面向毫米波手势识别的非均匀虚拟阵列快速角度估计OA

Fast and accurate angle estimation can significantly enhance the performance of gesture recognition in multiple input multiple output(MIMO)millimeter-wave radar systems.However,the non-uniform configuration of the virtual array of the radar inevitably introduces array manifold mismatch,which deteriorates the direction of arrival(DOA)estimation precision.Meanwhile,the conventional two-dimensional multiple signal classification(2D-MUSIC)algorithm entails a prohibitively high computational cost,hindering its deployment in real-time applications.To address these challenges,a non-uniform array DOA(NUA-DOA)estimation algorithm was proposed.The collected time-domain echo data were firstly preprocessed via static clutter cancellation to reinforce target echoes and suppress background interference.Then,a distance-velocity dimension fast Fourier transform(FFT)was performed on the preprocessed data to obtain the target two-dimensional spectrum.A structured array manifold was then constructed by incorporating the actual spatial geometry of the non-uniform array,and covariance subspace decomposition was subsequently performed to realize decoupling and dimensionality reduction estimation of two-dimensional angular parameters.Experimental results demonstrate that,by utilizing actual non-uniform array modeling and structured dimensionality reduction search,the proposed algorithm achieves angle estimation precision comparable to that of 2D-MUSIC while reducing the average computational time by approximately one order of magnitude,thus fulfilling the real-time processing requirements of gesture recognition systems.