Dimensionality-Reduced Virtual Transformation for DOA Estimation in Cylindrical Arrays

This paper addresses the challenges of high computational complexity in direction-of-arrival (DOA) estimation for cylindrical array configurations. We propose a tensor-based DOA estimation method that integrates a structured virtual array transformation with a dimensionality-reduced covariance representation. A third-order covariance tensor is first formulated within the tensor-tensor product (T-product) framework. Then, a customized compression strategy is developed to transform this covariance tensor into a compact, low-dimensional representation associated with a virtual array. The steering vector of the virtual array is mathematically characterized as the circular convolution of the uniform circular array (UCA) steering vector and the steering vector along the vertical axis. By exploiting the inherent periodicity of this virtual steering vector, the MUSIC spatial spectrum can be reformulated in terms of a discrete Fourier transform (DFT), leading to substantial computational savings. Moreover, the proposed compression matrix is specifically designed to preserve critical angular information, thereby enabling robust estimation performance even under challenging conditions with limited snapshots or low signal-to-noise ratios (SNRs). Extensive simulation results demonstrate that the proposed method achieves superior estimation accuracy and computational efficiency compared with existing subspace-based and compressed-sensing approaches, particularly in high-dimensional array scenarios.