Let $A\in\mathbb{C}^{m,n}$. I am looking to compute $x\in\mathbb{C}^n$ such that for $Ax=y$ it holds:
$|y_i|=1,~i=1,...,m$.
In practice, the columns of $A$ are transfer functions of $n$ sources, sampled at $m$ points in space. I would like to determine a somewhat optimal mixture of the sources such that the magnitude of the transfer function is close to 1 in a least-square sense. So I am looking to minimize:
$\sum_{i=1}^m(|\sum_{j=1}^nA_{i,j}x_j|-1)^2=\sum_{i=1}^m(|<a_i,\bar{x}>|-1)^2$,
where $a_i$ are the rows of $A$.
Any help at approaching this is greatly appreciated. Thank you!