定Qフィルタバンク出力の位相頻度に基づく標的パラメータの推定

抄録

A computational model of echolocation was investigated for unambiguous measurement of the range and velocity of a moving object. Based on the time-frequency structures of echolocation sounds utilized by some species of 'FM bats', a linear-period modulated (LPM) signal was used in the model. Motivated by the study reported by P.A.Saillant et al., we introduce in this paper a new mechanism for motion parameter estimation by adopting a voting mechanism according to output phase values across a bank of constant-Q filters. Resultant histograms of the phase values along the the temporal axis provide a two-dimensional map in which a maximum peak of the histograms gives a clue to perceive the existence of an object and estimate the corresponding motion parameters. Experimental results of motion parameter estimation performed in air are presented in comparison with a conventional correlation detection of chirp signals. Regardless of the relatively small time-bandwidth product of the LPM sound used in the experiments, the estimation accuracy was better than 0.05[m/s] for velocity and about 0.3[mm] for range. In contrast, reliable results were not obtained because of the ambiguous property in parameter estimation if the time-frequency structure of chirp sounds was inappropriately designed.