CN115065973A - Convolutional neural network-based satellite measurement and control ground station identity recognition method - Google Patents

Abstract

The embodiment of the invention discloses a convolutional neural network-based satellite measurement and control ground station identity recognition method, which comprises the steps of acquiring a radio frequency signal sent by equipment to be recognized, converting the radio frequency signal into a baseband signal, preprocessing the baseband signal to determine a target signal, and inputting the target signal into a pre-trained neural network model for recognition to determine a recognition result. Therefore, the radio frequency signal is converted into the baseband signal, and the baseband signal is identified according to the preprocessed baseband signal and the pre-trained neural network model, so that the key information of the radio frequency signal is kept, the accuracy of the identification result is improved, meanwhile, the model identification calculated amount can be reduced, and the identification efficiency is improved.

Description

A method for identification of satellite measurement and control ground stations based on convolutional neural network

technical field

The invention relates to the field of computer technology, in particular to a method for identifying a satellite measurement and control ground station based on a convolutional neural network.

Background technique

The radio frequency fingerprint identification technology analyzes the radio frequency signal of the wireless communication device, extracts the radio frequency fingerprint of the signal to identify the device, and realizes the identity authentication of the physical layer, thereby improving the security of the wireless communication.

The traditional radio frequency fingerprint identification method realizes identity verification by artificially selecting the features in the radio frequency signal and then designing a classifier. In addition, although the radio frequency identification method based on deep learning exceeds the traditional radio frequency fingerprint identification method in generality and identification accuracy, due to the large number of neural network parameters, the calculation amount is large, and the identification efficiency is low.

SUMMARY OF THE INVENTION

In view of this, the embodiments of the present invention provide a method for identifying a satellite measurement and control ground station based on a convolutional neural network, so as to improve the accuracy and efficiency of device identification results.

In a first aspect, an embodiment of the present invention provides a device identification method, the method includes:

Obtain the radio frequency signal sent by the device to be identified;

converting the radio frequency signal into a baseband signal;

preprocessing the baseband signal to determine a target signal;

The target signal is input into a pre-trained neural network model for identification to determine the identification result.

Further, converting the radio frequency signal into a baseband signal includes:

Down-conversion processing is performed on the radio frequency signal, and the down-converted in-phase quadrature signal is determined as the baseband signal.

Further, the neural network model includes:

a feature unit, used for dimensionality reduction and feature extraction on the target signal;

a convolution unit, the convolution unit includes a first preset number of convolution layers, for performing convolution processing on the feature information extracted by the feature unit;

A classification unit, the classification unit includes a second preset number of fully connected layers, and is used for classifying the processed feature information to determine a recognition result.

Further, the first preset number and the second preset number are determined based on a pre-training process corresponding to the neural network model.

Further, the baseband signal includes an in-phase baseband signal and a quadrature baseband signal, and the preprocessing of the baseband signal to determine the target signal includes:

Determine a steady-state segment signal of the baseband signal based on an energy detection method, where the steady-state segment signal includes an in-phase steady-state signal and a quadrature steady-state signal;

respectively normalizing the in-phase steady state signal and the quadrature steady state signal;

Perform data segmentation processing on the normalized in-phase steady-state signal and quadrature steady-state signal, and determine the in-phase target sub-signal and quadrature target sub-signal of at least one time period;

The in-phase target sub-signal and the quadrature target sub-signal in the same time period are respectively combined to determine the target signal in the form of two-dimensional data.

Further, the device to be identified is a measurement and control ground station, and the neural network model is deployed on a satellite terminal.

Further, the neural network model is determined based on the following steps:

get raw data;

Determine a training set, a verification set and a test set according to the original data;

inputting the data in the training set into a preset neural network model for training, and determining at least one set of model parameters;

Input the data corresponding to the verification set into the neural network models corresponding to each group of model parameters, so as to determine the target model parameters according to the recognition results corresponding to each neural network model;

Input the data corresponding to the test set into the neural network model corresponding to the target model parameter, and determine the accuracy of the current recognition result;

In response to the accuracy rate of the current recognition result satisfying a preset condition, the neural network model corresponding to the target model parameter is determined as the trained neural network model.

In a second aspect, an embodiment of the present invention provides a device identification device, the device comprising:

an acquisition unit, configured to acquire the radio frequency signal sent by the device to be identified;

a conversion unit for converting the radio frequency signal into a baseband signal;

a preprocessing unit for preprocessing the baseband signal to determine a target signal;

The identification unit is used for inputting the target signal into the pre-trained neural network model for identification, so as to determine the identification result.

In a third aspect, an embodiment of the present invention provides an electronic device, including a memory and a processor, where the memory is configured to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor Execute to implement the method as described in any of the above.

In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the method steps described in any of the above are implemented.

The technical solution of this embodiment converts the radio frequency signal sent by the device to be identified into a baseband signal for identification, and makes use of the unforgeable feature of the modulation domain characteristics of the physical layer of the signal. While retaining the key information of the radio frequency signal, it can avoid the identification process. Deciphered, counterfeited and influenced by environmental factors, thereby improving the accuracy of identification results. Furthermore, the pre-trained neural network model is used to process the baseband signal corresponding to the preprocessed radio frequency signal. In the process of determining the identity of the device to be identified, the identification result can be determined without manually extracting features, and the identification efficiency is higher.

Description of drawings

The above and other objects, features and advantages of the present invention will become more apparent from the following description of embodiments of the present invention with reference to the accompanying drawings, in which:

1 is a schematic diagram of a device identification method according to an embodiment of the present invention;

2 is a flowchart of determining a target signal according to an embodiment of the present invention;

3 is a schematic diagram of determining a target signal according to an embodiment of the present invention;

4 is a schematic diagram of a neural network model structure according to an embodiment of the present invention;

5 is a flowchart of determining a neural network model according to an embodiment of the present invention;

FIG. 6 is a flowchart of the specific implementation of the device identification method according to the embodiment of the present invention;

7 is a schematic diagram of a device identification device according to an embodiment of the present invention;

FIG. 8 is a schematic diagram of an electronic device according to an embodiment of the present invention.

Detailed ways

The present invention is described below based on examples, but the present invention is not limited to these examples only. In the following detailed description of the invention, some specific details are described in detail. The present invention can be fully understood by those skilled in the art without the description of these detailed parts. Well-known methods, procedures, procedures, components and circuits have not been described in detail in order to avoid obscuring the essence of the present invention.

Furthermore, those of ordinary skill in the art will appreciate that the drawings provided herein are for illustrative purposes and are not necessarily drawn to scale.

Unless clearly required by the context, words such as "including", "comprising" and the like in the specification should be construed in an inclusive rather than an exclusive or exhaustive sense; that is, in the sense of "including but not limited to".

In the description of the present invention, it should be understood that the terms "first", "second" and the like are used for descriptive purposes only, and should not be construed as indicating or implying relative importance. Also, in the description of the present invention, unless otherwise specified, "plurality" means two or more.

The tolerance of electronic devices in the production process will lead to subtle differences in the signals generated by different wireless communication devices. Even the same type of wireless communication devices from the same manufacturer will have a certain degree of difference due to the tolerance effect, and it is difficult to forge. characteristics, these differences form the unique RF fingerprint of the transmitter.

At present, the traditional radio frequency fingerprint identification method needs to manually select the features and then design the classifier. The accuracy of the identification results depends on the expert domain knowledge mastered by the operator. The degree of intelligence is low, and the identification accuracy is poor. In addition, the traditional radio frequency fingerprint identification method is greatly affected by the channel environment. In practical application scenarios, it is necessary to manually select features and design classifiers for specific environmental conditions, and the effect is poor when the environment changes.

With the development of big data and deep learning, the radio frequency fingerprinting method based on deep learning has surpassed the traditional method based on artificial features in both method versatility and recognition accuracy. At present, some researches have proposed a radio frequency fingerprinting method combining constellation map and deep learning, and a radio frequency fingerprinting method combining transform domain features and deep learning, but these methods have the problems of many neural network parameters and large amount of calculation, which limit their application.

In view of this, embodiments of the present invention provide a method for identifying a satellite measurement and control ground station based on a convolutional neural network, so as to improve the identification efficiency and accuracy of device identification.

The following description will be given by taking the radio frequency fingerprint identification process of the measurement and control ground station as an example. It should be understood that the method in this embodiment can also be applied to other scenarios where the identification is performed by radio frequency signals, which is not limited here.

FIG. 1 is a schematic diagram of a device identification method according to an embodiment of the present invention. As shown in FIG. 1 , the device identification method of this embodiment includes the following steps.

In step S110, the radio frequency signal sent by the device to be identified is acquired.

In this embodiment, the device to be identified may be a measurement and control ground station, or may be other devices that transmit radio frequency signals. Taking the measurement and control ground station as an example, the radio frequency signal sent by the device to be identified is obtained by collecting the radio frequency signal sent by the measurement and control ground station to the satellite terminal. Since the measurement and control ground station will continuously send radio frequency signals to the satellite terminal, in order to improve the identification efficiency, in this embodiment, the received continuous radio signals will be periodically sampled based on the sampling method to obtain the radio frequency signals sent by the device to be identified.

It should be understood that, in other optional implementation manners, a wireless signal for a continuous period of time may also be collected as a target radio frequency signal to perform subsequent processing.

Optionally, in this embodiment, by sampling the radio frequency signal sent by the measurement and control ground station according to the preset sampling time at a predetermined time, sampling the sampling signals of the preset sampling points, and sorting all the collected sampling signals in chronological order. Arrange to form a time series, and then obtain the radio frequency signal to be identified corresponding to the current measurement and control ground station.

Specifically, assuming that the preset sampling time is T, the preset number of sampling points is m, and each sampling corresponds to a segment of signal x(i), where i=1, 2, ···, m, then it is determined that the The radio frequency signal is X=[x(1), x(2), ···, x(m)].

Further, the duration corresponding to the preset sampling time and the number of preset sampling points in this embodiment are set and adjusted according to the actual usage scenario and the radio frequency fingerprint identity of the device to be identified itself. For example, when the radio frequency fingerprint identity of the device to be recognized contains a large amount of information, the duration corresponding to the preset sampling time may be longer, and the number of preset sampling points may be larger to further improve the accuracy of the identification result.

In step S120, the radio frequency signal is converted into a baseband signal.

In this embodiment, the baseband signal adopts an IQ signal (In-phase Quadrature, in-phase quadrature signal). Since the baseband signal carries modulation domain features such as IQ imbalance and phase offset, these feature information can still be maintained in the case of a large environmental impact, thereby avoiding the influence of environmental factors on the accuracy of the recognition results and improving the accuracy of the recognition results. . Based on this, in this embodiment, the radio frequency signal is down-converted, and the down-converted in-phase quadrature signal is determined as a baseband signal.

Down-conversion processing is also known as digital down-conversion (Digital Down Converters, DDC). It consists of mixing the intermediate frequency (IF) digital signal collected by AD with the local digital intermediate frequency carrier signal generated by the digitally controlled oscillator (NCO) in the DDC, and then obtains the baseband signal through a low-pass filter to realize the down-conversion function. The basic principle of digital down-conversion is the same as that of analog down-conversion, which is to multiply the input signal by the local oscillator signal, move the RF signal to the mid-frequency band through mixing, and then perform ADC sampling. The core process is to mix the intermediate frequency (IF) digital signal collected by the AD with the local digital intermediate frequency carrier signal generated by the digitally controlled oscillator (NCO) in the DDC, and down-convert the intermediate frequency signal to the baseband.

Further, in this embodiment, by performing down-conversion processing on the radio frequency signal, two baseband signals including an in-phase baseband signal and a quadrature baseband signal are simultaneously obtained. Therefore, by converting the radio frequency signal sent by the device to be identified into two-channel baseband signals, other transformation calculations can be reduced, thereby reducing the amount of calculation in the subsequent signal identification process, and improving the identification efficiency.

In step S130, the baseband signal is preprocessed to determine the target signal.

In this embodiment, the baseband signal includes an in-phase baseband signal and a quadrature baseband signal.

Optionally, as shown in FIG. 2 , the preprocessing of the baseband signal to determine the target signal in this embodiment includes the following steps.

In step S210, a steady-state segment signal of the baseband signal is determined based on the energy detection method. The steady-state segment signal includes an in-phase steady-state signal and a quadrature steady-state signal.

The energy detection method is a signal detection method, which uses the operation of modulo and squaring the sampled value of the time domain signal; Square to get the energy accumulation value of the signal in a certain frequency band. The comparison result is determined by comparing the energy accumulation value with the preset threshold value, and the existence state of the signal is determined according to the comparison result. In this embodiment, if the energy accumulation value of the signal is higher than the preset threshold value, it indicates that the signal is in the steady state segment.

In step S220, the in-phase steady state signal and the quadrature steady state signal are respectively normalized.

There are two forms of normalization, one is to change the number into a decimal between (0, 1), the other is to change the dimensional expression into a dimensionless expression. The data is mapped to the range of 0 to 1 through normalization processing, which can make the digital signal processing process more convenient and fast.

In step S230, data segmentation processing is performed on the normalized in-phase steady-state signal and the quadrature steady-state signal to determine the in-phase target sub-signal and the quadrature target sub-signal for at least one period of time.

In this embodiment, the in-phase steady-state signal and the quadrature steady-state signal are intercepted at a certain number of sampling points, respectively, so as to implement segmentation processing and determine the in-phase target sub-signal and quadrature target sub-signal for at least one period.

In step S240, the in-phase target sub-signals and the quadrature target sub-signals of the same time period are respectively combined to determine the target signal in the form of two-dimensional data.

In this embodiment, both the in-phase target sub-signal and the quadrature target sub-signal after the split processing correspond to one-dimensional data, and the data in the form of a two-dimensional matrix is formed by combining the in-phase target sub-signal and the quadrature target sub-signal of the same time period , and the combined two-dimensional matrix data is determined as the preprocessed target signal.

FIG. 3 is a schematic diagram of determining a target signal according to an embodiment of the present invention. As shown in Figure 3, the I channel signal is an in-phase signal, and the Q channel signal is a quadrature signal. In this embodiment, the in-phase steady-state signal and the quadrature steady-state signal are intercepted every M sampling points to obtain the divided in-phase target sub-signal and quadrature target sub-signal. After that, the signals corresponding to the same time period in the in-phase target sub-signal and the quadrature target sub-signal are combined to determine the target signal in the form of two-dimensional data.

For ease of understanding, as shown in Figure 3, it is assumed that the in-phase steady-state signal and the quadrature steady-state signal are intercepted every M sampling points, and the segmented in-phase target sub-signal and quadrature target sub-signal each include 5-segment signals. , and each segment of the signal includes M sampled signals. After combining the signals corresponding to the same time period in the in-phase target sub-signal and the quadrature target sub-signal, five target signals in the form of two-dimensional data are determined, including target signals S1, S2, S3, S4 and S5.

Optionally, in this embodiment, after the in-phase target sub-signal and the quadrature target sub-signal are combined, MATLAB or Python software is used to perform data enhancement on the combined signal data to increase the unique Doppler frequency shift of the measurement and control signal, Fading and other effects, and the enhanced signal is determined as the target signal. Therefore, by adopting the data enhancement method, the effective features in the target signal corresponding to the device to be identified are more obvious and easy to identify, and the accuracy of the identification result can be further improved.

In step S140, the target signal is input into the pre-trained neural network model for identification, so as to determine the identification result. The identification result is used to characterize the identity of the device to be identified.

In this embodiment, the neural network model is deployed on the satellite terminal. The identification result may be a label used to characterize the identity of the device to be identified. Thus, the radio frequency signal sent by the device to be identified is received through the satellite terminal, and the radio frequency signal sent by the device to be identified is obtained according to the received radio frequency signal, the radio frequency signal is converted into a baseband signal, and the baseband signal is preprocessed to determine The target signal is input into the pre-trained neural network model for identification, and then the identity of the device to be identified is determined.

Optionally, the satellite terminal in this embodiment can only receive the radio frequency signal sent by the legitimate measurement and control ground station, and identify the received radio frequency signal through the above processing process to determine the identity of the device to be identified.

Alternatively, the satellite terminal in this embodiment may receive radio frequency signals sent by various types of measurement and control ground stations, such as legal measurement and control ground stations and illegal measurement and control ground stations. Therefore, after the identity of the device to be identified is determined, the identification result is also judged in this embodiment. Specifically, the identification result output by the neural network model is matched with the legal ground station radio frequency fingerprint database pre-arranged on the satellite terminal. When the legal ground station radio frequency fingerprint database includes the radio frequency fingerprint corresponding to the measurement and control ground station represented by the identification result, It is determined that the measurement and control ground station corresponding to the identification result is a legal ground station. In this way, by matching the identification result with the radio frequency fingerprint database of the legal ground station, the validity of the identity of the measurement and control ground station can be determined.

The technical solution of this embodiment converts the radio frequency signal sent by the device to be identified into a baseband signal for identification, and makes use of the unforgeable feature of the modulation domain feature of the physical layer of the signal, which can avoid being deciphered, counterfeited and affected by environmental factors during the identification process. impact, thereby improving the accuracy of the recognition results. Furthermore, the pre-trained neural network model is used to process the baseband signal corresponding to the preprocessed radio frequency signal. In the process of determining the identity of the device to be identified, the identification result can be determined without manually extracting features, and the identification efficiency is higher. In addition, by optimizing the model structure and model parameters of the preset neural network, the amount of calculation in the identification process can be reduced, which further facilitates the deployment of the neural network model and improves the identification efficiency.

FIG. 4 is a schematic diagram of the structure of a neural network model according to an embodiment of the present invention. As shown in FIG. 4 , the neural network model structure in this embodiment includes a feature unit 1 , a convolution unit 2 and a classification unit 3 . Among them, the feature unit 1 is used to perform dimension reduction and feature extraction on the target signal. The convolution unit 2 includes a first preset number of convolution layers for performing convolution processing on the feature information extracted by the feature unit. The classification unit 3 includes a second preset number of fully connected layers for classifying the processed feature information to determine a recognition result. The identification result is used to represent the identity of the device to be identified, and the identity of the device to be identified may be represented by an identifier in the form of a label.

Optionally, the feature unit in this embodiment adopts a Conv(1, 2) convolution layer. Since the target signal includes two pieces of information, the in-phase target sub-signal and the quadrature target sub-signal, the input format of the neural network model in this embodiment is N*2. The feature unit 1 receives the target signal corresponding to the device to be identified in the form of N*2, and performs dimension reduction and IQ-related feature extraction on the data corresponding to the target signal by using a convolution layer with a convolution kernel size of (1, 2). The IQ-related features include the phase relationship and the amplitude relationship between the I-channel signal and the Q-channel signal in the target signal, and the like. Therefore, after the feature unit processes the target signal, it is convenient to subsequently extract different local features in the target signal through the first preset number of convolutional layers, and use the second preset number of fully connected layers to extract all local features. It becomes a global feature, and then the recognition result is determined according to the global feature.

Optionally, the first preset number and the second preset number in this embodiment are determined based on a pre-training process corresponding to the neural network model. Further, in this embodiment, the parameters in the model are tested by the control variable method, and the optimal parameters are determined as much as possible while fully considering the relationship between model parameters, recognition accuracy and robustness, so as to improve the accuracy of the output results of the neural network model. Accuracy, recognition efficiency and recognition stability.

Specifically, the specific values of the first preset number and the second preset number in this embodiment are the respective minimum values when both the accuracy of the recognition result and the recognition efficiency are taken into consideration. For example, if the output result of the feature unit is processed by X convolutional layers and Y fully connected layers, the recognition result accuracy and recognition efficiency can reach the preset value, or if the recognition result accuracy, recognition efficiency and network robustness are taken into account At the same time, the overall performance can reach the global optimum, the first preset number is X, and the second preset number is Y. Therefore, determining the number of convolution layers in the convolution unit and the fully connected layers in the classification unit through the above process can reduce the amount of model parameters in the neural network model, and reduce the amount of model calculation while ensuring the nonlinear mapping capability of features. .

Further, the convolution layer in the convolution unit in this embodiment adopts a small convolution kernel smaller than the preset convolution kernel size, so as to further reduce the amount of network parameters in the neural network model, which is beneficial to further improve the recognition efficiency. Among them, the specific value of the preset convolution kernel size is adjusted according to the requirements of the actual use scenario.

Further, as shown in FIG. 5 , in this embodiment, the neural network model is determined through the following steps.

In step S310, raw data is acquired.

In this embodiment, the satellite terminal receives radio frequency signals sent by multiple measurement and control ground stations, and obtains original data by sampling each radio frequency signal. The original data includes multiple sample data sets, and one measurement and control ground station corresponds to one sample data set.

Assuming that there are n measurement and control ground stations in total, when sampling the radio frequency signal, sampling is performed once every time T, and a total of m times are sampled. A sample data set Ai=[x(1), x(2), ···, x(m)] corresponding to the ground station, where i=1, 2, ···, n.

In step S320, a training set, a validation set and a test set are determined according to the original data.

In this embodiment, after the digital receiver of the satellite terminal obtains the original data, it first converts the radio frequency signal in each sample into a baseband signal through down-conversion processing, and then preprocesses the baseband signal corresponding to each sample to determine the corresponding baseband signal. target signal. Finally, all the target signals are divided to determine the training set, validation set and test set respectively.

Optionally, when preprocessing each baseband signal in this embodiment, the in-phase steady-state signal and the quadrature steady-state signal in the corresponding baseband signal are determined based on the energy detection method, and the in-phase steady-state signal and the quadrature steady-state signal are respectively analyzed. The signal is normalized, and then the normalized in-phase steady-state signal and quadrature steady-state signal are segmented based on a preset number of sampling points to determine the in-phase target sub-signal and the normal-phase target sub-signal in each period. , and finally combine the in-phase target sub-signals and quadrature target sub-signals of the same time period respectively to determine the target signal in the form of two-dimensional data.

Optionally, in this embodiment, after combining the in-phase target sub-signal and the quadrature target sub-signal into a two-dimensional matrix, the two-dimensional matrix is labeled with a sample to determine the target signal of the sample. Therefore, in this embodiment, the corresponding relationship between the target signal and the corresponding device is established by labeling, so as to adjust the model parameters based on the output result of the neural network model and the label information during the training process.

In step S330, the data in the training set is input into a preset neural network model for training, and at least one set of model parameters is determined.

In step S340, the data corresponding to the verification set is input into the neural network models corresponding to each group of model parameters, so as to determine target model parameters according to the recognition results corresponding to each neural network model.

In this embodiment, if the error between the recognition result outputted after the input data of the validation set is input into the neural network model and the output data corresponding to the input data of the validation set is within a preset range, it is determined that the validation of the neural network model under the current model parameters is successful. . After the verification is successful, the current model parameters are determined as the target model parameters.

In step S350, the data corresponding to the test set is input into the neural network model corresponding to the target model parameters, and the accuracy of the current recognition result is determined.

It should be understood that when training the neural network model through the corresponding data in the training set in this embodiment, a set of model parameters can be obtained in each training round, and the output results of the neural network model corresponding to each set of model parameters after inputting the corresponding data in the test set. Determine the target model parameters, and then determine the accuracy of the recognition results of the neural network models corresponding to each group of target model parameters through the validation set.

Optionally, in this embodiment, all target model parameter combinations can be obtained by training first, and then each group of target model parameters can be verified respectively; or a set of target model parameter combinations can be obtained by training, so as to determine the recognition accuracy of the group of target model parameters. , and then continue to train to obtain other target model parameter combinations, and determine the recognition accuracy corresponding to each group of target model parameters.

In step S360, in response to the accuracy rate of the current recognition result satisfying the preset condition, the neural network model corresponding to the target model parameter is determined as the trained neural network model.

Optionally, in this embodiment, the target model parameter combination with the highest recognition result accuracy rate is determined as the preset model parameter, and the preset model parameter is applied to the preset neural network model, and then the trained neural network model is determined. , and identify the radio frequency signal of the device to be identified through the trained neural network model to determine the identity of the device to be identified.

Optionally, in this embodiment, the target model parameter combination with the recognition result accuracy rate greater than the preset accuracy rate may also be determined as the preset model parameter. When there is a set of preset model parameters, the preset model parameters are applied to the preset neural network model, and then the trained neural network model is determined, and the identity of the device to be identified is determined through the trained neural network model, In order to improve the accuracy of identity recognition; when there are multiple sets of target model parameters whose accuracy satisfies the preset conditions, a set of model parameters with smaller model parameters, higher accuracy and better model robustness is preferentially selected as the final model. The preset model parameters are determined, and the preset model parameters are applied to the preset neural network model to identify the device to be identified, so as to improve the practical application performance of the identification method while ensuring the identification accuracy.

FIG. 6 is a flowchart when the device identification method according to the embodiment of the present invention is specifically implemented. As shown in FIG. 6 , in this embodiment, the identification result of the device to be identified is determined through the following steps.

In step S410, raw data is acquired.

In step S420, a training set, a validation set and a test set are determined according to the original data.

In step S430, the preset neural network model is trained, verified and tested based on the training set, the verification set and the test set, and the trained neural network model is determined.

In step S440, the radio frequency signal of the device to be identified is received, and the radio frequency signal is converted into a baseband signal.

In step S450, the baseband signal is preprocessed to determine the target signal.

In step S460, the target signal is input into the trained neural network model for identification, and the identification result is determined.

In step S470, the identification result is judged based on the radio frequency fingerprint database of the legitimate ground station, and the legitimacy of the device to be identified is determined.

It should be noted that, the specific implementation method of each step in this embodiment has been introduced above, and will not be repeated here.

FIG. 7 is a schematic diagram of an apparatus for identifying a device according to an embodiment of the present invention. As shown in FIG. 7 , the device identification device of this embodiment includes an acquisition unit 10 , a conversion unit 20 , a preprocessing unit 30 and an identification unit 40 . Wherein, the acquiring unit 10 is configured to acquire the radio frequency signal sent by the device to be identified. The conversion unit 20 is used for converting the radio frequency signal into a baseband signal. The preprocessing unit 30 is used for preprocessing the baseband signal to determine the target signal. The identification unit 40 is used for inputting the target signal into the pre-trained neural network model for identification, so as to determine the identification result.

Optionally, the acquisition unit 10 and the conversion unit 20 in this embodiment can be embedded in a receiver on the satellite terminal, so as to receive the radio frequency signal sent by the device to be identified through the receiver, and according to the radio frequency signal sent by the device to be identified Determine the identification result of the device to be identified. Further, the receiver architecture in this embodiment adopts a quadrature sampling zero-IF receiver, receives the wireless radio frequency signal sent by the device to be identified through the quadrature sampling zero-IF receiver, and quantifies the wireless radio frequency according to the preset sampling duration and sampling points. The signal is sampled, and the sampled signal obtained after sampling is determined as the radio frequency signal to be identified, and then the sampled radio frequency signal is down-converted through the internal structure of the receiver, and the radio frequency signal sampled by the device to be identified is converted into a baseband signal , so that the subsequent identification is performed based on the baseband signal, and then the identification result of the device to be identified is determined.

Optionally, the preprocessing unit 30 in this embodiment is specifically configured to determine the steady-state segment signal of the baseband signal based on the energy detection method, perform normalization processing on the in-phase steady-state signal and the quadrature steady-state signal respectively, and perform normalization on the normalized signal. The transformed in-phase steady-state signal and quadrature steady-state signal are subjected to data segmentation processing to determine the in-phase target sub-signal and quadrature target sub-signal of at least one period of time, and divide the in-phase target sub-signal and quadrature target sub-signal of each same period of time. They are combined separately to determine the target signal in the form of two-dimensional data. The steady-state segment signal includes an in-phase steady-state signal and a quadrature steady-state signal.

Optionally, the identification unit 40 in this embodiment is further configured to judge the identification result. Specifically, the identification result output by the neural network model is matched with the legal ground station radio frequency fingerprint database pre-arranged on the satellite terminal. When the legal ground station radio frequency fingerprint database includes the radio frequency fingerprint corresponding to the measurement and control ground station represented by the identification result, It is determined that the measurement and control ground station corresponding to the identification result is a legal ground station. In this way, by matching the identification result with the radio frequency fingerprint database of the legal ground station, the validity of the identity of the measurement and control ground station can be determined.

FIG. 8 is a schematic diagram of an electronic device according to an embodiment of the present invention. As shown in FIG. 8 , the electronic device in this embodiment includes a receiving end 41 and a processing system 42 . The receiving end 41 is configured to receive a radio frequency signal sent by the device to be identified. Processing system 42 includes receiver 421 and processor 422 . The receiver 421 is configured to, after receiving the radio frequency signal sent by the device to be identified, sample the radio frequency signal sent by the device to be identified based on the preset sampling duration and the preset number of sampling points to determine the sampled radio frequency signal. The processor 422 is configured to receive the sampled radio frequency signal sampled by the receiver 421, process and identify the sampled radio frequency signal as the radio frequency signal sent by the device to be identified, and then determine the identification result of the device to be identified.

Optionally, the receiving end in this embodiment adopts an RTL-SDR including an RTL2832U chip with a sampling accuracy of 7 bits, and receives the radio frequency signal to be sent by the video device through an omnidirectional antenna with a gain of 3 dB. The processing unit adopts Raspberry Pi 4B, which is connected to the receiving end through the USB interface. Further, the processing unit has built-in application software, and the application software adopts the open source software development tool suite of GNURadio. The software version is 3.8. The software sets the receiver frequency to 433MHz and the sampling rate to 2.0Sample/s, which is 4 times the maximum bandwidth of the signal.

Further, in the test stage, the LoRa wireless transmitter module is used as the signal transmitter (that is, the device to be identified) to simulate the transmission of the navigation satellite signal (that is, the wireless radio frequency signal). When transmitting the wireless radio frequency signal, the frequency point is 433MHz, the spreading factor is 7, the signal bandwidth is 500kHz, the transmission power is 11dBm, the antenna gain is 3dB, the content of the transmitted data is filled with random numbers, and the baud rate is 9600. Therefore, through the LoRa wireless transmitter module, RTL-SDR and Raspberry Pi 4B component RF fingerprint recognition real-time processing system, and by performing the above processing process to realize the identity recognition of the device to be recognized, the recognition accuracy is high, and the recognition efficiency is high.

It should be noted that the setting method of the receiving end and the processing unit in this embodiment is only an example in a specific test or usage scenario, and the specific setting of the receiving end and the processing unit is sufficient to realize the above-mentioned device identification method. No restrictions apply.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims (10)

1. A method for device identification, the method comprising:

acquiring a radio frequency signal sent by equipment to be identified;

converting the radio frequency signal into a baseband signal;

preprocessing the baseband signal to determine a target signal;

and inputting the target signal into a pre-trained neural network model for recognition so as to determine a recognition result.

2. The method of claim 1, wherein converting the radio frequency signal to a baseband signal comprises:

and performing down-conversion processing on the radio frequency signal, and determining the in-phase orthogonal signal after down-conversion processing as the baseband signal.

3. The method of claim 1, wherein the neural network model comprises:

the characteristic unit is used for carrying out dimension reduction and characteristic extraction on the target signal;

the convolution unit comprises a first preset number of convolution layers and is used for performing convolution processing on the feature information extracted by the feature unit;

and the classification unit comprises a second preset number of full connection layers and is used for classifying the processed characteristic information so as to determine an identification result.

4. The method of claim 3, wherein the first predetermined number and the second predetermined number are determined based on a pre-training process corresponding to the neural network model.

5. The method of claim 1, wherein the baseband signals comprise an in-phase baseband signal and a quadrature baseband signal, and wherein pre-processing the baseband signals to determine a target signal comprises:

determining a steady-state segment signal of the baseband signal based on an energy detection method, wherein the steady-state segment signal comprises an in-phase steady-state signal and a quadrature steady-state signal;

respectively carrying out normalization processing on the in-phase steady-state signal and the orthogonal steady-state signal;

carrying out data segmentation processing on the normalized in-phase steady-state signal and the normalized quadrature steady-state signal to determine an in-phase target sub-signal and a quadrature target sub-signal of at least one time period;

and respectively combining the in-phase target sub-signal and the orthogonal target sub-signal in each same time interval to determine a target signal in a two-dimensional data form.

6. The method according to claim 1, wherein the device to be identified is a measurement and control ground station, and the neural network model is deployed in a satellite terminal.

7. The method of claim 1, wherein the neural network model is determined based on the steps of:

acquiring original data;

determining a training set, a verification set and a test set according to the original data;

inputting the data in the training set into a preset neural network model for training, and determining at least one group of model parameters;

inputting the data corresponding to the verification set into the neural network models corresponding to each group of model parameters so as to determine target model parameters according to the identification results corresponding to each neural network model;

inputting data corresponding to the test set into a neural network model corresponding to the target model parameters, and determining the accuracy of the current identification result;

and determining the neural network model corresponding to the target model parameter as the trained neural network model in response to the accuracy of the current recognition result meeting a preset condition.

8. An apparatus for device identification, the apparatus comprising:

the device comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring a radio frequency signal sent by the device to be identified;

the conversion unit is used for converting the radio frequency signal into a baseband signal;

the preprocessing unit is used for preprocessing the baseband signal to determine a target signal;

and the recognition unit is used for inputting the target signal into a pre-trained neural network model for recognition so as to determine a recognition result.

9. An electronic device comprising a memory and a processor, wherein the memory is configured to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method of any of claims 1-7.

10. A computer-readable storage medium, characterized in that a computer program is stored in the computer-readable storage medium, which computer program, when being executed by a processor, carries out the method steps of any one of claims 1-7.

Images