One-dimensional and two-dimensional synthetic aperture radiometer fusion RFI positioning method based on covariance matrix

Abstract

The invention relates to a two-dimensional synthetic aperture radiometer fusion RFI positioning method based on a covariance matrix, and the method comprises the steps: a receiving end synchronously obtains a two-dimensional antenna array measurement covariance matrix, and carries out the fusion processing of the covariance matrix and the two-dimensional antenna array measurement covariance matrix to obtain a new covariance matrix; performing characteristic decomposition on the new covariance matrix, and dividing a signal subspace and a noise subspace according to characteristic value distribution characteristics; and generating a pseudo spectrum, and obtaining the azimuth estimation of the multi-radio frequency interference source through a spectrum peak. Different from a traditional strategy of fusion after single-load positioning, the measurement visibility of the one-dimensional antenna array and the two-dimensional antenna array is directly fused, the spatial characteristics of the RFI source can be more comprehensively captured, and main lobe interference and side lobe interference are distinguished through complementarity of direction response. Meanwhile, information of different dimensions is fused, and improvement of spatial resolution is achieved.

Description

TECHNICAL FIELD

[0001] The application relates to a one-dimensional synthetic aperture radiometer and a two-dimensional synthetic aperture radiometer, and relates to a high-precision RFI positioning method through fusion of a covariance matrix, which is suitable for carrying out RFI positioning data processing by using a synthetic aperture radiometer and belongs to the technical field of satellite microwave ocean remote sensing. BACKGROUND

[0002] The ocean salinity satellite adopts a new one-dimensional and two-dimensional synthetic aperture microwave radiometer synchronous measurement system, and the salinity joint detection precision is expected to reach 0.1 psu.

[0003] Since the satellite was first observed in orbit, multiple violations of radio frequency interference (RFI) have been detected. Locating these RFIs is an important basis for shutting them down. The traditional single-load super-resolution RFI positioning algorithm uses the MUSIC algorithm, which belongs to the subspace decomposition algorithm. The orthogonality of the signal and noise subspaces is used to separate the RFI source direction bit through eigenvalue decomposition.

[0004] The existing single-load MUSIC algorithm (whether one-dimensional or two-dimensional) has defects that cannot be compensated for by optimization: in a complex electromagnetic environment, relying on a single load severely limits the imaging quality and the accuracy of RFI positioning, and a low snapshot number further limits the effectiveness of the traditional single-load MUSIC algorithm. SUMMARY

[0005] The technical problem solved by the application is to overcome the shortcomings of the prior art and provide a one-two-dimensional synthetic aperture radiometer fusion RFI positioning method based on a covariance matrix.

[0006] The technical solution of the application is:

[0007] The one-two-dimensional synthetic aperture radiometer fusion RFI positioning method based on the covariance matrix comprises:

[0008] The receiving end synchronously acquires one-dimensional and two-dimensional antenna array measurement covariance matrices, and obtains a new covariance matrix after fusion processing of the two matrices;

[0009] Eigenvalue decomposition is performed on the new covariance matrix, and a signal subspace and a noise subspace are divided according to the eigenvalue distribution characteristics;

[0010] A pseudo-spectrum is generated, and the direction estimation of multiple radio frequency interference sources is obtained through the spectrum peak.

[0011] Preferably, the receiving end synchronously acquires one-dimensional and two-dimensional antenna array measurement covariance matrices, and obtains a new covariance matrix after fusion processing of the two matrices, and the method is as follows:

[0012] (1) the receiving end synchronously acquires the receiving signals of the one-dimensional antenna array and the two-dimensional antenna array, performs complex correlation operation on the receiving signals of the one-dimensional antenna array to obtain a measurement covariance matrix R1, and performs complex correlation operation on the receiving signals of the two-dimensional antenna array to obtain a measurement covariance matrix R2;

[0013] (2) the new covariance matrix R is obtained after fusion processing of the two; wherein a is the number of antennas of the one-dimensional antenna array, b is the number of antennas of the two-dimensional antenna array, the covariance matrix R is an (a+b)×(a+b) matrix, c represents a weight, and c is equal to the ratio of the sensitivity of the two-dimensional antenna array to the sensitivity of the one-dimensional antenna array.

[0014] Preferably, the element R(i,j) of R1 or R2 satisfies

[0015] i=j is autocorrelation, and i≠j is cross-correlation;

[0016] x i represents the output signal of antenna i, x j * represents the conjugate of the output signal of antenna j.

[0017] V(u ij ,v ij ) represents the visibility sample obtained by performing cross-correlation on the output signals of antenna i and antenna j.

[0018] Preferably, x i (t) represents the output signal of antenna i at time t, x j * (t) represents the conjugate of the output signal of antenna j at time t.

[0019] Preferably, the new covariance matrix is subjected to eigenvalue decomposition, and the signal subspace and the noise subspace are divided according to the eigenvalue distribution characteristics, and the specific method is as follows:

[0020] (1) the new covariance matrix R is subjected to eigenvalue decomposition to obtain eigenvalues and eigenvectors;

[0021] (2) the eigenvalues are arranged in descending order, a sliding window M is set, the value range of M is 1 to N, N is the total number of antennas of the one-dimensional antenna array and the two-dimensional antenna array, all eigenvalues are traversed through the sliding window, and the slopes of adjacent eigenvalues in each sliding window {Δλ k ,...,Δλ k+M-1} are calculated;

[0022] (3) the variance C r of each sliding window is calculated by using the slopes of adjacent eigenvalues in each sliding window, and the variances of all sliding windows are queued from large to small, and the last 25% C rAs a possible noise subspace, the mean μ of the possible noise subspace is calculated noise and the standard deviation σ noise ;

[0023] (4) The threshold K is calculated according to the mean μ of the possible noise subspace noise and the standard deviation σ noise ; r , K r = μ noise + 3σ noise ;

[0024] (5) Starting from the first sliding window, the variance of the sliding window is sequentially judged in order with the threshold K r , and the first sliding window satisfying the following condition is found: the variance of the sliding window is less than the threshold K r ;

[0025] (6) Assuming that the found sliding window is the kth sliding window, the demarcation point S' of the real signal subspace and the noise subspace is M(k-1).

[0026] Preferably, the signal subspace U s′ =[u1,u2,…,u S′ ] is composed of the eigenvectors corresponding to the first S' eigenvalues arranged in descending order, and the noise subspace U n =[u S′+1 ,u S′+2 ,…,u N ] is composed of the eigenvectors corresponding to the S'+1, S'+2, …, N eigenvalues arranged in descending order, and the two subspaces are independent and orthogonal to each other.

[0027] Preferably,

[0028]

[0029] Λ s′ , Λ n are diagonal matrices, and the diagonals are eigenvalues.

[0030] Preferably, the generated pseudo-spectrum is as follows:

[0031]

[0032] U n represents the noise subspace, and a(ξ,η) is the signal steering vector from the direction (ξ,η).

[0033] Preferably, the (ξ,η) corresponding to the peak value of P M (ξ,η) is the DOA estimation of the target signal.

[0034] Preferably, if in the case of multiple radio frequency interference sources, according to P M S peaks of (ξ,η) are used to estimate the azimuth of S RFI sources.

[0035] In order to fully utilize the advantages of multiple loads, the application proposes a one-dimensional and two-dimensional synthetic aperture radiometer fusion RFI positioning method of covariance matrix, which is a multi-load fusion MUSIC algorithm that synchronously processes one-dimensional and two-dimensional antenna arrays and has a variable threshold to reduce the influence of interference on angle of arrival estimation. Unlike the traditional strategy of fusing after single-load positioning, the application directly fuses the measurement visibility of one-dimensional and two-dimensional antenna arrays, can more comprehensively capture the spatial characteristics of RFI sources, and utilizes the complementarity of directional response to distinguish main lobe and side lobe interference. The application has the beneficial effects compared with the prior art:

[0036] 1. Improved spatial resolution

[0037] The method provided by the application can realize super-resolution. After merging one-dimensional and two-dimensional data, the number of elements of the receiving antenna increases, and the dimension of the signal subspace also increases. The resolution of the MUSIC algorithm is closely related to the dimension of the signal subspace. After increasing the dimension, the algorithm can more accurately estimate the direction of the signal source. This enables the MUSIC algorithm to distinguish multiple signal sources in a smaller angular interval, thereby improving the spatial resolution. The acceleration improves the ocean salinity detection satellite to reach 0.1 psu for the first time in the international community, laying a solid technical foundation for the commercialization of the salinity detection satellite.

[0038] 2. Stronger anti-interference capability

[0039] The traditional multi-load positioning fusion method needs to independently calculate the positioning results of each load, and then fuse the targets after the positioning process. This method not only has high computational complexity, but also cannot avoid the influence of interference. On the contrary, when high noise appears, multiple independent loads are inevitably disturbed by noise, thus introducing additional errors. The application directly fuses the measurement visibility of one-dimensional and two-dimensional antenna arrays, combines spatial information of different dimensions, and can more comprehensively capture the spatial characteristics of RFI sources. A one-dimensional array (such as a linear array) is sensitive to a specific direction (such as azimuth), while a two-dimensional array (such as a planar array) can simultaneously perceive azimuth and elevation. After combining the covariance matrices of the two, the MUSIC algorithm can utilize the complementarity of directional response to distinguish main lobe and side lobe interference through multi-dimensional signal feature fusion. The recognition and suppression capability of noise is improved. The new covariance matrix enhances the discrimination degree of signal and noise by fusing data of different dimensions, especially when the direction of noise and target signal is close, the signal can be more effectively distinguished.

[0040] 3. Stronger robustness, suppresses the influence of noise fluctuation through statistical principles

[0041] The application can adapt to different noise levels by setting a dynamic threshold. In a low noise scene: the noise variance is small, and the threshold is automatically reduced to avoid missing weak signals. In a high noise scene: the noise variance increases, and the threshold rises to suppress false alarms. At the same time, it can adapt to different signal strengths. Strong signal: the signal region Cr is significantly larger than the threshold, which can realize accurate detection of the signal. Weak signal: even if the signal Cr is close to the upper limit of noise fluctuation, according to the properties of normal distribution, the 3σ interval can still distinguish the weak signal.

[0042] 4. The number of snapshots is not high, and the calculation complexity is low

[0043] Compared with the traditional two-dimensional single load MUSIC algorithm, the application only adds a one-dimensional covariance matrix, and the threshold calculation is also more convenient, which is suitable for the scene of remote sensing satellite that needs real-time processing. At the same time, it can still distinguish the relatively close RFI source under the premise of single snapshot. One-dimensional and two-dimensional arrays capture instantaneous information of the same scene from different spatial perspectives, and joint processing is equivalent to multi-baseline interference, which improves the scene reconstruction capability of single snapshot. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 The synchronous processing of one-dimensional and two-dimensional antenna arrays and variable threshold multi-load fusion MUSIC algorithm flowchart for embodiment 1 of the application is provided.

[0045] Figure 2 The comprehensive aperture microwave radiometer complex correlation operation principle diagram used in embodiment 1 of the application.

[0046] Figure 3 The brightness temperature inversion diagram generated in the non-interference case of embodiment 1 of the application is used for comparison.

[0047] Figure 4 The brightness temperature inversion diagram generated in the interference case of embodiment 1 of the application is used for comparison.

[0048] Figure 5 The result diagram obtained after using the traditional single load MUSIC algorithm in embodiment 1 of the application.

[0049] Figure 6 The effect diagram obtained after using the synchronous processing of one-dimensional and two-dimensional antenna arrays and variable threshold multi-load fusion MUSIC algorithm of the application in embodiment 1 of the application. DETAILED DESCRIPTION

[0050] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application. In addition, the technical features involved in the various embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0051] Embodiment 1

[0052] As shown in the following steps: Figure 1

[0053] S1, the receiving end synchronously acquires a two-dimensional antenna array measurement covariance matrix, and obtains a new covariance matrix after fusion processing.

[0054] (1) The receiving end synchronously acquires the received signal of a two-dimensional antenna array, performs complex correlation operation on the received signal of a one-dimensional antenna array to obtain a measurement covariance matrix R1, and performs complex correlation operation on the received signal of a two-dimensional antenna array to obtain a measurement covariance matrix R2;

[0055] (2) After fusion processing, a new covariance matrix R is obtained, wherein Where a is the number of antennas of the one-dimensional antenna array, b is the number of antennas of the two-dimensional antenna array, then the one-dimensional antenna array and the two-dimensional antenna array are a×a matrix and b×b matrix respectively, the covariance matrix R is (a+b)×(a+b) matrix, c represents the weight, and c is equal to the ratio of the sensitivity of the two-dimensional antenna array to the sensitivity of the one-dimensional antenna array. The one-dimensional array with higher sensitivity is allocated a larger weight.

[0056] The element R(i,j) of R1 or R2 satisfies

[0057] i=j is autocorrelation, and i≠j is cross-correlation;

[0058] x i represents the output signal of antenna i, x j * represents the conjugate of the output signal of antenna j;

[0059] V(u ij ,v ij ) represents the visibility sample obtained by cross-correlation of the output signals of antenna i and antenna j.

[0060] The new covariance matrix is mainly obtained by receiving the thermal radiation signals of the target unit area by the satellite antennas respectively, and obtaining the visibility sample V(u,v) by performing complex correlation operation on the received signals.

[0061] In the synthetic aperture microwave radiometer, the specific implementation process of complex correlation sampling is:​

[0062] First, the output signals of the two antennas are respectively correlated with the same direction and the orthogonal direction, which are respectively: directly correlating the received signals of the two antennas to obtain the real part of the visibility function. The received signal of one of the antennas is phase-shifted by π / 2, and then correlated with the signal of the other antenna to obtain the imaginary part of the visibility function. The combination of the two is the complex correlation sampling of the synthetic aperture microwave radiometer.

[0063] After derivation, the expression of visibility under general conditions can be obtained as:

[0064]

[0065] where Ω i ,Ω j represent the solid angles of the receiving antennas i and j, θ, φ represent the zenith angle and azimuth angle of the signal source relative to the receiving antennas i and j, T B (θ, φ) represents the brightness temperature of the target unit area. F ni (θ, φ), F nj (θ, φ) represents the normalized voltage pattern of the two receiving antennas i and j in the correlation interferometer, r i (θ, φ) represents the propagation distance of the signal source (corresponding to the zenith angle θ and azimuth angle φ direction) to the receiving antenna i, r j (θ, φ) represents the propagation distance of the signal source (corresponding to the zenith angle θ and azimuth angle φ direction) to the receiving antenna j, f0 represents the center frequency of the signal, Δr represents the difference of the distance to the receiving antennas i and j. s" represents the unit area of the target.

[0066] represents the stripe removal function:

[0067]

[0068] H i (f) and H j (f) represent the frequency responses of antennas i and j, respectively, G i represents the maximum gain of channel i, B i represents the equivalent noise bandwidth of channel i.

[0069] Under far-field conditions, the same signal source reaching different antennas will produce a wave path difference, which will cause a phase difference. When the distance between the antennas is not greater than λ / 2, the phase difference caused by the distance d between the antennas is 2πdsinθ / λ, and the received signal envelope is: The expression of the signals emitted by multiple signal sources received by one antenna can be represented as:

[0070]

[0071] x l (t): the time-domain signal output by the lth receiving antenna at time t.

[0072] S: the total number of signal sources, used to determine the upper limit of summation, indicating the superposition of signals from S signal sources; s i (t): the time-domain signal emitted by the ith signal source, which is the original signal form without the influence of propagation phase, etc. ω0: the angular frequency of the signal. τ: the propagation delay from the signal source to the corresponding receiving antenna, representing the time delay caused by the difference in propagation path from the transmitting source to the receiving end. l (t): the time-domain noise signal at the lth receiving antenna, representing the interference introduced during the receiving process.

[0073] Assume there are S signal sources and N receiving antennas, the wavelength of the S signal sources is λ, the operating frequency is f, and the second assumption is that the signal is narrowband.

[0074] The output of each antenna can be expressed in matrix form as:

[0075]

[0076] x m (t)(m = 1, 2,..., N): the time-domain signal output by the mth receiving antenna at time t. N: the total number of receiving antennas, used to determine the number of matrix rows (output signal vector dimension), i.e. N: = the total number of receiving antennas (one or two dimensions). τ mn (m = 1,..., N; n = 1,..., S): the propagation delay from the nth signal source to the mth receiving antenna, representing the difference in propagation path from different signal sources to different receiving antennas. n (t)(n = 1, 2,..., S): the time-domain signal emitted by the nth signal source. m (t)(m = 1, 2,..., N): the time-domain noise signal at the mth receiving antenna, which is an element of the noise vector.

[0077] The N-element antenna receiving vector can be expressed as:

[0078]

[0079] The signals emitted by the S RFI sources are represented as s = [s(ξ1, η1), s(ξ2, η2),..., s(ξ S ,η S )], representing the collision signal vector in the (ξ, η) direction, where (ξ, η) = (sinθcosφ, sinθsinφ)) represents the direction cosine relative to the X and Y axes, n = [n1, n2,..., n N ]T is the noise vector at the receiver, a(ξ i ,η i ) is the signal steering vector from direction (ξ i ,η i ), and A is the steering matrix composed of the different steering vectors. The size of the steering matrix is equal to the number of antenna elements (N rows) times the number of signal sources (S columns). The steering vector for the mth RFI signal is given by:

[0080]

[0081] where (X n ,Y n ) are the coordinates of the nth antenna, and λ is the wavelength.

[0082] To compute the MUSIC spatial spectrum, the covariance matrix is first obtained. It has been shown that the visibility samples of a salinity satellite are equivalent to the elements of the covariance matrix. Both are obtained by correlating the outputs of each pair of antennas, so the covariance matrix elements can be obtained by:

[0083]

[0084] i = j is the autocorrelation and i ≠ j is the cross-correlation. x i represents the output signal of antenna i:

[0085]

[0086] The overall covariance matrix can be represented as:

[0087]

[0088] where R s is the autocorrelation matrix of the RFI signals. represents the noise power, X is the matrix form of the antenna outputs, and s represents the signals emitted by the RFI signal sources.

[0089] R: overall covariance matrix. X: signal vector of the receiving antenna outputs, dimension N x 1 (N is the number of antennas). n: noise vector. Noise power.

[0090] S2: perform eigenvalue decomposition on matrix R to obtain eigenvalues and eigenvectors. Among them, U s′ = [u1, u2, …, u S′ ] is composed of eigenvectors corresponding to the S' larger eigenvalues, called the signal subspace, and U n = [u S′+1 , u S′+2 , …, u Nis composed of N-S' eigenvectors corresponding to smaller eigenvalues, called noise subspace, the two subspaces are independent and orthogonal, Λ is a diagonal matrix, the diagonal line is the eigenvalue.

[0091]

[0092] Λ s′ : eigenvalue diagonal matrix corresponding to signal subspace. Λ n : eigenvalue diagonal matrix corresponding to noise subspace. U: complete eigenvector matrix. Λ: complete eigenvalue diagonal matrix.

[0093] The overall eigenvalues are arranged in descending order, and a sliding window M is set. According to experience, the window M = 5, and the value range of M is (1 to N, N is the number of antennas) Calculate the slope of each M group of eigenvalues {Δλ k ,...,Δλ k+4}(Δλ k = λ k -λ k+1 ), and slide the window to traverse the entire eigenvalue. Calculate the variance C r (var{Δλ k ,...,Δλ k+4}) of each five groups of slopes, and take the last 25% C r as the possible noise space, the mean μ noise and the standard deviation σ noise , set the threshold K r = μ noise +3σ noise , which can cover 99.7% of the noise fluctuations and reduce misjudgment. Starting from the first sliding window, judge the relationship between the variance of the sliding window and the threshold K r in order, find the first sliding window that meets the following conditions: the variance of the sliding window is less than the threshold K r ; Set the found sliding window as the kth sliding window, then the real signal subspace and noise subspace are divided by the point S' = M(k-1).

[0094] Preferably, the traditional single load MUSIC algorithm often uses MDL criterion or AIC criterion as the theoretical basis for source number estimation, but both methods cannot accurately estimate in the case of low signal-to-noise ratio and small number of snapshots. If K snapshots are used to estimate the covariance matrix, the AIC and MDL of the MUSIC algorithm are defined as:

[0095]

[0096] M s : represents the number of signal sources (sources, sources), which is the parameter to be estimated, by making (AIC(M s)) or (MDL(M s )) reaches minimum. K: number of snapshots, i.e., the number of independent data samples (snapshots) used for covariance matrix estimation.

[0097] R: number of array elements, representing the number of physical array elements of the receiving array.

[0098] λ i : eigenvalue, which is obtained after eigenvalue decomposition of the covariance matrix, where i = 1, 2,..., R, and it is generally assumed that the first Ms eigenvalues are large eigenvalues related to the signal, and the last N - Ms eigenvalues are small eigenvalues related to the noise.

[0099] represents the geometric mean of the eigenvalues from Ms + 1 to R, which is used to reflect the average level of the eigenvalues in the noise subspace.

[0100] represents the arithmetic mean of the eigenvalues from Ms + 1 to N, which is also used to reflect the average level of the eigenvalues in the noise subspace.

[0101] M s (2R - M s ), is derived according to the AIC and MDL criteria, which is used to construct the penalty term or correction term of the corresponding criterion function, and is related to the number of array elements, the number of sources, and the number of snapshots, etc., and is used to balance the goodness of fit and the model complexity in model selection (source number estimation).

[0102] The parameter Ms is estimated by the point at which the AIC (or MDL) reaches minimum.

[0103] The eigenvalues obtained by eigenvalue decomposition of the covariance matrix are arranged in descending order, and the variance of the slope between five consecutive eigenvalues is calculated, as follows:

[0104] C r (k) = Var{Δλ k , Δλ k+1 ,..., Δλ k+W-1}

[0105] C r = Cr signal + Cr noise

[0106] Cr(k): represents the "variance of the slope between W consecutive eigenvalues", and W is the sliding window.

[0107] αλ i : here Δλ i refers to the slope (change rate) between adjacent eigenvalues, such as Δλi = λ i+1 - λ i , Δλ i , Δλ i+1 , …, Δλ i+3 , Δλ i+4 , is the slope of the 5 consecutive eigenvalues.

[0108] Cr(k) represents the eigenvalue slope variance statistics of the signal-related region (the region corresponding to the eigenvalue containing the signal source information), that is, the part of the eigenvalue affected by the signal is calculated from the eigenvalue.

[0109] Cr(k) represents the eigenvalue slope variance statistics of the noise-related region (the region corresponding to the eigenvalue dominated by noise only), that is, the part of the pure noise eigenvalue is calculated, which will be used to fit the Gaussian distribution to assist in judging the noise region.

[0110] According to the analysis, for N eigenvalues, if there are S signal sources, the last N-S eigenvalues belong to noise. Since S << N, the last 25% of the data is likely to belong to the pure noise region. (When the noise is too large, 25% cannot be used as a judgment of the noise region, but at this time it is basically impossible to use the MUSIC algorithm for analysis). Therefore, the last 25% of Cr(k) is selected as the pure noise region.

[0111] The C r (k) of the pure noise region is represented by the symbol Cr noise (k), which is fitted according to the Gaussian noise distribution, The mean μ noise of the pure noise region and the standard deviation σ noise of the pure noise region are calculated as follows:

[0112]

[0113] According to the 3σ criterion of normal distribution, the Cr(k) of the noise region has a 99.7% probability of falling within the interval [μ noise -3σ noise , μ noise +3σ noise ].

[0114] Set the threshold:

[0115] K r = μ noise +3σ noise

[0116] Finally, compare C r (k) and K rThe relative size of the first C r (k) < K r The value S as a demarcation point between the signal subspace and the noise subspace.

[0117] To prove that the traditional single load MUSIC algorithm is extremely poor in the high noise environment, simulation experiments are carried out, and table 1 is set two-dimensional RFI source position and intensity information.

[0118] Table 1 Two-dimensional RFI source position and intensity information

[0119]

[0120] Table 2 is set one-dimensional RFI source position and intensity information, since the one-dimensional antenna array can only detect information in a single direction, it is assumed that only the ξ direction RFI source can be detected. Table 2 is as follows:

[0121] Table 2 One-dimensional RFI source position and intensity information

[0122]

[0123] Without interference, the brightness temperature inversion effect is as shown in Figure 3 , and after adding interference, the generated noise brightness temperature inversion figure is as shown in Figure 4 . It can be seen that the noise has a very serious effect on the brightness temperature inversion figure, and it is basically impossible to determine how many RFI source points there are.

[0124] When only the traditional single load MUSIC algorithm is used for analysis, the effect figure is as shown in Figure 5 , it can be seen that only one RFI source point is identified, and the interference condition is serious.

[0125] When the multi-load fusion MUSIC algorithm proposed in the application is used for analysis, the effect figure is as shown in Figure 6 , at least four RFI source points can be clearly identified, and the effect can reach the expectation.

[0126] S3, the spatial pseudo-spectrum (unitless) of MUSIC is given by the following formula:

[0127]

[0128] U n represents the noise subspace, P M (ξ,η) peak value corresponding to (ξ,η) is the incoming wave direction estimation of the target signal. a(ξ,η) is the signal steering vector from direction (ξ,η).

[0129]

[0130] If in the case of multiple RFI sources, according to PM The S peaks of (ξ, η) can provide azimuth estimates for the S RFI sources.

[0131] Those skilled in the art can easily understand that the above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A one- or two-dimensional synthetic aperture radiometer fusion RFI localization method based on covariance matrix, characterized in that, include: The receiving end synchronously acquires the measurement covariance matrix of the one-dimensional antenna array, and then fuses the two to obtain a new covariance matrix; The new covariance matrix is ​​subjected to eigenvalue decomposition, and the signal subspace and noise subspace are divided according to the eigenvalue distribution characteristics. A pseudo-spectrum is generated, and the azimuth estimation of multiple radio frequency interference sources is obtained through the spectral peaks.

2. The one- or two-dimensional synthetic aperture radiometer fusion RFI localization method based on covariance matrix according to claim 1, characterized in that, The receiving end synchronously acquires the measurement covariance matrix of a two-dimensional antenna array, and then fuses the two to obtain a new covariance matrix. The method is as follows: (1) The receiving end synchronously acquires the received signals of the one-dimensional antenna array, performs complex correlation operation on the received signals of the one-dimensional antenna array to obtain the measurement covariance matrix R1, and performs complex correlation operation on the received signals of the two-dimensional antenna array to obtain the measurement covariance matrix R2. (2) After fusing the two, a new covariance matrix R is obtained, where Where a is the number of antennas in the one-dimensional antenna array, b is the number of antennas in the two-dimensional antenna array, the covariance matrix R is an (a+b)×(a+b) dimensional matrix, c represents the weight, and c is equal to the ratio of the sensitivity of the two-dimensional antenna array to that of the one-dimensional antenna array.

3. The one- or two-dimensional synthetic aperture radiometer fusion RFI localization method based on covariance matrix according to claim 1, characterized in that, The elements R(i,j) of R1 or R2 satisfy i = j indicates autocorrelation, and i ≠ j indicates cross-correlation. x i x represents the output signal of antenna i. j * This represents the conjugate of the output signal of antenna j; V(u ij ,v ij ) represents the visibility sample obtained by cross-correlation of the output signals of antenna i and antenna j.

4. The one- or two-dimensional synthetic aperture radiometer fusion RFI localization method based on covariance matrix according to claim 1, characterized in that, x i (t) represents the output signal of antenna i at time t, x j * (t) represents the conjugate of the antenna j output signal at time t.

5. The one- or two-dimensional synthetic aperture radiometer fusion RFI localization method based on covariance matrix according to claim 1, characterized in that, The new covariance matrix is ​​subjected to eigenvalue decomposition, and the signal subspace and noise subspace are divided according to the eigenvalue distribution characteristics. The specific method is as follows: (1) Perform eigenvalue decomposition on the new covariance matrix R to obtain eigenvalues ​​and eigenvectors; (2) Sort the eigenvalues ​​in descending order, set a sliding window M, where M ranges from 1 to N, and N is the total number of antennas in the one-dimensional antenna array. Iterate through all eigenvalues ​​using the sliding window and calculate the slope {Δλ} of adjacent eigenvalues ​​within each sliding window. k ,...,Δλ k+M-1 }; (3) Calculate the variance C of each sliding window using the slope of adjacent feature values ​​within each sliding window. r All sliding windows are sorted by variance from largest to smallest, and the bottom 25% of C r As a possible noise subspace, the mean μ of the possible noise subspace is calculated. noise and standard deviation σ noise ; (4) Based on the mean μ of the possible noise subspace noise and standard deviation σ noise Calculate the threshold K r K r =μ noise +3σ noise ; (5) Starting from the first sliding window, sequentially evaluate the variance of each sliding window against the threshold K. r Find the first sliding window that satisfies the following condition: the variance of the sliding window is less than the threshold K. r ; (6) Suppose that the found sliding window is the kth sliding window, then the boundary point between the real signal subspace and the noise subspace is S′=M(k-1).

6. The one- or two-dimensional synthetic aperture radiometer fusion RFI localization method based on covariance matrix according to claim 5, characterized in that, Signal subspace U s′ =[u1,u2,…,u S′ The noise subspace U is composed of the eigenvectors corresponding to the first S′ eigenvalues ​​arranged in descending order. n =[u S′+1 ,u S′+2 ,…,u N The subspace is composed of the eigenvectors corresponding to the S′+1, S′+2, ..., Nth eigenvalues ​​arranged in descending order. The two subspaces are independent and orthogonal.

7. The one- or two-dimensional synthetic aperture radiometer fusion RFI localization method based on covariance matrix according to claim 6, characterized in that, Λ s′ Λ n It is a diagonal matrix, with the diagonal lines representing eigenvalues.

8. The one- or two-dimensional synthetic aperture radiometer fusion RFI localization method based on covariance matrix according to claim 1, characterized in that, The generated pseudo-spectrum is as follows: U n Let a(ξ,η) represent the noise subspace, and let a(ξ,η) be the signal turning vector from the direction (ξ,η).

9. The one- or two-dimensional synthetic aperture radiometer fusion RFI localization method based on covariance matrix according to claim 8, characterized in that, P M The (ξ,η) corresponding to the peak value is the estimated direction of arrival of the target signal.

10. The one- or two-dimensional synthetic aperture radiometer fusion RFI localization method based on covariance matrix according to claim 9, characterized in that, In the case of multiple radio frequency interference sources, according to P M The S peaks of (ξ,η) are used to estimate the orientation of S RFI sources.

Images