Multi-source target recognition method based on single-channel nonlinear blind source separation

Abstract

The disclosure relates to a multi-source target identification method based on single-channel nonlinear blind source separation. The method comprises the following steps: establishing a single-channel nonlinear blind source separation model by using multi-interference source information collected by a receiving device at the same time; converting the multi-interference source information into multi-source nonlinear aliasing signals; performing blind source separation processing on the multi-source nonlinear aliasing signals to obtain separated signals; and performing signal identification processing on the separated signals to obtain multi-source target identification results; wherein the blind source separation comprises a self-organizing feature mapping network and a radial basis function network. The multi-source target identification method based on single-channel nonlinear blind source separation can effectively solve the mode aliasing problem and the end effect problem, thereby improving the similarity between the separated signals and the multi-interference source information.

Description

Technical Field

[0001] This disclosure relates to the field of multi-source target recognition technology, and in particular to a multi-source target recognition method based on single-channel nonlinear blind source separation. Background Technology

[0002] Blind source separation is a type of blind signal processing technique. It refers to signal processing techniques that separate multiple mixed signals based solely on obtained observation data, without any prior knowledge of the source signals or transmission channels. Common classifications of blind source separation techniques include: based on the number of mixing channels, it can be divided into single-channel and multi-channel blind source separation; based on the mixing method of the source signals, it can be divided into linear mixing and nonlinear mixing, where linear mixing can be further divided into convolutional mixing and instantaneous mixing, and nonlinear mixing can be further divided into general nonlinear models and post-nonlinear aliasing models; based on the relationship between the number of source signals and the number of observed signals, it can be divided into underdetermined, positive-definite, and overdetermined blind source separation.

[0003] Single-channel blind source separation is a type of underdetermined blind source separation algorithm. It involves extracting observation data from a single receiving channel to estimate signals from multiple different sources. Solving single-channel blind source separation is relatively complex. This technique requires fewer receiving devices and enhances their flexibility when resolving multi-source targets. In practical environments, due to the dynamic nature of transmission channels, the mixed signals detected by sensors are often obtained through nonlinear mixing. While nonlinear mixing is efficient, it often leads to severe nonlinear distortion problems, such as mode aliasing and endpoint effects. Therefore, combining single-channel and nonlinear blind source methods to address these technical issues is of significant research importance.

[0004] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this disclosure is to provide a multi-source target recognition method based on single-channel nonlinear blind source separation, including the following steps:

[0006] A single-channel nonlinear blind source separation model is established by using information from multiple interference sources collected by the receiving device at the same time.

[0007] The information from the multiple interference sources is converted into a multi-source nonlinear aliasing signal;

[0008] The multi-source nonlinear aliasing signal is subjected to blind source separation processing to obtain a separated signal;

[0009] The separated signals are subjected to signal recognition processing to obtain multi-source target recognition results;

[0010] The blind source separation process includes a self-organizing feature map network and a radial basis function network.

[0011] In an exemplary embodiment of this disclosure, in the step of establishing a single-channel nonlinear blind source separation model using multi-interference source information collected by the receiving device at the same time, the multi-interference source information includes single-tone interference, multi-tone interference, frequency sweep interference, frequency modulation interference, pulse interference, BPSK broadband interference, BPSK narrowband interference, and spoofing interference.

[0012] In an exemplary embodiment of this disclosure, the step of converting the multi-interference source information into a multi-source nonlinear aliasing signal includes:

[0013] Each interference source information in the multi-interference source information is subjected to linear aliasing processing to obtain multiple linearly aliased interference source information;

[0014] The information from multiple linear aliasing interference sources is subjected to nonlinear aliasing processing to obtain the multi-source nonlinear aliasing signal.

[0015] In an exemplary embodiment of this disclosure, the step of performing blind source separation processing on the multi-source nonlinear aliasing signal to obtain a separated signal includes:

[0016] The multi-source nonlinear aliased signal is subjected to noise reduction preprocessing to obtain a multi-source denoised nonlinear aliased signal.

[0017] The multi-source denoised nonlinear aliased signal is subjected to empirical mode decomposition to obtain multiple IMF component signals;

[0018] The self-organizing feature map network is used to cluster multiple IMF component signals, and the result of the clustering process is input into the radial basis function network to obtain the separated signal output by the radial basis function network.

[0019] In an exemplary embodiment of this disclosure, the step of performing noise reduction preprocessing on the multi-source nonlinear aliasing signal includes:

[0020] The multi-source nonlinear aliased signal is subjected to zero-mean processing; and the multi-source nonlinear aliased signal after zero-mean processing is subjected to whitening processing; thus obtaining the multi-source denoised nonlinear aliased signal.

[0021] The formula for zero-mean normalization includes:

[0022]

[0023] The formula for the whitening treatment includes:

[0024]

[0025] in, This represents the multi-source noise reduction nonlinear aliasing signal after whitening; X i0 (t) represents the multi-source denoised nonlinear aliased signal after zero-mean processing; Q represents the whitening matrix, Q = Λ -1 / 2 U T ;X i (t) represents the mixed signal; Λ=diag(d1,d2,...,d m ), where represents a diagonal matrix consisting of m eigenvalues; U represents an eigenmatrix consisting of eigenvectors corresponding to the m eigenvalues; T represents the transpose of a vector; and t represents a certain time.

[0026] In an exemplary embodiment of this disclosure, the step of performing empirical mode decomposition on the multi-source noise-reduced nonlinear aliasing signal to obtain multiple IMF component signals includes:

[0027] Determine all local extrema in the multi-source denoised nonlinear aliasing signal; the local extrema include multiple local maxima and multiple local minima.

[0028] Connect all the local maxima to form the upper envelope x. max (t); Connect all the local minima to form the lower envelope x. min (t); and from this, the average envelope m is obtained. i (t);

[0029] The average envelope m i The formula for (t) includes:

[0030] m i (t)=(x max (t)+x min (t)) / 2 (3)

[0031] Where i = 1, 2, 3, ..., N;

[0032] Calculate the multi-source noise reduction nonlinear aliasing signal With the average envelope m i The difference between (t) is calculated, and the difference h is calculated. i The formula for (t) includes:

[0033]

[0034] Where i = 1, 2, 3, ..., N;

[0035] Using the difference h i(t) Replace And repeat the steps described above for calculating the average envelope and the difference until the difference h is reached. i If (t) meets the conditions to become the IMF component signal, then the difference h i (t) is IMF1, i = 1, denoted as C1(t);

[0036] Take the C1(t) from the After removing the aliasing, a new multi-source denoised nonlinear aliasing signal is obtained.

[0037]

[0038] For the obtained C1(t), repeat all the above steps until all the differences h that satisfy the condition of becoming the IMF component signal are obtained. i (t), and the remaining residual signal r n (t) is a monotonic function; that is,

[0039] Calculate the energy E of each of the IMF component signals. i The calculation formula includes:

[0040]

[0041] The total energy E of all IMF component signals is calculated using the following formulas:

[0042]

[0043] Calculate the relative energy η of each of the IMF component signals. k The calculation formula includes:

[0044] η k =E i / E×100% (9)

[0045] All the obtained IMF component signals are arranged according to the relative energy η. k Arrange in descending order and remove the remaining residual signals r. n (t), to obtain the intrinsic mode functions (IMFs) corresponding to the relative energies of all the IMF component signals. k All the intrinsic mode functions imf k With the whitened multi-source noise-reduced nonlinear aliasing signal A new multi-channel signal Z(t) is formed, Z(t) = [x(t), imf1,, imf2,... imf k ].

[0046] In an exemplary embodiment of this disclosure, the conditions for satisfying the IMF component signal include:

[0047] The number of local extrema and the number of zero-crossing points differ by at most one;

[0048] At any given time, the mean of the upper envelope and the lower envelope is zero.

[0049] In an exemplary embodiment of this disclosure, the step of performing blind source separation processing on the multi-source nonlinear aliasing signal to obtain a separated signal includes:

[0050] The multi-channel signal Z(t) is clustered using the self-organizing feature map network, and the clustering result is input into the radial basis function network to obtain the separated signal output by the radial basis function network;

[0051] The output formula of the radial basis function network includes:

[0052] y = WK(x,p) (10)

[0053] Where y = [y1, y2, ..., y n ] T This represents the corresponding n-dimensional output vector; This represents the output weight matrix; x = [x1, x2, ..., x...]. n ] T This represents an n-dimensional input signal vector; p = μ1, ..., μ N ;σ1,…,σ N The parameter vector representing the radial basis functions;

[0054] The radial basis function network includes radial basis functions, wherein when the radial basis functions are Gaussian functions...

[0055]

[0056] Where, μ i ,i=1,2,…,N, represents the centroid of the i-th cluster in the hidden layer; σ i Let represent the variance of the Gaussian function, and Cmax represents the maximum distance from the i-th cluster to the centroid; M represents the number of centroids in the cluster.

[0057] In an exemplary embodiment of this disclosure, the step of performing signal recognition processing on the separated signals to obtain multi-source target recognition results includes:

[0058] The separated signal is preprocessed for identification, resulting in a preprocessed identification signal.

[0059] The preprocessed identification signal is subjected to feature extraction, and then trained through a neural network to finally obtain the multi-source target identification result.

[0060] In one exemplary embodiment of this disclosure, the multi-source target identification result corresponds to the multi-interference source information.

[0061] The technical solution provided in this disclosure may include the following beneficial effects:

[0062] This disclosure proposes a multi-source target identification method based on single-channel nonlinear blind source separation. This method improves the similarity between the separated signal and the multi-interference source signal by combining a self-organizing feature map network and a radial basis function network in single-channel nonlinear blind source processing, thereby effectively solving the mode aliasing problem and the endpoint effect problem. Attached Figure Description

[0063] 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. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0064] Figure 1 This diagram illustrates the steps of a multi-source target recognition method based on single-channel nonlinear blind source separation in an exemplary embodiment of this disclosure.

[0065] Figure 2 A flowchart illustrating a multi-source target recognition method based on single-channel nonlinear blind source separation in an exemplary embodiment of this disclosure is shown.

[0066] Figure 3 This diagram illustrates a multi-source interference scenario in an exemplary embodiment of the present disclosure.

[0067] Figure 4 This diagram illustrates the upper envelope, lower envelope, and average envelope in an exemplary embodiment of the present disclosure.

[0068] Figure 5 This diagram illustrates the results of empirical mode decomposition in an exemplary embodiment of this disclosure.

[0069] Figure 6 This diagram illustrates the structure of a radial basis function network in an exemplary embodiment of this disclosure.

[0070] Figure 7 A schematic diagram of a single-tone interference signal in an exemplary embodiment of this disclosure is shown;

[0071] Figure 8A schematic diagram of a multi-tone interference signal in an exemplary embodiment of this disclosure is shown. Detailed Implementation

[0072] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that this disclosure will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0073] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0074] This example implementation provides a multi-source target recognition method based on single-channel nonlinear blind source separation, such as... Figure 1 and Figure 2 As shown, the following steps may be included:

[0075] Step S101: Utilize the multi-interference source information collected by the receiving device at the same time to establish a single-channel nonlinear blind source separation model;

[0076] Step S102: Convert the multi-interference source information into a multi-source nonlinear aliasing signal;

[0077] Step S103: Perform blind source separation processing on the multi-source nonlinear aliasing signal to obtain the separated signal;

[0078] Step S104: Perform signal recognition processing on the separated signals to obtain multi-source target recognition results.

[0079] Here, blind source separation processing includes self-organizing feature map networks and radial basis function networks.

[0080] This disclosure proposes a multi-source target identification method based on single-channel nonlinear blind source separation. This method improves the similarity between the separated signal and the multi-interference source signal by combining a self-organizing feature map network and a radial basis function network in single-channel nonlinear blind source processing, thereby effectively solving the mode aliasing problem and the endpoint effect problem.

[0081] The steps of the method described above in this example embodiment will now be explained in more detail.

[0082] In step S101, blind source separation, which is frequently used in practical environments, results in a mixed signal detected by the receiving device, such as a receiver, that is typically nonlinearly mixed due to the dynamic characteristics of its transmission channel. Nonlinearity can be... Figure 3 As shown, the receiver receives information from multiple interference sources simultaneously. These interference sources generally include single-tone interference, multi-tone interference, frequency sweep interference, frequency modulation interference, pulse interference, BPSK wideband interference, BPSK narrowband interference, and deception interference. Figure 1 Interference source 1, interference source 2, and so on are used to represent the various interferences mentioned above.

[0083] In this embodiment, a single-channel nonlinear blind source separation model was established by utilizing this multi-interference source information.

[0084] In step S102, as Figure 2 As shown, this step includes the following sub-steps:

[0085] Sub-step S1021: Perform linear aliasing processing on each interference source information in the multi-interference source information to obtain multiple linearly aliased interference source information.

[0086] Sub-step S1022: Perform nonlinear aliasing processing on the information from these multiple linear aliasing interference sources to obtain a multi-source nonlinear aliasing signal.

[0087] In step S103, blind source separation processing is performed on the multi-source nonlinear aliasing signal to obtain the separated signal. This step includes the following sub-steps:

[0088] Sub-step S1031: Perform noise reduction preprocessing on the obtained multi-source nonlinear aliasing signal to obtain a multi-source denoised nonlinear aliasing signal.

[0089] Specifically, the first step is to perform zero-mean processing on the multi-source nonlinear aliasing signal, and then perform whitening processing after zero-mean processing to finally obtain the multi-source denoised nonlinear aliasing signal.

[0090] Here, the formula for zero-mean normalization is:

[0091]

[0092] The formula for whitening treatment is:

[0093]

[0094] in, This represents the multi-source noise reduction nonlinear aliasing signal after whitening; X i0 (t) represents the multi-source denoised nonlinear aliased signal after zero-mean processing; Q represents the whitening matrix, Q = Λ -1 / 2 U T ;X i (t) represents the mixed signal; Λ=diag(d1,d2,...,d m ), where represents a diagonal matrix consisting of m eigenvalues; U represents an eigenmatrix consisting of eigenvectors corresponding to the m eigenvalues; T represents the transpose of a vector; and t represents a certain time.

[0095] Sub-step S1032: Perform Empirical Mode Decomposition (EMD) on the obtained multi-source denoised nonlinear aliasing signal to obtain multiple decomposed signals. These multiple separated signals are multiple Intrinsic Mode Function (IMF) component signals.

[0096] Sub-step S1032 specifically includes:

[0097] like Figure 4 As shown, all local extrema of the multi-source denoised nonlinear aliased signal are found; each local extrema includes multiple local maxima and multiple local minima; connecting all local maxima forms the upper envelope x. max (t); connect all local minima to form the lower envelope x. min (t); and from this, the average envelope m is calculated. i (t).

[0098] Here, the formula for calculating the average envelope is:

[0099] m i (t)=(x max (t)+x min (t)) / 2 (3)

[0100] Where i = 1, 2, 3, ..., N.

[0101] Then, the multi-source denoising nonlinear aliasing signal is calculated. With the average envelope m i The difference h of (t) i (t). Here, the difference h is calculated. i The formula for (t) includes:

[0102]

[0103] Where i = 1, 2, 3, ..., N.

[0104] Next, use this difference h i (t) Replace Repeat the steps above to calculate the average envelope and the difference until the difference h is reached. i If (t) meets the conditions to become an IMF component signal, then the difference h is... i (t) is IMF1, denoted as C1(t) when i = 1, and as C in other cases. i (t).

[0105] Then, C1(t) is removed from... After removing the aliasing, a new multi-source denoised nonlinear aliasing signal is obtained.

[0106] Right now:

[0107] For the obtained C1(t), repeat all the above steps until all the differences h that satisfy the condition of becoming IMF component signals are obtained. i (t), and the remaining residual signal r n (t) is a monotonic function;

[0108] Right now:

[0109] Calculate the energy E of each IMF component signal. i The calculation formula is as follows:

[0110]

[0111] The total energy E of all IMF component signals is calculated using the following formula:

[0112]

[0113] Calculate the relative energy η of each IMF component signal. k The calculation formula is as follows:

[0114] η k =E i / E×100% (9)

[0115] Finally, as Figure 5 As shown, all the obtained IMF component signals are sorted according to their relative energy η k Arrange in descending order and remove the remaining residual signal r. n (t), to obtain the intrinsic mode functions (IMFs) corresponding to the relative energies of all IMF component signals. k All intrinsic mode functions imf k With multi-source noise reduction nonlinear aliasing signal A new multi-channel signal Z(t) is formed, Z(t) = [x(t), imf1,, imf2,... imf k ].

[0116] Therefore, empirical mode decomposition is a method that recursively decomposes the signal itself on its time scale, making the complex signal decompose into a series of unknown but independent, distinct, orthogonal finite signal components arranged according to frequency magnitude. These signal components are called intrinsic mode functions (IMFs).

[0117] In this embodiment, the multi-source denoised nonlinear aliased signal is decomposed into different IMF component signals. The intrinsic mode functions (IMFs) mentioned in this embodiment must satisfy the following conditions:

[0118] The number of all local extrema and the number of zero-crossings differ by at most one;

[0119] At any given time, the mean of the local upper envelope and the local lower envelope is zero.

[0120] However, empirical mode decomposition has certain drawbacks: namely, mode aliasing and endpoint effects.

[0121] Mode aliasing refers to the phenomenon where an IMF component signal contains significantly different characteristic time scales, or similar characteristic time scales distributed across different IMFs, causing adjacent IMF waveforms to overlap, influence each other, and become difficult to distinguish. In EMD (Electronic Mode Decomposition), the upper envelope represents local maxima, and the lower envelope represents local minima. The average envelope is then derived from the upper and lower envelopes. During the calculation of the average envelope, the presence of anomalous events in the signal inevitably affects the selection of local extrema, resulting in an uneven distribution of extrema. This leads to the calculated average envelope being a combination of the local envelope of the anomalous event and the envelope of the true signal. Therefore, after calculating the average envelope, the filtered IMF component signals contain both the signal's inherent modes and anomalous events, or inherent modes from adjacent characteristic time scales, thus causing mode aliasing.

[0122] Endpoint effects are a major factor affecting the accuracy of empirical mode decomposition (EMD). During the selection process, the ends of the data sequences in the upper and lower envelopes diverge. Endpoint effects increase spurious components and consequently increase the total signal energy. To address this problem, various algorithms have been proposed, such as directly using data endpoints as extrema, polynomial fitting algorithms, neural network extension algorithms, and combining extrema with symmetric extension.

[0123] The similarity coefficient ρ between each component signal after EMD decomposition and its corresponding source signal undergoes shape distortion, causing endpoint effects and thus leading to inaccurate decomposition of each component signal. The effectiveness of a consistent endpoint effect suppression algorithm can be evaluated by comparing the similarity between each IMF component signal after EMD decomposition and the source signal. The following formula can be used for evaluation:

[0124]

[0125] In this formula, cov() represents the covariance; σ() represents the variance; and IMF represents the variance. i This represents the i-th mode component of the source signal after EMD decomposition; x i This represents the corresponding source signal components. It can be seen that the larger the ρ value, the better the suppression of the endpoint effect;

[0126] another,

[0127] In this formula, S represents the total number of source signals; x i (k) represents the i-th component of the source signal; IMF i (k) represents the corresponding component obtained after EMD decomposition. It can be seen that error_IMF i The smaller the value, the better the suppression of the endpoint effect.

[0128] To address the two issues mentioned above, two neural networks are utilized in sub-step S1033: a self-organizing feature mapping (SOFM) network and a radial basis function (RBF) network.

[0129] Blind source separation is a signal processing method that recovers or extracts the source signal from the mixed output signal of a sensor array or converter when the characteristics of the source signal and the transmission channel are unknown. In sub-step S1033, multiple IMF component signals are clustered using an SOFM network, and the clustering result is input into an RBF network to obtain the final separated signal output by the RBF network.

[0130] Here, the Self-Organizing Feature Map Network (SOFM) is an unsupervised network that does not require modeling and does not impose restrictions on the clustering of the output.

[0131] The basic principle of the SOFM network is as follows: When a certain pattern is input, a node in the output layer receives the maximum stimulus and wins. Nodes surrounding the winning node are also stimulated due to lateral effects. At this point, the network performs a learning operation, adjusting the connection weights of the winning node and its surrounding nodes in the direction of the input pattern. When the input pattern category changes, the winning node in the two-dimensional plane also moves from its original node to another node. In this way, the network adjusts its connection weights using a large amount of sample data through self-organization, ultimately ensuring that the feature map of the network's output layer reflects the distribution of the sample data.

[0132] like Figure 6 The diagram illustrates the schematic structure of a Radial Basis Function (RBF) network. As can be seen, an RBF network is an artificial neural network that uses radial basis functions as activation functions. The output of a RBF network is a linear combination of the input radial basis functions and the neuron parameters.

[0133] Specifically, in the steps of blind source separation using these two neural networks,

[0134] The multi-channel signal Z(t) is clustered using an SOFM network, and the clustering result is input into an RBF network to obtain the separated signal output by the RBF network.

[0135] Here, the output formula of the radial basis function network includes:

[0136] y = WK(x,p) (10)

[0137] Where y = [y1, y2, ..., y n ] T This represents the corresponding n-dimensional output vector; This represents the output weight matrix; x = [x1, x2, ..., x...]. n ] T Represents an n-dimensional input signal vector; p = (μ1, ..., μ) N ;σ1,…,σ N ) represents the parameter vector of the radial basis functions;

[0138] The radial basis function network includes radial basis functions, wherein when the radial basis functions are Gaussian functions...

[0139]

[0140] Where, μ i ,i=1,2,…,N, represents the centroid of the i-th cluster in the hidden layer; σ i Let represent the variance of the Gaussian function, and Cmax represents the maximum distance from the i-th cluster to the centroid; M represents the number of centroids in the cluster.

[0141] That is, when the radial basis function is a Gaussian function, the variance and cluster center of each node in the hidden layer are first solved by the output vector, and then the weight vector w of the output layer is solved by the input vector based on the values ​​of the center and variance.

[0142] Step S104 includes the following sub-steps:

[0143] Sub-step S1041: Perform identification preprocessing on the separated signal to obtain a preprocessed signal;

[0144] Sub-step S1042: Extract features from the obtained preprocessed signal and train it through a BP neural network to finally obtain the multi-source target recognition result.

[0145] The obtained multi-source target identification results correspond to the initial multi-interference source information.

[0146] It's important to clarify here that signal identification, or determining the category of the signal after blind source separation, employs feature extraction to use the unique signal characteristics of multiple interference sources as the basis for judgment. Through training and learning via a neural network, the separated signal can be identified.

[0147] For example, such as Figure 7 and Figure 8 As shown, the design principle of the feature parameters is to utilize the unique signal characteristics of multiple interference sources as parameters for extraction. Figure 7 and Figure 8 It can be seen that both single-tone interference and multi-tone interference have the characteristic that one or several spectral lines have significantly higher amplitudes than other parts of the spectrum. Therefore, based on this characteristic, the characteristic parameter of single-frequency energy concentration can be constructed to distinguish single-tone interference, multi-tone interference, and other interferences. Single-tone interference exhibits a certain amplitude impulse within the main frequency band of the navigation signal, while the interference power at other frequency points is zero. Other types of interference do not have this characteristic. Therefore, the ratio of the maximum to the second-largest amplitude of the single-tone interference spectrum must be larger than that of other types of interference, thus distinguishing single-tone interference from other interferences.

[0148] It should be noted that although several units of the system for executing actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units. Some or all of the units can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without any inventive effort.

[0149] 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 application 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 appended claims.

Claims

1. A multi-source target recognition method based on single-channel nonlinear blind source separation, characterized in that, Includes the following steps: A single-channel nonlinear blind source separation model is established by using multi-interference source information collected by the receiving device at the same time. The information from the multiple interference sources is converted into a multi-source nonlinear aliasing signal; Blind source separation processing is performed on the multi-source nonlinear aliasing signal to obtain a separated signal. The blind source separation processing includes a self-organizing feature map network and a radial basis function network, comprising: The multi-source nonlinear aliased signal is subjected to noise reduction preprocessing to obtain a multi-source denoised nonlinear aliased signal. The multi-source denoised nonlinear aliased signal is subjected to empirical mode decomposition to obtain multiple IMF component signals; The self-organizing feature map network is used to cluster multiple IMF component signals, and the result of the clustering is input into the radial basis function network to obtain the separated signal output by the radial basis function network. The separated signals are processed for signal recognition to obtain multi-source target recognition results.

2. The multi-source target recognition method for single-channel nonlinear blind source separation according to claim 1, characterized in that, In the step of establishing a single-channel nonlinear blind source separation model using multi-interference source information collected by the receiving device at the same time, the multi-interference source information includes single-tone interference, multi-tone interference, frequency sweep interference, frequency modulation interference, pulse interference, BPSK broadband interference, BPSK narrowband interference, and spoofing interference.

3. The multi-source target recognition method for single-channel nonlinear blind source separation according to claim 1, characterized in that, The steps of converting the multi-interference source information into a multi-source nonlinear aliasing signal include: Each interference source information in the multi-interference source information is subjected to linear aliasing processing to obtain multiple linearly aliased interference source information; The information from multiple linear aliasing interference sources is subjected to nonlinear aliasing processing to obtain the multi-source nonlinear aliasing signal.

4. The multi-source target recognition method for single-channel nonlinear blind source separation according to claim 1, characterized in that, The steps for noise reduction preprocessing of the multi-source nonlinear aliasing signal include: The multi-source nonlinear aliased signal is subjected to zero-mean processing; and the multi-source nonlinear aliased signal after zero-mean processing is subjected to whitening processing; thus obtaining the multi-source denoised nonlinear aliased signal. The formula for zero-mean normalization includes: The formula for the whitening treatment includes: in, This represents the multi-source noise reduction nonlinear aliasing signal after whitening processing; This represents a multi-source denoised nonlinear aliased signal after zero-mean processing. Represents the whitening matrix. ; Indicates a mixed signal; , represents a diagonal matrix consisting of m eigenvalues; This represents the feature matrix composed of the feature vectors corresponding to m feature values; T t represents the transpose of a vector; t represents a certain moment in time.

5. The multi-source target recognition method for single-channel nonlinear blind source separation according to claim 4, characterized in that, The steps of performing empirical mode decomposition on the multi-source denoised nonlinear aliased signal to obtain multiple IMF component signals include: Determine all local extrema in the multi-source denoised nonlinear aliased signal; the local extrema include multiple local maxima and multiple local minima. Connect all the local maxima to form the upper envelope. Connect all the local minima to form the lower envelope. And from this, the average envelope is obtained. ; The average envelope The formulas include: （3） in, i =1,2,3,...,N; Calculate the multi-source noise reduction nonlinear aliasing signal With the average envelope The difference is calculated. The formulas include: （4） in, i =1,2,3,...,N; Using the difference replace And repeat the steps described above for calculating the average envelope and the difference, until the difference is reached. If the conditions for becoming the IMF component signal are met, then the difference... For IMF1, i =1, denoted as ; The From the above After removing the aliasing, a new multi-source denoised nonlinear aliasing signal is obtained. ; (5) The obtained Repeat all the above steps until all the differences that satisfy the condition of becoming the IMF component signals are obtained. And the remaining residual signal It is a monotonic function; Right now, (6) Calculate the energy of each of the IMF component signals. The calculation formula includes: (7) The total energy of all IMF component signals was calculated. E The calculation formula includes: (8) Calculate the relative energy of each of the IMF component signals. The calculation formula includes: (9) All the obtained IMF component signals are arranged according to the relative energy. Arrange the signals in descending order and discard the remaining residual signals. The intrinsic mode functions corresponding to the relative energies of all the IMF component signals are obtained. All the aforementioned intrinsic mode functions With the multi-source noise reduction nonlinear aliasing signal To form a new multi-channel signal , .

6. The multi-source target recognition method for single-channel nonlinear blind source separation according to claim 5, characterized in that, The conditions for meeting the requirements of the IMF component signal include: The number of local extrema and the number of zero-crossing points differ by at most one; At any given time, the mean of the upper envelope and the lower envelope is zero.

7. The multi-source target recognition method for single-channel nonlinear blind source separation according to claim 5, characterized in that, In the step of performing blind source separation processing on the multi-source nonlinear aliasing signal to obtain the separated signal: The self-organizing feature mapping network is used to analyze the multi-channel signals. Clustering is performed, and the clustering results are input into the radial basis function network to obtain the separation signal output by the radial basis function network; The output formula of the radial basis function network include: (10) in, Indicates the corresponding 3D output vector; , This represents the output weight moments; express 3D input signal vector; The parameter vector representing the radial basis functions; The radial basis function network includes radial basis functions, wherein when the radial basis functions are Gaussian functions... (11) in, Indicates the first hidden layer The center of each basis function; Let represent the variance of the Gaussian function, and ; This represents the maximum distance from the i-th cluster to the center point; M This indicates the number of cluster centers.

8. The multi-source target recognition method based on single-channel nonlinear blind source separation according to claim 1, characterized in that, The steps of performing signal recognition processing on the separated signals to obtain multi-source target recognition results include: The separated signal is preprocessed for identification, resulting in a preprocessed identification signal. The preprocessed identification signal is subjected to feature extraction, and then trained through a neural network to finally obtain the multi-source target identification result.

9. The multi-source target recognition method based on single-channel nonlinear blind source separation according to claim 8, characterized in that, The multi-source target identification result corresponds to the multi-interference source information.

Images