1. A navigation satellite weak signal capturing method based on a deep convolutional neural network is characterized by comprising the following steps: the method comprises the following steps that firstly, a navigation satellite signal receiver obtains weak navigation satellite signals, the weak navigation satellite signals are processed by a radio frequency circuit, intermediate frequency signals r (t) are received, local generation signals s (t) are generated, and correlation operation is carried out between the intermediate frequency signals r (t) and the local generation signals s (t) to obtain correlation signals R (t) of the intermediate frequency signals r (t); detecting a correlation value of the signal by adopting an absolute position convolution neural network; thirdly, improving the absolute position precision of the correlation peak value by adopting a region coding method, training a data set and obtaining the PC-CNN_SAnd PP-CNN_SThe optimal model of (2).

2. The method for acquiring the weak signal of the navigation satellite based on the deep convolutional neural network as claimed in claim 1, wherein the step one is realized by the following substeps:

a) acquiring an intermediate frequency signal r (t); for a navigation satellite signal acquisition system of a deep convolutional neural network, after being processed by a radio frequency circuit, the received intermediate frequency signal is as follows:

r(t)＝AD_r(t)C_r(t-τ)×cos[2π(f_IF+f_d)t+φ₀]+n(t) (1)

in the formula (1), A is the signal amplitude, D_r(t) is modulated data, f_dIs the Doppler shift phi₀Is the original phase, C_r(t- τ) is a pseudo random code with a delay time;

b) the receiver generates a local generation signal s (t); for a deep convolutional neural network navigation satellite signal acquisition system, the receiver generates the locally generated signals as follows:

in the formula (2), the first and second groups,is provided with a delay timeThe local pseudo-random code of (a) is,is a local frequency shift, phi₁Is the initial phase;

c) obtaining a correlation signal R (t); performing correlation processing on the intermediate frequency signal r (t) and the local generation signal s (t), wherein the obtained correlation signals are as follows:

in the formula (3), the first and second groups,Δφ＝φ₀-φ₁，n_coris the equivalent noise that includes both gaussian noise and cross-correlation output.

3. The method for capturing the weak signal of the navigation satellite based on the deep convolutional neural network as claimed in claim 1, wherein the second step is realized by the following substeps:

d) the correlator processes the signal to obtain a correlation image and inputs the correlation image into the convolutional neural network; the method comprises the following steps of performing parallel processing on signals by adopting M correlators to reduce acquisition time, dividing a whole related image into a plurality of related sub-images, and inputting the sub-images into a convolutional neural network;

e)PC-CNN_Sidentifying a correlation peak in the image; the correlation peak classification in the absolute position convolutional neural network has S₁A convolutional layer, an A × A kernel, M₁A maximum pooling layer and E₁A full connection layer for identifying correlation peak in the image, since only one of the segmented images has correlation peak in the whole correlation image, PC-CNN_SThe high recognition accuracy of (2) can reduce the number of segmented images;

f)PP-CNN_Scarrying out absolute position identification on a correlation peak value of the segmented image; PP-CNN_SIn which is N₂Maximum pooling layer, S₂A convolutional layer and E₂A full connection layer, a ReLU as a nonlinear activation function behind each layer, and PP-CNN after peak classification_SAnd carrying out absolute position identification on the segmentation image with the correlation peak value, wherein the segmentation image without the correlation peak value is only used for assisting calculation.

4. The method for capturing the weak signal of the navigation satellite based on the deep convolutional neural network as claimed in claim 1, wherein the step three is implemented by the following sub-steps:

g) in order to improve the absolute position precision of the correlation peak value, a region coding method is adopted; the resolution of the absolute position of the pixel decreases with the increase in the image size in a linear relationship with the reciprocal of the image size, and when the two-dimensional coordinates are the training target of the convolutional neural network, the positioning accuracy is affected as the image size increases, for example, when the image size is X₁×X₁According to the result of normalization, the coordinate resolution ratio of any two pixels is 1/X₁For X₂×X₂The coordinate resolution ratio of any two pixels is 1/X₂In the same convolution neural network, the larger the image size is, the lower the coordinate identification precision is, and the region coding method is adopted to divide the input image into a plurality of regions with respective X₁×X₁Area of coordinates, all areas of which are limited in coordinate range to X₁To X₂To (c) to (d);

h) obtaining PC-CNN_SThe optimal model of (2); the correlation peak image is divided into a plurality of correlation sub-images, only one of which has a correlation peak and a plurality of noise-containing images, which can be in PC-CNN_STo ChinesePrecision detection of images without correlation peaks, precision classification of images with correlation peaksHigher, yielding PC-CNN_SThe optimal model of (2);

i) obtaining PP-CNN_SThe optimal model of (2); altering PP-CNN_SNumber of convolution layers S of frame₂The optimal absolute position precision can be found, and the PP-CNN is obtained according to the recognition rate and the absolute positioning precision of different models_SThe model has the highest position precision and determines the PP-CNN_SModel, absolute positioning accuracy of the whole correlation image isAnd obtaining the absolute position precision of the final correlation peak.

The invention has the beneficial effects that: the method for capturing the weak signals of the navigation satellite based on the deep convolutional neural network establishes a deep convolutional neural network framework for image correlation peak value classification and correlation peak value positioning, is applied to detection of the weak signals of the navigation satellite, determines optimized network parameters and outputs optimal values by performing data training on the deep convolutional neural network, and improves the performance of capturing the weak signals of the navigation satellite.

Background

Because the deep Convolutional Neural network has a high precision for capturing weak navigation satellite signals, the Convolutional Neural Network (CNNs) technology has gained wide attention in the industry and academia. In a fading environment, satellite navigation signals are very weak, and the detection probability of a receiver by adopting a traditional acquisition algorithm is very low. In a satellite navigation system such as GPS, a code division multiple access technique is used, which has a spread spectrum signal power of about-128.5 dBm at the surface of the earth. The receiver acquires the signal by spreading the gain so that the correlation peak is larger than the noise power. In fading environments such as urban canyons, vegetation, or indoors, the GPS signal is further attenuated. Typical GPS signal attenuation is about 5-15 dB under vegetation, about 10-30 dB in urban canyons and over 25dB underground, so the acquisition threshold of a usable GPS receiver is at least better than-145 dBm. The traditional navigation satellite signal acquisition algorithm mainly adopts extra hardware cost to improve the acquisition probability. In the documents "Research of week GPS Signal Acquisition algorithms" (s.tie, and y.pi,2008International Conference on Communications, Circuits and Systems,2008.pp.793-796May,2008.) and "a Double Dwell High Sensitivity GPS Acquisition Scheme Using binary coherent Neural Network" (z.wang, y.zhuang, j.yang, h.zhang, w.dong, m.wang, l.hua, b.liu, and l.shi, Sensors 181482,2018), differential correlation, spatial diversity, and Acquisition of GPS signals are used to achieve a Weak Signal Acquisition, but other hardware costs are calculated. The current navigation satellite weak signal capturing method based on the deep convolutional neural network generally relates to a deep convolutional neural network framework of image correlation Peak Classification and correlation Peak positioning, wherein the deep convolutional neural network comprises a Peak Classification (PC) and a Peak Positioning (PP), the Classification of correlation peaks and the identification of absolute positions are respectively carried out on correlation images, and the performance of obtaining weak satellite navigation signals is improved.

However, in the current method for capturing the weak signals of the navigation satellite based on the deep convolutional neural network, a method capable of fully utilizing the deep convolutional neural network is still lack of successful precedent. The deep convolutional neural network is not fully utilized to capture signals, and the existing navigation satellite weak signal capturing method has the defects of poor system reliability and low capturing efficiency.

In the invention, the inventor provides a navigation satellite weak signal capturing method based on a deep convolutional neural network aiming at capturing the navigation satellite weak signal. In the method, a deep convolutional neural network framework for image correlation peak classification and correlation peak positioning is established, the deep convolutional neural network framework is applied to detection of the weak signals of the navigation satellite, and the method for capturing the weak signals of the navigation satellite is improved by performing data training on the deep convolutional neural network, determining optimized network parameters and outputting optimal values.

Disclosure of Invention

In order to overcome the defects of the technical problems, the invention provides a navigation satellite weak signal capturing method based on a deep convolutional neural network.

The invention discloses a navigation satellite weak signal capturing method based on a deep convolutional neural network, which is characterized by comprising the following steps of: the method comprises the following steps that firstly, a navigation satellite signal receiver obtains weak navigation satellite signals, the weak navigation satellite signals are processed by a radio frequency circuit, intermediate frequency signals r (t) are received, local generation signals s (t) are generated, and correlation operation is carried out between the intermediate frequency signals r (t) and the local generation signals s (t) to obtain correlation signals R (t) of the intermediate frequency signals r (t); detecting a correlation value of the signal by adopting an absolute position convolution neural network; thirdly, improving the absolute position precision of the correlation peak value by adopting a region coding method, training a data set and obtaining the PC-CNN_SAnd PP-CNN_SThe optimal model of (2).

The invention discloses a navigation satellite weak signal capturing method based on a deep convolutional neural network, which is realized by the following steps:

a) acquiring an intermediate frequency signal r (t); for a navigation satellite signal acquisition system of a deep convolutional neural network, after being processed by a radio frequency circuit, the received intermediate frequency signal is as follows:

r(t)＝AD_r(t)C_r(t-τ)×cos[2π(f_IF+f_d)t+φ₀]+n(t) (1)

in the formula (1), A is the signal amplitude, D_r(t) is modulated data, f_dIs the Doppler shift phi₀Is the original phase, C_r(t- τ) is a pseudo random code with a delay time;

b) the receiver generates a local generation signal s (t); for a deep convolutional neural network navigation satellite signal acquisition system, the receiver generates the locally generated signals as follows:

in the formula (2), the first and second groups,is provided with a delay timeThe local pseudo-random code of (a) is,is a local frequency shift, phi₁Is the initial phase;

c) obtaining a correlation signal R (t); performing correlation processing on the intermediate frequency signal r (t) and the local generation signal s (t), wherein the obtained correlation signals are as follows:

in the formula (3), the first and second groups,Δφ＝φ₀-φ₁，n_coris the equivalent noise that includes both gaussian noise and cross-correlation output.

The invention discloses a navigation satellite weak signal capturing method based on a deep convolutional neural network, which is realized by the following steps:

d) the correlator processes the signal to obtain a correlation image and inputs the correlation image into the convolutional neural network; the method comprises the following steps of performing parallel processing on signals by adopting M correlators to reduce acquisition time, dividing a whole related image into a plurality of related sub-images, and inputting the sub-images into a convolutional neural network;

e)PC-CNN_Sidentifying a correlation peak in the image; the correlation peak classification in the absolute position convolutional neural network has S₁A convolutional layer, an A × A kernel, M₁A maximum pooling layer and E₁A full connection layer for identifying correlation peak in the image, since only one of the segmented images has correlation peak in the whole correlation image, PC-CNN_SThe high recognition accuracy of (2) can reduce the number of segmented images;

f)PP-CNN_Scarrying out absolute position identification on a correlation peak value of the segmented image; PP-CNN_SIn which is N₂Maximum pooling layer, S₂A convolutional layer and E₂A modified Linear Unit (ReLU) as a nonlinear activation function is arranged behind each full connection layer, and PP-CNN is used after peak classification_SAnd carrying out absolute position identification on the segmentation image with the correlation peak value, wherein the segmentation image without the correlation peak value is only used for assisting calculation.

The invention discloses a navigation satellite weak signal capturing method based on a deep convolutional neural network, which is realized by the following steps:

g) in order to improve the absolute position precision of the correlation peak value, a region coding method is adopted; the resolution of the absolute position of the pixel decreases with the increase in the image size in a linear relationship with the reciprocal of the image size, and when the two-dimensional coordinates are the training target of the convolutional neural network, the positioning accuracy is affected as the image size increases, for example, when the image size is X₁×X₁According to the result of normalization, the coordinate resolution ratio of any two pixels is 1/X₁For X₂×X₂The coordinate resolution ratio of any two pixels is 1/X₂In the same convolution neural network, the larger the image size is, the more the coordinate recognition is accurateThe lower the degree, the method of region coding is adopted to divide the input image into a plurality of X-ray regions₁×X₁Area of coordinates, all areas of which are limited in coordinate range to X₁To X₂To (c) to (d);

h) obtaining PC-CNN_SThe optimal model of (2); the correlation peak image is divided into a plurality of correlation sub-images, only one of which has a correlation peak and a plurality of noise-containing images, which can be in PC-CNN_STo ChinesePrecision detection of images without correlation peaks, precision classification of images with correlation peaksHigher, yielding PC-CNN_SThe optimal model of (2);

i) obtaining PP-CNN_SThe optimal model of (2); altering PP-CNN_SNumber of convolution layers S of frame₂The optimal absolute position precision can be found, and the PP-CNN is obtained according to the recognition rate and the absolute positioning precision of different models_SThe model has the highest position precision and determines the PP-CNN_SModel, absolute positioning accuracy of the whole correlation image isAnd obtaining the absolute position precision of the final correlation peak.

The invention has the beneficial effects that: the method for capturing the weak signals of the navigation satellite based on the deep convolutional neural network establishes a deep convolutional neural network framework for image correlation peak value classification and correlation peak value positioning, is applied to detection of the weak signals of the navigation satellite, determines optimized network parameters and outputs optimal values by performing data training on the deep convolutional neural network, and improves the performance of capturing the weak signals of the navigation satellite.

Drawings

FIG. 1 is a schematic diagram of a convolutional neural network navigation satellite signal acquisition framework system;

FIG. 2 is a block diagram of a framework of an absolute position convolutional neural network;

FIG. 3 is a schematic diagram of region-encoded coordinates;

FIG. 4 is a simulation diagram of the classification accuracy of the correlation peak values of different models in PC-CNNs;

FIG. 5 is a graph of absolute position accuracy simulations of models with different numbers of layers in PP-CNNs.

Detailed Description

The invention is further described with reference to the following figures and examples.

The invention provides a specific implementation method for capturing a weak signal of a navigation satellite based on a deep convolutional neural network, which aims at capturing and transmitting the weak signal of the navigation satellite because the deep convolutional neural network is widely arranged in a system for capturing the weak signal of the navigation satellite. In order to improve the performance of signal capture, the invention establishes a deep convolutional neural network framework for image correlation peak classification and correlation peak positioning, determines optimized network parameters and outputs optimal values by carrying out data training on the deep convolutional neural network, and improves the method for obtaining the performance of the weak signals of the navigation satellite.

The invention provides a navigation satellite weak signal capturing method based on a deep convolutional neural network, which analyzes the influence of different convolutional layer numbers and different training data sets on the detection precision of the convolutional neural network by establishing a deep convolutional neural network framework for correlation peak classification and correlation peak absolute position identification, determines optimized network parameters and outputs an optimal value, and improves the performance of the convolutional neural network for capturing the navigation satellite weak signal.

As shown in fig. 1, a schematic diagram of a convolutional neural network navigation satellite signal acquisition framework system is that for a deep convolutional neural network navigation satellite signal acquisition system, M related channels are simultaneously provided to acquire navigation satellite weak signals of different initial phases, the navigation satellite weak signals are processed by a radio frequency circuit, an intermediate frequency signal r (t) is received, a receiver generates a local generated signal s (t), and a related signal r (t) is obtained by performing correlation operation between the intermediate frequency signal r (t) and the local generated signal s (t).

In the method provided by the invention, a system for capturing the weak signal of the navigation satellite based on the deep convolutional neural network receives the intermediate frequency signal after being processed by a radio frequency circuit

r(t)＝AD_r(t)C_r(t-τ)×cos[2π(f_IF+f_d)t+φ₀]+n(t) (1)

In the formula (1), A is the signal amplitude, D_r(t) is modulated data, C_r(t- τ) is a pseudo-random code with a delay time, f_dIs the Doppler shift phi₀Is the original phase. On the other hand, the receiver generates a local generation signal s (t), and for a navigation satellite weak signal acquisition system based on the deep convolutional neural network, the receiver generates the local generation signal s (t)

In the formula (2), the first and second groups,is provided with a delay timeThe local pseudo-random code of (a) is,is a local frequency shift, phi₁Is the initial phase. By operation, the correlation processing is carried out between the intermediate frequency signal r (t) and the local generation signal s (t) to obtain a correlation signal of

In the formula (3), the first and second groups,Δφ＝φ₀-φ₁，n_coris to include Gaussian noise and cross-correlation output equivalent noiseAnd (4) sound.

Fig. 2 shows a schematic diagram of a framework module structure of the absolute position convolution neural network. The absolute position convolution neural network has two deep networks, one is correlation peak classification, and the other is correlation peak positioning. PC-CNN_SThere are 2 convolutional layers, one 3 x 3 kernel, 2 max pooling layers and 2 full-link layers to identify correlation peaks in the image. PC-CNN since only one correlation image has a correlation peak in the entire correlation image_SThe high recognition accuracy of (2) can reduce the number of its divided images. After peak classification, PP-CNN_SAnd carrying out absolute position identification on the segmentation image with the correlation peak value, wherein the segmentation image without the correlation peak value is only used for assisting calculation. PP-CNN_SThere are N largest pooling layers, S convolutional layers and 2 fully-connected layers. For absolute position it is important that there is a zero-padding process in each layer. Behind each layer there is a ReLU as a non-linear activation function.

Fig. 3 shows a schematic diagram of region coding coordinates. In order to improve the absolute position accuracy of the correlation peak, a region coding method is adopted, and the resolution of the absolute position of the pixel is reduced along with the increase of the image size and has a linear relation with the reciprocal of the image size. When the two-dimensional coordinates are the training target of the convolutional neural network, the positioning accuracy is affected as the size of the image increases. For example, when the image size is 8 × 8, the coordinate resolution ratio of any two pixels is 1/8 ═ 0.125 according to the result of normalization. For a 32 × 32 image, the coordinate resolution ratio of any two pixels is 1/32 ═ 0.03125. In the same convolutional neural network, the larger the image size, the lower the coordinate recognition accuracy. The input image is divided into a plurality of areas with respective coordinates of 8 x 8 by adopting a regional coding method, and the coordinate range of all the areas is limited to be between 0 and 7.

FIG. 4 shows different models in PC-CNN_SThe classification accuracy of the correlation peak in (1) is simulated. The correlation peak image is divided into 832 images of 32 × 32 pixels, of which only one image with a correlation peak and 831 noisy images are present. Can be in PC-CNN_STo ChinesePrecision detection of images without correlation peaks, precision classification of images with correlation peaksHigher. The model of the correlation peak classification can be obtained by changing the maximum pooling layer number and the convolution layer number, namely a model 1, a model 2 and a model 3, wherein the model 1 is the correlation peak classification model with 2 maximum pooling layers and 2 convolution layers, the model 2 is the correlation peak classification model with 4 maximum pooling layers and 4 convolution layers, and the model 3 is the correlation peak classification model with 3 maximum pooling layers and 3 convolution layers. When the signal power is weak, the classification accuracy of the model 1 and the model 2 is obviously reduced, the classification accuracy of the model 3 is higher than that of the model 1 and the model 2, and the PC-CNN is determined_SIs model 3.

FIG. 5 shows the model with different number of layers in PP-CNN_SAbsolute position accuracy simulation diagram in (1). Analyzing the absolute position model with different maximum pooling layers and convolution layers, and changing PP-CNN_SThe convolution layer number S of the frame obtains four different absolute position models to find the best absolute position precision and obtain the PP-CNN with the highest position precision_SAnd (4) modeling. Model 1 is an absolute position model with 4 max pooling layers and 4 convolutional layers, model 2 is an absolute position model with 10 max pooling layers and 4 convolutional layers, model 3 is an absolute position model with 13 max pooling layers and 4 convolutional layers, and model 4 is an absolute position model with 14 max pooling layers and 4 convolutional layers. When the signal power is weak, the absolute position accuracy of the model 1, the model 2 and the model 4 is obviously reduced, the absolute position accuracy of the model 3 is higher than that of the model 1, the model 2 and the model 4, and the model 3 is determined to be PP-CNN according to the recognition rate and the absolute positioning accuracy of different models_SThe optimal model of (2). The absolute positioning accuracy of the whole related image isAnd obtaining the absolute position precision of the final correlation peak.

In summary, according to the navigation satellite weak signal capturing method based on the deep convolutional neural network, the deep convolutional neural network framework of image correlation peak classification and correlation peak positioning is established, the deep convolutional neural network is subjected to data training, network parameters are determined and optimized, an optimal value is output, and the method for capturing the navigation satellite weak signal performance is improved.

The above-described embodiment is only one embodiment of the present invention, and it will be apparent to those skilled in the art that various modifications and variations can be easily made based on the application and principle of the present invention disclosed in the present application, and the present invention is not limited to the method described in the above-described embodiment of the present invention, so that the above-described embodiment is only preferred, and not restrictive.