Optical signal processing method and device, electronic equipment and nonvolatile storage medium

Abstract

The invention discloses an optical signal processing method and device, electronic equipment and a nonvolatile storage medium. The method comprises the following steps: acquiring a first optical signal; a complex numerical convolutional neural network is adopted to perform feature extraction on the first optical signal to obtain a first feature sequence, and the complex numerical convolutional neural network is used for removing noise in the first optical signal through feature extraction; and determining a target dependency relationship according to the first feature sequence by adopting a bidirectional long-short-term memory network, and eliminating intersymbol interference in the first optical signal according to the target dependency relationship to obtain a second optical signal, the target dependency relationship is used for representing the dependency relationship between each complex value symbol in the first optical signal and other adjacent complex value symbols, and the second optical signal is an optical signal after nonlinear equalization. The technical problems of high complexity and poor effect of a scheme for carrying out nonlinear equalization on a high-order optical signal in related technologies are solved.

Description

Technical Field

[0001] The present application relates to the field of optical transmission technology, and in particular, to an optical signal processing method, device, electronic device and non-volatile storage medium. Background Art

[0002] With the vigorous development of various emerging technologies, challenges have been raised to the transmission capacity of the underlying backbone network. The new generation of information transmission requires large capacity, good quality and high spectrum efficiency. In order to solve the above problems, the current mainstream solution is to use multiplexing technologies such as wavelength division multiplexing and space division multiplexing. Another solution is to start from the signal and further increase the modulation order of the signal, thereby improving the spectrum efficiency of the signal from the modulation format. However, the problem that follows is that the difficulty and complexity of digital signal processing for high-order signals are also greatly increased.

[0003] In coherent optical transmission systems, high-order signals are greatly affected by the transmission system, especially the nonlinearity. High-order signals are not only affected by the nonlinearity of the transmission link, but also by the devices, which seriously reduces the signal quality. On the one hand, it limits the stable long-distance transmission of the signal, and on the other hand, it also brings difficulties to the successful demodulation of the signal. However, most of the solutions to the nonlinearity of coherent optical transmission systems in the current related technologies have technical problems such as high complexity and poor effect.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present application provide an optical signal processing method, device, electronic device and non-volatile storage medium to at least solve the technical problems of high complexity and poor effect of nonlinear equalization solutions for high-order optical signals in related technologies.

[0006] According to one aspect of an embodiment of the present application, there is provided a method for processing an optical signal, comprising: obtaining a first optical signal, wherein the first optical signal is optical signal data in complex form, and the optical signal data includes a plurality of complex-valued symbols arranged in a time series order; using a complex-valued convolutional neural network to perform feature extraction on the first optical signal to obtain a first feature sequence, wherein the complex-valued convolutional neural network is used to remove noise in the first optical signal by performing feature extraction; using a bidirectional long short-term memory network to determine a target dependency relationship based on the first feature sequence, and based on the target dependency relationship, eliminate inter-symbol interference in the first optical signal to obtain a second optical signal, wherein the target dependency relationship is used to characterize the dependency relationship between each complex-valued symbol in the first optical signal and other complex-valued symbols immediately before and after, and the second optical signal is an optical signal after nonlinear equalization.

[0007] Optionally, a complex-valued convolutional neural network is used to perform feature extraction on the first optical signal to obtain a first feature sequence, including: using a real convolution kernel in the complex-valued convolutional neural network to perform a convolution operation on the real data of each complex-valued symbol in the first optical signal to obtain a real sub-feature; using an imaginary convolution kernel in the complex-valued convolutional neural network to perform a convolution operation on the imaginary data of each complex-valued symbol in the first optical signal to obtain an imaginary sub-feature; using a complex activation function to determine the first feature sequence based on the real sub-features and the imaginary sub-features.

[0008] Optionally, the bidirectional long short-term memory network includes: a first long short-term memory layer and a second long short-term memory layer, wherein the first long short-term memory layer and the second long short-term memory layer each include a plurality of long short-term memory recursive units; using the bidirectional long short-term memory network, based on the first feature sequence, determining the target dependency relationship includes: using the long short-term memory recursive units in the first long short-term memory layer to analyze the features corresponding to each complex-valued symbol in the first feature sequence in a direction of time order to obtain forward state information; using the long short-term memory recursive units in the second long short-term memory layer to analyze the features corresponding to each complex-valued symbol in the first feature sequence in a direction of time reverse order to obtain reverse state information; merging the forward state information and the reverse state information to obtain the target dependency relationship.

[0009] Optionally, the step of analyzing the features corresponding to the complex-valued symbols by the long short-term memory recursive unit includes: determining an input gate output value, a forget gate output value, and an output gate output value based on a first input value and a second external state value, wherein the first input value is a feature of the complex-valued symbol corresponding to the first long short-term memory recursive unit in the first feature sequence, the second external state value is an external state value output by the second long short-term memory recursive unit, and the second long short-term memory recursive unit is a previous long short-term memory recursive unit adjacent to the first long short-term memory recursive unit in the first long short-term memory layer or the second long short-term memory layer; updating the second memory unit state value based on the input gate output value and the forget gate output value to obtain a first memory unit state value corresponding to the first long short-term memory recursive unit, wherein the second memory unit state value is a memory unit state value output by the second long short-term memory recursive unit; determining the first external state value corresponding to the current long short-term memory recursive unit based on the output gate output value and the first memory unit state value.

[0010] Optionally, the target model includes: a complex-valued convolutional neural network and a bidirectional long short-term memory network; the training step of the target model includes: obtaining a training data set, wherein the training data set includes multiple groups of optical signal data under different optical signal-to-noise ratios; during the model training process, the complex-valued convolutional neural network in the initial model is used to extract features of the optical signal data in the training data set to obtain a feature sequence, and the bidirectional long short-term memory network in the initial model is used to perform nonlinear equalization on the optical signal data based on the feature sequence; based on the optical signal data after nonlinear equalization, a loss function value is determined, and an adaptive moment estimation algorithm is used to adjust the model parameters in the initial model based on the loss function value; and the initial model after the model parameters are adjusted is used to repeat the above training process until the target model is obtained.

[0011] Optionally, the above training process is repeated using the initial model after the model parameters are adjusted until the target model is obtained, including: when the loss function value is less than a preset error threshold, or the number of iterations of the model training process exceeds a preset threshold, the initial model after the model parameters are adjusted is determined as the target model.

[0012] Optionally, the first optical signal is optical signal data in a complex form transmitted in a coherent optical communication system, wherein the coherent optical communication system includes: a transmitting end and a receiving end, wherein the transmitting end is used to map a pseudo-random binary sequence into signals with different amplitude levels through orthogonal amplitude modulation encoding, and upsample and filter the signal; import the signal after upsampling and filtering into a waveform generator to obtain an analog level signal; load the analog level signal onto an orthogonal amplitude IQ modulator to obtain a modulation format signal; load the modulation format signal onto a narrow linewidth laser to obtain an optical signal, and send the optical signal to the receiving end through a transmission link; the receiving end is used to perform analog-to-digital conversion on the received optical signal, and perform digital signal processing on the optical signal after the analog-to-digital conversion to obtain a first optical signal, wherein the digital signal processing includes: resampling, matched filtering, IQ orthogonalization, dispersion compensation, and clock recovery.

[0013] According to another aspect of an embodiment of the present application, an optical signal processing device is also provided, including: a data acquisition module, used to acquire a first optical signal, wherein the first optical signal is optical signal data in complex form, and the optical signal data includes multiple complex-valued symbols arranged in a time series order; a feature extraction module, used to use a complex-valued convolutional neural network to extract features from the first optical signal to obtain a first feature sequence, wherein the complex-valued convolutional neural network is used to remove noise in the first optical signal by performing feature extraction; a signal equalization module, used to use a bidirectional long short-term memory network to determine a target dependency relationship based on the first feature sequence, and based on the target dependency relationship, eliminate inter-symbol interference in the first optical signal to obtain a second optical signal, wherein the target dependency relationship is used to characterize the dependency relationship between each complex-valued symbol in the first optical signal and other complex-valued symbols immediately before and after, and the second optical signal is an optical signal after nonlinear equalization.

[0014] According to another aspect of the embodiments of the present application, there is also provided an electronic device, including: a memory and a processor, wherein the processor is used to run a program stored in the memory, wherein the optical signal processing method is executed when the program is run.

[0015] According to another aspect of the embodiments of the present application, a non-volatile storage medium is provided, wherein the non-volatile storage medium includes a stored computer program, wherein a device where the non-volatile storage medium is located executes the optical signal processing method by running the computer program.

[0016] According to another aspect of the embodiments of the present application, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps of the optical signal processing method are implemented.

[0017] In an embodiment of the present application, a first optical signal is obtained, wherein the first optical signal is optical signal data in complex form, and the optical signal data includes multiple complex-valued symbols arranged in a time series order; a complex-valued convolutional neural network is used to extract features from the first optical signal to obtain a first feature sequence, wherein the complex-valued convolutional neural network is used to remove noise in the first optical signal by performing feature extraction; a bidirectional long short-term memory network is used to determine a target dependency relationship based on the first feature sequence, and based on the target dependency relationship, the inter-symbol interference in the first optical signal is eliminated to obtain a second optical signal, wherein the target dependency relationship is used to characterize the dependency relationship between each complex-valued symbol in the first optical signal and other complex-valued symbols immediately before and after, and the second optical signal is an optical signal after nonlinear equalization. By performing feature extraction based on a complex convolutional neural network and performing signal equalization in combination with a bidirectional long short-term memory network, the purpose of achieving nonlinear equalization at low complexity is achieved, thereby solving the technical problems of high complexity and poor effect of the scheme for nonlinear equalization of high-order optical signals in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0019] Figure 1 It is a hardware structure block diagram of a computer terminal (or electronic device) for implementing a method for optical signal processing provided in an embodiment of the present application;

[0020] Figure 2 is a schematic diagram of a method flow for optical signal processing provided according to an embodiment of the present application;

[0021] Figure 3 is a schematic diagram of the architecture of a coherent optical communication system provided according to an embodiment of the present application;

[0022] Figure 4 It is a schematic diagram of a CV-CNN+bi-LSTM network structure provided according to an embodiment of the present application;

[0023] Figure 5 It is a structural diagram of an LSTM recursive unit provided according to an embodiment of the present application;

[0024] Figure 6 It is a schematic diagram of comparing the results of different equalization methods provided according to an embodiment of the present application;

[0025] Figure 7 It is a structural schematic diagram of an optical signal processing device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] In order to facilitate those skilled in the art to better understand the embodiments of the present application, some technical terms or nouns involved in the embodiments of the present application are explained as follows:

[0029] CV-CNNs (Complex-Valued Convolutional Neural Networks): is a special type of convolutional neural network that processes data and parameters in complex form rather than traditional real values. In some application scenarios, such as signal processing, image processing (especially scenarios where phase information is important), radar signal analysis, etc., complex-valued CNNs can more effectively capture and utilize the amplitude and phase information of the signal.

[0030] LSTM (Long Short-Term Memory): is a special type of recurrent neural network (RNN). LSTM is designed to overcome the gradient vanishing or gradient exploding problems encountered by RNN during network training, especially in processing long-term dependencies of sequence data.

[0031] In the related technologies, the solutions to the nonlinearity of coherent optical transmission systems mainly include the following: the first is digital back propagation, which simulates the inverse process of the nonlinear Schrödinger equation by using the numerical method of distributed Fourier, and performs virtual fiber back propagation on the received signal to achieve the effect of nonlinear compensation. The whole calculation process involves a large number of Fourier transforms, which is very complex. Moreover, with the increase of distance, other effects such as dispersion will also interact with each other, which requires continuous simplification of the algorithm and performance optimization. The second is optical phase conjugation, which is to conjugate the optical signal in the optical fiber link. Yoke processing, theoretically, can offset the noise accumulated in the first half of signal transmission in the second half, but due to many influencing factors, the actual effect is general; the third type is the most commonly used Volterra nonlinear equalization (VNLE), which is also a nonlinear equalization algorithm more suitable for high-order signals. Since high-order signals are very sensitive to the nonlinearity of the device, the outer circle signal points of the signal constellation diagram are easily affected and distorted. However, as the system requirements become higher and higher, VNLE needs to perform high-order equalization for high-order signals. When the third-order nonlinear kernel is used, the complexity of the entire algorithm is very high.

[0032] In order to solve the above problems, a solution that can balance performance and complexity is provided in the embodiment of the present application. By building a neural network, it is only necessary to complete complex calculations in the training phase. The subsequent use complexity is low and the performance is excellent, which is conducive to the development of actual optical modules. The following is a detailed description.

[0033] According to an embodiment of the present application, an embodiment of a method for processing an optical signal is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0034] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal (or electronic device) for implementing an optical signal processing method. Figure 1As shown, the computer terminal 10 (or electronic device) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0035] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or electronic device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0036] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the optical signal processing method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the above-mentioned optical signal processing method is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0037] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0038] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or electronic device).

[0039] In the above operating environment, the embodiment of the present application provides an optical signal processing method. Figure 2 is a schematic diagram of a method flow for optical signal processing provided according to an embodiment of the present application, such as Figure 2 As shown, the method comprises the following steps:

[0040] Step S202, obtaining a first optical signal, wherein the first optical signal is optical signal data in a complex form, and the optical signal data includes a plurality of complex-valued symbols arranged in a time sequence;

[0041] Step S204, using a complex-valued convolutional neural network to perform feature extraction on the first optical signal to obtain a first feature sequence, wherein the complex-valued convolutional neural network is used to remove noise in the first optical signal by performing feature extraction;

[0042] Step S206, using a bidirectional long short-term memory network, according to the first feature sequence, determining the target dependency, and according to the target dependency, eliminating the inter-symbol interference in the first optical signal to obtain a second optical signal, wherein the target dependency is used to characterize the dependency between each complex-valued symbol in the first optical signal and other adjacent complex-valued symbols, and the second optical signal is an optical signal after nonlinear equalization.

[0043] Through the above steps, by performing feature extraction based on a complex convolutional neural network and combining a bidirectional long short-term memory network for signal equalization, the purpose of realizing nonlinear equalization at low complexity is achieved, thereby solving the technical problems of high complexity and poor effect of nonlinear equalization solutions for high-order optical signals in related technologies.

[0044] The optical signal processing method in steps S202 to S206 of the embodiment of the present application is further introduced below.

[0045] Related technology When using LSTM for nonlinear compensation, the transmitted M-QAM (M-ary QuadratureAmplitude Modulation, multi-level orthogonal amplitude modulation) signal sequence is used as the feature vector of the input network, which easily causes the network to fall into overfitting, and because the complexity calculation is related to the input features, the complexity is extremely high at this time, and it is not suitable for direct use; if the real and imaginary parts of the M-QAM signal are directly used as input, it is not considered that the pulse broadening caused by the nonlinear effect and the dispersion effect will introduce certain inter-symbol interference, so that there will be a certain correlation between the symbols, which may affect the nonlinear compensation of the neural network due to the lack of input features. Therefore, in the embodiment of the present application, by introducing CV-CNN to further fully extract the features of the original signal as the input of LSTM, the dependence of LSTM on signal features is compensated. At the same time, CV-CNN corrects the shortcomings of the original CNN in processing the signal sequence on one-dimensional convolution, and by directly processing complex signals, it is conducive to extracting more signal features.

[0046] On the other hand, in the process of coherent optical signal transmission, the output of the network at the current moment is not only related to the previous input information, but also to the subsequent input information. If only one side of the information is considered, it is likely to get the wrong result. Therefore, in the embodiment of the present application, it is considered to add a network layer arranged in reverse time to improve the accuracy of the network. Their inputs are the same, but the direction of information transmission is different. One layer is transmitted in chronological order and the other layer is transmitted in reverse time. The two hidden neuron states are processed by vector splicing operations and connected to the output layer neurons. Multiple paired neurons are the performance states of the same pair of neurons at different times. Therefore, the bi-LSTM network has a more superior performance for ultra-high-order coherent optical networks. It can more effectively handle the intersymbol interference (ISI) between the previous and next code elements.

[0047] In summary, the embodiment of the present application proposes a nonlinear equalization scheme based on a complex convolutional neural network (CV-CNN) combined with a bidirectional long short-term memory network (bi-LSTM) to solve the nonlinear damage of optical signal transmission. During the link transmission process, the signal sent at a certain moment will be affected by both linear and nonlinear factors, and the damage suffered by the symbol at the current moment is also related to multiple adjacent symbols. For the cumulative damage of such adjacent signals, LSTM has excellent performance in dealing with the problem of strong correlation of time series, and has excellent modeling capabilities, which can extract and learn the correlation between symbols. CNN has a much higher accuracy than other neural network models in feature extraction. Since high-order signal optical transmission is more susceptible to damage, the convolution layer of CNN can remove part of the signal noise and enhance the reception quality of the signal. According to the characteristics of the transmission signal, the system mainly transmits complex value sequences. This scheme selects a one-dimensional convolution structure for signal processing, calculates the delay accumulation characteristics of the signal, and does not perform convolution on the height.

[0048] The following is a detailed description of the high-order signal nonlinear equalization scheme based on a complex convolutional neural network combined with a bidirectional long short-term memory network in an embodiment of the present application.

[0049] First, a first optical signal that needs to be nonlinearly equalized is obtained. For example, in this embodiment, the first optical signal may be a signal recovered by a receiving end in a related optical communication transmission system, as described below.

[0050] In some embodiments of the present application, the first optical signal is optical signal data in a complex form transmitted in a coherent optical communication system, wherein the coherent optical communication system includes: a transmitting end and a receiving end, wherein the transmitting end is used to map a pseudo-random binary sequence into signals with different amplitude levels through orthogonal amplitude modulation encoding, and upsample and filter the signal; import the signal after upsampling and filtering into a waveform generator to obtain an analog level signal; load the analog level signal onto an orthogonal amplitude IQ modulator to obtain a modulation format signal; load the modulation format signal onto a narrow linewidth laser to obtain an optical signal, and send the optical signal to the receiving end through a transmission link; the receiving end is used to perform analog-to-digital conversion on the received optical signal, and perform digital signal processing on the optical signal after analog-to-digital conversion to obtain a first optical signal, wherein the digital signal processing includes: resampling, matched filtering, IQ orthogonalization, dispersion compensation, and clock recovery.

[0051] Specifically, Figure 3As shown, at the transmitting end, the pseudo-random binary (PRBS) sequence is mapped into amplitude level through QAM coding, and after probability shaping coding, the single symbol entropy is 7.8bit / s / Hz. After upsampling and root raised cosine filtering, the signal level is input into the arbitrary waveform generator to generate the analog signal level. Different amplitude levels are loaded into the IQ modulator to generate the transmitted 256QAM modulation format signal, which is then loaded into the laser with a line width of 1kHz, amplified by the erbium-doped fiber amplifier, and adjusted by the attenuator. The signal quality is then input into the characterizing single-mode optical fiber for transmission. After receiving, the receiver performs analog-to-digital conversion, resamples the signal, performs matched filtering, orthogonalization and normalization, dispersion compensation, clock recovery, and then recovers the carrier phase to obtain the initially processed raw data. At this point, the 256QAM signal (i.e., the first optical signal) that needs to be nonlinearly equalized is obtained.

[0052] Afterwards, the target model (CV-CNN+bi-LSTM network model) can be used to process the first optical signal, such as Figure 4 As shown, Figure 4 The upper part is a complex-valued convolutional neural network, and the lower part is a bidirectional long short-term memory network, which are introduced below.

[0053] First, a complex-valued convolutional neural network is constructed. The complex-valued convolutional neural network can be used to extract features of the first optical signal. The specific steps are as follows.

[0054] In some embodiments of the present application, a complex-valued convolutional neural network is used to perform feature extraction on the first optical signal to obtain a first feature sequence, including: using a real convolution kernel in the complex-valued convolutional neural network to perform a convolution operation on the real data of each complex-valued symbol in the first optical signal to obtain a real sub-feature; using an imaginary convolution kernel in the complex-valued convolutional neural network to perform a convolution operation on the imaginary data of each complex-valued symbol in the first optical signal to obtain an imaginary sub-feature; using a complex activation function to determine the first feature sequence based on the real sub-features and the imaginary sub-features.

[0055] Specifically, Figure 4 The original complex signal sequence A+iB in represents the transmitted 256QAM signal (i.e., the first optical signal). In this embodiment, the signal can be a string of 16*16 level complex values. The signal is input into the complex value convolutional neural network, and the real part and imaginary part of the signal are convolved with the real part convolution kernel and the imaginary part convolution kernel respectively to complete the preliminary feature extraction, as shown in the following formula:

[0056] W*h=(A*xB*y)+i(B*x+A*y)

[0057]

[0058] Wherein, W=A+iB represents the input 256QAM signal (first optical signal), and h=x+iy represents the complex convolution kernel. In this embodiment, the activation function of the complex-valued convolutional neural network can adopt a complex activation function, as shown in the following formula: The above complex-valued convolutional neural network can be built after complex weight initialization and batch normalization.

[0059] In addition, it is also necessary to build a bidirectional long short-term memory network, use the bidirectional long short-term memory network to perform nonlinear equalization, and eliminate the inter-symbol interference in the first optical signal. The specific steps are as follows.

[0060] In some embodiments of the present application, a bidirectional long short-term memory network includes: a first long short-term memory layer and a second long short-term memory layer, wherein the first long short-term memory layer and the second long short-term memory layer each include a plurality of long short-term memory recursive units; using the bidirectional long short-term memory network, determining the target dependency relationship based on the first feature sequence includes: using the long short-term memory recursive units in the first long short-term memory layer to analyze the features corresponding to each complex-valued symbol in the first feature sequence in a direction arranged in chronological order to obtain forward state information; using the long short-term memory recursive units in the second long short-term memory layer to analyze the features corresponding to each complex-valued symbol in the first feature sequence in a direction arranged in reverse chronological order to obtain reverse state information; merging the forward state information and the reverse state information to obtain the target dependency relationship.

[0061] Specifically, LSTM belongs to a recurrent neural network (RNN) with strong memory. Its recurrent core can be used to process time series data. The difference from conventional neural networks is that the output of LSTM at the current moment depends on the output at the current moment and the output of the hidden layer at the previous moment, that is, the neurons in the hidden layer are connected, and the transmission control of time series hidden information is realized by introducing a gating structure. It can selectively record and forget information to solve the gradient disappearance problem of RNN. In the process of coherent optical signal transmission, an information symbol at a moment is affected by multiple symbols before and after, and the dependency relationship between the symbols before and after needs to be considered. Therefore, when building a bidirectional long short-term memory network, this embodiment adds a network layer (i.e., the second long short-term memory layer) arranged in reverse time order on the basis of the network layer (i.e., the first long short-term memory layer) arranged in time order to improve the accuracy of the network. Their inputs are the same, but the information transmission directions are different. One layer is transmitted in time order, and the other layer is transmitted in reverse time order. The two hidden neuron states are processed by vector concatenation operation and connected to the output layer neurons. Multiple paired neurons are the performance states of the same pair of neurons at different times, that is, bi-LSTM is obtained. The bi-LSTM network shows superior performance for ultra-high-order coherent optical transmission systems, and it can more effectively handle inter-symbol interference ISI between previous and next code elements.

[0062] For any long short-term memory recursive unit in the bidirectional long short-term memory network (the first long short-term memory layer or the second long short-term memory layer), its structure and function are as follows.

[0063] In some embodiments of the present application, the step of analyzing the features corresponding to the complex-valued symbols by the long short-term memory recursive unit includes: determining the input gate output value, the forget gate output value, and the output gate output value based on the first input value and the second external state value, wherein the first input value is the feature of the complex-valued symbol corresponding to the first long short-term memory recursive unit in the first feature sequence, the second external state value is the external state value output by the second long short-term memory recursive unit, and the second long short-term memory recursive unit is the previous long short-term memory recursive unit adjacent to the first long short-term memory recursive unit in the first long short-term memory layer or the second long short-term memory layer; updating the second memory unit state value based on the input gate output value and the forget gate output value to obtain the first memory unit state value corresponding to the first long short-term memory recursive unit, wherein the second memory unit state value is the memory unit state value output by the second long short-term memory recursive unit; determining the first external state value corresponding to the current long short-term memory recursive unit based on the output gate output value and the first memory unit state value.

[0064] Specifically, Figure 4 The structure of an LSTM recursive unit in the bi-LSTM structure is as follows Figure 5 As shown, where it is the input gate, f t For the forget gate, o t is the output gate. Under the control of the three gates, the input state x t (i.e. the first input value mentioned above) and the previous state h t-1 (i.e. the second external state value mentioned above) determines the internal state c t (i.e. the state value of the first memory unit mentioned above) and the output h t (i.e. the first external state value mentioned above), the output formulas of the three gates are as follows:

[0065] i t =σ(W i x t +U i h t-1 +b i )

[0066] f t =σ(W f x t +U f h t-1 +b f )

[0067] o t =σ(W o x t +U o h t-1 +b o )

[0068] Among them, σ is the activation function, using the sigmoid function, x t is the current input (i.e. the first input value mentioned above), h t-1 is the previous external state (i.e., the second external state value mentioned above), W i , W f , W o belongs to the input weight, U i , U f , U o belongs to the recursive weight, and b i 、b f 、b o Belongs to the offset. Figure 5 As shown, first use the external state h at the previous moment t-1 (i.e. the second external state value mentioned above) and the current input x t (i.e. the first input value mentioned above), calculate the three gates (i.e. determine the input gate output value, the forget gate output value, and the output gate output value), and the candidate state Combined with forget gate f t and input gate i tTo update the memory cell c t , and finally combined with the output gate o t , passing the information of the internal state to the external state h t Through the LSTM recurrent unit, the entire network can establish long-distance time series dependencies.

[0069] The following is an introduction to the training process of the target model (CV-CNN+bi-LSTM network model) in the embodiment of the present application. After the complex-valued convolutional neural network and the bidirectional long short-term memory network are built, they still need to be trained and tested. The specific steps are as follows.

[0070] In some embodiments of the present application, the target model includes: a complex-valued convolutional neural network and a bidirectional long short-term memory network; the training step of the target model includes: obtaining a training data set, wherein the training data set includes multiple groups of optical signal data under different optical signal-to-noise ratios; during the model training process, the complex-valued convolutional neural network in the initial model is used to extract features of the optical signal data in the training data set to obtain a feature sequence, and the bidirectional long short-term memory network in the initial model is used to perform nonlinear equalization on the optical signal data based on the feature sequence; based on the optical signal data after nonlinear equalization, a loss function value is determined, and an adaptive moment estimation algorithm is used to adjust the model parameters in the initial model based on the loss function value; and the above training process is repeated using the initial model after the model parameters are adjusted until the target model is obtained.

[0071] In some embodiments of the present application, the above-mentioned training process is repeated using the initial model after the model parameters are adjusted until the target model is obtained, including: when the loss function value is less than a preset error threshold, or the number of iterations of the model training process exceeds a preset threshold, the initial model after the model parameters are adjusted is determined as the target model.

[0072] Specifically, the training data set can select multiple groups of restored signals under different optical signal-to-noise ratios. For example, in this embodiment, for an optical signal-to-noise ratio of 24-29dB, 10 groups of signals are selected for each value, and each group of signals includes 10,000 complex-valued symbols, which fully contains the information of the current symbol and the period before and after the symbol. Then it is first input into CV-CNN, CNN selects 8 convolution kernels, and the sliding window length is set to 7 for feature extraction, and then the feature-extracted data is input into LSTM.

[0073] For bi-LSTM, this embodiment sets the upper limit of the number of iterations to 1000 times (preset number threshold), the error function uses the mean square error (MSE), and the original data of the demodulated data is used as the reference set. The early stopping mechanism is set, that is, if the network does not decrease after 250 iterations, the training is stopped to prevent overfitting. In addition, due to the limitation of computing memory, the input layer only accepts a part of the total data each time, which is called a small batch. Too large a batch will cause it to fall into a saddle point, and too small a batch will cause the training to be time-consuming. Based on the above situation, LSTM selects 72 neurons, that is, 144 in both directions. The optimization algorithm can select the ADAM (Adaptive Moment Estimation) algorithm, set the initial learning rate to 0.008, and set the learning rate to the original 0.1 for every 500 iterations.

[0074] Execute the above steps to obtain a trained joint network (target model). For the trained joint network (target model), you can choose to migrate the network (target model) to the transmitter as a channel pre-compensation, or you can keep it at the back end as a nonlinear equalizer. In this embodiment, choose to place it in the back-end test, use the new experimental results as input, and use the trained parameters for equalization to obtain the experimental results.

[0075] Furthermore, in order to verify the effect of the above scheme, in the simulation experiment of 256QAM coherent optical transmission over 80km optical fiber, the signal light signal-to-noise ratio range is adjusted to 21-29dB, and the trained network is applied to the DSP (Digital Signal Processing) process at the receiving end. The simulation results of the CV-CNN+bi-LSTM scheme proposed in the embodiment of the present application and the traditional VNLE (Volterra Series Nonlinear Equalizer) scheme are compared as shown in the figure below. Figure 6As shown, the two curves in the figure, the solid line corresponds to the scheme proposed in this application, and the dotted line results represent the traditional VNLE method. It can be seen that in the range of 21-29dB for the optical signal-to-noise ratio, the result of the CV-CNN+bi-LSTM scheme is obviously better than the traditional VNLE scheme, and the effect is more obvious under high signal-to-noise ratio conditions. Under low signal-to-noise ratio conditions, the quality of high-order signals is poor, the damage is more serious, and the equalization effect is reduced. When the forward error correction code (FEC) overhead is 20%, the CV-CNN+bi-LSTM scheme has a performance advantage of 1.5dB compared to the traditional VNLE scheme. And from the heat map of each constellation diagram, it can be seen that the CV-CNN+bi-LSTM scheme makes the constellation diagram more convergent, and the high-level symbol divergence link further alleviates the linear and nonlinear damage of the system, and plays a better equalization effect.

[0076] This application proposes a solution based on a complex convolutional neural network combined with a bidirectional long short-term memory network to alleviate the nonlinear damage of optical signal transmission. It is mainly used for the digital signal processing algorithm of the optical transmission receiving end of the backbone network. The linear and nonlinear damage compensation of the transmission link is realized through the CV-CNN+bi-LSTM solution. The trained network can be migrated to the transmitting end for channel pre-equalization, or directly replace the traditional VNLE solution, with significant effects, and can ensure the transmission efficiency and stability of the system; based on the complex convolutional neural network algorithm, compared with the real value network, it has excellent performance in processing optical signal sequence problems, and is more conducive to feature extraction; using a bidirectional long short-term memory neural network, preliminary feature extraction is performed through CV-CNN, which can greatly reduce the complexity of the system and better realize signal equalization compared to the traditional VNLE solution. The network structure is simple, and the excellent algorithm performance is more conducive to the development of optical modules, which provides a good supplementary role in coping with future large-capacity optical transmission.

[0077] According to an embodiment of the present application, an embodiment of an optical signal processing device is also provided. Figure 7 is a schematic diagram of the structure of an optical signal processing device provided according to an embodiment of the present application. Figure 7 As shown, the device comprises:

[0078] The data acquisition module 70 is used to acquire a first optical signal, wherein the first optical signal is optical signal data in a complex form, and the optical signal data includes a plurality of complex value symbols arranged in a time sequence;

[0079] A feature extraction module 72, configured to use a complex-valued convolutional neural network to perform feature extraction on the first optical signal to obtain a first feature sequence, wherein the complex-valued convolutional neural network is configured to remove noise from the first optical signal by performing feature extraction;

[0080] The signal equalization module 74 is used to use a bidirectional long short-term memory network to determine the target dependency according to the first feature sequence, and eliminate the inter-symbol interference in the first optical signal according to the target dependency to obtain the second optical signal, wherein the target dependency is used to characterize the dependency between each complex-valued symbol in the first optical signal and other complex-valued symbols immediately before and after, and the second optical signal is an optical signal after nonlinear equalization.

[0081] Optionally, a complex-valued convolutional neural network is used to perform feature extraction on the first optical signal to obtain a first feature sequence, including: using a real convolution kernel in the complex-valued convolutional neural network to perform a convolution operation on the real data of each complex-valued symbol in the first optical signal to obtain a real sub-feature; using an imaginary convolution kernel in the complex-valued convolutional neural network to perform a convolution operation on the imaginary data of each complex-valued symbol in the first optical signal to obtain an imaginary sub-feature; using a complex activation function to determine the first feature sequence based on the real sub-features and the imaginary sub-features.

[0082] Optionally, the bidirectional long short-term memory network includes: a first long short-term memory layer and a second long short-term memory layer, wherein the first long short-term memory layer and the second long short-term memory layer each include a plurality of long short-term memory recursive units; using the bidirectional long short-term memory network, based on the first feature sequence, determining the target dependency relationship includes: using the long short-term memory recursive units in the first long short-term memory layer to analyze the features corresponding to each complex-valued symbol in the first feature sequence in a direction of time order to obtain forward state information; using the long short-term memory recursive units in the second long short-term memory layer to analyze the features corresponding to each complex-valued symbol in the first feature sequence in a direction of time reverse order to obtain reverse state information; merging the forward state information and the reverse state information to obtain the target dependency relationship.

[0083] Optionally, the step of analyzing the features corresponding to the complex-valued symbols by the long short-term memory recursive unit includes: determining an input gate output value, a forget gate output value, and an output gate output value based on a first input value and a second external state value, wherein the first input value is a feature of the complex-valued symbol corresponding to the first long short-term memory recursive unit in the first feature sequence, the second external state value is an external state value output by the second long short-term memory recursive unit, and the second long short-term memory recursive unit is a previous long short-term memory recursive unit adjacent to the first long short-term memory recursive unit in the first long short-term memory layer or the second long short-term memory layer; updating the second memory unit state value based on the input gate output value and the forget gate output value to obtain a first memory unit state value corresponding to the first long short-term memory recursive unit, wherein the second memory unit state value is a memory unit state value output by the second long short-term memory recursive unit; determining the first external state value corresponding to the current long short-term memory recursive unit based on the output gate output value and the first memory unit state value.

[0084] Optionally, the target model includes: a complex-valued convolutional neural network and a bidirectional long short-term memory network; the training step of the target model includes: obtaining a training data set, wherein the training data set includes multiple groups of optical signal data under different optical signal-to-noise ratios; during the model training process, the complex-valued convolutional neural network in the initial model is used to extract features of the optical signal data in the training data set to obtain a feature sequence, and the bidirectional long short-term memory network in the initial model is used to perform nonlinear equalization on the optical signal data based on the feature sequence; based on the optical signal data after nonlinear equalization, a loss function value is determined, and an adaptive moment estimation algorithm is used to adjust the model parameters in the initial model based on the loss function value; and the initial model after the model parameters are adjusted is used to repeat the above training process until the target model is obtained.

[0085] Optionally, the above training process is repeated using the initial model after the model parameters are adjusted until the target model is obtained, including: when the loss function value is less than a preset error threshold, or the number of iterations of the model training process exceeds a preset threshold, the initial model after the model parameters are adjusted is determined as the target model.

[0086] Optionally, the first optical signal is optical signal data in a complex form transmitted in a coherent optical communication system, wherein the coherent optical communication system includes: a transmitting end and a receiving end, wherein the transmitting end is used to map a pseudo-random binary sequence into signals with different amplitude levels through orthogonal amplitude modulation encoding, and upsample and filter the signal; import the signal after upsampling and filtering into a waveform generator to obtain an analog level signal; load the analog level signal onto an orthogonal amplitude IQ modulator to obtain a modulation format signal; load the modulation format signal onto a narrow linewidth laser to obtain an optical signal, and send the optical signal to the receiving end through a transmission link; the receiving end is used to perform analog-to-digital conversion on the received optical signal, and perform digital signal processing on the optical signal after the analog-to-digital conversion to obtain a first optical signal, wherein the digital signal processing includes: resampling, matched filtering, IQ orthogonalization, dispersion compensation, and clock recovery.

[0087] It should be noted that the various modules in the above-mentioned optical signal processing device can be program modules (for example, a set of program instructions that implement a certain specific function) or hardware modules. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.

[0088] It should be noted that the optical signal processing device provided in this embodiment can be used to perform Figure 2 The optical signal processing method shown, therefore, the relevant explanations and descriptions of the above optical signal processing method are also applicable to the embodiments of the present application and will not be repeated here.

[0089] An embodiment of the present application also provides a non-volatile storage medium, the non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the following optical signal processing method by running the computer program: obtaining a first optical signal, wherein the first optical signal is optical signal data in complex form, and the optical signal data includes multiple complex-valued symbols arranged in a time series order; using a complex-valued convolutional neural network to extract features from the first optical signal to obtain a first feature sequence, wherein the complex-valued convolutional neural network is used to remove noise in the first optical signal by performing feature extraction; using a bidirectional long short-term memory network to determine a target dependency relationship based on the first feature sequence, and based on the target dependency relationship, eliminate inter-symbol interference in the first optical signal to obtain a second optical signal, wherein the target dependency relationship is used to characterize the dependency relationship between each complex-valued symbol in the first optical signal and other complex-valued symbols immediately before and after, and the second optical signal is an optical signal after nonlinear equalization.

[0090] The embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the optical signal processing method described in each embodiment of the present application: obtaining a first optical signal, wherein the first optical signal is optical signal data in complex form, and the optical signal data includes multiple complex-valued symbols arranged in a time series order; using a complex-valued convolutional neural network to extract features from the first optical signal to obtain a first feature sequence, wherein the complex-valued convolutional neural network is used to remove noise in the first optical signal by performing feature extraction; using a bidirectional long short-term memory network to determine a target dependency relationship based on the first feature sequence, and based on the target dependency relationship, eliminate inter-symbol interference in the first optical signal to obtain a second optical signal, wherein the target dependency relationship is used to characterize the dependency relationship between each complex-valued symbol in the first optical signal and other complex-valued symbols immediately before and after, and the second optical signal is an optical signal after nonlinear equalization.

[0091] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0092] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0093] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0094] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0095] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0096] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0097] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for processing an optical signal, characterized in that: include: Acquire a first optical signal, wherein the first optical signal is optical signal data in complex form, and the optical signal data includes a plurality of complex-valued symbols arranged in a time sequence; Using a complex-valued convolutional neural network to perform feature extraction on the first optical signal to obtain a first feature sequence, wherein the complex-valued convolutional neural network is used to remove noise in the first optical signal by performing feature extraction; A bidirectional long short-term memory network is used to determine a target dependency relationship based on the first feature sequence, and based on the target dependency relationship, the inter-symbol interference in the first optical signal is eliminated to obtain a second optical signal, wherein the target dependency relationship is used to characterize the dependency relationship between each of the complex-valued symbols in the first optical signal and other complex-valued symbols immediately before and after, and the second optical signal is an optical signal after nonlinear equalization.

2. The optical signal processing method according to claim 1, characterized in that: The first optical signal is subjected to feature extraction using a complex-valued convolutional neural network, and the first feature sequence obtained includes: Using the real convolution kernel in the complex-valued convolutional neural network, a convolution operation is performed on the real part data of each of the complex-valued symbols in the first optical signal to obtain a real part sub-feature; Using the imaginary part convolution kernel in the complex-valued convolutional neural network, a convolution operation is performed on the imaginary part data of each of the complex-valued symbols in the first optical signal to obtain an imaginary part sub-feature; A complex activation function is used to determine the first feature sequence according to the real sub-feature and the imaginary sub-feature.

3. The optical signal processing method according to claim 1, characterized in that: The bidirectional long short-term memory network comprises: a first long short-term memory layer and a second long short-term memory layer, wherein the first long short-term memory layer and the second long short-term memory layer both comprise a plurality of long short-term memory recursive units; using the bidirectional long short-term memory network, according to the first feature sequence, determining the target dependency relationship comprises: Using the long short-term memory recursive unit in the first long short-term memory layer, the features corresponding to each of the complex-valued symbols in the first feature sequence are analyzed in a time-sequential direction to obtain positive state information; Using the long short-term memory recursive unit in the second long short-term memory layer, the features corresponding to each of the complex-valued symbols in the first feature sequence are analyzed in a direction of reverse time order to obtain reverse state information; The forward state information and the reverse state information are merged to obtain the target dependency relationship.

4. The optical signal processing method according to claim 3, characterized in that: in, The step of analyzing the features corresponding to the complex-valued symbols by the long short-term memory recursive unit comprises: Determine an input gate output value, a forget gate output value, and an output gate output value according to a first input value and a second external state value, wherein the first input value is a feature of the complex-valued symbol corresponding to the first long short-term memory recursive unit in the first feature sequence, the second external state value is an external state value output by the second long short-term memory recursive unit, and the second long short-term memory recursive unit is a previous long short-term memory recursive unit immediately adjacent to the first long short-term memory recursive unit in the first long short-term memory layer or the second long short-term memory layer; According to the input gate output value and the forget gate output value, the second memory unit state value is updated to obtain the first memory unit state value corresponding to the first long short-term memory recursive unit, wherein the second memory unit state value is the memory unit state value output by the second long short-term memory recursive unit; A first external state value corresponding to the current long short-term memory recursive unit is determined according to the output value of the output gate and the state value of the first memory unit.

5. The optical signal processing method according to claim 1, characterized in that: The target model includes: the complex-valued convolutional neural network and the bidirectional long short-term memory network; the training steps of the target model include: Acquire a training data set, wherein the training data set includes multiple groups of optical signal data under different optical signal-to-noise ratios; In the model training process, the complex-valued convolutional neural network in the initial model is used to extract features of the optical signal data in the training data set to obtain a feature sequence, and the bidirectional long short-term memory network in the initial model is used to perform nonlinear equalization on the optical signal data according to the feature sequence; Determine a loss function value according to the optical signal data after the nonlinear equalization, and use an adaptive moment estimation algorithm to adjust the model parameters in the initial model according to the loss function value; The above training process is repeated using the initial model after the model parameters are adjusted until the target model is obtained.

6. The optical signal processing method according to claim 5, characterized in that: Repeating the above training process using the initial model after adjusting the model parameters until the target model is obtained includes: When the loss function value is less than a preset error threshold, or the number of iterations of the model training process exceeds a preset threshold, the initial model after the model parameters are adjusted is determined as the target model.

7. The optical signal processing method according to claim 1, characterized in that: The first optical signal is a complex optical signal data transmitted in a coherent optical communication system, wherein the coherent optical communication system includes: a transmitting end and a receiving end, wherein: The transmitting end is used to map the pseudo-random binary sequence into signals with different amplitude levels through orthogonal amplitude modulation coding, and perform up-sampling and filtering processing on the signals; import the signals after the up-sampling and filtering processing into a waveform generator to obtain an analog level signal; load the analog level signal onto an orthogonal amplitude IQ modulator to obtain a modulation format signal; load the modulation format signal onto a narrow linewidth laser to obtain an optical signal, and send the optical signal to the receiving end through a transmission link; The receiving end is used to perform analog-to-digital conversion on the received optical signal, and perform digital signal processing on the optical signal after analog-to-digital conversion to obtain the first optical signal, wherein the digital signal processing includes: resampling, matched filtering, IQ orthogonalization, dispersion compensation, and clock recovery.

8. An optical signal processing device, characterized in that: include: A data acquisition module, configured to acquire a first optical signal, wherein the first optical signal is optical signal data in a complex form, and the optical signal data includes a plurality of complex value symbols arranged in a time sequence; a feature extraction module, configured to perform feature extraction on the first optical signal using a complex-valued convolutional neural network to obtain a first feature sequence, wherein the complex-valued convolutional neural network is configured to remove noise from the first optical signal by performing feature extraction; A signal equalization module is used to use a bidirectional long short-term memory network to determine a target dependency relationship based on the first feature sequence, and eliminate inter-symbol interference in the first optical signal based on the target dependency relationship to obtain a second optical signal, wherein the target dependency relationship is used to characterize the dependency relationship between each of the complex-valued symbols in the first optical signal and other complex-valued symbols immediately before and after, and the second optical signal is an optical signal after nonlinear equalization.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is used to run a program stored in the memory, wherein the optical signal processing method according to any one of claims 1 to 7 is executed when the program is run.

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the optical signal processing method according to any one of claims 1 to 7 by running the computer program.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the optical signal processing method according to any one of claims 1 to 7 are implemented.

Images