Solution of underdetermined Sylvester equations in sensor array signal processing (Q2564894)

Motivated by a practical example in signal processing, this paper studies numerical solution of an underdetermined Sylvester equation \((AP-PB) - (CQ-QD)= R\), all matrices \(n\times n\). Use of Kronecker products enables the problem to be reformulated as the underdetermined system \(Fz=r\), with \(F\) size \(n^2 \times 2n^2\). The required solution matrices \(P\) and \(Q\) can be recovered from the least squares solution \(z\) of the above system, the solution of which requires \(O(n^6)\) flops. To reduce computational effort, after a suitable transformation of the problem, the standard conjugate gradient method, based on matrices of size \(n\times n\), is implemented in a straightforward manner. This procedure reduces the work in finding \(P\) and \(Q\) to \(O(n^5)\) flops. A numerical example completes the paper.