CN104950290A - Large-scale phased-array antenna sub array division method based on weighted K average value clustering - Google Patents

Abstract

The invention discloses a large-scale phased-array antenna sub array division method based on weighted K average value clustering. When the method is used, under the condition of nonuniform array element stage weighting, the smaller weight vector approximate error is obtained, so that the static different wave beams with better performance can be integrated. According to the method, firstly, a subarray single pulse signal processing module is built; then, under the weight vector approximate criterion, the subarray optimization sum difference wave beam integration problem is converted into the reference sum different weight ratio clustering problem; the relationship between the weight vector approximate error and the clustering error is analyzed; finally, under the nonuniform array element stage weight condition, a weighted K average value clustering algorithm is adopted for optimizing the subarray structure; a least square method is used for optimizing the subarray weighing. Under the nonuniform array element stage weighting condition, the method provided by the invention can obtain smaller weight vector approximate error than a traditional clustering algorithm, and the static difference wave beam with better performance can be integrated.

Description

Based on the massive phased array antenna Antenna Subarray Division of weighting K mean cluster

Technical field

The present invention relates to Large Phased Array Radar technical field, be specifically related to a kind of massive phased array antenna Antenna Subarray Division based on weighting K mean cluster.

Background technology

In Large Phased Array Radar array antenna, generally there are hundreds of and even several ten thousand antenna elements, if fully adaptive digital beam froming will be realized, need to process separately the Received signal strength of each array element, so system can become very complicated, and hardware cost is very high.And the dimension of adaptive algorithm and complexity are very large, are difficult to requirement of real time.For this situation, often need to carry out Subarray partition to large-scale array.Subarray partition all array element is combined into several submatrixs on a small scale, then on Subarray, completes partial adaptivity process, thus reduce the required port number of reception and adaptive algorithm dimension, reduces hardware cost and realize difficulty, accelerating algorithm the convergence speed.The quality of Subarray partition structure directly affects the performance of the signal transacting such as Wave beam forming, Adaptive Anti-jamming, under certain condition (as given in array element number, array element arrangement, submatrix number and performance index), the handling property that the Subarray partition scheme how searching out a kind of optimum carrys out promotion signal is problem demanding prompt solution.

Under monopulse application background, the object of Subarray partition is the He er bu tong in order to comprehensively go out best performance, thus improves the precision of angle measurement.The basic framework of monopulse process forms optimum and wave beam in array element level, and the difference beam that Subarray is formed suboptimum, to approach the reference difference wave beam of array element level, therefore needs to be optimized subarray configuration and submatrix weighting to improve difference beam performance.Under weight vector approaches criterion, the optimization of subarray configuration can be converted into the clustering problem of reference and difference weighted ratio.When the weighting of array element level is even (being 1 entirely), weight vector approximate error equal Euclidean distance estimate under cluster error, use traditional clustering algorithm can obtain minimum approximate error.When the weighting of array element level is uneven (as carrying out Sidelobe Taylor weighting), weight vector approximate error no longer equals cluster error, uses traditional clustering algorithm that approximate error can not be made to minimize completely, have impact on the directional diagram performance of difference beam.

Summary of the invention

In view of this, the invention provides a kind of massive phased array antenna Antenna Subarray Division based on weighting K mean cluster, in the uneven situation of array element level weighting, less weight vector approximate error can be obtained, thus comprehensively can go out the better static difference wave beam of performance.

Massive phased array antenna Antenna Subarray Division based on weighting K mean cluster of the present invention, comprises the steps:

Step 1, sets up Subarray single pulse signal transaction module;

Step 2, approaches criterion according to the weight vector under 2 norm meanings, and subarray configuration Partitioning optimization problem can be described as follows:

$\begin{array}{c}\min \underset{n=1}{\overset{N}{\mathrm{\Σ}}}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{\left|\right|{(\frac{w_{\mathrm{ele}\_\mathrm{refn}}^{\mathrm{da}}}{w_{\mathrm{ele}\_\mathrm{refn}}^s},\frac{w_{\mathrm{ele}\_\mathrm{refn}}^{\mathrm{de}}}{w_{\mathrm{ele}\_\mathrm{refn}}^s})}^T-\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{\mathrm{\δ}}_{\mathrm{nl}}{(w_{\mathrm{subl}}^{\mathrm{da}},w_{\mathrm{subl}}^{\mathrm{de}})}^T\left|\right|}^2\\ =\min \underset{l=1}{\overset{L}{\mathrm{\Σ}}}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{\left|\right|h_n-c_l\left|\right|}^2\right)\\ =\min \underset{l=1}{\overset{L}{\mathrm{\Σ}}}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}\mathrm{dist}(h_n,c_l)\right)\end{array}$

Wherein represent reference and the difference weighted ratio vector of the n-th array element; represent the weight vectors of l submatrix; δ _nlrepresent dirichlet function, meet ${\mathrm{\δ}}_{\mathrm{nl}}=\left\{\begin{array}{cc}1,& g_n=l\\ 0,& g_n\≠l\end{array}\right.;\mathrm{dist}(h_n,c_l)={\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{\left|\right|h_n-c_l\left|\right|}^2$ Represent the vector h after weighting _nto c _lbetween distance; represent that the reference of the n-th array element and beam weighting, orientation are to reference difference beam weighting and pitching to reference difference beam weighting respectively; represent that the orientation of l submatrix is to the pitching of Subarray weighted sum to Subarray weighting respectively; g _n=l represents that the n-th array element belongs to l submatrix; N=1,2 ..., N; L=1,2 ..., L; N is array element sum; L is submatrix number; || || ²represent 2 norms;

Step 3, when the weighting of array element level is evenly time, use traditional clustering algorithm to solve the optimization problem of Subarray partition, obtain final Subarray partition result;

When the weighting of array element level is uneven namely time, use weighting K means clustering algorithm to be optimized subarray configuration, concrete steps are as follows:

Step 3.1, computing reference and difference weighted ratio vector set H={h ₁..., h _n..., h _n, wherein $h_n={(\frac{w_{\mathrm{ele}\_\mathrm{refn}}^{\mathrm{da}}}{w_{\mathrm{ele}\_\mathrm{refn}}^s},\frac{w_{\mathrm{ele}\_\mathrm{refn}}^{\mathrm{de}}}{w_{\mathrm{ele}\_\mathrm{refn}}^s})}^T$ As eigenvector to be clustered;

Step 3.2, makes iteration count t=0; The initial division of subarray configuration is g ⁽⁰⁾, g={g _n=l, l=1,2 ..., L, n=1,2 ..., N}; Initial L cluster centre of subarray configuration is wherein, the interior numerical value of subscript () is iterations;

$c_l^{\left(0\right)}=\frac{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}^{\left(0\right)}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{h}_n\right)}{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}^{\left(0\right)}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2\right)},l=1,...,L;$

Step 3.3, calculates each array element characteristic of correspondence vector h _ndistance between all L cluster centre, according to Nearest neighbor rule, the cluster result of the n-th array element for h _napart from all cluster centre c ₁, c ₂c _lin the class at nearest cluster centre place;

Step 3.4, according to the cluster result of array element each in step 3.3 recalculate new cluster centre $C^{(t+1)}={(c_1^{(t+1)},...,c_l^{(t+1)},...,c_L^{(t+1)})}^T:$

$c_l^{(t+1)}=\frac{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}^{(t+1)}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{h}_n\right)}{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}^{(t+1)}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2\right)},l=1,...,L;$

Step 3.5, t, from adding 1, repeats step 3.3, step 3.4 until cluster result no longer changes or iterations exceedes given maximum iteration time;

The cluster result g of last iteration gained is optimum Subarray partition structure; The cluster centre of last iteration gained is optimum Subarray weighting.

Further, initial division g ⁽⁰⁾obtained by taper function quantization method or adopt random device to produce.

Beneficial effect:

The inventive method is under weight vector approaches criterion, Subarray partition optimization problem is converted into the clustering problem of reference and difference weighted ratio, problem is simplified thus clustering algorithm can be used to solve, and to go forward side by side row iteration optimization by being weighted process to the clustering criteria in traditional K mean algorithm and cluster centre, less weight vector approximate error can be obtained in the uneven situation of array element level weighting; Use least square method to be optimized Subarray weighting simultaneously, calculate easy; With under the optimum prerequisite of wave beam, finally acquired can better difference beam, effectively improve Monopulse estimation precision.

Accompanying drawing explanation

Fig. 1 is Subarray signal transacting model schematic.

Fig. 2 is weighting K means clustering algorithm process flow diagram.

Fig. 3 is Circular Aperture array elements distribution schematic diagram.

Fig. 4 is that the subarray configuration that different demarcation method obtains compares; Wherein figure (a) evenly divides, and (b) is that taper function quantizes to divide, and (c) is the division of K mean cluster, and (d) is that weighting K mean cluster divides;

Fig. 5 is that the static difference beam pattern sectional view that different demarcation method obtains compares; Wherein (a) is for orientation is to difference beam sectional view, and (b) is for pitching is to difference beam sectional view.

Embodiment

To develop simultaneously embodiment below in conjunction with accompanying drawing, describe the present invention.

The invention provides a kind of massive phased array antenna Antenna Subarray Division based on weighting K mean cluster, first set up Subarray single pulse signal transaction module; Then, under weight vector approaches criterion, the optimum He er bu tong synthtic price index of Subarray is converted into the clustering problem of reference and difference weighted ratio, analyzes the relation between weight vector approximate error and cluster error; Last in the uneven situation of array element level weighting, adopt weighting K means clustering algorithm to be optimized subarray configuration, use least square method to be optimized submatrix weighting.In the uneven situation of array element level weighting, the present invention can obtain less weight vector approximate error compared to traditional clustering algorithm, and comprehensively can go out the better static difference wave beam of performance.

Specific implementation step is as follows:

Step one, set up Subarray single pulse signal transaction module.

Known array element adds up to N, needs the submatrix number divided to be L.Subarray partition structure vector g can be expressed as g={g _n=l, l=1,2 ..., L, n=1,2 ..., N}, wherein g _n=l represents that the n-th array element belongs to l submatrix.N × L ties up submatrix transition matrix T and can be expressed as

T＝diag(Φ ₀)diag(w _ele)T ₀(1)

Φ in formula ₀and w _elerepresent simulation phase shift and the weighting of array element level respectively, submatrix generator matrix T ₀meet:

$T_0\left[\mathrm{nl}\right]=\left\{\begin{array}{cc}1& g_n=l\\ 0& g_n\≠l\end{array}\right.,n=\mathrm{1,2},...,N;l=\mathrm{1,2},...,L---\left(2\right)$

So final beam pattern can be expressed as:

W in formula _{ele_eq}=TW _subfor the array element level weight vectors of equivalence, be two-dimensional guide vector on direction, position angle and the angle of pitch is represented respectively with θ.In order to simplify analog weighted network, typical Subarray monopulse process framework is formed and wave beam in array element level subarray forms orientation to difference beam with pitching to difference beam therefore the Subarray weighting network in model meets

$W_{\mathrm{sub}}=[w_{\mathrm{sub}}^s,w_{\mathrm{sub}}^{\mathrm{da}},w_{\mathrm{sub}}^{\mathrm{de}}]---\left(4\right)$

In formula represent that Subarray and beam weighting vector, orientation are tieed up to difference beam weight vectors and pitching to difference beam weight vectors in L × 1 respectively.Usually array element level weighting w is chosen _elefor with beam reference weighting namely thus meet with the weighting of wave beam Subarray

Step 2, under weight vector approaches criterion, set up the subarray configuration Optimized model based on cluster;

Array element level uses " direct and poor weighted method " usually select when carrying out Monopulse estimation with reference to weight vectors (as adopted Taylor weighting with wave beam, difference beam adopt Bayliss weighting), might as well set the desired reference weighting network of array element level needs realization as

$W_{\mathrm{ele}\_\mathrm{ref}}=[w_{\mathrm{ele}\_\mathrm{ref}}^s,w_{\mathrm{ele}\_\mathrm{ref}}^{\mathrm{da}},w_{\mathrm{ele}\_\mathrm{ref}}^{\mathrm{da}}]---\left(5\right)$

Wherein represent that array element level reference and beam weighting vector, orientation are tieed up to the pitching of reference difference beam weighting vector to reference difference beam weighting vector in N × 1 respectively.

The array element level weighting W that consideration weight vector is equivalent under approaching criterion and submatrix framework _{ele_eq}optimum array element level is approached with reference to weighting W under norm meaning _{ele_ref}, so the optimization problem of Subarray partition can be described as:

$\underset{w_{\mathrm{ele}},T_0,W_{\mathrm{sub}}}{\min }\left|\right|W_{\mathrm{ele}\_\mathrm{eq}}-W_{\mathrm{ele}\_\mathrm{ref}}\left|\right|=\underset{w_{\mathrm{ele}},T_0,W_{\mathrm{sub}}}{\min }\left|\right|{\mathrm{TW}}_{\mathrm{sub}}-W_{\mathrm{ele}\_\mathrm{ref}}\left|\right|$

(6)

s.t.T＝diag(Φ ₀)diag(w _ele)T ₀

Wherein by optimized variable w _ele, T ₀and W _subthus make the Norm minimum of the approximate error of weight vector, for convenience of calculating, generally get 2 norms.Because array element level is weighted to and beam reference weighting thus with the weighting of wave beam Subarray what so need optimization only has T ₀, three variablees.Optimization problem now under 2 norm meanings can be described as:

$\begin{array}{c}\underset{T_0,w_{\mathrm{sub}}^{\mathrm{da}},w_{\mathrm{sub}}^{\mathrm{de}}}{\min }{\left|\right|T(w_{\mathrm{sub}}^{\mathrm{da}},w_{\mathrm{sub}}^{\mathrm{de}})-(w_{\mathrm{ele}\_\mathrm{ref}}^{\mathrm{da}},w_{\mathrm{ele}\_\mathrm{ref}}^{\mathrm{de}})\left|\right|}^2\\ =\underset{T_0,w_{\mathrm{sub}}^{\mathrm{da}},w_{\mathrm{sub}}^{\mathrm{de}}}{\min }{\left|\right|T{w}_{\mathrm{sub}}^{\mathrm{da}}-w_{\mathrm{ele}\_\mathrm{ref}}^{\mathrm{da}}\left|\right|}^2+{\left|\right|T{w}_{\mathrm{sub}}^{\mathrm{de}}-w_{\mathrm{ele}\_\mathrm{ref}}^{\mathrm{de}}\left|\right|}^2\end{array}---\left(7\right)$

$s.t.T=\mathrm{diag}\left({\mathrm{\Φ}}_0\right)\mathrm{diag}\left(w_{\mathrm{ele}\_\mathrm{ref}}^s\right)T_0$

Work as beam position i.e. Φ ₀when=1, the objective function optimized in (7) formula is out of shape further and can be obtained

$\begin{array}{c}\min \underset{n=1}{\overset{N}{\mathrm{\Σ}}}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{\left|\right|{(\frac{w_{\mathrm{ele}\_\mathrm{refn}}^{\mathrm{da}}}{w_{\mathrm{ele}\_\mathrm{refn}}^s},\frac{w_{\mathrm{ele}\_\mathrm{refn}}^{\mathrm{de}}}{w_{\mathrm{ele}\_\mathrm{refn}}^s})}^T-\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{\mathrm{\δ}}_{\mathrm{nl}}{(w_{\mathrm{subl}}^{\mathrm{da}},w_{\mathrm{subl}}^{\mathrm{de}})}^T\left|\right|}^2\\ =\min \underset{l=1}{\overset{L}{\mathrm{\Σ}}}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{\left|\right|h_n-c_l\left|\right|}^2\right)\\ =\min \underset{l=1}{\overset{L}{\mathrm{\Σ}}}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}\mathrm{dist}(h_n,c_l)\right)\end{array}---\left(8\right)$

Wherein represent reference and the difference weighted ratio vector of the n-th array element, represent that the reference of the n-th array element and beam weighting, orientation are to reference difference beam weighting and pitching to reference difference beam weighting respectively; represent the weight vectors of l submatrix, represent that the orientation of l submatrix is to the pitching of Subarray weighted sum to Subarray weighting respectively; δ _nlrepresent that dirichlet function meets ${\mathrm{\δ}}_{\mathrm{nl}}=\left\{\begin{array}{cc}1,& g_n=l\\ 0,& g_n\≠l\end{array}\right.,$ G _n=l represents that the n-th array element belongs to l submatrix; represent the vector h after weighting _nto c _lbetween distance.

Finally, weight vector approximate error can be converted into the product of traditional cluster error and array element level weighted value square.

When the weighting of array element level is evenly time, the dist (h now in formula (8) _n, c _l)=|| h _n-c _l|| ², from the angle of cluster, h _nthe eigenvector of cluster can be regarded as, c _lcan cluster centre being regarded as, so making weight vector approximate error minimum equivalent in making cluster error minimum, traditional clustering algorithm (as K mean cluster) so just can be used to solve the optimization problem of Subarray partition.

Step 3, in the uneven situation of array element level weighting, weighting K means clustering algorithm is used to be optimized subarray configuration;

When the weighting of array element level is uneven namely time (as used Sidelobe windowing), now $\mathrm{dist}(h_n,c_l)={\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{\left|\right|h_n-c_l\left|\right|}^2\≠{\left|\right|h_n-c_l\left|\right|}^2,$ Consider weighted term existence, use traditional clustering algorithm error can not be made to minimize completely.For the situation that the weighting of array element level is uneven, use for reference the basic thought of traditional K means clustering algorithm, the key step of the Antenna Subarray Division based on weighting K mean cluster that the present invention carries is as follows:

(1) computing reference and difference weighted ratio vector set H={h ₁..., h _n..., h _n, wherein as feature vector set to be clustered;

(2) make iteration count t=0, given maximum iteration time is t _max.The cluster result using taper function quantization method to obtain or to produce at random is as initial division g ⁽⁰⁾, simultaneously according to subarray configuration initialization L cluster centre wherein initial cluster center computing formula be

$c_l^{\left(0\right)}=\frac{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}^{\left(0\right)}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{h}_n\right)}{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}^{\left(0\right)}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2\right)},l=1,...,L---\left(9\right)$

(3) each array element characteristic of correspondence vector h is calculated _ndistance between all cluster centres, according to Nearest neighbor rule, the cluster result of each array element can be expressed as

$g_n^{(t+1)}=\arg \underset{l}{\min }\mathrm{dist}(h_n,c_l^{\left(t\right)}),n=1,...,N;l=1,...,L---\left(10\right)$

(4) according to the cluster result of each array element in (3) recalculate new cluster centre $C^{(t+1)}={(c_1^{(t+1)},...,c_l^{(t+1)},...,c_L^{(t+1)})}^T:$

$c_l^{(t+1)}=\frac{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}^{(t+1)}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{h}_n\right)}{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}^{(t+1)}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2\right)},l=1,...,L---\left(11\right)$

(5) upgrade iterations t=t+1, repeat (3), (4) until cluster result no longer changes or iterations exceedes given maximum iteration time.The cluster result g of last iteration gained is optimum Subarray partition structure.

Step 4, to be optimized to fixing the weighting of antithetical phrase battle array at subarray configuration;

At the given (T of Subarray partition structure ₀under condition necessarily), the problems referred to above are convex optimization problems of submatrix weighting, and now Subarray optimal weighting can use least square method to try to achieve:

W _sub＝(T ^HT) ^-1T ^HW _{ele_ref}(12)

It is easy to show that, Subarray weighting W _subequal with the cluster centre C in weighting K means clustering algorithm, therefore the cluster centre of last iteration gained is optimum Subarray weighting.

Since then, just complete the Subarray partition optimization method based on weighting K mean cluster, by revising the account form of clustering criteria and cluster centre in traditional K means clustering method, through each iteration, weight vector approximate error can successively reduce.

In order to verify performance of the present invention, the Circular Aperture planar array shown in Fig. 3 is carried out to the emulation of subarray configuration division and static difference beam pattern.Convenient in order to describe, evenly divide and be called UPM method, taper function quantizes division methods and is called NM method, and the division methods based on K mean cluster is called KCM method, and the division methods based on weighting K mean cluster that the present invention proposes is called WKCM method.Wherein simulation parameter is as shown in table 1:

Table 1

Fig. 4 is that the subarray configuration that different demarcation method obtains compares, wherein scheme (a) (b) (c) and be followed successively by UPM method, the subarray configuration that NM method and KCM method divide, the subarray configuration that figure (d) divides for WKCM method that the present invention proposes.Be not difficult to find, it is adjacent for only having figure (c) and scheming between the submatrix that divides in (d), is more easy to Project Realization.

Fig. 5 is that the static difference beam pattern sectional view that different demarcation method obtains compares, and wherein scheme (a) for orientation is to difference beam sectional view, figure (b) is for pitching is to difference beam sectional view.In order to compare algorithm performance more intuitively, corresponding difference beam directional diagram performance index are as shown in table 2, and these indexs comprise weight vector approximate error, 3dB beam angle and maximum sidelobe levels.

Table 2

As can be seen from Fig. 5 and table 2, be no matter orientation to or pitching to, the beam angle of the directional diagram that different demarcation method obtains is relatively.Compared to the division methods of other classics, the weight vector approximate error that the WKCM method that the present invention carries obtains is minimum and maximum sidelobe levels that is difference beam directional diagram is minimum, the difference beam best performance that the method obtains is described and approaches the difference beam performance of reference.

In sum, these are only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (2)

1., based on a massive phased array antenna Antenna Subarray Division for weighting K mean cluster, it is characterized in that, comprise the steps:

Step 1, sets up Subarray single pulse signal transaction module;

Step 2, approaches criterion according to the weight vector under 2 norm meanings, and subarray configuration Partitioning optimization problem can be described as follows:

$\begin{array}{c}\min \underset{n=1}{\overset{N}{\mathrm{\Σ}}}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{\left|\right|{(\frac{w_{\mathrm{ele}\_\mathrm{refn}}^{\mathrm{da}}}{w_{\mathrm{ele}\_\mathrm{refn}}^s},\frac{w_{\mathrm{ele}\_\mathrm{refn}}^{\mathrm{de}}}{w_{\mathrm{ele}\_\mathrm{refn}}^s})}^T-\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{\mathrm{\δ}}_{\mathrm{nl}}{(w_{\mathrm{subl}}^{\mathrm{da}},w_{\mathrm{subl}}^{\mathrm{de}})}^T\left|\right|}^2\\ =\min \underset{l=1}{\overset{L}{\mathrm{\Σ}}}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{\left|\right|h_n-c_l\left|\right|}^2\right)\\ =\min \underset{l=1}{\overset{L}{\mathrm{\Σ}}}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}\mathrm{dist}(h_n,c_l)\right)\end{array}$

Wherein represent reference and the difference weighted ratio vector of the n-th array element; represent the weight vectors of l submatrix; δ _nlrepresent dirichlet function, meet ${\mathrm{\δ}}_{\mathrm{nl}}=\left\{\begin{array}{cc}1,& g_n=l\\ 0,& g_n\≠l\end{array}\right.;$ $\mathrm{dist}(h_n,c_l)={\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{\left|\right|h_n-c_l\left|\right|}^2$ Represent the vector h after weighting _nto c _lbetween distance; represent that the reference of the n-th array element and beam weighting, orientation are to reference difference beam weighting and pitching to reference difference beam weighting respectively; represent that the orientation of l submatrix is to the pitching of Subarray weighted sum to Subarray weighting respectively; g _n=l represents that the n-th array element belongs to l submatrix; N=1,2 ..., N; L=1,2 ..., L; N is array element sum; L is submatrix number; || || ²represent 2 norms;

Step 3, when the weighting of array element level is evenly time, use traditional clustering algorithm to solve the optimization problem of Subarray partition, obtain final Subarray partition result;

When the weighting of array element level is uneven namely time, use weighting K means clustering algorithm to be optimized subarray configuration, concrete steps are as follows:

Step 3.1, computing reference and difference weighted ratio vector set H={h ₁..., h _n..., h _n, wherein $h_n={(\frac{w_{\mathrm{ele}\_\mathrm{refn}}^{\mathrm{da}}}{w_{\mathrm{ele}\_\mathrm{refn}}^s},\frac{w_{\mathrm{ele}\_\mathrm{refn}}^{\mathrm{de}}}{w_{\mathrm{ele}\_\mathrm{refn}}^s})}^T$ As eigenvector to be clustered;

Step 3.2, makes iteration count t=0; The initial division of subarray configuration is g ⁽⁰⁾, g={g _n=l, l=1,2 ..., L, n=1,2 ..., N}; Initial L cluster centre of subarray configuration is $C^{\left(0\right)}={(c_1^{\left(0\right)},...,c_l^{\left(0\right)},...,c_L^{\left(0\right)})}^T$ Wherein, the interior numerical value of subscript () is iterations;

$c_l^{\left(0\right)}=\frac{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}^{\left(0\right)}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{h}_n\right)}{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}^{\left(0\right)}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2\right)},l=1,...,L;$

Step 3.3, calculates each array element characteristic of correspondence vector h _ndistance between all L cluster centre, according to Nearest neighbor rule, the cluster result of the n-th array element for h _napart from all cluster centre c ₁, c ₂c _lin the class at nearest cluster centre place;

Step 3.4, according to the cluster result of array element each in step 3.3 recalculate new cluster centre $C^{(t+1)}={(c_1^{(t+1)},...,c_l^{(t+1)},...,c_L^{(t+1)})}^T:$

$c_l^{(t+1)}=\frac{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}^{(t+1)}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2{h}_n\right)}{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left({\mathrm{\δ}}_{\mathrm{nl}}^{(t+1)}{\left|\right|w_{\mathrm{ele}\_\mathrm{refn}}^s\left|\right|}^2\right)},l=1,...,L;$

Step 3.5, t, from adding 1, repeats step 3.3, step 3.4 until cluster result no longer changes or iterations exceedes given maximum iteration time;

The cluster result g of last iteration gained is optimum Subarray partition structure; The cluster centre of last iteration gained is optimum Subarray weighting.

2., as claimed in claim 1 based on the massive phased array antenna Antenna Subarray Division of weighting K mean cluster, it is characterized in that, initial division g ⁽⁰⁾obtained by taper function quantization method or adopt random device to produce.