Research On Fast Antenna Array Beamforming Under Multiple Constraints

Antenna array beamforming technique has been widely used in radar,sonar and communication systems.However,in order to improve the anti-interference ability of the antenna array,such as generating the array pattern with deep nulls,low sidelobe and mainlobe preservation,it is necessary to establish a beamforming model with various array requirements.Consequently,the beamforming model turns out to be a complex nonlinear problem with multiple targets or constraints.For the nonlinear beamforming problem under multiple constraints or multiple requirements,the computation load will increase with the number of model constraints and the array elements,resulting in a long response time.And the long responding time will limit the application of antenna array in the rapid changing environment.Therefore,it is necessary to construct an accurate beamforming algorithm with fast response.In this dissertation,the array beamforming under multiple constraints is firstly constructed to satisfy the specific pattern requirement,and the optimization algorithm is studied to get a group of data set.Then,the neural network model is used to simulate the beamforming problem through off-line training.The fast on-line beamforming can be achieved by the trained neural network after off-line training.The main contributions of the dissertation are summarized as follows:(1)A self-adaptive constrained differential evolution algorithm is presented to solve the multi-constrained beamforming problem.For the sake of controlling the array pattern precisely,the beamforming is converted to the constraint problem,in which the null depth and the main lobe gain are set as the constraints.Then,the self-adaptive constrained differential evolution algorithm is designed to solve the constraint beamforming model.By adding the constraint violation,which is the sum of the violation of all constraint functions,the algorithm can distinguish the feasible solutions of the constraint problem.And through adaptively selecting the algorithm parameters and mutation strategy,local convergence is avoided and the convergence speed is improved.Compared to the conventional differential evolution algorithm,the proposed method successfully forming patterns under multi-constraints.(2)A phase only antenna array nulling model based on multi-objective radial basis function neural network is proposed.In order to realize the array nulling by the radial basis function neural network,the training and testing framework of radial basis function neural network is established,and then two different traditional radial basis function neural network are built to fit the array nulling model.Lastly,in view of the difficulties of the evaluation of the radial basis function neural network,a multi-objective radial basis function neural network is constructed.The multi-objective differential evolution algorithm is designed to optimize the centers and width of the network,in which the performance of the network is evaluated by the mean square error,the number of the hidden neurons and the failure rate of the pattern parameters realization.Simulation results show that the proposed beamformer has an accurate approximation and a good generalization with fewer hidden neurons.(3)The array nulling model with low sidelobe based on input pre-process coding multi-layer neural networks is proposed.For the sake of responding to different number of interferences in one network,the interference directions and the desired signal direction are converted to same size codes,which forms the input pre-process coding block before the neural network.Firstly,the array nulling model with low sidelobe is transformed into a convex model and solved using convex optimization,which achieves the efficient datasets construction.Then,based on the input pre-process coding process,the low-sidelobe interference nulling model based on a multilayer neural network,the low sidelobe monopulse array nulling model based on deep neural network and the minimized mainlobe array synthesis model based on deep neual network are established respectively.Simulation results show that the neural network works well and shows good generalization ability.With one neural network model,the array could respond to different scenarios,which combined with the advantage of the instant response(only a few milliseconds)makes the neural network-based beamformer very useful in practice.(4)A robust phase-only adaptive beamformer based on the deep learning is proposed.Firstly,in order to obtain a large enough database for the deep neural network efficiently,the phase-only array synthesis model with nulling operation and sidelobe control is relaxed to the convex problem and solved by direct iterative rank refinement.Then,the deep neural network is devised to emulate the phase array nulling behavior,in which the ideal interference covariance matrix is designed as the input of the network.After that,to improve the beamformer performance in practical application under low snapshots,the mutual coupling between the antenna array elements is considerd,and then the robust phase-only adaptive beamforming model that combining the denoising autoencoder and the deep neural network is established.In the model,the autoencoder is unified to reduce noise in the covariance matrix and extract the feature of the covariance matrix,the deep neural network is connected with the encoder to substitute the beamformer.Numerical results verified the robustness of the proposed model.