Signal reconstruction of fast moving sound sources using compressive beamforming

Abstract

Source signal is one of the main input parameters when auralizing moving sound sources in the Virtual Reality (VR) environments. This work utilizes compressive beamforming (CB) as a tool to reconstruct signals from fast moving sources. A pseudorandom microphone array is designed to meet the requirement of using CB and delay and sum beamforming (DSB), thus allowing for the signal reconstruction from the CB output and for the comparison between these two beamforming algorithms. Parameter studies through error analysis are conducted to evaluate how the reconstructed source signal is influenced by parameters, i.e. regularization parameter, window length, signal-to-noise ratio (SNR), basis mismatch and distance between the array and source trajectory. In general, CB outperforms DSB in signal reconstruction in terms of varying every parameter, except for the similar performance with SNR = 30 dB. We used the designed microphone array with both CB and DSB to reconstruct the signal of a known engine noise emitted by a loudspeaker installed on a moving car. The localization results delivered by CB are similar to DSB, which is in line with the simulation results. This behavior can result from potential coherence in the sensing matrix of CB due to similar time-domain transfer functions (TDTFs). However, CB still delivers lower reconstruction errors. Both simulation and measurement results indicate that CB is a viable option to reconstruct the signals of fast moving sound sources.