A method and system for monitoring optical performance of space division multiplexing signal
Through knowledge distillation technology, the knowledge of the teacher model is transferred to the student model, solving the problem of poor adaptability of neural network models in multi-channel optical networks, and achieving lightweight and efficient multi-channel optical performance monitoring.
Patent Information
- Application Number
- CN202111577464.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-22
- Publication Date
- 2025-05-09
- Estimated Expiration
- 2041-12-22
AI Technical Summary
The existing feature-based neural network model cannot adapt to the monitoring tasks of the current channel when the channel environment parameters change, making it difficult to achieve comprehensive optical performance monitoring in multi-channel optical networks.
Through the knowledge distillation method, the knowledge of the trained teacher model is transferred to the lightweight student model, realizing optical performance monitoring of multi-channel optical networks. The specific steps include building a training data set, training a teacher model, obtaining a soft target set from the teacher model and calculating weights based on the channel temperature, and training a student model to complete the optical performance monitoring task.
Only a lightweight student model can complete the optical performance monitoring task of all channels in the space-division multiplexing fiber, reducing training costs and deployment difficulties, while improving system response speed and reducing computing complexity.
Smart Images
Figure CN114358061B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of optical performance monitoring, and in particular relates to a method and system for monitoring the optical performance of a space-division multiplexing signal. Background Art
[0002] With the vigorous development of space-division multiplexing optical fiber communication technology, long-span optical transmission systems with space-division multiplexing optical fiber as the transmission medium have been widely used. The space-division multiplexing transmission system is a technical solution that can increase channel capacity by orders of magnitude and reduce the transmission cost per unit bandwidth at the same time. At the same time, due to the increasing number of signal types in the optical network, it is becoming increasingly difficult for the receiving end to process the signal. The optical network transmitter will dynamically change the signal transmission rate, modulation format and other information according to the link situation and business needs. At this time, the received signal is an unknown signal for the receiving end. Effective monitoring of the performance of complex and changeable large-capacity dynamic optical networks is a necessary technical means to achieve reasonable allocation of network resources and ensure efficient and stable operation of the network.
[0003] Optical performance monitoring in coherent optical communication systems based on few-mode multi-core fiber can achieve real-time perception of physical layer parameters. Algorithms such as channel equalization and carrier phase recovery in the back-end DSP of the coherent receiver require the modulation format of the signal as prior knowledge. At present, optical performance monitoring solutions based on machine learning are receiving more and more attention. This type of solution uses a neural network model to learn signal characteristics under different link conditions and realize intelligent optical network performance monitoring in complex environments. The current feature-based neural network model can only handle given tasks, but the optical network based on few-mode multi-core fiber has multiple transmission channels. If the environmental parameters of the channel change, such as channel change, the neural network model is no longer applicable to the monitoring task of the current channel. Summary of the invention
[0004] The purpose of the present invention is to provide a method and system for monitoring the optical performance of space-division multiplexing signals, which transfers the knowledge of a trained teacher model to a lightweight student model in the form of knowledge distillation, so as to realize the optical performance monitoring task of all channels in the space-division multiplexing optical fiber using only one lightweight neural network model.
[0005] In order to achieve the above object, the technical solution adopted by the present invention is:
[0006] On one hand, the present invention provides a method for monitoring the optical performance of a space division multiplexing signal, comprising:
[0007] The signal characteristics of the channel to be detected are collected and input into the trained student model to obtain the channel performance monitoring results;
[0008] The training method of the student model includes:
[0009] Acquire feature information from the channels of each optical performance monitoring task and construct a training data set;
[0010] The training data set is used to train the optical performance monitoring capability of the neural network model in a variety of ways. At the same time, the training process corrects the weights of each parameter of the neural network model through the loss function, and the trained neural network model is used as the teacher model.
[0011] A low-complexity lightweight neural network is established as the student model. The soft target set corresponding to each channel is obtained from the teacher model, and the weight of each soft target set is calculated according to the temperature T of the channel. The student model is trained using the weighted soft target set.
[0012] Preferably, the method for monitoring the optical performance of a training neural network model comprises:
[0013] The prior knowledge learned from k-1 channels is transferred to the current neural network model, and the feature information of the kth channel is learned through the current neural network model to obtain knowledge data; the knowledge data is refined and stored as prior knowledge for the next migration.
[0014] Preferably, the methods of training the neural network model include: knowledge retention, knowledge transfer and model expansion.
[0015] Preferably, the characteristic information includes: the Stokes vector of the space division multiplexing signal and the OSNR value of the corresponding link.
[0016] Preferably, the formula of the loss function is:
[0017]
[0018] Among them, L(θ) is the loss function under the current channel; θ represents the parameters of the neural network, including weights and biases, i represents the i-th channel, and θ i represents the updated parameters in the i-th channel, is the parameter before updating in the i-th channel; b i It means that for the parameter θ in the i-th channel i The importance of b i The larger the parameter θ is, the i The more important it is; λ is represented by the parameters of the training loss and accuracy results of the neural network model.
[0019] Preferably, the weight of each soft target set is calculated according to the temperature T of the channel, and the calculation formula is:
[0020]
[0021] Among them, parameter T controls the importance of each soft target, q irepresents the probability of the monitoring result of the signal in the i-th channel, Z i and Z j Represent the output vectors of the network's fully connected layer for the i-th channel and the j-th channel, respectively.
[0022] Another aspect of the present invention provides a system for monitoring the optical performance of a space division multiplexing signal, comprising:
[0023] An acquisition module is used to obtain characteristic information from the channels of each optical performance monitoring task and construct a training data set;
[0024] A training module is used to train the optical performance monitoring capability of the neural network model in a variety of ways using the training data set. At the same time, the training process corrects the weights of various parameters of the neural network model through a loss function, and uses the trained neural network model as a teacher model;
[0025] The knowledge distillation module is used to obtain the soft target set corresponding to each channel from the teacher model, calculate the weight of each soft target set according to the temperature T of the channel, and train the student model using the weighted soft target set;
[0026] The monitoring module is used to collect the signal characteristics of the channel to be detected and input them into the trained student model to obtain the channel performance monitoring results.
[0027] Compared with the prior art, the present invention has the following beneficial effects:
[0028] In the present invention, the soft target set corresponding to each channel is obtained from the teacher model, and the student model is trained using the soft target set; the signal characteristics of the channel to be detected are collected and input into the student model to obtain the channel performance monitoring results; only one lightweight student model is needed to complete all optical performance monitoring tasks in the space-division multiplexing optical fiber, which greatly reduces the training cost and actual deployment difficulty of the space-division multiplexing signal modulation format recognition solution based on the neural network. At the same time, the lightweight student model can improve the response speed of the system and reduce the computational complexity of the network.
[0029] In the present invention, the weights of each soft target set are calculated according to the temperature T of the channel, and the soft target set with weights is used to train the student model, thereby ensuring the correctness and reliability of the knowledge data acquired by the student model during the knowledge distillation process. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 A flow chart of a method for monitoring the optical performance of a space division multiplexing signal provided by the present invention;
[0031] Figure 2 It is a flow chart of teacher model training based on lifelong learning in the present invention;
[0032] Figure 3A flow chart of the student model provided by the present invention for monitoring the optical performance of space division multiplexing signals. DETAILED DESCRIPTION
[0033] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.
[0034] Embodiment 1
[0035] like Figure 1 As shown, a method for monitoring the optical performance of a space-division multiplexing signal includes: collecting signal characteristics of a channel to be detected and inputting them into a trained student model to obtain a channel performance monitoring result;
[0036] The training method of the student model includes:
[0037] A training data set is constructed based on the Stokes vector of the space-division multiplexing signal in the channel of each optical performance monitoring task and the OSNR value of the corresponding link;
[0038] The training data set is used to train the monitoring capability of the optical performance of the neural network model in a manner of knowledge retention, knowledge transfer and model expansion, wherein the process of training the monitoring capability of the optical performance of the neural network model includes: Figure 2 As shown, the prior knowledge learned from k-1 channels is transferred to the current neural network model, and the feature information of the kth channel is learned through the current neural network model to obtain knowledge data; the knowledge data is refined and stored as prior knowledge for the next migration.
[0039] At the same time, the training process corrects the weights of each parameter of the neural network model through the loss function, and uses the trained neural network model as the teacher model; the formula of the loss function is:
[0040]
[0041] Among them, L(θ) is the loss function under the current channel; θ represents the parameters of the neural network, including weights and biases, i represents the i-th channel, and θ i represents the updated parameters in the i-th channel, is the parameter before updating in the i-th channel; b i It means that for the parameter θ in the i-th channel i The importance of b i The larger the parameter θ is, the i The more important it is; λ is represented by the parameters of the training loss and accuracy results of the neural network model.
[0042] A low-complexity lightweight neural network is established as the student model. The soft target set corresponding to each channel is obtained from the teacher model, and the weight of each soft target set is calculated according to the temperature T of the channel. The student model is trained using the weighted soft target set.
[0043] The weight of each soft target set is calculated according to the temperature T of the channel. The calculation formula is:
[0044]
[0045] Among them, parameter T controls the importance of each soft target, q i represents the probability of the monitoring result of the signal in the i-th channel, Z i and Z j Represent the output vectors of the network's fully connected layer for the i-th channel and the j-th channel, respectively.
[0046] like Figure 3 As shown in the figure, transmitters Tx1 to Txn load the modulated signals on n different channels through the space division multiplexer, and transmit the signals to the space division multiplexer through the space division multiplexing optical fiber; the elastic optical network dynamically changes the modulation format, symbol rate and other parameters of the transmission signal according to the user's needs and the channel conditions, so as to achieve the effect of reasonable allocation of system resources. In the process of space division multiplexing optical fiber transmission, different transmission channels will produce random coupling, and the interference and channel transmission conditions of different channels are different; the space division multiplexer demultiplexes the light beam into n single-mode optical fibers.
[0047] Receivers Rx1 to Rxn need to use DSP units to perform corresponding compensation, equalization and other algorithm processing on the signals. DSP will first extract signal features and use them as input for the student model. Then, according to the scheme proposed by the present invention, the trained student model can directly monitor the signals in n different channels, obtain the modulation format of the received signal and the optical signal-to-noise ratio of the link, and feed back to the control layer of the optical network to achieve control of network parameters.
[0048] Embodiment 2
[0049] This embodiment provides a monitoring method that can be applied to the monitoring method described in Embodiment 1, a space division multiplexing signal optical performance monitoring system, including:
[0050] An acquisition module is used to obtain characteristic information from the channels of each optical performance monitoring task and construct a training data set;
[0051] A training module is used to train the optical performance monitoring capability of the neural network model in a variety of ways using the training data set. At the same time, the training process corrects the weights of various parameters of the neural network model through a loss function, and uses the trained neural network model as a teacher model;
[0052] The knowledge distillation module is used to obtain the soft target set corresponding to each channel from the teacher model, calculate the weight of each soft target set according to the temperature T of the channel, and train the student model using the weighted soft target set;
[0053] The monitoring module is used to collect the signal characteristics of the channel to be detected and input them into the trained student model to obtain the channel performance monitoring results.
[0054] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0055] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0056] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0057] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0058] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. A method for monitoring the optical performance of a space-division multiplexing signal, characterized in that: include: The signal characteristics of the channel to be detected are collected and input into the trained student model to obtain the channel performance monitoring results; The training method of the student model includes: Acquire feature information from the channels of each optical performance monitoring task and construct a training data set; The monitoring capability of the optical performance of the neural network model is trained by using the training data set in the manner of knowledge retention, knowledge transfer and model expansion, wherein the process of training the monitoring capability of the optical performance of the neural network model includes: transferring the prior knowledge learned from k-1 channels to the current neural network model, and learning the characteristic information of the kth channel through the current neural network model to obtain knowledge data; refining and storing the knowledge data as prior knowledge for the next transfer, and at the same time, the training process corrects the weights of each parameter of the neural network model through the loss function, and uses the trained neural network model as a teacher model; A low-complexity lightweight neural network is established as the student model. The soft target set corresponding to each channel is obtained from the teacher model, and the weight of each soft target set is calculated according to the temperature T of the channel. The student model is trained using the weighted soft target set.
2. The method for monitoring the performance of spatial division multiplexing signal light according to claim 1, characterized in that: The characteristic information includes: the Stokes vector of the space division multiplexing signal and the OSNR value of the corresponding link.
3. The method for monitoring the performance of spatial division multiplexing signal light according to claim 1, characterized in that: The formula of the loss function is: Among them, L(θ) is the loss function under the current channel; θ represents the parameters of the neural network, including weights and biases, i represents the i-th channel, and θ i represents the updated parameters in the i-th channel, is the parameter before updating in the i-th channel; b i It means that for the parameter θ in the i-th channel i The importance of b i The larger the parameter θ is, the i The more important it is; λ is represented by the parameters of the training loss and accuracy results of the neural network model.
4. The method for monitoring the performance of spatial division multiplexing signal light according to claim 1, characterized in that: The weight of each soft target set is calculated according to the channel temperature T, and the calculation formula is: Among them, parameter T controls the importance of each soft target, q i represents the probability of the monitoring result of the signal in the i-th channel, Z i and Z j Represent the output vectors of the network's fully connected layer for the i-th channel and the j-th channel, respectively.
5. A space division multiplexing signal optical performance monitoring system, characterized in that: include: An acquisition module is used to obtain characteristic information from the channels of each optical performance monitoring task and construct a training data set; A training module is used to train the optical performance monitoring capability of the neural network model in a variety of ways using the training data set. At the same time, the training process corrects the weights of various parameters of the neural network model through a loss function, and uses the trained neural network model as a teacher model; The knowledge distillation module is used to obtain the soft target set corresponding to each channel from the teacher model, calculate the weight of each soft target set according to the temperature T of the channel, and train the student model using the weighted soft target set; The monitoring module is used to collect the signal characteristics of the channel to be detected and input them into the trained student model to obtain the channel performance monitoring results; The training module uses the training data set to train the monitoring capability of the optical performance of the neural network model in the manner of knowledge retention, knowledge transfer and model expansion. The process of training the monitoring capability of the optical performance of the neural network model includes: transferring the prior knowledge learned from k-1 channels to the current neural network model, learning the characteristic information of the kth channel through the current neural network model to obtain knowledge data; and refining and storing the knowledge data as prior knowledge for the next migration.