CN114553650B - Multi-level neural network-based anti-mode coupling signal complex format analysis method - Google Patents

Abstract

The invention discloses an anti-mode coupled signal complex format analysis method based on a multi-level neural network. According to the received unknown signal, an unknown signal constellation diagram is generated; the unknown signal constellation diagram is input into a convolutional neural network model obtained by training, and the transmission mode and modulation format. Training to obtain a convolutional neural network model: the convolutional neural network model extracts high-dimensional information features from the labeled constellation map; according to the high-dimensional information features, it is determined to obtain the transmission mode and modulation format; Label comparison, and iteratively update the parameters of the convolutional neural network model. The invention solves the problem of inter-mode coupling in the mode division multiplexing, and can accurately identify the modulation format of the unknown signal constellation diagram under the condition of high coupling coefficient for the interference of modulation format identification, and reduce the interference of the mode coupling to the received signal. , the modulation format recognition results are more accurate and robust.

Description

Analysis method of complex format of anti-pattern coupling signal based on multi-level neural network

technical field

The invention relates to an anti-pattern coupling signal complex format analysis method based on a multi-level neural network, and belongs to the technical field of deep learning technology and signal complex format analysis.

Background technique

With the continuous improvement of the level of social informatization, emerging technologies such as video services, 5G, and the Internet of Things, as well as new services such as big data services and cloud computing, data services continue to grow at an explosive rate. Existing optical fiber transmission resources is being consumed rapidly. The current network traffic is approaching the limit of the existing transmission technology, and the need to expand the transmission bandwidth is becoming more and more urgent. It has become an urgent task to develop a new transmission technology to meet the requirements of future network development. However, the dimensions of light amplitude, phase, frequency, time slot and polarization in the optical fiber transmission network have been fully utilized, and only the spatial dimension still has a huge potential for development. Therefore, space division multiplexing technology based on spatial dimension has become a hot technology to solve the problem of channel capacity.

The mode division multiplexing technology is a kind of space division multiplexing technology, that is, using the orthogonality between the modes in the optical fiber, different modes are used as independent channels to carry different information, so that it is propagated in the optical fiber at the same time. Ideally, the different modes of the signal carried in the fiber are orthogonal to each other, and crosstalk does not occur during transmission. How many modes exist in the fiber, how much the corresponding channel transmission capacity can be expanded, and the modes in the fiber propagate independently without affecting each other. In the next-generation elastic optical network based on modulo-division multiplexing technology, the transmitter will dynamically change parameters such as the symbol rate or modulation format of the transmitted signal according to user services and system resources, while for the receiver, the received signal is unknown.

However, in the actual few-mode fiber, due to the limitations of the manufacturing process and the influence of external forces, random coupling of each mode signal will occur during the transmission process, which seriously affects the transmission performance. In order to solve the mode coupling problem, there are usually two schemes for the mode division multiplexing system. The first is to reduce the requirement of absolute orthogonality of the signals in the link, which can tolerate a certain degree of signal coupling, and use a MIMO equalizer at the receiving end to achieve signal recovery. However, for strong coupling, the complexity of the MIMO algorithm will be greatly increased. The second is to keep the modes in the channel as orthogonal as possible to reduce the crosstalk between modes, and this solution has stricter requirements on the optical devices on the link.

Mode coupling in the mode multiplexing system is one of the main damages of the mode division multiplexing system, which is introduced by the mode multiplexer and the few-mode fiber. As shown in Figure 1, ideally, the different modes of the signals carried in the few-mode fiber are orthogonal to each other, and crosstalk does not occur during transmission. However, in the actual few-mode fiber, as shown in Figure 2, due to the limitations of the manufacturing process and the influence of the external force on the fiber during use, the orthogonality of the modes in the few-mode fiber is destroyed, and the signals of each mode will occur during the transmission process. Random coupling seriously affects the transmission performance. The coupling between the degenerate mode and the non-degenerate mode makes the propagation constants of the modes no longer equal, and random energy migration causes crosstalk between signals, which affects all symbols and complicates the MIMO algorithm at the receiver.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to overcome the defects of the prior art, and to provide an anti-mode coupling signal complex format analysis method based on a multi-level neural network.

In order to achieve the above object, the present invention provides an anti-pattern coupling signal complex format analysis method based on a multi-level neural network, including:

Generate an unknown signal constellation diagram according to the received unknown signal;

Input the unknown signal constellation diagram into the convolutional neural network model obtained by training, and predict the transmission mode and modulation format;

Train a convolutional neural network model, including:

Input the training data set into the convolutional neural network model, the training data set includes the labeled constellation map;

The convolutional neural network model extracts high-dimensional informative features from the labeled constellation map;

According to high-dimensional information features, determine the transmission mode and modulation format;

Compare the transmission mode and modulation format obtained by the judgment with the label, and update the parameters of the convolutional neural network model;

Repeat the above steps until the preset number of iterations is reached to obtain the final convolutional neural network model;

According to the characteristics of high-dimensional information, it is determined to obtain the transmission mode and modulation format, including:

The convolutional neural network model includes the first-level convolutional neural network and the second-level convolutional neural network,

The first-level convolutional neural network extracts high-dimensional information features and discriminates the transmission mode;

The second-level convolutional neural network obtains the parameters output by the first-level convolutional neural network, and identifies the modulation format;

Both the first-level convolutional neural network and the second-level convolutional neural network include a first convolutional layer, a second convolutional layer, a first pooling layer, a second pooling layer, a first fully connected layer, and a second fully connected layer. The connection layer, the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the first fully connected layer and the second fully connected layer are connected in sequence.

Preferably, the transmission mode and modulation format obtained by the judgment are compared with the labels, and the parameters of the convolutional neural network model are updated and iterated, including:

Compare the transmission mode and modulation format obtained by the judgment with the tag, and obtain the determined data distribution and the correct data distribution in the tag;

Compute the mean squared error of the determined data distribution and the correct data distribution in the label;

Update parameters including neuron biases and weights in an iterative convolutional neural network model.

Preferentially, the labeled constellation diagrams are generated from optical signals with different known coupling coefficients in different known transmission modes.

Preferentially, the high-dimensional information features extracted by the first-stage convolutional neural network include the CCD spot mode field map of the transmitted signal, the imaging spectrum and the double Fourier transform sequence.

Preferentially, the parameters output by the first-stage convolutional neural network include signal constellation map, Stokes space spherical mapping parameters and higher-order cumulants.

Preferably, before inputting the unknown signal constellation into the convolutional neural network model obtained by training, the unknown signal constellation is averaged and normalized.

A storage medium on which a computer program is stored, and when the computer program is executed by a processor, implements the steps of any one of the methods described above.

The beneficial effects achieved by the present invention:

By constructing a multi-level convolutional neural network, the invention learns the characteristics of the constellation diagram of the transmission signals of different modulation formats in different modes when there is mode coupling, and uses the powerful fitting ability of the neural network to learn the effect of mode coupling on the signal constellation diagram in the few-mode fiber. Through the idea of "level by level", the first-level convolutional neural network is used to overcome the mode coupling in the few-mode fiber and discriminate the transmission mode of the unknown signal constellation; the second-level convolutional neural network is used to output The modulation format in this transmission mode can solve the problem of inter-mode coupling in the mode division multiplexing. For the interference of modulation format identification, the modulation format of the unknown signal constellation can be accurately identified when the coupling coefficient is high, and the reduction of The interference caused by the mode coupling to the received signal makes the identification result of the modulation format more accurate and robust; after the modulation format of the received optical signal is known, it is convenient to select different algorithms for the received optical signal according to different modulation formats. The signal undergoes a series of signal processing such as adaptive equalization, frequency offset recovery and carrier phase recovery.

Description of drawings

Figure 1 is a schematic diagram of the influence of mode coupling on the transmission signal under ideal transmission;

Fig. 2 is a schematic diagram of the influence of mode coupling on the transmission signal under actual transmission;

Fig. 3 is the flow chart of the present invention;

Fig. 4 is the principle block diagram of the present invention;

Fig. 5 is the structure diagram of the single-layer convolutional neural network of the present invention;

FIG. 6 is a flowchart of an embodiment of the present invention.

Detailed ways

The following examples are only used to illustrate the technical solutions of the present invention more clearly, and cannot be used to limit the protection scope of the present invention.

Example 1

A method for analyzing complex formats of anti-pattern coupling signals based on multi-level neural networks, including:

Generate an unknown signal constellation diagram according to the received unknown signal;

The unknown signal constellation diagram is input into the convolutional neural network model obtained by training, and the transmission mode and modulation format are obtained by prediction.

Further, in this embodiment, a convolutional neural network model is obtained by training, including:

Input the training data set into the convolutional neural network model, the training data set includes the labeled constellation map;

The convolutional neural network model extracts high-dimensional informative features from the labeled constellation map;

According to high-dimensional information features, determine the transmission mode and modulation format;

Compare the transmission mode and modulation format obtained by the judgment with the label, and update the parameters of the convolutional neural network model;

Repeat the above steps until the preset number of iterations is reached to obtain the final convolutional neural network model.

Further, in this embodiment, the transmission mode and modulation format obtained by the judgment are compared with the label, and the parameter update iteration is performed on the convolutional neural network model, including:

Compare the transmission mode and modulation format obtained by the judgment with the tag, and obtain the determined data distribution and the correct data distribution in the tag;

Compute the mean squared error of the determined data distribution and the correct data distribution in the label;

Update parameters including neuron biases and weights in an iterative convolutional neural network model.

Further, the constellation diagram with the label in this embodiment is generated according to the optical signal under different known transmission modes and different known coupling coefficients.

Further, in this embodiment, according to high-dimensional information features, it is determined to obtain the transmission mode and modulation format, including:

The convolutional neural network model includes the first-level convolutional neural network and the second-level convolutional neural network,

The first-level convolutional neural network extracts high-dimensional information features and discriminates the transmission mode;

The second-stage convolutional neural network obtains the parameters output by the first-stage convolutional neural network and identifies the modulation format.

Further, the high-dimensional information features extracted by the first-stage convolutional neural network in this embodiment include the CCD spot mode field map of the transmission signal, the imaging spectrum and the double Fourier transform sequence.

Further, the parameters output by the first-stage convolutional neural network in this embodiment include a signal constellation diagram, a Stokes space spherical mapping parameter, and a high-order cumulant.

Further, in this embodiment, both the first-level convolutional neural network and the second-level convolutional neural network include a first convolutional layer, a second convolutional layer, a first pooling layer, a second pooling layer, and a first convolutional layer. The fully connected layer and the second fully connected layer, the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the first fully connected layer and the second fully connected layer are sequentially connected.

Further, in this embodiment, before the unknown signal constellation is input into the convolutional neural network model obtained by training, the unknown signal constellation is averaged and normalized.

An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the steps of any one of the methods described above when the processor executes the program.

A storage medium on which a computer program is stored, and when the computer program is executed by a processor, implements the steps of any one of the methods described above.

The method proposed by the present invention, as shown in FIG. 3 , is divided into a training phase and a discrimination phase. Mode coupling in optical fiber communication systems can cause random migration of optical signal energy transmitted between different modes. In the training phase, the training dataset is used to perform supervised learning on the convolutional neural network model. The constellation diagrams in the training data set are generated at the receiving end according to the input known optical signals under different transmission modes and different coupling coefficients.

The two-stage convolutional neural network of the convolutional neural network model extracts high-dimensional information features from the labeled constellation map, respectively, and discriminates the transmission mode and modulation format of the input known optical signal.

The recognition results are compared with the labels, and the parameters of the convolutional neural network model are updated and iterated to satisfy the number of iterations, and the optimization of the convolutional neural network model is completed.

In the discrimination stage, the receiver receives an unknown signal, which is an optical signal with an unknown transmission mode and an unknown modulation format. The receiving end generates a constellation diagram according to the received unknown signal, and inputs the constellation diagram into the convolutional neural network model obtained by training. After simple preprocessing such as averaging and normalization, the constellation map is sent to the trained convolutional neural network model. The first-level convolutional neural network of the convolutional neural network model discriminates the transmission mode of the unknown signal according to the extracted high-dimensional information features, and the second-level convolutional neural network of the convolutional neural network model is based on the output of the first-level convolutional neural network. The parameter identifies the modulation format of the unknown signal and completes the complex format analysis of the unknown signal.

The scheme of the present invention solves the influence of mode crosstalk on the receiving end by means of a multi-level neural network, as shown in FIG. 4 . The first-level convolutional neural network includes an input layer, a first hidden layer and a first output layer, and the second-level convolutional neural network includes a second hidden layer and a second output layer. The layer receives a large amount of nonlinear input information. The input information is transmitted, analyzed and weighed in the neuron connection of the first hidden layer, and output in the first output layer. The first hidden layer is composed of the input layer and the first output layer. It consists of a large number of neuron connections, and the first hidden layer is connected by a nonlinear activation function.

The scheme of the present invention uses the labeled constellation diagram as the input of the first-level convolutional neural network. After the first-level convolutional neural network performs high-dimensional information feature extraction through the first hidden layer, it outputs the unknown signal constellation diagram/training data set The transmission mode of , and used as the input of the second stage convolutional neural network. The second hidden layer is used in the second-level convolutional neural network to analyze and process the high-dimensional information features of the constellation, and finally predict the modulation format of the unknown signal constellation/training data set. Since the mode coupling will cause energy transfer to the received unknown signal, the received unknown signals of different modes will be randomly coupled. The scheme of the present invention avoids the crosstalk caused by mode coupling in the received unknown signal by first identifying the transmission mode of the received unknown signal, and the second-layer convolutional neural network identifies the received unknown signal through the high-dimensional information features of the constellation diagram. modulation format.

In the solution of the present invention, the internal network structures of the first-level convolutional neural network and the second-level convolutional neural network are the same. The specific structure is shown in Figure 5. There are two convolutional layers, two pooling layers and two full The connection layers are the first convolutional layer, the second convolutional layer, the first pooling layer, the second pooling layer, the first fully connected layer and the second fully connected layer. The purpose of using the convolution layer is to extract the high-dimensional information features of the constellation map. The convolution operation can maintain the spatial relationship between pixels, turn the local sub-matrix of the input constellation map into an element, and complete the dimensionality reduction of the input image information. . The pooling layer can achieve the effect of sparse parameters and reduce the amount of data. The fully connected layer integrates all the extracted high-dimensional information features, integrates the class-distinguishing local information in the convolutional layer and the pooling layer, and finally gives the output result, which includes the transmission mode and modulation format.

The signal complex format analysis scheme proposed by the present invention utilizes a two-stage convolutional neural network to eliminate crosstalk between transmission signals of different modes in a mode division multiplexing communication system, and has good robustness. At the same time, the convolutional neural network has a strong feature extraction ability for the constellation diagram of the optical signal. Compared with the traditional recognition scheme, the convolutional neural network based on machine learning has a strong fitting ability. By supervised training on the training dataset, convolutional neural networks can achieve high recognition accuracy.

The modulation format of the signal changes, and the algorithms related to the modulation format in the receiver DSP also need to be changed. The invention completes the identification of the modulation format of the received unknown signal, and facilitates the subsequent digital signal processing of the optical signal.

Aiming at the problem of signal complex format analysis in the elastic optical network based on the few-mode fiber, the invention solves the modulation format identification problem in the few-mode fiber by using the multi-stage convolutional neural network, and reduces the interference of the mode coupling on the received signal, so that the modulation The format recognition results are more accurate and robust. Through the idea of "level by level", this scheme gives the transmission mode and modulation format of the optical signal through the convolutional neural network with the help of the constellation diagram of the received optical signal.

Embodiment 2

As shown in Figure 6, at the transmitter end of the mode division multiplexing communication system, transmitter Tx1, transmitter Tx2, transmitter Tx3 and transmitter Tx4 respectively load the modulated optical signals on LP01, LP11a, LP11b and LP21 on four modes, and transmit the optical signal through the few-mode fiber. The optical communication network system will dynamically change various parameters such as the modulation format and symbol rate of the transmitted optical signal at the transmitting end according to the user's needs and channel conditions, so as to achieve the effect of rationally configuring system resources. In the process of few-mode fiber transmission, different transmission modes will generate random coupling, resulting in the phenomenon of energy migration. At the receiving end of the mode division multiplexing communication system, the mode demultiplexer decomposes the beam into four modes: LP01, LP11a, LP11b and LP21, and the beam passes through receiver Rx1, receiver Rx2, receiver Rx3 and receiver Rx4. Input non-modulation format related algorithms and DSP units, mode multiplexers and mode demultiplexers will also introduce mode coupling due to incomplete energy conversion. At this time, the optical signal received by the receiver will affect the transmission performance due to mode coupling.

The receiving end needs to use the DSP unit to perform corresponding compensation or equalization algorithm processing on the received optical signal. The DSP unit will first perform non-modulation format related algorithms such as clock recovery and dispersion compensation. Next, according to the modulation format correlation algorithm scheme proposed by the present invention, the convolutional neural network model is used to identify the modulation format in the case of mode coupling. The first-level convolutional neural network CNN I gives the transmission mode of the received optical signal, and the second-level convolutional neural network CNNII gives the modulation format of the received optical signal. After the modulation format of the received optical signal is known, different algorithms can be selected to perform a series of signal processing on the received optical signal, such as adaptive equalization, frequency offset recovery, and carrier phase recovery, according to different modulation formats.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It will be understood that each flow and/or block in the flowcharts and/or block diagrams, and combinations of flows and/or blocks in the flowcharts and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in one or more of the flowcharts and/or one or more blocks of the block diagrams.

These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory result in an article of manufacture comprising instruction means, the instructions An apparatus implements the functions specified in a flow or flows of the flowcharts and/or a block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in one or more of the flowcharts and/or one or more blocks of the block diagrams.

The above are only the preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the technical principle of the present invention, several improvements and modifications can also be made. These improvements and modifications It should also be regarded as the protection scope of the present invention.

Claims (6)

1. The method for analyzing the complex format of the anti-mode coupling signal based on the multi-level neural network is characterized by comprising the following steps:

generating an unknown signal constellation diagram according to the received unknown signal;

inputting an unknown signal constellation diagram into a convolutional neural network model obtained by training, and predicting to obtain a transmission mode and a modulation format;

training to obtain a convolutional neural network model, comprising:

inputting a training data set into a convolutional neural network model, wherein the training data set comprises a constellation diagram with labels;

extracting high-dimensional information characteristics from the constellation map with the label by the convolutional neural network model;

according to the high-dimensional information characteristics, judging to obtain a transmission mode and a modulation format;

comparing the transmission mode and the modulation format obtained by judgment with the label, and performing parameter updating iteration on the convolutional neural network model;

repeating the steps until reaching the preset iteration times to obtain a final convolution neural network model;

according to the high-dimensional information characteristics, the transmission mode and the modulation format are judged and obtained, and the method comprises the following steps:

the convolutional neural network model comprises a first stage convolutional neural network and a second stage convolutional neural network,

extracting high-dimensional information characteristics and judging a transmission mode by a first-stage convolutional neural network;

the second-stage convolutional neural network acquires parameters output by the first-stage convolutional neural network and identifies a modulation format;

the first convolution layer, the first pooling layer, the second convolution layer, the second pooling layer, the first full-connection layer and the second full-connection layer are sequentially connected;

the labeled constellation diagram is generated according to the optical signal under different known coupling coefficients of different known transmission modes.

2. The method for resolving the anti-mode-coupling signal complex format based on the multi-level neural network according to claim 1, wherein the transmission mode and the modulation format obtained by the judgment are compared with the label, and the parameter updating iteration is performed on the convolutional neural network model, and comprises the following steps:

comparing the transmission mode and the modulation format obtained by judgment with the label to obtain the judged data distribution and the correct data distribution in the label;

calculating the mean square error of the determined data distribution and the correct data distribution in the label;

and updating parameters including neuron bias and weight in the iterative convolutional neural network model.

3. The multi-level neural network-based anti-mode coupling signal complex format resolving method as claimed in claim 1, wherein the high-dimensional information features extracted by the first-level convolutional neural network comprise a CCD light spot mode field pattern, an imaging frequency spectrum and a double Fourier transform sequence of the transmission signal.

4. The multi-level neural network-based anti-mode-coupling signal complex format parsing method as recited in claim 1, wherein the parameters output by the first-level convolutional neural network comprise a signal constellation, stokes space spherical mapping parameters and higher-order cumulants.

5. The method according to claim 1, wherein the unknown signal constellation is averaged and normalized before being input into the convolutional neural network model obtained by training.

6. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 5.

Images