Nonlinear adaptive prediction of speech with a pipelined recurrentneural network
IEEE Transactions on Signal Processing
Nonlinear adaptive prediction of nonstationary signals
IEEE Transactions on Signal Processing
| Hi-index | 0.00 |
A hybrid nonlinear time series predictor that consists a nonlinear subpredictor (NSP) and a linear sub-predictor (LSP) combined in a cascade form is proposed. A multilayer neural network is employed as the NSP and the algorithm used to update the NSP weights is Lyapunov stability-based backpropagation algorithm (LABP). The NSP can predict the nonlinearity of the input time series. The NSP prediction error is then further compensated by employing a LSP. Weights of the LSP are adaptively adjusted by the Lyapunov adaptive algorithm. Signals' stochastic properties are not required and the error dynamic stability is guaranteed by the Lyapunov Theory. The design of this hybrid predictor is simplified compared to existing hybrid or cascade neural predictors [1]-[2]. It is fast convergence and less computation complexity. The theoretical prediction mechanism of this hybrid predictor is further confirmed by simulation examples for real world data.