Radio astronomical signal radio frequency interference elimination method

Abstract

The invention relates to the technical field of radio astronomy, and discloses a radio astronomy signal radio frequency interference elimination method, which comprises the following steps: S1, a preprocessing step; s2, an independent component analysis step; s3, a signal separation step; and S4, a post-processing step. According to the radio astronomical signal radio frequency interference elimination method, a radio signal is expressed as a linear combination of random variables, and a source signal meets a specific condition; the fast independent component analysis estimates negentropy by maximizing non-Gaussian property, and separates out independent components; a threshold value does not need to be preset in the process, so that the problem of data loss or noise residue caused by improper threshold value selection is avoided; in addition, as the observation data is directly processed through rapid independent component analysis instead of depending on feature selection, the method is more stable when facing a time-varying celestial body such as a pulsar; finally, pulsar signals and radio frequency interference signals are identified and separated by calculating the column standard deviation and the median value of the hybrid matrix A, and generation of high-quality observation results is ensured.

Description

Technical Field

[0001] The invention relates to the technical field of radio astronomy, and in particular to a method for eliminating radio frequency interference of radio astronomy signals. Background Art

[0002] Radio astronomy has become an important means for human beings to explore the universe, especially in the field of pulsar observations. Due to their unique physical properties, pulsars provide an irreplaceable experimental environment for condensed matter physics, quantum chromodynamics, and stellar evolution. However, with the development of modern communication technology, the impact of radio frequency interference (RFI) on radio astronomical observations has become increasingly serious. These interference signals come from multiple sources such as television, FM radio transmissions, GPS, mobile phones, and aircraft navigation communications, which greatly affect the quality of observation data.

[0003] The currently widely used radio frequency interference elimination method based on threshold judgment has significant shortcomings; for example, although it is simple and efficient to identify and discard pixels exceeding the threshold by setting a threshold, it is a difficult problem to accurately determine the threshold, especially when dealing with time-varying celestial bodies such as pulsars, the threshold selection is particularly critical; in addition, this method will lead to a reduction in observation bandwidth and time, reducing observation efficiency; at the same time, although machine learning methods have made some progress, they still face many challenges in practical applications due to their reliance on feature selection and training sample construction. Therefore, we propose a method for eliminating radio frequency interference of radio astronomical signals. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the shortcomings of the prior art, the present invention provides a method for eliminating radio frequency interference of radio astronomy signals, which has the advantages of being able to effectively separate radio frequency interference signals and pulsar signals, and ensuring the integrity and high-quality retention of pulsar signals, and solves the problem that although setting a threshold to identify and discard pixels exceeding the threshold is simple and efficient, how to accurately determine the threshold is a difficult problem, especially when dealing with time-varying celestial bodies such as pulsars, the threshold selection is particularly critical; in addition, this method will lead to a reduction in observation bandwidth and time, reducing observation efficiency; at the same time, although machine learning methods have made certain progress, they still face many challenges in practical applications due to their reliance on feature selection and training sample construction.

[0006] (II) Technical solution

[0007] In order to achieve the above-mentioned method of effectively separating radio frequency interference signals and pulsar signals and ensuring the integrity and high quality retention of pulsar signals, the present invention provides the following technical solutions: a method for eliminating radio frequency interference of radio astronomy signals, comprising the following steps:

[0008] S1, preprocessing step, centering and whitening the observed radio signal to generate a new vector with zero mean and unit covariance;

[0009] S2, independent component analysis step, using the fast independent component analysis algorithm (FastICA) to estimate the negative entropy and separate the independent components by maximizing non-Gaussianity;

[0010] S3, signal separation step, identifying and separating pulsar signals and radio frequency interference signals according to the column standard deviation and median of the mixing matrix A;

[0011] S4, the post-processing step, converts the separated pulsar signals back to the original observation format to generate high-quality observation results.

[0012] Preferably, the pretreatment step further comprises:

[0013] Step 1: Centralization: Subtract the mean vector m=E{x} from the observed vector x to make x have zero mean;

[0014] Step 2: Whitening: By calculating the covariance matrix E{xx T}=EDE T , and obtain the whitened observation vector Where E is the orthogonal matrix of eigenvectors and D is the diagonal matrix of eigenvalues.

[0015] Preferably, the independent component analysis step further comprises:

[0016] Step 1: Establish an independent component analysis model: the observed radio signal is represented as a random variable x, and the source signal is represented as s, satisfying x = As, where A is a mixing matrix;

[0017] Step 2: Estimate independent components: through the objective function:

[0018]

[0019] A non-quadratic function G is selected and the objective function is optimized to obtain the independent component with maximum non-Gaussianity.

[0020] Preferably, the signal separation step further comprises:

[0021] Step 1: Calculate the column standard deviation std of the mixing matrix A j and median med j ;

[0022] Step 2: construct the discriminant vector d = med· / std, and select the column corresponding to the maximum value of the discriminant vector element as the pulsar signal;

[0023] Step 3: Modify the mixing matrix A so that it contains the most complete pulsar signal possible while minimizing radio frequency interference.

[0024] Preferably, the post-processing step further comprises:

[0025] Step 1: By transforming x p =A p s, and obtain the pulsar signal after removing the radio frequency interference;

[0026] Step 2: Correct the mixing matrix A p , so that the residual pulsar signal is included in the recovered signal as completely as possible, while containing as little radio frequency interference as possible.

[0027] A radio astronomy signal radio frequency interference elimination system, comprising:

[0028] A data acquisition module, used to receive the original radio signals observed by the radio telescope;

[0029] A pre-processing module, used to perform centering and whitening on the original radio signal;

[0030] An analysis module for performing independent component analysis to separate the radio frequency interference signal and the pulsar signal;

[0031] A signal separation module, used to identify and separate pulsar signals and radio frequency interference signals;

[0032] A post-processing module is used to convert the separated pulsar signals back to the original observation format.

[0033] Preferably, the analysis module further comprises:

[0034] Fast Independent Component Analysis Unit: used to estimate negative entropy and separate independent components by maximizing non-Gaussianity;

[0035] Objective function calculation unit: used to construct and optimize the objective function:

[0036]

[0037] Select a non-quadratic function G to optimize.

[0038] Preferably, the signal separation module further comprises:

[0039] Mixing matrix calculation unit: used to calculate the column standard deviation std of the mixing matrix A j and median med j ;

[0040] Discriminant vector construction unit: used to construct the discriminant vector d = med· / std, and select the column corresponding to the maximum value of the discriminant vector element as the pulsar signal;

[0041] Mixing matrix correction unit: used to correct the mixing matrix A so that it contains the most complete pulsar signal possible while minimizing radio frequency interference.

[0042] (III) Beneficial effects

[0043] Compared with the prior art, the present invention provides a method for eliminating radio frequency interference of radio astronomy signals, which has the following beneficial effects:

[0044] 1. The radio astronomy signal radio frequency interference elimination method effectively solves the threshold setting problem through the independent component analysis step, especially the fast independent component analysis algorithm; the radio signal is represented as a linear combination of random variables, in which the source signal meets specific conditions; the fast independent component analysis estimates the negative entropy by maximizing the non-Gaussianity and separates the independent components; this process does not require a preset threshold, avoiding data loss or noise residual problems caused by improper threshold selection; in addition, since the fast independent component analysis directly processes the observation data rather than relying on feature selection, the method is more robust when facing time-varying celestial bodies such as pulsars; finally, the pulsar signal and the radio frequency interference signal are identified and separated by calculating the column standard deviation and median of the mixing matrix A, ensuring the generation of high-quality observation results.

[0045] 2. This method of eliminating radio frequency interference of radio astronomy signals converts the separated pulsar signals back to the original observation format, thereby improving the observation efficiency without reducing the observation bandwidth and time; this module uses the corrected mixing matrix to restore the pulsar signal after removing the radio frequency interference to the optimal state, while minimizing the impact of radio frequency interference; specifically, by constructing a discriminant vector and selecting the column corresponding to its maximum value as the pulsar signal, the target signal can be effectively extracted from the complex signal; this method not only overcomes the information loss problem caused by the traditional threshold-based method, but also improves the integrity and clarity of the pulsar signal. Compared with the machine learning method, it reduces the dependence on training samples and enhances the feasibility of practical application. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic flow chart of the radio frequency interference elimination method of the present invention;

[0047] Figure 2 It is a schematic diagram of the structure of the radio frequency interference elimination system of the present invention;

[0048] Figure 3 This is the frequency domain histogram of the pulsar J0332+5434 of the present invention;

[0049] Figure 4 This is a schematic diagram of the J0332+5434 radio frequency interference elimination effect of the present invention;

[0050] Figure 5 This is a schematic diagram of the radio frequency interference elimination effect of the J0332+5434 integrated mean signal of the present invention;

[0051] Figure 6 The contour of the J0332+5434 pulsar after eliminating radio frequency interference using different methods of the present invention;

[0052] Figure 7 The figure is a curve diagram showing the influence of the k value on the signal-to-noise ratio of radio frequency interference elimination according to the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the embodiments of the present invention and the accompanying drawings to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0054] See also Figure 1-7 , a method for eliminating radio frequency interference of radio astronomy signals, comprising the following steps:

[0055] S1, preprocessing step, centering and whitening the observed radio signal to generate a new vector with zero mean and unit covariance;

[0056] S2, independent component analysis step, using the fast independent component analysis algorithm (FastICA) to estimate the negative entropy and separate the independent components by maximizing non-Gaussianity;

[0057] S3, signal separation step, identifying and separating pulsar signals and radio frequency interference signals according to the column standard deviation and median of the mixing matrix A;

[0058] S4, the post-processing step, converts the separated pulsar signals back to the original observation format to generate high-quality observation results.

[0059] A radio astronomy signal radio frequency interference elimination system, comprising:

[0060] A data acquisition module, used to receive the original radio signals observed by the radio telescope;

[0061] A pre-processing module, used to perform centering and whitening on the original radio signal;

[0062] An analysis module for performing independent component analysis to separate the radio frequency interference signal and the pulsar signal;

[0063] A signal separation module, used to identify and separate pulsar signals and radio frequency interference signals;

[0064] A post-processing module is used to convert the separated pulsar signals back to the original observation format.

[0065] Embodiment 1:

[0066] The preprocessing steps of radio astronomy signals include centering and whitening. In implementation, data preprocessing includes centering and whitening. The so-called centering is to process the observed radio signals into zero mean. It can be achieved by subtracting the mean vector m=E{x} from the observation vector x. At this time, taking the expectation on both sides of the formula x=As, it can be seen that the independent component s is also zero mean. The purpose of centering is to simplify the independent component analysis estimation. After estimating the mixing matrix A using the centered radio observation data, the mean vector A of s can be added to the zero mean independent component s, thereby completing the independent component estimation.

[0067] Whitening can reduce the parameters in independent component analysis; whitening is to transform the observed radio signal x into a new vector that is uncorrelated and has unit covariance Make Whitening can be achieved by eigenvalue decomposition of the covariance matrix of x:

[0068] E|xx T ∣=EDE T

[0069] Where E is E|xx T | The orthogonal matrix of the eigenvectors; D is the diagonal matrix of its eigenvalues, D = diag (d1, ..., d n ); note that the formula E|xx T ∣=EDE T The value of can be obtained by radio observation signals x(1),…,x(t) at different times, so the whitening can be written as:

[0070]

[0071] in, It can be verified that: The whitened observation vector Substituting into the formula x = As, we can get:

[0072]

[0073] in, is the new mixing matrix, which is:

[0074]

[0075] It can be seen that after whitening, the new mixing matrix It is an orthogonal matrix, so it has only n(n―1) / 2 degrees of freedom. If whitening is not performed, it is necessary to estimate the n of the mixing matrix A. 2 parameters.

[0076] Embodiment 2:

[0077] The present invention uses an independent component analysis model to estimate a mixing matrix A (or an unmixing matrix W) from a radio observation signal x, and then obtain a source signal s.

[0078] Depend on Figure 3 As shown, we can find that: (1) The data columns are aligned in phase, so each row of the pulsar observation signal (image) x can be regarded as a phase change signal: x1(t),…,x n (t), the estimated source signal vector s is also composed of the phase change signals s1(t),…,s m To ensure that A is a full-rank matrix, n ≥ m is required; (2) Compared with the RF interference signal, the pulsar signal has a strong broadband characteristic, that is, for the pulsar signal, there is a stronger correlation between the data rows; In addition, it can be seen from formula (2) that the row element of the matrix A is essentially the source signal s1(t),…,s m (t) group is mixed into the i-th frequency channel observation signal x i The jth column element corresponds to the weight of mixing the jth source signal s,(t) into the observation signals x1(t),…,x n (t) weight; From the meaning of each element of the mixing matrix A and the above discovery (2), we can know that if s j (t) corresponds to the pulsar signal, then the column of matrix A will show the characteristics of uniform distribution, and the variance or standard deviation can be used to measure the degree of uniform distribution of random variables; in addition, in practice, the number of RF interference sources is unpredictable, and generally greater than the number of RF interference sources, which will cause the standard deviation of some columns of matrix A to be very small and the median to be very small. At this time, these columns correspond to RF interference signals; according to the above analysis, the discriminant vector d can be obtained by dividing the standard deviation vector std by the median vector med, that is, the larger the median value of a column, the smaller the standard deviation, and the larger the value of the corresponding discriminant vector element:

[0079] d=med· / std

[0080] The standard deviation vector and median vector are respectively the standard deviations of each column of matrix A, std1,…,std m and the median meed1,...,medn, that is:

[0081]

[0082] Among them, μ j and N jRespectively represent the mean and number of elements in the j-th column of matrix A; median(.) represents the median function; if the j-th p Source signal Corresponding to the pulsar signal, the remaining s j (t)(j≠j p ) corresponds to the radio frequency interference signal; from the above analysis, it can be seen that the source signal corresponding to the column where the maximum value of the discriminant vector element is located should be selected as the pulsar signal; that is, j p for:

[0083] j p = argmax(d j ), j = 1, ..., m

[0084] Among them, d1,…,d m Constitute the discriminant vector d; Note that at this time Contains pulsar signal information for each frequency channel, not Figure 1 The form of display; From the meaning of each element of the mixing matrix A, it can be known that the following transformation can be used to obtain Figure 1 The pulsar signal after removing the RF interference is presented in the form of:

[0085] X p =A p s

[0086] Among them, the matrix A p It is obtained by retaining the elements of the Ath column of the mixing matrix A and setting the remaining column elements to 0.

[0087] Embodiment three:

[0088] Since FastlCA is an approximate solution for negative entropy, a small amount of pulsar signals are contained in other independent components s j (t)(j≠j p ) in the matrix A p Further correction makes the recovered signal x p Include the pulsar signal as completely as possible, while including as little radio frequency interference as possible; note that the residual pulsar signal is more evenly distributed in terms of frequency and pulse phase than radio frequency interference; experimentally found that the k-fold matrix A can be removed by removing the jth p The standard deviation after the column is used as the threshold, and the matrix A satisfies:

[0089]

[0090] Add the elements of A p In the middle, to correct A p ; Corrected A p By formula:

[0091] X p =Ap s

[0092] The pulsar signal can be restored relatively completely; formula:

[0093]

[0094] middle It represents the standard deviation of the matrix A after removing the elements in the jpth column; if the coefficient k is selected too small, a small amount of pulsar signals will be treated as radio frequency interference and eliminated, and if it is too large, the radio frequency interference will not be eliminated completely; it should be pointed out that the algorithm of the present invention is not sensitive to the k value, and taking 0.5≤k≤2.5 can obtain a pulsar integrated profile with a signal-to-noise ratio greater than 50, which meets the needs of improving pulsar observations; under the principle of eliminating radio frequency interference and keeping the pulsar signal as much as possible, k=1 is selected.

[0095] Embodiment 4:

[0096] In order to better understand the specific implementation of the present invention, the following is a detailed description with reference to the accompanying drawings and examples.

[0097] The experimental data of this embodiment are derived from the daily observation results of S-band pulsars by the 40-meter radio telescope of Yunnan Observatory; the observation terminal is the fourth-generation pulsar digital filter bank (PDFB4) introduced from the Commonwealth Scientific and Industrial Research Organization of Australia; PDFB4 performs incoherent de-dispersion and periodic profile folding processing on the observation data, with a sampling rate of 64μs and an observation center frequency of 2256MHz; the actual observation frequency band is 2170-2310MHz, and various types of interference that do not cause system saturation still exist within this bandwidth.

[0098] In order to verify the effect and feasibility of the present invention, the method is used to process the observation data of the J0332+5434 pulsar; the more representative observation data of the pulsar in the S-band daily observation in 2017 are selected; Figure 4 (a) shows very clearly that the observed interference frequencies are 2231~2241MHz, 2252~2258MHz, 2267~2273MHz, and 2300~2308MHz; the second and third interference frequencies are characterized by strong interference signals and long duration, while the first interference frequency is very weak. Figure 4 (a) is relatively obvious, the fourth interference frequency point shows uncertainty in phase. The experimental results are as follows Figure 4 (b) It can be seen from the figure that the radio frequency interference is basically eliminated, and only the PSRJ0332+5434 pulsar signal is retained; Figure 4 (c) shows Figure 4 The difference signal between (a) and (c) is the RF interference signal.

[0099] Through experiments, we can also conduct comparative analysis and discussion on the principal component analysis (PCA) method which is similar to independent component analysis in principle. Independent component analysis and principal component analysis belong to factor analysis, both of which maximize certain characteristics through linear combination. The principal component analysis method seeks to maximize the variance, while the independent component analysis seeks to maximize the non-Gaussianity. The principal component analysis method finds out the irrelevant parts of the signal, which corresponds to the second-order statistical analysis, and the independent component analysis finds out the independent parts of the signal, which corresponds to the high-order statistical analysis. However, the principal component analysis and independent component analysis have different uses. The principal component analysis method is currently a common method for data dimensionality reduction. If you only care about the energy or variance of the data, assuming that the interference signal is relatively weak, the principal component analysis method can be used to separate the main signal. However, in radio frequency interference elimination, strong and weak interferences are often mixed together, for example, Figure 4 (a) In the pulsar observation signal shown, the interference at 2252~2258MHz and 2267~2273MHz is obviously stronger than the pulsar signal, while the interference signal at 2231~2241MHz is much weaker than the pulsar signal, and the interference signal at 2300~2308MHz is close to the pulsar signal in intensity; using the principal component analysis method, if the signal strength is measured, it is difficult to determine which principal components correspond to radio frequency interference and which correspond to radio signals; in a sense, independent component analysis is smarter, it does not care about the energy or variance of the signal, but only looks at independence; given the mixed signal to be decomposed, any linear transformation will not affect the output result of the independent component analysis, but will seriously affect the result of the principal component analysis; of course, if other features besides energy can be found, and under this feature dimension, the radio signal can become the principal component and the radio frequency interference can become the secondary component, then the principal component analysis method can be used to decompose the radio signal and the radio frequency interference.

[0100] In practical applications, taking the average of the continuous observation signals of pulsars is an effective and commonly used method to eliminate radio frequency interference; Figure 5 (a) shows the result of taking the average of 96 consecutive sub-integral observation signals of the PSRJ0332+5434 pulsar. It can be clearly seen in the figure that while the pulsar signal is enhanced, the radio frequency interference signal is also enhanced. Figure 5 (b) is the result of eliminating radio frequency interference from 96 sub-integrated observation signals respectively by the method of the present invention, and then calculating the average value. It can be seen from the figure that the radio frequency interference signal is basically eliminated while the pulsar signal is well preserved; Figure 5 (c) Figure 5 The difference signal between (a) and (b), i.e., the RF interference signal after taking the average of the 96 sub-integrated observation signals; Figure 5(a) to (c) show the obvious effect of the method of the present invention on eliminating radio frequency interference in pulsar observation signals; however, upon closer inspection, Figure 5 (c) shows a small amount of pulsar signal residue; this is mainly because FastICA is an approximate solution to negative entropy, which makes it difficult to distinguish the independent components corresponding to the 2231-2241MHz weak radio frequency interference signal and the small amount of residual pulsar signal; but compared with the currently commonly used mean method, the method of the present invention significantly improves the pulsar observation signal; this effect can also be seen from Figure 6 obtained from the pulsar profile.

[0101] In order to more intuitively reflect the effectiveness of the method of the present invention, Figure 6 Shows the Figure 5 (a), (b) Pulsar profile after integration along the frequency domain; where, Figure 6 (a) is the integral contour diagram of the pulsar signal after eliminating radio frequency interference using the mean value method; Figure 6 (b) is the formula When k=1, the integrated contour diagram of the pulsar signal after the radio frequency interference is eliminated by the method of the present invention; the three-peak structure of the PSRJ0332+5434 pulsar can be clearly seen in both contour diagrams, but by comparing the elimination results of the mean value method, it can be seen that the method of the present invention effectively eliminates the radio frequency interference signal in the observed signal, and the pulsar signal is relatively completely preserved; in addition, the pulsar profile signal-to-noise ratio (SNR) is defined as:

[0102]

[0103] It can be calculated Figure 6 (a) Figure 6 The signal-to-noise ratios of (b) are 17.69 and 60.28 respectively; p , and W eq They are the mean, standard deviation and outline width of the non-pulse part of the pulsar outline diagram, respectively.

[0104] Figure 7 The formula is shown When k is 0-3, the method of the present invention is used to eliminate the radio frequency interference of 96 consecutive sub-integral observation signals of the PSRJ0332+5434 pulsar and calculate the average value. It can be seen from the figure that when k=0.69, the signal-to-noise ratio reaches a peak of 60.37 and then begins to decay. The reason is analyzed. When k is small, the radio frequency interference elimination result is calculated by the formula The contribution of the introduced residual pulsar signal to the improvement of the signal-to-noise ratio is greater than the impact of the introduced radio frequency interference on the signal-to-noise ratio, so the signal-to-noise ratio increases; but with the increase of k, the negative impact of excessive introduction of weak radio frequency interference increases, resulting in a decrease in the signal-to-noise ratio; it is worth noting that taking 0.5≤k≤2.5 can obtain a pulsar integral profile with a signal-to-noise ratio greater than 50; through the formula Including the complete pulsar signal as much as possible in the RF interference elimination result is a mutually balanced process. Therefore, when implementing it, under the principle of eliminating RF interference and maintaining the pulsar signal as much as possible, the k value when the signal-to-noise ratio reaches the peak is not selected, but k=1 is selected; compared with the currently commonly used method of taking the average of the continuous observation signal of the pulsar to improve the signal-to-noise ratio, the method of the present invention has a significant effect on improving the signal-to-noise ratio; this effect is Figures 4 to 6 This is also reflected in the experimental results.

[0105] In summary, the radio astronomy signal radio frequency interference elimination method effectively solves the threshold setting problem through the independent component analysis step, especially the fast independent component analysis algorithm; the radio signal is represented as a linear combination of random variables, in which the source signal meets specific conditions; the fast independent component analysis estimates the negative entropy by maximizing the non-Gaussianity and separates the independent components; this process does not require a preset threshold, avoiding data loss or noise residual problems caused by improper threshold selection; in addition, since the fast independent component analysis directly processes the observation data instead of relying on feature selection, the method is more robust when facing time-varying celestial bodies such as pulsars; finally, the pulsar signal and the radio frequency interference signal are identified and separated by calculating the column standard deviation and median of the mixing matrix A, ensuring the generation of high-quality observation results.

[0106] In addition, the radio astronomy signal radio frequency interference elimination method converts the separated pulsar signal back to the original observation format, thereby improving the observation efficiency without reducing the observation bandwidth and time; this module uses the modified mixing matrix to restore the pulsar signal after removing the radio frequency interference to the optimal state, while minimizing the influence of radio frequency interference; specifically, by constructing a discriminant vector and selecting the column corresponding to its maximum value as the pulsar signal, the target signal can be effectively extracted from the complex signal; this method not only overcomes the information loss problem caused by the traditional threshold-based method, but also improves the integrity and clarity of the pulsar signal. Compared with the machine learning method, it reduces the dependence on training samples, enhances the feasibility of practical application, and solves the problem of identifying and discarding pixels exceeding the threshold by setting a threshold. Although it is simple and efficient, how to accurately determine the threshold is a difficult problem, especially when dealing with time-varying celestial bodies such as pulsars, the threshold selection is particularly critical; in addition, this method will lead to a reduction in observation bandwidth and time, reducing the observation efficiency; at the same time, although the machine learning method has made some progress, it still faces many challenges in practical applications due to its dependence on feature selection and training sample construction.

[0107] The relevant modules involved in this system are all hardware system modules or functional modules that combine computer software programs or protocols with hardware in the prior art. The computer software programs or protocols involved in the functional modules are themselves technologies that are well known to those skilled in the art and are not improvements of this system. The improvements of this system are the interaction or connection relationships between the modules, that is, improvements to the overall structure of the system to solve the corresponding technical problems to be solved by this system.

[0108] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for eliminating radio frequency interference of radio astronomy signals, characterized in that: The following steps are involved: S1, preprocessing step, centering and whitening the observed radio signal to generate a new vector with zero mean and unit covariance; S2, independent component analysis step, using the fast independent component analysis algorithm (FastICA) to estimate the negative entropy and separate the independent components by maximizing non-Gaussianity; S3, signal separation step, identifying and separating pulsar signals and radio frequency interference signals according to the column standard deviation and median of the mixing matrix A; S4, the post-processing step, converts the separated pulsar signals back to the original observation format to generate high-quality observation results.

2. A method for eliminating radio astronomy signal radio frequency interference according to claim 1, characterized in that: The pre-processing step further comprises: Step 1: Centralization: Subtract the mean vector m=E{x} from the observed vector x to make x have zero mean; Step 2: Whitening: By calculating the covariance matrix E{xx T }=EDE T , and obtain the whitened observation vector Where E is the orthogonal matrix of eigenvectors and D is the diagonal matrix of eigenvalues.

3. The method for eliminating radio astronomy signal radio frequency interference according to claim 1, characterized in that: The independent component analysis step further comprises: Step 1: Establish an independent component analysis model: the observed radio signal is represented as a random variable x, and the source signal is represented as s, satisfying x = As, where A is a mixing matrix; Step 2: Estimate independent components: through the objective function: A non-quadratic function G is selected and the objective function is optimized to obtain the independent component with maximum non-Gaussianity.

4. The method for eliminating radio frequency interference of radio astronomy signals according to claim 1, characterized in that: The signal separation step further comprises: Step 1: Calculate the column standard deviation std of the mixing matrix A j and median med j ; Step 2: construct the discriminant vector d = med· / std, and select the column corresponding to the maximum value of the discriminant vector element as the pulsar signal; Step 3: Modify the mixing matrix A so that it contains the most complete pulsar signal possible while minimizing radio frequency interference.

5. The method for eliminating radio astronomy signal radio frequency interference according to claim 1, characterized in that: The post-processing step further comprises: Step 1: By transforming x p =A p s, and obtain the pulsar signal after removing the radio frequency interference; Step 2: Correct the mixing matrix A p , so that the residual pulsar signal is included in the recovered signal as completely as possible, while containing as little radio frequency interference as possible.

6. A radio astronomy signal radio frequency interference elimination system, characterized in that: include: A data acquisition module, used to receive the original radio signals observed by the radio telescope; A pre-processing module, used to perform centering and whitening on the original radio signal; An analysis module for performing independent component analysis to separate the radio frequency interference signal and the pulsar signal; A signal separation module, used to identify and separate pulsar signals and radio frequency interference signals; A post-processing module is used to convert the separated pulsar signals back to the original observation format.

7. The radio astronomy signal radio frequency interference elimination system according to claim 1, characterized in that: The analysis module further comprises: Fast Independent Component Analysis Unit: used to estimate negative entropy and separate independent components by maximizing non-Gaussianity; Objective function calculation unit: used to construct and optimize the objective function: Select a non-quadratic function G to optimize.

8. The radio astronomy signal radio frequency interference elimination system according to claim 1, characterized in that: The signal separation module further comprises: Mixing matrix calculation unit: used to calculate the column standard deviation std of the mixing matrix A j and median med j ; Discriminant vector construction unit: used to construct the discriminant vector d = med / std, and select the column corresponding to the maximum value of the discriminant vector element as the pulsar signal; Mixing matrix correction unit: used to correct the mixing matrix A so that it contains the most complete pulsar signal possible while minimizing radio frequency interference.

Images