Two forms of Friedland's separate bias estimation algorithm with U-D f
actorization of the covariance matrices are provided. Each is suited t
o implementation in a particular computing environment. (We consider M
ATLAB and compiled computer languages.) We reduce the computation time
substantially, primarily at the time propagation stage, by using a se
parated bias formulation, while retaining the desirable numerical prop
erties of the U-D factorization. The perecentage reduction typically i
ncreases with ratio of bias state dimension to dynamic state dimension
. A numerical evaluation is given for the MATLAB algorithm.