| The primary objective of this research is to improve the signal processing strategy applied in cochlear implant (CI) based on the fundamental work on modeling the active cochlea. The major original contributions in this thesis include that: (1) A cochlear model for the generation of otoacoustic emissions (OAES) is developed with physiological basis; (2) The active mechanisms of a cochlea is formulated more exactly; (3) A new time-frequency signal processing method, bionic wavelet transform (BWT), is developed based on our modeling of the behaviors of a cochlea; and (4) The potential application of BWT to the signal processing for the cochlear implants is explored.;In order to better understand the working mechanisms of a human's cochlea, first of all, in this thesis, a cochlear model is developed. This model, combining the characteristics of the classical transmission-line passive cochlear models, active cochlear models, active cochlear partition (CP) models, and single outer hair cell (OHC) models, is designed to have a clear physiology basis.;Based on this model, the cochlear active mechanisms are formulated descriptively. To be a practical mathematical cochlear model, the active mechanisms are simplified as two functions describing the equivalent resistance and capacitance of CP, respectively. This mathematical model successfully describes many active phenomena of cochlea, including the "tall and broad" basilar membrane response and OAEs, which strongly supports our simplified mathematical formulation and our basic understanding of the active mechanisms of a cochlea.;The two functions representing the active mechanisms of a cochlea are introduced into the traditional wavelet transform (WT), which results in a new time-frequency signal processing method---BWT---that mimics functions of a cochlea. Theoretical studies on the admissible condition, the inverse BWT (IBWT), and the constraint on constructing a BWT mother function are performed in the signal processing regime. The computerized simulations are conducted on both the simulated signals and the real speech signals. The convincing experimental results show that BWT does achieve successfully the optimal adjustment of time and frequency resolutions along these two axes simultaneously. In other words, a better "trade-off" between the time and frequency resolutions is resolved. The BWT can also represent the signal in time-frequency domain with concentric energy distribution. To be a practical method, the fast algorithm for BWT is also proposed.;The potential application of BWT to signal processing in Cl is then explored. The results of neural network simulations show that by using the BWT, the recognition rates of both vowels and consonants are improved by a factor over 20%. The BWT also has the abilities in (1) reducing the required number of channels selected, and (2) immunizing to the noise.;In summary, the research conducted above not only covers the extensive fundamental research work on modeling the active cochlea and developing the new signal processing method, but also includes practical considerations on the Cl application. Expectations for further studies are suggested at the end of this thesis. |
