Bat echolocation signals based on the time-varying autoregressive method

Bat echolocation is among the most efficient biological sonar known to human, and is highly valuable for biomimetic research. Most bats produce dynamically changing echolocation signals, which is the key to high task performance. Although considerable progress has been made in bat sonar bionics research, the working mechanism of the bat sonar system has not yet been fully revealed, mainly reflecting the imperfect parameterized model of the bat vocal system. This paper describe the bat echolocation signal production as a time-varying autoregressive (TV-AR) model, and the trajectory of model parameter changes is modeled as segmental constant and continuous change. Based on the two forms of parameter changes, this paper use the regularized least squares method and the basis function method for parameter solving, respectively. The TV-AR based system model realizes the simulation of bat vocal system with Gaussian white noise as input and bat echolocation signal as output. Using echolocation signals recorded from the Pratt's roundleaf bats performing an approach-and-land task in the laboratory, we show that naturalistic echolocation signals can be simulated from the proposed TV-AR with high quality. Preliminary simulation and analysis suggests that the model can also be extended to simulate echolocation signals of distinct bat species.