A STUDY ON COCHLEAR IMPLANT SYSTEM BASED ON A PHYSIOLOGICAL MODEL OF THE PERIPHERAL AUDITORY SYSTEM

本研究的目的是对人工耳蜗外周听觉模型进行反向处理和多脉冲编码。我们提出了一种递归算法，仅从波的正部分近似输入信号。在Meddis神经传导模型的基础上，通过求解神经传递物质的反向问题，实现了耳蜗非线性饱和特性的反向转换。并重新研究了Slaney可逆AGC（自动增益控制）在大输入时的不稳定性可能产生振荡。虽然在Meddis的模型中，对小信号的精度也很高，但它对大输入变得不稳定。由于作为线性系统处理，尽管反向转换不足，但仍实现了声音系统从内耳到外耳的传递特性。提出了一种基于语音编码原理的高效多通道脉冲转换算法，实现了对人工耳蜗电极刺激脉冲的有效转换。通过语音编码方法，对输入的声音进行分析，得到近似滤波器，在此问题上搜索给定滤波器集的最优组合级数。得到使滤波器设置标记的正/负脉冲序列作为压缩语音信息，并将其与电极脉冲匹配。通过编码解码来检验编码的准确性，发现在分析过程中，能量小的高频分量相对于低频分量出现了脱落和短缺的现象。虽然结合高频增强和频谱平坦化处理后的效果是可以接受的，但在音质改善方面是不够的。虽然这个问题还没有解决，但这个系统作为一个高质量的人工耳蜗植入语音编码的可能性得到了验证。

The purpose of this study is inverse processing of the peripheral auditory model and multi-pulse coding fro the cochlear implant. We proposed a recursive algorithm that approximates the input signal only from the positive part of the wave. Reverse conversion of the nonlinear saturation characteristic in cochlea was realized by solving a reverse problem about a nerve transfer substance on the basis of Meddis's nerve tranducion model. It also reexamined that an oscillation might occur in instability of the reversing AGC (automatic gain control) by Slaney for large inputs. Although the precision was high to small signals also in the Meddis's model, it became unstable to large inputs. Since it processed as a linear system, the transfer characteristic of the sound system which results in an inner ear from an outside ear was realized despite reverse conversion being inadequate.The efficient multi-channel pulsation algorithm based on the principle of the voice coding for efficient conversion to the cochlear implant electrode stimulation pulses was developed. By the voice coding method, analyzing an input sound and having obtained the approzimation filter, on this problem it searches for the optimum combination series about a given filter set. The pulse series of the positive/negative which makes a filter set a mark is obtained as compression voice information, and matches this with the electrode pulse. When accuracy of coding was checked by coding decoding, it was found that the high frequency components with small energy falls out and runs short in analysis process relatively compared with that of low frequency. Although the effect accepted when high frequency enhancement and spectrum flattening processing were incorporated, it was inadequate in respect of the tone-quality improvement. Although the unsolved problem had remained, the possibility of this system was checked as a quality voice coding for the cochlear implantation.