Parallel filter structures for RLS-type blind equalization algorithms

It is shown that several RLS-type blind equalization algorithms, e.g., decision-directed schemes as well as 'orthogonalized' constant modulus algorithms, possess a common algorithmic structure and are therefore straightforwardly implemented on an earlier developed triangular array (filter structure) for recursive least squares estimation with inverse updating. While the computational complexity for such algorithms is O(N2), where N is the problem size, the throughput rate for the array implementation is O(1), i.e., independent of the problem size. Such a throughput rate cannot be achieved with standard (Gentleman-Kung- type) RLS/QR-updating arrays because of feedback loops in the computational schemes.