Intelligent spectrum sensing positioning method and system based on neural network

Abstract

The embodiment of the application provides a kind of neural network-based intelligent spectrum sensing positioning method and system, first, the multi-source spectrum sensing signal set of the multiple frequency band radio frequency signal unit that different sensing nodes in target area are captured in the same period is gathered;Then the cross-node collaborative calibration processing is carried out to radio frequency signal unit, and collaborative sensing signal set is generated;Again, the hierarchical feature mining is executed to collaborative sensing signal set based on preset rule, and spectrum hole feature and channel attenuation feature are obtained;Then these features are input into the spectrum positioning neural network trained to carry out space-time correlation positioning analysis, and output initial positioning parameter set;Finally, the accurate spectrum positioning coordinate and spectrum sensing positioning instruction are generated by carrying out multi-node fusion optimization processing to initial positioning parameter set, and are transmitted to spectrum management terminal to complete positioning control.The application improves the precision and reliability of spectrum sensing positioning.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, and more specifically, to an intelligent spectrum sensing and positioning method and system based on neural networks. Background Technology

[0002] In the field of wireless communication, spectrum sensing positioning technology is crucial for effectively utilizing spectrum resources, ensuring communication quality, and achieving precise signal source management. Traditional spectrum sensing positioning methods have several limitations. On the one hand, some methods rely solely on signal acquisition from a single sensing node. Due to the limited sensing range and accuracy of a single node, it is difficult to comprehensively and accurately acquire spectrum information within the target area. Especially in complex electromagnetic environments, these methods are easily affected by local interference, leading to significant deviations in positioning results.

[0003] On the other hand, while some multi-node sensing methods collect signals from multiple nodes, they lack effective cross-node collaborative processing mechanisms. Signals from different nodes may have inconsistent baselines, leading to inaccurate feature extraction and analysis, which in turn affects positioning accuracy. Furthermore, existing spectrum sensing positioning technologies often employ relatively simple algorithms in feature extraction and positioning analysis, making it difficult to fully mine deep feature information in the spectrum signals. They cannot accurately identify key features such as spectrum holes and channel attenuation, and their ability to perform spatiotemporal correlation positioning analysis is insufficient, failing to meet the demand for high-precision spectrum sensing positioning in dynamically changing electromagnetic environments. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide an intelligent spectrum sensing and positioning method and system based on neural networks.

[0005] In conjunction with the first aspect of this application, a neural network-based intelligent spectrum sensing and positioning method is provided, applied to a neural network-based intelligent spectrum sensing and positioning system, the method comprising:

[0006] Collect a set of multi-source spectrum sensing signals within the target area, wherein the set of multi-source spectrum sensing signals includes multi-band radio frequency signal units captured by different sensing nodes at the same time period;

[0007] Cross-node collaborative calibration processing is performed on the radio frequency signal units in the multi-source spectrum sensing signal set to generate a collaborative sensing signal set with a unified spectrum reference.

[0008] Based on preset spectral feature extraction rules, hierarchical feature mining is performed on the collaborative sensing signal set to obtain the spectral hole features and channel attenuation features of each radio frequency signal unit;

[0009] The spectrum hole features and the channel attenuation features are input into the trained spectrum positioning neural network to perform spatiotemporal correlation positioning analysis and output the initial positioning parameter set of the radio frequency signal unit.

[0010] The initial positioning parameter set is subjected to multi-node fusion optimization processing to generate accurate spectrum positioning coordinates of the target area. Based on the accurate spectrum positioning coordinates, a spectrum sensing positioning command containing the signal source orientation identifier is generated. The spectrum sensing positioning command is transmitted to the spectrum management terminal to complete the positioning and control operation.

[0011] In conjunction with the second aspect of this application, a neural network-based intelligent spectrum sensing and positioning system is provided. The neural network-based intelligent spectrum sensing and positioning system includes a machine-readable storage medium and a processor. The machine-readable storage medium stores machine-executable instructions. When the processor executes the machine-executable instructions, the neural network-based intelligent spectrum sensing and positioning system implements the aforementioned neural network-based intelligent spectrum sensing and positioning method.

[0012] In conjunction with a third aspect of this application, a computer-readable storage medium is provided, wherein computer-executable instructions are stored therein, and when the computer-executable instructions are executed, the aforementioned neural network-based intelligent spectrum sensing and positioning method is implemented.

[0013] Combining any of the above aspects, by collecting a set of multi-source spectrum sensing signals within the target area and performing cross-node collaborative calibration processing, a collaborative sensing signal set with a unified spectrum reference is generated, effectively solving the problem of inconsistent signal references among different nodes. Based on preset spectrum feature extraction rules, hierarchical feature mining is performed on the collaborative sensing signal set, which can fully explore the spectrum hole features and channel attenuation features of each radio frequency signal unit. The extracted features are input into a trained spectrum positioning neural network for spatiotemporal correlation positioning analysis. Utilizing the powerful learning and analysis capabilities of the neural network, it can better handle complex spatiotemporal correlation relationships and output a more accurate set of initial positioning parameters. Finally, multi-node fusion optimization processing is performed on the initial positioning parameter set to further improve the positioning accuracy, generate precise spectrum positioning coordinates of the target area, and generate a spectrum sensing positioning command containing signal source orientation identifiers based on the precise spectrum positioning coordinates. This command is transmitted to the spectrum management terminal to complete the positioning and control operation, realizing a complete and efficient process from signal acquisition to precise positioning and control. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained in conjunction with these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating the neural network-based intelligent spectrum sensing and positioning method provided in this application embodiment. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0018] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0019] Figure 1 The diagram illustrates a flowchart of a neural network-based intelligent spectrum sensing and positioning method according to an embodiment of this application. It should be understood that in other embodiments, the order of some steps in the neural network-based intelligent spectrum sensing and positioning method of this embodiment can be shared based on actual needs, or some steps can be omitted or maintained. The detailed components of the neural network-based intelligent spectrum sensing and positioning method are as follows:

[0020] Step S110: Collect a set of multi-source spectrum sensing signals within the target area, wherein the set of multi-source spectrum sensing signals includes multi-band radio frequency signal units captured by different sensing nodes in the same time period.

[0021] In this embodiment, the selected target area is a complex electromagnetic environment area in the city center. This area contains multiple communication base stations, broadcast and television transmission towers, and various wireless devices, exhibiting radio frequency signals across multiple frequency bands and experiencing electromagnetic interference. To achieve accurate positioning of signal sources within this area, multiple sensing nodes are deployed in a grid-like distribution.

[0022] These sensing nodes are all dedicated devices with multi-band receiving capabilities. Each node is equipped with a high-sensitivity radio frequency receiving module, signal processing module, storage module, and communication module. All sensing nodes achieve time synchronization through the BeiDou satellite timing system, ensuring that all nodes can collect signals within the exact same time period.

[0023] Before signal acquisition begins, each sensing node performs system initialization. The initialization process includes scanning the frequency band of the RF receiving module to determine the receivable frequency range; configuring the parameters of the signal processing module, setting parameters such as bandwidth and gain for signal sampling; checking the storage module to ensure sufficient storage space for storing the acquired signal data; and testing the communication module to ensure normal communication with other nodes and the control center.

[0024] After initialization, the control center sends a data acquisition start command to all sensing nodes. Upon receiving the command, each node simultaneously begins data acquisition. During the acquisition process, the radio frequency receiving module of each sensing node continuously scans the preset frequency band range. When a radio frequency signal is received, it can be transmitted to the signal processing module.

[0025] The signal processing module performs preliminary processing on the received signal, including removing the DC component, filtering to eliminate out-of-band interference, and amplifying or attenuating the signal to bring it within a suitable amplitude range. The processed signal is then converted into a digital signal, and metadata such as the corresponding node identifier, acquisition timestamp, and signal frequency band information are added before being stored in the storage module.

[0026] After a period of data acquisition, the control center sends a stop command, and all nodes cease their acquisition operations. Subsequently, each sensing node uploads its stored radio frequency (RF) signal units to the control center via its communication module. The control center then aggregates the RF signal units uploaded by all nodes, forming a multi-source spectrum sensing signal set. This set contains RF signal units from multiple frequency bands captured by different sensing nodes within the same time period, and each signal unit carries complete metadata.

[0027] Step S120: Perform cross-node collaborative calibration processing on the radio frequency signal units in the multi-source spectrum sensing signal set to generate a collaborative sensing signal set with a unified spectrum reference.

[0028] Due to differences in hardware devices among different sensing nodes, such as variations in the frequency response, phase characteristics, and gain of the RF front-end, and potential differences in the electromagnetic interference conditions of the environments in which each node operates, the RF signal units in the multi-source spectrum sensing signal set exhibit deviations in their spectral references. These deviations affect the accuracy of subsequent feature extraction and localization analysis, thus necessitating cross-node collaborative calibration.

[0029] Step S121: Extract the spectral reference parameters of the radio frequency signal units corresponding to each sensing node in the multi-source spectrum sensing signal set. The spectral reference parameters include the signal frequency deviation value and the phase offset.

[0030] For each radio frequency (RF) signal unit in the multi-source spectrum sensing signal set, the signal units corresponding to different sensing nodes are distinguished based on the node identifier carried by the RF signal unit. For each sensing node, the spectral reference parameters of all its corresponding RF signal units are extracted.

[0031] When extracting the signal frequency deviation value, a high-precision standard frequency signal provided by the control center is used as a reference. For each radio frequency signal unit, its actual frequency is compared with the standard frequency, and the difference between the two is the signal frequency deviation value of that signal unit. In specific operation, a dedicated frequency comparison circuit is used to input the signal frequency received by the sensing node and the standard frequency into a comparator. The difference signal output by the comparator is processed to obtain the signal frequency deviation value.

[0032] When extracting the phase offset, a stable phase point within the RF signal unit is selected as a reference point, such as the signal's initial zero-crossing point. The phase of the signal captured by each sensing node at this reference point is compared with the phase of the standard signal at the same reference point; the difference between the two is the phase offset of that signal unit. Phase comparison is achieved through a phase detector. Two signals are input into the phase detector, and the phase difference signal output by the detector is processed to obtain the phase offset.

[0033] Using the above method, the signal frequency deviation value and phase offset are extracted for each radio frequency signal unit corresponding to each sensing node, forming a set of spectral reference parameters for each node.

[0034] Step S122: Calculate the deviation coefficients between the spectral reference parameters of different sensing nodes and establish the spectral deviation matrix between nodes.

[0035] Based on the spectral reference parameters of each sensing node extracted in step S121, the deviation coefficient between different sensing nodes is calculated. For the signal frequency deviation value, the frequency deviation coefficient between any two sensing nodes is calculated as follows: take the absolute value of the difference between the signal frequency deviation values ​​of the two nodes, and then divide it by the signal frequency deviation value of one of the nodes (select a non-zero signal frequency deviation value as the denominator; if the signal frequency deviation values ​​of both nodes are zero, then the frequency deviation coefficient is zero).

[0036] For phase offset, the phase deviation coefficient between two sensing nodes is calculated in a similar way: take the absolute value of the difference between the phase offsets of the two nodes, and then divide it by the phase offset of one of the nodes (again, choose a non-zero phase offset as the denominator; if both are zero, the phase deviation coefficient is zero).

[0037] The calculated frequency and phase deviation coefficients between all different nodes are arranged according to the node correspondence to construct a node spectral deviation matrix. The rows and columns of the matrix correspond to different sensing nodes, and each element in the matrix contains the frequency and phase deviation coefficients between the corresponding two nodes. For example, the element in the m-th row and n-th column of the matrix represents the frequency and phase deviation coefficients between the m-th and n-th sensing nodes.

[0038] Step S123: Based on the inter-node spectral deviation matrix, determine the master calibration node and the slave calibration node, wherein the master calibration node is the set of signal units corresponding to the sensing node with the most stable spectral reference parameters.

[0039] The inter-node spectral deviation matrix is ​​analyzed to evaluate the stability of the spectral reference parameters of each sensing node. Specifically, for each sensing node, the average value of its frequency deviation coefficient and the average value of its phase deviation coefficient compared to all other sensing nodes are calculated. These two average values ​​collectively reflect the overall deviation of the node's spectral reference parameters from those of other nodes.

[0040] The smaller the average value, the smaller the deviation of the spectral reference parameter of that node from other nodes, indicating that the spectral reference parameter of that node is more stable. These two average values ​​of all sensing nodes are compared, and the sensing node with the two smallest average values ​​is selected as the master calibration node, with its corresponding set of signal elements serving as the reference for the entire calibration process.

[0041] The remaining sensing nodes are identified as slave calibration nodes, whose signal units need to be calibrated with reference to the spectral reference of the master calibration node.

[0042] Step S124: Compare the spectral reference parameters of the RF signal unit of the slave calibration node with the spectral reference parameters of the master calibration node to generate the spectral calibration coefficients for each slave calibration node.

[0043] For each slave calibration node, the spectral reference parameters of all its radio frequency signal units are compared with the spectral reference parameters of the corresponding radio frequency signal units in the same frequency band of the master calibration node.

[0044] For the signal frequency deviation, the difference between the frequency deviation of the signal in a certain frequency band from the calibration node and the frequency deviation of the signal in the same frequency band from the master calibration node is calculated. A frequency calibration coefficient is generated based on this difference. The value of the frequency calibration coefficient is related to this difference, so that after adjusting the frequency deviation of the slave calibration node using this coefficient, the frequency deviation of the slave calibration node can be kept consistent with that of the master calibration node.

[0045] For phase offset, the difference between the phase offset of the signal in a certain frequency band from the calibration node and the phase offset of the signal in the same frequency band from the master calibration node is calculated. Based on this difference, a phase calibration coefficient is generated to adjust the phase offset of the calibration node so that it is consistent with the phase offset of the master calibration node.

[0046] Each signal unit in a different frequency band of a calibration node may correspond to a different spectral calibration coefficient. Therefore, it is necessary to generate corresponding frequency calibration coefficients and phase calibration coefficients for each signal unit in a frequency band of a calibration node.

[0047] Step S125: Perform frequency compensation and phase correction on the radio frequency signal unit of the slave calibration node according to the spectrum calibration coefficient, so that the radio frequency signal unit of the slave calibration node and the master calibration node maintain the same spectrum reference.

[0048] For each RF signal unit of each calibration node, frequency compensation is performed using the corresponding frequency calibration coefficient generated in step S124. Specifically, the compensation value determined by the frequency calibration coefficient is added to or subtracted from the frequency value of the RF signal unit of the calibration node, so that the compensated signal frequency is consistent with the frequency of the signal in the same frequency band of the main calibration node.

[0049] Simultaneously, phase correction is performed using the corresponding phase calibration coefficient. The correction value determined by the phase calibration coefficient is added to or subtracted from the phase value of the RF signal unit of the calibration node to ensure that the phase of the corrected signal is consistent with the phase of the signal in the same frequency band of the main calibration node.

[0050] During frequency compensation and phase correction, it is necessary to monitor the stability of the calibrated signal in real time to avoid overcompensation or correction that could lead to signal distortion. If an anomaly is detected in the calibrated signal, the generation process of the spectral calibration coefficients should be re-examined, adjusted, and then recalibrated.

[0051] Step S126: Merge the RF signal units of the calibrated slave calibration node with the RF signal units of the master calibration node to form a collaborative sensing signal set containing node identifiers. The spectral reference error of each RF signal unit in the collaborative sensing signal set is controlled within a preset range.

[0052] After all RF signal units from the calibration nodes have been calibrated, these calibrated signal units are merged with the signal units from the main calibration node. During the merging process, the original metadata of each signal unit, such as the node identifier, acquisition timestamp, and frequency band information, is retained so that the source and acquisition status of the signal can be traced in subsequent processing.

[0053] After merging, a collaborative sensing signal set is formed. Spectral reference error detection is performed on each radio frequency signal unit within this set. The detection method involves randomly selecting a certain number of signal units in the same frequency band from different nodes and calculating their frequency and phase deviations. If these deviations are all within a preset error range, the collaborative sensing signal set is considered to meet the requirements. If deviations exceed the error range, the corresponding signal units need to be recalibrated until the spectral reference error of all signal units is controlled within the preset range.

[0054] Step S130: Based on the preset spectral feature extraction rules, perform hierarchical feature mining on the collaborative sensing signal set to obtain the spectral hole features and channel attenuation features of each radio frequency signal unit.

[0055] The radio frequency signal units in the collaborative sensing signal set already possess a unified spectral benchmark. The next step is to extract spectral hole features and channel attenuation features that reflect the characteristics of the signal source. Hierarchical feature mining extracts features from the signal according to different dimensions and levels to comprehensively obtain the signal's feature information.

[0056] Step S131: Divide the radio frequency signal units in the collaborative sensing signal set into multiple sub-frequency band signal units according to the frequency band division rules. Each sub-frequency band signal unit corresponds to a specific frequency range.

[0057] A preset frequency band division rule is established based on the distribution of common signal frequency bands within the target area. For example, the entire receivable frequency range is divided into multiple consecutive sub-frequency band intervals according to certain frequency intervals.

[0058] For each radio frequency signal unit in the collaborative sensing signal set, it is divided into corresponding sub-frequency band intervals according to the frequency band information it carries, forming multiple sub-frequency band signal units. Each sub-frequency band signal unit contains all the signal units captured by all sensing nodes within that sub-frequency band interval.

[0059] During the partitioning process, it is necessary to ensure that the boundaries of each sub-band interval are clear to avoid unclear assignment of signal units. For signal units located on the boundaries of sub-band intervals, they are partitioned according to preset assignment rules, such as being assigned to a lower frequency sub-band interval.

[0060] Step S132: Perform power spectral density analysis on each sub-band signal unit, identify frequency ranges with power spectral density below a preset threshold, and use the start and end frequencies and duration of the frequency ranges as core parameters of the spectral hole characteristics.

[0061] For each sub-band signal unit, the power spectral density estimation method is used for analysis. Power spectral density estimation transforms the signal from the time domain to the frequency domain by performing a Fourier transform, thus obtaining the power distribution of the signal at different frequency points.

[0062] After obtaining the power spectral density distribution, it is compared with a preset threshold. The preset threshold is determined based on the normal power spectral density level of the sub-band, typically a certain proportion of the average power spectral density within that sub-band. All frequency ranges with power spectral density below the preset threshold are identified.

[0063] For each identified frequency interval, its start and end frequencies are recorded to determine the range of that interval. Simultaneously, based on the signal acquisition timestamp, the duration of that frequency interval within the acquisition period is calculated. These start and end frequencies and durations are used as the core parameters for the spectral hole characteristics of the signal unit in that sub-band.

[0064] Step S133: Extract the signal strength attenuation curve of the sub-band signal unit on the propagation path, calculate the slope change value of the attenuation curve in different distance intervals, and use the slope change value and the inflection point coordinates of the attenuation curve as the basic parameters of the channel attenuation characteristics.

[0065] For each sub-band signal unit, a signal strength attenuation curve is constructed along the propagation path by combining the location information of the sensing node and the possible propagation path of the signal source. The horizontal axis of the signal strength attenuation curve represents the propagation distance, and the vertical axis represents the signal strength.

[0066] To construct this curve, it is necessary to obtain the signal intensity values ​​at different propagation distances. By analyzing the intensity of the same sub-frequency band signal units captured by sensing nodes at multiple different locations, and combining the distance relationship between the nodes, a curve showing the change of signal intensity with propagation distance is obtained.

[0067] After obtaining the attenuation curve, the propagation distance is divided into multiple consecutive distance intervals. For each distance interval, the slope of the attenuation curve within that interval is calculated. The slope is calculated by taking the ratio of the signal strength difference between the two endpoints of the curve within the interval to the distance difference.

[0068] Compare the slopes of adjacent distance intervals to obtain the slope change values. Simultaneously, identify the inflection points in the attenuation curve—points where the slope changes significantly—and record the coordinates of these inflection points (including distance values ​​and corresponding signal strength values). Use these slope change values ​​and inflection point coordinates as the fundamental parameters for the channel attenuation characteristics of the signal unit in this sub-band.

[0069] Step S134: Extend the core parameters of the spectral hole feature by time dimension, count the frequency and interval of the same frequency range in continuous time period, and generate spectral hole features containing time distribution attributes.

[0070] The acquisition period is divided into multiple consecutive sub-periods, each with the same duration. For each frequency range in the core parameters of the spectral hole characteristics obtained in step S132, it is counted whether it appears in each sub-period.

[0071] Based on the statistical results, calculate the frequency of occurrence of this frequency interval within a continuous time period, i.e., the total number of sub-time periods. Simultaneously, calculate the interval period between occurrences of this frequency interval, i.e., the average number of sub-time periods between two consecutive occurrences.

[0072] By combining these occurrence frequencies and intervals with the original start and end frequencies and durations, a spectral hole feature containing temporal distribution attributes is generated. This feature not only reflects the frequency range and duration of the spectral hole but also its temporal distribution pattern.

[0073] Step S135: Spatial dimension correlation is performed on the basic parameters of the channel attenuation feature. Combined with the geographical distribution information of the sensing nodes, the attenuation difference coefficient of the same signal unit captured by different nodes is calculated to generate channel attenuation features containing spatial distribution attributes.

[0074] Acquire the geographic coordinates of all sensing nodes, which are pre-stored during node deployment. For the same sub-band signal element, collect the fundamental parameters of the channel attenuation characteristics of that signal element captured by different sensing nodes.

[0075] Based on the geographical distribution of sensing nodes, the distance between different nodes is calculated. The attenuation difference coefficient is calculated based on the slope changes and inflection point coordinates of the attenuation curves of the same signal unit captured by different nodes. The attenuation difference coefficient is calculated by taking the absolute value of the difference between corresponding parameters of two nodes and then dividing it by the distance between the two nodes (the distance is not zero).

[0076] The calculated attenuation difference coefficient is combined with the original slope change value and inflection point coordinates to generate a channel attenuation feature that includes spatial distribution attributes. This feature reflects the spatial variation of channel attenuation.

[0077] Step S136: Align the spectral hole features with temporal distribution attributes and the channel attenuation features with spatial distribution attributes in terms of dimensions to form a feature set that can be input into the neural network.

[0078] This analysis examines the number of dimensions and the meaning of each dimension in spectral hole features (which have temporal distribution attributes) and channel attenuation features (which have spatial distribution attributes). Spectral hole features may include dimensions such as frequency range, duration, frequency of occurrence, and interval period; channel attenuation features may include dimensions such as slope change value, inflection point coordinates, and attenuation difference coefficient.

[0079] By using feature mapping, the dimensions of two features are unified. For features with different numbers of dimensions, the feature with fewer dimensions is expanded by adding new dimensions and assigning reasonable default values; for dimensions with different meanings, feature transformation is used to make them comparable.

[0080] After dimensional alignment, the two features are combined to form a comprehensive feature set. Each element in this feature set contains information about the corresponding spectral hole feature and channel attenuation feature, and the dimensions of each feature are consistent, allowing them to be directly input into the neural network for processing.

[0081] Step S140: Input the spectrum hole feature and the channel attenuation feature into the trained spectrum positioning neural network, perform spatiotemporal correlation positioning analysis, and output the initial positioning parameter set of the radio frequency signal unit.

[0082] The spectrum localization neural network is a deep learning model specifically designed for signal source localization based on spectrum features. This model has been trained with a large amount of training data and possesses the ability to extract spatiotemporal correlation information from spectrum features and perform localization analysis. The spectrum hole features and channel attenuation features obtained in step S130 are input into this network, and through multi-layer processing of the network, the initial localization parameter set of the radio frequency signal unit is obtained.

[0083] Step S141: Input the spectral hole feature and the channel attenuation feature into the input layer of the spectral localization neural network, and convert them into a tensor form that the network can recognize through the feature mapping function of the input layer. The feature mapping function adopts a linear transformation method to map the feature parameter values ​​to a preset numerical range.

[0084] Spectral hole features and channel attenuation features are input as feature vectors to the input layer of the spectrum localization neural network. The input layer first transforms these feature vectors into a tensor form that the network can process. The dimension of the tensor is determined by the number and dimension of the features. For example, if the feature set contains multiple feature vectors, and each feature vector has multiple dimensions, then the dimension of the tensor is the number of feature vectors multiplied by the dimension of each vector.

[0085] Meanwhile, the input layer applies a feature mapping function to process the feature parameter values. The feature mapping function uses a linear transformation method, and the transformation process is as follows: for each feature parameter value, subtract the minimum value of the feature parameter in the training data, and then divide by the difference between the maximum and minimum values ​​of the feature parameter in the training data, thus mapping the feature parameter value to a preset numerical range (usually between 0 and 1).

[0086] The linear transformation described above ensures that feature parameter values ​​of different magnitudes fall within the same numerical range, preventing excessively large or small feature parameter values ​​from negatively impacting the network's training performance and prediction accuracy. The transformed tensor-form features will then be passed to the next layer of the network for further processing.

[0087] Step S142: Input the tensor form features output by the input layer into the spatiotemporal fusion layer of the spectral localization neural network. The spatiotemporal fusion layer includes a parallel temporal feature extraction sublayer and a spatial feature extraction sublayer.

[0088] The tensor-form features output from the input layer are simultaneously fed into the temporal feature extraction sublayer and the spatial feature extraction sublayer of the spatiotemporal fusion layer. These two sublayers work in parallel, extracting features from different dimensions respectively.

[0089] The temporal feature extraction sublayer focuses on extracting time-related information from features, analyzing the patterns and relationships of feature changes over time; the spatial feature extraction sublayer focuses on extracting spatially related information, analyzing the distribution and relationships of features in the spatial dimension. Through this parallel processing approach, both temporal and spatial information within the features can be captured simultaneously.

[0090] Step S143: The time feature extraction sub-layer adopts a recurrent neural network structure to perform sequence modeling on the time distribution attributes of spectral hole features and generate time-related feature vectors.

[0091] The recurrent neural network structure of the time feature extraction sublayer includes input gate, forget gate, cell state and output gate, which can process sequence data and capture the dependencies in time series.

[0092] Step S1431: Determine the number of hidden layer neurons and the time step of the recurrent neural network. The time step corresponds to the sampling interval of the temporal distribution attribute of the spectral hole feature.

[0093] The number of neurons in the hidden layer of a recurrent neural network is determined based on the complexity of the temporal distribution attributes of the spectral hole features. A higher number of neurons results in greater expressive power but also higher computational complexity; a balance must be struck between the two.

[0094] The determination of the time step is related to the sampling interval of the temporal distribution attribute of the spectral hole feature. The sampling interval is the duration of the sub-period divided in step S134. The size of the time step is equal to the number of sampling intervals, that is, the number of sub-periods into which the entire acquisition period is divided.

[0095] Step S1432: Arrange the temporal distribution attribute parameters of the spectral hole features in chronological order to form a time series feature matrix, with each time step corresponding to a set of spectral hole feature parameters.

[0096] The temporal distribution attributes of the spectral hole features, such as frequency ranges and occurrence frequencies within each sub-period, are arranged in chronological order. Each sub-period corresponds to a time step, and the parameters of each time step form a vector. The vectors of all time steps are combined to form a time series feature matrix. The number of rows in this matrix represents the time step length, and the number of columns represents the number of feature parameters at each time step.

[0097] Step S1433: Input the time series feature matrix into the input gate of the recurrent neural network. The input gate calculates the feature update weights based on the feature parameters of the current time step and the hidden state of the previous time step.

[0098] The time-series feature matrix is ​​sequentially input into the input gate of the recurrent neural network at each time step. For each time step, the input gate receives the feature parameters of the current time step and the hidden state of the previous time step. The input gate processes these inputs through a sigmoid activation function and outputs feature update weights. The feature update weights range from 0 to 1 and are used to control the degree to which the feature parameters of the current time step update the cell state. The closer the weight is to 1, the greater the influence of the current feature parameters.

[0099] Step S1434: Calculate the forgetting weights of historical features based on the feature parameters of the current time step and the hidden state of the previous time step using the forgetting gate of the recurrent neural network.

[0100] The forget gate and input gate both receive the feature parameters of the current time step and the hidden state of the previous time step. They are also processed by the sigmoid activation function, outputting the forgetting weights of the historical features. The forgetting weights range from 0 to 1, controlling the degree to which historical features are retained in the cell state of the previous time step. The closer the weight is to 0, the more historical features are forgotten.

[0101] Step S1435: Combine the outputs of the input gate and the forget gate to update the cell state of the recurrent neural network and obtain the cell state value at the current time step.

[0102] The forgetting weights output by the forget gate are multiplied by the cell state from the previous time step to obtain the historical features that need to be retained. The feature update weights output by the input gate are multiplied by the tanh function-processed value of the feature parameters at the current time step to obtain the new features. These two parts are then added together to obtain the cell state value at the current time step, thus updating the cell state.

[0103] Step S1436: Calculate the output weights based on the cell state and current feature parameters at the current time step through the output gate of the recurrent neural network, and generate the hidden state vector at the current time step.

[0104] The output gate receives the cell state and current feature parameters at the current time step and calculates the output weights using the sigmoid activation function. The cell state at the current time step is then processed by the tanh function and multiplied by the output weights to obtain the hidden state vector for the current time step. This vector integrates the feature information from the current time step and historical feature information, reflecting the time series characteristics up to the current time step.

[0105] Step S1437: Perform global pooling on the hidden state vectors of all time steps, extract key features in the time dimension, and generate a time-related feature vector containing time dependencies.

[0106] After the hidden state vectors are generated at all time steps, a global pooling operation is performed on them. Global pooling can be performed using average pooling or max pooling. Average pooling calculates the average value of the corresponding positions of all hidden state vectors, while max pooling takes the maximum value of the corresponding positions of all hidden state vectors.

[0107] Global pooling can extract the most critical temporal features from the hidden state vectors of multiple time steps, resulting in a fixed-length vector that contains temporal correlation feature vectors containing temporal dependencies.

[0108] Step S144: The spatial feature extraction sub-layer adopts a convolutional neural network structure to extract local features of the spatial distribution attributes of channel attenuation features and generate spatial correlation feature vectors.

[0109] The convolutional neural network structure of the spatial feature extraction sublayer contains multiple convolutional layers and pooling layers, which can effectively extract local and global spatial features from the data.

[0110] Step S1441: Convert the spatial distribution attribute parameters of the channel attenuation features into a two-dimensional feature map. The horizontal and vertical axes of the map correspond to the geographical coordinates of the sensing node, and the map element values ​​correspond to the channel attenuation feature parameters.

[0111] Based on the geographic coordinates of the sensing nodes, the location of each node is determined on a two-dimensional plane. Spatial distribution parameters of channel attenuation characteristics, such as the attenuation difference coefficient for each node, are assigned to the corresponding node locations on the two-dimensional plane, forming a two-dimensional feature map. For locations without sensing nodes, appropriate feature parameter values ​​are assigned through interpolation, ensuring that the feature map fully reflects the spatial distribution of channel attenuation characteristics across the entire target area.

[0112] Step S1442: Input the two-dimensional feature map into the first convolutional layer of the convolutional neural network, and perform sliding window convolution through multiple convolutional kernels of different sizes to generate multiple sets of local spatial feature maps.

[0113] After the two-dimensional feature map is input into the first convolutional layer, the layer performs convolution operations on it using multiple convolutional kernels of different sizes. Each convolutional kernel slides on the feature map with a preset stride. At each sliding position, the convolutional kernel is multiplied and summed with the elements of the corresponding region of the feature map to obtain a feature value.

[0114] Convolutional kernels of different sizes can extract local spatial features of different ranges. For example, smaller kernels extract local detail features, while larger kernels extract features from a wider area. Through convolution operations with multiple kernels, multiple sets of local spatial feature maps are generated, with each set of feature maps corresponding to the features extracted by one kernel.

[0115] Step S1443: Perform batch normalization on the multiple sets of local spatial feature maps to adjust the feature value distribution range.

[0116] Batch normalization is performed on each set of local spatial feature maps, calculating the mean and variance of all eigenvalues ​​in that set. Then, each eigenvalue is subtracted from the mean and divided by the square root of the variance (with a minimum value added to avoid dividing by zero) to obtain the normalized eigenvalue.

[0117] Batch normalization can make the feature values ​​of each feature map distributed within a similar range, which can speed up the training and convergence of the network and reduce the occurrence of overfitting.

[0118] Step S1444: Input the batch-normalized local spatial feature map into the pooling layer of the convolutional neural network, and use max pooling for downsampling to retain key features and reduce dimensionality.

[0119] The pooling layer downsamples the batch-normalized local spatial feature maps. The max pooling operation divides the feature map into multiple non-overlapping regions, each with a preset pooling window size, and takes the maximum value within each region as the pooling result for that region.

[0120] Max pooling can reduce the dimensionality of feature maps and lower the computational complexity of the network while preserving key information in local spatial features.

[0121] Step S1445: Input the feature map output by the pooling layer into the second convolutional layer of the convolutional neural network, and perform deep convolution on the feature map through more convolutional kernels to extract higher-level spatial correlation features.

[0122] The second convolutional layer uses a greater number of convolutional kernels than the first convolutional layer to convolve the feature maps output by the pooling layer. The size of these convolutional kernels can be the same as or different from that of the first convolutional layer, and they are designed to extract higher-level, more abstract spatial correlation features from the downsampled feature maps.

[0123] The convolution process is similar to that of the first convolutional layer. New feature maps are generated through sliding window convolution, which reflect more complex spatial relationships between features.

[0124] Step S1446: Flatten the feature map output by the second convolutional layer, converting the two-dimensional feature map into a one-dimensional feature vector, which serves as the spatial correlation feature vector.

[0125] The feature map output by the second convolutional layer is still two-dimensional and needs to be converted into a one-dimensional feature vector before it can be fused with the temporally correlated feature vector. The flattening operation arranges all the elements of the two-dimensional feature map in a certain order (such as row-major) into a one-dimensional vector, the length of which is the number of rows multiplied by the number of columns of the two-dimensional feature map.

[0126] The resulting one-dimensional feature vector is the spatial correlation feature vector, which contains key information in the spatial distribution attributes of channel attenuation features.

[0127] Step S145: Input the time-related feature vector and the spatial-related feature vector into the feature spatiotemporal fusion layer's feature splicing module, and fuse them according to the feature importance weight to generate a spatiotemporal joint feature vector.

[0128] After the temporal and spatial correlation feature vectors are input into the feature concatenation module, their feature importance weights are first determined. These weights are based on the contribution of each feature to the localization analysis. By analyzing the training data, the influence of different features on the localization results is calculated, with features having a greater impact assigned higher weights.

[0129] Then, each element of the temporally correlated feature vector is multiplied by its corresponding weight, and each element of the spatially correlated feature vector is also multiplied by its corresponding weight. Finally, the two weighted vectors are concatenated sequentially to form a longer one-dimensional vector, namely the spatiotemporal joint feature vector. This vector simultaneously contains temporal and spatial correlated feature information, comprehensively reflecting the spatiotemporal characteristics of the signal source.

[0130] Step S146: Input the spatiotemporal joint feature vector into the positioning output layer of the spectrum positioning neural network, perform nonlinear mapping through a multilayer perceptron structure, and output an initial positioning parameter set containing signal source distance estimates and azimuth angle estimates.

[0131] The spatiotemporal joint feature vector input localization output layer adopts a multilayer perceptron structure, containing multiple fully connected layers. Neurons in each fully connected layer are connected to all neurons in the previous layer, and the input features are nonlinearly transformed through weight parameters.

[0132] During the processing of the multilayer perceptron, the spatiotemporal joint feature vector is transformed through multiple fully connected layers, gradually mapping to the parameter space related to localization. The output of the last fully connected layer is the initial localization parameter set, which contains the distance estimate and azimuth estimate of the signal source corresponding to each radio frequency signal unit.

[0133] The distance estimate represents the distance between the signal source and the reference point (usually a sensing node or control center), while the azimuth estimate represents the azimuth angle of the signal source relative to the reference point.

[0134] Step S150: Perform multi-node fusion optimization processing on the initial positioning parameter set to generate accurate spectrum positioning coordinates of the target area, and generate a spectrum sensing positioning command containing the signal source orientation identifier based on the accurate spectrum positioning coordinates. Transmit the spectrum sensing positioning command to the spectrum management terminal to complete the positioning control operation.

[0135] The initial set of positioning parameters is obtained based on the feature analysis of a single node, which may contain some errors. By using multi-node fusion optimization processing to integrate the positioning information of multiple nodes, the accuracy of positioning can be improved, resulting in precise spectral positioning coordinates. Then, positioning commands are generated based on these coordinates to achieve positioning and control of the signal source.

[0136] Step S151: Extract the signal source distance estimate and azimuth estimate corresponding to each sensing node in the initial positioning parameter set.

[0137] From the initial set of positioning parameters, based on the node identifier corresponding to each parameter, the estimated distance and azimuth of the signal source corresponding to each sensing node are extracted. Each sensing node has a set of corresponding distance and azimuth estimates, which reflect the node's estimation of the signal source's location.

[0138] During the extraction process, it is necessary to ensure that each parameter correctly corresponds to its corresponding sensing node to avoid parameter confusion. After extraction, the estimated values ​​of each node are organized to form a set of estimated values ​​categorized by node.

[0139] Step S152: Based on the known geographic coordinates of each sensing node, convert the signal source distance estimate and azimuth estimate of each sensing node into the coordinates of candidate positioning points in three-dimensional space.

[0140] Step S1521: Obtain the known geographic coordinates of each sensing node in the three-dimensional coordinate system, including longitude, latitude and altitude.

[0141] During deployment, the geographic coordinates of each sensing node are measured and stored using high-precision positioning equipment. These coordinates are based on a preset three-dimensional coordinate system (such as the WGS84 coordinate system). The longitude, latitude, and altitude values ​​of each sensing node are read from the storage device as the reference coordinates for transformation.

[0142] Step S1522: Decompose the azimuth estimate of each sensing node into horizontal azimuth and vertical azimuth.

[0143] The estimated azimuth angle is usually a composite angle that needs to be decomposed into horizontal and vertical angles. The horizontal azimuth angle represents the orientation of the signal source relative to the sensing node on the horizontal plane, with true north as the reference and rotated clockwise. The vertical azimuth angle represents the pitch angle of the signal source relative to the sensing node on the vertical plane, with the horizontal direction as the reference and either upward or downward.

[0144] The decomposition process is achieved through trigonometric function relationships. Based on the estimated azimuth angle and the preset decomposition rules, the horizontal and vertical azimuth angles are calculated.

[0145] Step S1523: Based on trigonometric relationships, convert the estimated distance to the signal source, the horizontal azimuth angle, and the vertical azimuth angle into three-dimensional coordinate offsets relative to the sensing node, including offsets in the east-west, north-south, and vertical directions.

[0146] Based on the horizontal azimuth angle, the unit vectors in the east-west and north-south directions are calculated using cosine and sine functions; based on the vertical azimuth angle, the unit vector in the vertical direction is calculated using a sine function. The estimated distance to the signal source is then multiplied by these three unit vectors to obtain the east-west, north-south, and vertical offsets relative to the sensing node.

[0147] Step S1524: Superimpose the three-dimensional coordinate offset with the known geographic coordinates of the sensing node to obtain the absolute coordinate value of the signal source relative to the three-dimensional coordinate system.

[0148] The east-west and north-south offsets obtained in step S1523 are added to the longitude and latitude values ​​of the sensing node, respectively, to obtain the estimated longitude and latitude values ​​of the signal source; the vertical offset is added to the altitude value of the sensing node to obtain the estimated altitude value of the signal source. These three estimates together constitute the absolute coordinates of the signal source relative to the three-dimensional coordinate system.

[0149] Step S1525: Perform coordinate system normalization on the absolute coordinate values ​​to convert them into coordinate values ​​under the preset target coordinate system, which are used as candidate positioning point coordinates.

[0150] Since different application scenarios may use different coordinate systems, it is necessary to convert absolute coordinate values ​​into coordinate values ​​in a preset target coordinate system. Coordinate transformation is achieved through preset coordinate transformation formulas, converting longitude, latitude, and altitude values ​​into X, Y, and Z coordinate values ​​in the target coordinate system.

[0151] After the conversion is completed, the resulting coordinate values ​​are the coordinates of the candidate positioning point corresponding to the sensing node.

[0152] Step S1526: Record the sensing node identifier and signal acquisition timestamp corresponding to the coordinates of each candidate positioning point.

[0153] To trace the source and acquisition time of candidate location point coordinates in subsequent processing, the coordinates of each candidate location point are associated with and recorded along with the corresponding sensing node identifier and signal acquisition timestamp. This information will be used to determine the validity and relevance of candidate points during clustering and fusion optimization.

[0154] Step S153: Calculate the spatial distance between the coordinates of all candidate positioning points and construct the candidate point distance matrix.

[0155] For all candidate location point coordinates obtained in step S152, calculate the spatial distance between any two candidate points. The spatial distance is calculated using the distance formula in three-dimensional space, that is, subtract the X, Y, and Z coordinate values ​​of the two points respectively, take the square root of the sum of the squares.

[0156] All calculated spatial distances are arranged according to the correspondence between candidate points to construct a candidate point distance matrix. The rows and columns of the matrix correspond to different candidate positioning points, and each element in the matrix represents the spatial distance between two corresponding candidate points.

[0157] Step S154: Use density clustering algorithm to cluster the candidate location point coordinates, and group the candidate location points with a spatial distance less than a preset threshold into the same cluster.

[0158] Step S1541: Set the neighborhood radius parameter and minimum number of contained points parameter for the density clustering algorithm. The neighborhood radius is determined based on the distribution density of the candidate location point coordinates, and the minimum number of contained points is determined based on the number of sensing nodes.

[0159] The neighborhood radius parameter needs to be set considering the distribution of candidate points. If the candidate points are densely distributed, the neighborhood radius can be set smaller; if the distribution is sparse, the neighborhood radius can be set larger. The minimum number of points to include parameter is usually set as a certain proportion of the number of sensing nodes to ensure that each cluster contains a sufficient number of candidate points to guarantee its reliability.

[0160] Step S1542: Calculate the number of other candidate positioning points within the neighborhood radius of each candidate positioning point coordinate, and mark the candidate positioning points with a number greater than or equal to the minimum number of contained points as core points.

[0161] For each candidate localization point, draw a sphere with that point as the center and the radius as the neighborhood radius, and count the number of other candidate localization points contained within the sphere. If this number is greater than or equal to the minimum number of contained points, then mark the candidate localization point as a core point. A core point indicates that there are a sufficient number of candidate points around it, forming a dense region.

[0162] Step S1543: Assign candidate locations whose spatial distance from the core point is less than the neighborhood radius to the cluster where the core point is located, thus forming the initial cluster.

[0163] For each core point, all candidate locations (including the core point itself) whose spatial distance from the core point is less than the neighborhood radius are grouped into the same cluster, forming an initial cluster. These candidate points are spatially close to each other and may correspond to the location of the same signal source.

[0164] Step S1544: For candidate locations that are not marked as core points and do not belong to any initial cluster, determine whether their spatial distance to a core point in an initial cluster is less than the neighborhood radius. If so, they are assigned to that initial cluster.

[0165] There are some candidate locations that are not core points themselves, but may be close to a core point. For these candidate points, calculate their spatial distance to the core points in each initial cluster. If the distance to a core point is less than the neighborhood radius, then assign it to the initial cluster containing that core point.

[0166] Step S1545: Repeat the above steps until all candidate localization points are assigned to the corresponding cluster or marked as noise points.

[0167] Check if there are any unprocessed candidate localization points. If so, repeat steps S1543 and S1544 until all candidate localization points are either assigned to a cluster or marked as noise points. Noise points are candidate localization points that are neither core points nor close to any core points. These points may be outliers caused by measurement errors or interference.

[0168] Step S1546: Remove clusters containing fewer candidate location points than the minimum number of points, and retain the remaining clusters as the final result.

[0169] The resulting clusters are filtered out, removing those containing fewer candidate location points than the minimum number of points, as these clusters may be unreliable. The remaining clusters are used as the final clustering result, with each cluster representing a possible signal source location region.

[0170] Step S155: Calculate the geometric mean of the coordinates of all candidate location points in each cluster to obtain the coordinates of the center location point of the cluster.

[0171] For each retained cluster, the coordinates of all candidate points are collected. For the X, Y, and Z coordinate components, the average value of the corresponding components for all candidate points is calculated to obtain the X, Y, and Z coordinates of the center point of the cluster. The geometric mean reflects the central tendency of the candidate points in the cluster and serves as the representative coordinate of the cluster.

[0172] Step S156: Count the number of candidate positioning points contained in each cluster, and use the coordinates of the center positioning point of the cluster with the most candidate positioning points as the initial fusion positioning coordinates.

[0173] The number of candidate localization points in each cluster is counted. The higher the number, the more reliable the cluster is, and the more likely it is to correspond to the actual signal source location. The cluster with the most candidate localization points is selected, and its center localization point coordinates are used as the initial fused localization coordinates.

[0174] Step S157: Combine the signal strength confidence parameters of each sensing node to perform weighted correction on the preliminary fused positioning coordinates. The higher the signal strength confidence parameter, the greater the weight of the corresponding candidate positioning point in the correction.

[0175] The signal strength reliability parameter measures the reliability of the signal captured by the sensing node. The more stable the signal strength and the less interference, the higher the reliability parameter. This parameter can be calculated based on factors such as the signal-to-noise ratio and the range of signal strength fluctuations.

[0176] For each candidate location point in the cluster corresponding to the initially fused location coordinates, a weight is assigned to the candidate point based on the signal strength confidence parameter of its respective sensing node; the higher the confidence parameter, the greater the weight. Then, the coordinate value of each candidate point is multiplied by its corresponding weight, all weighted coordinate values ​​are summed, and finally divided by the sum of the weights to obtain the corrected location coordinates.

[0177] Step S158: Use the corrected positioning coordinates as the precise spectral positioning coordinates of the target area, wherein the error range of the precise spectral positioning coordinates is lower than a preset threshold.

[0178] The corrected positioning coordinates are evaluated for error. The average distance between the coordinates and the coordinates of each candidate positioning point in the cluster is calculated. If the average distance is lower than a preset threshold, the coordinates are considered reliable and used as the accurate spectral positioning coordinates of the target area. If the error exceeds the preset threshold, the clustering and fusion process needs to be re-examined to find the cause of the excessive error and make adjustments until accurate spectral positioning coordinates that meet the error requirements are obtained.

[0179] Step S159: Generate a spectrum sensing positioning command containing the signal source orientation identifier based on the precise spectrum positioning coordinates, and transmit the spectrum sensing positioning command to the spectrum management terminal to complete the positioning and control operation.

[0180] Based on precise spectrum positioning coordinates, the specific location of the signal source in the target area is determined, and a location identifier is generated. This identifier can be a directional description or coordinate mark relative to a reference point. The precise spectrum positioning coordinates and location identifier are integrated into the spectrum sensing positioning command. The command can also include additional information such as the frequency band information and signal strength of the signal source.

[0181] The spectrum sensing and positioning command is transmitted to the spectrum management terminal through the communication network. After receiving the command, the spectrum management terminal parses the information and marks the location and orientation of the signal source on the terminal's display interface. It then performs corresponding control operations according to the preset control rules, such as adjusting the operating parameters of relevant equipment to avoid signal interference, thereby completing the entire positioning and control process.

[0182] In the above embodiments, the neural network-based intelligent spectrum sensing and positioning system for performing the above method embodiments has at least one processor, a control module (chipset) coupled to at least one of the processors, a memory coupled to the control module, a non-volatile memory (NVM) / storage device coupled to the control module, at least one load to / output device coupled to the control module, and a network interface coupled to the control module.

[0183] The processor may include at least one single-core or multi-core processor, and may include any combination of general-purpose processors or special-purpose processors (e.g., graphics processors, application processors, baseband processors, etc.). For some alternative implementations, a neural network-based intelligent spectrum sensing positioning system can serve as the gateway or other electronic device described in the embodiments of this application.

[0184] In some alternative implementations, a neural network-based intelligent spectrum sensing and positioning system may include at least one computer-readable medium (e.g., a memory or NVM / storage device) having instructions and at least one processor fused with the at least one computer-readable medium and configured to execute the instructions to implement the module thereby performing the actions described in this disclosure.

[0185] In one embodiment, the control module may include any suitable interface controller to provide any suitable interface to at least one of the processors and / or any suitable device or component communicating with the control module.

[0186] The control module may include a memory controller module to provide an interface to the memory. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0187] The memory can be used, for example, to load and store data and / or instructions for a neural network-based intelligent spectrum sensing positioning system. In one embodiment, the memory may include any suitable volatile memory, such as suitable DRAM.

[0188] In one embodiment, the control module may include at least one load-to-output controller to provide an interface to the NVM / storage device and (at least one) load-to-output device.

[0189] For example, an NVM / storage device can be used to store data and / or instructions. An NVM / storage device may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable (at least one) non-volatile storage device (e.g., at least one hard disk drive (HDD), at least one optical disc (CD) drive, and / or at least one digital universal optical disc (DVD) drive).

[0190] NVM / storage devices may include storage resources that are physically part of a device mounted on which a neural network-based intelligent spectrum-aware positioning system is installed, or that can be accessed by the device without needing to be part of the device. For example, an NVM / storage device may be accessed over a network via at least one load-to-output device.

[0191] At least one loading / output device may provide an interface for the neural network-based intelligent spectrum sensing and positioning system to communicate with any other suitable device. The loading / output device may include communication components, pinyin components, sensor components, etc. A network interface may provide an interface for the neural network-based intelligent spectrum sensing and positioning system to communicate over at least one network. The neural network-based intelligent spectrum sensing and positioning system may wirelessly communicate with at least one component of a wireless network based on at least one wireless network prior and / or protocol, such as accessing a wireless network based on communication priors.

[0192] In one embodiment, at least one of the processors may be integrated with the logic of at least one controller of the control module (e.g., a memory controller module). In one embodiment, at least one of the processors may be integrated with the logic of at least one controller of the control module to form a system-level integration. In one embodiment, at least one of the processors may be fused with the logic of at least one controller of the control module on the same die. In one embodiment, at least one of the processors may be fused with the logic of at least one controller of the control module on the same die to form a system-on-a-chip (SoC).

[0193] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

[0194] This invention discloses a computer read storage medium that stores a computer program for electronic data interchange, wherein the computer program causes a computer to execute the steps in the neural network-based intelligent spectrum sensing and positioning method described in the foregoing embodiments.

[0195] This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps in the neural network-based intelligent spectrum sensing and positioning method described in the foregoing embodiments.

[0196] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any inventive effort.

[0197] Finally, it should be noted that the above-disclosed embodiments are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A neural network-based intelligent spectrum sensing and positioning method, characterized in that, The method includes: The system collects a set of multi-source spectrum sensing signals within the target area. The set of multi-source spectrum sensing signals includes radio frequency signal units of multiple frequency bands captured by different sensing nodes in the same time period. The radio frequency signal units are digital signals with added metadata, which includes the corresponding node identifier, acquisition timestamp, and signal frequency band information. Cross-node collaborative calibration processing is performed on the radio frequency signal units in the multi-source spectrum sensing signal set to generate a collaborative sensing signal set with a unified spectrum reference. Based on preset spectral feature extraction rules, hierarchical feature mining is performed on the collaborative sensing signal set to obtain the spectral hole features and channel attenuation features of each radio frequency signal unit; The spectrum hole features and the channel attenuation features are input into the trained spectrum positioning neural network to perform spatiotemporal correlation positioning analysis and output the initial positioning parameter set of the radio frequency signal unit. The initial positioning parameter set is subjected to multi-node fusion optimization processing to generate accurate spectrum positioning coordinates of the target area, and a spectrum sensing positioning command containing the signal source orientation identifier is generated based on the accurate spectrum positioning coordinates. The spectrum sensing positioning command is then transmitted to the spectrum management terminal to complete the positioning and control operation. The method, based on preset spectral feature extraction rules, performs hierarchical feature mining on the collaborative sensing signal set to obtain the spectral hole features and channel attenuation features of each radio frequency signal unit, including: According to the frequency band division rules, the radio frequency signal units in the collaborative sensing signal set are divided into multiple sub-frequency band signal units, and each sub-frequency band signal unit corresponds to a specific frequency range. Power spectral density analysis is performed on each sub-band signal unit to identify frequency ranges with power spectral density below a first preset threshold. The start and end frequencies and duration of the frequency ranges are used as the core parameters of the spectral hole characteristics. The signal strength attenuation curve of the sub-band signal unit on the propagation path is extracted, the slope change value of the attenuation curve in different distance intervals is calculated, and the slope change value and the inflection point coordinate of the attenuation curve are used as the basic parameters of the channel attenuation characteristics. The core parameters of the spectral hole feature are extended in the time dimension, and the frequency and interval of the same frequency range in a continuous time period are statistically analyzed to generate spectral hole features containing time distribution attributes. The basic parameters of the channel attenuation characteristics are correlated spatially, and combined with the geographical distribution information of the sensing nodes, the attenuation difference coefficient of the same radio frequency signal unit captured by different nodes is calculated to generate channel attenuation characteristics that include spatial distribution attributes. By dimensionally aligning the spectral hole features with temporal distribution attributes and the channel attenuation features with spatial distribution attributes, a feature set that can be input into a neural network is formed.

2. The intelligent spectrum sensing and positioning method based on neural networks according to claim 1, characterized in that, The step of performing cross-node collaborative calibration processing on the radio frequency signal units in the multi-source spectrum sensing signal set to generate a collaborative sensing signal set with a unified spectrum reference includes: Extract the spectral reference parameters of the radio frequency signal units corresponding to each sensing node in the multi-source spectrum sensing signal set. The spectral reference parameters include the signal frequency deviation value and the signal phase offset. Calculate the deviation coefficients between the spectral reference parameters of different sensing nodes and establish the spectral deviation matrix between nodes; Based on the inter-node spectrum deviation matrix, the master calibration node and slave calibration node are determined. The master calibration node is the set of radio frequency signal units corresponding to the sensing node with the most stable spectrum reference parameters. The spectrum reference parameters of the radio frequency signal unit of the slave calibration node are compared with the spectrum reference parameters of the master calibration node to generate the spectrum calibration coefficients for each slave calibration node. Based on the spectral calibration coefficients, frequency compensation and phase correction are performed on the radio frequency signal units of the slave calibration node to ensure that the radio frequency signal units of the slave calibration node and the master calibration node maintain spectral reference consistency. The calibrated radio frequency signal units of the slave calibration node are merged with the radio frequency signal units of the master calibration node to form a collaborative sensing signal set containing node identifiers. The spectral reference error of each radio frequency signal unit in the collaborative sensing signal set is controlled within a preset range.

3. The intelligent spectrum sensing and positioning method based on neural networks according to claim 1, characterized in that, The process involves inputting the spectral hole features and the channel attenuation features into a trained spectral localization neural network, performing spatiotemporal correlation localization analysis, and outputting an initial set of localization parameters for the radio frequency signal unit, including: The spectral hole feature and the channel attenuation feature are input into the input layer of the spectral localization neural network. The feature mapping function of the input layer is used to convert them into a tensor form that the network can recognize. The feature mapping function adopts a linear transformation method to map the feature parameter values ​​to a preset numerical range. The tensor-form features output from the input layer are input into the spatiotemporal fusion layer of the spectral localization neural network, which includes parallel temporal feature extraction sublayers and spatial feature extraction sublayers. The temporal feature extraction sublayer adopts a recurrent neural network structure to perform sequence modeling on the temporal distribution attributes of spectral hole features and generate temporally correlated feature vectors. The spatial feature extraction sublayer adopts a convolutional neural network structure to extract local features from the spatial distribution attributes of channel attenuation characteristics and generate spatial correlation feature vectors. The time-related feature vector and the spatial-related feature vector are input into the feature spatiotemporal fusion layer's feature splicing module, and then weighted and fused according to feature importance weights to generate a spatiotemporal joint feature vector. The spatiotemporal joint feature vector is input into the positioning output layer of the spectrum positioning neural network, and nonlinear mapping is performed through a multilayer perceptron structure to output an initial positioning parameter set containing signal source distance estimates and azimuth angle estimates.

4. The intelligent spectrum sensing and positioning method based on neural networks according to claim 3, characterized in that, The time feature extraction sublayer employs a recurrent neural network structure to perform sequence modeling of the temporal distribution attributes of spectral hole features, generating time-related feature vectors, including: Determine the number of hidden layer neurons and the time step of the recurrent neural network. The time step corresponds to the sampling interval of the temporal distribution attribute of the spectral hole feature. Arrange the temporal distribution attribute parameters of the spectral hole features in chronological order to form a time series feature matrix, with each time step corresponding to a set of spectral hole feature parameters. The time series feature matrix is ​​input into the input gate of the recurrent neural network. The input gate calculates the feature update weights based on the feature parameters of the current time step and the hidden state of the previous time step. The forgetting weights of historical features are calculated using the forgetting gate of a recurrent neural network, based on the feature parameters of the current time step and the hidden state of the previous time step. The outputs of the input gate and the forget gate are combined to update the cell state of the recurrent neural network, thus obtaining the cell state value at the current time step. By using the output gate of the recurrent neural network, the output weights are calculated based on the cell state and current feature parameters at the current time step, and the hidden state vector at the current time step is generated. Global pooling is performed on the hidden state vectors of all time steps to extract key features in the time dimension and generate time-related feature vectors containing time dependencies.

5. The intelligent spectrum sensing and positioning method based on neural networks according to claim 3, characterized in that, The spatial feature extraction sublayer employs a convolutional neural network structure to extract local features from the spatial distribution attributes of channel attenuation characteristics, generating a spatially correlated feature vector, including: The spatial distribution attribute parameters of the channel attenuation feature are converted into a two-dimensional feature map. The horizontal and vertical axes of the map correspond to the geographical coordinates of the sensing node, and the element values ​​of the map correspond to the channel attenuation feature parameters. The two-dimensional feature map is input into the first convolutional layer of the convolutional neural network, and sliding window convolution is performed through multiple convolutional kernels of different sizes to generate multiple sets of local spatial feature maps. Batch normalization is performed on the multiple sets of local spatial feature maps to adjust the distribution range of feature values; The batch-normalized multiple sets of local spatial feature maps are input into the pooling layer of the convolutional neural network, and max pooling is used for downsampling to retain key features and reduce dimensionality. The feature map output from the pooling layer is input into the second convolutional layer of the convolutional neural network. More convolutional kernels are used to perform deep convolution on the feature map to extract higher-level spatial correlation features. The feature map output by the second convolutional layer is flattened, converting the two-dimensional feature map into a one-dimensional feature vector, which serves as the spatial correlation feature vector.

6. The intelligent spectrum sensing and positioning method based on neural networks according to claim 1, characterized in that, The step of performing multi-node fusion optimization processing on the initial positioning parameter set to generate accurate spectral positioning coordinates of the target area includes: Extract the signal source distance estimate and azimuth estimate corresponding to each sensing node in the initial positioning parameter set; Based on the known geographic coordinates of each sensing node, the estimated distance and azimuth of the signal source for each sensing node are converted into the coordinates of candidate positioning points in three-dimensional space. Calculate the spatial distance between the coordinates of all candidate positioning points and construct the candidate positioning point distance matrix; Density clustering algorithm is used to cluster the coordinates of candidate positioning points, and candidate positioning points whose spatial distance is less than the second preset threshold are grouped into the same cluster. Calculate the geometric mean of the coordinates of all candidate location points in each cluster to obtain the coordinates of the center location point of that cluster; The number of candidate localization points contained in each cluster is counted, and the coordinates of the center localization point of the cluster with the most candidate localization points are used as the initial fused localization coordinates. By combining the signal strength confidence parameters of each sensing node, the preliminary fused positioning coordinates are weighted and corrected. The higher the signal strength confidence parameter, the greater the weight of the corresponding candidate positioning point in the correction. The corrected positioning coordinates are used as the precise spectral positioning coordinates of the target area, and the error range of the precise spectral positioning coordinates is lower than a third preset threshold.

7. The intelligent spectrum sensing and positioning method based on neural networks according to claim 6, characterized in that, The process of converting the signal source distance estimate and azimuth estimate of each sensing node into candidate location point coordinates in three-dimensional space based on the known geographic coordinates of each sensing node includes: Obtain the known geographic coordinates of each sensing node in the three-dimensional coordinate system, including longitude, latitude and altitude values; The azimuth estimate of each sensing node is decomposed into horizontal azimuth and vertical azimuth; Based on trigonometric relationships, the estimated distance to the signal source, the horizontal azimuth angle, and the vertical azimuth angle are converted into three-dimensional coordinate offsets relative to the sensing node, including offsets in the east-west, north-south, and vertical directions. The three-dimensional coordinate offset is superimposed with the known geographic coordinates of the sensing node to obtain the absolute coordinate value of the signal source relative to the three-dimensional coordinate system. The absolute coordinate values ​​are normalized to a coordinate system and converted into coordinate values ​​in a preset target coordinate system, which are then used as candidate positioning point coordinates. Record the sensor node identifier and signal acquisition timestamp corresponding to the coordinates of each candidate positioning point.

8. The intelligent spectrum sensing and positioning method based on neural networks according to claim 6, characterized in that, The step of using density clustering algorithm to cluster candidate location point coordinates, grouping candidate location points with a spatial distance less than a second preset threshold into the same cluster, includes: Set the neighborhood radius parameter and the minimum number of points to be included parameter for the density clustering algorithm. The neighborhood radius parameter is determined based on the distribution density of the candidate location point coordinates, and the minimum number of points to be included parameter is determined based on the number of sensing nodes. Calculate the number of other candidate points within the neighborhood radius of each candidate point's coordinates, and mark the candidate points whose number is greater than or equal to the minimum number of contained points as core points; Candidate locations whose spatial distance from the core point is less than the neighborhood radius are assigned to the cluster where the core point is located, forming the initial cluster. For candidate locations that are not marked as core points and do not belong to any initial cluster, determine whether their spatial distance to a core point in an initial cluster is less than the neighborhood radius. If so, they are assigned to that initial cluster. Repeat the above steps until all candidate localization points are assigned to the corresponding cluster or marked as noise points; Clusters containing fewer candidate locations than the minimum number of locations are removed, and the remaining clusters are retained as the final result.

9. A neural network-based intelligent spectrum sensing and positioning system, characterized in that, The method includes a processor and a computer-readable storage medium storing machine-executable instructions, which, when executed by a computer, implement the neural network-based intelligent spectrum sensing and positioning method according to any one of claims 1-8.

Images