CN-122027055-A - Intelligent frequency agility method and system based on frequency spectrum cognition and deep learning

CN122027055ACN 122027055 ACN122027055 ACN 122027055ACN-122027055-A

Abstract

The invention discloses an intelligent frequency agility method and system based on frequency spectrum cognition and deep learning, wherein the method comprises the steps of generating a plurality of baseband linear frequency modulation signals with different center frequencies to obtain a broadband radio frequency linear frequency modulation signal set; the method comprises the steps of inputting signals into a digital phased array receiving channel simultaneously to obtain a plurality of paths of baseband data sets, carrying out integral multiple clock period delay on each signal data in the baseband data sets according to the real-time total delay of the receiving channel to obtain a delay linear frequency modulation signal set, determining the peak position of each data in a related peak data set according to the plurality of paths of baseband data sets and the delay linear frequency modulation signal set through a peak searching algorithm model to further calculate a phase difference set, and removing first-order linear components and direct-current components in the phase difference set by adopting a declivity algorithm to obtain a phase nonlinear set of the digital phased array receiving channel. The invention can realize nonlinear, efficient, accurate and low-cost measurement of the digital phased array receiving channel phase.

Inventors

REN BO
WAN JIAHUI
ZHANG JINPING

Assignees

南京华成微波技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260326

Claims (10)

1. An intelligent frequency agility method based on spectrum cognition and deep learning, which is characterized by comprising the following steps: Capturing radio frequency signals of the current electromagnetic environment in real time through a broadband receiver, and performing time-frequency analysis processing on the radio frequency signals to generate real-time frequency spectrum waterfall diagram data; inputting the spectrum waterfall diagram data into a pre-trained convolutional neural network model, performing feature extraction and classification recognition on a spectrum diagram by using the convolutional neural network model, and outputting a signal type recognition result and signal occupation state probability distribution in a current spectrum environment; constructing a frequency spectrum cost function based on the signal type identification result and the signal occupation state probability distribution, and calculating the frequency spectrum agility value of each frequency point in the candidate frequency point set; According to the frequency spectrum agility cost value, selecting a frequency point with the minimum cost value from the candidate frequency point set as a target agility frequency point, and controlling a radio frequency front end to be switched to the target agility frequency point for communication; And monitoring the quality of a communication link in the communication process, and carrying out online updating or parameter fine adjustment on the convolutional neural network model according to a monitoring result.
2. The intelligent frequency agility method based on spectrum cognition and deep learning according to claim 1, wherein the feature extraction and classification recognition are performed on a spectrogram by using the convolutional neural network model, and a signal type recognition result and a signal occupation state probability distribution in a current spectrum environment are output, and the method comprises the following steps: Normalizing the frequency spectrum waterfall diagram data into a two-dimensional image matrix, and inputting the two-dimensional image matrix into an input layer of the convolutional neural network model; Local feature extraction is carried out through a plurality of convolution layers of the convolution neural network model, and an activation function layer and a pooling layer are connected behind each convolution layer so as to extract time-frequency texture features of a spectrum signal; mapping the extracted features to a signal classification space through a full connection layer, and outputting multi-classification probability vectors including communication signals, broadcast signals, background noise and interference signals; And meanwhile, outputting the occupied energy probability value of each frequency point in a preset time window through a regression output branch of the convolutional neural network model.
3. The intelligent frequency agility method based on spectrum cognition and deep learning according to claim 1, wherein the constructing a spectrum cost function based on the signal type recognition result and the signal occupation state probability distribution, and calculating the spectrum agility cost value of each frequency point in the candidate frequency point set comprises: Defining a spectrum fusion index for indicating the similarity degree of the signal characteristics of the candidate frequency points and the background noise statistical characteristics; Defining an anti-interference coverage index for indicating the risk probability that the candidate frequency points are predicted and covered by the enemy intelligent jammer; The frequency spectrum cost function is a weighted sum of the negative weight of the frequency spectrum fusion index and the positive weight of the anti-interference coverage index; the spectrum fusion index is obtained by calculating the divergence distance between the signal power spectrum density distribution at the candidate frequency point and the background noise power spectrum density distribution.
4. The intelligent frequency agility method based on spectrum cognition and deep learning of claim 3, wherein the method for obtaining the anti-interference coverage index comprises: Judging the working mode of the jammer by utilizing the type of the jammer signal identified by the convolutional neural network model; If the jammer is tracking type jammer, predicting the range of the jammer frequency point at the next moment according to the historical jammer track; if the jammer is the blocking type interference, calculating the energy duty ratio of the candidate frequency point in the blocking frequency band; and mapping the predicted energy duty ratio in the interference frequency point range or the blocking frequency band into risk probability as the anti-interference coverage index.
5. The intelligent frequency agility method based on spectrum cognition and deep learning of claim 1, wherein the online updating or parameter fine tuning of the convolutional neural network model according to the monitoring result comprises: recording the communication error rate, the signal-to-interference-and-noise ratio and the link interruption times on the target frequency agility point; If the communication error rate or the link interruption times exceed a preset threshold value, marking the current frequency spectrum waterfall diagram data as a difficult sample; Adding the difficult sample into a training data set, and carrying out local gradient descent updating on the weight parameters of the convolutional neural network model by utilizing a transfer learning algorithm; And if the communication link quality is normal, freezing the model parameters, and only updating the frequency spectrum environment statistical database.
6. An intelligent frequency agility system based on spectrum cognition and deep learning, comprising: The frequency spectrum data acquisition module is used for capturing radio frequency signals of the current electromagnetic environment in real time through the broadband receiver, performing time-frequency analysis processing on the radio frequency signals and generating real-time frequency spectrum waterfall diagram data; The intelligent cognitive processing module is used for inputting the frequency spectrum waterfall diagram data into a pre-trained convolutional neural network model, extracting features and classifying and identifying the frequency spectrum diagram by utilizing the convolutional neural network model, and outputting a signal type identification result and a signal occupation state probability distribution in the current frequency spectrum environment; The agility decision calculation module is used for constructing a frequency spectrum cost function based on the signal type identification result and the signal occupation state probability distribution and calculating the frequency spectrum agility value of each frequency point in the candidate frequency point set; the frequency execution control module is used for selecting a frequency point with the minimum cost value from the candidate frequency point set as a target agile frequency point according to the frequency spectrum agile cost value, and controlling the radio frequency front end to be switched to the target agile frequency point for communication; and the model online optimization module is used for monitoring the quality of a communication link in the communication process and carrying out online updating or parameter fine adjustment on the convolutional neural network model according to a monitoring result.
7. The intelligent frequency agile system based on spectrum cognition and deep learning of claim 6, wherein the intelligent cognition processing module comprises: The data preprocessing unit is used for normalizing the frequency spectrum waterfall diagram data into a two-dimensional image matrix and inputting the two-dimensional image matrix into an input layer of the convolutional neural network model; The characteristic extraction unit is used for extracting local characteristics through a plurality of convolution layers of the convolution neural network model, and an activation function layer and a pooling layer are connected behind each convolution layer so as to extract time-frequency texture characteristics of the spectrum signals; the classification and identification unit is used for mapping the extracted characteristics to a signal classification space through the full connection layer and outputting multi-classification probability vectors including communication signals, broadcast signals, background noise and interference signals; The state regression unit is used for outputting the occupied energy probability value of each frequency point in a preset time window through the regression output branch of the convolutional neural network model.
8. The intelligent frequency agility system based on spectrum cognition and deep learning of claim 6, wherein the agility decision computing module comprises: The fusion degree calculation unit is used for defining a spectrum fusion degree index and representing the similarity degree of the candidate frequency point signal characteristics and the background noise statistical characteristics; the risk prediction unit is used for defining an anti-interference coverage index and representing the risk probability that the candidate frequency points are predicted and covered by the enemy intelligent jammer; the cost function construction unit is used for taking the weighted sum of the negative weight of the spectrum fusion index and the positive weight of the anti-interference coverage index as the spectrum cost function; The fusion degree calculating unit is specifically used for calculating the divergence distance between the signal power spectrum density distribution at the candidate frequency points and the background noise power spectrum density distribution.
9. The intelligent frequency agility system based on spectrum cognition and deep learning of claim 8, wherein the risk prediction unit is specifically configured to: Judging the working mode of the jammer by utilizing the type of the jammer signal identified by the convolutional neural network model; If the jammer is tracking type jammer, predicting the range of the jammer frequency point at the next moment according to the historical jammer track; if the jammer is the blocking type interference, calculating the energy duty ratio of the candidate frequency point in the blocking frequency band; and mapping the predicted energy duty ratio in the interference frequency point range or the blocking frequency band into risk probability as the anti-interference coverage index.
10. The intelligent frequency agility system based on spectrum cognition and deep learning of claim 6, wherein the model online optimization module is specifically configured to: recording the communication error rate, the signal-to-interference-and-noise ratio and the link interruption times on the target frequency agility point; If the communication error rate or the link interruption times exceed a preset threshold value, marking the current frequency spectrum waterfall diagram data as a difficult sample; Adding the difficult sample into a training data set, and carrying out local gradient descent updating on the weight parameters of the convolutional neural network model by utilizing a transfer learning algorithm; And if the communication link quality is normal, freezing the model parameters, and only updating the frequency spectrum environment statistical database.

Description

Intelligent frequency agility method and system based on frequency spectrum cognition and deep learning Technical Field The invention relates to the technical field of wireless communication, in particular to an intelligent frequency agility method and system based on spectrum cognition and deep learning. Background In the field of intelligent frequency agility measurement based on spectrum cognition and deep learning, the existing technical scheme mainly comprises the following common methods: Conventional frequency agility methods are typically based on pseudo random sequence control. The method controls the carrier frequency to jump among a plurality of frequency points according to a preset rule through a preset frequency hopping pattern. However, it has significant limitations in application to interference-free communications in complex electromagnetic environments. The pseudo-random sequence has certain randomness, but the generation algorithm is usually fixed, and once the enemy electronic reconnaissance equipment intercepts the period of the frequency hopping pattern or the generation algorithm through long-time monitoring, the following interference or predictive interference can be implemented, so that the communication link is interrupted. In addition, the method has low test efficiency, needs to preset a large number of frequency hopping patterns, and has high cost, because a complex synchronization mechanism is needed to ensure the consistency of frequencies of the two transceivers. Another prior art is the cognitive radio frequency agility method based on energy detection. This approach typically requires a high sensitivity spectrum sensor and fast decision logic. The method comprises the steps of firstly, detecting the energy of a frequency spectrum, and judging to be idle and accessed when the energy of a certain frequency point is lower than a threshold value. However, this approach requires high hardware requirements, not only a very high dynamic range of the sensor, but also a powerful processor to process the spectral data in real time. This not only increases the complexity of the system, but also increases the cost. In addition, because the signal distinguishing capability requirement on the system is high only based on energy detection, otherwise, extra measurement errors are introduced, background noise, legal communication signals and malicious interference signals cannot be distinguished, and the interference frequency band is easy to be wrongly entered. Yet another prior art is a spectrum sensing method based on simple machine learning. This approach utilizes traditional machine learning algorithms (e.g., support vector machines, decision trees, etc.) to classify spectral features. However, this method requires manual extraction of features, and the quality of feature extraction directly determines the classification effect. In complex electromagnetic environments, signal features are often difficult to define manually, resulting in poor classification accuracy. In addition, the traditional machine learning algorithm is difficult to process high-dimensional frequency spectrum data, has poor generalization capability and is difficult to adapt to the continuously-changing electromagnetic environment. Therefore, when the intelligent frequency agility based on spectrum cognition and deep learning is measured in the prior art, the problems that the frequency hopping rule is easy to predict, the spectrum cognition granularity is coarse, the anti-interference strategy is single, the requirement on hardware is high, the dynamic environment is difficult to adapt to and the like exist, and the requirements for efficient, accurate and low-cost measurement on the intelligent frequency agility based on spectrum cognition and deep learning in practical application are difficult to meet. Disclosure of Invention The embodiment of the invention provides an intelligent frequency agility method and system based on spectrum cognition and deep learning, which can realize the intelligent frequency agility efficient, accurate and low-cost measurement based on spectrum cognition and deep learning. An embodiment of the present invention provides an intelligent frequency agility method based on spectrum cognition and deep learning, including: Capturing radio frequency signals of the current electromagnetic environment in real time through a broadband receiver, and performing time-frequency analysis processing on the radio frequency signals to generate real-time frequency spectrum waterfall diagram data; inputting the spectrum waterfall diagram data into a pre-trained convolutional neural network model, performing feature extraction and classification recognition on a spectrum diagram by using the convolutional neural network model, and outputting a signal type recognition result and signal occupation state probability distribution in a current spectrum environment; constructing a frequency spectrum cost f