Abstract

An algorithm of angular superresolution based on the Cholesky decomposition, which is a modification of the Capon algorithm, is proposed. It is shown that the proposed algorithm makes it possible to abandon the inversion of the covariance matrix of input signals. The proposed algorithm is compared with the Capon algorithm by the number of operations. It is established that the proposed algorithm, with a large dimension of the problem, provides some gain both when implemented on a single-threaded and multithreaded computer. Numerical estimates of the performance of the proposed and original algorithm using the Compute Unified Device Architecture (CUDA) NVidia parallel computing technology are obtained. It is established that the proposed algorithm saves GPU computing resources and is able to solve the problem of constructing a spatial spectrum when the dimensionality of the covariance matrix of input signals is almost doubled.

中文翻译：

---

基于Cholesky分解的角度超分辨率算法及其基于并行计算技术的实现

摘要

提出了一种基于Cholesky分解的角度超分辨率算法，该算法是Capon算法的改进。结果表明，该算法可以放弃输入信号协方差矩阵的求逆。所提出的算法与 Capon 算法在运算次数上进行了比较。可以确定的是，所提出的算法对于问题的大维度来说，在单线程和多线程计算机上实现时都提供了一些增益。使用统一计算设备架构 (CUDA) NVidia 并行计算技术对所提出的算法和原始算法的性能进行了数值估计。结果表明，该算法节省了GPU计算资源，并且能够解决输入信号协方差矩阵维数几乎加倍时的空间谱构建问题。