Receiver interference response modeling method and system based on temporal convolutional network

Abstract

Disclosed are a receiver interference response modeling method, and a system, based on a temporal convolutional network. Nonlinear characteristics of a receiver are modeled by means of a temporal convolutional network, a temporal convolutional network is trained by using data of a receiver port under a part of typical interference conditions, then a trained model is used for predicting an interference response under another interference condition, and finally, a frequency domain signal of an output end of the receiver is obtained by means of a Fourier transform, and functionality of equivalent modeling of the receiver is achieved. The present invention uses a temporal convolutional network, and effectively solves problems present in an existing modeling method by means of expanding a receptive field and using causal convolution, accurately fitting nonlinear characteristics caused by filtering and frequency mixing of the signal in a receiver under conditions in which an internal parameter of the receiver is unknown.

Description

A receiver interference response modeling method and system based on a time series convolution network TECHNICAL FIELD

[0001] The application belongs to the technical field of electromagnetic interference prediction, and particularly relates to a receiver interference response modeling method and system based on a time series convolution network. BACKGROUND

[0002] At present, behavior-level modeling for radio frequency microwave equipment is divided into two categories. The first category is polynomial models such as Volterra series models, Wiener models and Hammerstein models. However, these two modeling methods have the defect that it is difficult to improve the prediction accuracy of the polynomial model in the case of complex and variable modulation modes of interference signals and increased complexity of data.

[0003] In recent years, due to the excellent performance of neural networks in nonlinear function approximation, many scholars have begun to study behavior-level modeling based on neural networks, such as recurrent neural networks, long short-term memory networks and convolutional neural networks. However, these models such as recurrent neural networks and long short-term memory networks will face the problem of gradient vanishing or explosion when processing long sequences, and although convolutional neural networks perform well in feature extraction, their fixed kernel size limits their expression ability in the time dimension when processing sequence data.

[0004] In order to overcome these limitations, a time series convolution network is proposed, which effectively makes up for the problems existing in the existing modeling methods by expanding the receptive field and using causal convolution. The time series convolution network can expand the receptive field without increasing the number of deep network layers, so that it can understand the complex information coupled in the time domain by interference signals of different frequency components.

[0005] SUMMARY

[0006] The technical problem to be solved by the application is to provide a receiver interference response modeling method and system based on a time series convolution network, which is used for equivalent modeling of a receiver.

[0007] The technical scheme adopted by the application to solve the above technical problem is: a receiver interference response modeling method based on a time series convolution network, comprising the following steps:

[0008] S1: constructing a time series convolution network model of the interference response of the nonlinear characteristics of the receiver;

[0009] S2: collecting receiver port data, and dividing training samples and test samples;

[0010] S3: initializing the time series convolution network, training the time series convolution network using the training samples, and obtaining the weight parameters of the time series convolution network model;

[0011] S4: input the receiver port data in the interference state to be predicted into the trained time sequence convolution network model to obtain time domain characteristics of the interference response in the interference state;

[0012] S5: perform Fourier transform on the time domain characteristics of the interference response to obtain a frequency domain signal of the interference response.

[0013] According to the above scheme, in step S1, the time sequence convolution network model comprises an input layer, a causal convolution network, an inflation convolution network, a residual connection part and an output layer.

[0014] Further, in step S1, the number of neurons of the input layer is 8000; the causal convolution network comprises a two-layer structure, the number of convolution kernels of the first layer is 8, and the number of convolution kernels of the second layer is 16; the inflation factor of the inflation convolution network is 2; the residual connection part is the result connection of the input layer and the inflation convolution network, and is connected to the output layer through a fully connected layer.

[0015] According to the above scheme, in step S2, the receiver port data comprises antenna aperture surface time domain signals, video end time domain signals and frequency domain signals in different typical interference situations.

[0016] Further, in step S2, the specific steps are:

[0017] The antenna aperture surface time domain signals are taken as the input of the time sequence convolution network model, the video end time domain signals are taken as the output of the time sequence convolution network model, and the above data are divided into training samples and test samples in a certain proportion.

[0018] Further, in step S3, the specific steps are:

[0019] S3.1: randomly set all parameters of the time sequence convolution network model to complete initialization of the time sequence convolution network model;

[0020] S3.2: input the antenna aperture surface time domain signals of the training samples into the initialized time sequence convolution network model, let F(·) represent one forward propagation of the time sequence convolution network model, and then the predicted value Y' obtained through the forward propagation calculation of the time sequence convolution network model is: Y' train =F(X train );

[0021] S3.3: let Y train represent the video end time domain signals of the training samples, Y' train represent the predicted value of the video end time domain signals of the training samples obtained through the time sequence convolution network model, and n represent the nth sampling point, then the loss function L(·) of the training time sequence convolution network model is represented as the mean square error MSE:

[0022] S3.4: updating the weight parameters of the time convolution network model by using the error back propagation algorithm;

[0023] S3.5: repeating steps S3.2-S3.4 until the time convolution network model converges, completing the training and obtaining the parameters of the time convolution network model.

[0024] Further, in step S3.1, the parameters of the time convolution network model include weight values and bias values.

[0025] Further, in step S4, the specific steps are as follows:

[0026] S4.1: inputting the antenna aperture plane time domain signal of the test sample as the receiver port data under the interference state to be predicted into the trained time convolution network model, and obtaining the predicted value Y' of the video end time domain signal of the test sample through forward propagation calculation test ;

[0027] S4.2: let Y pre represent the predicted value of the time convolution network model, Y true represent the actual observed value, Y′ true represent the average value of the actual observed value, then use the determinable coefficient R 2 to evaluate the performance of the time convolution network model, and the calculation formula is:

[0028] The value range of R2 is (-∞, 1], and the closer R2 is to 1, the closer the predicted waveform of the time convolution network model is to the actual waveform, and the better the prediction effect is.

[0029] A receiver interference response system based on a time convolution network, comprising a processor and a memory, the memory storing computer instructions, and the processor being configured to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the system implements the steps of a receiver interference response modeling method based on a time convolution network.

[0030] A computer memory storing a computer program executable by a computer processor, the computer program executing a receiver interference response modeling method based on a time convolution network.

[0031] The beneficial effects of the present application are:

[0032] 1.A method and system for modeling receiver interference response based on time convolutional network, which models the non-linear characteristics of the receiver through time convolutional network, trains the time convolutional network using the data of the receiver port under some typical interference conditions, then uses the trained model to predict the interference response under other interference conditions, and finally gets the frequency domain signal of the receiver output end through Fourier transform, thereby realizing the function of equivalent modeling of the receiver.

[0033] 2.The method uses time convolutional network to effectively make up for the problems existing in the current modeling method by expanding the receptive field and using causal convolution, and accurately fits the non-linear characteristics of the signal caused by filtering and mixing in the receiver under the condition that the internal parameters of the receiver are unknown. BRIEF DESCRIPTION OF DRAWINGS

[0034] Fig. 1 is a flowchart of an embodiment of the present application.

[0035] Fig. 2 is a comparison chart of the time domain results of the receiver interference response prediction based on TCN and the simulation results of an embodiment of the present application.

[0036] Fig. 3 is a partial enlarged view of Fig. 2.

[0037] Fig. 4 is a comparison chart of the frequency domain results of the receiver interference response prediction based on TCN and the simulation results of an embodiment of the present application. DETAILED DESCRIPTION

[0038] The present application will be further described in detail below in combination with the drawings and specific embodiments.

[0039] Referring to Fig. 1, an embodiment of the present application includes the following steps:

[0040] Step 1: Construct a time convolutional network model for the sensitive response of the receiver, which includes an input layer, a causal convolutional network, an inflation convolutional network, a residual connection and an output layer.

[0041] The time convolutional network model includes five parts: an input layer, a causal convolution, an inflation convolution, a residual connection and an output layer. The number of neurons in the input layer is 8000, the causal convolutional network includes two layers, the number of convolution kernels in the first layer is 8, and the number of convolution kernels in the second layer is 16. The inflation factor of the inflation convolutional network is 2, and the residual connection part refers to the connection of the input layer and the result of the inflation convolution. Then, it is connected with the output layer through a fully connected layer.

[0042] Step 2: Collect the port data and divide the training samples and test samples.

[0043] The collected receiver port data includes time-domain data of the antenna aperture under different interference conditions, and time-domain and frequency-domain data of the video end. Among them, the time-domain data of the antenna aperture is used as the input of the temporal convolutional network model, and the time-domain data of the video end is used as the output of the temporal convolutional network model. They are divided into training samples and test samples in an 8:2 ratio.

[0044] Step 3: Initialize the Temporal Convolutional Network (TCN), train the TCN using the training sample space, and obtain the weight parameters of the TCN model;

[0045] Step 3.1: Randomly set all parameters of the TCN network, including weight values ​​and bias values, to complete the network initialization;

[0046] Step 3.2: Input the temporal domain data of the antenna aperture surface from the training data obtained in Step 2 into the TCN network of Step 3-1. After forward propagation of the network, the predicted value Y′ is obtained. The calculation formula is: Y′ train =F(X) train (1)

[0047] Where F(·) represents one forward propagation of the neural network.

[0048] Step 3.3: Set the loss function (loss) for training the neural network to the mean squared error (MSE). The goal of training the network is to reduce the loss function to an acceptable range, indicating that the network has converged. The formula for the loss function is:

[0049] Among them, Y train Y' represents the time-domain signal of the training samples at the video end. train This represents the video temporal signal predicted based on the TCN model, where n represents the nth sampling point;

[0050] Step 3.4: Update the network weight parameters using the backpropagation algorithm;

[0051] Step 3.5: Repeat steps S3.2 to 3.4 until the neural network converges, completing the training and obtaining the parameters of the network model;

[0052] Step 4: Input the antenna aperture surface time-domain signal under the interference state to be predicted, and obtain the time-domain characteristics of the interference response under the interference state through the trained temporal convolutional network model.

[0053] Step 4.1: Input the test data of the test set into the trained TCN network, and obtain the predicted value Y' of the video temporal signal through forward propagation. test ;

[0054] Step 4.2: Evaluate the model performance by using the coefficient R2, which is calculated by:

[0055] where Y pre represents the predicted value of the model, Y true represents the actual observed value, and Y' true represents the average value of the actual observed value. The value range of R2 is (-∞, 1], the closer R2 is to 1, the closer the predicted waveform of the model is to the actual waveform, and the better the prediction effect is.

[0056] Step 5: Obtain the frequency domain data of the sensitive response under the interference state by using Fourier transform.

[0057] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0058] In order to verify the modeling and prediction results of the receiver interference response based on the TCN model, the receiver model is constructed as shown in Table 1, the typical interference signal training sample data parameters are shown in Table 2, and the test sample data are shown in Table 3:

[0059] Table 1: Receiver model case

[0060] Table 2: Training sample data parameters

[0061] Table 3: Test sample data parameters

[0062] Fig. 2, Fig. 3 and Fig. 4 show the results of typical cases of the present patent implementation. Fig. 2 is the observed value of the receiver video end time domain signal (black diamonds) and the predicted value of the receiver video end time domain signal based on the TCN model (gray triangles) simulated in Matlab. Fig. 3 is a partial enlarged view of Fig. 2, from which it can be seen that the predicted time domain waveform based on the TCN model has a high degree of similarity with the receiver video end time domain waveform simulated in Matlab, and the calculated coefficient reaches 0.9997, indicating that the time domain fitting degree of the predicted value based on the TCN model and the actual observed value reaches 99.97%. Fig. 4 is a comparison of the predicted value and the actual value of the frequency domain characteristics of the interference response based on the TCN model, and it can be seen that the fitting degree of the frequency domain characteristics is high. From the test sample data parameters, it can be seen that the bandwidth of the test sample is 2.5 MHz, the center frequency offset is 6 MHz, the amplitude of the interference signal is 6 dBm, and the test sample is an unlearned interference signal. It shows that after training the port data under certain typical interference signals, the model has generalization ability and can predict unlearned interference responses.

[0063] The above examples are only used to illustrate the design idea and characteristics of the present application, and the purpose is to enable those skilled in the art to understand the content of the present application and implement it, and the protection scope of the present application is not limited to the above examples. Therefore, any equivalent changes or modifications made in accordance with the principles and design ideas disclosed by the present application are within the protection scope of the present application.

Claims

1. A method of modeling receiver interference response based on a temporal convolutional network, the method comprising: The method comprises the following steps: S1: constructing a time convolution network model of interference response of non-linear characteristics of a receiver; S2: collecting receiver port data, dividing training samples and test samples; S3: initializing the time convolution network, training the time convolution network by using the training samples, and obtaining each weight parameter of the time convolution network model; S4: inputting receiver port data in an interference state to be predicted into the trained time convolution network model to obtain time domain characteristics of the interference response in the interference state; S5: performing Fourier transform on the time domain characteristics of the interference response to obtain a frequency domain signal of the interference response.

2. The method of claim 1, wherein the method is based on a time-sequential convolutional network. In the step S1, the time convolution network model comprises an input layer, a causal convolution network, an expanded convolution network, a residual connection part, and an output layer.

3. The method of claim 2, wherein: In the step S1, the number of neurons of the input layer is 8000; the causal convolution network comprises a two-layer structure, the number of convolution kernels of the first layer is 8, and the number of convolution kernels of the second layer is 16; the expansion factor of the expanded convolution network is 2; and the residual connection part is the result connection of the input layer and the expanded convolution network, and is connected to the output layer through a full connection layer.

4. The method of claim 1, wherein: In the step S2, the receiver port data comprises antenna aperture surface time domain signals, video end time domain signals, and frequency domain signals in different typical interference situations.

5. The method of claim 4, wherein: In the step S2, the specific steps are as follows: The antenna aperture surface time domain signals are taken as the input of the time convolution network model, the video end time domain signals are taken as the output of the time convolution network model, and the above data are divided into training samples and test samples in a certain proportion.

6. The method of claim 4, wherein: In the step S3, the specific steps are as follows: S3.1: randomly setting all parameters of the time convolution network model to complete the initialization of the time convolution network model; S3.2: inputting the antenna aperture surface time domain signals of the training samples into the initialized time convolution network model, where F(·) represents one forward propagation of the time convolution network model, and the predicted value Y' obtained through the forward propagation of the time convolution network model is: Y' train = F(X train ); S3.3: Set Y train represents the video end time domain signal of the training sample, Y train represents the prediction value of the video end time domain signal of the training sample obtained by the time series convolution network model, n represents the nth sampling point, and the loss function L(·) for training the time series convolution network model is represented as the mean square error MSE: S3.4: updating the weight parameters of the time convolution network model by using the error back propagation algorithm; S3.5: repeating steps S3.2-S3.4 until the time convolution network model converges, completing the training, and obtaining the parameters of the time convolution network model.

7. The receiver-interference-response system based on a temporal convolutional network according to claim 6, characterized in that: In the step S3.1, the parameters of the time convolution network model include weight values and bias values. 8.The method of claim 4, wherein: In the step S4, the specific steps are as follows: S4.1: input the antenna aperture plane time domain signal of the test sample into the trained time convolution network model as the receiver port data under the interference state to be predicted, and calculate the predicted value Y' of the video end time domain signal of the test sample through forward propagation. test ; S4.2: Set Y pre Y represents the predicted value of the time series convolution network model true Y' represents the actual observation value true Y represents the mean value of the actual observation value, and the coefficient R 2 The performance of the time series convolution network model is evaluated, and the calculation formula is: The value range of R2 is (-∞, 1], and the closer R2 is to 1, the closer the predicted waveform of the time convolution network model is to the actual waveform, and the better the prediction effect is.

9. A receiver-interference response system based on a temporal convolutional network, the system comprising a processor and a memory, characterized in that: The memory stores computer instructions, and the processor is configured to execute the computer instructions stored in the memory, so that the system implements the steps of the method for modeling the interference response of a receiver based on a time convolution network according to any one of claims 1 to 8.

10. A computer memory, characterized by: The memory stores a computer program executable by a computer processor, and the computer program executes the method for modeling the interference response of a receiver based on a time convolution network according to any one of claims 1 to 8.

Images