir network of the four-channel F-DFBS array. The distributed feedback laser in the first The average carrier number density corresponding to each consecutive time period, where α is the linewidth enhancement factor. Represents the differential gain coefficient It is the gain compression factor Represents the transparent carrier density. For photon lifetime, For carrier lifetime It is the feedback intensity. Represents the laser angular frequency. Indicates injection intensity. Indicates the first The distributed feedback laser in the first One of the sampled signals received in a continuous time period.

7. The fiber compensation method for a four-channel F-DFBS array parallel photonic reservoir according to claim 1, characterized in that, The evaluation results include the mean squared error (MSE) and the bit error rate (BER); the MSE is obtained according to the following formula: ； in, This represents the mean square error value. Indicates the number of signal sample values. The first undamaged original signal corresponding to the damaged fiber optic signal represents the... Each signal sample value, The first of the compensation signals Each signal sample value.

8. The fiber compensation method for a four-channel F-DFBS array parallel photonic reservoir according to claim 1, characterized in that, The bit error rate is obtained according to the following formula: ； in, This represents the bit error rate. This indicates the number of bits that were incorrectly received or detected during signal transmission. This indicates the total number of bits during signal transmission.

Images