Artificial Neural Network-based Adaptive Filtering

The thesis based on the research subject about the adaptive filter realtime procede, mainly study using ANN to reach adaptive filter.In the paper,Artificial neural network (ANN) with high speed parallel process ability is introduced to meet the need of the adaptive filter real-time procede, thus changes the traditional serial arithmetic, and raises the operation speed greatly. Theory of the adaptive filter by ANN is analysed ,then a bond network is built to implement stable alternate structure adaptive filters for high speed real-time applications. Especially a variable step modified model is put forword ,and successfully applied to adaptive prediction with lattice filer and adaptive identification with lattice-ladder filter or Power wave digital filter(PWDF). Illustrative examples have shown that this parallel method results in much improved adaptation properties compared with the already existing approaches. The adaptive filter implement methed based on ANN will bring good reference to relevant intergrated circuit.On the basis of recursive gradients' algorithm,an adaptive PWDF is developed. The algorithm is more tractable to achieve the stability and suitable to construct the hardware. The robust performance of the algorithm is verified by simulation with the application in adaptive system identification.