Bat Echolocation Scan Pattern Reconstruction using Convolutional Sparse Coding

Compressive sensing enables the detection and allocation of sparse signals at a sub-Nyquist sampling rate. For this reason, it is particularly interesting in the case of high sample rate applications. Monitoring bat echolocation signals using an ultrasonic microphone array is a high sample rate application. To evaluate the methods proposed in this work, Nyquist-compliant sample data is undersampled to simulate compressive sensing (CS) and reconstructed using convolutional sparse coding with a dictionary set trained on a bat’s echolocation calls. This paper evaluates the robustness of the proposed method for extracting key acoustic properties from bat echolocation signals. It compares these properties to those extracted with a Nyquist-compliant dataset that serves as a ground truth reference.