The Least Squares Lattice Algorithm: An Alternate Derivation with a Discussion of Numerical Conditioning.
Final technical rept. Nov 83-Jan 84,
NAVAL OCEAN SYSTEMS CENTER SAN DIEGO CA
Pagination or Media Count:
A new derivation of the least squares lattice algorithm is given in which the filter coefficients are solved for directly by a Gram-Schmidt orthogonalization of the data. This approach shows that the unnormalized and normalized least squares lattice algorithms have the same numerical conditioning as the so-called normal equations associated with least squares problems. Thus, contrary to some of the literature, the normalized lattice is not numerically superior to the unnormalized lattice for ill-conditioned problems. Author
- Statistics and Probability