Key points are not available for this paper at this time.
A cochlea-based preprocessor for speech recognition emulates the fluid ducts with two 4680-element diffusive grids, the basilar membrane with 360 2 nd -order sections, and the auditory nerve with 2160 pulse-frequency modulators. Integrated in 10.9mm 2 in 0.25μm CMOS and consuming 52mW, this silicon cochlea employs active bidirectional coupling, a selective amplification mechanism that sharpens tuning (Q 10 is 2.7) and controls gain (24dB compression)
Wen et al. (Sun,) studied this question.