Ground wave radar amplitude and phase calibration method based on oil and gas platform

By using the amplitude ratio between array antennas to screen the arrival angle signal on the oil and gas platform and combining with the MUSIC iterative method, the problem of calibration difficulties in the special-shaped array is solved, and efficient amplitude phase calibration is achieved, suitable for complex environments.

CN118914991APending Publication Date: 2024-11-08ZNPL OCEAN DETECTION SYST ENG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410913277.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-09
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Due to the array-type high-frequency ground wave radar deployed on the oil and gas platform, it is difficult to find a suitable translational unchanged array element pair, which leads to the failure of the traditional amplitude-phase calibration algorithm, affecting the radar's inversion ability to sea conditions.

Method used

The amplitude-phase calibration method that does not rely on a specific array is adopted. The amplitude-phase calibration method is used to filter the amplitude ratio between array antennas, calculate the amplitude error, and solve the phase error using array dimension reduction and MUSIC iterative methods, and calibrate iteratively with the Adadelta gradient descent algorithm.

Benefits of technology

It realizes amplitude phase calibration without the need for specific arrays in complex construction environments, improves the radar's ability to identify sea conditions information and is suitable for oil and gas platforms and other sites.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118914991A_ABST
    Figure CN118914991A_ABST
Patent Text Reader

Abstract

The invention relates to a ground wave radar amplitude and phase calibration method based on an oil and gas platform. The method is used for solving the calibration problem of an oil and gas platform ground wave radar. The method comprises the following steps: firstly, screening out a single-angle-of-arrival echo signal by using an amplitude ratio between array antennas; and then analyzing the receiving matrix of the antenna signals, and calculating the amplitude error of each antenna. And performing initial value estimation on the phase error in an array dimension reduction mode, and finally accurately solving the phase error of the antenna by using a MUSIC iteration method to complete calibration. Compared with a traditional calibration method depending on an antenna array shape, the calibration method has no special requirements for the array shape, and has good adaptability to places with complex construction environments such as oil and gas platforms.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of radio ocean remote sensing and digital signal processing, and in particular relates to an amplitude and phase calibration method of a ground wave radar based on an oil and gas platform. Background Art

[0002] Detecting ocean wind speed, wave height and current information is crucial to the development of marine resources, environmental monitoring and maritime traffic safety. As a long-range and large-scale ocean monitoring technology, ground wave radar can effectively provide these key data and play an important role in predicting changes in the marine environment, assessing marine disaster risks and guiding marine operations.

[0003] Oil and gas platforms are important facilities for offshore resource development. Because they are usually far from the coastline, they are difficult to be covered by traditional marine monitoring methods. Therefore, deploying ground wave radar on oil and gas platforms can not only expand the detection range and realize the monitoring of offshore sea conditions, but also grasp the environmental conditions of the sea area around the platform in real time, providing decision support for offshore operations.

[0004] However, due to the inconsistency of the amplitude and phase characteristics between antennas, array-type high-frequency ground wave radars have amplitude and phase errors, which will reduce the DOA resolution capability of the antenna array, and thus seriously affect the radar's inversion of sea conditions. For shore-based high-frequency ground wave radars, a passive amplitude and phase calibration method based on ocean echoes is usually adopted. This method does not require an auxiliary calibration source and is stable and reliable. However, this method requires the antenna array to have a translation-invariant subarray to screen single-arrival ocean echoes. When laying out the array, shore-based ground wave radars can fully consider this layout requirement due to the smaller site restrictions. For example, the shore-based model of the OSMAR071G high-frequency ground wave radar usually adopts a symmetrical array of six in the front and two in the back, in which there are multiple translation-invariant subarrays.

[0005] In contrast, deploying ground wave radar on oil and gas platforms faces different challenges. Since the main function of the platform is to exploit seabed resources, the platform is already equipped with many production and detection equipment, and it is difficult to find a flat site to deploy a uniform symmetrical array similar to shore-based radar. Therefore, special-shaped arrays can only be deployed based on the hardware site conditions of the platform. However, it is difficult to find suitable translation-invariant array element pairs for most special-shaped arrays, resulting in the failure of traditional amplitude and phase calibration algorithms.

[0006] Therefore, for the array-type high-frequency ground wave radar on oil and gas platforms, it is necessary to design a passive amplitude and phase calibration method that does not require special antenna array layout requirements. This has become a problem that needs to be solved urgently. Summary of the invention

[0007] In order to solve the calibration problem of the above-mentioned oil and gas platform ground wave radar, the present invention proposes an amplitude and phase calibration method that does not rely on a specific array. The method first uses the amplitude ratio between array antennas to screen out single arrival angle echo signals. Then, the receiving matrix of the antenna signal is analyzed to calculate the amplitude error of each antenna. Then, the initial value of the phase error is estimated by array dimensionality reduction, and finally the MUSIC iterative method is used to accurately calculate the phase error of the antenna to complete the calibration.

[0008] The specific technical solutions are as follows:

[0009] A ground wave radar amplitude and phase calibration method based on an oil and gas platform, comprising:

[0010] Step 1: Filter out single arrival angle signals from echo signals by utilizing the amplitude ratio aggregation of single arrival echo signals between array elements;

[0011] Step 2: Compare the single angle of arrival signal received by each array element with the single angle of arrival signal received by the reference array element, so as to obtain the amplitude error of the corresponding array element;

[0012] Step 3: On the basis of eliminating the amplitude error, the initial phase error value of the entire array element is gradually determined by using array dimension reduction;

[0013] Step 4: Combine the MUSIC cost function with the Adadelta gradient descent algorithm, and through cyclic iterative optimization, accurately obtain the phase error value of each array element to complete the amplitude and phase calibration of the ground wave radar.

[0014] Further, step 1 includes the following steps:

[0015] a) Signal preprocessing: First, perform fast Fourier transform (FFT) on the received echo signal to convert the echo signal into the frequency domain; then, use short-time Fourier transform (STFT) to further analyze the signal and obtain a snapshot matrix for each echo signal, where the number of rows M of the matrix represents the number of array elements in the array, and the number of columns T represents the number of time windows selected in the STFT;

[0016] b) Calculate the ratio of signals from different array elements: For each snapshot matrix of the echo signal, divide it into blocks by rows to form M complex vectors of length T, each complex vector corresponding to a signal received by an array element; then, divide the vectors corresponding to the other array elements except the reference array element by the vector corresponding to the reference array element, let C=M-1, and obtain a C×T complex ratio matrix; average the complex ratio matrix by rows, and calculate its amplitude or modulus, so as to obtain a real vector of length C; when the signal is a single arrival angle signal, ignore the influence of noise and the ratio is a constant.

[0017] c) For all signal snapshot matrices that meet the signal-to-noise ratio requirements, perform the above processing steps to construct a real ratio matrix, where the number of rows in the matrix is ​​equal to C, and the number of columns H is the total number of echo signals that meet the signal-to-noise ratio requirements;

[0018] d) In the above C×H ratio matrix, treat each column as a point and calculate the Euclidean distance between any two different points;

[0019] e) Use the local clustering algorithm to find the density center point Q with a radius of h, where the radius h is the estimated number of single arrival angle echoes, usually a certain proportion of the total number of observed samples;

[0020] f) Taking Q as the center, h points closest to Q are screened out, and these points are judged as single arrival angle signals.

[0021] Furthermore, for each array element, a statistical average value of amplitude errors calculated from a plurality of single angle-of-arrival signals is taken as the amplitude error of the array element to improve accuracy.

[0022] Further, step 3 includes the following steps:

[0023] a) First, MN array elements are selected from all array elements to form a nonlinear sub-array, where N>0 and N≤M-3, and the reference array element is removed. For the remaining array elements in the nonlinear sub-array, all possible phase error values ​​are searched in the range of [-π,π) with a fixed step size, and a set of phase error values ​​is determined;

[0024] b) Based on the phase error value, the covariance matrix of each single arrival angle signal snapshot matrix is ​​calculated and the eigenvalue decomposition is performed. When the number of signal sources is 1, the noise subspace is directly obtained.

[0025] c) using the determined phase error value and noise subspace, the arrival angle θ corresponding to each single arrival angle signal and its corresponding MUSIC spectrum value are obtained by searching through the MUSIC expression;

[0026] d) Accumulate the inverse of the MUSIC spectrum values ​​corresponding to all single arrival angle signals to obtain the value of the cost function P;

[0027] e) Traverse all possible phase error combinations in the range of [-π,π), repeat steps b to d, and calculate the values ​​of all cost functions P; when the P value is the smallest, obtain the initial phase error value and single arrival angle signal angle of the MN array elements;

[0028] f) On the basis of the sub-array with determined phase error, gradually add an array element to form a new sub-array, and set the phase error of the newly added array element to be

[0029] g) Yes Traverse in the range [-π,π) with a fixed step size, keep the previously calculated array element phase error unchanged, substitute it into the MUSIC summation cost function, when P reaches the minimum value, the corresponding That is the initial value of the phase error of the array element;

[0030] h) Repeat steps f and g until the initial phase error values ​​of all M array elements are obtained.

[0031] Further, step 4 includes the following steps:

[0032] a) First, set the hyperparameter values ​​of the Adadelta algorithm, then substitute the previously obtained initial phase error value and single angle of arrival signal angle into the MUSIC cost function to calculate the initial cost function value;

[0033] b) The MUSIC cost function takes the partial derivative of the phase error and obtains the gradient

[0034] c) Update the next phase error value according to the Adadelta algorithm update rule

[0035] d) According to the updated phase error Search the corresponding angle θ with the MUSIC expression, and then recalculate the cost function value P t ;

[0036] e) Compare the current cost function value P t and the previous cost function value P t-1 If P t-1 -P t >ε, ε is a minimum value greater than 0, and the iteration continues;

[0037] f) Repeat steps b to f until the cost function value P no longer decreases significantly, that is, P t-1 -P t <∈; when the convergence condition is met, the loop is exited, and the phase error at this time This is the final phase error value.

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] 1. Compared with the traditional calibration method that relies on the antenna array formation, this calibration method has no special requirements for the formation. It has good adaptability to sites with complex construction environments such as oil and gas platforms.

[0040] 2. The amplitude ratio distribution of the single arrival echo signal received by the antenna has the characteristics of clustering. At the same time, combined with the multi-element multi-dimensional clustering algorithm, the actual single arrival angle signal is better screened out.

[0041] 3. This method combines the MUSIC cost function with the Adadelta gradient descent algorithm, and achieves efficient iteration of the phase error by adaptively adjusting the learning rate, while ensuring the stability of the optimization process in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 , flow chart of the method of the present invention;

[0043] Figure 2 , Statistical diagram of antenna signal ratios corresponding to 3-dimensional space elements. DETAILED DESCRIPTION

[0044] The specific steps and related principles of the present invention are as follows:

[0045] Step 1: Filter out the single arrival angle signal in the echo signal. When the ground wave radar of the oil and gas platform has M array elements, the coordinates of the array element i are expressed as (x i ,y i )(i=1,2,…,M), where antenna No. 1 is the origin coordinate (x1=0,y1=0). After performing two Fast Fourier Transforms (FFT) on the received signal, the two-dimensional echo signal on channel i can be described as:

[0046]

[0047] Where m is the discrete frequency in the distance dimension; o is the discrete frequency in the velocity dimension (Doppler shift); g i is the amplitude error of array element i relative to array element 1 (reference array element); is the phase error of element i relative to element 1; S k The arrival angle received by array element i is θ k Two-dimensional echo signal; n i is additive noise; a i (θ k ) is the time when the array element i is relative to the array element 1, θ k The corresponding directional complex response or steering vector can be expressed as:

[0048] a(θ k )=[1 a2(θ k ) … a M (θ k )] T (2)

[0049] in λ is the wavelength of radar electromagnetic wave.

[0050] Ignoring the influence of noise, when the signal echo comes from multiple directions, that is, when K≠1, the value of the signal received by element i is modulo the value of the signal received by element 1, as shown below:

[0051]

[0052] Obviously, when K≠1 (K is greater than 1), the ratio cannot be reduced due to the influence of multiple arrival angles, resulting in a non-unique ratio result.

[0053] When K=1, the ratio is:

[0054]

[0055] Therefore, when the echo signal is a single arrival angle signal, the above ratio result is exactly the amplitude error g of the antenna relative to antenna No. 1 (reference antenna). i , is a fixed value.

[0056] To perform statistics on a large number of echo signals, use |Z i | is greater than |Z1|, for a single angle of arrival signal, the ratio will be towards g i For non-single arrival angle signals, the final result is also related to the signal echo direction, and the ratio distribution will be more random. (When noise is not considered, for all single arrival echo signals, theoretically |Z i |The value of |Z1| is g i ; When noise is taken into account, the ratio will be g i Gather. )

[0057] The above-mentioned focusing effect can be used to effectively screen single-arrival angle signals. When the signal-to-noise ratio is high enough, theoretically only two antennas are needed to easily identify the signal. However, the signal-to-noise ratio of ground wave radar echoes is usually only about 20dB, or even lower. If only two antennas are used for screening, the focusing effect will be weakened. At this time, multiple antennas can be considered for joint screening.

[0058] Assume that H echo signals are received within a period of time. For each signal, the amplitude response on the ith antenna is divided by the amplitude response on the 1st antenna, as At this time, the M array elements can calculate M-1 ratios for the signal. Define an M-1 dimensional space and Considered as a dimension of the space. Then the M-1 ratios corresponding to each signal constitute a point in the above space. Let C = M-1, for all H echo signals, a C×H space point set will be formed. The dimension of the C-dimensional space, H is the number of points in the point set.

[0059] For this spatial point set, a local clustering algorithm is used to calculate the density center position Q of the point set. The specific method is as follows:

[0060] 1. Traverse each point in the point set and find the h points closest to it. Calculate the Euclidean distance between the point and the nearest h points, and add them up to calculate the sum of the distances, recorded as d i (i=1,2,…,H). h is empirical and can be regarded as the search radius, and its value is usually selected as 20% of the total number of echo signals H, that is, h=0.2H.

[0061] 2. For any point, if its corresponding d i When the minimum value is obtained, this point is the density center point Q.

[0062] 3. Take Q as the center and find the h points closest to it. The echo signals corresponding to the h points can be considered as single-arrival echo signals.

[0063] Figure 2 When only a 3-element matrix is ​​used, the corresponding ratio statistics are obtained in the two-dimensional space:

[0064] Step 2: Calculate the amplitude error of the array antenna. After the single arrival angle signal is screened out, the amplitude error of array element i can be calculated using formula (4) in step 1. In practical applications of signal processing, in order to improve the sampling accuracy, short-time Fourier transform (STFT) is usually used to replace the second fast Fourier transform (FFT). STFT moves the time window with a fixed step size and finally generates an M×T signal snapshot matrix, where T represents the total number of sliding time windows in STFT.

[0065] At the same time, since the actual echo signal contains noise, the amplitude errors calculated for different single-angle-of-arrival signals on the same antenna will be different. Taking the statistical average of the amplitude errors calculated for multiple single-angle-of-arrival signals can further improve the accuracy:

[0066]

[0067] Among them, Z i represents the echo signal received by the ith antenna, l represents the lth single arrival angle signal source; t represents a specific signal snapshot. Z in formula (6) and formula (1) i The physical meanings expressed are the same, except that formula (1) observes the signal from the distance and speed dimensions, while formula (6) expresses it through the number of signal sources and the number of snapshots.

[0068] Step 3: Perform an initial value estimation of the phase error by array dimensionality reduction. After amplitude correction of the echo data, the array covariance matrix corresponding to any single arrival angle signal can be expressed as:

[0069] R = E[X(t)X(t) H ]=Φa(θl )σ l 2 a H (θ l )Φ H +σ n 2 I (7)

[0070] Among them, X(t) is the receiving matrix corresponding to signal l, (.) H represents conjugate transpose, Φ is the phase error value The diagonal matrix of l is the echo arrival angle of the lth single source signal, a(θ l ) is θ l The corresponding steering vector, σ l 2 is the signal power; σ n 2 is the noise power. Perform eigenvalue decomposition on R, and we have:

[0071]

[0072] E s (l) is the signal subspace matrix, which consists of the eigenvectors of the signal, Λ s is the signal’s eigenvalue diagonal matrix, W n (l) is the noise subspace matrix, which consists of the eigenvectors of the noise, Λ n is the diagonal matrix of eigenvalues ​​of the noise. Since the noise subspace is orthogonal to the steering vector, there exists E n H (l)[Φa(θ l )]=0, then the expression of the corresponding MUSIC spectrum (multiple signal classification method) is:

[0073]

[0074] Furthermore, the MUSIC summation cost function based on the orthogonality principle of noise subspace under single arrival angle condition is:

[0075]

[0076] L is the number of single information sources, a(θ l ) is θ l The corresponding steering vector. In order to solve the cost function P, it can be transformed into a quadratic iterative optimization problem as follows:

[0077]

[0078] θ is the phase error condition of the previous iteration (denoted as ) under the MUSIC azimuth estimation angle; the phase error generated by the new round of iteration is ω=[1 0 0 ..... 0] T In the above iterative optimization process, J will gradually decrease, and J ≥ 0, and the optimization process can be guaranteed to converge to a local optimum. If it is to converge to the global optimum, the initial phase error value (k = 1) must be close to the true phase error.

[0079] Since estimating the phase errors and signal angles of all antennas at the same time will result in too much calculation, the array dimension reduction and iterative search method is used to gradually calculate the initial phase error values ​​of all antennas. When the number of array elements is M, if the cost function P is used to estimate all antennas at the same time, too many parameters need to be estimated at the same time. To simplify the calculation, the phase errors of MN antennas are first estimated (N>0 and N≤M-3), and then the number of estimated antennas is gradually increased, one at a time, until the initial phase error estimation of all M antennas is completed. The specific method is as follows:

[0080] 1. From all array elements, select MN array elements to form a nonlinear subarray; the reference array element phase error is Assume that the phase error of the other MN-1 array elements is And perform a traversal search on [-π,π) with a fixed step size to first determine the first set of phase values.

[0081] 2. Then calculate the covariance matrix of each received single arrival angle signal snapshot matrix and perform eigenvalue decomposition. When the number of signal sources is determined to be 1, the noise subspace can be directly obtained At this time, since the phase error and the noise subspace are known, the arrival angle θ corresponding to each signal and the MUSIC spectrum value corresponding to θ can be directly searched according to the MUSIC expression.

[0082] 3. According to formula (10), the inverse of the MUSIC spectrum values ​​corresponding to all θ are accumulated to obtain the value of the cost function P;

[0083] 4. Then traverse all possible phase error combinations for the MN array with a fixed step size on [-π,π) Repeat steps 2 and 3 to calculate the values ​​of all cost functions P. When the value of P is the smallest, the initial phase value of the MH array and the arrival angle of the single signal source are obtained.

[0084] 5. Add another array element to the entire array to form a new subarray; let the phase error of the added array element be right Traverse with a fixed step size on [-π,π), keeping the previously calculated array phase error unchanged; substitute the new sub-array phase error determined by each traversal into equation (10), when P reaches the minimum value, the corresponding This is the initial value of the phase error of the array element.

[0085] 6. Repeat step 5 until the initial phase error values ​​of all array elements are obtained.

[0086] Step 4: Based on the initial phase values ​​of all arrays, the MUSIC cost function is combined with the Adadelta gradient descent method to iteratively obtain the final phase error value. Formula (10) is used as the cost function to be optimized, the phase error of each antenna and the single angle of arrival signal are used as parameters, and a set of parameters is found to minimize the value of the cost function P. When P is the smallest, this set of parameters is considered to be the optimal solution.

[0087] During the iteration process, the Adadelta algorithm can adaptively adjust the learning rate of each parameter to ensure that it can effectively converge to the global minimum during the optimization process. In each iteration, the gradient under the current parameter is first calculated, and then the parameter value (phase error value) is adjusted according to the Adadelta update rule. This process is repeated until the cost function value P no longer decreases significantly, then the algorithm stops to obtain the final phase error value. The specific method is as follows:

[0088] 1. Substitute the initial phase value and the single angle of arrival signal obtained in step 3 into equation (10) to calculate the P value, and at the same time, take the partial derivative of equation (10) to obtain the gradient.

[0089]

[0090] 2. Let the gradient To phase To update, the update method is as follows:

[0091]

[0092] Where ∈ and ρ are two hyperparameters in the Adadelta method, which are two extremely small fixed values. t With s t It is a parameter updated by the method, and its initial value is 0. ⊙ is the Hadamard Product, also known as the element-wise product.

[0093] 3. Get the updated phase After that, the corresponding angle θ is searched according to the MUSIC expression, and then the new cost function value P is calculated according to formula (10).

[0094] 4. When P t-1-P t >ε (ε is the minimum value greater than 0), repeat steps 2 to 4. t-1 -P t <ε, then the loop is exited to obtain the final phase error value

[0095] The technical solution of the present invention is further described in detail below in combination with actual practice.

[0096] Step 1: Filter out the single arrival angle signal in the echo signal. When the antenna array has 8 elements, first perform a fast Fourier transform (FFT) on the received echo signal to convert it to the frequency domain. Then, the short-time Fourier transform (STFT) is used to further analyze the signal, and the time window of the STFT moves with a fixed step size. At this time, the signal of an effective frequency point is an 8x17 signal snapshot matrix, where 8 represents the number of elements in the antenna array, and 17 represents the number of time windows selected in the STFT. Then process each signal as follows:

[0097] The snapshot matrix of each signal is divided into 8 complex vectors of length 17, each of which corresponds to a signal received by an antenna. Subsequently, the vectors corresponding to antennas 2 to 8 are divided by the vector corresponding to antenna 1 to obtain a 7x17 complex ratio matrix. The matrix is ​​then averaged row by row and its amplitude (or modulus) is calculated to obtain a 7x1 real vector. When the signal is a single arrival angle signal, the noise effect is ignored, the ratio is a constant, and there will be aggregation when a large number of echoes are counted.

[0098] For all signal snapshot matrices that meet the SNR requirements, perform the above processing steps. When there are 1000 echo signals that meet the requirements, a 7×1000 ratio matrix can be constructed. In this matrix, the number of columns 1000 can represent the total number of observation samples, and the number of rows 7 can represent the 7 dimensions of a point in the matrix space. For example, the distance between point p and point q can be expressed as:

[0099]

[0100] Furthermore, the local clustering algorithm is used to find the center position Q with the highest density and radius h. Here, the radius h is 200, which is obtained by multiplying 1000 by 20%. 20% is an empirical ratio, which means that about 20% of all signals are single-angle-of-arrival signals. At this time, with Q as the center, the 200 points closest to Q are screened out, and these points are judged to be single-angle-of-arrival signals.

[0101] Step 2: Calculate the amplitude error of the array antenna. After the single angle of arrival signal is screened out, the amplitude ratio of the signal received by each antenna can be calculated by comparing it with the echo signal received by the reference antenna (antenna No. 1) to obtain the amplitude error of the corresponding antenna. Taking the statistical average of the amplitude errors calculated for multiple single angle of arrival signals can further improve the accuracy:

[0102]

[0103] Among them, z i represents the echo signal received by the i-th antenna, l represents the l-th single arrival angle signal source, t represents the t-th signal snapshot, To calculate the amplitude error value of the i-th antenna.

[0104] Step 3: Perform an initial value estimate of the phase error by array dimensionality reduction. After the amplitude correction of the echo data is implemented, antennas 1 to 4 are selected to form a nonlinear subarray of 4 elements, considering the computational efficiency and time complexity. First, the most appropriate phase error value of these 4 elements is estimated by traversal, and then the number of estimated antennas is gradually increased, one at a time, until the initial value estimation of the error of all 8 elements is completed. The specific method is as follows:

[0105] 1. In the initial 4-element configuration, excluding the first antenna, the phase errors of the remaining 3 antennas will be traversed in the range of [-π,π) with a fixed step size of 5 degrees to explore all possible phase error values.

[0106] 2. When the phases of the four array elements are at a certain value, such as (0°, 20°, 20°, 20°), perform eigenvalue decomposition on the snapshot matrix of each received single arrival angle signal. Since the signal is a single source signal at this time, the noise subspace can be directly obtained. At this time, since the phase error and noise subspace are known, the arrival angle θ of the corresponding signal and its corresponding MUSIC spectrum value can be directly searched through the MUSIC expression.

[0107] 3. According to the MUSIC algorithm, a summation cost function with phase parameters is defined, and the expression is as follows:

[0108]

[0109] The reciprocals of the MUSIC spectrum values ​​corresponding to all single-arrival angle signals are accumulated to obtain the value of the cost function P.

[0110] 4. In the interval [-π,π), for all possible phase error combinations of the 4 array elements obtained by traversal, steps 2 and 3 are repeated to calculate all cost function values ​​P. When the P value is minimized, the phase value at this time is used as the initial phase value of the 4 array elements.

[0111] 5. Add another array element to the entire array to form a new subarray; let the phase error of the added array element be right Traverse with a fixed step size (5 degrees) on [-π,π), keeping the previously calculated array phase error unchanged; substitute the phase error determined in each traversal into the MUSIC summation cost function, when P reaches the minimum value, the corresponding This is the initial value of the phase error of the array element.

[0112] 6. Repeat step 5 until the initial phase error values ​​of all array elements are obtained.

[0113] Step 4: Combine the MUSIC cost function and the Adadelta gradient descent method to iterate and obtain the final phase error value. During the iteration process, the Adadelta algorithm can adaptively adjust the learning rate of each parameter to ensure that it can effectively converge to the global minimum during the optimization process. The specific method is as follows:

[0114] 1. Set the hyperparameter values ​​of the Adadelta algorithm, ρ to 0.95, ∈ to 1e-6. Set the initial phase values ​​(0°, 25°, 30°, 145°, 90°, 40°, 55°, 65°) obtained in step 3 and the angle θ of each single arrival angle signal i , bring in the MUSIC summation cost function, calculate the P value, and at the same time Derivative, get the gradient

[0115] 2. Then use the Adadelta method to compare the gradient and To update, the update method is as follows:

[0116]

[0117] Δx t With s t It is a parameter updated by the method, with an initial value of 0.

[0118] 3. Get the updated phase After that, the corresponding angle θ is searched according to the MUSIC expression, and then the new cost function value P is calculated.

[0119] 4. Compare the error between the tth and t-1th P values. t-1 -P t>ε (ε is 0.0001), repeat steps 2 to 4. t-1 -P t <ε, then the loop is exited to obtain the final phase error value

[0120] The present invention screens out the receiving matrix of a single source through the aggregation of the amplitude ratio of the single-arrival echo signal between array antennas, and then iteratively obtains the amplitude and phase errors in combination with the MUSIC method, and calibrates the errors. The present invention breaks through the traditional dependence on the antenna array formation, can be implemented without a specific formation, and is particularly suitable for complex construction environments such as oil and gas platforms, showing good adaptability; by relying on the aggregation of the amplitude ratio of the single-arrival echo signal received by the antenna, combined with a multi-element multidimensional clustering algorithm, the actual single-arrival angle signal is effectively screened out, and the accuracy of signal recognition is improved; in addition, the present invention combines the MUSIC cost function with the Adadelta gradient descent algorithm, and realizes efficient iteration of the phase error by adaptively adjusting the learning rate, ensuring the stability of the optimization process in a complex environment, and providing a flexible solution for the amplitude and phase calibration of ground wave radar.

Claims

1. A ground wave radar amplitude and phase calibration method based on an oil and gas platform, characterized in that: Step 1: Filter out single arrival angle signals from echo signals by utilizing the amplitude ratio aggregation of single arrival echo signals between array elements; Step 2: Compare the single angle of arrival signal received by each array element with the single angle of arrival signal received by the reference array element, so as to obtain the amplitude error of the corresponding array element; Step 3: On the basis of eliminating the amplitude error, the initial phase error value of the entire array element is gradually determined by using array dimension reduction; Step 4: Combine the MUSIC cost function with the Adadelta gradient descent algorithm, and through cyclic iterative optimization, accurately obtain the phase error value of each array element to complete the amplitude and phase calibration of the ground wave radar.

2. The amplitude and phase calibration method of ground wave radar based on an oil and gas platform according to claim 1, characterized in that: Step 1 further comprises the following steps: 1.1) Signal preprocessing: First, perform fast Fourier transform (FFT) on the received echo signal to convert the echo signal into the frequency domain; then, use short-time Fourier transform (STFT) to further analyze the signal and obtain a snapshot matrix for each echo signal, where the number of rows M of the matrix represents the number of array elements in the array, and the number of columns T represents the number of time windows selected in the STFT; 1.2) Calculate the ratio of signals from different array elements: For each snapshot matrix of the echo signal, divide it into blocks by row to form M complex vectors of length T, each complex vector corresponds to a signal received by an array element; then, use the vectors corresponding to the other array elements except the reference array element to compare with the vector corresponding to the reference array element, let C = M-1, and obtain a C×T complex ratio matrix; take the average of the complex ratio matrix by row, and calculate its amplitude or modulus, so as to obtain a real vector of length C; when the signal is a single arrival angle signal, ignore the influence of noise, and the ratio is a constant; 1.3) For all signal snapshot matrices that meet the SNR requirement, perform the above processing steps to construct a real ratio matrix, where the number of rows in the matrix is ​​equal to C, and the number of columns H is the total number of echo signals that meet the SNR requirement; 1.4) In the above C×H ratio matrix, treat each column as a point and calculate the Euclidean distance between any two different points; 1.5) Use the local clustering algorithm to find the density center point Q with a radius of h. The radius h is the estimated number of single arrival angle echoes, which is usually a certain proportion of the total number of observed samples. 1.6) Taking Q as the center, select the h points closest to Q. These points are judged as single arrival angle signals.

3. The amplitude and phase calibration method of ground wave radar based on an oil and gas platform according to claim 1 is characterized in that: For each array element, the amplitude errors calculated from multiple single arrival angle signals are statistically averaged and used as the amplitude error of the array element to improve accuracy.

4. The amplitude and phase calibration method of ground wave radar based on an oil and gas platform according to claim 1 is characterized in that: Step 3 further comprises the following steps: 3.1) First, MN array elements are selected from all array elements to form a nonlinear sub-array, where N>0 and N≤M-3. The reference array element is removed, and all possible phase error values ​​are searched for the remaining array elements in the nonlinear sub-array in the range [-π,π) with a fixed step size, and a set of phase error values ​​is determined; 3.2) Based on the phase error value, the covariance matrix of each single arrival angle signal snapshot matrix is ​​calculated and the eigenvalue decomposition is performed. When the number of signal sources is 1, the noise subspace is directly obtained. 3.3) Using the determined phase error value and noise subspace, the arrival angle θ corresponding to each single arrival angle signal and its corresponding MUSIC spectrum value are obtained through the MUSIC expression search; 3.4) Accumulate the inverse of the MUSIC spectrum values ​​corresponding to all single arrival angle signals to obtain the value of the cost function P; 3.5) Traverse all possible phase error combinations in the range of [-π,π), repeat steps 3.2)-3.4), and calculate the values ​​of all cost functions P; when the P value is the smallest, obtain the initial phase error value and single arrival angle signal angle of the MN array elements; 3.6) Based on the sub-array with determined phase error, gradually add an array element to form a new sub-array, and set the phase error of the newly added array element to be 3.7) Yes Traverse in the range [-π,π) with a fixed step size, keep the previously calculated array element column phase error unchanged, substitute it into the MUSIC summation cost function, when P reaches the minimum value, the corresponding That is the initial value of the phase error of the array element; 3.8) Repeat steps 3.6)-3.7) until the initial phase error values ​​of all P array elements are obtained.

5. According to any one of the ground wave radar amplitude and phase calibration methods based on oil and gas platforms described in claims 1 to 4, it is characterized by: Step 4 further includes the following steps: 4.1) First, set the hyperparameter values ​​of the Adadelta algorithm, then substitute the previously obtained initial phase error value and single angle of arrival signal angle into the MUSIC cost function to calculate the initial cost function value; 4.2) The MUSIC cost function takes the partial derivative of the phase error and obtains the gradient 4.3) According to the Adadelta optimization algorithm, the current g t Bring in the update rule to update the next phase error value 4.4) According to the updated phase error Search the corresponding angle θ with the MUSIC expression, and then recalculate the cost function value P t ; 4.5) Compare the current cost function value P t and the previous cost function value P t-1 If P t-1 -P t >ε, ε is a minimum value greater than 0, and the iteration continues; 4.6) Repeat steps 4.2) to 4.6) until the cost function value P no longer decreases significantly, that is, P t-1 -P t <∈; when the convergence condition is met, the loop is exited, and the phase error at this time This is the final phase error value.