A Method for Bandwidth Extension of Satellite-Ground Channel Detection Using Autoencoder

Abstract

The present invention relates to the technical field of signal detection, and particularly to a method for extending the detection bandwidth of a satellite-ground channel using an autoencoder. The method includes: preprocessing the time-domain data signal to obtain power spectral density data; using a preset spectral prediction network model to extract the power spectral density features of the power spectral density data, wherein the preset spectral prediction network model is obtained by training a fully connected neural network with broadband data; based on the power spectral density features, using the preset spectral prediction network model to predict and complete the missing part in the residual spectral signal and restore the original spectral shape. Thus, by learning the spectrum of broadband data in offline training and using a fully connected neural network to implement residual spectrum detection and spectral prediction and completion after extracting spectral features, the problem in the prior art that the spectral prediction result is inaccurate due to factors such as the ADC receiving sampling rate and broadband resolution is solved, and the spectral prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of signal detection, and particularly to a method for extending the detection bandwidth of satellite-ground channels using an autoencoder. Background Art

[0002] Driven by the global demand for information application, the demand for data transmission rate in hot areas of satellite communication such as satellite navigation, space-based Internet, global Internet of Things, and 6G communication is increasing day by day. However, achieving a higher data transmission rate usually requires a corresponding increase in the signal bandwidth. To effectively sample the output signal of a power amplifier (PA), a high-speed analog-to-digital converter (ADC) is usually required.

[0003] The ADC sampling rate of traditional signal transmitters needs to reach several times the original signal bandwidth. The maximum reception sampling rate of the ADC is limited by its hardware design. Once the signal bandwidth exceeds this range, the signal will suffer varying degrees of attenuation, thus affecting the normal reception of the signal. Similarly, in the face of the demand for broadband high-resolution spectrum detection, existing spectrum sensing technologies can only be implemented within the maximum reception sampling bandwidth of the ADC. For channels beyond the ADC sampling bandwidth, it is often difficult to accurately obtain the signal state or even impossible to sense. In the field of satellite communication, the ADC reception bandwidth and power of spaceborne signal processing receivers are strictly limited, which poses a great challenge to the broadband signal spectrum sensing in the satellite communication process and urgently needs to be solved. Summary of the Invention

[0004] The present invention provides a method for extending the detection bandwidth of satellite-ground channels using an autoencoder to solve the problem that the spectrum prediction result is inaccurate due to factors such as the ADC reception sampling rate and broadband resolution in the prior art, and to improve the spectrum prediction accuracy.

[0005] To achieve the above object, the first aspect embodiment of the present invention proposes a method for extending the detection bandwidth of satellite-ground channels using an autoencoder, including the following steps:

[0006] Obtain a time-domain data signal, and perform preprocessing on the time-domain data signal to obtain power spectral density data;

[0007] Use a preset spectrum prediction network model to extract the power spectral density features of the power spectral density data, where the preset spectrum prediction network model is obtained by training a fully connected neural network with broadband data;

[0008] Based on the power spectral density features, use the preset spectrum prediction network model to predict and complete the missing part in the residual spectrum signal, and restore the original spectrum shape.

[0009] According to an embodiment of the present invention, before extracting the power spectral density features of the power spectral density data by using the preset spectrum prediction network model, the following steps are further included:

[0010] Obtain the broadband data, and construct a training and test sample data group by using the broadband data, where the broadband data is sampled by an analog-to-digital converter (ADC);

[0011] Preprocess the training and test sample data group, and obtain a training set and a validation set based on the preprocessed training and test sample data group;

[0012] Based on a preset loss function, train the fully connected neural network by using the training set to obtain an initial neural network model, and verify the initial neural network model by using the validation set until the initial neural network model meets the preset criteria, and end the iterative training of the fully connected neural network to obtain the preset spectrum prediction network model; otherwise, adjust the model parameters and continue the iterative training.

[0013] According to an embodiment of the present invention, the step of preprocessing the training and test sample data group and obtaining a training set and a validation set based on the preprocessed training and test sample data group includes:

[0014] Divide the training and test sample data group into low-speed ADC-sampled broadband data and high-speed ADC-sampled broadband data at corresponding times;

[0015] Based on a preset frame length and a preset step size, perform segmentation processing on the low-speed ADC-sampled broadband data and the high-speed ADC-sampled broadband data at corresponding times respectively, use the segmented low-speed ADC-sampled broadband data as training data, and use the segmented high-speed ADC-sampled broadband data as validation data;

[0016] Perform fast Fourier transform processing with a first preset length on each row of data in each data group of the training data and the validation data respectively to obtain a broadband spectrum;

[0017] Take the modulus value of the broadband spectrum to obtain a power spectral density estimation result;

[0018] Divide the power spectral density estimation result at intervals of a second preset length to obtain the training set and the validation set.

[0019] According to an embodiment of the present invention, after obtaining the training set and the validation set based on the preprocessed training and test sample data group, the following steps are further included:

[0020] Perform normalization processing on the training set to obtain a normalized training set;

[0021] Activate each training sample in the normalized training set to obtain the first feature vector of each training sample;

[0022] Activate the channel dimension of the first feature vector of each training sample respectively to obtain the second feature vector of each training sample;

[0023] Integrate the first feature vector of each training sample and the second feature vector of the corresponding training sample to obtain the channel feature result of each training sample.

[0024] According to an embodiment of the present invention, after obtaining the channel feature result of each training sample, it further includes:

[0025] Activate the channel feature result of each training sample respectively to obtain the third feature vector of each training sample;

[0026] Activate the channel dimension of the third feature vector of each training sample respectively to obtain the fourth feature vector of each training sample;

[0027] Integrate and predict the third feature vector of each training sample and the fourth feature vector of the corresponding sample to obtain the predicted channel power spectral density of each training sample;

[0028] Based on a preset splicing strategy, splice the predicted channel power spectral density of each training sample to obtain a complete power spectral density result.

[0029] According to an embodiment of the present invention, the preset loss function is:

[0030]

[0031] Wherein, is the complete power spectral density result, is the validation set, is the preset frame length, is the number of rows of the power spectral density result.

[0032] A method for extending the bandwidth of satellite-ground channel detection of an autoencoder according to an embodiment of the present invention can obtain a time-domain data signal, and preprocess the time-domain data signal to obtain power spectral density data; then, using a preset spectral prediction network model, the power spectral density features of the power spectral density data can be extracted; based on the power spectral density features, using the preset spectral prediction network model, the missing part in the residual spectral signal is predicted and complemented, so as to restore the original spectral shape. Thus, by learning the spectrum of broadband data in offline training and using a fully connected neural network to implement residual spectrum detection and spectrum prediction and complementation after extracting spectral features, the problem that the spectrum prediction result is inaccurate due to factors such as the ADC reception sampling rate and broadband resolution in the prior art is solved, and the spectrum prediction accuracy is improved.

[0033] To achieve the above object, an embodiment of the second aspect of the present invention proposes a device for extending the bandwidth of satellite-ground channel detection of an autoencoder, including:

[0034] An acquisition module, configured to acquire a time-domain data signal, and preprocess the time-domain data signal to obtain power spectral density data;

[0035] An extraction module, configured to use a preset spectral prediction network model to extract the power spectral density features of the power spectral density data, wherein the preset spectral prediction network model is obtained by training a fully connected neural network with broadband data;

[0036] A prediction module, configured to predict and complement the missing part in the residual spectral signal based on the power spectral density features, and restore the original spectral shape by using the preset spectral prediction network model.

[0037] According to an embodiment of the present invention, before using the preset spectral prediction network model to extract the power spectral density features of the power spectral density data, the extraction module includes:

[0038] A construction unit, configured to acquire the broadband data, and use the broadband data to construct a training and test sample data group, wherein the broadband data is sampled by an analog-to-digital converter (ADC);

[0039] A preprocessing unit, configured to preprocess the training and test sample data group, and obtain a training set and a validation set based on the preprocessed training and test sample data group;

[0040] A training unit, which is used to train the fully-connected neural network based on a preset loss function by using the training set to obtain an initial neural network model, and verify the initial neural network model by using the validation set until the initial neural network model meets the preset criteria, and then end the iterative training of the fully-connected neural network to obtain the preset spectrum prediction network model; otherwise, adjust the model parameters and continue the iterative training.

[0041] According to an embodiment of the present invention, the preprocessing unit is specifically configured to:

[0042] Divide the training and test sample data group into low-speed ADC sampled broadband data and high-speed ADC sampled broadband data at corresponding times;

[0043] Based on a preset frame length and a preset step size, perform segmentation processing on the low-speed ADC sampled broadband data and the high-speed ADC sampled broadband data at corresponding times respectively, use the segmented low-speed ADC sampled broadband data as training data, and use the segmented high-speed ADC sampled broadband data as validation data;

[0044] Perform fast Fourier transform processing with a first preset length on each row of data in each data group of the training data and the validation data respectively to obtain a broadband spectrum;

[0045] Take the modulus value of the broadband spectrum to obtain a power spectral density estimation result;

[0046] Divide the power spectral density estimation result at intervals of a second preset length to obtain the training set and the validation set.

[0047] According to an embodiment of the present invention, after obtaining the training set and the validation set based on the preprocessed training and test sample data group, the preprocessing unit further includes:

[0048] A processing subunit, which is used to perform normalization processing on the training set to obtain a normalized training set;

[0049] A first activation subunit, which is used to activate each training sample in the normalized training set to obtain a first feature vector of each training sample;

[0050] A second activation subunit, which is used to activate the channel dimension of the first feature vector of each training sample respectively to obtain a second feature vector of each training sample;

[0051] An integration subunit, which is used to integrate the first feature vector of each training sample and the second feature vector of the corresponding training sample to obtain a channel feature result of each training sample.

[0052] According to an embodiment of the present invention, after obtaining the channel feature results of each of the training samples, the integration subunit is further configured to:

[0053] Activate the channel feature results of each of the training samples respectively to obtain the third feature vector of each of the training samples;

[0054] Activate the channel dimension of the third feature vector of each of the training samples respectively to obtain the fourth feature vector of each of the training samples;

[0055] Integrate and predict the third feature vector of each of the training samples and the fourth feature vector of the corresponding sample to obtain the predicted channel power spectral density of each of the training samples;

[0056] Based on a preset splicing strategy, splice the predicted channel power spectral density of each of the training samples to obtain a complete power spectral density result.

[0057] According to an embodiment of the present invention, the preset loss function is:

[0058]

[0059] Wherein, is the complete power spectral density result, is the validation set, is the preset frame length, is the number of rows of the power spectral density result.

[0060] A satellite-ground channel detection bandwidth extension device of an autoencoder proposed according to an embodiment of the present invention can obtain power spectral density data by acquiring a time-domain data signal and preprocessing the time-domain data signal; then, using a preset spectrum prediction network model, the power spectral density features of the power spectral density data can be extracted; based on the power spectral density features, using the preset spectrum prediction network model, predict and complete the missing part in the residual spectrum signal, so as to restore the original spectrum shape. Thus, by learning the spectrum of broadband data in offline training and using a fully connected neural network to implement residual spectrum detection and spectrum prediction and completion after extracting spectrum features, the problem that the spectrum prediction result is inaccurate due to factors such as the ADC receiving sampling rate and broadband resolution in the prior art is solved, and the spectrum prediction accuracy is improved.

[0061] To achieve the above object, an embodiment of the third aspect of the present invention proposes an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the program to implement a satellite-ground channel detection bandwidth extension method of an autoencoder as described in the above embodiment.

[0062] To achieve the above object, an embodiment of the fourth aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored, and the program is executed by a processor to implement a method for extending the bandwidth of satellite-ground channel detection of an autoencoder as described in the above embodiment.

[0063] To achieve the above object, an embodiment of the fifth aspect of the present invention provides a computer program product, which includes a computer program, and when the program is executed by a processor, it is used to implement a method for extending the bandwidth of satellite-ground channel detection of an autoencoder as described in the above embodiment.

[0064] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. Description of the Drawings

[0065] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the following description of the embodiments in conjunction with the drawings, where:

[0066] Figure 1 FIG. is a flowchart of a method for extending the bandwidth of satellite-ground channel detection of an autoencoder according to an embodiment of the present invention;

[0067] Figure 2 FIG. is a schematic diagram of the working process of a method for extending the bandwidth of satellite-ground channel detection of an autoencoder according to an embodiment of the present invention;

[0068] Figure 3 FIG. is a flowchart of the training process of a spectrum prediction network model according to an embodiment of the present invention;

[0069] Figure 4 FIG. is a schematic diagram of the effect of spectrum detection bandwidth extension according to an embodiment of the present invention;

[0070] Figure 5 FIG. is a schematic diagram of the effect of spectrum detection bandwidth extension according to another embodiment of the present invention;

[0071] Figure 6 FIG. is a block diagram of a device for extending the bandwidth of satellite-ground channel detection of an autoencoder according to an embodiment of the present invention;

[0072] Figure 7 FIG. is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Embodiments

[0073] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention, but should not be construed as limiting the present invention.

[0074] A method for extending the bandwidth of satellite-ground channel detection of an autoencoder according to an embodiment of the present invention will be described below with reference to the accompanying drawings.

[0075] Figure 1 It is a flowchart of a method for extending the bandwidth of satellite-ground channel detection of an autoencoder according to an embodiment of the present invention.

[0076] Exemplarily, as Figure 1 shown, the method for extending the bandwidth of satellite-ground channel detection of an autoencoder includes the following steps:

[0077] In step S101, time-domain data signals are acquired, and the time-domain data signals are preprocessed to obtain power spectral density data.

[0078] Specifically, first, time-domain data signals can be collected from various sensors or data acquisition devices. Since these signals usually contain various noises and interferences, further processing can be performed to ensure the quality and accuracy of the data. Next, the acquired time-domain data signals are preprocessed. For example, the signals are filtered to remove high-frequency noises or low-frequency interferences, thereby improving the signal quality; the signals are normalized to be within a specific range for subsequent analysis and processing; the signals are detrended to eliminate the long-term trend in the signals and make them more stable, etc. After these preprocessing steps, more reliable time-domain data signals can be obtained. Sampling the preprocessed signals using an ADC can obtain broadband data, and performing a fast Fourier transform on the broadband data to convert it from the time domain to the frequency domain. Through the fast Fourier transform, the frequency components of the signal can be obtained, and the power spectral density data can be calculated. The power spectral density data reflects the energy distribution of the signal at different frequencies, which lays a foundation for further signal analysis and processing.

[0079] In step S102, a power spectral density feature of the power spectral density data is extracted by using a preset spectral prediction network model, where the preset spectral prediction network model is obtained by training a fully connected neural network with broadband data.

[0080] Among them, the Power Spectral Density (PSD) is a function that describes the power distribution of a signal varying with frequency; a fully connected neural network refers to a network structure in deep learning where each neuron is connected to all neurons in the previous layer and can process complex data patterns.

[0081] That is to say, by using broadband data to train a fully connected neural network, a preset spectrum prediction network model can be obtained. This preset spectrum prediction network model is essentially an artificial intelligence algorithm. Through the fully connected spectrum feature extraction module in the preset spectrum prediction network model, the power spectral density features in the power spectral density data can be effectively extracted. Here, the spectral information that is severely attenuated in the power spectral density is mainly concerned, thereby accurately locating the position and shape of the spectral signal before attenuation.

[0082] In step S103, based on the power spectral density features, using the preset spectrum prediction network model, predict and complete the missing part in the residual spectral signal to restore the original spectral shape.

[0083] It can be understood that after obtaining the power spectral density features, further call the fully connected spectrum feature prediction and extension module in the preset spectrum prediction network model, infer the data of the missing part according to the known spectral information, so as to restore a complete original spectral shape and realize the prediction and generation of the residual spectral signal to complete the complement.

[0084] For the convenience of understanding, the training process of the preset spectrum prediction network model will be elaborated in detail below.

[0085] As a possible implementation method, in some embodiments, before using the preset spectrum prediction network model to extract the power spectral density features of the power spectral density data, it further includes: obtaining broadband data and constructing a training and test sample data group using the broadband data, where the broadband data is sampled by an analog-to-digital converter (ADC); preprocessing the training and test sample data group and obtaining a training set and a validation set based on the preprocessed training and test sample data group; training a fully connected neural network using the training set based on a preset loss function to obtain an initial neural network model, and validating the initial neural network model using the validation set until the initial neural network model meets the preset criteria, ending the iterative training of the fully connected neural network to obtain the preset spectrum prediction network model, otherwise adjusting the model parameters and continuing the iterative training.

[0086] Specifically, in order to train a preset spectral prediction network model, first, the broadband data sampled by ADC can be used to construct a training and test sample data set for inputting into a fully connected neural network. These sample data sets are the basis for subsequent analysis and model training. Next, preprocessing is performed on these training and test sample data sets, such as data cleaning, normalization, denoising, etc., to ensure the quality and consistency of the data. After the preprocessing is completed, an optimized training set and validation set can be obtained. Based on these training set and validation set obtained after preprocessing, a preset loss function can be used to guide the training process of the model, that is, using a fully connected neural network as the basic model and training it with the training set. During the training process, the model will continuously adjust its internal parameters and update the model to minimize the value of the loss function.

[0087] To verify the effectiveness and accuracy of the model, we will use the validation set to evaluate the initial neural network model obtained during the training process. If the performance of the model meets the preset criteria, such as the value of the loss function being lower than a certain threshold or the accuracy being higher than a certain level, then it can be considered that the model has achieved the expected performance, and the iterative training process ends. However, if the performance of the model fails to meet the preset criteria, the parameters of the model, such as the learning rate, number of layers, number of neurons, etc., need to be adjusted, and then the iterative training continues until the performance of the model meets the preset criteria, thus obtaining a spectral prediction network model with excellent performance.

[0088] Among them, in some embodiments, the preset loss function is:

[0089]

[0090] Among them, is the complete power spectral density result, is the validation set, is the preset frame length, is the number of rows of the power spectral density result.

[0091] In the embodiments of the present invention, in order to ensure that the power spectral results generated in step S102 and step S103 are correct and stable, the above loss function can be used to update the model during the model training process. The validation set in the preprocessed training and test sample data set is compared with the result obtained each time the training set is input into the model for training to calculate its mean square error loss result , and the data is used to update the model parameters. When tends to 0, the model update becomes more and more accurate. With the learning of the data set, the ability of the model to generate accurate power spectra is further improved. Among them, both the input and output of the fully connected neural network are of length One-dimensional vector.

[0092] The following details how to obtain the training set and the validation set.

[0093] As a possible implementation, in some embodiments, preprocess the training and test sample data group, and obtain the training set and the validation set based on the preprocessed training and test sample data group, including: dividing the training and test sample data group into low-speed ADC sampled wideband data and high-speed ADC sampled wideband data at corresponding times; based on a preset frame length and a preset step size, respectively perform segmentation processing on the low-speed ADC sampled wideband data and the high-speed ADC sampled wideband data at corresponding times, use the segmented low-speed ADC sampled wideband data as training data, and use the segmented high-speed ADC sampled wideband data as validation data; respectively perform fast Fourier transform processing with a length of a first preset length on each row of data in the training data and the validation data to obtain a wideband spectrum; take the modulus value of the wideband spectrum to obtain a power spectral density estimation result; divide the power spectral density estimation result at intervals of a second preset length to obtain the training set and the validation set.

[0094] Specifically, first divide the training and test sample data group into two main categories, namely, low-speed ADC sampled wideband data and high-speed ADC sampled wideband data at corresponding times; next, based on a preset frame length and step size parameters, perform segmentation operations on these two types of data respectively, where needs to be an integer multiple of . Set the division starting point at the position of the first sampling point of the low-speed ADC sampled wideband data, and intercept a total of data with a frame length of from the starting point as the first row of training data in the data set, then move the division starting point backward by the step size sampling points, and intercept data with a frame length of from the th sampling point as the second row of training data in the data set. The subsequent th row of training data in the data set is intercepted from the th sampling point backward with a frame length of until all the low-speed ADC sampled wideband data is divided, thus obtaining the training data ; the segmentation operation of the high-speed ADC sampled wideband data is the same as that of the low-speed ADC sampled wideband data, thereby obtaining the validation data .

[0095] To extract the signal spectrum features and make the data in the dataset conform to the designed model size, after the segmentation process is completed, the training data in each obtained data group and the verification data can be processed separately. For the data in a single data group, it has several rows with a frame length of data. For the training data and the verification data in each data group, fast Fourier transform processing can be performed on each row of data in each data group to obtain the Fourier transform result in complex form, that is, the broadband spectrum; then, a modulus operation is performed on the broadband spectrum to convert the complex spectrum into a real power spectrum, thereby obtaining the power spectral density estimation result; finally, the power spectral density estimation result is divided at intervals of the second preset length to obtain a matrix with the number of columns , that is, the training set and the verification set , and they are saved for subsequent model training and verification processes.

[0096] Through the above operations, not only the accuracy and consistency of data processing are ensured, but also a high-quality dataset is provided for model training and verification.

[0097] Furthermore, in some embodiments, after obtaining the training set and the verification set based on the preprocessed training and test sample data groups, it further includes: performing normalization processing on the training set to obtain the normalized training set; activating each training sample in the normalized training set to obtain the first feature vector of each training sample; respectively activating the channel dimension of the first feature vector of each training sample to obtain the second feature vector of each training sample; integrating the first feature vector of each training sample and the second feature vector of the corresponding training sample to obtain the channel feature result of each training sample.

[0098] Next, a fully connected spectrum feature extraction module is constructed, and the fully connected spectrum feature extraction module is used to process the training set to extract features from the signals with severe attenuation in the spectrum and further generate the spectrum features of the spectrum information.

[0099] Specifically, the fully connected spectrum feature extraction module divides the broadband spectrum into channels, and only processes one of the channels at a time. Taking one of the channels as an example, first, the training data in the first training data group is normalized to obtain the corresponding normalized training data . Among them, is the number of rows of the power spectral density result input to the neural network, is the number of channels of the signal, represents mathematical multiplication, a function represents taking the maximum value of the absolute values of a vector. Then, the input training data is activated by module E1 in the fully connected spectrum feature extraction module, which includes a fully connected layer, a normalization layer (Batch Normalization, abbreviated as BN), and an activation layer to extract the spectrum features in the channel, obtaining the first feature vector , and then the channel dimension of the first feature vector is activated by module E2, which includes a fully connected layer, a normalization layer, and an activation layer to obtain the second feature vector , and finally, the signals in the channel dimension are integrated and generated by the feature enhancement generation module E3, which includes a fully connected layer, to obtain the enhanced channel feature result , represents the current channel enhanced feature extracted, which is used as a reference for subsequent recovery and generation of channel feature content.

[0100] Among them,

[0101]

[0102]

[0103]

[0104] Among them, the detailed structure of module E1 is: a fully connected layer with a dimension of , a normalization layer, a RELU (Rectified Linear Unit) activation function layer, and a Dropout layer (a technique for reducing overfitting in neural networks); the detailed structure of module E2 is: a fully connected layer, a normalization layer, a RELU activation function layer, and a Dropout layer with the output dimension of module E1; the detailed structure of module E3 is: a fully connected layer with an input dimension of the output dimension of module E2 and an output dimension of N.

[0105] Furthermore, in some embodiments, after obtaining the channel feature results of each training sample, it further includes: activating the channel feature results of each training sample respectively to obtain the third feature vector of each training sample; activating the channel dimension of the third feature vector of each training sample respectively to obtain the fourth feature vector of each training sample; integrating and predicting the third feature vector of each training sample and the corresponding fourth feature vector of each sample to obtain the predicted channel power spectral density of each training sample; and splicing the predicted channel power spectral density of each training sample according to a preset splicing strategy to obtain the complete power spectral density result.

[0106] Next, construct a fully connected spectrum feature prediction and extension module, and use the fully connected spectrum feature prediction and extension module to process the channel enhancement feature results to analyze the signal features with severe attenuation in the spectrum, and generate the spectrum features of the spectrum information in a predictive manner.

[0107] Specifically, taking one of the channels as an example, first use the module D1 including the fully connected layer, normalization layer, and activation layer in the fully connected spectrum feature prediction and extension module to activate the enhanced feature data of the input channel, extract the enhanced channel spectrum features, and obtain the third feature vector , and then use the module D2 including the fully connected layer, normalization layer, and activation layer to activate the channel dimension of the third feature vector to obtain the fourth feature vector , and finally, integrate and predict the signals in the channel dimension through the spectrum generation module D3 including the fully connected layer to obtain the predicted channel power spectral density , which

[0108] represents the predicted current channel power spectral density result and is used for subsequent modules to perform spectrum recognition and detection.

[0109]

[0110]

[0111]

[0112] Among them, the detailed structure of module D1 is: a fully connected layer with dimension N, a normalization layer, a RELU activation function layer, and a Dropout layer; the detailed structure of module D2 is: a fully connected layer with the output dimension of module D1, a normalization layer, a RELU activation function layer, and a Dropout layer; the detailed structure of module D3 is a fully connected layer with the input dimension of the output dimension of D2.

[0113] Each time the fully connected spectrum feature prediction and extension module processes, it will generate the power spectral density of one of the channels in the broadband spectrum. Therefore, it is necessary to process segments of data, and based on the preset splicing strategy, change the dimension order of the predicted channel power spectral density each time and splice it into a complete power spectral density result , where is the length of the continuous signal before division, that is, the frame length .

[0114] As Figure 2 shown, Figure 2(a) is a schematic diagram of the broadband spectrum in the input model; Figure 2 (b) is a schematic diagram of the fully connected spectrum feature extraction module and the fully connected spectrum feature prediction and extension module; Figure 2 (c) is a schematic diagram of the complete power spectral density result output by the model.

[0115] For the convenience of those skilled in the art to further understand a satellite-ground channel detection bandwidth extension method of an autoencoder proposed in an embodiment of the present invention, the following will be further elaborated in conjunction with Figures 3 - 5 to make a further elaboration.

[0116] This satellite-ground channel detection bandwidth extension method of an autoencoder mainly adopts a spectrum prediction network model based on end-to-end training, learns the signal spectrum attenuated outside the ADC sampling bandwidth during offline training, and uses a fully connected neural network to achieve residual spectrum detection and spectrum prediction and completion after the spectrum features are extracted by the autoencoder neural network. Among them, as Figure 3 shown, the training process of the spectrum prediction network model may include the following steps:

[0117] Step S301, construct a training set of ADC sampling broadband signal data.

[0118] Step S302, construct a fully connected spectrum feature extraction module.

[0119] Step S303, construct a fully connected spectrum feature prediction and extension module.

[0120] Step S304, train the spectrum prediction network model.

[0121] Step S305, train and verify the model effect multiple times.

[0122] Further, as Figure 4 shown, Figure 4 (a) is a schematic diagram of the spectrum detection result after inputting the model (the two ends of the frequency band are suppressed), Figure 4 (b) is a schematic diagram of the ideal recovery result based on Figure 4 (a), Figure 4 (c) is a schematic diagram of the recovery result output by the embodiment of the present invention; Figure 5 The same as Figure 4 , for the sake of avoiding redundancy, it will not be elaborated here.

[0123] Thus, it can be seen that a satellite-ground channel detection bandwidth extension method of an autoencoder proposed in an embodiment of the present invention has the following advantages:

[0124] (1)A method for extending the bandwidth of satellite-ground channel detection of an autoencoder proposed by an embodiment of the present invention can accurately locate the position and shape of the spectrum signal before attenuation by using a generative neural network to recover the severely attenuated spectrum information, realize the prediction and generation of the residual spectrum signal for completion, improve the spectrum prediction accuracy, and achieve wideband spectrum sensing under the condition of low-speed ADC sampling.

[0125] (2)A method for extending the bandwidth of satellite-ground channel detection of an autoencoder proposed by an embodiment of the present invention decomposes the spectrum information into channels for decomposition processing, reduces the required network processing depth, and significantly reduces the network implementation complexity.

[0126] (3)A method for extending the bandwidth of satellite-ground channel detection of an autoencoder proposed by an embodiment of the present invention enriches the types of the data set and greatly reduces the amount of original data required for neural network training by using a preprocessing method of segmenting and intercepting the wideband data sampled by the ADC.

[0127] According to a method for extending the bandwidth of satellite-ground channel detection of an autoencoder proposed by an embodiment of the present invention, by acquiring a time-domain data signal and preprocessing the time-domain data signal, power spectral density data can be obtained; then, by using a preset spectrum prediction network model, the power spectral density features of the power spectral density data can be extracted; based on the power spectral density features, using the preset spectrum prediction network model, the missing part in the residual spectrum signal is predicted and completed, so as to restore the original spectrum shape. Thus, by learning the spectrum of wideband data in offline training and using a fully connected neural network to implement residual spectrum detection and spectrum prediction completion after extracting spectrum features, the problem that the spectrum prediction result is inaccurate due to factors such as the ADC receiving sampling rate and wideband resolution in the prior art is solved, and the spectrum prediction accuracy is improved.

[0128] Next, a satellite-ground channel detection bandwidth extension device of an autoencoder proposed by an embodiment of the present invention is described with reference to the accompanying drawings.

[0129] Figure 6 is a block diagram of a satellite-ground channel detection bandwidth extension device of an autoencoder according to an embodiment of the present invention.

[0130] As Figure 6 shown, the satellite-ground channel detection bandwidth extension device 10 of an autoencoder includes: an acquisition module 100, an extraction module 200, and a prediction module 300.

[0131] Among them, the acquisition module 100 is used to acquire a time-domain data signal and preprocess the time-domain data signal to obtain power spectral density data;

[0132] An extraction module 200 is configured to extract the power spectral density features of power spectral density data by using a preset spectral prediction network model, where the preset spectral prediction network model is obtained by training a fully-connected neural network with broadband data;

[0133] A prediction module 300 is configured to predict and complete the missing part in the residual spectral signal based on the power spectral density features by using the preset spectral prediction network model, and restore the original spectral shape.

[0134] Further, in some embodiments, before extracting the power spectral density features of the power spectral density data by using the preset spectral prediction network model, the extraction module 200 includes:

[0135] A construction unit is configured to obtain broadband data and construct a training and test sample data set by using the broadband data, where the broadband data is sampled by an analog-to-digital converter (ADC);

[0136] A preprocessing unit is configured to preprocess the training and test sample data set and obtain a training set and a validation set based on the preprocessed training and test sample data set;

[0137] A training unit is configured to train the fully-connected neural network based on a preset loss function by using the training set to obtain an initial neural network model, and verify the initial neural network model by using the validation set until the initial neural network model meets the preset criteria, end the iterative training of the fully-connected neural network to obtain the preset spectral prediction network model, otherwise adjust the model parameters and continue the iterative training.

[0138] Further, in some embodiments, the preprocessing unit is specifically configured to:

[0139] Divide the training and test sample data set into low-speed ADC-sampled broadband data and high-speed ADC-sampled broadband data at corresponding times;

[0140] Based on a preset frame length and a preset step size, perform segmentation processing on the low-speed ADC-sampled broadband data and the high-speed ADC-sampled broadband data at corresponding times respectively, use the segmented low-speed ADC-sampled broadband data as training data, and use the segmented high-speed ADC-sampled broadband data as validation data;

[0141] Perform fast Fourier transform processing with a first preset length on each row of data in each data group of the training data and the validation data respectively to obtain broadband spectra;

[0142] Take the modulus value of the broadband spectra to obtain a power spectral density estimation result;

[0143] Divide the power spectral density estimation result at intervals of a second preset length to obtain a training set and a validation set.

[0144] Further, in some embodiments, after obtaining the training set and the validation set based on the preprocessed training and test sample data sets, the preprocessing unit further includes:

[0145] A processing subunit, configured to perform normalization processing on the training set to obtain a normalized training set;

[0146] A first activation subunit, configured to activate each training sample in the normalized training set to obtain a first feature vector of each training sample;

[0147] A second activation subunit, configured to respectively activate the channel dimension of the first feature vector of each training sample to obtain a second feature vector of each training sample;

[0148] An integration subunit, configured to integrate the first feature vector of each training sample and the second feature vector of the corresponding training sample to obtain a channel feature result of each training sample.

[0149] Further, in some embodiments, after obtaining the channel feature result of each training sample, the integration subunit is further configured to:

[0150] Respectively activate the channel feature result of each training sample to obtain a third feature vector of each training sample;

[0151] Respectively activate the channel dimension of the third feature vector of each training sample to obtain a fourth feature vector of each training sample;

[0152] Integrate and predict the third feature vector of each training sample and the fourth feature vector of the corresponding sample to obtain the predicted channel power spectral density of each training sample;

[0153] Based on a preset splicing strategy, splice the predicted channel power spectral densities of each training sample to obtain a complete power spectral density result.

[0154] Further, in some embodiments, the preset loss function is:

[0155]

[0156] Wherein, is the complete power spectral density result, is the validation set, is the preset frame length, is the number of rows of the power spectral density result.

[0157] It should be noted that the foregoing explanatory description of the embodiment of the method for extending the bandwidth of satellite-ground channel detection of an autoencoder also applies to the device for extending the bandwidth of satellite-ground channel detection of an autoencoder in this embodiment, and will not be elaborated here.

[0158] A device for extending the bandwidth of satellite-ground channel detection of an autoencoder according to an embodiment of the present invention can obtain a time-domain data signal, and through preprocessing the time-domain data signal, power spectral density data can be obtained; then, by using a preset spectrum prediction network model, the power spectral density features of the power spectral density data can be extracted; based on the power spectral density features, by using the preset spectrum prediction network model, the missing part in the residual spectrum signal is predicted and complemented, so as to restore the original spectrum shape. Thus, by learning the spectrum of broadband data in offline training, and after extracting the spectrum features, using a fully connected neural network to implement residual spectrum detection and spectrum prediction and complementation, the problem that the spectrum prediction result is inaccurate due to factors such as the ADC receiving sampling rate and broadband resolution in the prior art is solved, and the spectrum prediction accuracy is improved.

[0159] Figure 7 The structure diagram of the electronic device provided by the embodiment of the present invention. The electronic device may include:

[0160] A memory 701, a processor 702, and a computer program stored on the memory 701 and executable on the processor 702.

[0161] When the processor 702 executes the program, it implements the method for extending the bandwidth of satellite-ground channel detection of an autoencoder provided in the foregoing embodiment.

[0162] Furthermore, the electronic device further includes:

[0163] A communication interface 703, used for communication between the memory 701 and the processor 702.

[0164] The memory 701 is used for storing a computer program executable on the processor 702.

[0165] The memory 701 may include a high-speed RAM (Random Access Memory) memory, and may also include a non-volatile memory, such as at least one disk memory.

[0166] If the memory 701, the processor 702, and the communication interface 703 are implemented independently, the communication interface 703, the memory 701, and the processor 702 can be interconnected through a bus and communicate with each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of simplicity in representation, Figure 7 only a thick line is used in Figure 7 , but it does not mean that there is only one bus or one type of bus.

[0167] Optionally, in specific implementation, if the memory 701, the processor 702, and the communication interface 703 are integrated on a single chip, the memory 701, the processor 702, and the communication interface 703 can communicate with each other through an internal interface.

[0168] The processor 702 may be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention.

[0169] The embodiments of the present invention also provide a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements a method for extending the bandwidth of satellite-ground channel detection of an autoencoder as described above.

[0170] The embodiments of the present invention also propose a computer program product, which includes a computer program, and when the program is executed by a processor, it is used to implement a method for extending the bandwidth of satellite-ground channel detection of an autoencoder as described above.

[0171] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In the description of the present invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0172] In the description of this specification, the descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc., mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0173] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for extending the detection bandwidth of the satellite-ground channel of an autoencoder, characterized in that Including the following steps: Obtain a time-domain data signal, and perform preprocessing on the time-domain data signal to obtain power spectral density data; Use a preset spectral prediction network model to extract the power spectral density features of the power spectral density data, where the preset spectral prediction network model is obtained by training a fully-connected neural network with broadband data; Based on the power spectral density features, use the preset spectral prediction network model to predict and complete the missing part in the residual spectral signal, and restore the original spectral shape; Wherein, before using the preset spectral prediction network model to extract the power spectral density features of the power spectral density data, it further includes: obtaining the broadband data, and using the broadband data to construct a training and test sample data group, where the broadband data is sampled by an analog-to-digital converter (ADC), performing preprocessing on the training and test sample data group, and obtaining a training set and a validation set based on the preprocessed training and test sample data group, based on a preset loss function, using the training set to train the fully-connected neural network to obtain an initial neural network model, and using the validation set to verify the initial neural network model until the initial neural network model meets the preset criteria, ending the iterative training of the fully-connected neural network to obtain the preset spectral prediction network model, otherwise adjusting the model parameters and continuing the iterative training; Wherein, performing preprocessing on the training and test sample data group and obtaining a training set and a validation set based on the preprocessed training and test sample data group includes: dividing the training and test sample data group into low-speed ADC-sampled broadband data and high-speed ADC-sampled broadband data at corresponding times, based on a preset frame length and a preset step size, respectively performing segmentation processing on the low-speed ADC-sampled broadband data and the high-speed ADC-sampled broadband data at corresponding times, using the segmented low-speed ADC-sampled broadband data as training data, and using the segmented high-speed ADC-sampled broadband data as validation data, respectively performing fast Fourier transform processing with a first preset length on each row of data in each data group of the training data and the validation data to obtain a broadband spectrum, taking the modulus value of the broadband spectrum to obtain a power spectral density estimation result, and dividing the power spectral density estimation result at intervals of a second preset length to obtain the training set and the validation set.

2. The method for extending the bandwidth of satellite-ground channel detection of an autoencoder according to claim 1, wherein, After obtaining the training set and the validation set based on the preprocessed training and test sample data group, it further includes: Performing normalization processing on the training set to obtain a normalized training set; Activating each training sample in the normalized training set to obtain a first feature vector of each training sample; Respectively activating the channel dimension of the first feature vector of each training sample to obtain a second feature vector of each training sample; Integrating the first feature vector of each training sample and the second feature vector of the corresponding training sample to obtain a channel feature result of each training sample.

3. A method for extending the detection bandwidth of the satellite-ground channel of an autoencoder according to claim 2, characterized in that, After obtaining the channel feature result of each training sample, it further includes: Activate the channel feature results of each of the training samples to obtain the third feature vector of each of the training samples; Activate the channel dimension of the third feature vector of each of the training samples to obtain the fourth feature vector of each of the training samples; Integrate and predict the third feature vector of each of the training samples and the fourth feature vector of the corresponding sample to obtain the predicted channel power spectral density of each of the training samples; Based on a preset splicing strategy, splice the predicted channel power spectral density of each of the training samples to obtain a complete power spectral density result.

4. A method for extending the detection bandwidth of the satellite-ground channel of an autoencoder according to claim 1, characterized in that, The preset loss function is: Among them, is the complete power spectral density result, is the validation set, is the preset frame length, is the number of rows of the power spectral density result.

5. A satellite-ground channel detection bandwidth extension device for an autoencoder, characterized in that, Including: An acquisition module, configured to acquire a time-domain data signal, and preprocess the time-domain data signal to obtain power spectral density data; An extraction module, configured to use a preset spectrum prediction network model to extract the power spectral density features of the power spectral density data, where the preset spectrum prediction network model is obtained by training a fully-connected neural network with broadband data; A prediction module, configured to, based on the power spectral density features, use the preset spectrum prediction network model to predict and complement the missing part in the residual spectrum signal to restore the original spectrum shape; Wherein, before using the preset spectrum prediction network model to extract the power spectral density features of the power spectral density data, the extraction module includes: a construction unit, configured to acquire the broadband data, and use the broadband data to construct a training and test sample data group, where the broadband data is sampled by an analog-to-digital converter (ADC), a preprocessing unit, configured to preprocess the training and test sample data group, and obtain a training set and a validation set based on the preprocessed training and test sample data group, a training unit, configured to, based on a preset loss function, use the training set to train the fully-connected neural network to obtain an initial neural network model, and use the validation set to verify the initial neural network model until the initial neural network model meets a preset standard, and end the iterative training of the fully-connected neural network to obtain the preset spectrum prediction network model, otherwise adjust the model parameters and continue the iterative training; Wherein, the preprocessing unit is specifically configured to: divide the training and test sample data group into low-speed ADC-sampled broadband data and high-speed ADC-sampled broadband data at corresponding moments, perform segmentation processing on the low-speed ADC-sampled broadband data and the high-speed ADC-sampled broadband data at corresponding moments respectively based on a preset frame length and a preset step length, use the segmented low-speed ADC-sampled broadband data as training data, and use the segmented high-speed ADC-sampled broadband data as validation data, perform fast Fourier transform processing with a length of a first preset length on each row of data in each data group of the training data and the validation data respectively to obtain a broadband spectrum, take the modulus value of the broadband spectrum to obtain a power spectral density estimation result, and divide the power spectral density estimation result at intervals of a second preset length to obtain the training set and the validation set.

6. An electronic device, characterized in that, Including: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement a method for extending the bandwidth of satellite-ground channel detection of an autoencoder as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement a method for extending the bandwidth of satellite-ground channel detection of an autoencoder as described in any one of claims 1-4.

8. A computer program product, characterized in that, It includes a computer program which, when executed by the processor, is used to implement a method for extending the bandwidth of satellite-ground channel detection of an autoencoder as described in any one of claims 1-4.

Images