A similarity transformation that introduces zero coefficients in the A matrix of a local state-space (LSS) realization is presented. The transformation is orthogonal and thus preserves

. This guarantees that the number of multipliers is reduced without sacrificing zero input stability to nonlinearities satisfying

. It is shown that the transformation does not affect the sensitivity. The results are extended to the generalized asymmetric half-plane state-space model. An example of a general

second-order filter is given.