Transform domain neural filters

A neural filter is effective for the system identification of a nonlinear system and the noise reduction in a nonlinear signal. However, the neural filter requires large number of iterations for convergence. This paper presents new structures of the multi-layered neural filter (Transform Domain Neural Filter; TDNF) where the orthonormal transform is introduced to accelerate the convergence speed. In the TDNF (i), the input signal is transformed by the orthonormal transform, then led to the input layer of the neural filter. The TDNF (ii) adopts the orthonormal transform in all inter-layers. Through the computer simulation in the nonlinear system identification, it is confirmed that the introduction of the orthonormal transform is effective for the speed-up of convergence in the neural filter