Stability is one of the most important issues in designing recursive digital adaptive filters. It is well known that the conventional transfer coefficient (or direct-form) Parameterization yields a simple implementation but suffers from potential instability problems. In this paper, based on an input-balanced realization a new adaptive filter structure is developed. It is shown that this new Parameterization has some nice numerical properties and possesses very good convergence behavior. More importantly, with this Parameterization the stability can be ensured easily and no stability monitoring is needed.

In this correspondence, a model for multicomponent signal representation is proposed. This model can avoid the difficulty in dealing with the time-varying phases and allow us to carry out an optimization procedure easily such that the synthetic signal can be made as close to the original one as possible. This model is successfully applied to speech analysis/synthesis.