Test excitation signal strain generator based on functional link resource autonomous matching

By using a test excitation signal strain generator based on autonomous matching of functional link resources, combined with deep neural network technology, the problems of realism and scene adaptability in existing electromagnetic signal simulation have been solved, achieving high-precision and high-efficiency signal simulation.

CN116054985BActive Publication Date: 2026-11-10CHINA ELECTRONIS TECH INSTR CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211705390.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-12-29
Publication Date
2026-11-10
Estimated Expiration
2042-12-29

AI Technical Summary

Technical Problem

Existing electromagnetic signal simulation generation technology is limited by ideal mathematical formulas and the experience of testers, and cannot meet the requirements of high fidelity and high efficiency in complex scenarios. Furthermore, the signal simulation has large errors compared to the real signal, and its application scenarios are limited.

Method used

A test excitation signal strain generator based on autonomous matching of functional link resources is adopted. It combines a deep neural network feature processing module, a signal generation module, and a signal processing training module. A signal feature model library is established through a deep neural network with Fourier expansion, and a noise suppression loss function is introduced to perform autonomous signal matching and optimization.

Benefits of technology

It achieves high-precision simulation of complex electromagnetic signals, breaks through the limitations of ideal mathematical formulas, improves the realism of signal simulation and the ability to adapt to application scenarios, and reduces the dependence on the experience of testers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116054985B_ABST
    Figure CN116054985B_ABST
Patent Text Reader

Abstract

This invention discloses a test excitation signal strain generator based on autonomous matching of functional link resources, comprising a deep neural network feature processing module, a signal generation module, and a signal processing training module. The deep neural network feature processing module is configured to perform feature processing using a deep neural network based on Fourier expansion, establish a signal feature model library based on signal feature learning, generate simulated signal data through autonomous matching of test scenarios, and autonomously construct the implicit feature space of the signal. The signal generation module is configured to generate radio frequency signals based on the simulated signal data. The signal processing training module is configured to compare the signal generated by the signal generation module with the real signal to complete the training and optimization of deep neural network parameters. This invention solves the problems of low simulation fidelity and limited application scenarios caused by the high dependence of electromagnetic space construction on traditional ideal mathematical formulas.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of electronic testing technology, specifically relating to a test excitation signal strain generator based on autonomous matching of functional link resources. Background Technology

[0002] Complex electromagnetic signal simulation technology is a key technology in radar signal simulation, electronic countermeasures, and complex electromagnetic environment simulation, aiming to generate various electromagnetic signals applicable to different fields. Traditional digital domain simulation is based on ideal mathematical formulas for specific types of electromagnetic signals. It involves selecting formulas and parameter sets related to the target signal according to the tester's knowledge, generating simulation data under specific conditions, and then generating the electromagnetic signal through analog-to-digital conversion and I / Q modulation. This approximates the real signal under ideal assumptions. This method is widely used in the simulation of electromagnetic signals in various fields, especially in general-purpose signal generators. Although it solves the problem of generating various typical electromagnetic signal types, it cannot meet the requirements of high-fidelity signal simulation and efficient generation in complex scenarios due to limitations imposed by ideal mathematical formulas and the experience of testers.

[0003] Currently, the simulation of electromagnetic signals is typically based on ideal mathematical expressions, generating simulation data through various pre-set parameters. This technology is an approximation of real signals under ideal assumptions. Taking the simulation of electromagnetic radiation signals as an example, users generate waveform data through various parameters to simulate and generate signals, the principle of which is as follows: Figure 1 As shown, users can set relevant parameters on the control panel according to their needs. The software parses the set data set, compiles it to generate electromagnetic radiation signals of the specified type and parameters, and stores them in the system storage space as files for users to retrieve and call. Alternatively, the generated waveform data can be directly transmitted to a signal generator via LAN communication to achieve radio frequency output. It is evident that this solution simulates and generates data based on ideal mathematical formulas for specific types of electromagnetic signals, approximating the real signal under ideal assumptions. This leads to a large error between the simulated electromagnetic radiation signal and the real space, making evaluation inconvenient and failing to meet the relevant requirements for electromagnetic signal simulation. The types of parameters set through the panel are limited and fixed. If other types of excitation signals are to be generated, the entire test process must be repeated, relying entirely on manual calculation and modeling by the testers. This method is objectively limited by the fixed mechanism of ideal mathematical formulas and parameter sets, and is also greatly influenced by the knowledge level and experience of the testers, posing potential risks to the scientific rigor and efficiency of related testing work. Summary of the Invention

[0004] To address the aforementioned technical problems in the prior art, this invention provides a test excitation signal strain generator based on autonomous matching of functional link resources. The design is reasonable, overcomes the shortcomings of the prior art, and has excellent performance.

[0005] To achieve the purpose of the invention, the following technical solution is adopted:

[0006] A test excitation signal strain generator based on autonomous matching of functional link resources includes a deep neural network feature processing module, a signal generation module, and a signal processing training module.

[0007] The deep neural network feature processing module is configured to perform feature processing through a deep neural network based on Fourier expansion, establish a signal feature model library based on signal feature learning, generate signal simulation data through autonomous matching of test scenarios, and autonomously construct the feature implicit space of the signal.

[0008] The signal generation module is configured to generate radio frequency signals based on signal analog data;

[0009] The signal processing training module is configured to perform deep neural network parameter training and optimization by comparing the signal generated by the signal generation module with the real signal.

[0010] Furthermore, the device introduces a loss function with noise suppression effect, expressed as:

[0011]

[0012] in and These represent the feature input, real signal label, and network-generated signal of the deep neural network, respectively. Θ represents the set of all weights and bias parameters of the deep neural network. λ>0 is the smoothing factor of the loss function, used to control the smoothness of the loss function and suppress the influence of noise in the input features. β>0 is the regularization factor of the network parameters.

[0013] Furthermore, by training a deep neural network through a loss function, its hidden layers learn to achieve autonomous adaptation and matching of features, transforming the input features into the feature implicit space, and providing suitable feature input for the signal generation module.

[0014] The beneficial technical effects of this invention are as follows:

[0015] 1. This invention organically combines the broadband radio frequency signal generation mechanism with the deep neural network algorithm, processes and implicitly constructs the input features, establishes a signal feature model library based on signal feature learning, realizes high-precision simulation generation of complex electromagnetic signals, and breaks through the limitations of existing signal generator signal simulation methods based on ideal mathematical formulas of electromagnetic signals, which are restricted by the fixed mechanism and parameter set of ideal mathematical formulas. It can achieve better fidelity than signals in the actual test environment.

[0016] 2. This invention enables autonomous matching of functional link resources for multiple application scenarios, breaking through the limitation that existing signal generators rely heavily on the experience and operational accuracy of testers for signal simulation. It achieves autonomous learning and correction based on the signal feature library of actual environmental signals, thereby improving the simulation efficiency and accuracy of environmental signals. Attached Figure Description

[0017] Figure 1 This is a block diagram illustrating the principle of electromagnetic radiation signal simulation generation in existing technologies.

[0018] Figure 2 This is a schematic diagram of the test excitation signal strain generator based on autonomous matching of functional link resources in this invention.

[0019] Figure 3 This is a schematic diagram of the deep neural network based on the Fourier activation function in this invention. Detailed Implementation

[0020] The specific embodiments of the present invention will be further described below with reference to specific examples:

[0021] This invention provides a test excitation signal strain generator based on autonomous matching of functional link resources, such as... Figure 2 As shown, it includes a deep neural network feature processing module, a signal generation module, and a signal processing training module;

[0022] The deep neural network feature processing module is configured to perform feature processing through a deep neural network (DNN) based on Fourier expansion, establish a signal feature model library based on signal feature learning, generate signal simulation data through autonomous matching of test scenarios, and autonomously construct the implicit feature space of the signal.

[0023] The signal generation module is configured to generate radio frequency signals based on signal analog data;

[0024] The signal processing training module is configured to compare the signal generated by the signal generation module with the real signal to complete the training and optimization of the deep neural network parameters, thereby enabling the device to achieve autonomous learning and feature matching.

[0025] The key to neural networks processing input features lies in the application of Fourier mapping activation functions, which aligns with the sine and cosine representations in signal generation techniques. Based on the mathematical principles of deep neural networks, Fourier mapping-based neural networks simulate the Fourier expansion of arbitrarily complex functions, such as... Figure 3 As shown.

[0026] The deep neural network feature processing module employs fully connected artificial neural network technology to implicitly construct input features, providing the necessary features for the signal generation module. Generally, the components of the input features differ significantly in characteristics and scale. After Fourier mapping, these features are smoothed and scale-separated, facilitating nonlinear combination of these outputs by subsequent DNN modules and enhancing the nonlinear expressive power of the neural network. Furthermore, the signal generated by the input features is often affected by various unpredictable noises, which not only makes optimizing the loss function between the generated and real signals difficult but also causes underfitting, impacting network performance. Therefore, this invention introduces a loss function with noise suppression capabilities to improve the network's ability to filter unpredictable noise. The loss function is expressed as:

[0027]

[0028] in and These represent the feature input, real signal label, and network-generated signal of the deep neural network, respectively. Θ represents the set of all weights and bias parameters of the deep neural network. λ>0 is the smoothing factor of the loss function, used to control the smoothness of the loss function and suppress the influence of noise in the input features. β>0 is the regularization factor of the network parameters.

[0029] Deep neural networks are trained using a loss function. Their hidden layers learn to autonomously adapt and match features, transforming input features into a latent feature space to provide suitable feature input for the signal generation module. Through iterative training, the resulting network model can effectively process input features and automatically generate excitation signals for various application scenarios. Especially when feature input is incomplete, the network can automatically construct the input for the signal generation module, heuristically generating the signal pattern desired by the user. Furthermore, some features of complex electromagnetic signals generated by the adaptive network can also serve as input, providing training data for further optimization of the network model. This enables the network to possess self-learning capabilities, gradually improving the model's robustness and generalization performance.

[0030] The intelligent complex electromagnetic signal generation method based on functional link resources proposed in this invention realizes the iterative optimization process from feature to signal, and solves the problems of low simulation fidelity and limited application scenarios caused by the high dependence of electromagnetic space construction on traditional ideal mathematical formulas.

[0031] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A test excitation signal strain generator based on autonomous matching of functional link resources, characterized in that, It includes a deep neural network feature processing module, a signal generation module, and a signal processing training module; The deep neural network feature processing module is configured to perform feature processing through a deep neural network based on Fourier expansion, establish a signal feature model library based on signal feature learning, generate signal simulation data through autonomous matching of test scenarios, and autonomously construct the feature implicit space of the signal. The signal generation module is configured to generate radio frequency signals based on signal analog data; The signal processing training module is configured to perform deep neural network parameter training and optimization by comparing the signal generated by the signal generation module with the real signal. The device introduces a loss function with noise suppression effect, expressed as: ; in , and These represent the feature input, the real signal label, and the network-generated signal of the deep neural network, respectively. Represents the set of all weights and bias parameters of a deep neural network. It is a smoothing factor in the loss function, used to control the smoothness of the loss function and suppress the influence of noise in the input features. It is the regularization factor for network parameters; By training a deep neural network using a loss function, its hidden layers learn to autonomously adapt and match features, transforming the input features into a feature implicit space to provide feature input for the signal generation module.