The design of a kind of chirp-like mother wavelet by neural network

Methods are presented for adaptive design of mother wavelets for the representation and processing of chirp-like signals using neural networks. A new kind of network structure is proposed and its training algorithm is explored. The concept of "wavelet atoms" is introduced to construct a new mother wavelet. Numerical simulations show that our approach is always effective in finding a "satisfactory" mother wavelet for applications. In fact, our methods can also be used in the processing of other kinds of signals.