Narrowband frequency selective monitoring method, apparatus, device, medium and program product

Abstract

The application relates to a narrow-band frequency selection monitoring method, device, equipment, medium and program product. The method comprises the following steps: performing spectrum analysis and candidate frequency point evaluation processing on a narrow-band spectrum scanning signal of a target region, dynamically screening and determining a target frequency point; acquiring spectrum identification information of the target frequency point, narrow-band time sequence monitoring data of the target frequency point and type identification information corresponding to multiple signal types; calculating a confidence degree corresponding to each signal type of the target frequency point according to frequency point description characteristics, spectrum quality characteristics and type identification information characteristics; determining a target frequency point type of the target frequency point according to the confidence degree; and performing a corresponding monitoring treatment strategy according to the target frequency point type and / or the spectrum quality characteristics. The method can adapt to the communication stability requirement in a complex electromagnetic environment.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a narrowband frequency selective monitoring method, apparatus, device, medium, and program product. Background Technology

[0002] With the development of narrowband communication technology, the intensive use of spectrum resources has led to a complex spectrum environment. Traditional narrowband frequency selection monitoring methods rely on manual experience or simple rules, which have problems such as single spectrum analysis dimension, low accuracy of signal type identification, and rigid monitoring strategies, making it difficult to meet the communication stability requirements in complex electromagnetic environments. Summary of the Invention

[0003] Therefore, it is necessary to provide a narrowband frequency selective monitoring method, device, equipment, medium, and program product that can adapt to the communication stability requirements under complex electromagnetic environments, addressing the aforementioned technical problems.

[0004] Firstly, this application provides a narrowband frequency-selective monitoring method. The method includes:

[0005] Perform spectrum analysis and candidate frequency point evaluation on the narrowband spectrum scanning signal of the target area, and dynamically screen and determine the target frequency point;

[0006] Acquire the spectrum identification information of the target frequency point, the narrowband timing monitoring data of the target frequency point, and the type identification information corresponding to multiple signal types;

[0007] The first encoder performs feature extraction processing on the spectrum identification information to obtain frequency point description features;

[0008] The spectral quality features are obtained by performing feature extraction processing on the narrowband timing monitoring data based on the second encoder.

[0009] The type identification information is extracted by the third encoder to obtain the type identification information features. The first encoder, the second encoder and the third encoder are obtained by joint training based on the same sample data set.

[0010] Calculate the confidence level of the target frequency point and each signal type based on the frequency point description features, the spectral quality features, and the type identification information features;

[0011] The target frequency type of the target frequency point is determined based on the confidence level;

[0012] Based on the target frequency type and / or the spectrum quality characteristics, execute the corresponding monitoring and handling strategy.

[0013] In some embodiments of the method, the second encoder includes a first network branch and a second network branch, and the step of performing feature extraction processing on the narrowband timing monitoring data according to the second encoder to obtain spectral quality features includes:

[0014] The monitoring index data in the narrowband time-series monitoring data is obtained, and the monitoring index data is expanded in feature space according to the first mapping matrix and the first offset vector to obtain the monitoring index feature vector.

[0015] The inherent identifier of the frequency point in the narrowband timing monitoring data is obtained, and the inherent identifier of the frequency point is encoded to obtain the encoded monitoring index data.

[0016] Based on the second mapping matrix and the second offset vector, the encoded monitoring index data is expanded in feature space to obtain an inherent identifier feature vector.

[0017] The deep feature representation is obtained by combining the feature vector of the monitoring indicator with the feature vector of the inherent identifier;

[0018] Based on the second network branch, the deep feature representation is reconstructed to obtain the spectral quality features.

[0019] In some embodiments of the method, calculating the confidence level of the target frequency point corresponding to each signal type based on the frequency point description features, the spectral quality features, and the type identification information features includes:

[0020] The spectrum identification information is parsed to obtain the identification resolution result;

[0021] Based on the identifier resolution result, determine the feature gain value corresponding to the frequency point description feature and the spectral quality feature;

[0022] Based on the characteristic gain value, the frequency point description feature and the spectral quality feature are concatenated to obtain the concatenated feature;

[0023] Calculate the spectral matching degree value between the splicing feature and the type identification information feature;

[0024] The confidence level of the target frequency point corresponding to each signal type is determined based on the spectral matching degree value.

[0025] In some embodiments of the method, calculating the spectral matching degree value between the splicing feature and the type identification information feature includes:

[0026] The splicing features are aligned according to the feature space of the type identification information features to obtain the target splicing features;

[0027] A spectral feature matching operation is performed on the target splicing feature and the type identification information feature to obtain a spectral matching degree value.

[0028] In some embodiments of the method, the step of performing feature extraction processing on the spectrum identification information according to the first encoder to obtain frequency point description features includes:

[0029] If the data dimension of the spectrum identification information is lower than the preset frequency band feature dimension, the spectrum identification information is filled with spectrum according to the preset frequency band feature dimension to obtain the target spectrum information;

[0030] If the data dimension of the spectrum identification information is not lower than the preset frequency band feature dimension, the spectrum identification information is truncated based on the preset frequency band feature dimension to obtain the target spectrum information;

[0031] The first encoder performs feature extraction processing on the target spectrum information to obtain frequency point description features.

[0032] In some embodiments of the method, the monitoring and handling strategy includes:

[0033] If the target frequency point type is a preset normal type and the spectrum quality characteristic indicates that the quality meets the preset conditions, then maintain the current monitoring status or output a frequency point health report.

[0034] If the target frequency point type is a preset interference type or an unknown type, or if the spectrum quality characteristics indicate abnormal quality, an alarm signal is triggered.

[0035] If the target frequency type is a preset avoidance type or the spectrum quality characteristics indicate severe interference, then the adjacent frequency avoidance strategy is triggered and the historical spectrum quality database is updated.

[0036] In some embodiments of the method, before acquiring the spectrum identification information of the target frequency point, the narrowband timing monitoring data of the target frequency point, and the type identification information corresponding to multiple signal types, the method further includes:

[0037] Acquire a sample data set, which includes sample spectrum identification information of sample frequency points, sample narrowband timing monitoring data of sample frequency points, real type target value of sample frequency points, and type identification information corresponding to multiple signal types;

[0038] The first encoder performs feature extraction processing on the sample spectrum identification information to obtain sample frequency point description features;

[0039] The sample narrowband time-series monitoring data is processed by the second encoder to obtain the sample spectral quality features;

[0040] The sample frequency point description features and the sample spectral quality features are concatenated to obtain the sample concatenated features;

[0041] The third encoder performs feature extraction processing on the type identification information to obtain sample type identification features;

[0042] Error information is calculated based on the sample splicing features, the sample type identification features, and the target value of the true type, and the parameters of the first encoder, the second encoder, and the third encoder are optimized based on the error information.

[0043] Repeat the steps of acquiring sample datasets, merging them, and optimizing the parameters of the first encoder, the second encoder, and the third encoder based on the acquired sample datasets, and stop the process after the training of the first encoder, the second encoder, and the third encoder has stabilized.

[0044] In some embodiments of the method, the step of calculating error information based on the sample splicing features, the sample type identification features, and the true type target value, and optimizing the parameters of the first encoder, the second encoder, and the third encoder based on the error information, includes:

[0045] The target sample splicing features are obtained by optimizing the feature space alignment of the sample splicing features according to the fourth encoder.

[0046] Perform spectral feature matching operation on the target sample splicing features and the sample type identification features to obtain the sample predicted spectral type;

[0047] Error information is calculated based on the predicted spectrum type of the sample and the target value of the true type, and the parameters of the first encoder, the second encoder, the third encoder and the fourth encoder are optimized based on the error information;

[0048] The step of repeatedly executing the process of acquiring sample datasets and optimizing the parameters of the first encoder, the second encoder, and the third encoder based on the acquired sample datasets, and stopping after the training of the first encoder, the second encoder, and the third encoder has stabilized, includes:

[0049] Repeat the steps of acquiring sample datasets and optimizing the parameters of the first encoder, second encoder, third encoder, and fourth encoder based on the acquired sample datasets, and stop after the training of the first encoder, second encoder, third encoder, and fourth encoder has stabilized.

[0050] In some embodiments of the method, the step of performing spectral analysis and candidate frequency point evaluation processing on the narrowband spectral scanning signal of the target region, and dynamically filtering and determining the target frequency point, includes:

[0051] Acquire the narrowband spectrum scan signal of the target region;

[0052] The narrowband spectrum scanning signal is subjected to spectrum analysis according to the preset scanning range and step, the potential candidate frequency point set is identified and the noise floor signal characteristics and receiving sensitivity of each candidate frequency point are extracted.

[0053] Based on the pre-set co-channel and adjacent channel interference assessment model and historical spectrum quality database, the receiving sensitivity and potential adjacent channel interference of each candidate frequency point are pre-assessed, and a frequency point priority list is generated.

[0054] Based on the frequency optimization conditions set according to the network quality optimization goals, the target frequency is dynamically selected from the frequency priority list.

[0055] According to a second aspect of the present disclosure, a narrowband frequency selective monitoring device is provided. The device includes:

[0056] The first module is used to perform spectrum analysis and candidate frequency point evaluation on the narrowband spectrum scanning signal of the target area, and dynamically filter and determine the target frequency point.

[0057] The second module is used to acquire the spectrum identification information of the target frequency point, the narrowband timing monitoring data of the target frequency point, and the type identification information corresponding to multiple signal types;

[0058] The third module is used to perform feature extraction processing on the spectrum identification information based on the first encoder to obtain frequency point description features;

[0059] The fourth module is used to perform feature extraction processing on the narrowband timing monitoring data based on the second encoder to obtain spectral quality features;

[0060] The fifth module is used to perform feature extraction processing on the type identification information according to the third encoder to obtain type identification information features. The first encoder, the second encoder, and the third encoder are obtained based on the same sample data set through joint training.

[0061] The sixth module is used to calculate the confidence level of the target frequency point corresponding to each signal type based on the frequency point description features, the spectral quality features, and the type identification information features;

[0062] The seventh module is used to determine the target frequency type of the target frequency point based on the confidence level;

[0063] The eighth module is used to execute corresponding monitoring and handling strategies based on the target frequency type and / or the spectrum quality characteristics.

[0064] According to a third aspect of the present disclosure, a computer device is provided. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the narrowband frequency selective monitoring method described above.

[0065] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the narrowband frequency selective monitoring method described above.

[0066] According to a fifth aspect of the present disclosure, a computer program product is provided. The computer program product includes a computer program that, when executed by a processor, implements the narrowband frequency selective monitoring method described above.

[0067] The narrowband frequency selection monitoring scheme provided in this application can realize intelligent monitoring and management of narrowband spectrum. It can extract multi-dimensional features through three jointly trained encoders and achieve accurate signal type identification by combining confidence calculation, providing an efficient and reliable technical solution for spectrum management in different scenarios. By integrating functional modules such as spectrum analysis, feature extraction, type identification, and policy execution, a complete narrowband frequency selection monitoring system is constructed, which can be widely used in spectrum management scenarios of various narrowband communication networks.

[0068] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0069] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0070] Figure 1 This is a flowchart illustrating a narrowband frequency selective monitoring method according to an exemplary embodiment;

[0071] Figure 2 This is a schematic flowchart illustrating a narrowband frequency selective monitoring method according to an exemplary embodiment;

[0072] Figure 3 This is a structural block diagram of a narrowband frequency selective monitoring device according to an exemplary embodiment;

[0073] Figure 4 This is a diagram illustrating the internal structure of a computer device according to an exemplary embodiment. Detailed Implementation

[0074] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0075] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure. The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitations, the presence of other identical or equivalent elements in a process, method, product, or apparatus that includes said elements is not excluded. For example, the use of terms such as "first," "second," etc., to denote names does not indicate any specific order.

[0076] In some embodiments provided in this disclosure, the execution of the narrowband frequency selective monitoring method can be controlled by a unified controller or by multiple controllers. These controllers may include controllers on local terminals or controllers on remote servers. In some embodiments, the controllers on local terminals and servers may work together to complete the narrowband frequency selective monitoring control processing. The local terminal mentioned in this disclosure may include, but is not limited to, various robotic devices, vehicle-mounted devices, personal computers, laptops, smartphones, tablets, wearable devices, medical devices, VR (Virtual Reality) devices, etc. The server may also be a server, server cluster, distributed subsystem, cloud processing platform, server containing blockchain nodes, or a combination thereof. The controllers described in this disclosure may include various control units capable of implementing logic processing functions, including but not limited to CPU (Central Processing Unit), PLC (Programmable Logic Controller), ECU (Electronic Control Unit), MCU (Microcontroller Unit), FPGA (Field Programmable Gate Array), and CPLD (Complex Programmable Logic Device), as well as controllers composed of one or more logic function units, chips, etc.

[0077] In some embodiments of this disclosure, a narrowband frequency-selective monitoring method is provided, such as... Figure 1 As shown, the method includes:

[0078] S20. Perform spectrum analysis and candidate frequency point evaluation on the narrowband spectrum scanning signal of the target area, and dynamically screen and determine the target frequency point.

[0079] The server first receives narrowband spectrum scanning signals through M spectrum monitoring nodes deployed in the target area (distributed according to regional coverage requirements). The scanning range covers multiple pre-defined licensed frequency band sets {B1, B2, ..., Bk}, with the scanning step set to 1 / 10 of the system's minimum bandwidth (ensuring the spectral resolution meets signal recognition requirements). The server performs spectral analysis on the original scanning signal: converting the time-domain signal into frequency-domain data using a Fast Fourier Transform, setting a noise floor threshold T0 (determined based on regional environmental noise statistics), and identifying a set of potential candidate frequency points {C1, C2, ..., Cm} whose signal energy exceeds T0. Preliminary analysis identifies m candidate frequency points, each containing basic parameters such as frequency band range and center frequency.

[0080] Next, the server extracts the noise floor signal characteristics (including statistical features such as mean noise floor and variance of fluctuation) and receiver sensitivity index of each candidate frequency point, and performs a pre-evaluation based on a preset co-channel and adjacent-channel interference assessment model and a historical spectrum quality database. The co-channel and adjacent-channel interference assessment model integrates the probability distribution of various interference events recorded in historical data, and the historical spectrum quality database stores data such as interference records and quality index change trends of each frequency point within a preset period. The server calculates the interference risk score of each candidate frequency point through the model (with a value range of [0,1], where a lower value indicates a lower interference risk), and generates a frequency point priority list in conjunction with the receiver sensitivity index.

[0081] Subsequently, the server dynamically filters target frequencies from the frequency priority list based on the frequency optimization conditions set for the network quality optimization goals (including frequency band priority, interference risk threshold, and receiver sensitivity threshold). For example, if the current optimization goal is to prioritize frequencies with the highest frequency band priority, an interference risk score below threshold Th, and a receiver sensitivity better than threshold Ts, the server iterates through the frequency priority list and selects the frequency that simultaneously meets the above conditions and has the highest priority as the target frequency F.

[0082] S21. Obtain the spectrum identification information of the target frequency point, the narrowband timing monitoring data of the target frequency point, and the type identification information corresponding to multiple signal types.

[0083] In some implementations, after determining the target frequency F, the server can obtain spectrum identification information, narrowband timing monitoring data, and type identification information corresponding to multiple signal types through a multi-source data interface. Specifically, for the spectrum identification information, the server sends a target frequency query request to the spectrum monitoring node via a standard communication protocol to obtain the physical and protocol identification information of the frequency, including but not limited to: unique frequency identifier, frequency band range, modulation scheme, frequency band bandwidth, center frequency, signal bandwidth, and other structured parameters. For the narrowband timing monitoring data, the server retrieves the timing monitoring data of the target frequency within a preset monitoring period from the edge computing node, including but not limited to: signal strength, signal-to-noise ratio, bit error rate, frame error rate, and other key indicators. The sampling period for the timing data is set according to the monitoring accuracy requirements, typically 1 minute / time, and the monitoring period is typically 24 hours, forming multi-dimensional time series data. For the type identification information, the server loads a preset signal type set {T1, T2, ..., Tn} from a local configuration file, where each signal type contains a unique type identifier and corresponding feature description information. Preset signal types typically include: normal type (for legitimate communication signals), interference type (for various types of interference signals), unknown type (for signals that cannot match preset characteristics), and avoidance type (for signals that have long-term severe interference).

[0084] S22. Perform feature extraction processing on the spectrum identification information according to the first encoder to obtain frequency point description features.

[0085] The server first performs structured parsing on the spectrum identification information, converting the text-based identification information into structured data. The parsing process includes extracting key dimensions such as frequency point type, frequency band parameters, and modulation parameters to form an initial feature vector.

[0086] S23. Perform feature extraction processing on the narrowband timing monitoring data according to the second encoder to obtain spectral quality features.

[0087] Spectral quality characteristics are key indicators used to evaluate the performance of a signal in the frequency domain and are widely used in fields such as communication, audio, and vibration analysis.

[0088] S24. Perform feature extraction processing on the type identification information according to the third encoder to obtain type identification information features. The first encoder, the second encoder and the third encoder are obtained based on the same sample data set through joint training.

[0089] After the server loads the preset signal type identification information, it can call the third encoder to perform feature extraction processing on the identification information of various signal types. The third encoder is a text feature extraction network, specifically adopting an embedding layer plus a convolutional neural network structure, which can convert textual type description information into high-dimensional feature vectors.

[0090] In some implementations, the server can convert the identification information (including type identifiers and feature description text) for each signal type into a text sequence, and then convert the text sequence into a word vector sequence using a pre-trained word embedding model. Subsequently, the word vector sequence is input into the convolutional neural network of a third encoder, which extracts key features from the text description through convolution operations. After processing by pooling and fully connected layers, the output is a type identifier feature vector with dimension F3.

[0091] The first, second, and third encoders are acquired through joint training using the same set of sample data, ensuring that the feature vectors output by the three encoders reside in the same feature space and are comparable and fusionable. During joint training, by sharing sample data and collaboratively optimizing objectives, the three encoders learn a consistent feature representation.

[0092] S25. Calculate the confidence level of the target frequency point and each signal type based on the frequency point description features, the spectrum quality features, and the type identification information features.

[0093] The confidence level of the target frequency point and each signal type can be calculated based on the frequency point description features, spectral quality features and type identification information features obtained above.

[0094] S26. Determine the target frequency type of the target frequency point based on the confidence level.

[0095] After obtaining the confidence values ​​of the target frequency and each signal type, the server can determine the target frequency type using a maximum value selection strategy. Specifically, the confidence values ​​of all signal types are compared, and the signal type with the highest confidence value is selected as the final target frequency type. If multiple signal types have the same and highest confidence value, a secondary selection is performed based on a preset type priority rule to ensure the uniqueness of the target frequency type. In some examples, the confidence values ​​corresponding to the signal type set {T1, T2, T3, T4} can be set as {0.92, 0.05, 0.02, 0.01}, in which case the server selects T1 as the target frequency type; if the confidence value is {0.45, 0.45, 0.05, 0.05}, then T1 is selected as the target frequency type according to the preset type priority (e.g., T1>T2>T3>T4).

[0096] S27. Execute the corresponding monitoring and handling strategy according to the target frequency type and / or the spectrum quality characteristics.

[0097] The server executes a pre-defined monitoring and handling strategy based on the determined target frequency type and extracted spectrum quality characteristics. This strategy is a set of rules pre-configured according to network management needs, defining corresponding actions for different signal types and spectrum quality conditions. It mainly includes the following types of strategies: normal state strategies, alarm strategies, and avoidance strategies. For the normal state strategy, if the target frequency type is a pre-defined normal type and the spectrum quality characteristics indicate that the quality meets the standards (all quality indicators are within the pre-defined threshold range), the server maintains the current monitoring state and outputs a frequency health report periodically. The health report includes information such as frequency identifier, type, quality indicator statistics, and potential risk warnings, is stored in the historical spectrum quality database, and pushed to the network management platform for maintenance personnel to view.

[0098] In some embodiments of this disclosure, intelligent monitoring and management of narrowband spectrum can be achieved. Multi-dimensional features can be extracted through three jointly trained encoders, and accurate signal type identification is achieved by combining confidence level calculations. This provides an efficient and reliable technical solution for spectrum management in different scenarios. By integrating functional modules such as spectrum analysis, feature extraction, type identification, and policy execution, a complete narrowband frequency selection monitoring system is constructed, which can be widely applied to spectrum management scenarios in various narrowband communication networks.

[0099] In some embodiments of this disclosure, S22 includes:

[0100] If the data dimension of the spectrum identification information is lower than the preset frequency band feature dimension, the spectrum identification information is filled with spectrum according to the preset frequency band feature dimension to obtain the target spectrum information;

[0101] If the data dimension of the spectrum identification information is not lower than the preset frequency band feature dimension, the spectrum identification information is truncated based on the preset frequency band feature dimension to obtain the target spectrum information;

[0102] The first encoder performs feature extraction processing on the target spectrum information to obtain frequency point description features.

[0103] In some examples, the initial feature vector can be adjusted according to the input dimension requirements of the first encoder (preset frequency band feature dimension D1): if the initial dimension is lower than D1, relevant extended features (such as frequency band usage, standard protocols, etc.) are added according to preset rules; if the initial dimension is higher than D1, secondary dimensions are truncated based on feature importance ranking, and core dimensions are retained, finally obtaining target spectrum information with dimension D1.

[0104] Subsequently, the server invokes the first encoder to perform feature extraction processing on the target spectrum information. The first encoder is a deep learning-based feature extraction network, specifically employing a multi-layer fully connected neural network structure. The input layer has a dimension of D1, contains multiple hidden layers, and the output layer has a dimension of F1 (a preset feature dimension). The activation function is ReLU. This encoder is jointly trained based on a sample dataset, which contains a large amount of historical frequency point identification information and their corresponding type labels. The server inputs the numerical vector of the target spectrum information into the first encoder, and through forward propagation, calculates a frequency point description feature vector of dimension F1. This vector contains abstract representations of the frequency point's physical characteristics, protocol features, etc.

[0105] In some embodiments of this disclosure, the second encoder includes a first network branch and a second network branch, as shown in the reference. Figure 2 S23 includes:

[0106] S231. Obtain monitoring index data from the narrowband time-series monitoring data, and expand the feature space of the monitoring index data according to the first mapping matrix and the first offset vector to obtain the monitoring index feature vector.

[0107] S232. Obtain the inherent frequency identifier in the narrowband timing monitoring data, and encode the inherent frequency identifier to obtain the encoded monitoring index data.

[0108] S233. Expand the feature space of the encoded monitoring index data according to the second mapping matrix and the second offset vector to obtain the inherent identifier feature vector.

[0109] S234. Combine the feature vector of the monitoring indicator with the feature vector of the inherent identifier to obtain a deep feature representation;

[0110] S235. Based on the second network branch, the deep feature representation is reconstructed to obtain the spectral quality features.

[0111] In some implementations, the second encoder includes a first network branch (processing monitoring index data) and a second network branch (processing frequency point inherent identifiers) for extracting spectral quality features from time-series data.

[0112] First, the server processes the monitoring indicator data through the first network branch: it concatenates multi-dimensional time-series monitoring data (including time series of indicators such as signal strength and signal-to-noise ratio) into a matrix form, and expands the feature space using a first mapping matrix and a first offset vector, mapping the low-dimensional time-series data to a high-dimensional feature space to obtain the monitoring indicator feature vector. This process can capture the dynamic change characteristics of the time-series data, such as the periodic fluctuations in signal strength and the abrupt changes in signal-to-noise ratio.

[0113] Secondly, the server processes the inherent identifiers of the frequency point through the second network branch: the inherent identifiers of the frequency point include the hardware attributes and environmental parameters of the frequency point (such as antenna parameters, transmit power, transmission loss, etc.). The server encodes these inherent identifiers and converts them into numerical vectors. Then, it expands the feature space through the second mapping matrix and the second offset vector to obtain the inherent identifier feature vector. This vector reflects the impact of the physical hardware characteristics of the frequency point on the spectrum quality.

[0114] The server combines the monitoring indicator feature vector with the inherent identifier feature vector to form a deep feature representation. This deep feature representation integrates temporal dynamic features and physical attribute features, with a dimension of D². Subsequently, the server inputs this deep feature representation into the feature reconstruction module of the second network branch (containing a Long Short-Term Memory network and a fully connected network). The Long Short-Term Memory network is used to capture long-term dependencies in the temporal data, while the fully connected network is used for feature dimensionality reduction and fusion. The final output is a spectral quality feature vector with a dimension of F², which contains abstract features related to the quality of the spectral signal, such as stability, noise level, and interference probability.

[0115] In some examples, after the server determines the target frequency F, it acquires its narrowband timing monitoring data (a 24-hour timing sequence including signal strength, signal-to-noise ratio, bit error rate, and frame error rate, with a sampling period of 1 minute and a total of 1440 data points) and the inherent identifiers of the frequency point (antenna gain 2dBi, transmit power 20dBm, feeder loss 1.5dB, and vertical polarization). The first network branch of the second encoder processes the monitoring index data: it concatenates the four timing sequences into a 4×1440 matrix, and expands the feature space through a preset first mapping matrix (4×64 dimensions, training parameters) and a first offset vector (64 dimensions), mapping the low-dimensional timing data into a 64×1440 monitoring index feature vector to capture timing features such as periodic fluctuations in signal strength and dynamic changes in signal-to-noise ratio. The second network branch processes the inherent identifiers of frequency points: The inherent identifiers are encoded (antenna gain normalized to 0.2, transmit power normalized to 0.5, feeder loss normalized to 0.15, polarization encoded to 1), and expanded into a 32-dimensional inherent identifier feature vector through the second mapping matrix (4×32-dimensional, training parameters) and the second offset vector (32-dimensional), reflecting the impact of hardware attributes on spectrum quality. The server concatenates the monitoring indicator feature vector (64×1440) with the inherent identifier feature vector (32-dimensional) by channel to obtain a 96×1440 deep feature representation, fusing temporal dynamic features and physical attribute features, laying the foundation for subsequent spectrum quality feature extraction.

[0116] In some implementations, when processing narrowband timing monitoring data for target frequency F, the server first retrieves the inherent identifier of that frequency from the device configuration database. This identifier includes: antenna gain of 2dBi (hardware parameter, characterizing the antenna signal amplification capability), transmit power of 20dBm (output power of the device's transmitting end), feeder loss of 1.5dB (line loss during signal transmission), and polarization mode (vertical, the direction of antenna signal polarization). Next, the server encodes the inherent identifier, converting the physical parameters into numerical vectors. The specific encoding rules are as follows: antenna gain is normalized to its maximum value of 10dBi, with 2dBi corresponding to an encoded value of 0.2 (2 / 10); transmit power is normalized to its maximum value of 40dBm, with 20dBm corresponding to an encoded value of 0.5 (20 / 40); feeder loss is normalized to its maximum value of 10dB, with 1.5dB corresponding to an encoded value of 0.15 (1.5 / 10); and polarization mode uses binary encoding, with vertical polarization encoded as 1 (horizontal polarization encoded as 0). After encoding, 4-dimensional encoded monitoring index data is obtained: [0.2, 0.5, 0.15, 1]. Subsequently, the server uses the second mapping matrix and the second offset vector to expand the feature space of the encoded monitoring index data. The second mapping matrix is ​​a 4×32-dimensional training parameter matrix (obtained through joint training with sample data and stored in the server's model parameter library), and the second offset vector is a 32-dimensional bias parameter (also obtained through training). The server multiplies the 4-dimensional encoded data with the second mapping matrix (0.2×1st row of the matrix + 0.5×2nd row of the matrix + 0.15×3rd row of the matrix + 1×4th row of the matrix), then superimposes the second offset vector, and processes it through the ReLU activation function to obtain a 32-dimensional intrinsic identifier feature vector. This vector, for example, is [0.02, 0.05, 0.015, 0.1, 0.03, ..., 0.08] (partial element values), which contains abstract representations related to hardware attributes such as the influence of antenna gain on signal coverage and the contribution of transmit power to signal strength, providing a foundation for subsequent deep feature fusion.

[0117] In some embodiments of this disclosure, S25 includes:

[0118] The spectrum identification information is parsed to obtain the identification resolution result;

[0119] Based on the identifier resolution result, determine the feature gain value corresponding to the frequency point description feature and the spectral quality feature;

[0120] Based on the characteristic gain value, the frequency point description feature and the spectral quality feature are concatenated to obtain the concatenated feature;

[0121] Calculate the spectral matching degree value between the splicing feature and the type identification information feature;

[0122] The confidence level of the target frequency point corresponding to each signal type is determined based on the spectral matching degree value.

[0123] In some implementations, the spectrum identification information is first parsed to obtain the identification resolution result. The identification resolution process includes verifying indicators such as frequency band compliance, modulation scheme matching degree, and frequency band utilization rate of the frequency point. Based on these indicators, the feature gain value G (range [0,1]) corresponding to the frequency point description features and spectrum quality features is determined. This value reflects the reliability and effectiveness of the current frequency point identification information.

[0124] Secondly, the frequency description feature and the spectral quality feature are concatenated based on the feature gain value G. Specifically, the frequency description feature vector and the spectral quality feature vector are summed element-wise with weights (the weights are the feature gain value G) to obtain a weighted feature vector. Then, the weighted feature vector is concatenated with the frequency description feature vector and the spectral quality feature vector respectively to form a concatenated feature vector with dimensions F1+F2+F1.

[0125] Next, the spectral matching degree value between the spliced ​​features and the type identification information features is calculated. Since the spliced ​​features and the type identification information features may have dimensionality differences, the server first performs feature alignment on the spliced ​​features according to the feature space of the type identification information features: the spliced ​​features are reduced to the same dimension as the type identification information features through a linear mapping layer to obtain the target spliced ​​features.

[0126] In some examples, after the server obtains the frequency description features (a 32-dimensional vector, such as [0.12, 0.35, 0.08, ..., 0.21], containing abstract features such as bandwidth and modulation type) and spectral quality features (a 32-dimensional vector, such as [0.12, 0.85, 0.03, ..., 0.18], containing quality features such as signal stability and noise level) of the target frequency point F, it performs feature concatenation: the 32-dimensional frequency description features and the 32-dimensional spectral quality features are concatenated in channel order to obtain a 64-dimensional concatenated feature vector, for example, [0.12, 0.35, ..., 0.21, 0.12, 0.85, ..., 0.18] (the first 32 dimensions are the frequency description features, and the last 32 dimensions are the spectral quality features). Next, the server calculates the spectral matching degree value between the concatenated features and the features of each type of identification information. The type identification information features are 32-dimensional feature vectors corresponding to preset signal types (T1: normal type, T2: interference type, T3: unknown type, T4: avoidance type), for example, the feature vector of T1 is [0.05, 0.82, 0.03, ..., 0.11]. Since the splicing feature (64 dimensions) and the type feature (32 dimensions) have different dimensions, the server reduces the dimensionality of the splicing feature to the target 32-dimensional splicing feature through a linear mapping layer (64→32 dimensions, training parameter matrix), for example, [0.15, 0.78, 0.06, ..., 0.22]. Subsequently, the server performs cosine similarity calculation (spectral matching value) on the target splicing feature and each type feature. For example, the dot product of the target splicing feature and the T1 feature vector is 28.5, and the product of their magnitudes is 31.0, resulting in a matching degree of 0.92. The matching degree with the T2 feature vector (e.g., [0.78, 0.04, 0.15, ..., 0.02]) is 0.05, and the matching degrees with T3 and T4 are 0.02 and 0.01, respectively. Finally, the server directly uses the spectral matching degree as the confidence level (set to 0 when the matching degree is negative) to determine the confidence level of the target frequency point F with each signal type: 0.92 for T1, 0.05 for T2, 0.02 for T3, and 0.01 for T4.

[0127] Before concatenating the frequency description features (32-dimensional, including bandwidth, modulation type, etc.) and spectral quality features (32-dimensional, including signal stability, noise level, etc.) of the target frequency point F, the server first obtains the feature gain values ​​of both. The server parses the spectral identification information (frequency range, modulation method, etc.) of the target frequency point F to obtain the gain evaluation index of the frequency description features: frequency band compliance (complies with the preset authorized frequency band, score 1.0) and modulation matching degree (GFSK modulation matches the system supported type, score 0.95). The gain value of the frequency description features is calculated as 0.98 (1.0 × 0.6 + 0.95 × 0.4) according to the formula "gain value = 0.6 × compliance + 0.4 × matching degree". Simultaneously, the server analyzes historical data on spectrum quality characteristics: the compliance rate of quality indicators over the past 7 days was 98% (score 0.98), and the variance of indicator fluctuation was <2dB (stability score 0.95). Using the formula "gain value = 0.5 × compliance rate + 0.5 × stability", the gain value of the spectrum quality characteristics is calculated to be 0.965 (0.98 × 0.5 + 0.95 × 0.5). Subsequently, the server weights and concatenates the two features based on the gain value: each element of the frequency point description feature vector (e.g., [0.12, 0.35, ... 0.21]) is multiplied by 0.98 to obtain the weighted frequency point features (e.g., [0.1176, 0.343, ... 0.2058]); each element of the spectrum quality feature vector (e.g., [0.12, 0.85, ... 0.18]) is multiplied by 0.965 to obtain the weighted quality features (e.g., [0.1158, 0.8203, ... 0.1737]). Finally, the weighted frequency features (32-dimensional) and the weighted quality features (32-dimensional) are concatenated in sequence to obtain a 64-dimensional concatenated feature vector, which is used for subsequent spectrum matching degree calculation.

[0128] After obtaining the spectrum identification information of the target frequency point F (including the unique frequency point identifier "F-001", frequency band range "B1-B2", modulation scheme "M1", frequency band bandwidth "W1", center frequency "C1", and signal bandwidth "S1"), the server performs identifier resolution and extracts key evaluation indicators as the identifier resolution results. First, the server resolves the indicators related to the frequency point description characteristics: Frequency band compliance: Check whether the frequency band range "B1-B2" is within the system's preset list of authorized frequency bands (e.g., the set of authorized frequency bands is {B1-B2, B3-B4}). If "B1-B2" is confirmed to be an authorized frequency band, the compliance score is set to 1.0 (out of 1.0, unauthorized frequency bands score 0.0); Modulation matching degree: Verify whether the modulation scheme "M1" is a modulation type supported by the system (the system supports modulation sets of {M1, M2}). If "M1" is confirmed to be within the support list, the matching degree score is set to 0.95 (complete match 1.0, partial match 0.5, no match 0.0). Next, the server analyzes the spectrum quality characteristics related indicators: Frequency band utilization: Query the historical spectrum database to obtain the average device access rate of the target frequency point "F-001" over the past 30 days, which is 0.75 (number of access devices / maximum capacity), and the utilization rate score is set to 0.75 (out of 1.0); Historical stability: Extract the quality indicator fluctuation data of this frequency point over the past 30 days, calculate the noise floor fluctuation variance as 2.3dB (threshold ≤5dB is stable), and set the stability score to 0.9 (fluctuation variance ≤2dB score 1.0, 2-5dB score 0.9, >5dB score 0.5). Based on the above identifier resolution results, the server determines the feature gain values ​​for the frequency point description feature and the spectrum quality feature: Frequency point description feature gain value: calculated using the weighting rule of "frequency band compliance × 0.6 + modulation matching degree × 0.4", i.e., 1.0 × 0.6 + 0.95 × 0.4 = 0.6 + 0.38 = 0.98; Spectrum quality feature gain value: calculated using the weighting rule of "frequency band utilization × 0.5 + historical stability × 0.5", i.e., 0.75 × 0.5 + 0.9 × 0.5 = 0.375 + 0.45 = 0.825. Finally, the server stores the frequency point description feature gain value of 0.98 and the spectrum quality feature gain value of 0.825 in memory for subsequent weighted concatenation processing of the two features.

[0129] In some embodiments of this disclosure, calculating the spectral matching degree value between the splicing feature and the type identification information feature includes:

[0130] The splicing features are aligned according to the feature space of the type identification information features to obtain the target splicing features;

[0131] A spectral feature matching operation is performed on the target splicing feature and the type identification information feature to obtain a spectral matching degree value.

[0132] In some implementations, spectral feature matching operations can be performed on the target splicing features and the features of various types of identification information, using cosine similarity as the matching degree index. The calculation formula can be referred to as the following formula (1):

[0133]

[0134] In equation (1), To construct feature vectors for the target, For type identification information feature vectors, Let be the cosine similarity between the two, with a value in the range [-1, 1].

[0135] Finally, the confidence level of the target frequency point corresponding to each signal type is determined based on the spectral matching degree value. The spectral matching degree value is normalized (if the matching degree is negative, the confidence level is set to 0) to obtain the confidence level value of each signal type, with a value range of [0,1]. The higher the value, the greater the probability that the target frequency point belongs to that signal type.

[0136] In some implementations, after the server obtains the 64-dimensional splicing features of the target frequency point F (obtained by weighted splicing of 32-dimensional frequency point description features and 32-dimensional spectral quality features, such as [0.1176, 0.343, ..., 0.2058, 0.1158, 0.8203, ..., 0.1737]), it first performs feature alignment on the splicing features according to the feature space of the type identification information features. The type identification information features are 32-dimensional feature vectors corresponding to preset signal types (T1-T4) (stored in the server model library). For example, the feature vector of normal type T1 is [0.05, 0.82, 0.03, ..., 0.11] (32-dimensional). Because the splicing features (64-dimensional) and type features (32-dimensional) have inconsistent dimensions, the server calls a preset linear mapping layer (a 4×32-dimensional parameter matrix, jointly trained using sample data) to perform dimensionality reduction and alignment on the splicing features: multiplying the 64-dimensional splicing features by the mapping matrix (each row of elements corresponds to a weighted sum), and then superimposing a 32-dimensional bias vector (training parameters) to obtain the 32-dimensional target splicing features, for example [0.15, 0.78, 0.06, 0.21, ..., 0.22]. Subsequently, the server performs spectral feature matching operations on the target splicing features and the type identification information features, using cosine similarity as the spectral matching degree value. Taking the normal type T1 as an example, the server loads the 32-dimensional feature vector [0.05, 0.82, 0.03, ..., 0.11] of T1 from memory, concatenates it with the target feature [0.15, 0.78, 0.06, ..., 0.22], and performs a vector dot product operation: (0.15 × 0.05) + (0.78 × 0.82) + (0.06 × 0.03) + ... + (0.22 × 0.11) = 0.0075 + 0.6396 + 0.0018 + ... + 0.0242 = 28.5. Simultaneously, the modulus of the target splicing feature (the square root of the sum of the squares of each element) is calculated as √(0.15²+0.78²+...+0.22²)=√30.8≈5.55, and the modulus of the T1 feature vector is √(0.05²+0.82²+...+0.11²)=√5.5≈2.35. The product of the modulus lengths is 5.55×2.35≈13.04. The final spectral matching degree value is 28.5 / 13.04≈0.92, meaning the matching degree between the target splicing feature and the T1 type feature is 0.92. The server repeats the above calculation for the feature vectors of interference type T2, unknown type T3, and evasion type T4, obtaining matching degree values ​​of 0.05, 0.02, and 0.01 respectively, completing the calculation of the spectral matching degree value.

[0137] In some embodiments of this disclosure, prior to S21, the method further includes:

[0138] Acquire a sample data set, which includes sample spectrum identification information of sample frequency points, sample narrowband timing monitoring data of sample frequency points, real type target value of sample frequency points, and type identification information corresponding to multiple signal types;

[0139] The first encoder performs feature extraction processing on the sample spectrum identification information to obtain sample frequency point description features;

[0140] The sample narrowband time-series monitoring data is processed by the second encoder to obtain the sample spectral quality features;

[0141] The sample frequency point description features and the sample spectral quality features are concatenated to obtain the sample concatenated features;

[0142] The third encoder performs feature extraction processing on the type identification information to obtain sample type identification features;

[0143] Error information is calculated based on the sample splicing features, the sample type identification features, and the target value of the true type, and the parameters of the first encoder, the second encoder, and the third encoder are optimized based on the error information.

[0144] Repeat the steps of acquiring sample datasets, merging them, and optimizing the parameters of the first encoder, the second encoder, and the third encoder based on the acquired sample datasets, and stop the process after the training of the first encoder, the second encoder, and the third encoder has stabilized.

[0145] In some examples, the server needs to complete the joint training of the first encoder, second encoder, and third encoder before performing narrowband frequency selective monitoring. First, the server obtains a sample data set from the historical spectrum database, containing 5000 sets of sample frequency point data. Each sample set includes: sample spectrum identification information: such as "Frequency Point ID: S-001, Frequency Band Range: B1-B2, Modulation Method: M1, Bandwidth: 1MHz"; sample narrowband time-series monitoring data: monitoring indicators (signal strength, signal-to-noise ratio, bit error rate, etc.) for the corresponding sample frequency point over the past 24 hours; real type target value: manually labeled signal type (such as "normal type" or "interference type"); type identification information: text descriptions of preset signal type sets (normal, interference, unknown, avoidance type). The server calls the first encoder to perform feature extraction on the sample spectrum identification information: parsing text information such as "frequency band range B1-B2" and "modulation mode M1" into 10-dimensional structured data (after spectrum padding / truncation normalization), inputting it into the first encoder (3-layer fully connected network), and outputting 32-dimensional sample frequency point description features, for example, [feature vector of S-001: 0.12, 0.35, ..., 0.21]. At the same time, the server processes the sample narrowband time-series monitoring data through the second encoder: expanding the signal strength time-series sequence (such as [-75, -73, ..., -78]) and the inherent frequency point identifier (antenna gain 2dBi, etc.) into a 96×1440 deep feature representation through the first and second network branches, and then outputting 32-dimensional sample spectrum quality features through LSTM and fully connected layers, for example, [quality features of S-001: 0.12, 0.85, ..., 0.18]. The server concatenates the sample frequency description features (32-dimensional) and sample spectral quality features (32-dimensional) into a 64-dimensional sample concatenated feature. It then calls the third encoder to perform text feature extraction on the type identification information (e.g., "Normal type: stable communication, SNR > 25dB"), outputting a 32-dimensional sample type identification feature (e.g., normal type features: 0.05, 0.82, ..., 0.11). Subsequently, the server calculates the error information: it reduces the 64-dimensional sample concatenated feature to a 32-dimensional target concatenated feature using a linear mapping layer, calculates the cosine similarity (prediction confidence) with the sample type identification feature, and then calculates the error (e.g., initial error value 2.3) with the true type target value (e.g., "Normal type" corresponding to label 1) using the cross-entropy loss function. Based on the error information, the server uses the Adam optimizer (learning rate 0.001) for backpropagation to update the parameters of the first, second, and third encoders (e.g., adjusting the weights of the fully connected layers and the biases of the LSTM hidden layers). Repeat the above steps: input 200 samples per batch, iterate for 50 rounds, and when the validation set error decreases to <0.01 for 3 consecutive rounds (e.g., stabilizes after decreasing from 2.3 to 0.08), determine that the training is stable, stop optimization and save the encoder parameters to the database.

[0146] In some embodiments of this disclosure, the step of calculating error information based on the sample splicing features, the sample type identification features, and the true type target value, and optimizing the parameters of the first encoder, the second encoder, and the third encoder based on the error information, includes:

[0147] The target sample splicing features are obtained by optimizing the feature space alignment of the sample splicing features according to the fourth encoder.

[0148] Perform spectral feature matching operation on the target sample splicing features and the sample type identification features to obtain the sample predicted spectral type;

[0149] Error information is calculated based on the predicted spectrum type of the sample and the target value of the true type, and the parameters of the first encoder, the second encoder, the third encoder and the fourth encoder are optimized based on the error information;

[0150] The step of repeatedly executing the process of acquiring sample datasets and optimizing the parameters of the first encoder, the second encoder, and the third encoder based on the acquired sample datasets, and stopping after the training of the first encoder, the second encoder, and the third encoder has stabilized, includes:

[0151] Repeat the steps of acquiring sample datasets and optimizing the parameters of the first encoder, second encoder, third encoder, and fourth encoder based on the acquired sample datasets, and stop after the training of the first encoder, second encoder, third encoder, and fourth encoder has stabilized.

[0152] In some examples, the server introduces a fourth encoder to optimize feature space alignment during the joint training of the first, second, and third encoders. First, the server selects a set of typical samples from the sample dataset (such as sample frequency point "S-123", with the target value of the true type being "normal type"). The concatenated feature of these samples is a 64-dimensional vector: [0.12, 0.35, 0.08, ..., 0.21, 0.12, 0.85, ..., 0.18] (the first 32 dimensions are the sample frequency point description features, and the last 32 dimensions are the sample spectral quality features).

[0153] The server invokes the fourth encoder to optimize the feature space alignment of the sample splicing features. The fourth encoder is a two-layer fully connected neural network (64 neurons in the input layer, 48 neurons in the hidden layer, 32 neurons in the output layer, and ReLU activation function), with parameters initialized through pre-training using sample data. The server inputs the 64-dimensional sample splicing features into the fourth encoder, which, after weighted summation and ReLU activation in the hidden layer (48 neurons), outputs 32-dimensional target sample splicing features: [0.15, 0.78, 0.06, 0.21, ..., 0.22] (consistent with the dimension of the sample type identifier feature, facilitating spatial matching).

[0154] Next, the server performs spectral feature matching operations on the target sample splicing features and sample type identifier features. The sample type identifier features are 32-dimensional vectors output by the third encoder (e.g., the feature vector of "normal type" T1: [0.05, 0.82, 0.03, 0.18, ..., 0.11]). The server uses the softmax function to convert the cosine similarity between the target sample splicing features and each type feature into a probability distribution: the cosine similarity between the target sample splicing features and T1 is 0.92, with T2 (interference type) it is 0.05, with T3 (unknown type) it is 0.02, and with T4 (avoidance type) it is 0.01. After softmax calculation, the probability distribution is [0.92, 0.05, 0.02, 0.01]. The T1 with the highest probability is selected as the sample predicted spectral type.

[0155] Subsequently, the server calculates error information based on the predicted spectrum type (T1) and the true type target value (T1) of the sample. The cross-entropy loss function is used, with the formula "Loss=-y_true・log(y_pred)", where y_true is the one-hot encoding of the true type target value ([1,0,0,0]) and y_pred is the predicted probability distribution ([0.92,0.05,0.02,0.01]). The calculated Loss is -(1×log0.92+0×log0.05+...)≈0.083.

[0156] The server optimizes the parameters of the first, second, third, and fourth encoders using a backpropagation algorithm: error information is propagated backward from the loss function layer to each encoder, adjusting network weights (such as the weights of the fully connected layer in the first encoder, the weights of the LSTM layer in the second encoder, the convolutional kernel parameters in the third encoder, and the hidden layer biases in the fourth encoder). The optimizer uses Adam with a learning rate of 0.001 and a batch size of 64, and the parameters are updated after each iteration.

[0157] The server repeatedly executes the training process: 200 samples are randomly selected from the sample dataset in each round, and the parameters are optimized after alignment, spectrum matching, and error calculation by the fourth encoder. During training, the validation set loss (decreasing from the initial 1.2 to 0.08) and accuracy (increasing from 65% to 96%) are monitored in real time. When the validation set loss fluctuation is <0.001 for 5 consecutive rounds (e.g., the losses in rounds 45-49 are 0.081, 0.080, 0.082, 0.081, and 0.080 respectively) and the accuracy is stable above 95%, the training of the four encoders is considered stable, the server stops training, and saves the parameters to the database for subsequent feature extraction and type recognition in narrowband frequency selection monitoring.

[0158] In some embodiments of this disclosure, the step of performing spectrum analysis and candidate frequency point evaluation processing on the narrowband spectrum scanning signal of the target region, and dynamically filtering and determining the target frequency point, includes:

[0159] Acquire the narrowband spectrum scan signal of the target region;

[0160] The narrowband spectrum scanning signal is subjected to spectrum analysis according to the preset scanning range and step, the potential candidate frequency point set is identified and the noise floor signal characteristics and receiving sensitivity of each candidate frequency point are extracted.

[0161] Based on the pre-set co-channel and adjacent channel interference assessment model and historical spectrum quality database, the receiving sensitivity and potential adjacent channel interference of each candidate frequency point are pre-assessed, and a frequency point priority list is generated.

[0162] Based on the frequency optimization conditions set according to the network quality optimization goals, the target frequency is dynamically selected from the frequency priority list.

[0163] In some examples, the server targets a smart manufacturing park (with 500 narrowband devices operating in frequency bands B1 to B4) and performs narrowband spectrum scanning signal acquisition and target frequency selection. First, the server receives the narrowband spectrum scanning signal through four distributed spectrum monitoring nodes (node ​​IDs: N1 to N4, deployed at the four corners of the park). These monitoring nodes use RTL-SDR devices with a sampling rate of 2MHz and a scan cycle of 5 minutes per scan. The raw I / Q signals are transmitted back to the server via wired Ethernet (TCP / IP protocol, port 5000) in JSON format, with each signal approximately 20MB in size, and are stored in real-time in the server's local cache. Next, the server invokes the spectrum analysis module to perform an FFT transformation (8192 points) on the raw signal according to a preset scan range (B1: 860-870MHz, B2: 910-920MHz, 100kHz step), converting it into frequency domain power spectral density data. A noise floor threshold of -100dBm (based on the campus environment noise baseline) was set to identify frequency bands with signal energy exceeding the threshold, resulting in 12 potential candidate frequency points, such as "F01 (frequency band 867.5-868.5MHz, center frequency 868MHz, signal bandwidth 250kHz)" and "F02 (frequency band 914.2-915.2MHz, center frequency 914.7MHz, signal bandwidth 250kHz)". For each candidate frequency point, the server extracted noise floor signal characteristics (e.g., the noise floor fluctuation variance of F01 is 2.1dB, and the noise floor fluctuation variance of F02 is 2.5dB, both satisfying the threshold ≤5dB) and receiver sensitivity (the receiver sensitivity of F01 is -120dBm, and the receiver sensitivity of F02 is -118dBm, determined by test values ​​at BER=1e-5). Subsequently, the server pre-evaluated the candidate frequency points based on a preset co-channel and adjacent-channel interference assessment model and a historical spectrum quality database (stored interference events from the past 3 months). The co-channel interference assessment model integrates historical interference probabilities (0.03 for B1 band, 0.05 for B2 band) and real-time parameters. For example, the interference risk score for F01 is calculated as follows: 0.4 × historical interference probability (0.03) + 0.6 × (receiver sensitivity / interference intensity) = 0.4 × 0.03 + 0.6 × (-120 / -80) = 0.012 + 0.9 = 0.912 (the lower the score, the lower the interference risk). The interference risk score for F02 is 0.935 (historical interference probability 0.05, receive sensitivity -118dBm, interference intensity -75dBm). Based on the receive sensitivity index, the server generates a frequency priority list, where F01 has priority 1, F02 has priority 2, and other candidate frequencies are ranked sequentially. Finally, the server dynamically selects the target frequency from the frequency priority list based on the frequency optimization conditions set for the network quality optimization goals (current goals: "prioritize the B1 band, interference risk score < 1.0, and receiver sensitivity ≤ -115dBm").F01 belongs to the B1 band, with an interference risk score of 0.912 < 1.0, a receiver sensitivity of -120dBm ≤ -115dBm, and the highest priority. Therefore, the server determines F01 as the target frequency point and writes its frequency band range, center frequency, and other parameters into the system configuration file for subsequent feature extraction and monitoring analysis.

[0164] In some embodiments of this disclosure, the monitoring and handling strategy includes:

[0165] If the target frequency point type is a preset normal type and the spectrum quality characteristic indicates that the quality meets the preset conditions, then maintain the current monitoring status or output a frequency point health report.

[0166] If the target frequency point type is a preset interference type or an unknown type, or if the spectrum quality characteristics indicate abnormal quality, an alarm signal is triggered.

[0167] If the target frequency type is a preset avoidance type or the spectrum quality characteristics indicate severe interference, then the adjacent frequency avoidance strategy is triggered and the historical spectrum quality database is updated.

[0168] In some implementations, the server executes a preset monitoring and handling strategy based on the determined target frequency type and extracted spectrum quality characteristics. The monitoring and handling strategy is a set of rules pre-configured according to network management needs, defining corresponding actions for different signal types and spectrum quality conditions. It mainly includes the following types of strategies: normal state strategies, alarm strategies, and avoidance strategies. For the normal state strategy, if the target frequency type is a preset normal type and the spectrum quality characteristics indicate that the quality meets the standards (all quality indicators are within preset threshold ranges), the server maintains the current monitoring state and outputs a frequency health report periodically. The health report includes information such as frequency identifier, type, quality indicator statistics, and potential risk warnings, is stored in the historical spectrum quality database, and pushed to the network management platform for maintenance personnel to view.

[0169] For alarm policies, if the target frequency point is a preset interference type or an unknown type, or if the spectrum quality characteristics indicate abnormal quality (at least one quality indicator exceeds the preset threshold range), the server triggers an alarm signal. The alarm signal includes the alarm level (graded according to the degree of abnormality), frequency point information, abnormal indicators, and suggested handling measures, and is sent to the network management platform through a standard interface. It can also notify relevant responsible persons via SMS, email, or other means.

[0170] For the avoidance strategy, if the target frequency is of a preset avoidance type, or if the spectrum quality characteristics indicate severe interference (key quality indicators significantly exceed the threshold range and the duration exceeds the preset duration), the server triggers the adjacent frequency avoidance strategy. Specifically, this includes: reselecting a new frequency that meets the conditions from the candidate frequency list as the target frequency, updating the system's current operating frequency configuration, and recording the severe interference events and avoidance results of the original frequency to the historical spectrum quality database to provide data support for subsequent spectrum planning.

[0171] In some examples, the server can execute monitoring and handling strategies based on the target frequency type and spectrum quality characteristics. When the target frequency F01 is of the preset normal type (T1), and the spectrum quality characteristics indicate that the quality meets the standards (the signal stability characteristic value 0.12≤0.2 and the noise level characteristic value 0.85≥0.8 in the spectrum quality characteristic vector both meet the preset thresholds), the server maintains the current monitoring status and generates a frequency health report. The report includes the frequency ID (F01), type (normal type), quality indicators (mean signal strength -74dBm, mean signal-to-noise ratio 29dB, bit error rate <1e-5), and potential risk warnings. The report is stored in the historical spectrum quality database and pushed to the network operation and maintenance platform visualization interface.

[0172] If the target frequency F01 is of the preset interference type (T2) or an unknown type (T3), or if the spectrum quality characteristics indicate abnormal quality (such as noise level characteristic value 0.6 < 0.8, exceeding the threshold), the server triggers an alarm signal: it sends alarm information to the operation and maintenance platform via the SNMP protocol, including alarm level (level 2, medium risk), frequency information (F01, 867.5-868.5MHz), abnormal indicators (noise level 0.6, below the threshold 0.8), and suggested handling measures (manual verification of the interference source), and at the same time sends an alarm email to the responsible person via the email interface.

[0173] When the target frequency F01 is of the preset avoidance type (T4), or the spectrum quality characteristics indicate severe interference (burst interference probability characteristic value 0.8 > 0.05, and duration exceeding 1 hour), the server triggers the adjacent channel avoidance strategy: selects the next candidate frequency F02 (priority 2, interference risk score 0.935) from the frequency priority list as the new target frequency, updates the current working frequency configuration of the system (active_freq=F02), and records the severe interference event of F01 (interference intensity -70dBm, lasting 90 minutes) and the avoidance result (switching to F02) to the historical spectrum quality database for subsequent spectrum planning analysis.

[0174] The narrowband frequency selection monitoring methods disclosed herein enable intelligent monitoring and management of narrowband spectrum. They extract multi-dimensional features through three jointly trained encoders and, combined with confidence level calculations, achieve accurate signal type identification, providing an efficient and reliable technical solution for spectrum management in various scenarios. By integrating functional modules such as spectrum analysis, feature extraction, type identification, and policy execution, a complete narrowband frequency selection monitoring system is constructed, which can be widely applied to spectrum management scenarios in various narrowband communication networks.

[0175] It is understood that the various embodiments of the methods described in this specification are presented in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. Related details can be found in the descriptions of other method embodiments.

[0176] It should be understood that although the steps in the flowcharts shown in the accompanying drawings are displayed sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the accompanying drawings may include multiple steps or stages, which are not necessarily completed at the same time, but may be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but may be performed alternately or in turn with other steps or at least a portion of the steps or stages of other steps.

[0177] Based on the description of the narrowband frequency selective monitoring method embodiments described above, this disclosure also provides a narrowband frequency selective monitoring device for implementing the narrowband frequency selective monitoring method involved above. The device may include a system (including a distributed system), software (application), module, component, controller, server, terminal, etc., using the method described in the embodiments of this specification, combined with necessary implementation hardware. Based on the same innovative concept, the devices in one or more embodiments provided by the embodiments of this disclosure are as described in the following embodiments. Since the implementation schemes and methods for solving the problem by the devices are similar, the implementation of specific devices in the embodiments of this specification can refer to the implementation of the foregoing method, and repeated details will not be repeated. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0178] Figure 3This is a schematic block diagram of a narrowband frequency selective monitoring device according to an exemplary embodiment. The device can be the aforementioned terminal, a server, or a module, component, device, control unit, etc., integrated into the terminal. For details, please refer to... Figure 3 The device 100 may include: a first module, a second module, a third module, a fourth module, a fifth module, a sixth module, a seventh module, and an eighth module. The system comprises the following modules: a first module for performing spectrum analysis and candidate frequency point evaluation on the narrowband spectrum scanning signal of the target area, dynamically filtering and determining the target frequency point; a second module for acquiring the spectrum identification information of the target frequency point, the narrowband timing monitoring data of the target frequency point, and the type identification information corresponding to multiple signal types; a third module for performing feature extraction processing on the spectrum identification information using a first encoder to obtain frequency point description features; a fourth module for performing feature extraction processing on the narrowband timing monitoring data using a second encoder to obtain spectrum quality features; a fifth module for performing feature extraction processing on the type identification information using a third encoder to obtain type identification information features, wherein the first encoder, the second encoder, and the third encoder are obtained through joint training based on the same sample data set; a sixth module for calculating the confidence level of the target frequency point corresponding to each signal type based on the frequency point description features, the spectrum quality features, and the type identification information features; a seventh module for determining the target frequency point type based on the confidence level; and an eighth module for executing corresponding monitoring and handling strategies based on the target frequency point type and / or the spectrum quality features.

[0179] Each module in the aforementioned narrowband frequency selective monitoring device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0180] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a narrowband frequency selective monitoring method.

[0181] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0182] Based on the foregoing description of the relevant methods and apparatus embodiments, this disclosure also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the narrowband frequency selective monitoring method described in any embodiment of this specification.

[0183] Based on the foregoing description of the relevant methods and apparatus embodiments, this disclosure also provides a computer-readable storage medium that, when the instructions in the computer-readable storage medium are executed by the processor of a computer device, enables the computer device to implement the narrowband frequency selective monitoring method as described in any embodiment of this disclosure.

[0184] Based on the foregoing description of the relevant methods and apparatus embodiments, this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the narrowband frequency selective monitoring method described in any embodiment of this specification.

[0185] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, hardware + program embodiments are relatively simple in description because they are fundamentally similar to method embodiments; relevant parts can be referred to the descriptions in the method embodiments.

[0186] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0187] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0188] It should be noted that the apparatus, computer equipment, storage medium, and computer program products described above may also include other implementation methods according to the description of the method embodiments. Specific implementation methods can be found in the description of the relevant method embodiments. Furthermore, new embodiments formed by combinations of features from various methods, apparatuses, devices, and server embodiments still fall within the scope of this disclosure and will not be elaborated upon here.

[0189] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, when implementing one or more of these specifications, the functions of each module can be implemented in the same or different software and / or hardware, or a module that performs the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are merely illustrative. For example, the division of modules or units is only a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling and communication connections between the devices or units shown or described can be implemented through direct and / or indirect coupling / connection, through standard or custom interfaces or protocols, and can be implemented electrically, mechanically, or in other forms.

[0190] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0191] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A narrowband frequency-selective monitoring method, characterized in that, The method includes: Perform spectrum analysis and candidate frequency point evaluation on the narrowband spectrum scanning signal of the target area, and dynamically screen and determine the target frequency point; Acquire the spectrum identification information of the target frequency point, the narrowband timing monitoring data of the target frequency point, and the type identification information corresponding to multiple signal types; The first encoder performs feature extraction processing on the spectrum identification information to obtain frequency point description features; The spectral quality features are obtained by performing feature extraction processing on the narrowband timing monitoring data based on the second encoder. The type identification information is extracted by the third encoder to obtain the type identification information features. The first encoder, the second encoder and the third encoder are obtained by joint training based on the same sample data set. Calculate the confidence level of the target frequency point and each signal type based on the frequency point description features, the spectral quality features, and the type identification information features; The target frequency type of the target frequency point is determined based on the confidence level; Based on the target frequency type and / or the spectrum quality characteristics, execute the corresponding monitoring and handling strategy.

2. The method according to claim 1, characterized in that, The second encoder includes a first network branch and a second network branch. The step of performing feature extraction processing on the narrowband time-series monitoring data based on the second encoder to obtain spectral quality features includes: The monitoring index data in the narrowband time-series monitoring data is obtained, and the monitoring index data is expanded in feature space according to the first mapping matrix and the first offset vector to obtain the monitoring index feature vector. The inherent identifier of the frequency point in the narrowband timing monitoring data is obtained, and the inherent identifier of the frequency point is encoded to obtain the encoded monitoring index data. Based on the second mapping matrix and the second offset vector, the encoded monitoring index data is expanded in feature space to obtain an inherent identifier feature vector. The deep feature representation is obtained by combining the feature vector of the monitoring indicator with the feature vector of the inherent identifier; Based on the second network branch, the deep feature representation is reconstructed to obtain the spectral quality features.

3. The method according to claim 1, characterized in that, The step of calculating the confidence level of the target frequency point corresponding to each signal type based on the frequency point description features, the spectral quality features, and the type identification information features includes: The spectrum identification information is parsed to obtain the identification resolution result; Based on the identifier resolution result, determine the feature gain value corresponding to the frequency point description feature and the spectral quality feature; Based on the characteristic gain value, the frequency point description feature and the spectral quality feature are concatenated to obtain the concatenated feature; Calculate the spectral matching degree value between the splicing feature and the type identification information feature; The confidence level of the target frequency point corresponding to each signal type is determined based on the spectral matching degree value.

4. The method according to claim 3, characterized in that, The calculation of the spectral matching degree value between the splicing feature and the type identification information feature includes: The splicing features are aligned according to the feature space of the type identification information features to obtain the target splicing features; A spectral feature matching operation is performed on the target splicing feature and the type identification information feature to obtain a spectral matching degree value.

5. The method according to claim 1, characterized in that, The step of performing feature extraction processing on the spectrum identification information according to the first encoder to obtain frequency point description features includes: If the data dimension of the spectrum identification information is lower than the preset frequency band feature dimension, the spectrum identification information is filled with spectrum according to the preset frequency band feature dimension to obtain the target spectrum information; If the data dimension of the spectrum identification information is not lower than the preset frequency band feature dimension, the spectrum identification information is truncated based on the preset frequency band feature dimension to obtain the target spectrum information; The first encoder performs feature extraction processing on the target spectrum information to obtain frequency point description features.

6. The method according to claim 1, characterized in that, The monitoring and response strategies include: If the target frequency point type is a preset normal type and the spectrum quality characteristic indicates that the quality meets the preset conditions, then maintain the current monitoring status or output a frequency point health report. If the target frequency point type is a preset interference type or an unknown type, or if the spectrum quality characteristics indicate abnormal quality, an alarm signal is triggered. If the target frequency type is a preset avoidance type or the spectrum quality characteristics indicate severe interference, then the adjacent frequency avoidance strategy is triggered and the historical spectrum quality database is updated.

7. A narrowband frequency-selective monitoring device, characterized in that, The device includes: The first module is used to perform spectrum analysis and candidate frequency point evaluation on the narrowband spectrum scanning signal of the target area, and dynamically filter and determine the target frequency point. The second module is used to acquire the spectrum identification information of the target frequency point, the narrowband timing monitoring data of the target frequency point, and the type identification information corresponding to multiple signal types; The third module is used to perform feature extraction processing on the spectrum identification information based on the first encoder to obtain frequency point description features; The fourth module is used to perform feature extraction processing on the narrowband timing monitoring data based on the second encoder to obtain spectral quality features; The fifth module is used to perform feature extraction processing on the type identification information according to the third encoder to obtain type identification information features. The first encoder, the second encoder, and the third encoder are obtained based on the same sample data set through joint training. The sixth module is used to calculate the confidence level of the target frequency point corresponding to each signal type based on the frequency point description features, the spectral quality features, and the type identification information features; The seventh module is used to determine the target frequency type of the target frequency point based on the confidence level; The eighth module is used to execute corresponding monitoring and handling strategies based on the target frequency type and / or the spectrum quality characteristics.

8. A computer device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, It stores a computer program thereon, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

Images