Synchronization head positioning method and device and signal synchronization method

Abstract

The invention discloses a synchronization head positioning method, a synchronization head positioning device and a signal synchronization method. The synchronization head positioning method comprises the following steps: acquiring a transmission signal; inputting the transmission signal into a synchronous head positioning neural network, carrying out feature extraction on the transmission signal through the synchronous head positioning neural network, generating a synchronous head guide mask according to the extracted features, and respectively carrying out calculation according to the synchronous head guide mask and the extracted different scale features to obtain synchronous head masks; calculating a mean value of all synchronous head masks; and determining the position of the synchronization head in the transmission signal according to the mean value of the synchronization head mask. According to the method, the transmission signal is input into the synchronous head positioning neural network to extract the features, the synchronous head guidance mask with the same scale as the transmission signal can be generated according to the extracted features, and the synchronous head mask calculated based on the synchronous head guidance mask can effectively enhance the representation of the synchronous head. The interference of noise on the synchronous head is suppressed to a certain extent, and the accuracy of positioning the synchronous head is improved.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a synchronization head positioning method, device and signal synchronization method. Background Art

[0002] In most communication systems, time synchronization is a fundamental and crucial step, serving as the first step in completing communication. Time synchronization of communication signals enables the receiver to determine the starting position of the code element (or transmitted waveform) within the received signal. Inaccurate time synchronization can lead to a range of issues at the receiving end, such as inter-symbol interference (ISI) and inter-symbol crosstalk (ISC), which can even lead to disconnection between communication devices.

[0003] Existing synchronization header detection methods typically use external synchronization. This involves inserting a synchronization header (also known as a sequence) with special correlation properties at the beginning of the transmitted signal. The receiver then locates the synchronization header through autocorrelation or cross-correlation. However, during signal transmission, the transmitted signal and synchronization header are affected by various noises, resulting in significant positioning errors and making accurate signal timing difficult to achieve. Summary of the Invention

[0004] The present invention provides a synchronization head positioning method, device and signal synchronization method to solve the technical problem that in the existing synchronization head positioning method, the transmission signal and the synchronization head are affected by various noises during the signal transmission process, resulting in large positioning errors and difficulty in accurately synchronizing the signal time.

[0005] The present invention provides a synchronization head positioning method, comprising: Acquire transmission signals; Inputting the transmission signal into a synchronization head positioning neural network, performing feature extraction on the transmission signal through the synchronization head positioning neural network, generating a synchronization head guidance mask based on the extracted features, and respectively calculating synchronization head masks based on the synchronization head guidance mask and the extracted features of different scales, and calculating the mean of all the synchronization head masks; The position of the synchronization header in the transmission signal is determined according to the average value of the synchronization header mask.

[0006] Furthermore, the synchronization head positioning neural network includes a feature extraction module, a mask generation module and a mask guidance module; The feature extraction module is configured to extract features from the transmission signal to obtain multi-scale features, and transmit the multi-scale features to the mask generation module and the mask guidance module; The mask generation module is configured to generate a synchronization header guidance mask having the same scale as the transmission signal according to the multi-scale features; The mask guidance module is configured to calculate a synchronization header mask based on the synchronization header mask and the multi-scale features.

[0007] Furthermore, the feature extraction module includes a first multi-scale feature extraction module, a first downsampling layer, a second multi-scale feature extraction module, a second downsampling layer, and a third multi-scale feature extraction module connected in series, and the output end of the first downsampling layer and the output end of the third multi-scale feature extraction module are respectively connected to the input end of the mask generation module; The number of output channels of the first multi-scale feature extraction module, the first downsampling layer, the second multi-scale feature extraction module, the second downsampling layer, and the third multi-scale feature extraction module increases sequentially.

[0008] Furthermore, each multi-scale feature extraction module includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a first dilated convolutional layer, a second dilated convolutional layer, and a third dilated convolutional layer; The input end of the first convolutional layer is used to receive the transmission signal / signal feature, and the output end of the first convolutional layer is connected to the input end of each hole convolutional layer and the input end of the second convolutional layer respectively; The output end of each of the dilated convolutional layers is connected to the input end of the second convolutional layer; The second convolutional layer, the third convolutional layer, and the fourth convolutional layer are connected in series, and the fourth convolutional layer outputs multi-scale features; the multi-scale feature extraction module is used to: After the first convolutional layer receives the transmission signal, the signal features of the transmission signal are extracted according to the first convolutional layer; different scale features of the signal features are obtained through the first dilated convolutional layer, the second dilated convolutional layer, and the third dilated convolutional layer, and the signal features and the different scale features are spliced; After compressing the number of channels of the spliced features through the second convolution layer, the spliced features are fused through the third convolution layer and the fourth convolution layer; The number of output channels of the first convolutional layer, the first dilated convolutional layer, the second convolutional layer, the third convolutional layer, and the fourth convolutional layer increases sequentially, and the number of output channels of the first dilated convolutional layer, the second dilated convolutional layer, and the third dilated convolutional layer are the same.

[0009] Furthermore, the mask generation module includes a fifth convolutional layer, a sixth convolutional layer, a first upsampling layer, a seventh convolutional layer, an eighth convolutional layer and a ninth convolutional layer; An input end of the fifth convolutional layer is connected to an output end of the first downsampling layer; An input end of the sixth convolutional layer is connected to an output end of the third multi-scale feature extraction module; An input end of the first upsampling layer is connected to an output end of the sixth convolutional layer; The output ends of the fifth convolutional layer and the first upsampling layer are respectively connected to the input end of the seventh convolutional layer, the seventh convolutional layer, the eighth convolutional layer and the ninth convolutional layer are connected in series, and the output end of the ninth convolutional layer outputs a synchronization header guidance mask; The mask generation module is further configured to: After the fifth convolution layer and the sixth convolution layer respectively receive different multi-scale features transmitted by the feature extraction module, the number of channels of the multi-scale features is compressed, the first upsampling layer adjusts the size of the compressed multi-scale features, and the features after the fifth convolution layer and the first upsampling layer are spliced. The seventh convolution layer and the eighth convolution layer are used to extract mask information from the spliced multi-scale features, and the ninth convolution layer is used to resize the extracted mask information, and a sigmoid function is used to constrain the value range to generate a synchronization header guidance mask with the same scale as the transmission signal; The number of output channels of the fifth convolutional layer is less than the number of output channels of the sixth convolutional layer, the number of output channels of the seventh convolutional layer and the number of output channels of the eighth convolutional layer increase successively, and the number of output channels of the ninth convolutional layer is 1.

[0010] Further, the mask guidance module includes a first mask guidance module, a second mask guidance module and a third mask guidance module; Each mask guidance module is connected to the output end of the mask generation module, the first mask guidance module and the second mask guidance module are connected to the output end of the first downsampling layer and the second downsampling layer respectively, and the third mask guidance module is connected to the output end of the third multi-scale feature extraction module; Each of the mask guidance modules includes: a third downsampling layer, a tenth convolutional layer, a first pooling layer, an eleventh convolutional layer, and a twelfth convolutional layer; The input end of the third downsampling layer is used to receive the synchronization header guidance mask; An output end of the third downsampling layer is connected to the tenth convolutional layer; An input end of the tenth convolutional layer is used to receive the multi-scale features and the small-size synchronization header guidance mask output by the third downsampling layer; An output end of the tenth convolutional layer is connected to an input end of the first pooling layer and an input end of the twelfth convolutional layer respectively; the first pooling layer, the eleventh convolutional layer and the twelfth convolutional layer are connected in series, and an output end of the twelfth convolutional layer outputs the synchronization header mask; The mask guidance module is used to: Multiplying the small-size synchronization header guidance mask output by the third downsampling layer by the input multi-scale feature, and then adding the mask to the multi-scale feature to obtain a feature to be processed; Compressing the number of channels of the feature to be processed according to the tenth convolutional layer to obtain a further feature to be processed; Passing the further processed features through the first pooling layer and then entering the eleventh convolutional layer, and applying a sigmoid function to constrain the value range of the result of the eleventh convolutional layer to obtain local channel attention; After multiplying the local channel attention and the feature to be processed, the channel number is compressed through the twelfth convolutional layer to obtain the synchronization header mask; The number of output channels of the tenth convolutional layer is greater than the number of output channels of the twelfth convolutional layer, and the number of output channels of the tenth convolutional layer is 1.

[0011] Furthermore, the training of the synchronization head positioning neural network includes: Acquire a plurality of training samples, each of the training samples corresponds to a sample label, and each of the sample labels is a synchronization head mask for indicating a synchronization head position in the training sample; Iteratively training a neural network using the training sample as input and the synchronization head mask mean and synchronization head guidance mask corresponding to the training sample as output, calculating a loss function value between the synchronization head mask mean, the synchronization head guidance mask, and the true synchronization head mask during the training process, and adjusting the parameters of the neural network according to the loss function value; The trained neural network is used as the synchronization head positioning neural network.

[0012] Furthermore, each training of the neural network includes: Inputting a number of training samples into the neural network so that the neural network outputs a synchronization head mask mean and a synchronization head guidance mask corresponding to each training sample; The total loss function value of the neural network is calculated based on the true synchronization head mask of each training sample, the synchronization head mask mean of the network output, and the synchronization head guidance mask; Determine whether the total loss function value of the neural network is less than a preset value. If so, complete the training of the neural network; if not, update the network parameters of the neural network according to the total loss function value of the neural network.

[0013] Furthermore, the loss function used to train the synchronization head positioning neural network includes: LOSS=λ1(L BCE_P +L IOU_P )+λ2(L BCE_G +L IOU_G )+λ3(Loffset_P_StartPoint +L offset_P_EndPoint )+λ4(L offset_G_StartPoint +L offset_G_EndPoint ) Among them, L BCE_P is the cross entropy loss function between the synchronization header mask mean and the sample label, L IOU_P is the IoU loss function between the synchronization head mask mean and the sample label, L BCE_G is the cross entropy loss function between the synchronization head guidance mask and the sample label, L IOU_G is the IoU loss function between the synchronization head guidance mask and the sample label, L offset_P_StartPoint is the absolute value of the difference between the synchronization header mask mean and the starting point of the sample label, L offfset_P_EndPoint is the absolute value of the difference between the synchronization header mask mean and the endpoint of the sample label, L offset_G_StartPoint is the absolute value of the difference between the synchronization header guidance mask and the starting point of the sample label, L offfset_G_EndPoint is the absolute value of the difference between the synchronization header guidance mask and the end point of the sample label. λ1, λ2, λ3 and λ4 are adjustable parameters.

[0014] The present invention also provides a synchronization head positioning device, comprising: A transmission signal acquisition module, used to acquire the transmission signal; a synchronization head mask mean calculation module, configured to input the transmission signal into a synchronization head positioning neural network, perform feature extraction on the transmission signal through the synchronization head positioning neural network, generate a synchronization head guidance mask based on the extracted features, calculate synchronization head masks based on the synchronization head guidance mask and different extracted features, and calculate the mean of all the synchronization head masks; A synchronization head positioning module is used to determine the position of the synchronization head in the transmission signal according to the average value of the synchronization head mask.

[0015] The present invention also provides a signal synchronization method, comprising: Acquire a transmission signal, and locate the synchronization head of the transmission signal according to the synchronization head positioning method as described above; The transmission signal is synchronized according to the synchronization header obtained by positioning.

[0016] The present invention inputs the transmission signal into the synchronization head positioning neural network to extract features. According to the extracted features, a synchronization head guidance mask with the same scale as the transmission signal can be generated. The synchronization head mask calculated based on the synchronization head guidance mask can effectively enhance the representation of the synchronization head and reduce the interference of noise on the synchronization head to a certain extent, thereby improving the accuracy of locating the synchronization head, and further effectively improving the accuracy of signal synchronization. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 1 is a flow chart of a synchronization head positioning method provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a synchronous head positioning neural network provided by an embodiment of the present invention; Figure 3 1 is a schematic structural diagram of a multi-scale feature extraction module provided by an embodiment of the present invention; Figure 4 1 is a structural diagram of a mask generation module provided by an embodiment of the present invention; Figure 5 1 is a structural diagram of a mask guidance module provided by an embodiment of the present invention; Figure 6 It is a structural schematic diagram of a synchronization head positioning device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0019] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0020] In the description of this application, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in this application based on the specific circumstances.

[0021] See also Figure 1 An embodiment of the present invention provides a synchronization head positioning method, comprising: S1. Obtain transmission signal; In the embodiment of the present invention, the transmission signal is provided with a synchronization header, and a sine wave with the highest frequency supported by the system sampling rate can be selected as the synchronization header inserted into the transmission signal to reduce noise interference on the synchronization header.

[0022] In one embodiment, other forms of synchronization headers may also be inserted, such as: (1) Preamble in IEEE 802.11, which consists of Short Training Field (STF) and Long Training Field (LTF); (2) Maximum-Length Sequence (m Sequence); (3) Gold sequence; (Gold is a person's name) (4) Zadoff–Chu sequence; (Zadoff and Chu are names of people) (5) Pseudo-random noise (PN); (6) Sine waves, square waves, triangle waves, etc. with different frequencies, S2. Input the transmission signal into the synchronization head positioning neural network, extract features of the transmission signal through the synchronization head positioning neural network, generate a synchronization head guidance mask based on the extracted features, and calculate synchronization head masks based on the synchronization head guidance mask and the extracted features of different scales, and calculate the mean of all synchronization head masks; In an embodiment of the present invention, a synchronization head guidance mask with the same scale as the transmission signal can be generated based on the extracted features. The synchronization head mask calculated based on the synchronization head guidance mask can enhance the representation of the synchronization head and reduce the interference of noise on the synchronization head to a certain extent, thereby reducing the positioning error of the synchronization head.

[0023] S3. Determine the position of the synchronization header in the transmission signal according to the mean value of the synchronization header mask.

[0024] In the embodiment of the present invention, the mean value of the synchronization header mask can completely display the position of the synchronization header in the transmission signal, so that the synchronization header can be accurately located according to the mean value of the synchronization header mask to complete signal synchronization.

[0025] In an embodiment of the present invention, a plurality of different synchronization head masks can be calculated based on the synchronization head guidance mask. These synchronization head masks act on features of different scales. By calculating the mean of all different synchronization head masks, a more accurate synchronization head mask can be obtained. The synchronization head mask can effectively enhance the representation of the synchronization head and suppress the interference of noise on the synchronization head to a certain extent, thereby effectively improving the accuracy of locating the synchronization head.

[0026] In an embodiment of the present invention, a transmission signal is input into a synchronization head positioning neural network to extract features. A synchronization head guidance mask having the same scale as the transmission signal can be generated based on the extracted features. The synchronization head mask calculated based on the synchronization head guidance mask can effectively enhance the representation of the synchronization head and suppress the interference of noise on the synchronization head to a certain extent, thereby improving the accuracy of positioning the synchronization head, and further effectively improving the accuracy of signal synchronization.

[0027] In one embodiment, the mean of the synchronization header mask may be binarized to obtain the corresponding mask, specifically: values of the synchronization mask mean greater than a preset threshold are set to 1, and values less than the preset threshold are set to 0. The preset threshold may be set to 0.55.

[0028] See also Figure 2 ,In one embodiment, the synchronous head positioning neural network includes a feature extraction module 10, a mask generation module 20, and a mask guidance module 30; The feature extraction module 10 is used to extract features from the transmission signal to obtain multi-scale features, and transmit the multi-scale features to the mask generation module 20 and the mask guidance module 30; A mask generation module 20 is configured to generate a synchronization header guidance mask having the same scale as the transmission signal based on the multi-scale features; The mask guidance module 30 is configured to calculate a synchronization header mask based on the synchronization header mask and the multi-scale features.

[0029] In one embodiment, the feature extraction module 10 includes a first multi-scale feature extraction module 101, a first downsampling layer 102, a second multi-scale feature extraction module 103, a second downsampling layer 104, and a third multi-scale feature extraction module 105 connected in series. The output end of the first downsampling layer 102 and the output end of the third multi-scale feature extraction module 105 are respectively connected to the input end of the mask generation module 20; The numbers of output channels of the first multi-scale feature extraction module 101 , the first downsampling layer 102 , the second multi-scale feature extraction module 103 , the second downsampling layer 104 and the third multi-scale feature extraction module 105 increase in sequence.

[0030] See also Figure 3 In one embodiment, each multi-scale feature extraction module includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a first atrous convolutional layer, a second atrous convolutional layer, and a third atrous convolutional layer; The input end of the first convolutional layer is used to receive the transmission signal / signal feature, and the output end of the first convolutional layer is connected to the input end of each hole convolutional layer and the input end of the second convolutional layer respectively; The output of each dilated convolutional layer is connected to the input of the second convolutional layer; The second convolutional layer, the third convolutional layer, and the fourth convolutional layer are connected in series, and the fourth convolutional layer outputs multi-scale features; The multi-scale feature extraction module is used to: After the first convolutional layer receives the transmission signal, the signal features of the transmission signal are extracted according to the first convolutional layer; The first, second and third dilated convolutional layers are used to obtain the different scale features of the signal features, and the signal features and the different scale features are spliced together; After the number of channels of the spliced features is compressed by the second convolutional layer, the spliced features are fused through the third and fourth convolutional layers; The number of output channels of the first convolutional layer, the first dilated convolutional layer, the second convolutional layer, the third convolutional layer, and the fourth convolutional layer increases successively, and the number of output channels of the first dilated convolutional layer, the second dilated convolutional layer, and the third dilated convolutional layer are the same.

[0031] In this embodiment of the present invention, when the multi-scale feature extraction module receives a transmission signal, it extracts shallow features or further features of the transmission signal through a convolution layer with a convolution kernel of 3. The first, second, and third dilated convolution layers are three dilated convolution layers with different expansion coefficients. These three dilated convolution layers are connected in parallel to form a dilated convolution layer group, which can observe the distribution of signal features from different perception fields of the features transmitted from the first convolution layer and obtain further features at different scales.

[0032] After the features output by the first convolutional layer and the features generated by the three dilated convolutional layers are spliced together, the number of channels increased after splicing is reduced by the second convolutional layer with a convolution kernel of 1. The features at different scales are fused through the third and fourth convolutional layers with convolution kernels of 3 to obtain multi-scale features.

[0033] See also Figure 4 In one embodiment, the mask generation module 20 includes a fifth convolutional layer, a sixth convolutional layer, a first upsampling layer, a seventh convolutional layer, an eighth convolutional layer, and a ninth convolutional layer; The input end of the fifth convolutional layer is connected to the output end of the first downsampling layer 102; The input end of the sixth convolutional layer is connected to the output end of the third multi-scale feature extraction module 105; The input of the first upsampling layer is connected to the output of the sixth convolutional layer; The output ends of the fifth convolutional layer and the first upsampling layer are respectively connected to the input end of the seventh convolutional layer. The seventh convolutional layer, the eighth convolutional layer, and the ninth convolutional layer are connected in series. The output end of the ninth convolutional layer outputs a synchronization header guidance mask. The mask generation module 20 is further configured to: After the fifth and sixth convolutional layers receive different multi-scale features transmitted by the feature extraction module, the number of channels of the multi-scale features is compressed, the first upsampling layer adjusts the size of the compressed multi-scale features, the features of the fifth and first upsampling layers are spliced, the seventh and eighth convolutional layers are used to extract mask information from the spliced multi-scale features, the ninth convolutional layer is used to resize the extracted mask information, and the sigmoid function is used to constrain the value range to generate a synchronization header guidance mask with the same scale as the transmitted signal; The number of output channels of the fifth convolutional layer is smaller than that of the sixth convolutional layer. The number of output channels of the seventh convolutional layer and the eighth convolutional layer increases successively. The number of output channels of the ninth convolutional layer is 1.

[0034] In an embodiment of the present invention, the mask generation module 20 can generate a synchronization header guidance mask based on the multi-scale features output by the first downsampling layer 102 and the third multi-scale feature extraction module 105. The multi-scale features output by the third multi-scale feature extraction module 105 are spliced with the previous multi-scale features after passing through the first upsampling layer, so as to integrate the contour information contained in the shallow features with the semantic information contained in the deep features.

[0035] In this embodiment of the present invention, the seventh and eighth convolutional layers with convolution kernels of 3 are used to extract mask information of the spliced features, and the ninth convolutional layer with a convolution kernel of 1 is used for size processing. The sigmoid function is used to constrain the value range to generate a synchronization header guidance mask with the same scale as the transmission signal.

[0036] The synchronization header guidance mask generated in the embodiment of the present invention is transmitted to the mask guidance module 30 .

[0037] Please continue reading Figure 2 ,In one embodiment, the mask guidance module 30 includes a first mask guidance module 301, a second mask guidance module 302 and a third mask guidance module 303; Each mask guidance module is connected to the output end of the mask generation module respectively, the first mask guidance module and the second mask guidance module are connected to the output ends of the first downsampling layer and the second downsampling layer respectively, and the third mask guidance module is connected to the output end of the third multi-scale feature extraction module.

[0038] In the embodiment of the present invention, the synchronization header guidance mask is transmitted to each mask guidance module respectively to generate a corresponding synchronization header mask.

[0039] See also Figure 5 ,In one embodiment, each mask guidance module includes: a third downsampling layer, a tenth convolutional layer, a first pooling layer, an eleventh convolutional layer, and a twelfth convolutional layer; An input end of the third downsampling layer is used to receive a synchronization header guidance mask; The output of the third downsampling layer is connected to the tenth convolutional layer; The input end of the tenth convolutional layer is used to receive the multi-scale features and the small-size synchronization header guidance mask output by the third downsampling layer; the output end of the tenth convolutional layer is connected to the input end of the first pooling layer and the input end of the twelfth convolutional layer respectively; the first pooling layer, the eleventh convolutional layer and the twelfth convolutional layer are connected in series, and the output end of the twelfth convolutional layer outputs the synchronization header mask; The mask guidance module 30 is used to: The small-size synchronization header guidance mask output by the third downsampling layer is multiplied by the input multi-scale feature and then added to the multi-scale feature to obtain the feature to be processed; According to the tenth convolutional layer, the number of channels of the features to be processed is compressed to obtain further features to be processed; The features to be further processed are passed through the first pooling layer and then enter the eleventh convolutional layer. The sigmoid function is used to constrain the value range of the result of the eleventh convolutional layer to obtain local channel attention; After multiplying the local channel attention with the feature to be processed, the number of channels is compressed through the twelfth convolutional layer to obtain the synchronization header mask; The number of output channels of the tenth convolutional layer is greater than that of the twelfth convolutional layer, and the number of output channels of the tenth convolutional layer is 1.

[0040] In this embodiment of the present invention, the synchronization head guidance mask is multiplied by the input multi-scale feature -i (i=1, 2, 3) after the downsampling layer and added to the original input multi-scale feature -i to enhance the weight of the synchronization head in the features obtained by the feature extraction network. In order to further enhance the feature representation, this embodiment of the present invention introduces a local attention mechanism (GAP) to explore key feature channels. The difference between this structure and the traditional attention mechanism calculation is that: (1) Reducing one convolutional layer after GAP to reduce network depth and computational complexity; (2) The mask guidance module introduces a local attention mechanism to calculate local attention and reduce complexity. For example, only the k neighbors of each channel are considered, the channel attention is multiplied by the input feature, and the final feature is obtained by reducing the number of channels through the twelfth convolutional layer.

[0041] It should be noted that the attention strategy in the mask guidance module of the embodiment of the present invention can highlight the key channels and suppress redundant channels or noise, so as to obtain a clearer synchronization head mask -k (k=1, 2, 3).

[0042] In one embodiment, training of the synchronization head positioning neural network includes: Obtaining a plurality of training samples, each training sample corresponds to a sample label, and each sample label is a synchronization head mask used to represent the synchronization head position in the training sample; The neural network is iteratively trained with training samples as input and the synchronization head mask mean and synchronization head guidance mask corresponding to the training samples as output. During the training process, the loss function value of the synchronization head mask mean, synchronization head guidance mask and the true synchronization head mask is calculated, and the parameters of the neural network are adjusted according to the loss function value. The trained neural network is used as the synchronization head positioning neural network.

[0043] In one embodiment, each training of the neural network includes: Inputting a number of training samples into the neural network so that the neural network outputs a synchronization head mask mean and a synchronization head guidance mask corresponding to each training sample; The total loss function value of the neural network is calculated based on the true synchronization head mask, the synchronization head mask mean and the synchronization head guidance mask of each training sample; Determine whether the total loss function value of the neural network is less than the preset value. If so, complete the training of the neural network; if not, update the network parameters of the neural network according to the total loss function value of the neural network.

[0044] In one embodiment, the loss function used to train the synchronization head positioning neural network includes: LOSS=λ1(L BCE_P +L IOU_P )+λ2(L BCE_G +L IOU_G )+λ3(L offset_P_StartPoint +L offset_P_EndPoint )+λ4(L offset_G_StartPoint +L offset_G_EndPoint ) Among them, L BCE_P is the cross entropy loss function between the synchronization header mask mean and the sample label, L IOU_P is the IoU loss function between the synchronization head mask mean and the sample label, L BCE_G is the cross entropy loss function between the synchronization head guidance mask and the sample label, L IOU_G is the IoU loss function between the synchronization head guidance mask and the sample label, L offset_P_StartPoint is the absolute value of the difference between the synchronization header mask mean and the starting point of the sample label, L offset_P_EndPoint is the absolute value of the difference between the synchronization header mask mean and the endpoint of the sample label, L offset_G_StartPoinnt is the absolute value of the difference between the synchronization header guidance mask and the starting point of the sample label, L offset_G_EndPoint is the absolute value of the difference between the synchronization header guidance mask and the endpoint of the sample label. λ1, λ2, λ3 and λ4 are adjustable parameters and can take values of 1, 1, 2, and 2 respectively.

[0045] In an embodiment of the present invention, the loss function is used for training a neural network. The goal of neural network training is to make the loss function as small as possible. The smaller the loss function, the smaller the error between the synchronization head position located by the neural network and the actual position of the synchronization head of the training sample.

[0046] The neural network in the embodiment of the present invention includes two types of supervision: the synchronization head mask output by the network (the mean of the synchronization head masks) and the synchronization head guidance mask. For these two types of results, the cross entropy loss function, the intersection-over-union (IoU) loss function, the loss function that calculates the absolute value of the difference between the starting points of the two, and the loss function that calculates the absolute value of the difference between the end points of the two are used.

[0047] The implementation of the present invention has the following beneficial effects: In an embodiment of the present invention, a transmission signal is input into a synchronization head positioning neural network to extract features. A synchronization head guidance mask with the same scale as the transmission signal can be generated based on the extracted features. The synchronization head mask calculated based on the synchronization head guidance mask can effectively enhance the representation of the synchronization head and suppress the interference of noise on the synchronization head to a certain extent, thereby improving the positioning accuracy of the synchronization head, and further effectively improving the accuracy of signal synchronization.

[0048] See also Figure 6 Based on the same inventive concept as the above embodiment, the present invention further provides a synchronization head positioning device, comprising: The transmission signal acquisition module 100 is used to acquire the transmission signal; The synchronization head mask mean calculation module 200 is used to input the transmission signal into the synchronization head positioning neural network, extract features of the transmission signal through the synchronization head positioning neural network, generate a synchronization head guidance mask based on the extracted features, and calculate synchronization head masks based on the synchronization head guidance mask and the extracted features of different scales, and calculate the mean of all synchronization head masks; The synchronization header positioning module 300 is configured to determine the location of the synchronization header in the transmission signal according to the mean value of the synchronization header mask.

[0049] In one embodiment, the synchronization head positioning neural network includes a feature extraction module, a mask generation module 20 and a mask guidance module; The feature extraction module 10 is used to extract features from the transmission signal to obtain multi-scale features, and transmit the multi-scale features to the mask generation module 20 and the mask guidance module 30; A mask generation module 20 is configured to generate a synchronization header guidance mask having the same scale as the transmission signal based on the multi-scale features; The mask guidance module 30 is configured to calculate a synchronization header mask based on the synchronization header mask and the multi-scale features.

[0050] In one embodiment, the feature extraction module 10 includes a first multi-scale feature extraction module 101, a first downsampling layer 102, a second multi-scale feature extraction module 103, a second downsampling layer 104, and a third multi-scale feature extraction module 105 connected in series, and the output end of the first downsampling layer 102 and the output end of the third multi-scale feature extraction module 105 are respectively connected to the input end of the mask generation module; The numbers of output channels of the first multi-scale feature extraction module 101 , the first downsampling layer 102 , the second multi-scale feature extraction module 103 , the second downsampling layer 104 and the third multi-scale feature extraction module 105 increase in sequence.

[0051] In one embodiment, each multi-scale feature extraction module includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a first dilated convolutional layer, a second dilated convolutional layer, and a third dilated convolutional layer; The input end of the first convolutional layer is used to receive the transmission signal / signal feature, and the output end of the first convolutional layer is connected to the input end of each hole convolutional layer and the input end of the second convolutional layer respectively; The output of each dilated convolutional layer is connected to the input of the second convolutional layer; The second convolutional layer, the third convolutional layer, and the fourth convolutional layer are connected in series, and the fourth convolutional layer outputs multi-scale features; The multi-scale feature extraction module is used to: After the first convolutional layer receives the transmission signal, the signal features of the transmission signal are extracted according to the first convolutional layer; The first, second and third dilated convolutional layers are used to obtain the different scale features of the signal features, and the signal features and the different scale features are spliced together; After the number of channels of the spliced features is compressed by the second convolutional layer, the spliced features are fused through the third and fourth convolutional layers; The number of output channels of the first convolutional layer, the first dilated convolutional layer, the second convolutional layer, the third convolutional layer, and the fourth convolutional layer increases successively, and the number of output channels of the first dilated convolutional layer, the second dilated convolutional layer, and the third dilated convolutional layer are the same.

[0052] In one embodiment, the mask generation module 20 includes a fifth convolutional layer, a sixth convolutional layer, a first upsampling layer, a seventh convolutional layer, an eighth convolutional layer, and a ninth convolutional layer; The input end of the fifth convolutional layer is connected to the output end of the first downsampling layer 102; The input end of the sixth convolutional layer is connected to the output end of the third multi-scale feature extraction module; The input of the first upsampling layer is connected to the output of the sixth convolutional layer; The output ends of the fifth convolutional layer and the first upsampling layer are respectively connected to the input end of the seventh convolutional layer. The seventh convolutional layer, the eighth convolutional layer, and the ninth convolutional layer are connected in series. The output end of the ninth convolutional layer outputs a synchronization header guidance mask. The mask generation module is also used to: After the fifth and sixth convolutional layers receive different multi-scale features transmitted by the feature extraction module, they compress the number of channels of the multi-scale features. The first upsampling layer adjusts the size of the compressed multi-scale features. The features of the fifth and first upsampling layers are spliced together. The seventh and eighth convolutional layers are used to extract mask information from the spliced multi-scale features. The ninth convolutional layer resizes the extracted mask information and uses a sigmoid function to constrain the value range to generate a synchronization header guidance mask with the same scale as the transmitted signal. The number of output channels of the fifth convolutional layer is smaller than that of the sixth convolutional layer. The number of output channels of the seventh convolutional layer and the eighth convolutional layer increases successively. The number of output channels of the ninth convolutional layer is 1.

[0053] In one embodiment, the mask guidance module includes a first mask guidance module 301, a second mask guidance module 302, and a third mask guidance module 303; Each mask guidance module is connected to the output end of the mask generation module respectively, the first mask guidance module and the second mask guidance module are connected to the output end of the first downsampling layer and the second downsampling layer respectively, and the third mask guidance module is connected to the output end of the third multi-scale feature extraction module; Each mask guidance module includes: the third downsampling layer, the tenth convolutional layer, the first pooling layer, the eleventh convolutional layer and the twelfth convolutional layer; An input end of the third downsampling layer is used to receive a synchronization header guidance mask; The output of the third downsampling layer is connected to the tenth convolutional layer; The input end of the tenth convolutional layer is used to receive the multi-scale features and the small-size synchronization header guidance mask output by the third downsampling layer; the output end of the tenth convolutional layer is connected to the input end of the first pooling layer and the input end of the twelfth convolutional layer respectively; the first pooling layer, the eleventh convolutional layer and the twelfth convolutional layer are connected in series, and the output end of the twelfth convolutional layer outputs the synchronization header mask; The mask guidance module is used to: The small-scale synchronization head guidance mask output by the third downsampling layer is multiplied by the input multi-scale feature and then added to the multi-scale feature to obtain the feature to be processed; According to the tenth convolutional layer, the number of channels of the features to be processed is compressed to obtain further features to be processed; The features to be further processed are passed through the first pooling layer and then enter the eleventh convolutional layer. The sigmoid function is used to constrain the value range of the result of the eleventh convolutional layer to obtain local channel attention; After multiplying the local channel attention with the feature to be processed, the number of channels is compressed through the twelfth convolutional layer to obtain the synchronization header mask; The number of output channels of the tenth convolutional layer is greater than that of the twelfth convolutional layer, and the number of output channels of the tenth convolutional layer is 1.

[0054] In one embodiment, training of the synchronization head positioning neural network includes: Obtaining a plurality of training samples, each training sample corresponds to a sample label, and each sample label is a synchronization head mask used to represent the synchronization head position in the training sample; The neural network is iteratively trained with training samples as input and the synchronization head mask mean and synchronization head guidance mask corresponding to the training samples as output. During the training process, the loss function value of the synchronization head mask mean, synchronization head guidance mask and the true synchronization head mask is calculated, and the parameters of the neural network are adjusted according to the loss function value. The trained neural network is used as the synchronization head positioning neural network.

[0055] In one embodiment, each training of the neural network includes: Inputting a number of training samples into the neural network so that the neural network outputs a synchronization head mask mean and a synchronization head guidance mask corresponding to each training sample; The total loss function value of the neural network is calculated based on the true synchronization head mask of each training sample, the synchronization head mask mean of the network output, and the synchronization head guidance mask; Determine whether the total loss function value of the neural network is less than the preset value. If so, complete the training of the neural network; if not, update the network parameters of the neural network according to the total loss function value of the neural network.

[0056] In one embodiment, the loss function used to train the synchronization head positioning neural network includes: LOSS=λ1(L BCE_P +L IOU_P )+λ2(L BCE_G +L IOU_G )+λ3(L offset_P_StartPoint +L offset_P_EndPoint )+λ4(L offset_G_StartPoint +L offset_G_EndPoint ) Among them, L BCE_P is the cross entropy loss function between the synchronization header mask mean and the sample label, L IOU_P is the IoU loss function between the synchronization head mask mean and the sample label, L BCE_Gis the cross entropy loss function between the synchronization head guidance mask and the sample label, L IOU_G is the IoU loss function between the synchronization head guidance mask and the sample label, L offset_P_StartPoint is the absolute value of the difference between the synchronization header mask mean and the starting point of the sample label, L offset_P_EndPoint is the absolute value of the difference between the synchronization header mask mean and the endpoint of the sample label, L offset_G_StartPoint is the absolute value of the difference between the synchronization header guidance mask and the starting point of the sample label, L offset_G_EndPoint is the absolute value of the difference between the synchronization header guidance mask and the end point of the sample label. λ1, λ2, λ3 and λ4 are adjustable parameters.

[0057] The present invention also provides a signal synchronization method, comprising: Acquire a transmission signal, and locate the synchronization head of the transmission signal according to the synchronization head positioning method; The transmission signal is synchronized according to the synchronization head obtained by positioning.

[0058] An embodiment of the present invention further provides a computer-readable storage medium, which includes a stored computer program. When the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the synchronization head positioning method as described above.

[0059] An embodiment of the present invention further provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the synchronization head positioning method described above is implemented.

[0060] An embodiment of the present invention further provides a computer program product. When the computer program product is run on a terminal device, the terminal device is enabled to execute the synchronization head positioning method as described above.

[0061] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A synchronization head positioning method, characterized in that: include: Acquire transmission signals; Inputting the transmission signal into a synchronization head positioning neural network, performing feature extraction on the transmission signal through the synchronization head positioning neural network, generating a synchronization head guidance mask based on the extracted features, and respectively calculating synchronization head masks based on the synchronization head guidance mask and the extracted features of different scales, and calculating the mean of all the synchronization head masks; The position of the synchronization header in the transmission signal is determined according to the average value of the synchronization header mask.

2. The synchronization head positioning method according to claim 1, wherein: The synchronization head positioning neural network includes a feature extraction module, a mask generation module and a mask guidance module; The feature extraction module is configured to extract features from the transmission signal to obtain multi-scale features, and transmit the multi-scale features to the mask generation module and the mask guidance module; The mask generation module is configured to generate a synchronization header guidance mask having the same scale as the transmission signal according to the multi-scale features; The mask guidance module is configured to calculate a synchronization header mask based on the synchronization header mask and the multi-scale features.

3. The synchronization head positioning method according to claim 2, wherein: The feature extraction module includes a first multi-scale feature extraction module, a first downsampling layer, a second multi-scale feature extraction module, a second downsampling layer, and a third multi-scale feature extraction module connected in series, wherein the output end of the first downsampling layer and the output end of the third multi-scale feature extraction module are respectively connected to the input end of the mask generation module; The number of output channels of the first multi-scale feature extraction module, the first downsampling layer, the second multi-scale feature extraction module, the second downsampling layer, and the third multi-scale feature extraction module increases sequentially.

4. The synchronization head positioning method according to claim 3, wherein: Each multi-scale feature extraction module includes a first convolution layer, a second convolution layer, a third convolution layer, a fourth convolution layer, a first dilated convolution layer, a second dilated convolution layer, and a third dilated convolution layer; The input end of the first convolutional layer is used to receive the transmission signal / signal feature, and the output end of the first convolutional layer is connected to the input end of each hole convolutional layer and the input end of the second convolutional layer respectively; The output end of each of the dilated convolutional layers is connected to the input end of the second convolutional layer; The second convolutional layer, the third convolutional layer, and the fourth convolutional layer are connected in series, and the fourth convolutional layer outputs multi-scale features; The multi-scale feature extraction module is used to: After the first convolutional layer receives the transmission signal, extracting signal features of the transmission signal according to the first convolutional layer; Acquire different-scale features of the signal feature through the first atrous convolution layer, the second atrous convolution layer, and the third atrous convolution layer, and concatenate the signal feature and the different-scale features; After compressing the number of channels of the spliced features through the second convolution layer, the spliced features are fused through the third convolution layer and the fourth convolution layer; The number of output channels of the first convolutional layer, the first dilated convolutional layer, the second convolutional layer, the third convolutional layer, and the fourth convolutional layer increases sequentially, and the number of output channels of the first dilated convolutional layer, the second dilated convolutional layer, and the third dilated convolutional layer are the same.

5. The synchronization head positioning method according to claim 3, wherein: The mask generation module includes a fifth convolutional layer, a sixth convolutional layer, a first upsampling layer, a seventh convolutional layer, an eighth convolutional layer and a ninth convolutional layer; An input end of the fifth convolutional layer is connected to an output end of the first downsampling layer; An input end of the sixth convolutional layer is connected to an output end of the third multi-scale feature extraction module; An input end of the first upsampling layer is connected to an output end of the sixth convolutional layer; The output ends of the fifth convolutional layer and the first upsampling layer are respectively connected to the input end of the seventh convolutional layer, the seventh convolutional layer, the eighth convolutional layer and the ninth convolutional layer are connected in series, and the output end of the ninth convolutional layer outputs a synchronization header guidance mask; The mask generation module is further configured to: After the fifth convolution layer and the sixth convolution layer respectively receive different multi-scale features transmitted by the feature extraction module, the number of channels of the multi-scale features is compressed, the first upsampling layer adjusts the size of the compressed multi-scale features, and the features after the fifth convolution layer and the first upsampling layer are spliced. The seventh convolution layer and the eighth convolution layer are used to extract mask information from the spliced multi-scale features, and the ninth convolution layer is used to resize the extracted mask information, and a sigmoid function is used to constrain the value range to generate a synchronization header guidance mask with the same scale as the transmission signal; The number of output channels of the fifth convolutional layer is less than the number of output channels of the sixth convolutional layer, the number of output channels of the seventh convolutional layer and the number of output channels of the eighth convolutional layer increase successively, and the number of output channels of the ninth convolutional layer is 1.

6. The synchronization head positioning method according to claim 3, wherein: The mask guidance module includes a first mask guidance module, a second mask guidance module and a third mask guidance module; Each mask guidance module is connected to the output end of the mask generation module, the first mask guidance module and the second mask guidance module are connected to the output end of the first downsampling layer and the second downsampling layer respectively, and the third mask guidance module is connected to the output end of the third multi-scale feature extraction module; Each of the mask guidance modules includes: a third downsampling layer, a tenth convolutional layer, a first pooling layer, an eleventh convolutional layer, and a twelfth convolutional layer; The input end of the third downsampling layer is used to receive the synchronization header guidance mask; An output end of the third downsampling layer is connected to the tenth convolutional layer; An input end of the tenth convolutional layer is used to receive the multi-scale features and the small-size synchronization header guidance mask output by the third downsampling layer; An output end of the tenth convolutional layer is connected to an input end of the first pooling layer and an input end of the twelfth convolutional layer, respectively; the first pooling layer, the eleventh convolutional layer, and the twelfth convolutional layer are connected in series, and an output end of the twelfth convolutional layer outputs the synchronization header mask; The mask guidance module is used to: Multiplying the small-size synchronization header guidance mask output by the third downsampling layer by the input multi-scale feature, and then adding the mask to the multi-scale feature to obtain a feature to be processed; Compressing the number of channels of the feature to be processed according to the tenth convolutional layer to obtain a further feature to be processed; Passing the further processed features through the first pooling layer and then entering the eleventh convolutional layer, and applying a sigmoid function to constrain the value range of the result of the eleventh convolutional layer to obtain local channel attention; After multiplying the local channel attention and the feature to be processed, the channel number is compressed through the twelfth convolutional layer to obtain the synchronization header mask; The number of output channels of the tenth convolutional layer is greater than the number of output channels of the twelfth convolutional layer, and the number of output channels of the tenth convolutional layer is 1.

7. The synchronization head positioning method according to any one of claims 1 to 6, characterized in that: The training of the synchronization head positioning neural network includes: Acquire a plurality of training samples, each of the training samples corresponds to a sample label, and each of the sample labels is a synchronization head mask for indicating a synchronization head position in the training sample; Iteratively training a neural network using the training sample as input and the synchronization head mask mean and synchronization head guidance mask corresponding to the training sample as output, calculating a loss function value between the synchronization head mask mean, the synchronization head guidance mask, and the true synchronization head mask during the training process, and adjusting the parameters of the neural network according to the loss function value; The trained neural network is used as the synchronization head positioning neural network.

8. The synchronization head positioning method according to any one of claims 1 to 6, characterized in that: Each training of a neural network includes: Inputting a number of training samples into the neural network so that the neural network outputs a synchronization head mask mean and a synchronization head guidance mask corresponding to each training sample; The total loss function value of the neural network is calculated based on the true synchronization head mask of each training sample, the synchronization head mask mean of the network output, and the synchronization head guidance mask; Determine whether the total loss function value of the neural network is less than a preset value. If so, complete the training of the neural network; if not, update the network parameters of the neural network according to the total loss function value of the neural network.

9. The synchronization head positioning method according to claim 1, wherein: The loss function used to train the synchronization head positioning neural network includes: LOSS=λ1(L BCE_P +L IOU_P )+λ2(L BCE_G +L IOU_G )+λ3(L offset_P_StartPoint +L offset_P_EndPoint )+λ4(L offset_G_StartPoint +L offset_G_EndPoint ) Among them, L BCE_P is the cross entropy loss function between the synchronization header mask mean and the sample label, L IOU_P is the IoU loss function between the synchronization head mask mean and the sample label, L BCE_G is the cross entropy loss function between the synchronization head guidance mask and the sample label, L IOU_G is the IoU loss function between the synchronization head guidance mask and the sample label, L offset_P_startPoint is the absolute value of the difference between the synchronization header mask mean and the starting point of the sample label, L offset_P_EndPoint is the absolute value of the difference between the synchronization header mask mean and the endpoint of the sample label, L offset_G_startPoint is the absolute value of the difference between the synchronization header guidance mask and the starting point of the sample label, L offset_G_EndPoint is the absolute value of the difference between the synchronization header guidance mask and the end point of the sample label. λ1, λ2, λ3 and λ4 are adjustable parameters.

10. A synchronous head positioning device, characterized in that: include: Transmission signal acquisition module, used to obtain transmission signals: a synchronization head mask mean calculation module, configured to input the transmission signal into a synchronization head positioning neural network, perform feature extraction on the transmission signal through the synchronization head positioning neural network, generate a synchronization head guidance mask based on the extracted features, calculate synchronization head masks based on the synchronization head guidance mask and different extracted features, and calculate the mean of all the synchronization head masks; A synchronization head positioning module is used to determine the position of the synchronization head in the transmission signal according to the average value of the synchronization head mask.

11. A signal synchronization method, characterized in that: include: Acquire a transmission signal, and locate the synchronization head of the transmission signal according to the synchronization head positioning method according to any one of claims 1 to 9; The transmission signal is synchronized according to the synchronization header obtained by positioning.

Images