Convolutional neural network-based modulation format identification method and system

Abstract

The invention discloses a modulation format recognition method and system based on a convolutional neural network, and relates to the technical field of cognitive optical networks, and the method comprises the following steps: obtaining to-be-transmitted data, inputting the to-be-transmitted data into a pre-constructed IQ modulator, coupling composite light into the to-be-transmitted data, obtaining coupled data, and transmitting the coupled data to a cognitive optical network; inputting the coupling data into a pre-constructed coherent receiver for demodulation, and obtaining an optical signal through analog-to-digital conversion; the modulation format recognition method comprises the steps of preprocessing an optical signal to obtain a processed optical signal, normalizing the processed optical signal, inputting the normalized optical signal into a pre-established light-weight convolutional neural network model based on meta-learning assistance, and outputting to obtain a modulation format recognition result.

Description

Technical Field

[0001] The present invention relates to the field of cognitive optical network technology, and in particular to a modulation format recognition method and system based on a convolutional neural network. Background Art

[0002] In recent years, emerging data services such as social applications, the Internet of Things, cloud computing, and virtual reality have emerged one after another, and the number of Internet users has continued to grow, driving the optical network to flourish in the direction of ultra-large capacity. In order to meet the needs of business transmission, the optical network has evolved and upgraded accordingly, and its structure has become more and more diversified. Network operators are facing the huge challenge of supporting a large number of services. However, simply increasing the transmission capacity is not enough. Due to the lack of flexibility in the allocation of optical network resources, the fixed bandwidth of each wavelength cannot be fully utilized. According to the different service quality requirements of each business, operators also need to provide flexible transmission technologies (such as modulation format, coding or data rate) for their optical transmission networks.

[0003] In order to break the limitation that optical networks cannot adapt to dynamic environmental changes, cognitive optical networks are proposed. Cognitive optical networks (CON) are a new generation of intelligent optical networks based on machine learning. They can sense the external environment and adjust the network configuration in real time through understanding and learning the external environment. They can intelligently adapt to changes in the external environment and play a very important role in the development of network technology. The cognitive function of cognitive optical networks can further guide directional planning, and can be actually designed according to the current network function environment to ensure more reasonable network function application. After the introduction of cognitive optical networks, not only can the optical network configuration be automatically optimized, but also the optical signal-to-noise ratio can be improved, the optical path quality can be predicted, the optical layer recovery time can be reduced, the self-optimization switching of different modulation modes can be realized, and the overall performance of the all-optical network can be improved. However, with the development of space division multiplexing technology, more new high-order complex format solutions are adopted in optical fiber transmission systems. The more complex the modulation format, the more devices, subsystems and algorithms are required. The modulation format recognition technology of the signal at the receiving end of multi-core optical fiber is crucial for signal demodulation and recovery in the receiver. Therefore, while realizing the self-optimization switching of different modulation modes in the cognitive optical network system, how to accurately identify the modulation format at the receiving end is one of the key issues.

[0004] The deep learning model training process has very strict requirements on computing hardware, especially when manually adjusting hyperparameters, a lot of calculations are required. Its mechanism is to use statistical models learned on pre-prepared training samples to make predictions about future data. In most cases, the generalization ability of the model is guaranteed by a sufficient number of training samples. However, in the modulation format recognition task, a large amount of data (such as constellation diagrams, eye diagrams, etc.) is required to train the learning model. After switching to another type of modulation format recognition task, the model needs to be retrained. Such a model has a very large amount of calculation and is very time-consuming. Because the number and type of signal samples obtained by the transmission system are sometimes limited, the optical network system model based on deep learning is limited in practical applications, and it is impossible to achieve a balance between high accuracy and high efficiency in modulation format recognition. Summary of the invention

[0005] In order to solve the deficiencies mentioned in the above background technology, the purpose of the present invention is to provide a modulation format recognition method and system based on convolutional neural network.

[0006] In a first aspect, the purpose of the present invention can be achieved by the following technical solution: a modulation format recognition method based on a convolutional neural network, the method comprising the following steps:

[0007] Acquire the data to be transmitted, input the data to be transmitted into a pre-built IQ modulator, couple the composite light into the data to be transmitted to obtain coupled data, input the coupled data into a pre-built coherent receiver for demodulation, and obtain an optical signal through analog-to-digital conversion;

[0008] The optical signal is preprocessed to obtain a processed optical signal, the processed optical signal is normalized, and the normalized optical signal is input into a pre-established lightweight convolutional neural network model assisted by meta-learning, and the modulation format recognition result is output.

[0009] In combination with the first aspect, in some implementations of the first aspect, the method further includes: a process of acquiring the data to be transmitted:

[0010] First, it is necessary to collect radio signals or simulated signal data. It can be collected through a real wireless channel environment, or use simulation software (such as MATLAB, GNU Radio, etc.) to generate signals of different modulation formats. Then perform data preprocessing. The collected signals are generally time domain or frequency domain data, which need to undergo certain preprocessing. The preprocessing process can include operations such as denoising, normalization, and downsampling. For CNN, one-dimensional signal data is usually converted into a two-dimensional form, such as a time-amplitude matrix or a time-phase matrix, to adapt to the input format of CNN. After that, data annotation is performed. In the modulation format recognition task, each sample needs to be labeled with the corresponding modulation type so that CNN can perform supervised learning. The data is then divided into a training set, a validation set, and a test set.

[0011] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: inputting the data to be transmitted into a pre-built IQ modulator, loading it onto the I and Q paths of the two polarized lights through the IQ modulator, and using a combiner to couple the multiplexed light into the few-mode optical fiber for data transmission.

[0012] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: coupling the composite light into the data to be transmitted, coupling the multiplexed light into the few-mode optical fiber through a combiner, sending the signal light and the local oscillator light into a coherent receiver for demodulation, the coherent receiver completes polarization demultiplexing and converts the analog signal into a digital signal through a digital-to-analog converter ADC and enters the DSP to obtain an optical signal.

[0013] In combination with the first aspect, in some implementations of the first aspect, the method further includes: pre-processing the optical signal to perform dispersion compensation of the optical fiber channel, clock recovery of the signal, and CMA equalization.

[0014] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the pre-established lightweight convolutional neural network model based on meta-learning assistance includes:

[0015] Meta-learning module: This is the core module used to learn how to adapt to new tasks efficiently. The meta-learning module optimizes the parameter initialization of the model so that the model can quickly adjust to new tasks with small samples or a small number of iterations. Common meta-learning methods include model-invariant meta-learning (MAML), relational networks, and gradient-based meta-learning methods.

[0016] Lightweight convolutional neural network: This part of the network is the basic model for modulation recognition. By introducing lightweight technology, the computational complexity and model size of the network are reduced. Lightweight CNN models usually use the following technologies:

[0017] Depthwise separable convolution: decomposes the standard convolution into depthwise convolution and pointwise convolution to reduce the amount of computation (.

[0018] Pruning and quantization: Prune unimportant parameters and quantize floating-point numbers (e.g., 8-bit quantization) to reduce model size and speed up inference.

[0019] Knowledge distillation: By transferring the knowledge of the large model (teacher model) to the small model (student model), the complexity of the model can be reduced while maintaining accuracy.

[0020] Feature extraction module: This module is used to extract features from the input signal. It can be a stack of convolutional layers, or a combination of convolutional blocks and pooling layers to extract multi-scale features. With the help of meta-learning, the parameters of this feature extraction module will be optimized to make it more adaptable to new modulation modes.

[0021] Task Adaptation Module: This module is used to quickly adjust model parameters in each new task. Through meta-learning algorithms (such as MAML), the network can be quickly fine-tuned, allowing lightweight CNNs to perform well in different modulation modes.

[0022] Modulation format classifier: This is the last layer or output layer of the model, which is used to classify different modulation formats. With the support of meta-learning, the lightweight CNN will quickly adjust the classifier according to the new task data to achieve efficient and accurate recognition.

[0023] Online adaptation mechanism (optional): In order to cope with the dynamic changes of modulation formats, an online adaptation module can be added to enable the model to continuously update parameters in a real-time environment to adapt to new data distributions. Online adaptation combined with meta-learning enables lightweight CNN to adapt to changing radio environments, enhancing its real-time and robustness.

[0024] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the pre-established lightweight convolutional neural network model based on meta-learning assistance is expressed as follows:

[0025] Suppose there is a lightweight CNN model f θ , whose parameters are θ. The goal of meta-learning is to find an initial parameter θ so that the model can adapt quickly to new tasks after a small number of updates. Each modulation format recognition task is defined as T i , each task T i Contains a support set and a queryset

[0026] Meta-objective: The goal of meta-learning is to minimize the loss on the new task. Assuming the loss function is Υ, the meta-objective can be defined as:

[0027]

[0028] Among them, θ′ is the i The updated parameters.

[0029] Gradient-based update: In each task T i In , we first use the support set data to fine-tune the model parameters θ. The update method can be expressed as:

[0030]

[0031] Among them, α is the learning rate, θ i ' represents the updated task-specific parameters.

[0032] Meta-update: Based on multiple tasks, the initial parameters θ are updated:

[0033]

[0034] Among them, β is the learning rate of meta-learning.

[0035] In a second aspect, in order to achieve the above-mentioned object, the present invention discloses a modulation format recognition system based on a convolutional neural network, comprising:

[0036] A signal processing module, used for acquiring data to be transmitted, inputting the data to be transmitted into a pre-built IQ modulator, coupling the composite light into the data to be transmitted to obtain coupled data, inputting the coupled data into a pre-built coherent receiver for demodulation, and obtaining an optical signal through analog-to-digital conversion;

[0037] The format recognition module is used to preprocess the optical signal to obtain a processed optical signal, normalize the processed optical signal, input the normalized optical signal into a pre-established lightweight convolutional neural network model assisted by meta-learning, and output a modulation format recognition result.

[0038] In another aspect of the present invention, in order to achieve the above-mentioned purpose, a terminal device is disclosed, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the memory stores a computer program capable of running on the processor, and when the processor loads and executes the computer program, a modulation format recognition method based on a convolutional neural network as described above is adopted.

[0039] In another aspect of the present invention, in order to achieve the above-mentioned purpose, a computer-readable storage medium is disclosed, in which a computer program is stored. When the computer program is loaded and executed by a processor, a modulation format recognition method based on a convolutional neural network as described above is adopted.

[0040] Beneficial effects of the present invention:

[0041] The present invention adopts a multi-task training strategy based on meta-learning, and obtains a cross-task signal recognition capability through a large number of different task training networks, so that the network can quickly adapt to new signal categories with only a small number of samples. At the same time, the training network model is improved, and a more efficient network calculation method is implemented using a lightweight convolutional neural network SqueezeNet. On the basis of not sacrificing accuracy, high-efficiency and low-complexity recognition tasks are completed. The learning method of meta-learning is used to train the neural network model to "learn how to learn", thereby realizing the modulation format recognition of a small number of samples by identifying different constellation diagrams, reducing the pressure of obtaining training samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 It is a schematic flow chart of the method of the present invention;

[0044] Figure 2 It is a schematic diagram of the meta-learning principle of the present invention;

[0045] Figure 3 It is a schematic diagram of the compression and expansion structure of the present invention;

[0046] Figure 4 It is a schematic diagram of the overall structure of SqueezeNet of the present invention;

[0047] Figure 5 It is a schematic diagram of an experimental model of the modulation format identification method of the present invention;

[0048] Figure 6 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] Embodiment 1:

[0051] The following is an introduction to the relevant terms involved in the embodiments of the present application:

[0052] Convolutional Neural Networks (CNN) are a type of feedforward neural networks that include convolutional calculations and have a deep structure. They are one of the representative algorithms of deep learning. Convolutional neural networks have the ability of representation learning and can perform shift-invariant classification on input information according to their hierarchical structure. Therefore, they are also called "Shift-Invariant Artificial Neural Networks (SIANN)". The study of convolutional neural networks began in the 1980s and 1990s. Time delay networks and LeNet-5 were the earliest convolutional neural networks. After the 21st century, with the introduction of deep learning theory and the improvement of numerical computing equipment, convolutional neural networks have developed rapidly and have been applied to computer vision, natural language processing and other fields. Convolutional neural networks are constructed by imitating the visual perception mechanism of organisms and can perform supervised and unsupervised learning. The sharing of convolution kernel parameters within the hidden layer and the sparsity of inter-layer connections enable convolutional neural networks to learn grid-like topology features such as pixels and audio with a small amount of computation, with stable results and no additional feature engineering requirements for the data.

[0053] Meta-learning as applied to robotic control is a method that enables robots to quickly adapt to new tasks or environments, even with only a small number of examples or experience, by leveraging prior knowledge learned from other tasks. This is critical for robotics, as real-world robots are often required to handle a wide variety of tasks and environments, often with limited training data. Meta-learning addresses this challenge by training models on multiple tasks, allowing them to generalize and quickly adapt to new tasks with minimal retraining.

[0054] like Figure 1 As shown, a modulation format recognition method based on a convolutional neural network comprises the following steps:

[0055] S101: Acquire data to be transmitted, input the data to be transmitted into a pre-built IQ modulator, couple the composite light into the data to be transmitted to obtain coupled data, input the coupled data into a pre-built coherent receiver for demodulation, and obtain an optical signal through analog-to-digital conversion;

[0056] The process of obtaining the data to be transmitted:

[0057] First, it is necessary to collect radio signals or simulated signal data. It can be collected through a real wireless channel environment, or use simulation software (such as MATLAB, GNU Radio, etc.) to generate signals of different modulation formats. Then perform data preprocessing. The collected signals are generally time domain or frequency domain data, which need to undergo certain preprocessing. The preprocessing process can include operations such as denoising, normalization, and downsampling. For CNN, one-dimensional signal data is usually converted into a two-dimensional form, such as a time-amplitude matrix or a time-phase matrix, to adapt to the input format of CNN. After that, data annotation is performed. In the modulation format recognition task, each sample needs to be labeled with the corresponding modulation type so that CNN can perform supervised learning. Then divide the data into training set, validation set, and test set. The training set is used for model training, the validation set is used for parameter adjustment and to prevent overfitting, and the test set is used for performance evaluation of the final model. When the amount of data is small, data augmentation technology can be used to expand the data set to improve the generalization ability of the model. Although CNN can automatically extract features, in some cases, designers can also perform some feature engineering before input, such as calculating instantaneous amplitude, phase and other features as input to CNN.

[0058] The data to be transmitted is input into a pre-built IQ modulator, loaded onto the I path and Q path of two polarized lights through the IQ modulator, and the multiplexed light is coupled into the few-mode optical fiber by a combiner for data transmission.

[0059] The composite light is coupled into the data to be transmitted, and the multiplexed light is coupled into the few-mode optical fiber through the combiner. The signal light and the local oscillator light are sent to the coherent receiver for demodulation. The coherent receiver completes polarization demultiplexing and converts the analog signal into a digital signal through the digital-to-analog converter ADC and enters the DSP to obtain an optical signal.

[0060] S102: Preprocess the optical signal to obtain a processed optical signal, normalize the processed optical signal, input the normalized optical signal into a pre-established lightweight convolutional neural network model assisted by meta-learning, and output a modulation format recognition result.

[0061] The optical signal is pre-processed to compensate for the dispersion of the optical fiber channel, recover the clock of the signal, and perform CMA equalization.

[0062] The pre-established meta-learning-assisted lightweight convolutional neural network model includes:

[0063] Meta-learning module: This is the core module used to learn how to adapt to new tasks efficiently. The meta-learning module optimizes the parameter initialization of the model so that the model can quickly adjust to new tasks with small samples or a small number of iterations. Common meta-learning methods include model-invariant meta-learning (MAML), relational networks, and gradient-based meta-learning methods.

[0064] Lightweight convolutional neural network: This part of the network is the basic model for modulation recognition. By introducing lightweight technology, the computational complexity and model size of the network are reduced. Lightweight CNN models usually use the following technologies:

[0065] Depthwise separable convolution: decomposes the standard convolution into depthwise convolution and pointwise convolution to reduce the amount of computation (.

[0066] Pruning and quantization: Prune unimportant parameters and quantize floating-point numbers (e.g., 8-bit quantization) to reduce model size and speed up inference.

[0067] Knowledge distillation: By transferring the knowledge of the large model (teacher model) to the small model (student model), the complexity of the model can be reduced while maintaining accuracy.

[0068] Feature extraction module: This module is used to extract features from the input signal. It can be a stack of convolutional layers, or a combination of convolutional blocks and pooling layers to extract multi-scale features. With the help of meta-learning, the parameters of this feature extraction module will be optimized to make it more adaptable to new modulation modes.

[0069] Task Adaptation Module: This module is used to quickly adjust model parameters in each new task. Through meta-learning algorithms (such as MAML), the network can be quickly fine-tuned, allowing lightweight CNNs to perform well in different modulation modes.

[0070] Modulation format classifier: This is the last layer or output layer of the model, which is used to classify different modulation formats. With the support of meta-learning, the lightweight CNN will quickly adjust the classifier according to the new task data to achieve efficient and accurate recognition.

[0071] Online adaptation mechanism (optional): In order to cope with the dynamic changes of modulation formats, an online adaptation module can be added to enable the model to continuously update parameters in a real-time environment to adapt to new data distributions. Online adaptation combined with meta-learning enables lightweight CNN to adapt to changing radio environments, enhancing its real-time and robustness.

[0072] The pre-established meta-learning-assisted lightweight convolutional neural network model is expressed as follows:

[0073] Suppose there is a lightweight CNN model f θ , whose parameters are θ. The goal of meta-learning is to find an initial parameter θ so that the model can adapt quickly to new tasks after a small number of updates. Each modulation format recognition task is defined as T i , each task T i Contains a support set and a queryset

[0074] Meta-objective: The goal of meta-learning is to minimize the loss on the new task. Assuming the loss function is Υ, the meta-objective can be defined as:

[0075]

[0076] Among them, θ′ is the i The updated parameters.

[0077] Gradient-based update: In each task T i In , we first use the support set data to fine-tune the model parameters θ. The update method can be expressed as:

[0078]

[0079] Among them, α is the learning rate, θ i ' represents the updated task-specific parameters.

[0080] Meta-update: Based on multiple tasks, the initial parameters θ are updated:

[0081]

[0082] Specifically, the present invention is further described below by way of embodiments:

[0083] Meta-learning model architecture

[0084] The goal of meta-learning is to train a model on a large number of different tasks so that it can quickly adapt to new tasks and solve new problems using a very small amount of training data (i.e., small samples) and a very small number of gradient descent steps. Different from ordinary machine learning methods, meta-learning does not just divide the data into training sets and test sets, but divides the data into meta-training sets and meta-test sets. For any sample in the meta-training set and meta-test set, it is no longer a piece of data in ordinary machine learning, but one of the tasks in a few-sample learning task. The goal of meta-learning is to complete the accurate classification required by the task when a new task (test task) appears, and there is no need to re-learn using a large number of samples. The classification task in each meta-training set also contains the training set and test set in the traditional machine learning method, which are used for each neural network training and parameter adjustment. Meta-learning hopes to train a set of parameters so that a small amount of updates on a new task can make the model converge, so that the meta-learning model can "learn how to learn".

[0085] Assume that we want to perform an N-ways / K-shot (the data contains N categories / each category contains K images) modulation constellation classification task. Meta-learning requires us to build many N-ways / K-shot tasks as its task source. We train meta-learning through many N-ways / K-shot training tasks so that the model learns "prior knowledge" (initialized parameters). This "prior knowledge" can perform better on new N-ways / K-shot tasks, accurately identify constellation categories, and make correct modulation format judgments. The specific principle diagram is as follows: Figure 2 shown.

[0086] Lightweight convolutional neural network SqueezeNet feature extraction module

[0087] The classic network model AlexNet proposed in 2012 and a series of improved models thereafter all aim to improve recognition accuracy. The main means adopted include deepening the network structure and enhancing the function of the convolution module. At the same time, this also makes the entire network more and more complex, and the required memory and computing power are greatly increased. The introduction of the lightweight convolutional neural network SqueezeNet has opened up another direction: to maximize the computing speed while ensuring that the model accuracy is not reduced. It can achieve an effect similar to AlexNet on the ImageNet dataset, but the number of parameters is 50 times less than that of AlexNet. In the meta-learning training structure of the previous chapter, the present invention uses the SqueezeNet network as the neural network training model. When designing the SqueezeNet network, the following strategies are mainly followed:

[0088] Strategy (1): Replace 3×3 filters with 1×1 filters. Since 1×1 filters have 9 times fewer parameters than 3×3 filters, we will set most filters to 1×1.

[0089] Strategy (2): Reduce the number of input channels of the 3×3 filter. When a convolutional layer composed entirely of 3×3 filters is used, the total number of parameters of this layer is (number of input channels)*(number of filters)*(3*3). In order to ensure that the network has a small number of parameters, not only the number of 3×3 filters should be reduced, but also the number of input channels of the 3×3 filters should be reduced.

[0090] Strategy (3): Postpone downsampling, that is, postpone the pooling layer to expand the capture field of view and obtain as many features as possible. In a convolutional network, each convolutional layer generates an output activation map with a spatial resolution of at least 1×1 and usually much larger than 1×1. The height and width of these activation maps are controlled by the following factors: (1) the size of the input data and (2) the choice of layers in the CNN architecture that perform downsampling. The most common approach is to design downsampling into the CNN architecture by setting stride>1 in some convolutional or pooling layers. If the early layers of the network have a large stride, then most layers will have small activation maps. Conversely, if most layers in the network have a stride of 1 and strides greater than 1 are concentrated at the end of the network, then many layers in the network will have large activation maps. Under the same conditions, large activation maps (due to delayed downsampling) can lead to higher classification accuracy.

[0091] Strategies (1) and (2) are about judiciously reducing the number of parameters in a CNN while trying to maintain accuracy. Strategy (3) is about maximizing accuracy under a limited parameter budget. Next, we describe the firemodule, which is the building block of our CNN architecture that enables us to successfully use strategies (1), (2), and (3). The firemodule consists of two parts: a squeeze layer and an expand layer. Figure 3 shown.

[0092] The design idea of ​​the SqueezeNet network structure is mainly to use stacked convolution operations, but the stacked structure here is the compression expansion structure proposed by this network. The overall structure of SqueezeNet is as follows Figure 4 As shown in the figure. Because a large number of 1×1 convolutions are used, the number of input channels is reduced, and the convolution kernel parameters and computational complexity are also reduced. In addition, the rectified linear unit (ReLU) nonlinear activation function is used. Compared with the traditional sigmoid and tanh functions, ReLU can effectively solve the gradient vanishing problem while maintaining the calculation speed, making the training more efficient.

[0093] The system model described in the present invention is as follows Figure 5 As shown, at the transmitting end of the coherent fiber-optic communication network, the communication system based on the cognitive optical network can sense the external environment and adjust the network configuration in real time by understanding and learning the external environment. It can also dynamically change the modulation format, transmission rate and other parameters of the transmission signal according to user needs and channel conditions, and reasonably configure system resources to maximize efficiency.

[0094] Embodiment 2: In the second aspect, as Figure 6 As shown, in order to achieve the above-mentioned purpose, the present invention discloses a modulation format recognition system based on a convolutional neural network, comprising:

[0095] The signal processing module 11 is used to obtain the data to be transmitted, input the data to be transmitted into a pre-built IQ modulator, couple the composite light into the data to be transmitted to obtain coupled data, input the coupled data into a pre-built coherent receiver for demodulation, and obtain an optical signal through analog-to-digital conversion;

[0096] The format recognition module 12 is used to preprocess the optical signal to obtain a processed optical signal, normalize the processed optical signal, input the normalized optical signal into a pre-established lightweight convolutional neural network model assisted by meta-learning, and output a modulation format recognition result.

[0097] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0098] It needs to be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program is executed by a processor to execute the above method. The storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0099] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0100] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure may have various changes and improvements, and these changes and improvements fall within the scope of the present disclosure to be protected.

Claims

1. A modulation format recognition method based on convolutional neural network, characterized in that: The method comprises the following steps: Acquire the data to be transmitted, input the data to be transmitted into a pre-built IQ modulator, couple the composite light into the data to be transmitted to obtain coupled data, input the coupled data into a pre-built coherent receiver for demodulation, and obtain an optical signal through analog-to-digital conversion; The optical signal is preprocessed to obtain a processed optical signal, the processed optical signal is normalized, and the normalized optical signal is input into a pre-established lightweight convolutional neural network model assisted by meta-learning, and the modulation format recognition result is output.

2. The modulation format recognition method based on convolutional neural network according to claim 1, characterized in that: The process of obtaining the data to be transmitted: Collect radio signals or simulated signal data through wireless channel environment, or use simulation software to generate signals of different modulation formats, and perform data preprocessing, including denoising, normalization, and downsampling. For CNN, convert one-dimensional signal data into two-dimensional form, and then perform data annotation. Each sample needs to be labeled with the corresponding modulation type and then divided into training set, validation set and test set. The training set is used for model training, the validation set is used to adjust parameters and prevent overfitting, and the test set is used to evaluate the performance of the final model.

3. The modulation format recognition method based on convolutional neural network according to claim 1, characterized in that: The data to be transmitted is input into a pre-built IQ modulator, loaded onto the I path and Q path of two polarized lights through the IQ modulator, and the multiplexed light is coupled into the few-mode optical fiber by a combiner for data transmission.

4. The modulation format recognition method based on convolutional neural network according to claim 1, characterized in that: The composite light is coupled into the data to be transmitted, and the multiplexed light is coupled into the few-mode optical fiber through the combiner. The signal light and the local oscillator light are sent to the coherent receiver for demodulation. The coherent receiver completes polarization demultiplexing and converts the analog signal into a digital signal through the digital-to-analog converter ADC and enters the DSP to obtain an optical signal.

5. The modulation format recognition method based on convolutional neural network according to claim 1, characterized in that: The optical signal is pre-processed to compensate for the dispersion of the optical fiber channel, recover the clock of the signal, and perform CMA equalization.

6. The modulation format recognition method based on convolutional neural network according to claim 1, characterized in that: The pre-established lightweight convolutional neural network model based on meta-learning assistance includes a meta-learning module: used to learn how to efficiently adapt to new tasks and optimize the parameter initialization of the model so that the model can be quickly adjusted to new tasks under small samples or a small number of iterations; Lightweight Convolutional Neural Network: By introducing lightweight technology to reduce the computational complexity and model size of the network, the lightweight CNN model uses the following technologies: Depthwise separable convolution: decomposes the standard convolution into depthwise convolution and pointwise convolution; Pruning and quantization: pruning unimportant parameters and quantizing floating-point numbers; Knowledge distillation: by transferring the knowledge of the large model to the small model; Feature extraction module: used to extract features from the input signal, through the stacking of convolutional layers, or through the combination of convolutional blocks and pooling layers to extract multi-scale features Task Adaptation Module: used to quickly adjust model parameters in each new task through a meta-learning algorithm; Modulation format classifier: used to classify different modulation formats.

7. The modulation format recognition method based on convolutional neural network according to claim 6, characterized in that: The pre-established meta-learning-assisted lightweight convolutional neural network model is expressed as follows: Assume lightweight CNN model f θ , whose parameter is θ, and each modulation format recognition task is defined as T i , each task T i Contains a support set and a queryset Meta-objective: Let the loss function be Υ, then the meta-objective is defined as: Among them, θ′ is the i The updated parameters; Gradient-based update: In each task T i In the above example, the support set data is used to fine-tune the model parameters θ, and the update method is expressed as: Among them, α is the learning rate, θ i ' represents the updated task-specific parameters; Meta-update: Based on multiple tasks, the initial parameters θ are updated: Among them, β is the learning rate of meta-learning.

8. A modulation format recognition system based on convolutional neural network, characterized in that: include: A signal processing module, used for acquiring data to be transmitted, inputting the data to be transmitted into a pre-built IQ modulator, coupling the composite light into the data to be transmitted to obtain coupled data, inputting the coupled data into a pre-built coherent receiver for demodulation, and obtaining an optical signal through analog-to-digital conversion; The format recognition module is used to preprocess the optical signal to obtain a processed optical signal, normalize the processed optical signal, input the normalized optical signal into a pre-established lightweight convolutional neural network model assisted by meta-learning, and output a modulation format recognition result.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: The memory stores a computer program that can be run on a processor. When the processor loads and executes the computer program, a modulation format recognition method based on a convolutional neural network as described in any one of claims 1 to 7 is adopted.

10. A computer-readable storage medium having a computer program stored therein, characterized in that: When the computer program is loaded and executed by the processor, a modulation format recognition method based on a convolutional neural network as described in any one of claims 1 to 7 is adopted.

Images