Study On Near-field Beamforming And Node Intelligent Selection Of Ultra-sparse Array

The ultra-sparse array adopts the form of wide-area transceiver nodes(such as square-square-scale array),and its array aperture is large,which makes it have high spatial resolution and strong battlefield survivability compared to dual/multi-base radar.The centralized transmission and reception,ultra-sparse array sparsity is several dozen times that of the traditional dense array,easy to install,easy to deploy on the existing infrastructure,to some extent overcome the influence of the traditional array by terrain constraints,ultra-sparse array is sent and received Centralized decentralized,and through decentralized node arrangement,its transmission and reception angles are numerous,making the ultra-sparse array transceiver work system very flexible,anti-radiation and anti-interference ability is greatly enhanced,and has the potential to greatly improve anti-stealth capability;In addition,the ultra-sparse array has a wide range of nodes,and compared with the centralized radar,each node has small transmission power,low side lobes,and good anti-interception ability.However,although the ultra-sparse array has many of the above advantages,it brings the following challenges.Generally,the target within the line of sight range is a near-field detection problem for our ultra-sparse array,not only with the target's azimuth pitch angle.Relevant,and related to the distance of the target;so we have to study the near-field beamforming technology of ultra-sparse arrays.In addition,a universal broadband RF receiving/transmitting node connected to an agile network with an integrated time-frequency space reference is redundantly distributed for tasks such as single detection,dry,probe,and communication,and needs to be called for different nodes according to specific task calls.Configuration and invocation,thus making the ultra-sparse array face the challenge of enriching the resource "choosing confusion",that is,how to select the optimal detection resource to meet the detection task requirements and reduce the pressure of subsequent data transmission and processing.In view of the above existing problems,combined with radar detection tasks,this thesis studies the near-field beamforming and intelligent resource scaling of ultra-sparse arrays.Firstly,the parameter analysis of the phased array radar system is carried out,and the array size is determined.Then,the study on the ultra-sparse array near-field beamforming is carried out in combination with the terrain of a certain area in the southeast coast of China.Based on this,the distance and height dependence are carried out.Analysis;then,when considering different target azimuths,the optimal array required is different.In order to meet the requirements of omnidirectional coverage and multi-tasking(such as radar detection or electronic reconnaissance,etc.),ultra-sparse arrays face rich node resources.Choosing the puzzle of confusion.Aiming at the problem of large amount of data and high system complexity caused by a large number of nodes in ultra-sparse array radar,this thesis proposes an intelligent selection method for ultra-sparse nodes based on enhanced learning Monte Carlo tree search tree(like Alpha Go).The Monte Carlo tree search is used to obtain the high probability characteristics of the global optimal solution.The beam space gain maximization objective function is designed to optimize the selection of some nodes to reduce the system complexity under the condition of no loss of spatial gain.The Monte Carlo tree search algorithm combines the cutting-edge technology of machine learning and artificial intelligence,and has the advantage of better obtaining the global optimal solution compared with traditional intelligent optimization algorithms(such as genetic algorithm and simulated annealing).Firstly,the principle of Monte Carlo algorithm is introduced.Then,the spatial gain maximization objective function is designed.Then the Monte Carlo tree algorithm is used to realize the node selection.Finally,the beamforming performance of the selected node and the full-dimensional node is analyzed and compared.The simulation results show that the cost function is obtained close to the global optimal node selection combination.In principle,the algorithm establishes an asymmetric search tree based on a large number of training samples through the strategy network.The search tree obtains a large number of different selection results according to the target orientation and characteristics,and scores the results through the value network.The merits of each selection result will affect the next decision-making process.Finally,after a large number of training iterations,under the premise of satisfying the constraints,the node combination that makes the cost function close to the global optimal is selected.By filtering the finite resource nodes,the algorithm effectively compresses the data volume of the nodes and reduces the signal processing pressure;reduces the spatial gain loss of the radar detection and avoids the waste of node resources.