A method for reinforcing beidou time

Abstract

The application relates to a Beidou time reinforcement method and belongs to the technical field of satellite navigation signal processing. The application comprises a server module, an antenna receiving module, a real position calculating module, a real position storage module, a position difference calculating module, a remote upgrading module, an FPGA module, an output interface module, a fraud state and switching module, a constant temperature crystal oscillator and crystal oscillator temperature sensing module and a crystal oscillator working time length module. The feature signal acquisition method is simple, can accurately capture the subtle features of the fraud signal, realizes high accuracy and low false alarm rate, reduces the risk of false detection and missed detection, effectively solves the identification difficulty of the non-sensing fraud signal, has low calculation complexity, reduces the demand for calculation resources, enhances real-time performance, reduces cost, improves the practicability and reliability of the system, and expands the application scenarios.

Description

TECHNICAL FIELD

[0001] The application relates to a Beidou time reinforcement method and belongs to the technical field of satellite navigation signal processing. BACKGROUND

[0002] The Beidou system uses weak electromagnetic signals for positioning, timing, speed measurement and the like. Due to the electromagnetic signal characteristics of the Beidou system, the system is vulnerable to various spoofing attacks. In particular, the non-inductive spoofing technology gradually guides the receiver clock to produce a cumulative deviation from the real time through subtle shifts, which is difficult to detect in a short time, but long-term accumulation will lead to serious time synchronization errors. This causes great harm to national defense, production and social life.

[0003] At present, there are a large number of technologies and methods for anti-spoofing of the Beidou system, which can be roughly divided into the following categories:

[0004] (1) Single difference of signal characteristics is used to identify spoofing signals, such as carrier-to-noise ratio, signal spectrum, etc. This method is simple to process, but as the spoofing technology develops, the degree of simulating real signal characteristics improves, leading to the inability to distinguish spoofing signals.

[0005] (2) Therefore, people have comprehensively used the correlation of multiple signals to improve the technical level of Beidou reinforcement. For example, the cross ambiguity function (CAF) calculates the correlation of Doppler frequency and code delay to find the existence of spoofing signals. If the normal signal and the spoofing signal are not significantly different, the reinforcement effect of this method is not ideal. For another example, the spoofing detection method based on CNN and using particle swarm optimization algorithm (PSO) gets better results in detecting spoofing attacks, but requires higher computing resources, which is not convenient for scenarios with weak computing resources.

[0006] (3) Therefore, people have adopted BP neural networks to detect spoofing signals. This method has relatively reduced computational complexity, but its limitations are also more obvious. Its accuracy needs to be improved, and the complexity can be further reduced.

[0007] (4) In addition, people select multiple signals at the radio frequency end. This selection has two defects, one is the high complexity of signal acquisition, and the other is the weak signal strength. After being output to different signal processing modules through radio frequency power division, the already weak satellite signal strength is weakened, which causes unnecessary impact on the reception and processing of normal signals in the absence of interference. Third, it cannot cover all potential interference and spoofing. SUMMARY

[0008] The present application aims at solving the above-mentioned problems of the prior art, discarding the conventional reinforcement technology at the radio frequency end, and proposing strong signal end processing. A Beidou time reinforcement method is provided. The feature signal acquisition method of the present application is simple, can accurately capture the subtle features of the deceptive signal, realizes high accuracy and low false alarm rate, reduces the risk of false detection and missed detection, effectively solves the identification problem of non-inductive deceptive signals, has low computational complexity, reduces the demand for computing resources, enhances real-time performance, reduces cost, improves the practicality and reliability of the system, and expands the application scenarios.

[0009] To achieve the above object, the present application provides the following technical solutions:

[0010] The Beidou time reinforcement method of the present application comprises:

[0011] The server module, the antenna receiving module, the real position calculation module, the real position storage module, the position difference calculation module, the remote upgrade module, the FPGA module, the output interface module, the deception state and switching module, the constant temperature crystal oscillator and crystal oscillator temperature sensing module, and the crystal oscillator working time module.

[0012] The server module is used for data preprocessing and BP neural network model construction, training and testing.

[0013] The antenna receiving module is used for receiving Beidou signals to obtain positioning and time data.

[0014] The real position calculation module is used for calibrating the accurate antenna position.

[0015] The real position storage module is used for storing the real position.

[0016] The position difference calculation module is used for calculating the difference between the real position and the positioning of the received Beidou.

[0017] The remote upgrade module is used for deploying the BP neural network model to the FPGA module.

[0018] The FPGA module is used for receiving the Beidou time from the antenna receiving module, the positioning difference from the position difference calculation module, the environment temperature provided by the crystal oscillator temperature sensing module, and the crystal oscillator working time provided by the crystal oscillator working time module, performing BP neural network inference calculation, and distinguishing normal signals and deceptive signals.

[0019] The output interface module receives the processing results from the FPGA module, including synchronization time data or deception signal detection state signals, and outputs the processed time synchronization signals or alarm signals to the deception state and switching module.

[0020] The fraud state and the switching module: if the fraud state is accepted, an alarm signal is output to remind the downstream device, and if the non-fraud state is accepted, time information is output;

[0021] The constant temperature crystal oscillator and crystal oscillator temperature sensing module is used to provide an oscillation source and the environmental temperature of the crystal oscillator;

[0022] The crystal oscillator working time length module is used to accumulate the working time length of the crystal oscillator.

[0023] Preferably, the server module includes a data acquisition and preprocessing module, a neural network training module, and a neural network testing module; the data acquisition and preprocessing module is used to obtain positioning difference, crystal oscillator temperature, and working time length data, normalize the data, and divide the data; the neural network training module is used to train the data, iteratively adjust the network parameters multiple times, optimize the weights and biases of the neural network, and minimize the loss function; the neural network testing module uses test data to evaluate the performance of the trained neural network model.

[0024] The working process of the data acquisition and preprocessing module is as follows:

[0025] 1) The positioning difference dataset is the difference between the high-precision positioning of the ground-based augmentation system and the positioning received by the receiver; the crystal oscillator temperature and working time length dataset is obtained through the crystal oscillator temperature sensor and the recorded crystal oscillator working time length;

[0026] 2) The normalization function used is mapminmax, which scales the data to the [0, 1] interval:

[0027]

[0028] where x min is the minimum value in the dataset, x max is the maximum value, and x' is the normalized data;

[0029] 3) After normalization, the dataset is divided into a training set, a validation set, and a test set;

[0030] The working process of the neural network training module is as follows:

[0031] 1) Initialize the network parameters module: set the number of layers in each layer, as well as its activation function, optimizer, learning rate, and evaluation metric;

[0032] 2) Input layer module: 3 parameters, including positioning difference, temperature, and working time length;

[0033] 3) Hidden layer calculation module: set the number of hidden layers to 1, the number of neurons to 12, and the activation function to ReLU:

[0034] ​

[0035] y ReLU (x)=max(0,x)

[0036] where x is the input signal of the neuron;

[0037] 4) Output layer calculation module: the output layer is 1, and the activation function uses Sigmoid:

[0038]

[0039] where x is the net input of the output layer neuron, which is calculated based on the output of the previous layer and the weight and bias of the current layer;

[0040] 5) Loss value calculation module: the loss function is binary cross-entropy, and the cross-entropy loss can improve the stability and convergence speed of training:

[0041]

[0042] where, is the predicted probability, y i is the true label (0 or 1);

[0043] 6) Training round judgment module: train the model for 200 cycles, each batch contains 32 samples, and return the input layer when the specified number of rounds is not reached, and retrain;

[0044] 7) Back propagation module: pass the output to the input layer, and use the output error value as the basis for correcting the connection weights of each neuron. When the error value is added to each neuron again, the output error value decreases after the forward propagation process, that is, the minimum value of the error function is found;

[0045] 8) Training end module: if the specified number of training rounds is reached, the training is ended, and the model is saved as a file.

[0046] The working process of the neural network test module is as follows:

[0047] 1) Input layer: receive input data and convert it to a format that the model can process;

[0048] 2) Hidden layer calculation: perform nonlinear transformation on the input data, extract features, and pass them to the output layer;

[0049] 3) Output layer calculation: output the final prediction result according to the binary classification task;

[0050] 4) Calculate model performance indicators: calculate the final test set indicators to judge the generalization ability and performance of the model on unseen data;

[0051] Precision: The probability that the model correctly detects a BD spoofing signal as a spoofing signal:

[0052]

[0053] Recall: The probability that the model correctly identifies all spoofing signals:

[0054]

[0055] False alarm rate: The probability that a normal BD signal is incorrectly predicted as a spoofing signal:

[0056]

[0057] Accuracy: The probability that both correct signals and spoofing signals are correctly detected as real signals and spoofing signals:

[0058]

[0059] Where TP is the number of correct signals that the model correctly predicts as correct signals, FP is the number of spoofing signals that the model incorrectly predicts as correct signals, FN is the number of correct signals that the model incorrectly predicts as spoofing signals, and TN is the number of spoofing signals that the model correctly predicts as spoofing signals.

[0060] Preferably, the antenna receiving module receives satellite signals through an antenna, which include real satellite signals, as well as spoofing signals or interference signals. After processing by the antenna receiving module, the signals are transmitted to the input data processing module in the FPGA module. The antenna receiving module includes a positioning processing module and a time processing module.

[0061] Positioning processing module: calculates the received positioning.

[0062] Time processing module: processes time information.

[0063] Preferably, the real position calculation module runs in the initial running stage of the device and includes a ground-based enhancement module, a significant spoofing interference detection module, and a real position calculation module.

[0064] The ground-based enhancement module: receives Beidou ground-based enhancement signals.

[0065] The significant spoofing interference detection module: uses a single signal feature to identify spoofing signals with significant differences.

[0066] The real position calculation module: if Beidou ground-based enhancement signals are received and there is no significant spoofing interference, then the accurate antenna position result is calculated based on the ground-based enhancement positioning data and the Beidou positioning data, and is transmitted to the real position storage module.

[0067] Preferably, the FPGA module is deployed with a BP neural network model trained by the server module, and the received data enters the BP neural network model for analysis and processing, the process being as follows:

[0068] 1) Input data processing module: cache and preprocess the received data;

[0069] 2) Input layer module: receive preprocessed data and pass it to the hidden layer;

[0070] 3) Hidden layer calculation module: composed of multiple parallel processing units, each processing unit is responsible for calculating the output of a neuron, these processing units include multipliers, activation function units;

[0071] The hidden layer has 3 neurons;

[0072] Weight matrix W (1) : 12x3 matrix, used to connect the input layer and the hidden layer;

[0073] Bias vector b (1) : 12x1 vector;

[0074] Linear calculation of the hidden layer:

[0075] z (1) = W (1) x + b (1)

[0076] Input z (1) to the ReLU activation function, which performs a nonlinear transformation on the output of each node;

[0077] a (1) = ReLU(z (1) )

[0078] 4) Output layer calculation module: calculate the final prediction result from the output of the hidden layer;

[0079] Output layer weight vector W (2) : 1x1 matrix;

[0080] Bias scalar b (2) ;

[0081] Linear calculation of the output layer:

[0082] z (2) = W (2) a (1) + b (2)

[0083] Input z (2) to the Sigmoid activation function, and y is the final output result;

[0084] 5) Network output result module: including a data converter, for converting the result of processing into a format that can be understood by an external system, and then discriminating whether there is a cheating signal;

[0085] 6) Time delay compensation module: when the processing result is a normal signal, the time information output by the BP neural network model is transmitted to this module for time delay compensation processing;

[0086] 7) Communication interface module: the time information and the cheating state after time delay compensation are transmitted to the output interface module through this module.

[0087] The beneficial effects of the present application are:

[0088] (1) Abandoning the conventional reinforcement technology at the radio frequency end

[0089] The signal strength at the radio frequency end is weak, which affects the reinforcement result. After passing through the radio frequency power division module, the signal is further weakened, and in the case of no interference cheating, the normal signal reception and processing are weakened. It cannot cover all potential interference and cheating. It cannot effectively identify the non-sensing cheating signal.

[0090] (2) Proposing strong signal end processing

[0091] By introducing the working characteristics of crystal oscillator aging and environmental change, and the calibrated antenna position, the detection ability of the signal micro-deviation is enhanced, and the influence of the radio frequency end interference on the system is reduced.

[0092] (3) Multi-dimensional feature joint analysis to improve anti-cheating ability

[0093] The present application adopts the signal features of positioning difference, temperature and working time, introduces artificial intelligence technology, and combines multi-dimensional feature training to make up for the limitations of single difference relying on signal features. It can accurately capture the subtle features of the cheating signal, quickly distinguish between real signals and cheating signals, and effectively solve the recognition problem of non-sensing cheating signals.

[0094] (4) Low computational complexity, reducing the demand for computing resources

[0095] The traditional method of comprehensively utilizing the correlation of multiple signals (such as CNN) has complex network, high computational complexity, complex process, and requires high computing resources. The present application reduces the demand for computing resources while ensuring the improvement of accuracy.

[0096] (5) Real-time enhancement

[0097] Because the computing resources required by the present application are small, the neural network can still maintain the real-time performance of signal processing when it is transplanted to a smaller FPGA hardware platform.

[0098] (6) reduce the cost, expand the application scenario

[0099] Because the required FPGA platform of the application is small, the cost is reduced, and the application scenario is expanded.

[0100] (7) the difficulty of obtaining characteristic signal is low

[0101] The application selects time and temperature affecting the frequency of the crystal oscillator and positioning accuracy as characteristic input, and reduces the difficulty of signal acquisition.

[0102] In summary, the application shows significant advantages for BD time reinforcement in the non-inductive fraud scene, and provides an important guarantee for the safe application of Beidou system. BRIEF DESCRIPTION OF DRAWINGS

[0103] Figure 1 is a structure diagram of a Beidou time reinforcement device of the application;

[0104] Figure 2 is a schematic diagram of a server module of the application;

[0105] Figure 3 is an antenna receiving module diagram of the application;

[0106] Figure 4 is a calculation of real position module diagram of the application;

[0107] Figure 5 is an FPGA module diagram of the application. DETAILED DESCRIPTION

[0108] The technical solutions in the embodiments of the application will be described below with reference to the drawings in the embodiments of the application.

[0109] Example 1

[0110] The Beidou time reinforcement method of the embodiment comprises a server module, an antenna receiving module, a real position calculating module, a real position storage module, a remote upgrade module, an FPGA module, an output interface module, a spoofing state and switching module, a constant temperature crystal oscillator and crystal oscillator temperature sensing module, and a crystal oscillator working time length module. The server module is used for data preprocessing and BP neural network model construction, training and testing. The antenna receiving module is used for receiving Beidou signals to obtain positioning and time data. The real position calculating module is used for calibrating accurate antenna positions. The real position storage module is used for storing real positions. The remote upgrade module is used for deploying the BP neural network model to the FPGA module. The FPGA module is used for receiving Beidou time from the antenna receiving module, positioning differences from the position difference calculating module, environmental temperature provided by the crystal oscillator temperature sensing module, and crystal oscillator working time length provided by the crystal oscillator working time length module, and then performing BP neural network inference calculation to distinguish normal signals and spoofing signals. The output interface module receives processing results from the FPGA module, including synchronous time data or spoofing signal detection state signals, and outputs the processed time synchronization signals or alarm signals to the spoofing state and switching module. The spoofing state and switching module outputs an alarm signal to remind downstream devices if a spoofing state is accepted, and outputs time information if a non-spoofing state is accepted. The constant temperature crystal oscillator and crystal oscillator temperature sensing module are used for providing an oscillation source and the environmental temperature of the crystal oscillator. The crystal oscillator working time length module is used for cumulative calculation of the crystal oscillator working time length.

[0111] The server module comprises a data acquisition and preprocessing module, a neural network training module, and a neural network testing module. The data acquisition and preprocessing module is used for obtaining positioning differences, crystal oscillator temperature and working time length data, normalizing the data, and dividing the data. The neural network training module is used for training data, adjusting network parameters multiple times, optimizing the weights and biases of the neural network, and minimizing the loss function. The neural network testing module uses test data to evaluate the performance of the trained neural network model.

[0112] The working process of the data acquisition and preprocessing module is as follows:

[0113] 1) The positioning difference data set is the difference between the high-precision positioning of the ground-based augmentation system and the positioning received by the receiver. The crystal oscillator temperature and working time length data set is obtained through a crystal oscillator temperature sensor and recorded crystal oscillator working time length.

[0114] 2) The normalization function used is mapminmax, which scales the data to the [0, 1] interval:

[0115]

[0116] wherein, x min is the minimum value in the data set, x maxis the maximum value, x' is the normalized data;

[0117] 4) After normalization, the data set is divided into training set, validation set, test set;

[0118] The working process of the neural network training module is as follows:

[0119] 1) Initialization of network parameters module: set the number of layers of each layer and its activation function, optimizer, learning rate, evaluation index;

[0120] 2) Input layer module: 3 parameters, including positioning difference, temperature and working time;

[0121] 3) Hidden layer calculation module: the number of hidden layers is set to 1, the number of neurons is 12, and the activation function

[0122] is ReLU:

[0123] y ReLU (x) = max(0, x)

[0124] Where x is the input signal of the neuron;

[0125] 4) Output layer calculation module: the output layer is 1, and the activation function uses Sigmoid:

[0126]

[0127] Where x is the net input of the output layer neuron, which is calculated based on the output of the previous layer and the weight and bias of the current layer;

[0128] 5) Loss value calculation module: the loss function is binary cross entropy, and the cross entropy loss can improve the stability and convergence speed of training:

[0129]

[0130] Where, is the predicted probability, y i is the true label (0 or 1);

[0131] 6) Training round judgment module: train the model for 200 cycles, each batch contains 32 samples, and return the input layer when the specified number of rounds is not reached, and retrain;

[0132] 7) Back propagation module: pass the output to the input layer, and use the output error value as the basis for correcting the connection weights of each neuron. When the error value is added to each neuron again, the output error value decreases after the forward propagation process again, that is, the minimum value of the error function is found;

[0133] 8) Training end module: if the set number of training rounds is reached, the training ends, and the model is saved as a file.

[0134] The working process of the neural network test module is as follows:

[0135] 1) Input layer: receives input data and converts it into a format that the model can process;

[0136] 2) Hidden layer calculation: performs a nonlinear transformation on the input data, extracts features, and passes them to the output layer;

[0137] 3) Output layer calculation: outputs the final prediction result according to the binary classification task;

[0138] 4) Calculate model performance indicators: calculate the final indicators of the test set to judge the generalization ability and performance of the model on unseen data;

[0139] Accuracy: the probability that the model correctly detects a BD spoof signal as a spoof signal:

[0140]

[0141] Recall: the probability that the model correctly identifies all spoof signals:

[0142]

[0143] False alarm rate: the probability that a normal BD signal is incorrectly predicted as a spoof signal:

[0144]

[0145] Accuracy: the probability that correct signals and spoof signals are correctly detected as true signals and spoof signals:

[0146]

[0147] where TP is the number of correct signals correctly predicted by the model as correct signals, FP is the number of spoof signals incorrectly predicted by the model as correct signals, FN is the number of correct signals incorrectly predicted by the model as spoof signals, and TN is the number of spoof signals correctly predicted by the model as spoof signals.

[0148] The antenna receiving module receives satellite signals through an antenna, which includes real satellite signals, spoof signals, or interference signals. After processing by the antenna receiving module, the signals are transmitted to the input data processing module in the FPGA module. The antenna receiving module includes a positioning processing module and a time processing module.

[0149] Positioning processing module: calculates the received positioning;

[0150] Time processing module: processing time information.

[0151] Real position calculation module, which runs in the initial running stage of the device, including ground-based enhancement module, significant spoofing interference detection module, real position calculation module;

[0152] Ground-based enhancement module: for receiving Beidou ground-based enhancement signals;

[0153] Significant spoofing interference detection module: using a single signal feature to identify a significantly different spoofing signal;

[0154] Real position calculation module: if Beidou ground-based enhancement signals are received and there is no significant spoofing interference, then according to the ground-based enhancement positioning data and the Beidou positioning data calculated, the accurate antenna position result is calculated and transmitted to the real position storage module.

[0155] Calculate the position difference module: calculate the difference between the real position and the Beidou positioning received;

[0156] The BP neural network model trained by the server module is deployed on the FPGA module, and the received data enters the BP neural network model for analysis and processing, the process is as follows:

[0157] 1) Input data processing module: cache and preprocess the received data;

[0158] 2) Input layer module: receives preprocessed data and passes it to the hidden layer;

[0159] 3) Hidden layer calculation module: the module is composed of multiple parallel processing units, each processing unit is responsible for calculating the output of a neuron, these processing units include multipliers, activation function units;

[0160] The hidden layer has 3 neurons;

[0161] Weight matrix W (1) : 12x3 matrix, used to connect the input layer and the hidden layer;

[0162] Bias vector b (1) : 12x1 vector;

[0163] Linear calculation of hidden layer:

[0164] z (1) = W (1) x+b (1)

[0165] Input z (1) to the ReLU activation function to perform nonlinear transformation on the output of each node;

[0166] a (1)= ReLU(z (1) )

[0167] 4) Output layer calculation module: calculate the final prediction result with the output of hidden layer;

[0168] Output layer weight vector W (2) : 1x1 matrix;

[0169] Bias scalar b (2) ;

[0170] Linear calculation of output layer:

[0171] z (2) = W (2) a (1) +b (2)

[0172] Input z (2) to Sigmoid activation function, y is the final output result;

[0173] 5) Network output result module: including a data converter, used to convert the processed result into a format that can be understood by external system, and then identify whether there is a fraud signal;

[0174] 6) Time delay compensation module: when the processing result is normal signal, the time information output by BP neural network model is transmitted to this module for time delay compensation processing;

[0175] 7) Communication interface module: the time information and fraud state after time delay compensation are transmitted to output interface module through this module.

[0176] Although the embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A Beidou time reinforcement device, characterized in that include: Server module, antenna receiving module, real position calculation module, real position storage module, position difference calculation module, remote upgrade module, FPGA module, output interface module, deception status and switching module, constant temperature crystal oscillator and crystal oscillator temperature sensor module, crystal oscillator working time module; The server module is used for data preprocessing and BP neural network model construction, training and testing; The antenna receiving module is used to receive Beidou signals and obtain positioning and time data; The module for calculating the real position is used to calibrate the precise antenna position; The real position storage module is used to store the real position; The position difference calculation module is used to calculate the difference between the real position and the positioning received by Beidou; the positioning difference data set is the difference between the high-precision positioning of the ground-based augmentation system and the positioning received by the receiver; The remote upgrade module is used to deploy the BP neural network model to the FPGA module; The FPGA module is used to receive the Beidou time from the antenna receiving module, the positioning difference from the position difference calculation module, the ambient temperature provided by the crystal oscillator temperature sensor module, and the crystal oscillator working time provided by the crystal oscillator working time module, perform BP neural network reasoning calculation, and distinguish normal signals from deceptive signals; The output interface module receives the processing results from the FPGA module, including the synchronization time data or the deception signal detection status signal, and outputs the processed time synchronization signal or alarm signal to the deception status and switching module; The deception status and switching module: if a deception status is received, it outputs an alarm signal to remind the downstream device; if a non-deception status is received, it outputs time information; The constant temperature crystal oscillator and crystal oscillator temperature sensing module are used to provide an oscillation source and the ambient temperature of the crystal oscillator; The crystal oscillator working time module is used to cumulatively calculate the crystal oscillator working time.

2. A BeiDou time reinforcement device according to claim 1, characterized in that The server module includes a data acquisition and preprocessing module, a neural network training module, and a neural network testing module; the data acquisition and preprocessing module is used to obtain positioning error, crystal oscillator temperature, and operating time data, and normalize and divide the data; the neural network training module is used to train data, iteratively adjust network parameters, optimize the weights and biases of the neural network, and minimize the loss function; the neural network testing module uses test data to evaluate the performance of the trained neural network model; The working process of the data acquisition and preprocessing module is as follows: 1) The positioning difference data set is the difference between the high-precision positioning of the ground-based augmentation system and the positioning received by the receiver; the crystal oscillator temperature and operating time data set is obtained through the crystal oscillator temperature sensor and the recorded crystal oscillator operating time; 2) The normalization function used is mapminmax, which scales the data to the [0,1] interval: Among them, x min is the minimum value in the data set, x max is the maximum value, and x' is the normalized data; 3) After normalization, the dataset is divided into training set, validation set, and test set; The working process of the neural network training module is as follows: 1) Initialize the network parameter module: set the number of layers, activation function, optimizer, learning rate, and evaluation index of each layer; 2) Input layer module: 3 parameters, including positioning error, temperature, and working time; 3) Hidden layer calculation module: The number of hidden layers is set to 1, the number of neurons is 12, and the activation function is ReLU: y ReLU (x)=max(0,x) Among them, x is the input signal of the neuron; 4) Output layer calculation module: The output layer is 1, and the activation function uses Sigmoid: Where x is the net input to the output layer neurons, which is calculated based on the output of the previous layer and the weights and biases of the current layer; 5) Loss calculation module: The loss function is binary cross entropy. Cross entropy loss can improve the stability and convergence speed of training: in, is the predicted probability, y i is the true label 0 or 1; 6) Training round number judgment module: The model is trained for 200 cycles, with each batch containing 32 samples. If the specified number of rounds is not reached, it returns to the input layer and retrains; 7) Backward propagation module: The output is passed to the input layer, and the output error value is used as the basis for correcting the connection weights of each neuron. When the error value is added back to each neuron, it goes through the forward propagation process again, and the output error value decreases, that is, the minimum value of the error function is found; 8) Training end module: If the set number of training rounds is reached, the training ends and the model is saved as a file; The working process of the neural network test module is as follows: 1) Input layer: receives input data and converts it into a format that the model can process; 2) Hidden layer calculation: Perform nonlinear transformation on the input data, extract features, and pass them to the output layer; 3) Output layer calculation: Output the final prediction result based on this binary classification task; 4) Calculate model performance indicators: Calculate various indicators of the final test set to determine the model's generalization ability and performance on unseen data; Precision: The probability that the model correctly detects a BD spoofing signal as a spoofing signal: Recall: The probability that the model can correctly identify all deceptive signals: False alarm rate: The probability that a normal BD signal is mistakenly predicted as a spoofing signal: Accuracy: The probability of correctly detecting the correct signal and the spoofed signal as the real signal and the spoofed signal: Among them, TP is the number of correct signals correctly predicted by the model as correct signals, FP is the number of spoofed signals incorrectly predicted by the model as correct signals, FN is the number of correct signals incorrectly predicted by the model as spoofed signals, and TN is the number of spoofed signals correctly predicted by the model as spoofed signals.

3. A BeiDou time reinforcement device according to claim 1, characterized in that The antenna receiving module receives satellite signals through the antenna. These signals include real satellite signals, deceptive signals or interference signals. After being processed by the antenna receiving module, they are transmitted to the input data processing module in the FPGA module. The antenna receiving module includes a positioning processing module and a time processing module. Positioning processing module: calculates the received positioning; Time processing module: processes time information.

4. A BeiDou time reinforcement device according to claim 1, characterized in that The module for calculating the real position, which runs in the initial operation stage of the device, includes a ground-based enhancement module, a significant deception interference detection module, and a module for calculating the real position; The ground-based augmentation module is used to receive Beidou ground-based augmentation signals; The significant deceptive interference detection module uses a single signal feature to identify deceptive signals with significant differences; The module for calculating the true position: if the Beidou ground-based augmentation signal is received and there is no significant deception interference, the module calculates the precise antenna position result based on the ground-based augmentation positioning data and the Beidou-solved positioning data, and transmits it to the true position storage module.

5. A BeiDou time reinforcement device according to claim 1, characterized in that The BP neural network model trained by the server module is deployed on the FPGA module. The received data enters the BP neural network model for analysis and processing. The process is as follows: 1) Input data processing module: caches and preprocesses the received data; 2) Input layer module: receives preprocessed data and passes it to the hidden layer; 3) Hidden layer calculation module: The module consists of multiple parallel processing units, each of which is responsible for calculating the output of a neuron. These processing units include multipliers and activation function units. The hidden layer has 3 neurons; Weight matrix W (1) : A 12×3 matrix used to connect the input layer and the hidden layer; Bias vector b (1) : 12×1 vector; Linear calculation of hidden layer: z (1) =W (1) x+b (1) z (1) Input to the ReLU activation function to perform nonlinear transformation on the output of each node; and (1) =ReLU(z (1) ) 4) Output layer calculation module: calculates the final prediction result based on the output of the hidden layer; Output layer weight vector W (2) : 1×1 matrix; Bias scalar b (2) ; Linear calculation of the output layer: z (2) =W (2) a (1) +b (2) z (2) Input to the Sigmoid activation function, y is the final output result; 5) Network output module: This includes a data converter that converts the processed results into a format understandable by the external system and then identifies whether there are any fraudulent signals. 6) Delay compensation module: When the processing result is a normal signal, the time information output by the BP neural network model is transmitted to this module for delay compensation processing; 7) Communication interface module: The time information and deception status after delay compensation are transmitted to the output interface module through this module.

Images