Collaborative multi-layer identification and node selection method for spectral aliasing wireless signals

Abstract

The present invention discloses a collaborative multi-layer identification and node selection method for spectrum aliasing wireless signals, comprising the following steps: decomposing the result of the aliasing signal after short-time Fourier transform to obtain a basis matrix for signal decomposition and a representation matrix of the aliasing spectrum signal type, and constructing a cascade matrix with the latter and an indicator factor; performing HL test on the elements in different cascade matrices to obtain the optimal anchor node group, and then splitting them by column to obtain several coefficient matrices; performing multi-perspective information fusion on each coefficient matrix to determine the signal category; decomposing each coefficient matrix into a distance representation matrix of the monitoring node, a signal power and position coupling matrix, and then decomposing the latter into a power coefficient matrix for power estimation and a radian information matrix for positioning. The present invention realizes the identification of the category, position and power estimation of the aliasing spectrum signal in sequence by designing a three-layer deep decomposition structure, effectively reducing the computational complexity and accelerating the iterative convergence speed.

Description

Technical Field

[0001] The present invention relates to the field of radio spectrum monitoring, and in particular to a collaborative multi-layer recognition and node selection method for spectrum-aliased wireless signals. Background Art

[0002] With the rapid development of wireless communication technology, spectrum resources are becoming increasingly scarce. To effectively manage and utilize spectrum resources, comprehensive monitoring and control of the electromagnetic environment is necessary. Organizations such as the International Telecommunication Union (ITU) are promoting the development of spectrum monitoring technology through the development of standards and reports. Spectrum monitoring technology has evolved from simple signal detection to complex signal separation and multi-mode positioning. Modern spectrum monitoring technology not only detects weak signals but also separates co-frequency signals and combines multiple positioning techniques for precise positioning. Radio communication systems are continuously and rapidly evolving, exemplified by software-defined radio and cognitive radio systems. This requires future spectrum monitoring systems to have monitoring capabilities for a variety of emerging radio communication technologies and systems.

[0003] To improve the coverage and accuracy of spectrum monitoring, distributed monitoring networks are widely deployed. These networks consist of multiple monitoring nodes, which can be fixed, mobile, or portable. These nodes are connected to a central node via wireless communication modules, enabling real-time data transmission and processing. In distributed spectrum monitoring, data collected from different monitoring points must be fused to generate regional electromagnetic field intensity distribution data. Feature identification and location results for the same radiation source must also be fused to form a feature vector for that source. This source identification is then performed based on a feature library. This involves the fusion of spectrum data in four domains: time, space, frequency, and energy, achieved through techniques such as structured data tables and tensor decomposition algorithms. In TDOA (Time Difference of Arrival) positioning, the spatial geometry of monitoring nodes significantly impacts the spatial resolution and location accuracy of the signal source. Therefore, optimizing the spatial placement of monitoring nodes is an effective means of improving location accuracy. Furthermore, due to limitations in sensor characteristics, research is needed to investigate sensor selection techniques to select the optimal sensor subset for signal source location. To improve location accuracy and reduce computational complexity, existing research must consider node deployment and selection optimization algorithms that account for signal source uncertainty and measurement error.

[0004] In terms of blind source separation, non-negative matrix factorization (NMF) was proposed as an important method by Lee and Seung in "Learning the parts of objects by non-negative matrix factorization" and published in "Nature", but NMF requires that the number of monitoring nodes is greater than or equal to the number of source signals. Professor Zhou Mu's team from Chongqing University of Posts and Telecommunications studied the use of information between different nodes to improve the accuracy of device-to-device cooperative positioning and select appropriate anchor nodes for position conversion in "Device-to-device cooperative positioning via matrix completion and anchor selection". In terms of power estimation, Liu Zhaoting's team from Hangzhou Dianzi University studied the use of measurements on a subset of nodes to study the power estimation of random signals on all sensor network nodes in "Signal Power Estimation of All Sensor Network Nodes With Measurements From a Subset of Network Nodes". Summary of the Invention

[0005] The purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and to provide a collaborative multi-layer identification and node selection method for spectrum-aliased wireless signals. Existing studies regard signal detection, separation, and reconstruction as independent tasks, and lack a unified analysis framework to integrate these tasks. The present invention aims to construct a collaborative multi-layer signal recognition framework that can simultaneously handle signal detection, separation, and parameter identification to improve the efficiency and accuracy of spectrum monitoring. At the same time, in order to obtain optimal performance, different signal sources may need to select different spectrum monitoring nodes. Traditional algorithms usually regard all information under a single node as a single entity, thereby ignoring the unique needs of each signal source. In order to solve the problem of selecting different sets of monitoring nodes for different identification parameters, the present invention aims to develop an optimization algorithm to intelligently select and deploy monitoring nodes to ensure that spectrum signals with different parameters can be effectively monitored and identified.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A collaborative multi-layer identification and node selection method for spectrum-aliased wireless signals includes the following steps:

[0008] Step 1: Perform NMF decomposition on the short-time Fourier transform result of the aliased signal received by the monitoring node to obtain the basis matrix of signal decomposition and the representation matrix of the aliased spectrum signal type;

[0009] Step 2: Construct a cascade matrix based on the characterization matrix of the aliased spectrum signal type and the indicator factor;

[0010] Step 3: Perform HL test on the elements in different cascade matrices to obtain the optimal anchor node group;

[0011] Step 4: Split the optimal anchor node group by columns to obtain several coefficient matrices;

[0012] Step 5: Use the DempsterShafer theory to perform multi-perspective information fusion on each coefficient matrix to achieve the first layer of collaborative perception to determine the category of the spectrally aliased wireless signal;

[0013] Step 6: Decompose each coefficient matrix NMF into the distance representation matrix, signal power and position coupling matrix of the monitoring node;

[0014] Step 7: The signal power and position coupling matrix is further decomposed by NMF into a power coefficient matrix for power estimation and a radian information matrix for positioning.

[0015] In step 3, the HL test is performed according to the following formula, and the combination with the largest HL test calculation value is considered to be the optimal anchor node group :

[0016] ;

[0017] in, Indicates that the When building the anchor node group for the first time, the location signal source Ranking results; ; is the number of anchor nodes, is the number of signal templates in the library.

[0018] In step 6, the distance representation matrix of the monitoring node Sum signal power and position coupling matrix , the particle swarm optimization algorithm is used to obtain the distance representation matrix of the monitoring nodes Optimal initial value and introduction of consensus result matrix , update according to the following formula:

[0019] ;

[0020] ;

[0021] ;

[0022] in, is a hyperparameter, Represents the multiplication operation of the corresponding elements of the matrix, is the matrix transpose operation.

[0023] In step 7, the signal power and position coupling matrix Decomposed into a power coefficient matrix Radian information matrix , introduce the consensus result matrix , update according to the following formula:

[0024] ;

[0025] ;

[0026] ;

[0027] Among them, the power coefficient matrix Represents the power estimate of the spectral aliased wireless signal, radian information matrix Represents the positioning of spectrum-aliased wireless signals, is a hyperparameter.

[0028] At the same time, the present invention provides:

[0029] A server includes a processor and a memory, wherein the memory stores at least one program, and the program is loaded and executed by the processor to implement the above-mentioned collaborative multi-layer identification and node selection method for spectrum-aliased wireless signals.

[0030] A computer-readable storage medium stores at least one program, which is loaded and executed by a processor to implement the above-mentioned collaborative multi-layer identification and node selection method for spectrum-aliased wireless signals.

[0031] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0032] 1. This paper proposes a CMSR architecture based on the multi-layer NMF theory. By designing a three-layer deep decomposition structure, the architecture sequentially realizes the category, location and power estimation of the aliased spectrum signal, effectively reducing the computational complexity and accelerating the iterative convergence speed.

[0033] 2. For different signal sources, the present invention selects different monitoring node sets, and the independence between signal sources is guaranteed.

[0034] 3. In terms of signal classification and recognition, the present invention designs a fusion strategy based on the Dempster-Shafer (DS) rule to fuse information from multiple monitoring points and assign weights.

[0035] 4. The present invention avoids the problem of decreased computational efficiency caused by increased matrix operations in Deep-NMF, reduces computational complexity and accelerates iterative convergence speed.

[0036] 5. For different parameter identification tasks and different signal sources, the selection method of the monitoring node of the present invention is different, ensuring the unique requirements of each task.

[0037] 6. In the case of high noise, increasing the number of monitoring nodes will affect the overall performance. The present invention reduces the number of nodes, which not only reduces the computational complexity but also improves the system performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Flowchart of a collaborative multi-layer identification and node selection method for spectrally aliased wireless signals.

[0039] Figure 2 Schematic diagram of the structure of the spectrum aliasing signal monitoring model.

[0040] Figure 3 Schematic diagram of node deployment.

[0041] Figure 4 Schematic diagram of the node selection process.

[0042] Figure 5 The comparison chart of the error cumulative distribution function of positioning error under different monitoring ranges.

[0043] Figure 6 Comparison diagram of the error cumulative distribution function of power estimation error under different monitoring ranges. DETAILED DESCRIPTION

[0044] The present invention will be described in further detail below with reference to the embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0045] like Figure 1 , a collaborative multi-layer identification and node selection method for spectrum-aliased wireless signals, comprising the following steps:

[0046] Step 1: Perform NMF decomposition on the short-time Fourier transform result of the aliased signal received by the monitoring node to obtain the basis matrix of signal decomposition and the representation matrix of the aliased spectrum signal type;

[0047] Step 2: Construct a cascade matrix based on the characterization matrix of the aliased spectrum signal type and the indicator factor;

[0048] Step 3: Perform HL test on the elements in different cascade matrices to obtain the optimal anchor node group;

[0049] Step 4: Split the optimal anchor node group by columns to obtain several coefficient matrices;

[0050] Step 5: Use the DempsterShafer (DS) theory to perform multi-perspective information fusion on each coefficient matrix to achieve the first layer of collaborative perception to determine the category of the spectrally aliased wireless signal;

[0051] Step 6: Decompose each coefficient matrix NMF into the distance representation matrix, signal power and position coupling matrix of the monitoring node;

[0052] Step 7: The signal power and position coupling matrix is further decomposed by NMF into a power coefficient matrix for power estimation and a radian information matrix for positioning.

[0053] Node selection is reflected in step 3. The optimal node combination is obtained through the HL test and applied to signal recognition, positioning, and power estimation. Therefore, node selection serves the purpose of multi-layer recognition.

[0054] Furthermore, the collaborative multi-layer identification and node selection method for spectrum-aliased wireless signals includes the following steps:

[0055] S1. First, Perform NMF decomposition: ,in, is the result of short-time Fourier transform of the aliased signal received by the monitoring node. is the basis matrix of signal decomposition, is the characterization matrix of the aliased spectrum signal type;

[0056] S2. Initialize the matrix , update according to the following formula :

[0057] ;

[0058] S3. Define indicator factors ,use and Constructed cascade matrix;

[0059] S4. Perform HL test according to the following formula to improve the accuracy and efficiency of the anchor node selection process. The combination with the largest HL test calculation value is considered to be the optimal anchor node group. ;

[0060] ;

[0061] in, Indicates that the When building the anchor node group for the first time, the location signal source Ranking results; ; is the number of anchor nodes, is the number of signal templates in the library; ; Node deployment diagram as shown Figure 3 , the node selection process diagram is as follows Figure 4 .

[0062] The combination with the largest HL test calculation value can be considered as a group of anchor nodes with the weakest correlation. Node combinations with weak correlation can obtain more comprehensive information, which is beneficial for subsequent multi-layer recognition.

[0063] S5. Group the best anchor nodes Split by column to get coefficient matrix ; Taking the spectrum detection of each individual monitoring node as a perspective, the DempsterShafer (DS) theory is used to calculate the coefficient matrix of each Perform multi-perspective information fusion to achieve first-layer collaborative perception to determine the category of spectrally mixed wireless signals;

[0064] S6. The coefficient matrix Decomposed into the distance representation matrix of monitoring nodes Sum signal power and position coupling matrix ; Use particle swarm optimization algorithm (PSO) to get the distance representation matrix of monitoring nodes Optimal initial value and introduction of consensus result matrix , update according to the following formula:

[0065] ;

[0066] ;

[0067] ;

[0068] in, is a hyperparameter, Represents the multiplication operation of the corresponding elements of the matrix, is the transpose operation of the matrix;

[0069] S7, signal power and position coupling matrix Decomposed into a power coefficient matrix Radian information matrix , introduce the consensus result matrix , update according to the following formula:

[0070] ;

[0071] ;

[0072] ;

[0073] Among them, the power coefficient matrix Represents the power estimate of the spectral aliased wireless signal, radian information matrix Represents the positioning of spectrum-aliased wireless signals, is a hyperparameter.

[0074] The time it takes for each monitoring node to complete the detection of a frequency band is ,The spectrum data return time of this frequency band is , then the total time to complete a frequency band monitoring task is . Assume that there is An unknown wireless signal source, the monitoring node has no prior information about its signal type, location, transmission power, etc. The monitoring nodes are evenly distributed. The spectrum data generated by the spectrum scanning can be transmitted to the edge computing node by each monitoring node through wired or wireless means. The separation, positioning and power estimation of the aliased signal are completed in the edge computing node. The edge computing node receives the relevant data of each monitoring node. The monitoring node does not judge and process the aliased signal. Instead, the edge computing node performs statistics and analysis on the data to make the final decision. The monitoring network model is as follows: Figure 2 shown.

[0075] No. The monitoring node is The spectrum mixed signal received in a spectrum sampling time slot is:

[0076] ;

[0077] in For the The aliased signal received by each monitoring node is For the The original signal of the transmitting source, For the An unknown signal source in the time slot t The transmission power, For the unknown signal source to the The channel gain of each monitoring node, For the The noise of monitoring nodes and obey This project mainly considers the large-scale fading of wireless propagation in outdoor rich scattering environments, so the channel gain coefficient is defined as:

[0078] ;

[0079] Among them, the channel constant , is the operating center frequency of the unknown signal, is the speed of light, and Respectively monitoring nodes and The antenna gain of an unknown signal. and Respectively monitoring nodes and The channel response and distance between unknown signals are:

[0080] ;

[0081] in and ) Respectively monitoring nodes and An unknown signal source is The coordinates in the coordinate system of the coordinates in the monitoring range, , Therefore, let The polar coordinates of an unknown signal source are ,in , . exist Through the short-time Fourier transform (STFT), a typical time-frequency transform analysis method, the time series is multiplied by a window function to convert the signal into the time-frequency domain. After short-time Fourier transform, , yes The result of the short-time Fourier transform, that is, the unknown signal source The power spectral density of the signal is defined as the standard spectrum dictionary for identifying the signal and ,in It is The spectral feature matrix of the signal, where is the number of sampling frequencies of the monitoring point in this frequency band, is the spectral feature representation dimension and , is the number of signal templates, and .

[0082] In order to reduce the computational complexity in integrated recognition and accelerate the iterative convergence speed, the present invention proposes a collaborative multi-layer signal recognition (CMSR) architecture based on the multi-layer NMF theory.

[0083] Aiming at the problem of identifying the category, location and power estimation of multi-parameter aliased spectrum signals, a deep NMF method is used to extract multiple features from the aliased spectrum signals received by all monitoring nodes. Breaks down into:

[0084] ;

[0085] Using the Matrix As the basis matrix of signal decomposition. Considering the The aliased signal received by the monitoring point, let the coefficient matrix of the aliased signal be , the condition is ,in represents a non-negative value, As a characterization matrix of the aliased spectrum signal type. It is multi-parameter and multi-angle coupled, and it is difficult to reach a consensus directly. Therefore, a new matrix was redesigned ,in, Indicates the Class signal The monitoring points were not detected, and The opposite is true. Defined as: ;

[0086] In the formula is the characteristic index threshold. In the process of signal classification and recognition, signals with too low characteristic index do not exist, and vice versa.

[0087] In addition, let the matrix To indicate the consistent results of signal category recognition. The higher the value, the better The higher the value, the greater the possibility of achieving higher recognition accuracy for each monitoring node, while lower values indicate a lower possibility of accurate recognition.

[0088] Taking the spectrum detection of each individual monitoring node as a perspective lays the foundation for establishing a multi-viewpoint joint NMF decomposition model, as shown below:

[0089] ;

[0090] ;

[0091] Due to the problem In the alternating solution and is convex, respectively and The update rule is obtained by differentiation:

[0092] ;

[0093] Redefine a selection node by Constructed cascade matrix ,in Represents the number of selected nodes. Represents the total number of cascade matrices that can be formed, In order to facilitate the calculation, Cascade matrix Defining indicator factors ,in ,and .therefore It can be expressed as:

[0094] ;

[0095] in It means that the elements that are not equal to 0 are extracted and combined into a new vector. , for each signal The corresponding operations will be performed. Represents the index of the reconstructed node. For different signals, The value of the original index will also be different. The goal is to find the optimal set of monitoring point nodes. and the optimal number of nodes The Hodges–Lehmann (HL) rank sum test is used to compare three or more independent combinations.

[0096] In order to obtain the optimal set and number of nodes, first calculate About the mean of , and decentralized

[0097] ;

[0098] is the adjusted element value. Then calculate the average ranking of each combination. For the Under these combinations, the unknown signal source The ranking results are calculated as follows:

[0099] ;

[0100] in is an indicator function, if , which means the condition is true, if it is 0, it means the condition is false. The HL test statistic obtained is as follows:

[0101] ;

[0102] Through the HL test, the combination of weakly correlated nodes is beneficial for joint positioning. It's about function, so the objective function of finding the optimal node can be written as:

[0103] ;

[0104] Because the above objective function is an integer optimization problem and Usually it is not too large, so you can directly use the search method to solve it. For subsequent processing, we redefine is the coefficient matrix under the selected node, the same For different unknown signal sources in the original index The following values may be different.

[0105] In the subsequent signal recognition process, Normalization is as follows:

[0106] ;

[0107] As the judgment basis of each monitoring node, it is used for integration to reach a consensus; different The values need to be analyzed by combining their information. The DempsterShafer (DS) theory is used to implement the first layer of collaborative perception for determining the signal category.

[0108] Define the existence state set of each unknown signal source , where 1 indicates the target exists and 0 indicates the target does not exist. It is Monitoring node The trust ratio of each URS.

[0109] ;

[0110] The fusion formula for every two monitoring nodes is as follows. The total fusion required is Second-rate:

[0111] ;

[0112] Each element in is defined as follows:

[0113] ;

[0114] in is a hyperparameter, Used to make fusion decisions on signal types;

[0115] For the second layer, define the diagonal matrix As the first The distance representation matrix of monitoring nodes is Indicates that for the signal source The selected monitoring node estimates the The arc of an unknown signal source. It is a A function used to extract the arc features from the aggregated position information. Represents the coupling coefficient matrix of signal power and position information. Then, the matrix Breaking it down like this:

[0116] ;

[0117] The second-layer NMF decomposition model is constructed based on the above formula, and the consensus regularization term is added as follows:

[0118] ;

[0119] ;

[0120] The above optimization problem can be transformed into three optimization sub-problems, 、 and Optimize and fix the other optimization sub-problem at the same time. The iterative formula of the three is as follows:

[0121] ;

[0122] ;

[0123] ;

[0124] In order to speed up model optimization, it is very necessary and important to initialize the parameters of each layer for pre-training.

[0125] Since the basis matrix represents the attributes of each layer, its physical meaning and dimension selection have a great influence on the optimization results. At the same time, a reasonable selection of the basis matrix can also reduce the convergence time of the model. The particle swarm optimization (PSO) algorithm is used as the initialization algorithm for the basis matrix. For a multi-transmitter system, the objective function is:

[0126] ;

[0127] Obtained through optimization Used to get the The monitoring node estimates the The arc of a URS , thus completing the Initialization.

[0128] Further decomposition of the matrix , and obtain the power coefficient matrix , radian information matrix , the third layer can be decomposed into:

[0129] ;

[0130] The third-layer NMF decomposition model is constructed based on the above formula, and the consensus regularization term is added to express it as:

[0131] ;

[0132] ;

[0133] The above optimization problem can also be transformed into three optimization sub-problems, respectively 、 and Optimize and fix the other optimization sub-problem at the same time. The iterative formula of the three is as follows:

[0134] ;

[0135] ;

[0136] .

[0137] like Figure 5 ,The positioning performance of the maximum value test method is better than the ,minimum value test and random selection methods in the monitoring range of ,50m*50m and 100m*100m.

[0138] like Figure 6 ,The power estimation performance of the maximum value test method is better than the ,minimum value test and random selection methods under the ,monitoring range of 50m*50m and 100m*100m.

[0139] At the same time, the present invention provides:

[0140] A server includes a processor and a memory, wherein the memory stores at least one program, and the program is loaded and executed by the processor to implement the above-mentioned collaborative multi-layer identification and node selection method for spectrum-aliased wireless signals.

[0141] A computer-readable storage medium stores at least one program, which is loaded and executed by a processor to implement the above-mentioned collaborative multi-layer identification and node selection method for spectrum-aliased wireless signals.

[0142] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A collaborative multi-layer identification and node selection method for spectrum-aliased wireless signals, characterized in that: The following steps are involved: Step 1: Perform NMF decomposition on the short-time Fourier transform result of the aliased signal received by the monitoring node to obtain the basis matrix of signal decomposition and the representation matrix of the aliased spectrum signal type; Step 2: Construct a cascade matrix based on the characterization matrix of the aliased spectrum signal type and the indicator factor; Step 3: Perform HL test on the elements in different cascade matrices to obtain the optimal anchor node group; The HL test is performed according to the following formula. The combination with the largest HL test value is considered to be the optimal anchor node group: : ； in, Indicates that the When building the anchor node group for the first time, the location signal source Ranking results; ; is the number of anchor nodes, is the number of signal templates in the library; Step 4: Split the optimal anchor node group by columns to obtain several coefficient matrices; Step 5: Use the DempsterShafer theory to perform multi-perspective information fusion on each coefficient matrix to achieve the first layer of collaborative perception to determine the category of the spectrally aliased wireless signal; Step 6: Decompose each coefficient matrix NMF into the distance representation matrix, signal power and position coupling matrix of the monitoring node; The distance representation matrix of the monitoring node Sum signal power and position coupling matrix , the particle swarm optimization algorithm is used to obtain the distance representation matrix of the monitoring nodes Optimal initial value and introduction of consensus result matrix , update according to the following formula: ； ； ； in, is a hyperparameter, Represents the multiplication operation of the corresponding elements of the matrix, is the transpose operation of the matrix; Step 7: The signal power and position coupling matrix is further decomposed by NMF into a power coefficient matrix for power estimation and a radian information matrix for positioning.

2. The collaborative multi-layer identification and node selection method for spectrum-aliased wireless signals according to claim 1, characterized in that: In step 7, the signal power and position coupling matrix Decomposed into a power coefficient matrix Radian information matrix , introduce the consensus result matrix , update according to the following formula: ； ； ； Among them, the power coefficient matrix Represents the power estimate of the spectral aliased wireless signal, radian information matrix Represents the positioning of spectrum-aliased wireless signals, is a hyperparameter.

3. A server comprising a processor and a memory, characterized in that: At least one program is stored in the memory, and the program is loaded and executed by the processor to implement the collaborative multi-layer identification and node selection method for spectrum-aliased wireless signals as described in any one of claims 1 or 2.

4. A computer-readable storage medium storing at least one program, characterized in that: The program is loaded and executed by a processor to implement the collaborative multi-layer identification and node selection method for spectrum-aliased wireless signals as described in any one of claims 1 or 2.

Images