Signal extraction method and device based on ICA-R algorithm and medium

Abstract

The invention discloses a signal extraction method and device based on an ICA-R algorithm and a medium, and aims to solve the problem of same-frequency signal interference. According to the scheme, the cosine pulse signal with the same carrier frequency as the target signal is generated as the reference signal, so that effective separation can be still performed by using the carrier frequency under the condition of lacking symbol information of the target signal; a characteristic parameter threshold value is set by utilizing prior information of a target signal, after an estimation signal is obtained, whether the estimation signal meets a preset requirement is judged through the characteristic parameter threshold value, and the accuracy of an extraction result can be continuously optimized by introducing the threshold value and dynamically verifying an extraction signal; and until the preset number of optimization times is reached or the estimated signal meets the requirement of the target signal. According to the method, the defect of insufficient accuracy in the prior art is overcome to a certain extent, the robustness of accurately extracting the target signal by the ICA-R algorithm is improved, and the high requirement of the communication field on the signal extraction quality is met.

Description

Technical Field

[0001] The present invention relates to the field of blind source separation algorithm optimization, and in particular to a signal extraction method, device and medium based on ICA-R algorithm. Background Art

[0002] With the rapid development of electronic technology and communication technology, a large number of frequency-using equipment such as radio stations, mobile phones, and satellite communication terminals have emerged. The electromagnetic environment of radio communications has become increasingly complex. It is very likely that there will be mutual interference or even co-frequency interference between various frequency-using equipment, which will seriously affect the quality of communication. Traditional anti-interference methods based on filtering, frequency hopping, direct sequence spread spectrum, and antenna zeroing cannot effectively deal with the problem of co-frequency interference. As a digital signal processing technology based on the assumption of signal independence, the ICA (Independent Component Analysis) algorithm can achieve blind separation of co-frequency signals under the conditions of unknown signal sources and signal mixing methods. Using some prior information of the target signal, such as carrier frequency, phase, symbol rate, etc., a reference signal can be constructed to constrain the optimization target of the ICA algorithm and directly extract one or more target signals of interest.

[0003] The ICA algorithm with reference signal constraint is called ICA-R (Independent Component Analysis with Reference). The original ICA-R algorithm uses the sign function of the target signal as the reference signal, which can achieve more than 90% target signal extraction accuracy. However, in application fields such as communication and reconnaissance, the receiver cannot know the sign information of the target signal in advance, so it is impossible to construct the required reference signal. If a cosine pulse signal with the same frequency as the carrier of the target signal is used as the reference signal, the target signal extraction accuracy will be reduced to about 15%. This accuracy cannot meet the engineering application requirements in fields such as communication and reconnaissance.

[0004] To solve the above technical problems, the prior art uses a non-random initialization separation vector method or a meta-heuristic optimization algorithm to optimize the objective function of ICA-R to improve the accuracy of extracting the target signal. However, the accuracy of the target signal obtained by using these two methods in the prior art is still low. Summary of the invention

[0005] The purpose of the present invention is to provide a signal extraction method, device and medium based on the ICA-R algorithm, which to a certain extent overcomes the defect of insufficient accuracy in the current technology, improves the robustness of the ICA-R algorithm in accurately extracting the target signal, and meets the high requirements of the communication field for signal extraction quality.

[0006] In order to solve the above technical problems, the present invention provides a signal extraction method based on ICA-R algorithm, comprising:

[0007] receiving an input signal, wherein the input signal includes a target signal and an interference signal;

[0008] Generate a reference signal according to the carrier frequency of the target signal, wherein the reference signal is a cosine pulse signal having the same carrier frequency as the target signal;

[0009] Calculating characteristic parameters according to the prior information of the target signal and setting characteristic parameter thresholds;

[0010] Extracting an estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal;

[0011] Extracting characteristic parameters from the estimated signal, and determining whether the estimated signal meets preset requirements according to the characteristic parameters and the characteristic parameter threshold;

[0012] If not, the step of extracting an estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal is repeated until the estimated signal meets the preset requirement.

[0013] Preferably, calculating the characteristic parameters according to the prior information of the target signal includes:

[0014] Performing Fourier transform on the target signal to obtain a frequency domain signal;

[0015] Calculating the power spectral density of the frequency domain signal;

[0016] A power spectrum density symmetry coefficient is calculated according to the power spectrum density.

[0017] Preferably, the expression of the power spectrum density symmetry coefficient is:

[0018] ; Wherein, PSI is the symmetry coefficient of the power spectrum density, M is the number of sampling points in the power spectrum density that deviate from the center frequency, k represents the kth sampling point of the deviation, and Y L (k) is the power spectrum density value of the kth sampling point in the part where the power spectrum density of the target signal is less than the center frequency, Y R (k) is the power spectrum density value of the kth sampling point in the part where the power spectrum density of the target signal is greater than the center frequency, (k) is Y R (k) horizontal flip, Represents the modulus of a vector.

[0019] Preferably, the expression of the power spectrum density symmetry coefficient is:

[0020] ; Wherein, PSI is the symmetry coefficient of the power spectrum density, M is the number of sampling points in the power spectrum density that deviate from the center frequency, k represents the kth sampling point of the deviation, and Y L (k) is the power spectrum density value of the kth sampling point in the part where the power spectrum density of the target signal is less than the center frequency, Y R (k) is the power spectrum density value of the kth sampling point in the part where the power spectrum density of the target signal is greater than the center frequency, (k) is Y R (k) horizontal flip, represents the modulus of a vector, Represents the second-order normal form.

[0021] Preferably, calculating the characteristic parameter threshold according to the prior information of the target signal includes:

[0022] Perform Fourier transform on the target signal to obtain a frequency domain signal.

[0023] Calculating the power spectral density of the frequency domain signal;

[0024] Calculating a first power value within a signal bandwidth according to the distribution of the power spectrum density;

[0025] Calculating a second power value within the same bandwidth outside the signal bandwidth according to the distribution of the power spectrum density;

[0026] The ratio between the first power value and the second power value is calculated to obtain an in-band and out-band power ratio.

[0027] Preferably, the characteristic parameter is a power spectrum density symmetry coefficient, and determining whether the estimated signal meets a preset requirement according to the characteristic parameter and the characteristic parameter threshold comprises:

[0028] Determining whether a power spectral density coefficient in the estimated signal is greater than a power spectral density coefficient threshold;

[0029] If it is greater than the power spectrum density coefficient threshold, it is determined that the estimated signal meets the preset requirement;

[0030] If it is not greater than the power spectrum density coefficient threshold, it is determined that the estimated signal does not meet the preset requirement.

[0031] Preferably, after determining that the preset requirement is not met, the method further includes:

[0032] Determine whether the number of times the step of extracting the estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal is executed is greater than a set maximum number of optimization searches;

[0033] If it is greater than the set maximum number of optimization times, determining that the estimated signal is the target signal;

[0034] If it is not greater than the set maximum number of optimization times, the step of extracting an estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal is re-entered.

[0035] Preferably, the ICA-R algorithm includes a built-in number of iterations, and after determining that the number of times the step of extracting the estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal is performed is not greater than the set maximum number of optimization searches, further includes:

[0036] The built-in iteration number of the ICA-R algorithm is reset to 0, and the step of extracting the estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal is re-entered.

[0037] In order to solve the above technical problems, the present invention provides a signal extraction device based on ICA-R algorithm, comprising:

[0038] Memory for storing computer programs;

[0039] The processor is used to implement the steps of the signal extraction method based on the ICA-R algorithm described above when executing the computer program.

[0040] To solve the above technical problem, the present invention provides a non-volatile storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the signal extraction method based on the ICA-R algorithm are implemented.

[0041] The present invention provides a signal extraction method, device and medium based on ICA-R algorithm, aiming to solve the problem of co-frequency signal interference. This scheme generates a cosine pulse signal with the same carrier frequency as the target signal as a reference signal, and can still use the carrier frequency for effective separation in the absence of target signal symbol information; the characteristic parameter threshold is set using the prior information of the target signal, and after the estimated signal is obtained, the characteristic parameter threshold is used to determine whether the estimated signal meets the preset requirements. The introduction of the threshold and the dynamic verification of the extracted signal ensure that the accuracy of the extraction result can be continuously optimized until the preset number of optimizations is reached or the estimated signal meets the requirements of the target signal. This application not only improves the reliability of signal extraction, but also overcomes the defect of insufficient accuracy in the current technology to a certain extent, improves the robustness of the ICA-R algorithm in accurately extracting the target signal, and meets the high requirements of the communication field for signal extraction quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A schematic diagram of a flow chart of a signal extraction method based on the ICA-R algorithm provided by the present invention;

[0044] Figure 2 A schematic diagram of calculating the symmetry coefficient of power spectrum density provided by the present invention;

[0045] Figure 3 A block diagram of a target signal blind extraction system based on the ICA-R algorithm provided by the present invention;

[0046] Figure 4 A schematic diagram of an ICA-R algorithm processing module provided by the present invention;

[0047] Figure 5 A flow chart of an algorithm for extracting a target signal using prior information of the target signal provided by the present invention;

[0048] Figure 6 A curve diagram showing the variation of the target signal extraction accuracy with the number of times the algorithm is re-run when the threshold is set using PSI as a characteristic parameter provided by the present invention;

[0049] Figure 7 A schematic diagram of a signal extraction device based on the ICA-R algorithm provided by the present invention. DETAILED DESCRIPTION

[0050] The core of the present invention is to provide a signal extraction method, device and medium based on the ICA-R algorithm, which to a certain extent overcomes the defect of insufficient accuracy in the current technology, improves the robustness of the ICA-R algorithm in accurately extracting the target signal, and meets the high requirements of the communication field for signal extraction quality.

[0051] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] First, the ICA algorithm in the prior art is introduced. Specifically, the ICA algorithm with reference signal constraints is called ICA-R (Independent Component Analysis with Reference). The ICA-R algorithm contains equality and inequality constraints. The Lagrange multiplier method is usually used to transform the constrained optimization problem into an unconstrained optimization problem, and then a Newton-like optimization method with a learning rate is used to solve the optimal value of the separation vector. Because the Newton-like optimization method is sensitive to the learning rate and the initialization value of the separation vector, inappropriate initial values ​​may cause the algorithm to fail to converge to the global optimum or even diverge. Only when the initial point is close enough to the global optimal solution can the global convergence of the algorithm be guaranteed. Therefore, when the initial value of the separation vector is generated in a random manner, there is a certain probability that the ICA-R algorithm can accurately extract a certain expected target signal. Therefore, different ICA-R algorithms have different robustness in accurately extracting a certain target signal.

[0053] The original ICA-R algorithm uses the sign function of the target signal as the reference signal, which can achieve a target signal extraction accuracy of more than 90%. However, in application fields such as communications and reconnaissance, the receiver cannot know the sign information of the target signal in advance, so it is impossible to construct the required reference signal. If a cosine pulse signal with the same frequency as the carrier of the target signal is used as the reference signal, the target signal extraction accuracy will be reduced to about 15%. This accuracy cannot meet the engineering application requirements in fields such as communications and reconnaissance.

[0054] To solve the above technical problems, the prior art uses a method of non-randomly initializing separation vectors to improve the extraction accuracy. One method is to assume that the reference signal r(t) is the convergence target of the ICA-R algorithm. According to the signal extraction formula , through the Moore-Penrose generalized inverse operation, the initialization formula of the separation vector can be obtained as:

[0055] ;

[0056] in The Moore-Penrose generalized inverse matrix of X is called X, where X is the observed signal after whitening. Compared with the original ICA-R algorithm with random initialization, this method uses reference prior information not only in the contrast function, but also in the initialization of the separation vector. This method can increase the accuracy of extracting the target signal to about 95%. Nevertheless, for some communication occasions with high reliability requirements, this accuracy still needs to be further improved. In addition, this method needs to calculate the pseudo-inverse of the signal matrix, which has a high computational complexity, is more complicated for engineering implementation, and requires a large amount of memory resources.

[0057] Another solution is to use a meta-heuristic optimization algorithm to optimize the objective function of ICA-R. A meta-heuristic optimization algorithm is a type of algorithm that solves complex optimization problems by simulating phenomena or processes in nature. This method has the advantages of strong global search capabilities and good robustness, but it has shortcomings such as slow convergence, difficulty in parameter adjustment, high consumption of computing resources, and easy to fall into local optimality.

[0058] In order to solve the above technical problems, the present invention provides a signal extraction method based on ICA-R algorithm, such as Figure 1 As shown, Figure 1 A schematic flow chart of a signal extraction method based on the ICA-R algorithm provided by the present invention, the method comprising:

[0059] S11: receiving an input signal, where the input signal includes a target signal and an interference signal;

[0060] In the process of signal reception, the input signal usually contains the target signal (i.e. the useful information to be extracted) and the interference signal (which may be an interference signal or noise). This is because the communication environment in the real world is complex and changeable, full of various unpredictable electromagnetic waves and physical phenomena. These phenomena may originate from nature, such as lightning and solar radiation, or from human activities, such as the operation of other electronic devices, radio broadcasting, industrial equipment, etc. The electromagnetic waves generated by these external factors are mixed with the target signal to form an interference signal. In addition, the signal may be affected by physical effects such as attenuation, reflection, refraction, and scattering during transmission, resulting in a decrease in signal quality, which further increases the component of the interference signal. Therefore, the received input signal will inevitably contain the target signal and the interference signal, and the processing of the input signal needs to effectively separate them so that the target signal can be identified and extracted in the subsequent steps. The quality of this step directly affects the accuracy of subsequent signal processing, and ensuring the integrity and accuracy of the input signal is the key.

[0061] S12: Generate a reference signal according to the carrier frequency of the target signal, where the reference signal is a cosine pulse signal having the same carrier frequency as the target signal;

[0062] According to the carrier frequency of the target signal, a reference signal is generated, which is a cosine pulse signal with the same carrier frequency as the target signal. This reference signal provides a benchmark, which is helpful for signal separation in the subsequent ICA-R algorithm. The generated reference signal should have good time domain and frequency domain characteristics to retain the characteristics of the target signal to the greatest extent and provide effective support for subsequent blind signal separation. As a preferred embodiment, the relationship between the cosine reference signal is: r(t)=cos(2πf c t); where r(t) is the reference signal, f c is the carrier frequency of the target signal, and t is the time.

[0063] S13: Calculate characteristic parameters according to the prior information of the target signal and set characteristic parameter thresholds;

[0064] The feature parameters are calculated using the prior information of the target signal. These parameters may include the amplitude, phase, frequency and other characteristics of the signal. Based on these feature parameters, the threshold of the feature parameters is set. The selection of the threshold is crucial for the effective extraction of the target signal. A reasonable feature parameter threshold can effectively suppress interference signals and improve the extraction accuracy of the target signal.

[0065] As a preferred embodiment, the characteristic parameters are calculated according to the prior information of the target signal, including: performing Fourier transform on the target signal to obtain a frequency domain signal; calculating the power spectrum density of the frequency domain signal; and calculating the power spectrum density symmetry coefficient according to the power spectrum density. In a preferred embodiment, in order to calculate the characteristic parameters according to the prior information of the target signal, the target signal is firstly Fourier transformed to convert it from the time domain to the frequency domain. The Fourier transform can reveal the frequency components of the signal and its amplitude and phase characteristics at different frequencies. Then the power spectrum density of the frequency domain signal is calculated. This process obtains the energy distribution of the signal at each frequency by statistically analyzing the amplitude square of the frequency domain signal. The power spectrum density not only reflects the main frequency components of the signal, but also reveals potential noise and interference. Finally, the power spectrum density symmetry coefficient is calculated according to the power spectrum density, aiming to analyze the spectral characteristics and symmetry of the signal and help identify the nature and quality of the signal. The value of the symmetry coefficient can indicate whether the signal is evenly distributed in the positive and negative frequency ranges, which is crucial for further signal processing and extraction of the target signal. Through this series of steps, the frequency domain features can be fully utilized to provide an effective basis for the extraction of the target signal.

[0066] As a preferred embodiment, the characteristic parameter is a power spectrum density symmetry coefficient, and determining whether the estimated signal meets the preset requirements according to the characteristic parameter and the characteristic parameter threshold includes: judging whether the power spectrum density coefficient in the estimated signal is greater than the power spectrum density coefficient threshold; if it is greater than the power spectrum density coefficient threshold, judging that the estimated signal meets the preset requirements; if it is not greater than the power spectrum density coefficient threshold, judging that the estimated signal does not meet the preset requirements. Specifically, firstly, it is judged whether the power spectrum density coefficient in the estimated signal is greater than the set power spectrum density coefficient threshold. The setting of this threshold is based on the results of the previous signal analysis, and is intended to provide a benchmark for distinguishing valid signals from interference or noise signals. If the power spectrum density coefficient of the estimated signal is greater than the threshold, it is judged that the signal meets the preset requirements, which means that the characteristics of the signal are consistent with the target signal and have good signal quality and availability. On the contrary, if the power spectrum density coefficient is not greater than the threshold, it is judged that the estimated signal does not meet the preset requirements, which may indicate that the signal is affected by interference or noise and needs to be further processed or re-extracted. This judgment process provides an effective quality control mechanism for signal extraction, ensuring that the final extracted signal has high reliability and accuracy.

[0067] As a preferred embodiment, the expression of the power spectrum density symmetry coefficient is:

[0068] ; Where PSI is the symmetry coefficient of the power spectrum density, M is the number of sampling points in the power spectrum density that deviate from the center frequency, k represents the kth sampling point of the deviation, and Y L (k) is the power spectrum density value of the kth sampling point in the part where the power spectrum density of the target signal is less than the center frequency, Y R (k) is the power spectrum density value of the kth sampling point in the part where the power spectrum density of the target signal is greater than the center frequency, (k) is Y R (k) horizontal flip, Represents the modulus of a vector.

[0069] Figure 2 A schematic diagram of calculating the symmetry coefficient of power spectrum density provided by the present invention, wherein Indicates the carrier frequency of the target signal, Y L Indicates the part where the power spectrum density of the target signal is less than the center frequency, Y R Indicates that the power spectrum density of the target signal is greater than the center frequency. Indicates the distance the signal frequency deviates from the center frequency.

[0070] Assuming that the target signal is a digitally modulated communication signal, then for the receiver, the carrier frequency of the communication signal is is known prior information. The power spectral density symmetry coefficient (PSI) of the signal is defined as follows:

[0071] ;

[0072] in yes Horizontal flip.

[0073] In the absence of noise and interference signals, the power spectral density of a pure digital modulated communication signal is The signal is symmetrical around the center, that is, PSI = 0. However, after the signal is superimposed with Gaussian white noise, due to the random disturbance of Gaussian white noise, the power spectral density of the mixed signal will no longer be The PSI value will be greater than 0 if the power spectrum density is not centered on For a centrally symmetrical interference signal, the PSI value of the mixed signal will become larger. Therefore, the PSI value under a certain signal-to-noise ratio condition can be calculated in advance, and then used as a characteristic parameter threshold to determine whether the estimated signal output by the ICA-R algorithm is the target signal.

[0074] As a preferred embodiment, the expression of the power spectrum density symmetry coefficient is:

[0075] ; Where PSI is the symmetry coefficient of the power spectrum density, M is the number of sampling points in the power spectrum density that deviate from the center frequency, k represents the kth sampling point of the deviation, and Y L (k) is the power spectrum density value of the kth sampling point in the part where the power spectrum density of the target signal is less than the center frequency, Y R (k) is the power spectrum density value of the kth sampling point in the part where the power spectrum density of the target signal is greater than the center frequency, (k) is Y R (k) horizontal flip, represents the modulus of a vector, Represents the second-order normal form. The value range of the index PSI is [0,1]. The larger the index, the stronger the spectrum symmetry of the signal. Ideally, PSI is 0, and it becomes a non-zero value after superimposing noise. This value can be calculated in advance and used as a parameter threshold to determine whether the algorithm accurately extracts the target signal.

[0076] As a preferred embodiment, the characteristic parameter threshold is calculated according to the prior information of the target signal, including: performing Fourier transform on the target signal to obtain a frequency domain signal. Calculating the power spectrum density of the frequency domain signal; calculating the first power value within the signal bandwidth according to the distribution of the power spectrum density; calculating the second power value within the same bandwidth outside the signal bandwidth according to the distribution of the power spectrum density; calculating the ratio between the first power value and the second power value to obtain the in-band and out-band power ratio.

[0077] The in-band power ratio (PRIO) is another signal characteristic parameter that can be considered. The so-called in-band power ratio refers to the ratio between the power value within the signal bandwidth and the power value within the same bandwidth outside the signal bandwidth calculated based on the power spectrum density distribution of the signal. The in-band power ratio of the target signal when the channel noise is superimposed is used as the threshold (PRIO Thr ), the PRIO value of the extracted signal is calculated at the end of the ICA-R algorithm (labeled as PRIOy) and compared with PRIO Thr The relationship between PRIOy and PRIOy is that if PRIOy is greater than PRIO Thr , then restart the optimization process of the ICA-R algorithm until PRIOy is no less than PRIO Thr , or the number of re-optimization times m exceeds the set maximum value m max .

[0078] S14: extracting an estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal;

[0079] In this step, the estimated signal of the target signal is extracted from the input signal using the ICA-R algorithm and the reference signal generated previously. The ICA-R algorithm can achieve blind separation of signals under conditions of unknown signal sources and mixing modes by assuming the independence of the signals. This process aims to maximize the extraction of the target signal while reducing the impact of interference and noise. The specific process can be referred to Figure 4 .

[0080] S15: extracting characteristic parameters from the estimated signal, and determining whether the estimated signal meets the preset requirements according to the characteristic parameters and the characteristic parameter threshold; if not, returning to S14 until the estimated signal meets the preset requirements;

[0081] S16: End.

[0082] Extract characteristic parameters from the estimated signal and compare them with the set characteristic parameter threshold to determine whether the estimated signal meets the preset requirements. This step is the quality control link of signal extraction. Through the analysis of characteristic parameters, it is ensured that the quality of the extracted signal meets the actual application requirements, thereby achieving effective signal verification.

[0083] If the estimated signal does not meet the preset requirements, it will return to step S14 and re-extract the target signal. This feedback mechanism ensures the flexibility and reliability of the signal extraction process. Through multiple iterations and optimizations, the extraction effect of the target signal can be continuously improved until the set quality standard is finally met. This process not only improves the accuracy of signal extraction, but also ensures the application effect in complex electromagnetic environments.

[0084] As a preferred embodiment, after determining that the preset requirements are not met, it also includes: determining whether the number of times the step of extracting an estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal is executed is greater than the set maximum number of optimization searches; if it is greater than the set maximum number of optimization searches, determining that the estimated signal is the target signal; if it is not greater than the set maximum number of optimization searches, re-entering the step of extracting an estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal.

[0085] Among them, the maximum number of optimization times is Figure 5 m max In this preferred embodiment, when it is determined that the estimated signal does not meet the preset requirements, it is further determined whether the number of times the step of extracting the estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal has been executed exceeds the set maximum number of optimization times (m max ). The purpose of this mechanism is to control the use of processing time and resources while ensuring the signal extraction effect. If the number of extraction steps exceeds the set maximum number of optimizations, the current estimated signal will be directly determined as the target signal, although it may not fully meet the preset requirements. This step is intended to prevent excessive iterations from causing waste of resources and time delays, while ensuring that a relatively usable signal can still be obtained within a certain number of times. On the contrary, if the maximum number of optimizations is not exceeded, the extraction step of S14 will be re-entered to continue optimizing the estimated signal. In this way, a balance is achieved between effective extraction of the target signal and resource utilization, ensuring that a quick response and available signal extraction results can be provided in practical applications.

[0086] As a preferred embodiment, the ICA-R algorithm includes a built-in number of iterations. After determining that the number of times the step of extracting an estimated signal of a target signal from an input signal according to the ICA-R algorithm and a reference signal is executed is not greater than a set maximum number of optimization searches, it also includes: resetting the built-in number of iterations of the ICA-R algorithm to 0, and re-entering the step of extracting an estimated signal of a target signal from an input signal according to the ICA-R algorithm and a reference signal.

[0087] Specifically, the ICA-R algorithm has a built-in iteration mechanism to enhance the accuracy and efficiency of signal extraction. After determining that the number of times the extraction step is executed does not exceed the set maximum number of optimizations, the built-in number of iterations (k) of the ICA-R algorithm is reset to 0, thereby restarting the extraction process. Specifically, the ICA-R algorithm undergoes multiple iterations during execution, and each iteration increases the value of k to track the progress of the current iteration. When k reaches its maximum number of iterations, the ICA-R algorithm ends its execution. This design enables the system to run the extraction process from scratch, which helps to improve the accuracy of signal extraction and avoid the problem of insufficient signal extraction due to insufficient iterations. The mechanism of resetting the number of iterations ensures that the algorithm can fully explore possible signal features each time it is re-extracted, thereby optimizing the extraction results. This process is similar to the process of the ICA-R algorithm in the prior art, and this application will not be repeated here to ensure the effective extraction of target signals in complex electromagnetic environments.

[0088] Furthermore, to better illustrate Figure 1 How does the target signal blind extraction module based on ICA-R algorithm use the reference signal r(t) to separate and extract the target signal s(t)? Please refer to Figure 3 and Figure 4 ,in, Figure 3 The block diagram of the target signal blind extraction system based on the ICA-R algorithm provided by the present invention, wherein s(t) is the target signal, j(t) is the interference signal in the input signal, x1(t) and x2(t) are two antenna signals, n1(t) and n2(t) are two noise signals superimposed at the receiver, r(t) is the reference signal, y(t) is the extraction signal, b(n) is the information bit of the transmitting source, b'(n) is the information bit recovered by the receiving end, and A represents the channel mixing matrix. Of course, the representation of the wireless channel is not limited to the use of a matrix, and other implementation methods can also be used, and this application does not make any special restrictions here.

[0089] The full-rank matrix of 2 rows and 2 columns in this embodiment can complete the transmission of signals, and the implementation method is simple and reliable.

[0090] Here is a general description of the principle of the ICA-R (Independent Component Analysis with Reference) algorithm:

[0091] The blind signal extraction method based on ICA-R can well solve the shortcomings of the blind source separation method based on the ICA algorithm. By introducing a reference signal containing prior information of the target signal, the blind source separation method based on the ICA-R algorithm can accurately separate and extract the only target signal. The key to the success of this method is to select a suitable reference signal and initialize the separation vector appropriately. The reference signal must be a non-Gaussian signal containing prior information of the target signal. Through a survey of existing public literature, it can be seen that the methods for generating reference signals mainly include the following: based on the sign function of the signal, based on the statistical characteristics of the target signal, based on the signal frequency, etc. The initialization methods of the separation vector mainly include random initialization methods and non-random initialization methods assisted by prior information.

[0092] Please refer to Figure 4 , Figure 4 This is a schematic diagram of an ICA-R algorithm processing module provided by the present invention. The main idea is to introduce the proximity measure of the output signal y(t) and the reference signal r(t) into the difference function of the traditional ICA, and then complete the learning of the separation vector in a certain cyclic iteration mode. After the separation vector converges, it is multiplied with the observed signal x(t) to achieve the separation and extraction of the target signal s(t). The specific steps are as follows:

[0093] (1) Initialize Lagrange parameters ( ), learning rate ( ), randomly generate separation vectors ( ), and set the convergence criteria;

[0094] (2) Select the proximity metric function ( ), compares the proximity of the output signal y(t) and the reference signal r(t). When the output signal y(t) is the expected communication signal s(t), the proximity metric reaches the minimum value. It can be selected in the following three forms:

[0095] , (Formula 1);

[0096] in, represents mathematical expectation.

[0097] (3) Construct inequality constraints ( ). The proximity between the output signal and the reference signal is controlled within a certain threshold ( ), which is used to distinguish the target signal from other communication signals, and its expression is: , (Formula 2);

[0098] (4) Constructing equality constraints ( ). This constraint can ensure that the separated and extracted signal has unit variance, and its expression is: , (Formula 3);

[0099] (5) Construct the optimization objective function ( ). It can be a difference function based on negative entropy, and its expression is: , (Formula 4);

[0100] in, is a positive constant, is an arbitrary non-quadratic function, and v is a Gaussian variable with zero mean and unit variance.

[0101] (6) Using the augmentation method, the optimization problem under the above constraints is transformed into an optimization problem with Lagrangian parameters ( ) is the extreme value problem of the augmented Lagrangian function. The extreme value is:

[0102] , (Formula 5);

[0103] In the formula and is the Lagrange multiplier corresponding to the two constraints, is the penalty parameter, is the Euclidean norm.

[0104] (7) By solving equation (5) for The extreme value of The iterative learning formula is: , (Formula 6);

[0105] in, and L pairs The first and second order partial derivatives of R zz is the covariance matrix of the whitened observation matrix z, is the learning rate. At the same time, the Lagrange parameter and The iterative update formula is:

[0106] , (Formula 7);

[0107] (8) Iterate formula 6 and formula 7 to determine Whether the convergence criteria are met, if so, stop the loop calculation and save the final separation vector , otherwise continue iterative calculation.

[0108] (9) Through the formula By calculation, the expected independent component y(t) closest to the reference signal r(t) can be directly obtained, that is, the restored sample of the target signal s(t), and the separation and extraction of the target signal is completed.

[0109] In order to ensure the stability and reliability of the signal during the calculation process, this embodiment first centers and whitens the N antenna signals before (1).

[0110] It can be seen that the above calculation method in this embodiment can extract the communication signal from the N antenna signals, and the implementation method is simple and reliable, thereby achieving anti-interference during the communication signal transmission process.

[0111] Specifically, this embodiment aims to provide a specific implementation method for utilizing signal prior information to improve the robustness of the ICA-R algorithm in extracting the target signal. The prior information in this embodiment is not limited to the power spectral density symmetry coefficient (PSI) mentioned in the above implementation mode. As long as it can reflect the uniqueness of the target signal, it is not particularly limited here.

[0112] Please refer to Figure 5 , Figure 5 The present invention provides an algorithm flow chart for extracting a target signal using the prior information of the target signal. The portion of the figure that does not contain a gray background is a flow chart of a conventional ICA-R algorithm, and the portion with a gray background is the innovative improvement of the present invention. Specifically, an estimated value y(t) of the extracted signal is obtained on the basis of the conventional ICA-R algorithm, and then the characteristic parameter value of the signal is calculated and compared with the characteristic parameter threshold of the target signal s(t) under the set signal-to-noise ratio condition. If it is greater than the threshold, it means that the signal estimated by the algorithm this time is not the target signal, and the learning step k of the single optimization of the ICA-R algorithm needs to be reset to 0, so that the algorithm restarts the optimal separation vector. The search task is completed until the characteristic parameter value of the estimated signal y(t) is not less than or equal to the set threshold, or the number of algorithm restarts m exceeds the set maximum value m Max .

[0113] Please refer to Figure 6 , Figure 6 The present invention provides a curve diagram of the target signal extraction accuracy rate versus the number of algorithm re-runs when the threshold is set using PSI as a characteristic parameter, wherein Figure 6 The separation vector ω corresponding to the diagram on the left p Initialized by random generation. Figure 6 The separation vector ω corresponding to the diagram on the right p Initialization is performed using a non-random generation method.

[0114] get Figure 6 The simulation parameters for the experimental results shown are set as follows.

[0115] The target signal is set to a BPSK modulated digital bandpass communication signal, whose carrier frequency is fc1 =40kHz, the signal bandwidth is 16kHz, and the up / down sampling multiple is 16. To avoid inter-symbol interference, a root raised cosine filter with a roll-off factor of 1 is used for shaping filtering before up-conversion, and the same filter is used for matched filtering at the receiving end. The interference signal is set to a Gaussian narrowband noise signal with a center frequency of f c2 =36kHz, signal bandwidth is 4kHz. The power ratio of the two signals, i.e., signal-to-interference ratio, is set to -30dB. Obviously, the spectrum of the interference signal is within the working bandwidth of the communication signal, and the power is much greater than that of the communication signal, so this is a typical partial band suppression interference scenario.

[0116] The channel mixing matrix is ​​set to A = [0.64598 0.93228; 0.93647 0.1598], the receiver sampling rate is set to 160kHz, and after channel noise addition, E b N0 reaches 7dB, which is the demodulation threshold of speech signals transmitted on the additive white Gaussian noise (AWGN) channel. The reference signal is set to a cosine pulse function with the same carrier frequency as the target signal, that is, r(t)=cos(2πf c t), which is easy to occur in actual communication systems.

[0117] Since digitally modulated communication signals are all super-Gaussian signals, the non-quadratic function is selected as , the proximity measurement function adopts the mean square error function, that is, , the feature parameter threshold is set to The Lagrange multipliers and penalty factors are initialized as , The learning rate for the iterative calculation of the separation vector is set to .

[0118] During the iteration process, the following formula is used to determine whether the separation vector has reached the convergence standard: ; In simulation Set to 10 -3 .

[0119] Finally, the proportion of the number of effective transmission source information bit sequences in N transmission tests (accuracy) is used to measure the robustness of the algorithm in separating and extracting the target signal. That is, the accuracy is defined as follows:

[0120] ;

[0121] in , and They are the total number of tests, the number of times the bit error rate does not exceed the theoretical value, and the number of times the bit error rate exceeds the theoretical value. The value range is between 0 and 1. The larger the value, the better the stability of the ICA-R algorithm. According to the theoretical bit error rate calculation formula for BPSK modulated signals transmitted on a Gaussian white noise channel, E b The bit error rate when N0=7dB is 7.73×10 -4 , that is, the number of bit errors in transmitting 10,000 information bits does not exceed 8. Therefore, in the simulation experiment, we use the number of bit errors in transmitting 10,000 information bits not exceeding 8 as the statistical threshold for the bit error rate not exceeding the theoretical value. The value is 50000.

[0122] The threshold of the feature parameter PSI is set to 0.26.

[0123] like Figure 6 It can be seen that when the power spectrum density estimation symmetry coefficient (PSI) is used as the characteristic parameter to set the threshold, regardless of the separation vector ω p Regardless of the initialization method, the accuracy of the ICA-R algorithm in extracting the target communication signal and the number of times the algorithm is re-run are all in an S-shaped exponential function relationship. Figure 6 As shown in the diagram on the left, the separation vector ω p When using random generation for initialization, if the maximum number of times the algorithm is re-run is set to more than 20 times, the accuracy can be increased from 15.11% in a single run to more than 90%, an increase of more than 75 percentage points. p When using a non-random generation method for initialization, if the maximum number of times the algorithm is re-run is set to more than 16 times, the accuracy can be improved from 95.76% in a single run to more than 99%, an improvement of more than 3 percentage points.

[0124] Figure 6 The experimental results show that the signal extraction method based on the ICA-R algorithm provided by the present invention can improve the robustness of the algorithm.

[0125] To solve the above technical problems, Figure 7 As shown, the present invention provides a signal extraction device based on ICA-R algorithm, comprising:

[0126] A memory 71, used for storing computer programs;

[0127] The processor 72 is used to implement the steps of the above-mentioned signal extraction method based on the ICA-R algorithm when executing the computer program. For the introduction of the signal extraction device based on the ICA-R algorithm, please refer to the above-mentioned embodiment, and this application will not repeat it here.

[0128] In order to solve the above technical problems, the present invention provides a non-volatile storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the signal extraction method based on the ICA-R algorithm are implemented. For the introduction of the non-volatile storage medium, please refer to the above embodiment, and this application will not repeat it here.

[0129] It should also be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0130] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A signal extraction method based on ICA-R algorithm, characterized in that: include: receiving an input signal, wherein the input signal includes a target signal and an interference signal; Generate a reference signal according to the carrier frequency of the target signal, wherein the reference signal is a cosine pulse signal having the same carrier frequency as the target signal; Calculating characteristic parameters according to the prior information of the target signal and setting characteristic parameter thresholds; Extracting an estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal; Extracting characteristic parameters from the estimated signal, and determining whether the estimated signal meets preset requirements according to the characteristic parameters and the characteristic parameter threshold; If not, the step of extracting an estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal is repeated until the estimated signal meets the preset requirement.

2. The signal extraction method based on the ICA-R algorithm according to claim 1, characterized in that: Calculating characteristic parameters according to the prior information of the target signal includes: Performing Fourier transform on the target signal to obtain a frequency domain signal; Calculating the power spectral density of the frequency domain signal; A power spectrum density symmetry coefficient is calculated according to the power spectrum density.

3. The signal extraction method based on the ICA-R algorithm as claimed in claim 2, characterized in that: The expression of the power spectrum density symmetry coefficient is: ; Wherein, PSI is the symmetry coefficient of the power spectrum density, M is the number of sampling points in the power spectrum density that deviate from the center frequency, k represents the kth sampling point of the deviation, and Y L (k) is the power spectrum density value of the kth sampling point in the part where the power spectrum density of the target signal is less than the center frequency, Y R (k) is the power spectrum density value of the kth sampling point in the part where the power spectrum density of the target signal is greater than the center frequency, (k) is Y R (k) horizontal flip, Represents the modulus of a vector.

4. The signal extraction method based on the ICA-R algorithm as claimed in claim 2, characterized in that: The expression of the power spectrum density symmetry coefficient is: ; Wherein, PSI is the symmetry coefficient of the power spectrum density, M is the number of sampling points in the power spectrum density that deviate from the center frequency, k represents the kth sampling point of the deviation, and Y L (k) is the power spectrum density value of the kth sampling point in the part where the power spectrum density of the target signal is less than the center frequency, Y R (k) is the power spectrum density value of the kth sampling point in the part where the power spectrum density of the target signal is greater than the center frequency, (k) is Y R (k) horizontal flip, represents the modulus of a vector, Represents the second-order normal form.

5. The signal extraction method based on ICA-R algorithm according to claim 1, characterized in that: Calculating a characteristic parameter threshold according to the prior information of the target signal includes: Perform Fourier transform on the target signal to obtain a frequency domain signal. Calculating the power spectral density of the frequency domain signal; Calculating a first power value within a signal bandwidth according to the distribution of the power spectrum density; Calculating a second power value within the same bandwidth outside the signal bandwidth according to the distribution of the power spectrum density; The ratio between the first power value and the second power value is calculated to obtain an in-band and out-band power ratio.

6. The signal extraction method based on ICA-R algorithm according to claim 1, characterized in that: The characteristic parameter is a power spectrum density symmetry coefficient, and determining whether the estimated signal meets a preset requirement according to the characteristic parameter and the characteristic parameter threshold includes: Determining whether a power spectral density coefficient in the estimated signal is greater than a power spectral density coefficient threshold; If it is greater than the power spectrum density coefficient threshold, it is determined that the estimated signal meets the preset requirement; If it is not greater than the power spectrum density coefficient threshold, it is determined that the estimated signal does not meet the preset requirement.

7. The signal extraction method based on the ICA-R algorithm according to any one of claims 1 to 6, characterized in that: After determining that the preset requirement is not met, the method further includes: Determine whether the number of times the step of extracting the estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal is executed is greater than a set maximum number of optimization searches; If it is greater than the set maximum number of optimization times, determining that the estimated signal is the target signal; If it is not greater than the set maximum number of optimization times, the step of extracting an estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal is re-entered.

8. The signal extraction method based on ICA-R algorithm as claimed in claim 7, characterized in that: The ICA-R algorithm includes a built-in number of iterations, and after determining that the number of times the step of extracting the estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal is performed is not greater than the set maximum number of optimization searches, further includes: The built-in iteration number of the ICA-R algorithm is reset to 0, and the step of extracting the estimated signal of the target signal from the input signal according to the ICA-R algorithm and the reference signal is re-entered.

9. A signal extraction device based on ICA-R algorithm, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the signal extraction method based on the ICA-R algorithm as described in any one of claims 1 to 8 when executing a computer program.

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the signal extraction method based on the ICA-R algorithm as described in any one of claims 1 to 8 are implemented.

Images