観測雑音を考慮したブロック直交射影アルゴリズムの特性改善に関する一考察

抄録

Much work has been done, concerning the adaptive algorithms to adjust filter's coefficients. Among them, Block Orthogonal Projection Algorithm (BOPA) already proposed by authors is effective one, from viewpoints of convergence speed and computational requirements. Unfortunately, the convegence characteristics of BOPA lower down, when the BOPA is applied to the important system of use in which the input signal is colored one and observed noises are added at output of unknown sysycem.
First, this paper makes it clear that the convergence characteristic of BOPA can be strongly influenced by the dispersion of singuar value distribution of input state matrix at any block, and secondly presents a block adaptive algorithm based on singular value decomposition, which can realize effective parameter estimation under the condition mentioned above. The proposed algorithm can obtain stable convegence characteristics by putting some smaller singular values of state matrix into zero. Furthermore, it is shown that the proposed algorithm is capable for application to practical systems by computer simulations.