A method, device and storage medium for determining a type of interference source

Abstract

This application provides a method, apparatus, device, and storage medium for determining interference source types. The method includes acquiring a fusion feature matrix of a cell under test; inputting the fusion feature matrix of the cell under test into an interference source type identification model for identification processing; and determining the interference source type of the cell under test based on the output of the interference source type identification model. This method can comprehensively capture the complex characteristics of interference signals by acquiring a multi-feature fusion matrix, reducing the risk of misjudgment based on a single feature. The interference source type identification model replaces traditional manual feature engineering, reduces the influence of network engineers' subjective experience on the identification results, and provides objective and quantitative output results, avoiding ambiguity in manual interpretation, thereby improving the accuracy of interference source type determination.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device and storage medium for determining the type of interference source. Background Technology

[0002] With the increasing variety and number of network devices, network interference problems are becoming more and more prominent. There are various types of interference, such as interference from video surveillance equipment, fake base stations, and jammers. There is an urgent need to quickly identify the type of interference to solve the problem of cell interference. The methods for determining the type of interference in related technologies usually rely on manual analysis, requiring network engineers to manually collect and process a large amount of interference data from cells. This method depends on the subjective experience and technical level of network engineers to make judgments, so the accuracy of identifying the type of interference source is low. Summary of the Invention

[0003] This application provides a method, apparatus, device, and storage medium for determining the type of interference source, in order to solve the problem of low accuracy in identifying the type of interference source in existing methods for determining the type of interference source.

[0004] To solve the above-mentioned technical problems, this application is implemented as follows:

[0005] In a first aspect, embodiments of this application provide a method for determining the type of interference source, the method comprising:

[0006] Obtain the fusion feature matrix of the cell to be tested;

[0007] The fusion feature matrix of the cell under test is input into the interference source type identification model for identification processing;

[0008] The type of interference source for the cell under test is determined based on the output of the interference source type identification model.

[0009] Optionally, obtaining the fusion feature matrix of the cell to be tested includes:

[0010] Acquire at least one input data, the input data including a cell label and the time-domain information, frequency-domain information and spatial information corresponding to the cell label;

[0011] Construct a two-dimensional time-frequency matrix corresponding to each input data based on each input data;

[0012] Each of the two-dimensional time-frequency matrices is decomposed into a one-dimensional vector by singular value decomposition;

[0013] The fusion feature matrix of the cell to be tested is constructed based on the at least one spatial information and at least one one-dimensional vector.

[0014] Optionally, the at least one input data includes input data of the cell to be tested and input data of N neighboring cells, where N is a positive integer;

[0015] The N neighboring cells include the N cells that are closest to the cell to be tested, as determined by spatial clustering analysis.

[0016] Optionally, the interference source type identification model includes an interference source type identification model based on an algorithm model and trained on a training dataset;

[0017] The training dataset includes fusion feature matrices and interference source type data under different interference sources.

[0018] Optionally, the algorithm model is a densely connected convolutional network;

[0019] The densely connected convolutional network includes N+1 consecutive feature delivery modules and N bottleneck layers, each of which includes a convolutional layer and a transition layer.

[0020] Optionally, when N is greater than 1, a bottleneck layer is set between every two feature transfer modules, and the bottleneck layer is used to reduce computational complexity.

[0021] Secondly, embodiments of this application also provide an interference source type determination device. This interference source type determination device includes:

[0022] The first acquisition module acquires the fusion feature matrix of the cell to be tested; the first acquisition module is used to acquire the first parameter value, the second parameter value, the third parameter value, and the fourth parameter value;

[0023] The input module is used to input the fusion feature matrix of the cell under test into the interference source type identification model for identification processing;

[0024] The output module is used to determine the interference source type of the cell under test based on the output of the interference source type identification model.

[0025] Optionally, the first acquisition module includes:

[0026] The second acquisition module is used to acquire at least one input data, the input data including a cell label and the time domain information, frequency domain information and spatial information corresponding to the cell label;

[0027] The first construction module is used to construct a two-dimensional time-frequency matrix corresponding to each input data according to each input data;

[0028] A decomposition module is used to decompose each of the two-dimensional time-frequency matrices into a one-dimensional vector through singular value decomposition.

[0029] The second construction module is used to construct a fusion feature matrix of the cell to be tested based on the at least one spatial information and at least one one-dimensional vector.

[0030] Optionally, the at least one input data includes input data of the cell under test and input data of N neighboring cells, where N is a positive integer;

[0031] The N neighboring cells include the N cells that are closest to the cell to be tested, as determined by spatial clustering analysis.

[0032] Optionally, the interference source type identification model includes an interference source type identification model based on an algorithm model and trained on a training dataset;

[0033] The training dataset includes fusion feature matrices and interference source type data under different interference sources.

[0034] Optionally, the algorithm model is a densely connected convolutional network;

[0035] The densely connected convolutional network includes N+1 consecutive feature delivery modules and N bottleneck layers, each of which includes a convolutional layer and a transition layer.

[0036] Optionally, when N is greater than 1, a bottleneck layer is set between every two feature transfer modules, and the bottleneck layer is used to reduce computational complexity.

[0037] Thirdly, embodiments of this application provide an electronic device, including a transceiver and a processor, wherein the processor is used for:

[0038] Obtain the fusion feature matrix of the cell to be tested;

[0039] The fusion feature matrix of the cell under test is input into the interference source type identification model for identification processing;

[0040] The type of interference source for the cell under test is determined based on the output of the interference source type identification model.

[0041] Optionally, the processor is further configured to:

[0042] Acquire at least one input data, the input data including a cell label and the time-domain information, frequency-domain information and spatial information corresponding to the cell label;

[0043] Construct a two-dimensional time-frequency matrix corresponding to each input data based on each input data;

[0044] Each of the two-dimensional time-frequency matrices is decomposed into a one-dimensional vector by singular value decomposition;

[0045] The fusion feature matrix of the cell to be tested is constructed based on the at least one spatial information and at least one one-dimensional vector.

[0046] Optionally, the at least one input data includes input data of the cell to be tested and input data of N neighboring cells, where N is a positive integer;

[0047] The N neighboring cells include the N cells that are closest to the cell to be tested, as determined by spatial clustering analysis.

[0048] Optionally, the interference source type identification model includes an interference source type identification model based on an algorithm model and trained on a training dataset;

[0049] The training dataset includes fusion feature matrices and interference source type data under different interference sources.

[0050] Optionally, the algorithm model is a densely connected convolutional network;

[0051] The densely connected convolutional network includes N+1 consecutive feature delivery modules and N bottleneck layers, each of which includes a convolutional layer and a transition layer.

[0052] Optionally, when N is greater than 1, a bottleneck layer is set between every two feature transfer modules, and the bottleneck layer is used to reduce computational complexity.

[0053] Fourthly, embodiments of this application provide an electronic device, including: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, it implements the steps of the interference source type determination method as described in the first aspect above.

[0054] Fifthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the interference source type determination method described in the first aspect above.

[0055] In a sixth aspect, embodiments of this application provide a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the interference source type determination method as described in the first aspect above.

[0056] The interference source type determination method of this application includes acquiring a fused feature matrix of a cell under test; inputting the fused feature matrix of the cell under test into an interference source type identification model for identification processing; and determining the interference source type of the cell under test based on the output of the interference source type identification model. This method, by acquiring a multi-feature fusion matrix, can comprehensively capture the complex characteristics of interference signals and reduce the risk of misjudgment based on a single feature; the interference source type identification model replaces traditional manual feature engineering, reducing the influence of network engineers' subjective experience on the identification results, and the output results are objectively quantified, avoiding ambiguity from manual interpretation, thereby improving the accuracy of interference source type determination. Attached Figure Description

[0057] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart of the interference source type determination method provided in the embodiments of this application;

[0059] Figure 2 This is an image showing the interference identification output results provided in an embodiment of this application;

[0060] Figure 3 This is a two-dimensional time-frequency matrix diagram provided in the embodiments of this application;

[0061] Figure 4 This is a singular value decomposition graph provided in an embodiment of this application;

[0062] Figure 5 This is a time-frequency spatial domain feature fusion map provided in the embodiments of this application;

[0063] Figure 6 This is a diagram of the DenseNet network structure provided in the embodiments of this application;

[0064] Figure 7 This is a diagram of the internal structure of the Dense Block provided in the embodiments of this application;

[0065] Figure 8 This is the second flowchart of the interference source type determination method provided in the embodiments of this application;

[0066] Figure 9 This is the third flowchart of the interference source type determination method provided in the embodiments of this application;

[0067] Figure 10 This is a structural diagram of an interference source type determination device provided in an embodiment of this application;

[0068] Figure 11 This is a structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0069] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0070] This application provides a method for determining the type of interference source. See also... Figure 1 , Figure 1 This is a flowchart of the interference source type determination method provided in the embodiments of this application, such as... Figure 1 As shown, it includes the following steps:

[0071] Step 101: Obtain the fusion feature matrix of the cell to be tested;

[0072] In this step, the aforementioned fused feature matrix is ​​a matrix that integrates multiple different types of features. These features can originate from various aspects of the cell, such as signal strength, signal quality, frequency distribution, and interference intensity. For example, information such as signal strength values ​​in different frequency bands and interference intensity values ​​at different time periods can be integrated to form a matrix. This method allows for a unified representation of various cell feature information, facilitating subsequent processing and analysis.

[0073] Step 102: Input the fused feature matrix of the cell to be tested into the interference source type identification model for identification processing;

[0074] In this step, the fused feature matrix of the cell under test is input into the interference source type identification model. This interference source type identification model can be trained on a large amount of data, and it can learn the feature patterns corresponding to different interference source types. For example, for different types of interference sources such as co-channel interference, adjacent-channel interference, and external signal interference, the model can identify the characteristics of their respective feature matrices. When the fused feature matrix of the cell under test is input, the model will analyze and judge based on the learned patterns. In the embodiments of this application, the specific form of the interference source type identification model is not limited. The interference source type identification model can be a decision tree. For example, for interference source type identification, each internal node can represent a feature in the fused feature matrix, such as signal strength or frequency. Branches are formed based on the value of this feature, and finally, a leaf node is reached to determine the interference source type. In some optional embodiments, the interference source type identification model can also be a binary or multi-class classification model. In interference source type identification, the fused feature matrix is ​​used as input, and different types of interference sources are distinguished by solving for the optimal hyperplane. The interference source type identification model can also be a neural network model, a random forest model, etc.

[0075] Step 103: Determine the interference source type of the cell under test based on the output of the interference source type identification model.

[0076] In this step, the interference source type identification model, after processing the input fused feature matrix, outputs a corresponding result. This result directly points to the type of interference source that may exist in the cell under test. For example, if the model output represents "co-channel interference," the interference source type of the cell under test can be determined to be co-channel interference based on the output. In this way, the type of interference source faced by the cell under test can be quickly and accurately determined using the model's output, providing a basis for subsequent interference problem-solving. In some optional implementations, the output of the above interference source type identification model can be a fixed-length binary string representing the interference type, with each bit corresponding to a type of interference. For example, 1 indicates the presence of the interference, and 0 indicates its absence. For instance, assuming there are 6 predefined interference types (such as fake base station interference, video surveillance interference, Multi-channel Multipoint Distribution Service (MMDS) interference, co-channel interference, jammer interference, and other interference), the output format is a 6-bit binary code:

[0077] 101000: Indicates the presence of fake base station interference and MMDS interference.

[0078] 000000: Indicates no interference or no known interference type identified.

[0079] In other alternative implementations, the output of the interference source type identification model can also be a probability vector, where each element corresponds to the probability value of one interference type, and the sum is 1. For example, assuming the model is trained with 5 interference types (A, B, C, D, E), the output is a probability vector of length 5. Furthermore, the output can also be in the form of a label string or a matrix; different output formats do not affect the implementation of the basic functions of the embodiments of this application.

[0080] In some other alternative implementations, such as Figure 2 As shown, the output of the interference source type identification model can be encoded using 6-bit binary codes, with each bit corresponding to a type of interference. When the network model determines that the target cell is experiencing a certain type of interference, the corresponding bit will be set to 1. Considering that in actual network deployment and operation environments, multiple factors and devices may simultaneously interfere with 5G signals, this multi-bit storage design that simultaneously detects and records multiple interference types is particularly important.

[0081] In the interference source type determination method of this application embodiment, the complex characteristics of the interference signal can be fully captured by obtaining the multi-feature fusion matrix, reducing the risk of misjudgment of a single feature; the interference source type identification model replaces the traditional manual feature engineering, reduces the influence of network engineers' subjective experience on the identification results, and outputs objective and quantitative results, avoiding ambiguity in manual interpretation, thereby improving the accuracy of interference source type determination.

[0082] Optionally, obtaining the fusion feature matrix of the cell to be tested includes:

[0083] Acquire at least one input data, the input data including a cell label and the time-domain information, frequency-domain information and spatial information corresponding to the cell label;

[0084] Construct a two-dimensional time-frequency matrix corresponding to each input data based on each input data;

[0085] Each of the two-dimensional time-frequency matrices is decomposed into a one-dimensional vector by singular value decomposition;

[0086] The fusion feature matrix of the cell to be tested is constructed based on the at least one spatial information and at least one one-dimensional vector.

[0087] In the interference source type determination method of this application embodiment, the aforementioned cell label can be understood as a unique identifier for the cell, used to associate spatiotemporal information. For example, it can be a Cell Global Identifier (CGI) or other forms of label. The aforementioned time-domain information can be understood as the signal characteristics of the cell in the time dimension; the aforementioned frequency-domain information can be understood as the signal characteristics of the cell in the frequency dimension; the aforementioned spatial information can be understood as the geographical location information of the cell. The aforementioned information together form a quadruple of information, that is, the aforementioned input data.

[0088] The aforementioned two-dimensional time-frequency matrix can be understood as combining the time-domain information and frequency-domain information in the input data into a matrix form. For example, for each input data, the Physical Resource Block (PRB) vector can be arranged in time order to form a matrix with time on the horizontal axis and frequency on the vertical axis.

[0089] The singular value decomposition described above can be understood as decomposing the matrix to be decomposed into the product of three matrices and extracting the singular values ​​as a one-dimensional feature vector. The construction of the fusion feature matrix of the cell under test based on at least one spatial information and at least one one-dimensional vector can be understood as concatenating the spatial information with the one-dimensional vector to form a matrix containing spatiotemporal features.

[0090] In some optional implementations, data cleaning or normalization can be performed on time-frequency-space related fields. The appropriate method can be selected according to the actual business scenario. In this regard, the embodiments of this application do not make specific limitations.

[0091] For example, to obtain all batches of PRB vectors for a target cell within a specified time range (e.g., one day), for a specific batch of data collection for a specific 5G cell:

[0092]

[0093] -124≤PRB i ≤-40

[0094] Furthermore, a two-dimensional time-frequency matrix is ​​constructed in chronological order (the horizontal axis represents time, and the vertical axis represents frequency), as follows: Figure 3 Furthermore, through singular value decomposition, the two-dimensional time-frequency matrix is ​​decomposed into a one-dimensional vector:

[0095] The singular value decomposition algorithm is shown below. Its main function is to reduce the dimensionality and compress data.

[0096] Assume A is a two-dimensional matrix to be processed:

[0097] A=U∑V T

[0098] Where U is an m×m orthogonal matrix (A×A^T), whose column vectors are called left singular vectors; Σ is an m×n diagonal matrix (but usually only has non-zero values ​​on the diagonal and a portion of the top left corner, and the rest are 0), the elements on the diagonal are called singular values, which are arranged in descending order; and V is an n×n orthogonal matrix (A^T×A), whose column vectors are called right singular vectors.

[0099] A one-dimensional feature vector of the 5G cell can be obtained from the singular value matrix Σ through Singular Value Decomposition (SVD), such as... Figure 4 .

[0100] In this implementation, by fusing time-domain, frequency-domain, and spatial-domain features to cover the multi-dimensional characteristics of interference signals, a multi-dimensional feature fusion matrix can be constructed as input to improve representation capabilities. By using singular value decomposition to achieve noise reduction and dimensionality reduction, the use of original PRB data, which has a long model training time and is susceptible to noise, is avoided to some extent, thereby accelerating model training and inference.

[0101] Optionally, the at least one input data includes input data of the cell to be tested and input data of N neighboring cells, where N is a positive integer;

[0102] The N neighboring cells include the N cells that are closest to the cell to be tested, as determined by spatial clustering analysis.

[0103] In the interference source type determination method of this application embodiment, the aforementioned cell under test can be understood as the core object of the interference source type identification work, that is, the specific cell whose interference source type needs to be determined. The aforementioned neighboring cells can be understood as cells that are spatially close to the cell under test. The signal conditions of these cells may be affected by the same interference source, or there may be a correlation between their signal interference. The aforementioned spatial clustering analysis can be understood as a data analysis method that can use the spatial information of the cells, such as latitude and longitude coordinates, to group spatially close cells into one category. In this scenario, it is used to find the cells closest to the cell under test. The aforementioned use of spatial clustering analysis to determine the N cells closest to the cell under test can, for example, collect the spatial information of each cell, and then use a spatial clustering algorithm, such as a density-based spatial algorithm, to cluster these cells; in the clustering results, the N cells closest to the cell under test are selected as neighboring cells.

[0104] For example, to determine the type of interference source in cell A, cell A is the cell under test. By collecting spatial information of surrounding cells and using spatial clustering analysis, it is found that cells B, C, and D are spatially closest to cell A. Therefore, cells B, C, and D are the neighboring cells, with N equal to 3. The input data obtained then includes the input data for cells A, B, C, and D, which covers cell labels and corresponding time-domain, frequency-domain, and spatial information.

[0105] In some alternative implementations, the spatial clustering analysis described above can be an unsupervised learning method designed to reveal hidden patterns and structures in data by classifying data points into different categories. For example, after data preprocessing, Geographic Information System (GIS) fields in the cell data can be extracted as spatial domain features, and appropriate clustering algorithms and distance metrics (epsilon) between cells can be selected to ultimately obtain cell clusters based on geographic coordinates. In 5G cell interference identification, density-based spatial (DBSCAN) clustering is highly recommended, as it does not require pre-specifying the number of clusters and can identify and process noise points.

[0106] For example, a DBSCAN clustering algorithm function can be constructed as follows:

[0107] The function ST_ClusterDBSCAN originates from the PostgreSQL extension PostGIS, and its definition is:

[0108] integer ST_ClusterDBSCAN(geometry winset geom, float8 eps, integerminpoints);

[0109] In practical use, the first parameter `geom` needs to be specified as the set of geometric objects of the input cells, i.e., the geographic coordinates of the cells calculated from their latitude and longitude; the second parameter `eps` is set as the maximum distance threshold for two cells to be considered to be in the same cluster; and the third parameter `minpoints` specifies the minimum number of points required to form a cluster. The output of this function is a set of integers, where each element represents the cluster ID (CID) to which the corresponding input cell belongs. Based on these CIDs, core points can be further identified and all their nearest boundary points can be obtained, thus yielding statistical information on the spatial distribution of the cells.

[0110] In some optional implementations, N can be 4. For each cell to be identified, its four nearest neighbors are obtained and their feature vectors are extracted, centered on the cell and sorted according to spatial distance. Subsequently, these feature vectors are merged (concatenated and fused) in groups of five cells to construct a fused feature matrix containing multi-source temporal, frequency, and spatial information, such as... Figure 5 As shown, Cell0 is the cell to be identified (the central cell), and Cells 1 to 4 are the four nearest neighbor cells in the same cluster. The smaller the subscript number of a Cell, the closer the spatial distance between that cell and the central cell.

[0111] In this implementation, since interference signals may propagate in space and affect surrounding cells, by considering the input data of the cell under test and neighboring cells, the characteristics and propagation patterns of interference signals can be captured more comprehensively. By analyzing the data of multiple cells, the type of interference source can be determined more accurately.

[0112] Optionally, the interference source type identification model includes an interference source type identification model based on an algorithm model and trained on a training dataset;

[0113] The training dataset includes fusion feature matrices and interference source type data under different interference sources.

[0114] In the interference source type determination method of this application embodiment, the above-mentioned algorithm model is the basic framework for constructing the above-mentioned interference source identification model. This application embodiment does not specifically limit the specific form of the algorithm model. The above-mentioned algorithm model may include decision trees, support vector machines, and neural networks, such as multilayer perceptrons, convolutional neural networks, and recurrent neural networks. Different algorithm models have different characteristics and applicable scenarios, and can be selected according to specific needs. The row number training dataset is a data set used to train the interference source type identification model. It contains fusion feature matrices under different interference sources and corresponding interference source type data. By allowing the model to learn the relationship between the features in the training dataset and the corresponding interference source types, the model can gradually master the ability to identify interference source types. The above-mentioned interference source type data clearly identifies the interference source type corresponding to each fusion feature matrix, such as fake base station interference, co-channel interference, adjacent channel interference, external signal interference, etc. The training dataset is input into the established model for training. During the training process, the model will continuously adjust its parameters so that the interference source type predicted by the model for the fusion feature matrix in the training dataset is as close as possible to the actual interference source type data. This process typically employs optimization algorithms, such as gradient descent, to minimize the error between the model's predictions and the actual results. Training continues for multiple rounds until the model's performance reaches a satisfactory level.

[0115] For example, suppose we need to build an interference source type identification model to identify three types of interference sources: fake base station interference, co-channel interference, and adjacent-channel interference. First, we need to prepare a training dataset, collecting relevant data from different cells under the conditions of fake base station interference, co-channel interference, and adjacent-channel interference. For each cell, we construct its fusion feature matrix. For example, the fusion feature matrix includes the signal strength changes of the cell at different times of the day (time domain information), the signal energy distribution of different frequency bands (frequency domain information), and the cell's latitude and longitude coordinates (spatial information). Simultaneously, we label each fusion feature matrix with the corresponding interference source type, such as "fake base station interference," "co-channel interference," or "adjacent-channel interference." After selecting a suitable algorithm model, such as a Multilayer Perceptron (MLP), we input the prepared training dataset into the MLP for training. During training, the model continuously adjusts the weight parameters of the hidden layer and output layer based on the error between the predicted results and the actual labeled interference source types. After multiple rounds of training, the model gradually learns to identify the corresponding interference source types from the fusion feature matrix.

[0116] In some optional implementations, a training set, validation set, and test set for the network model can be constructed, with a recommended split ratio of 6:2:2. To improve the stability and generalization ability of the model, recognized methods such as k-fold cross-validation or bootstrap sampling can be appropriately adopted. This application does not specifically limit the specific implementation of these methods.

[0117] In this implementation, by training the model with a large number of fused feature matrices and corresponding interference source type data under different interference sources, the model can learn the feature patterns of various interference sources. This enables it to more accurately identify the interference source type when facing a new cell to be tested. In addition, the training dataset covers a variety of different interference source situations, allowing the model to adapt to different real-world scenarios and improving its generalization ability. Furthermore, the interference source type identification model is scalable. If new interference source types appear in the future, it is only necessary to collect the corresponding fused feature matrices and interference source type data and further train the model to expand its recognition capabilities.

[0118] Optionally, the algorithm model is a densely connected convolutional network;

[0119] The densely connected convolutional network includes N+1 consecutive feature delivery modules and N bottleneck layers, each of which includes a convolutional layer and a transition layer.

[0120] In the interference source type determination method of this application embodiment, the aforementioned densely connected convolutional network (DenseNet) can be further improved by building an improved two-dimensional convolutional neural network (CNN) based on DenseNet. Unlike traditional CNNs or residual networks (ResNet), the core idea of ​​DenseNet is to establish dense connections (channel connections) between each layer and all preceding layers to maximize feature reuse. Feature reuse and bypass settings from the perspective of features make information flow and feature transfer in the network more efficient, reducing the number of parameters in the network model and mitigating the gradient vanishing problem to some extent. Figure 6 As shown, the DenseNet network structure mainly consists of dense blocks and transition layers.

[0121] The aforementioned Dense Block is a core component of DenseNet, such as... Figure 7 As shown, it consists of multiple convolutional layers, with the input of each layer being the concatenation of the outputs of all preceding convolutional layers along the channel dimension. This dense connectivity allows for full utilization of feature information, and the network's expressive power gradually increases with the number of layers. Figure 7 In this diagram, X0 represents the input feature layer, BN+ReLU+Conv2D refers to regularization, activation functions, and convolutional layers, and Concat represents the concatenation layer, which reuses all the previous layers. The input features of X0 are convolved to obtain X1, which is then stacked with the original X0 (the feature maps must be of equal size) to obtain a new feature layer X1. This process is repeated to obtain X2, and so on, until all feature layers have participated in the stacking, finally outputting Y0.

[0122] Transition Layer: The transition layer is mainly used to connect two adjacent Dense Blocks, integrate the features obtained from the previous Dense Block, and reduce the size of the feature map through pooling operations. Typically, an average pooling 2D with a stride of 2 is used to reduce the width and height of the feature layer.

[0123] In terms of usage details, the DenseNet network model provides three adjustable parameters:

[0124] Growth Rate (k): In a Dense Block, each convolutional layer outputs k feature maps. These feature maps are concatenated with the feature maps from all previous layers along the channel dimension to serve as the input for the next layer. As the number of layers increases, the number of input feature maps to the Dense Block gradually increases. Depending on the actual business scenario, the Growth Rate can be set to a relatively small value, such as around 12.

[0125] Compression Rate: In the Transition Layer, the degree to which the number of feature maps is reduced can be controlled by adjusting the compression rate. A compression rate less than 1 indicates that the number of feature maps is reduced in the Transition Layer; a compression rate equal to 1 indicates that the number of feature maps remains unchanged.

[0126] In this implementation, a balance between efficient extraction of interference features and computational complexity is achieved through densely connected convolutional networks. This structure not only improves the model's efficiency in utilizing multi-source features but also enhances the ability to identify regional interference through a neighboring cell feature transfer module, ultimately reducing model deployment costs while ensuring recognition accuracy.

[0127] Optionally, when N is greater than 1, a bottleneck layer is set between every two feature transfer modules, and the bottleneck layer is used to reduce computational complexity.

[0128] In the interference source type determination method of this application embodiment, the bottleneck layer can be understood as reducing computational load and improving network efficiency. By adding a 1x1 convolutional layer before the convolutional layer of the Dense Block, the number of input feature maps is reduced, thereby reducing computational complexity. In practical use, it is recommended to add a bottleneck layer between every two Dense Blocks, which can reduce the number of parameters by about 30%. The above N can be understood as the number of neighboring cells, and here it is limited to a positive integer greater than 1. The data of neighboring cells are used together with the data of the cell to be tested to construct a fused feature matrix, which serves as the input to the interference source type identification model. The bottleneck layer is a special network layer structure, usually composed of 1x1 convolutions. Its main function is to reduce the number of channels in the feature map without losing too much information, thereby reducing the complexity of subsequent calculations. When the number of neighboring cells N is greater than 1, a bottleneck layer is set between every two feature transfer modules. This is because when there are multiple neighboring cells, the feature information input into the model will be richer and more complex, and the number of channels in the feature map may be relatively large. Without proper handling, subsequent computations will increase significantly, leading to reduced efficiency in model training and inference, and potentially even overfitting. The main function of the bottleneck layer is to reduce computational complexity. Specifically, it compresses the number of channels in the input feature map through 1x1 convolution operations. For example, assuming the feature map output by the previous feature transfer module has 128 channels, after the 1x1 convolution in the bottleneck layer, the number of channels may be compressed to 64. This reduces the number of features that need to be processed in subsequent feature transfer modules, thus reducing computational load. Furthermore, since 1x1 convolution can integrate and abstract features to some extent, it reduces computational complexity without losing too much important feature information.

[0129] For example, if we want to identify the type of interference source in a cell under test, and select 3 neighboring cells (N=3), the constructed densely connected convolutional network has 4 feature transfer modules. Then, 3 bottleneck layers will be set between these 4 feature transfer modules, because one is set between every two feature transfer modules.

[0130] After the first feature transfer module processes the input data, the output feature map has 128 channels.

[0131] Next, we enter the first bottleneck layer, where the number of channels is compressed to 64 using 1x1 convolution.

[0132] The compressed feature map is then input into the second feature transfer module for further feature extraction and processing.

[0133] After the second feature transfer module finishes processing, it passes through the second bottleneck layer to compress the number of channels again before being input into the third feature transfer module, and so on.

[0134] In some alternative implementations, the improved DenseNet network model obtained after the above training can be as follows: This model can consist of five consecutive feature transfer modules, each module can embed a Dense Block and a Transition Layer, and a Bottleneck Layer can be set between every two modules to optimize information flow. The input layer stride is set to 2, the initial convolutional kernel size can be 5x5 to ensure sufficient receptive field, the connection stride within the Dense Block can be 1, the internal convolutional kernel size is uniformly 1x1, the growth rate is 12 to control the growth rate of the feature maps, and the compression rate is 1 to maintain information integrity.

[0135] In this implementation, reducing the number of channels in the feature map lowers the complexity of subsequent calculations, enabling the model to complete computational tasks faster during training and inference. For interference source type identification tasks that require processing large amounts of data, this can save time and computational resources to some extent. Furthermore, reducing the number of features can prevent the model from becoming overly complex, thereby reducing the risk of overfitting. Overfitting causes a model to perform well on training data but poorly on new data, while setting a bottleneck layer helps improve the model's generalization ability.

[0136] In some alternative implementations, see [link to implementation details]. Figure 8 and Figure 9 The process of this application embodiment can be divided into three parts: feature fusion, model training, and interference identification. Feature fusion is achieved by extracting singular vectors of the time-frequency space matrix through SVD decomposition and fusing them into a feature matrix. Model training is performed by training the DenseNet network with cell data containing labels (CGI) and geographic information. Interference identification is based on the trained DenseNet network model to determine interference in suspected interference cells and identify the type of interference.

[0137] See Figure 10 , Figure 10 This is a structural diagram of an interference source type determination device provided in another embodiment of this application.

[0138] like Figure 10 As shown, the interference source type determination device 1000 includes:

[0139] The first acquisition module 1001 acquires the fusion feature matrix of the cell to be tested; the first acquisition module is used to acquire the first parameter value, the second parameter value, the third parameter value and the fourth parameter value;

[0140] Input module 1002 is used to input the fusion feature matrix of the cell under test into the interference source type identification model for identification processing;

[0141] The output module 1003 is used to determine the interference source type of the cell under test based on the output of the interference source type identification model.

[0142] Optionally, the first acquisition module includes:

[0143] The second acquisition module is used to acquire at least one input data, the input data including a cell label and the time domain information, frequency domain information and spatial information corresponding to the cell label;

[0144] The first construction module is used to construct a two-dimensional time-frequency matrix corresponding to each input data according to each input data;

[0145] A decomposition module is used to decompose each of the two-dimensional time-frequency matrices into a one-dimensional vector through singular value decomposition.

[0146] The second construction module is used to construct a fusion feature matrix of the cell to be tested based on the at least one spatial information and at least one one-dimensional vector.

[0147] Optionally, the at least one input data includes input data of the cell under test and input data of N neighboring cells, where N is a positive integer;

[0148] The N neighboring cells include the N cells that are closest to the cell to be tested, as determined by spatial clustering analysis.

[0149] Optionally, the interference source type identification model includes an interference source type identification model based on an algorithm model and trained on a training dataset;

[0150] The training dataset includes fusion feature matrices and interference source type data under different interference sources.

[0151] Optionally, the algorithm model is a densely connected convolutional network;

[0152] The densely connected convolutional network includes N+1 consecutive feature delivery modules and N bottleneck layers, each of which includes a convolutional layer and a transition layer.

[0153] Optionally, when N is greater than 1, a bottleneck layer is set between every two feature transfer modules, and the bottleneck layer is used to reduce computational complexity.

[0154] For details, see Figure 11As shown in the figure, this application embodiment also provides an electronic device, including a bus 1101, a transceiver 1102, an antenna 1103, a bus interface 1104, a processor 1105, and a memory 1106.

[0155] Processor 1105, used for:

[0156] Obtain the fusion feature matrix of the cell to be tested;

[0157] The fusion feature matrix of the cell under test is input into the interference source type identification model for identification processing;

[0158] The type of interference source for the cell under test is determined based on the output of the interference source type identification model.

[0159] exist Figure 11 In this document, a bus architecture (represented by bus 1101) is used. Bus 1101 may include any number of interconnected buses and bridges, linking various circuits including one or more processors represented by processor 1105 and memory represented by memory 1106. Bus 1101 may also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 1104 provides an interface between bus 1101 and transceiver 1102. Transceiver 1102 may be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 1105 is transmitted over a wireless medium via antenna 1103, which further receives data and transmits it to processor 1105.

[0160] Processor 1105 is responsible for managing bus 1101 and general processing, and can also provide various functions, including timing, peripheral interface, voltage regulation, power management, and other control functions. Memory 1106 can be used to store data used by processor 1105 during operation.

[0161] Optionally, the processor 1105 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a complex programmable logic device (CPLD).

[0162] Optionally, the monitoring information includes distance information and speed information corresponding to the at least one target object, and the processor 1105 is specifically used for:

[0163] Acquire at least one input data, the input data including a cell label and the time-domain information, frequency-domain information and spatial information corresponding to the cell label;

[0164] Construct a two-dimensional time-frequency matrix corresponding to each input data based on each input data;

[0165] Each of the two-dimensional time-frequency matrices is decomposed into a one-dimensional vector by singular value decomposition;

[0166] The fusion feature matrix of the cell to be tested is constructed based on the at least one spatial information and at least one one-dimensional vector.

[0167] Optionally, the at least one input data includes input data of the cell to be tested and input data of N neighboring cells, where N is a positive integer;

[0168] The N neighboring cells include the N cells that are closest to the cell to be tested, as determined by spatial clustering analysis.

[0169] Optionally, the interference source type identification model includes an interference source type identification model based on an algorithm model and trained on a training dataset;

[0170] The training dataset includes fusion feature matrices and interference source type data under different interference sources.

[0171] Optionally, the algorithm model is a densely connected convolutional network;

[0172] The densely connected convolutional network includes N+1 consecutive feature delivery modules and N bottleneck layers, each of which includes a convolutional layer and a transition layer.

[0173] Optionally, when N is greater than 1, a bottleneck layer is set between every two feature transfer modules, and the bottleneck layer is used to reduce computational complexity.

[0174] It should be noted that the electronic device provided in this application embodiment is a device capable of executing the above-described interference source type determination method. Therefore, all implementation methods in the above-described interference source type determination method embodiments are applicable to this electronic device and can achieve the same or similar beneficial effects. To avoid repetition, this embodiment will not elaborate further.

[0175] This invention also provides an electronic device, including: a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it implements the various processes of the above-described interference source type determination method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0176] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, this computer program implements the various processes of the above-described interference source type determination method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0177] This application also provides a computer program product, including computer instructions. When executed by a processor, the computer instructions implement the various processes of the above-described interference source type determination method embodiment and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0178] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0179] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0180] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for determining the type of interference source, characterized in that, The method includes: Obtain the fusion feature matrix of the cell to be tested; The fusion feature matrix of the cell under test is input into the interference source type identification model for identification processing; The type of interference source for the cell under test is determined based on the output of the interference source type identification model.

2. The method according to claim 1, characterized in that, The process of obtaining the fusion feature matrix of the cell to be tested includes: Acquire at least one input data, the input data including a cell label and the time-domain information, frequency-domain information and spatial information corresponding to the cell label; Construct a two-dimensional time-frequency matrix corresponding to each input data based on each input data; Each of the two-dimensional time-frequency matrices is decomposed into a one-dimensional vector by singular value decomposition; The fusion feature matrix of the cell to be tested is constructed based on the at least one spatial information and at least one one-dimensional vector.

3. The method according to claim 2, characterized in that, The at least one input data includes input data of the cell to be tested and input data of N neighboring cells, where N is a positive integer; The N neighboring cells include the N cells that are closest to the cell to be tested, as determined by spatial clustering analysis.

4. The method according to claim 1, characterized in that, The interference source type identification model includes an interference source type identification model based on an algorithm model and trained on a training dataset. The training dataset includes fusion feature matrices and interference source type data under different interference sources.

5. The method according to claim 4, characterized in that, The algorithm model is a densely connected convolutional network; The densely connected convolutional network includes N+1 consecutive feature delivery modules and N bottleneck layers, each of which includes a convolutional layer and a transition layer.

6. The method according to claim 5, characterized in that, When N is greater than 1, a bottleneck layer is set between every two feature transfer modules, and the bottleneck layer is used to reduce computational complexity.

7. A device for determining the type of interference source, characterized in that, The device includes: The first acquisition module acquires the fusion feature matrix of the cell to be tested; The input module is used to input the fusion feature matrix of the cell under test into the interference source type identification model for identification processing; The output module is used to determine the interference source type of the cell under test based on the output of the interference source type identification model.

8. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the interference source type determination method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the interference source type determination method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps of the interference source type determination method as described in any one of claims 1 to 6.

Images