Efficient Implementation of Linear System Solution Block using LDL<SUP>T</SUP> Factorization

초록

Two implementation forms of linear system solution are described in this paper. Cholesky LLT decomposition require square roots, whereas LDLT decomposition can avoid taking square roots by one more forward-substitute computation. Through computational complexity and hardware size analysis, it is shown that matrix inversion using LDLT decomposition is faster than cholesky decomposition using square root function. Moreover the LDLT decomposition has better performance of 31.4% in hardware size.

키워드