Single-burst multi-dimensional parameter high-precision estimation method based on deep learning

Abstract

The invention provides a single-burst multi-dimensional parameter high-precision estimation method based on deep learning, belongs to the technical field of radio signal processing, and solves the problem of low precision of existing single-burst time / frequency / space / energy multi-dimensional parameter estimation. The method comprises the following steps: constructing a data set of a corresponding dimension according to data characteristics of a signal sample in each dimension; aiming at the data set of each dimension, constructing a corresponding specific parameter estimation model; training a corresponding specific parameter estimation model by using the data set of each dimension; taking the trained specific parameter estimation model of each dimension as a branch, connecting the feature extraction module and the parameter estimation module of each branch by using an attention module, constructing a multi-branch depth estimation model, and finely adjusting the multi-branch depth estimation model by using a multi-dimensional data set; and inputting the data characteristics of a signal to be estimated in different dimensions into the multi-branch depth estimation model, and predicting and outputting a time / frequency / space / energy multi-dimensional parameter estimation result.

Description

Technical Field

[0001] The present invention relates to the technical field of radio signal processing, and in particular, to a method for high-precision estimation of single-burst multi-dimensional parameters based on deep learning. Background Art

[0002] In the field of modern communication, signal parameter estimation is a fundamental and crucial step in radio signal analysis and processing. This technology analyzes the received signal to accurately infer the key characteristic parameters of the signal, such as the direction of arrival, carrier frequency offset, and signal-to-noise ratio, etc., providing a basis for subsequent signal processing and demodulation.

[0003] Traditional parameter estimation methods are usually designed for specific parameters and signal types, and are easily affected by factors such as noise and multipath fading, unable to meet the requirements of high precision and high efficiency for parameter estimation in complex electromagnetic environments. With the development of artificial intelligence technology, deep learning technology has demonstrated superior performance in the field of signal processing. It can automatically extract the deep features of the signal, overcome the limitations of traditional parameter estimation methods in complex environments, and become an effective way to solve high-precision parameter estimation.

[0004] However, existing deep learning-based signal parameter estimation technologies usually rely on single-feature input or single-task learning, lacking the modeling of the correlation between features, and it is difficult to fully exploit the multi-dimensional characteristics of the signal in the time domain, frequency domain, spatial domain, and energy domain, resulting in limited accuracy of parameter estimation. Therefore, aiming at the problems of low accuracy of parameter estimation in complex electromagnetic environments and the need for traditional estimation methods to be designed for specific parameters and signal types, how to achieve high-precision unified estimation of single-burst time / frequency / spatial / energy multi-dimensional parameters is an urgent problem to be solved at present. Summary of the Invention

[0005] In view of the above analysis, the embodiments of the present invention aim to provide a method for high-precision estimation of single-burst multi-dimensional parameters based on deep learning to solve the problem of low accuracy of existing single-burst time / frequency / spatial / energy multi-dimensional parameter estimation.

[0006] The present invention discloses a method for high-precision estimation of single-burst multi-dimensional parameters based on deep learning, and the method includes: Construct a data set for each dimension according to the data characteristics of the signal samples in each dimension; For the data set of each dimension, construct a corresponding specific parameter estimation model; and train the corresponding specific parameter estimation model by using the data set of each dimension respectively; wherein, the specific parameter estimation model is composed of a feature extraction module and a parameter estimation module; Take the trained specific parameter estimation model for each dimension as a branch, use the attention module to connect the feature extraction module and the parameter estimation module of each branch, construct a multi-branch depth estimation model, and use a multi-dimensional dataset to fine-tune the multi-branch depth estimation model; Input the data features of the newly received signal to be estimated in different dimensions into the trained multi-branch depth estimation model, and the multi-branch depth estimation model predicts and outputs the multi-dimensional parameter estimation results of time / frequency / space / energy.

[0007] On the basis of the above solution, the present invention has also made the following improvements: Further, when training the corresponding specific parameter estimation model using the dataset of each dimension, perform: Use the data features in the dataset of the corresponding dimension as the input of the feature extraction model in the specific parameter estimation model, and use the corresponding label as the prediction output of the parameter estimation model in the specific parameter estimation model to train the specific parameter estimation model to obtain the trained specific parameter estimation model.

[0008] Further, the feature extraction module and the parameter estimation module are connected in sequence. The feature extraction module is used to extract the deep features of the data features, and the parameter estimation module is used to estimate the signal parameters from the extracted deep features.

[0009] Further, when constructing the multi-branch depth estimation model, perform: Take the specific parameter estimation model of each dimension as a branch respectively. Add the feature vectors output by the feature extraction modules of each branch and then input them into the attention module. The vectors output by the attention module are respectively input into the parameter estimation modules of each branch to construct a multi-branch depth estimation model.

[0010] Further, when using the multi-dimensional dataset to fine-tune the multi-branch depth estimation model, perform: Transfer the model parameters of the specific parameter estimation model of each dimension to the multi-branch depth estimation model; Use the multi-dimensional dataset to fine-tune the network parameters of the multi-branch depth estimation model.

[0011] Further, when the multi-branch depth estimation model predicts and outputs the multi-dimensional parameter estimation results of time / frequency / space / energy, perform: Input the feature data of each dimension into the corresponding dimension's feature extraction module in the trained multi-branch depth estimation model, and the corresponding dimension's parameter estimation module in the multi-branch depth estimation model predicts and outputs the parameter prediction results of the corresponding dimension; Combine the parameter prediction results of different dimensions to obtain the multi-dimensional parameter estimation results of time / frequency / space / energy.

[0012] Further, the prediction parameters for each dimension are respectively: the direction of arrival of the signal, the start time and duration of the signal, the carrier frequency offset of the signal, the signal-to-noise ratio, and the signal bandwidth; among them, The start time and duration of the signal are time-domain parameters, the carrier frequency offset and signal bandwidth of the signal are frequency-domain parameters, the direction of arrival of the signal is a spatial-domain parameter, and the signal-to-noise ratio is an energy-domain parameter.

[0013] Further, the datasets constructed for each dimension include: the signal IQ dataset, the signal amplitude dataset, the signal phase dataset, the signal covariance eigenvalue dataset, and the signal power spectrum dataset.

[0014] Further, to construct the datasets for each dimension, execute: Use the IQ matrix of each signal sample as the data feature and the direction of arrival of the signal as the corresponding label to construct the signal IQ dataset; Use the instantaneous amplitude of each signal sample as the data feature and the start time and duration of the signal as the label to construct the signal amplitude dataset; Use the instantaneous phase of each signal sample as the data feature and the carrier frequency offset of the signal as the label to construct the signal phase dataset; Use the eigenvalue matrix corresponding to the covariance matrix of each signal sample as the data feature and the signal-to-noise ratio as the label to construct the signal covariance eigenvalue dataset; Use the power spectrum matrix constructed from the power spectrum of each signal sample as the data feature and the signal bandwidth as the label to construct the signal power spectrum dataset.

[0015] Further, determine the signal samples in the following manner: Adopt a multi-antenna reception method or a simulation method to obtain radio signals of different modulation styles and frequencies, and extract the burst segment and then process it to obtain signal samples with a fixed length.

[0016] Compared with the prior art, the present invention can at least achieve one of the following beneficial effects: (1) The present invention adopts various forms of input, fully integrates the key features of the signal in the multi-dimensional space of time, frequency, space, and energy, realizes the unified high-precision estimation of single-burst multi-dimensional parameters, and significantly improves the estimation accuracy and system robustness.

[0017] (2) The present invention proposes a high-precision estimation method for single-burst multi-dimensional parameters based on deep learning. It applies a neural network to learn deep features from signal data, overcomes the problems that traditional estimation methods estimate for specific parameters and signal types and are easily affected by factors such as noise and multipath fading, and realizes the high-precision unified estimation of time / frequency / space / energy multi-dimensional parameters.

[0018] (3) The present invention adopts a specific parameter estimation model weight migration method, which effectively reduces the training time and difficulty of the multi-branch depth estimation model.

[0019] In the present invention, the above technical solutions can also be combined with each other to achieve more preferred combination schemes. Other features and advantages of the present invention will be described in the following specification, and some advantages can be made obvious from the specification, or understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained from the content specifically pointed out in the specification and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings are only for the purpose of showing specific embodiments, and are not considered as a limitation of the present invention. Throughout the drawings, the same reference signs denote the same components; Figure 1 It is a flowchart of a high-precision single-burst multi-dimensional parameter estimation method based on deep learning provided by an embodiment of the present invention; Figure 2 It is a structural diagram of a multi-branch depth estimation model provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] The following will specifically describe the preferred embodiments of the present invention with reference to the drawings. The drawings form a part of this application and are used together with the embodiments of the present invention to explain the principles of the present invention, but are not used to limit the scope of the present invention.

[0022] A specific embodiment of the present invention discloses a high-precision single-burst multi-dimensional parameter estimation method based on deep learning. The flowchart of this method is as Figure 1 shown, and this method includes the following steps.

[0023] Step S1: Obtain radio signals with different modulation styles and frequencies, extract the burst segments, and post-process to obtain signal samples with a fixed length.

[0024] Preferably, in this embodiment, a multi-antenna reception method is used to collect radio signals with different modulation styles and frequencies multiple times, or a simulation method matching the multi-antenna reception method is used to obtain radio signals with different modulation styles and frequencies. The multi-antenna reception method is used to obtain the direction of arrival of the radio signals, and collecting different modulation styles and frequencies is to change the modulation style and frequency during signal transmission, aiming to improve the diversity of the signal samples.

[0025] Generally, the obtained radio signals contain multiple burst segments. Each burst segment in the originally continuous radio signal containing multiple burst segments is separately extracted, and after zero-padding for each single burst, signal samples with a fixed length are obtained. Therefore, each signal sample contains a complete burst.

[0026] Step S2: Construct a data set for each dimension according to the data characteristics of the signal samples in each dimension.

[0027] Preferably, in this embodiment, data sets in 5 dimensions are set, including: signal IQ data set, signal amplitude data set, signal phase data set, signal covariance eigenvalue data set, and signal power spectrum data set.

[0028] Obtain the signal arrival direction, signal start time and duration, signal carrier frequency offset, signal-to-noise ratio, and bandwidth of each signal sample respectively, and construct each data set.

[0029] (1) Construction of the signal IQ data set Extract the IQ components of each signal sample respectively to form an IQ matrix; use the IQ matrix of each signal sample as the data feature and the signal arrival direction as the corresponding label to construct the signal IQ data set. .

[0030] Exemplarily, the IQ matrix of the signal sample is expressed as: (1) where , respectively represent the I component and Q component of the th sampling point in the th antenna in the signal sample; , represents the number of antennas; , represents the number of sampling points of the signal sample.

[0031] Therefore, the signal IQ data set can be expressed as: (2) where , respectively represent the IQ matrix and signal arrival direction of the th signal sample, is the number of signal samples.

[0032] (2) Construction of the signal amplitude data set Use the instantaneous amplitude of each signal sample as the data feature and the signal start time and duration as the label to construct the signal amplitude data set. .

[0033] The instantaneous amplitude of the signal sample is expressed as: (3) Among them, represents the th sampling point in the signal sample.

[0034] The signal amplitude data set is expressed as: (4) Among them, represents the instantaneous amplitude of the th signal sample, , respectively represent the start time and duration of the signal of the th signal sample.

[0035] (3) Construction of the signal phase data set Using the instantaneous phase of each signal sample as the data feature and the signal carrier frequency offset as the label, construct the signal phase data set .

[0036] The instantaneous phase of the signal sample is expressed as: (5) The signal phase data set is expressed as: (6) Among them, , respectively represent the instantaneous phase and the signal carrier frequency offset of the th signal sample.

[0037] (4) Construction of the signal covariance eigenvalue data set Using the eigenvalue matrix corresponding to the covariance matrix of each signal sample (i.e., the covariance matrix eigenvalue) as the data feature and the signal-to-noise ratio as the label, construct the signal covariance eigenvalue data set .

[0038] Shift and intercept each point of the signal sample in sequence according to the length and stack them to obtain the matrix , which is expressed as: (7) Calculate the sample covariance matrix , which is expressed as: (8) Perform eigenvalue decomposition on to obtain the eigenvalue matrix .

[0039] The signal covariance eigenvalue data set is expressed as: (9) Among them, and respectively represent the eigenvalue matrix and the signal-to-noise ratio corresponding to the covariance matrix of the th signal sample.

[0040] (5) Construction of the signal power spectrum data set Construct a signal power spectrum data set by using the power spectrum matrix constructed from the power spectrum of each signal sample as data features and the signal bandwidth as labels. .

[0041] The signal at the th sampling point in the signal sample The power spectrum of is expressed as: (10) Among them, is the Fourier transform. Extract the real part and the imaginary part of to construct a power spectrum matrix, which is expressed as: (11) The signal power spectrum data set is expressed as: (12) Among them, and respectively represent the power spectrum matrix of the th signal sample and the signal bandwidth.

[0042] Step S3: For each dimension of the data set, construct a corresponding specific parameter estimation model; and respectively use each dimension of the data set to train the corresponding specific parameter estimation model; among them, the specific parameter estimation model is composed of a feature extraction module and a parameter estimation module.

[0043] In the specific parameter estimation model, the feature extraction module and the parameter estimation module are connected in sequence. The feature extraction module is used to extract the deep features of the data features, and the parameter estimation module is used to estimate specific signal parameters from the extracted deep features. In the specific implementation process, an appropriate specific parameter estimation model can be selected according to the characteristics of each dimension of the data set. Exemplarily, the network structures in the feature extraction module and the parameter estimation module of a certain dimension can be designed in the manner of Table 1.

[0044] Table 1 Network structures in the feature extraction module and the parameter estimation module

[0045] As shown in Table 1, the feature extraction module includes a convolutional layer Conv, a max pooling layer MaxPooling, and two residual blocks Resblock. Among them, 、 respectively represent the size of the convolutional kernel and the pooling window, which are adjusted according to the dimension of the input data, represents the stride, and represents the number of convolutional kernels. The parameter estimation module is composed of three fully connected layers (FC) with different numbers of neurons cascaded, representing the number of neurons in the fully connected layer. It should be noted that only a simple example of the estimation model is given here. In the actual application process, a batch normalization layer and an activation function should also be added after the convolutional layer of the feature extraction module.

[0046] Corresponding to the 5-dimensional dataset constructed in step S2, in this step, the specific parameter estimation models of the 5 dimensions constructed are respectively represented as: the specific parameter estimation model for the direction-of-arrival estimation dimension, the specific parameter estimation model for the start and duration estimation dimension, the specific parameter estimation model for the carrier frequency offset estimation dimension, the specific parameter estimation model for the signal-to-noise ratio estimation dimension, and the specific parameter estimation model for the bandwidth estimation dimension. Therefore, the dataset is used as the dataset for training the specific parameter estimation model for the direction-of-arrival estimation dimension, the dataset is used as the dataset for training the specific parameter estimation model for the start and duration estimation dimension, the dataset is used as the dataset for training the specific parameter estimation model for the carrier frequency offset estimation dimension, the dataset is used as the dataset for training the specific parameter estimation model for the signal-to-noise ratio estimation dimension, and the dataset is used as the dataset for training the specific parameter estimation model for the bandwidth estimation dimension.

[0047] Specifically, in the process of separately training the parameter estimation model for each dimension, the data features in the corresponding dimension dataset are used as the input of the feature extraction module in the specific parameter estimation model, and the corresponding label is used as the prediction output of the parameter estimation module in the specific parameter estimation model to train the specific parameter estimation model to obtain the trained parameter estimation model.

[0048] Exemplarily, the mean square error can be used as the loss function. Assume that the result predicted by the parameter estimation model for the th signal sample is , and the corresponding true label is . At this time, the loss function is expressed as: (13) When the training loss reaches the minimum, the parameter estimation model training is completed, and the model parameters of the specific parameter estimation model that has completed training are saved.

[0049] Step S4: Take the specific parameter estimation model of each dimension that has completed training as a branch, use the attention module to connect the feature extraction modules of each branch and the parameter estimation network, construct a multi-branch depth estimation model, and use a multi-dimensional data set to fine-tune the multi-branch depth estimation model.

[0050] In this embodiment, after the specific parameter estimation models of each dimension are trained, the specific parameter estimation model of each dimension is taken as a branch respectively (that is, one dimension corresponds to one branch). The feature vectors output by the feature extraction modules of each branch are added and then input into the attention module. The vectors output by the attention module are respectively input into the parameter estimation modules of each branch to achieve parameter estimation in multiple dimensions. The constructed multi-branch depth estimation model is as Figure 2 shown. In the constructed multi-branch depth estimation model, the model parameters of the specific parameter estimation models of each dimension are also correspondingly migrated to the multi-branch depth estimation model. That is, according to the corresponding relationship between different branches in the multi-branch depth estimation model and the specific parameter estimation model, the model parameters of the specific parameter estimation model saved after training are imported.

[0051] Preferably, in this embodiment, the dimensions of the feature vectors output by the feature extraction models of different dimensions are the same, and the feature vectors of different dimensions are summed through matrix addition.

[0052] More specifically, the feature extraction modules of all branches can be called a multi-branch feature extraction module. Refer to Figure 2 , the multi-branch feature extraction module includes five branches. The architectures of the feature extraction modules of branch one, branch two, branch three, branch four, and branch five are the same as those of the specific parameter estimation models in the direction-of-arrival estimation dimension, start and duration estimation dimension, carrier frequency offset estimation dimension, signal-to-noise ratio estimation dimension, and bandwidth estimation dimension respectively, and correspond to five different training data, namely the IQ matrix, instantaneous amplitude, instantaneous phase, eigenvalue matrix corresponding to the covariance matrix, and power spectrum matrix. In addition, the parameter estimation modules of all branches can also be called a multi-branch parameter estimation module. Refer to Figure 2 , the multi-branch parameter estimation module also includes five branches. The architectures of each branch are the same as those of the parameter estimation modules of the above five specific parameter estimation models (the structure and the parameters of each layer are the same), and output the direction-of-arrival, start and duration, carrier frequency offset, signal-to-noise ratio, and bandwidth estimation results respectively.

[0053] In this embodiment, the attention module adopts the QKV attention mechanism. Assume the feature vectors output by the multi-branch feature extraction module , represents the batch size, represents the sequence length, represents the dimension of the feature vector. Through linear transformation of three groups of trainable weight matrices , , we get: (14) where, , , , is the dimension of attention calculation. The calculated attention score is: (15) The vector output by the attention module is expressed as: (16) The role of the attention module is to dynamically allocate the importance weights of different features, enabling the model to focus on the most critical features in different branches. In subsequent fine-tuning training, the attention module will be trained. The vector output by the attention module is divided into five paths and input into the parameter estimation modules of five dimensions in the multi-branch depth estimation model respectively.

[0054] Preferably, in this embodiment, a multi-dimensional dataset is used to fine-tune the multi-branch depth estimation model, and the following steps are performed: Step S41: Transfer the model parameters of the specific parameter estimation model of each dimension to the multi-branch depth estimation model.

[0055] Step S42: Fine-tune the network parameters of the multi-branch depth estimation model using the multi-dimensional dataset.

[0056] Specifically, five constructed datasets are used to fine-tune and train the multi-branch depth estimation model, and the weighted mean square error is used as the loss function. Specifically, the constructed datasets , , , and are used to fine-tune and train the multi-branch depth estimation model. The mean square error of different branches is , , then the loss function of the multi-branch depth estimation model is expressed as: (17) where, , , , , represent the weights of five dimensions.

[0057] Step S5: Input the data features of the newly received signal to be estimated in different dimensions into the trained multi-branch depth estimation model, and the multi-branch depth estimation model predicts the multi-dimensional parameter estimation results of time / frequency / space / energy when outputting.

[0058] Specifically, the parameter estimation modules in different dimensions of the multi-branch depth estimation model can respectively predict and output the direction of arrival of the signal, the start time and duration of the signal, the carrier frequency offset of the signal, the signal-to-noise ratio, and the signal bandwidth. Among them, the start time and duration of the signal are time-domain parameters, the carrier frequency offset and signal bandwidth of the signal are frequency-domain parameters, the direction of arrival of the signal is a space-domain parameter, and the signal-to-noise ratio is an energy-domain parameter.

[0059] Specifically, Step S5 is specifically implemented in the following manner.

[0060] Step S51: For the newly received signal to be estimated, according to the construction method of the data sets in each dimension, respectively obtain the feature data in each dimension, including: IQ matrix, instantaneous amplitude, instantaneous phase, eigenvalue matrix corresponding to the covariance matrix, and power spectrum matrix.

[0061] Step S52: Input the feature data in each dimension into the corresponding dimension feature extraction module in the trained multi-branch depth estimation model, and the parameter estimation module in the corresponding dimension of the multi-branch depth estimation model predicts and outputs the parameter estimation results in the corresponding dimension; combine the parameter estimation results in different dimensions to obtain the multi-dimensional parameter estimation results of time / frequency / space / energy.

[0062] In summary, the high-precision estimation method for single-burst multi-dimensional parameters based on deep learning provided in this embodiment collects signal data of different modulation styles and frequencies by using multi-antenna reception, extracts and processes signal bursts, constructs data sets of different dimensions and corresponding specific parameter estimation models, and uses the corresponding data sets for separate training. A multi-dimensional parameter estimation model with a multi-branch structure is constructed based on the specific parameter estimation model, and the multi-branch depth estimation model is fine-tuned and trained using weight migration. By introducing the multi-branch depth estimation model, this embodiment can fuse the features extracted by the specific models of each dimension, alleviate the information isolation existing in the separate training of the specific parameter estimation models of each dimension, and improve the generalization ability. At the same time, it can avoid the problem that the specific parameter estimation models of each dimension need to be separately loaded and independently inferred during the inference stage, and simplify the deployment process. Subsequently, the signal data to be estimated is input into the trained multi-branch depth estimation model to obtain the multi-dimensional parameter estimation result. The solution of the embodiment of the present invention uses a neural network to learn the deep features of the signal and can achieve unified high-precision estimation of the time / frequency / space / energy multi-dimensional parameters of the signal.

[0063] Those skilled in the art can understand that all or part of the processes for implementing the methods of the above embodiments can be completed by instructing relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory, etc.

[0064] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A high-precision estimation method for single-burst multi-dimensional parameters based on deep learning, characterized in that The method includes: Construct datasets for corresponding dimensions according to the data characteristics of signal samples in each dimension; For each dimension's dataset, construct a corresponding specific parameter estimation model; and respectively train the corresponding specific parameter estimation model using each dimension's dataset; wherein, the specific parameter estimation model consists of a feature extraction module and a parameter estimation module; Take the trained specific parameter estimation model of each dimension as a branch, use an attention module to connect the feature extraction modules and parameter estimation modules of each branch, construct a multi-branch depth estimation model, and fine-tune the multi-branch depth estimation model using a multi-dimensional dataset; Input the data characteristics of the newly received signal to be estimated in different dimensions into the trained multi-branch depth estimation model, and the multi-branch depth estimation model predicts and outputs the multi-dimensional parameter estimation results of time / frequency / space / energy.

2. The high-precision estimation method for single-burst multi-dimensional parameters based on deep learning according to claim 1, characterized in that When respectively training the corresponding specific parameter estimation model using each dimension's dataset, perform: Use the data characteristics in the corresponding dimension's dataset as the input of the feature extraction model in the specific parameter estimation model, and use the corresponding label as the predicted output of the parameter estimation model in the specific parameter estimation model to train the specific parameter estimation model, and obtain the trained specific parameter estimation model.

3. The high-precision estimation method for single-burst multi-dimensional parameters based on deep learning according to claim 2, characterized in that The feature extraction module and the parameter estimation module are connected in sequence. The feature extraction module is used to extract the deep features of data characteristics, and the parameter estimation module is used to estimate signal parameters from the extracted deep features.

4. The high-precision estimation method for single-burst multi-dimensional parameters based on deep learning according to claim 2, wherein When constructing the multi-branch depth estimation model, perform: Respectively take the specific parameter estimation model of each dimension as a branch, add the feature vectors output by the feature extraction modules of each branch and then input them into the attention module, and the vectors output by the attention module are respectively input into the parameter estimation modules of each branch to construct a multi-branch depth estimation model.

5. The high-precision estimation method for single-burst multi-dimensional parameters based on deep learning according to claim 4, wherein When fine-tuning the multi-branch depth estimation model using a multi-dimensional dataset, perform: Migrate the model parameters of the specific parameter estimation model of each dimension to the multi-branch depth estimation model; Use the multi-dimensional dataset to fine-tune the network parameters of the multi-branch depth estimation model.

6. The high-precision estimation method for single-burst multi-dimensional parameters based on deep learning according to any one of claims 1-5, characterized in that When the multi-branch depth estimation model predicts and outputs the multi-dimensional parameter estimation results of time / frequency / space / energy, perform: Input the feature data of each dimension correspondingly into the feature extraction module of the corresponding dimension in the trained multi-branch depth estimation model, and the parameter estimation module of the corresponding dimension in the multi-branch depth estimation model predicts and outputs the parameter prediction results of the corresponding dimension; Combine the parameter prediction results of different dimensions to obtain the multi-dimensional parameter estimation results of time / frequency / space / energy.

7. The high-precision estimation method for single-burst multi-dimensional parameters based on deep learning according to claim 6, wherein The predicted parameters of each dimension are respectively: signal arrival direction, signal start time and duration, signal carrier frequency offset, signal-to-noise ratio, and signal bandwidth; wherein, The signal start time and duration are time-domain parameters, the signal carrier frequency offset and signal bandwidth are frequency-domain parameters, the signal arrival direction is a space-domain parameter, and the signal-to-noise ratio is an energy-domain parameter.

8. The high-precision estimation method for single-burst multi-dimensional parameters based on deep learning according to claim 7, characterized in that, The constructed datasets for each dimension include: signal IQ dataset, signal amplitude dataset, signal phase dataset, signal covariance eigenvalue dataset, and signal power spectrum dataset.

9. The high-precision estimation method for single-burst multi-dimensional parameters based on deep learning according to claim 8, characterized in that When constructing the datasets for each dimension, perform: Construct a signal IQ dataset by using the IQ matrix of each signal sample as the data feature and the direction of arrival of the signal as the corresponding label; Construct a signal amplitude dataset by using the instantaneous amplitude of each signal sample as the data feature and the start time and duration of the signal as the labels; Construct a signal phase dataset by using the instantaneous phase of each signal sample as the data feature and the carrier frequency offset of the signal as the label; Construct a signal covariance eigenvalue dataset by using the eigenvalue matrix corresponding to the covariance matrix of each signal sample as the data feature and the signal-to-noise ratio as the label; Construct a signal power spectrum dataset by using the power spectrum matrix constructed from the power spectrum of each signal sample as the data feature and the signal bandwidth as the label.

10. The high-precision estimation method for single-burst multi-dimensional parameters based on deep learning according to claim 1, characterized in that, Determine the signal samples in the following manner: Obtain radio signals of different modulation styles and frequencies by using a multi-antenna reception method or a simulation method, and extract the burst segment and then process it to obtain signal samples of a fixed length.

Images