Real-time implementation of recursive DFT on PDA platforms for cochlear implant studies.

摘要

The signal processor used in a typical cochlear implant is programmed to emulate the function of the inner ear by dividing an input speech signal into different frequency bands in order to obtain a signal envelope or strength in each band. There are two main strategies that are adopted in commercially available cochlear implants. The first strategy utilizes a bank of bandpass filters or a filterbank to divide the signals into different bands, and the second strategy utilizes Fast Fourier Transform (FFT). In the second strategy, the computation of FFT needs to be carried out at high rates in order to have high stimulation rates of implanted electrodes. This creates an implementation bottleneck on portable platforms that have relatively limited resources as compared to desktop platforms. In this thesis, a recursive DFT update method is utilized to address this bottleneck problem. The Discrete Fourier Transform (DFT) of an input windowed speech signal is updated at the input sampling rate by shifting the window by one sample at a time. This update is achieved in real-time on PDA platforms. The developed method is shown to be computationally more efficient as compared to computing FFT for every window. Basically, this thesis involves the implementation of an interactive cochlear implant system on PDA platforms using the recursive DFT approach as well as the filterbank approach. The software developed is portable and runs in real-time on any PDA platform. A real-time throughput is achieved by using the PDA module and the hybrid programming mode of the LabVIEW graphical programming environment together with appropriate code optimizations. The results obtained show that in addition to achieving a real-time throughput, the performance is not adversely deteriorated.