CN114236577B - GNSS signal capturing method based on artificial neural network - Google Patents

Abstract

The invention relates to a GNSS signal acquisition method based on an artificial neural network. The maximum correlation peak value generated during GNSS signal acquisition and the maximum correlation peak value data of nearby frequency points are used as data sets to train a multi-layer perceptron neural network to obtain an optimal Neural network structure and parameter values, according to the obtained neural network structure and parameters, carry out high-precision carrier frequency prediction. The invention utilizes the artificial neural network to predict the spatial distribution of correlation peaks, shortens the search time of the carrier frequency, realizes the rapid capture of the signal, and improves the capture precision and the capture speed.

Description

A GNSS Signal Acquisition Method Based on Artificial Neural Network

technical field

The invention relates to a GNSS signal acquisition method based on an artificial neural network, and belongs to the technical field of satellite positioning and navigation.

Background technique

Satellite navigation and positioning systems involve political, economic, military and other fields, and are widely used in vehicle navigation, aviation navigation, geographic mapping, mass consumption, etc., and can provide users with location and time information. Transmit radio navigation signals to achieve positioning and navigation functions for end users.

The receiver is the core part of the navigation and positioning system, usually composed of an antenna, a radio frequency front end, and a baseband signal processing part. The antenna receives satellite signals broadcast by visible satellites in the space constellation; the RF front-end converts the RF signal received by the antenna into an easy-to-handle digital IF signal through amplification, down-conversion, filtering, and A/D sampling and quantization, and then sends it to the baseband signal processing part; The baseband signal processing part obtains pseudorange information by capturing and tracking the digital intermediate frequency signal, and at the same time demodulates the satellite position information, satellite operating status information, clock error correction information and ionospheric correction information contained in the signal, and finally the solution is obtained. Calculate the receiver position. Among them, acquisition is the core step of baseband signal processing. As the initial part of signal synchronization performed by the navigation and positioning receiver, its performance directly affects the accuracy and processing speed of the subsequent tracking loop. The acquisition process includes demodulation and signal correlation processing. The main task is to identify satellites visible to the current receiver, roughly obtain the carrier frequency and code phase of the received signal, and provide parameter estimation for the subsequent tracking module. Therefore, how to improve the capture performance has become a hot issue in the field of satellite positioning and navigation. The acquisition of satellite signals is actually a three-dimensional search process about visible satellites, carrier frequency and pseudo-random code phase, and the receiver needs to perform a two-dimensional search of the maximum range of each satellite signal when it starts up. Among them, the maximum Doppler frequency shift caused by the relative motion of the user receiver and the satellite in the direction of the connection between the two is ±10KHz. The 20KHz indefinite interval centered on the carrier nominal frequency f is usually used as the receiver startup The frequency search range used to acquire satellite signals. After determining the search range of the signal Doppler frequency shift, the receiver needs to start from the starting value of the frequency search range, and sequentially search with a certain search step size until the signal is finally detected or all frequency ranges are searched. When the step size of the carrier frequency search is set smaller, the frequency error is smaller, but the amount of related operations will increase, resulting in a longer acquisition time and affecting the performance of the receiver; when the step size setting of the carrier frequency search is larger, the frequency error will be smaller. The larger the signal component, the weaker the signal component output by the correlator, thereby increasing the missed alarm rate of signal detection, reducing the sensitivity of signal acquisition, and when the estimation accuracy of the carrier frequency is too large, it will increase the burden of dynamic adjustment of the tracking loop and increase the navigation rate. The data demodulation time affects the performance of the receiver. In severe cases, the tracking loop cannot pull the signal to a locked state, resulting in errors in the demodulated navigation data.

In the implementation of the traditional acquisition algorithm, in order to meet the acquisition accuracy problem, after the frequency domain search is completed, the search step size is reduced in a small range and the search is repeated several times, and the estimation accuracy of the carrier frequency is tens of hertz. The acquisition time is increased, and it is difficult to obtain a high-precision carrier frequency. Therefore, it is urgent to design an acquisition method with good performance, which can reduce the acquisition and search time while acquiring high-precision carrier frequencies.

SUMMARY OF THE INVENTION

Aiming at the deficiencies of the prior art, the present invention proposes a GNSS signal acquisition method based on an artificial neural network.

The core idea of the present invention is to combine the parallel code phase search and capture algorithm with the artificial neural network optimization algorithm, and through the training of the artificial neural network model, the accurate carrier frequency can be quickly predicted, the capture time can be greatly reduced, and the capture time and capture time can be taken into consideration. accuracy, optimizing receiver performance.

Terminology Explanation:

1. GNSS signal, GNSS is the abbreviation of Global Navigation Satellite System, called global satellite navigation system signal or global navigation satellite system signal.

2. PRN is an abbreviation for pseudo random noise code.

3. Levenberg-Marquardt algorithm, Chinese is Levenberg-Marquardt algorithm, referred to as LM algorithm. The least squares fitting used to calculate nonlinear functions has a wide range of applications in parameter fitting. It is an algorithm that uses gradients to find the maximum (small) value, and has the advantages of both the gradient method and the Newton method.

The technical scheme of the present invention is:

A GNSS signal acquisition method based on artificial neural network, the specific implementation steps include:

Step A: Get the dataset, including:

Step 1: Find the correlation peak in the time domain;

Step 2: Use the average correlation peak detection method to detect the correlation peak in the time domain obtained in Step 1, and determine whether the current satellite is captured; it means: if the ratio of the maximum correlation peak to the average correlation peak is greater than the set threshold, go to Step 3 ; otherwise, go to step 4; wherein, the maximum correlation peak value refers to the maximum value in the correlation peak value in the time domain obtained in step 1, and the average correlation peak value is the average value of the correlation peak value in the time domain obtained in step 1;

Step 3: Reduce the frequency search step size, perform step 1, and determine whether the carrier frequency search is completed. If the search is completed, intercept the maximum correlation peak of the current frequency point and the maximum correlation peak of the nearby frequency points as a data set; the data set includes training. set and test set; otherwise, proceed to step 1;

Step 4: determine whether the carrier frequency search is completed, if the search is completed, consider that the current satellite PRN has not been captured, adjust the next satellite PRN, and go to step 5, otherwise, go to step 1;

Step 5: determine whether the satellite PRN search is completed, if the search is completed, the GNSS signal acquisition is completed, otherwise, step 1 is performed;

Step B: Train the artificial neural network model, including:

Step 6: Input the training set obtained in Step A into the artificial neural network model for training and maximum prediction;

Specifically, it refers to: training the artificial neural network model, the training cut-off condition is that the number of iterations is completed or the error meets the requirements, and according to the local optimal principle, the optimal neural network structure and parameters are obtained, that is, the trained artificial neural network model;

Input the test data in the test set into the trained artificial neural network model, test the detection effect of the trained artificial neural network model, and predict the position where the maximum value appears;

Step C: Capture GNSS signals through the trained artificial neural network model;

During the test, obtain the relevant peak test set data generated when the GNSS signal is captured through steps 1-5, input the test set data into the trained artificial neural network model, and perform calculations in the artificial neural network model along the direction of data flow. , until the data is transmitted to the output layer and output, a prediction is completed and GNSS signal acquisition is realized.

The training of the multilayer perceptron neural network is based on the Levenberg-Marquardt learning algorithm, and the optimal neural network structure and parameter values are obtained to improve the capture accuracy and precision.

Preferably according to the present invention, the specific implementation steps of step 1 include:

Step 1.1: Mix the received digital intermediate frequency signal with a replica sine carrier wave and a replica cosine carrier signal of a certain frequency to obtain a baseband complex signal;

Step 1.2: Perform Fourier transform on the baseband complex signal obtained in step 1.1;

Step 1.3: Perform Fourier transform on the local copy pseudocode, and multiply the conjugate value after Fourier transform with the result obtained in step 1.2;

Step 1.4: Perform inverse Fourier transform on the result obtained in step 1.3;

Step 1.5: The result obtained in step 1.4 is modulo squared to obtain the correlation peak in the time domain.

Preferably according to the present invention, in step 2, the average correlation peak detection method is used to detect the correlation peak in the time domain obtained in step 1, which specifically refers to:

Maximum correlation peak A _peak =max(A);

Average correlation peak

N represents the length of the Fourier transform sequence; A refers to the two-dimensional array generated after modulo squaring with the pseudorandom code phase index value as the X axis and the correlation peak value corresponding to each pseudorandom code phase index value as the Y axis ;

Among them, m represents the size of the frequency index value, f _MP represents the maximum Doppler frequency shift, f _IF represents the nominal carrier frequency of the digital intermediate frequency signal, and f _bin represents the carrier frequency search step size;

n represents the size of the pseudo code phase index value, t represents the pseudo code length, t _c represents the maximum pseudo code phase shift, and t _bin is the pseudo code phase search step size;

o represents the size of the frequency index value after reducing the frequency search step size, f _C represents the center carrier frequency predicted by the artificial neural network model, f _MC represents the maximum search range predicted by the artificial neural network model, and f _cmin represents the predicted value of the artificial neural network model. Carrier frequency search step, N represents the satellite PRN sequence to be acquired.

floor为向下取整函数。Preferably according to the present invention, in step 2, the maximum correlation peak value of the current frequency point and the maximum correlation peak value of the nearby frequency points are intercepted as the data set, which specifically refers to: taking the current frequency point as the center, extending p frequencies on the left and right point, obtain the search matrix A _s composed of the maximum correlation peak value A _peak corresponding to each frequency point as the data set,

floor is the round-down function.

According to the preferred embodiment of the present invention, the artificial neural network model includes an input layer, a hidden layer and an output layer, the Levenberg-Marquardt optimization algorithm is used for the optimization algorithm, the sigmoid activation function is used for the hidden layer, and the linear activation function is used for the output layer.

Preferably according to the present invention, in step 6, the training process of the artificial neural network model is as follows:

Using the supervised learning training method, after the training set is input into the input layer, it flows from each neuron to the corresponding neuron in the next layer, summed and transmitted in the hidden layer, and finally reaches the output layer for processing;

Once the artificial neural network model calculates the output corresponding to one of the inputs, the loss function calculates the error vector, and the loss function is the mean square error function: as shown in formula (I):

In formula (I), x is the input vector in the training set, y(x) is the output generated by the artificial neural network, y is the desired output, n is the size of the training set, w is the weight vector, and b is the bias;

Use the back-propagation algorithm to obtain the gradient of all parameters of the artificial neural network model, and update all the parameters of the artificial neural network model through the Levenberg-Marquardt optimization algorithm;

When the loss function converges to a certain degree or the number of iterations is completed, the training ends, and the parameters of the trained artificial neural network model are saved to obtain a trained artificial neural network model.

A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the processor implements the steps of a GNSS signal acquisition method based on an artificial neural network.

A computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, implements the steps of an artificial neural network-based GNSS signal acquisition method.

The beneficial effects of the present invention are:

1. A GNSS signal acquisition method based on artificial neural network of the present invention uses artificial neural network to express the nonlinear relationship between the maximum correlation peak value and the carrier frequency, and predicts the spatial distribution of the correlation peak value according to this relationship, and shortens the frequency of the carrier frequency. Search time, achieve fast signal acquisition, and improve receiver processing efficiency.

2. The present invention is a GNSS signal capture method based on artificial neural network, which trains a multi-layer perceptron neural network based on artificial neural network functions, obtains the optimal network structure and parameter values according to the local optimal principle, and improves the capture accuracy and precision. .

3. The present invention is a GNSS signal acquisition method based on artificial neural network, which combines parallel code phase search and acquisition algorithm and artificial neural network technology, taking into account both acquisition accuracy and acquisition efficiency, and realizes fast and high-precision GNSS signal acquisition.

Description of drawings

Fig. 1 is the flow chart that the present invention obtains the correlation peak value in the time domain;

Fig. 2 is the overall flow chart of the GNSS signal acquisition method based on artificial neural network of the present invention;

FIG. 3 is a schematic structural diagram of the artificial neural network model of the present invention.

Detailed ways

The present invention is further described below in conjunction with the accompanying drawings and embodiments of the description, but is not limited to this:

Example 1

A GNSS signal acquisition method based on artificial neural network, as shown in Figure 2, because the correlation peaks of different frequency search units on the same code band are the sampling points of the |sinc| function curve, so based on the artificial neural network function, the GNSS The maximum correlation peak generated during signal capture and the maximum correlation peak data of nearby frequency points are used as the data set to train the multi-layer perceptron neural network to obtain the optimal neural network structure and parameter values. According to the obtained neural network structure and parameters, high-precision carrier frequency prediction is carried out. The specific implementation steps include:

Step A: Get the dataset, including:

Step 1: Find the correlation peak in the time domain;

The correlation peak in the time domain refers to: two sequences x(n) and y(n) are correlated in the time domain, which is equivalent to their Fourier transforms X(k) and Y ^* (k) (Y ^* (k ) is the conjugate of Y(k)) to perform the product operation in the frequency domain, and the inverse Fourier transform of the product X(k)Y ^* (k) is exactly the correlation value at each code phase that needs to be detected. The result of this inverse Fourier transform is modulo squared, and then the maximum value of the sequence after modulo squaring is found to obtain the correlation peak value in the time series here.

Step 2: Use the average correlation peak detection method to detect the correlation peak in the time domain obtained in step 1, and determine whether the current satellite is captured; it means: if the ratio of the maximum correlation peak to the average correlation peak is greater than the set threshold, the set The setting of the threshold is related to the distribution probability of the correlation value. When the satellite signal is strong, the threshold is 25; when the satellite signal is weak, the threshold is 15. Go to step 3; otherwise, go to step 4; wherein, the maximum correlation peak refers to the maximum value of the correlation peaks in the time domain obtained in step 1, and the average correlation peak is the average value of the correlation peaks in the time domain obtained in step 1;

Step 3: Reduce the frequency search step size, perform step 1, and determine whether the carrier frequency search is completed. If the search is completed, intercept the maximum correlation peak of the current frequency point and the maximum correlation peak of the nearby frequency points as a data set; the data set includes training. set and test set; otherwise, proceed to step 1;

Step 4: determine whether the carrier frequency search is completed, if the search is completed, consider that the current satellite PRN has not been captured, adjust the next satellite PRN, and go to step 5, otherwise, go to step 1;

Step 5: determine whether the satellite PRN search is completed, if the search is completed, the GNSS signal acquisition is completed, otherwise, step 1 is performed;

Step B: Train the artificial neural network model, including:

Step 6: input the training set obtained in step A into the artificial neural network model for training and maximum prediction; the structure of the artificial neural network model is preset according to the length of the intercepted data set;

Specifically, it refers to: training the artificial neural network model, the training cut-off condition is that the number of iterations is completed or the error meets the requirements, and according to the local optimal principle, the optimal neural network structure and parameters are obtained, that is, the trained artificial neural network model;

Input the test data in the test set into the trained artificial neural network model, test the detection effect of the trained artificial neural network model, and predict the position where the maximum value appears;

Step C: Capture GNSS signals through the trained artificial neural network model;

During the test, obtain the relevant peak test set data generated when the GNSS signal is captured through steps 1-5, input the test set data into the trained artificial neural network model, and perform calculations in the artificial neural network model along the direction of data flow. , until the data is transmitted to the output layer and output, a prediction is completed and GNSS signal acquisition is realized.

Example 2

A GNSS signal acquisition method based on artificial neural network according to Embodiment 1, the difference is:

As shown in Figure 1, the specific implementation steps of step 1 include:

Step 1.1: Mix the received digital intermediate frequency signal with a replica sine carrier wave and a replica cosine carrier signal of a certain frequency to obtain a baseband complex signal;

Step 1.2: Perform Fourier transform on the baseband complex signal obtained in step 1.1;

Step 1.3: Perform Fourier transform on the local copy pseudocode, and multiply the conjugate value after Fourier transform with the result obtained in step 1.2;

Step 1.4: Perform inverse Fourier transform on the result obtained in step 1.3;

Step 1.5: The result obtained in step 1.4 is modulo squared to obtain the correlation peak in the time domain.

In step 2, the average correlation peak detection method is used to detect the correlation peak in the time domain obtained in step 1, which specifically refers to:

Maximum correlation peak A _peak =max(A);

Average correlation peak

N represents the length of the Fourier transform sequence; A refers to the two-dimensional array generated after modulo squaring with the pseudo-random code phase index value as the X-axis and the correlation peak value corresponding to each pseudo-random code index value as the Y-axis; The size of X is 2 ^N , and the spatial distribution of the correlation peaks is a |sinc| function curve. Using the artificial neural network to predict the frequency index value corresponding to the maximum correlation peak value, the carrier frequency to be captured can be obtained.

Among them, m represents the size of the frequency index value, f _MP represents the maximum Doppler frequency shift, f _IF represents the nominal carrier frequency of the digital intermediate frequency signal, and f _bin represents the carrier frequency search step size;

n represents the size of the pseudo code phase index value, t represents the pseudo code length, t _c represents the maximum pseudo code phase shift, and t _bin is the pseudo code phase search step size;

o represents the size of the frequency index value after reducing the frequency search step size, f _C represents the center carrier frequency predicted by the artificial neural network model, f _MC represents the maximum search range predicted by the artificial neural network model, and f _cmin represents the predicted value of the artificial neural network model. Carrier frequency search step, N represents the satellite PRN sequence to be acquired.

floor为向下取整函数。In step 2, the maximum correlation peak value of the current frequency point and the maximum correlation peak value of the nearby frequency points are intercepted as the data set, which specifically refers to: taking the current frequency point as the center, extending p frequency points to the left and right, and obtaining each frequency point. The search matrix A _s composed of the corresponding maximum correlation peak A _peak is used as the data set,

floor is the round-down function.

As shown in Figure 3, the artificial neural network model includes input layer, hidden layer and output layer. The optimization algorithm uses the Levenberg-Marquardt optimization algorithm, the hidden layer uses the Sigmoid activation function, and the output layer uses the linear activation function.

In step 6, the training process of the artificial neural network model is as follows:

Using the supervised learning training method, after the training set is input into the input layer, it flows from each neuron to the corresponding neuron in the next layer, summed and transmitted in the hidden layer, and finally reaches the output layer for processing;

Once the artificial neural network model calculates the output corresponding to one of the inputs, the loss function calculates the error vector, and the loss function is the mean square error function: as shown in formula (I):

In formula (I), x is the input vector in the training set, y(x) is the output generated by the artificial neural network, y is the desired output, n is the size of the training set, w is the weight vector, and b is the bias;

Use the back-propagation algorithm to obtain the gradient of all parameters of the artificial neural network model, and update all the parameters of the artificial neural network model through the Levenberg-Marquardt optimization algorithm;

When the loss function converges to a certain degree or the number of iterations is completed, the training ends, and the parameters of the trained artificial neural network model are saved to obtain a trained artificial neural network model. In the training phase, the number of nodes in the input layer, output layer and hidden layer, weight vector w and bias b in the artificial neural network structure are modified by the optimal algorithm, so that the loss function is minimized.

Example 3

A GNSS signal acquisition method based on artificial neural network according to embodiment 1 or 2, the difference is:

The digital intermediate frequency signal output by the RF front-end of the GNSS receiver is processed. In the signal acquisition stage, the GNSS receiver searches all PRNs in turn, saves the maximum value of the correlation peak value of the search frequency point, and generates a frequency index value of the X-axis and a frequency index value of each PRN. The correlation peak corresponding to each frequency index value is a two-dimensional array data set A _s of the Y-axis.

Taking the application of Beidou signal for digital intermediate frequency signal processing as an example, m is [1, 41], f _MP is 10KHz, f _IF is 0.098MHz, f _bin is 500Hz, t is 1023, t _bin is 1, and f _MC is 100Hz, f _cmin is 100Hz, o is [1, 3], PRN_N is [1, 37]. Include the following steps:

Step 1: Set the current satellite number PRN∈PRN_N to be captured, set the frequency f=f _IF -f _MP +(i-1)f _bin , i∈m generated by the local carrier generator, and set the received digital intermediate frequency signal S _IF (n) is respectively mixed with the sine carrier signal i and the cosine carrier signal q generated by the local carrier generator to obtain the baseband complex signal s(n)=I+jQ.

Step 2: Perform Fourier transform on the baseband complex signal s(n) obtained in step 1 to obtain a transform result X(k).

Step 3: The local pseudocode generator generates a pseudocode signal c(n) according to the current PRN, performs Fourier transform to obtain C(k), and takes the conjugate to obtain C ^* (k), which is the same as the result X(k) obtained in step 2. Multiply to get Y(k).

Step 4: Perform inverse Fourier transform on the result Y(k) obtained in Step 3 to obtain y(n).

Step 5: Modulo square the result y(n) obtained in step 4 to obtain the correlation peak array A in the time domain, and obtain the average value of the correlation peak array A

认为当前卫星PRN已被捕获，得到f_C＝f_IF-f_MP+(i-1)f_bin，进行步骤7，否则进行步骤8。Step 6: Use the average correlation peak detection method to detect the result of Step 5, if

It is considered that the current satellite PRN has been captured, and f _C =f _IF -f _MP +(i-1)f _bin is obtained, and step 7 is performed; otherwise, step 8 is performed.

Step 7: Set the frequency generated by the local carrier generator f=f _C -f _MC +L/-1)f _cmin , j∈o, if j<o, then j+1, repeat steps 1 to 5, save each The maximum value of the correlation peak value of the frequency point is searched, and if j=o, a search matrix A _s composed of A _peak is obtained, and step 10 is performed.

Step 8: If i<m, then i+1, repeat steps 1 to 6, if i=m, consider that the current PRN satellite is not visible, and go to step 9.

Step 9: If PRN<PRN_N, then PRN+1, repeat steps 1 to 6, if PRN=PRN_N, the Beidou signal acquisition is completed.

Step 10: The matrix A _s is handed over to the artificial neural network model for training and maximum value prediction, and the structure of the artificial neural network model is preset according to the size of the matrix A _s .

Step 11: Train the multi-layer perceptron neural network based on the Levenberg Marquardt learning algorithm. The training cut-off condition is that the number of iterations is completed or the error meets the requirements. According to the principle of local optimization, the optimal neural network structure and parameters are obtained.

Step 12: After the training is completed, input test data to test the detection effect of the artificial neural model. Analyze and test the test data, including skewness and peak value, obtain the spatial distribution of the peak value, and predict the position pos of the maximum value.

Step 13: The Beidou signal acquisition is completed, and the predicted high-precision carrier frequency f _PRN =f _C -f _MC +(pos-1)f _{cmin is obtained} .

In this embodiment, seven test sets are used to compare the carrier frequency predicted based on the artificial neural network and the carrier frequency captured by the traditional acquisition algorithm. Table 1 lists the carrier frequency of the simulated intermediate frequency data in this Predicted carrier frequency, error and Matlab simulation time. Table 2 lists the carrier frequency of the simulated intermediate frequency data in this embodiment, the carrier frequency captured based on the traditional capture algorithm, the error, and the Matlab simulation time.

Table 1

Table 2

Combining Table 1 and Table 2, it can be seen that the GNSS signal acquisition method based on artificial neural network is better than the carrier frequency based on the traditional acquisition algorithm in acquisition accuracy and precision, and the time is about five times that based on the traditional acquisition algorithm. One, the efficiency is greatly improved. This shows that the GNSS signal acquisition method based on artificial neural network takes both acquisition accuracy and acquisition efficiency into consideration, and can achieve fast and high-precision GNSS signal acquisition.

Example 4

A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the processing of the artificial neural network-based GNSS signal acquisition method described in any of Embodiments 1-3 is realized. step.

Example 5

A computer-readable storage medium having a computer program stored thereon, the computer program implementing the steps of the artificial neural network-based GNSS signal acquisition method described in any one of Embodiments 1-3 when the computer program is executed by a processor.

Claims (8)

1. A GNSS signal capturing method based on an artificial neural network is characterized by comprising the following concrete implementation steps:

step A: acquiring a data set comprising:

step 1: obtaining a correlation peak value in a time domain;

step 2: detecting the correlation peak value in the time domain obtained in the step 1 by adopting an average correlation peak detection method, and judging whether the current satellite is captured or not; the method comprises the following steps: if the ratio of the maximum correlation peak value to the average correlation peak value is larger than a set threshold value, entering the step 3; otherwise, entering step 4; wherein, the maximum correlation peak value refers to the maximum value in the correlation peak values in the time domain obtained in step 1, and the average correlation peak value is the average value of the correlation peak values in the time domain obtained in step 1;

and 3, step 3: reducing the frequency searching step length, executing step 1, judging whether the carrier frequency searching is finished, and if the carrier frequency searching is finished, intercepting the maximum correlation peak value of the current frequency point and the maximum correlation peak value of the frequency points nearby as a data set; the data set comprises a training set and a test set; otherwise, continuing to execute the step 1;

and 4, step 4: judging whether the carrier frequency searching is finished or not, if so, determining that the current satellite PRN is not captured, adjusting the next satellite PRN, and performing the step 5, otherwise, performing the step 1;

and 5: judging whether the satellite PRN search is finished, if so, finishing the acquisition of the GNSS signal, otherwise, executing the step 1;

and B: training an artificial neural network model, comprising:

step 6: inputting the training set obtained in the step A into an artificial neural network model for training and maximum value prediction; the method specifically comprises the following steps: training an artificial neural network model, wherein the training cutoff condition is that iteration times are finished or errors meet requirements, and obtaining an optimal neural network structure and parameters, namely the trained artificial neural network model, according to a local optimal principle;

inputting the test data in the test set into the trained artificial neural network model, testing the detection effect of the trained artificial neural network model, and predicting the position of the maximum value;

and C: GNSS signal capture is carried out through the trained artificial neural network model;

during testing, the relevant peak value test set data generated during GNSS signal capture is obtained through the steps 1-5, the test set data is input into the trained artificial neural network model, calculation is carried out in the artificial neural network model along the data flowing direction and output, one-time prediction is completed, and GNSS signal capture is achieved.

2. The method for capturing GNSS signals based on artificial neural network according to claim 1, wherein the step 1 is implemented by:

step 1.1: mixing the received digital intermediate frequency signal with a copy sine carrier signal and a copy cosine carrier signal of a certain frequency respectively to obtain a baseband complex signal;

step 1.2: carrying out Fourier transform on the baseband complex signal obtained in the step 1.1;

step 1.3: carrying out Fourier transform on the local copy pseudo code, and multiplying the conjugate value after Fourier transform by the result obtained in the step 1.2;

step 1.4: performing inverse Fourier transform on the result obtained in the step 1.3;

step 1.5: and (4) carrying out modulus square on the result obtained in the step 1.4 to obtain a correlation peak value in a time domain.

3. The method for capturing GNSS signals based on an artificial neural network according to claim 1, wherein in step 2, an average correlation peak detection method is used to detect the correlation peak value in the time domain obtained in step 1, specifically:

maximum correlation peak value a _peak ＝max(A)；

Mean correlation peak

N represents the length of the sequence subjected to fourier transform; a is a two-dimensional array which is generated after modular squaring and takes a pseudo-random code phase index value as an X axis and takes a correlation peak value corresponding to each pseudo-random code index value as a Y axis;

where m denotes the size of the frequency index value, f _MP Representing the maximum Doppler shift, f _IF Indicating the nominal carrier frequency, f, of the digital intermediate frequency signal _bin Representing a carrier frequency search step; n represents the magnitude of the pseudo code phase index value, t represents the pseudo code length,t _c representing the maximum pseudo-code phase shift, t _bin Searching step length for the pseudo code phase; o denotes the size of the frequency index value after reducing the frequency search step size, f _C Representing the center carrier frequency, f, for artificial neural network model prediction _MC Maximum search Range, f, representing Artificial neural network model prediction _cmin Representing the carrier frequency search step predicted by the artificial neural network model, and N representing the satellite PRN sequence to be acquired.

4. The method as claimed in claim 1, wherein in step 2, the maximum correlation peak of the current frequency point and the maximum correlation peak of the frequency points in the vicinity thereof are intercepted as data sets, specifically: the current frequency point is taken as the center, the left side and the right side are respectively extended by p frequency points, and the maximum correlation peak value A corresponding to each frequency point is obtained _peak Composed search matrix A _s As a set of data, it is possible to,

floor is a floor rounding function.

5. The GNSS signal capturing method based on artificial neural network of claim 1, characterized in that the artificial neural network model comprises an input layer, a hidden layer and an output layer, the optimization algorithm uses Levenberg-Marquardt optimization algorithm, the hidden layer uses Sigmoid activation function, the output layer uses linear activation function.

6. The method as claimed in claim 5, wherein in step 6, the artificial neural network model is trained as follows:

after the training set is input into the input layer by adopting a supervised learning training mode, flowing into corresponding neurons in the next layer from each neuron, summing and transmitting in the hidden layer, and finally reaching the output layer for processing;

once the artificial neural network model calculates the output corresponding to one of the inputs, a loss function calculates an error vector, the loss function being a mean square error function: as shown in formula (I):

in formula (I), x is the input vector in the training set, y (x) is the output generated by the artificial neural network, y is the desired output, n is the size of the training set, w is the weight vector, and b is the offset;

obtaining gradients of all parameters of the artificial neural network model by using a back propagation algorithm, and updating all parameters of the artificial neural network model by using a Levenberg-Marquardt optimization algorithm;

and when the loss function is minimum or the iteration times are finished, finishing the training, and storing the parameters of the trained artificial neural network model to obtain the trained artificial neural network model.

7. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor when executing the computer program performs the steps of the artificial neural network-based GNSS signal acquisition method of any of claims 1 to 6.

8. A computer-readable storage medium, on which a computer program is stored, wherein the computer program, when being executed by a processor, implements the steps of the artificial neural network-based GNSS signal acquisition method of any one of claims 1 to 6.

Images