The performance of a blind source separation system based on a custom microphone array is explored. The system prioritizes artifact-free processing over source separation effectiveness and extracts source signals using a quadratically constrained least-squares fit based on estimated source arrival directions. The level of additive noise present in extracted source signals is computed empirically for various numbers of microphones used and different degrees of uncertainty in knowledge of microphone locations. The results are presented in comparison to analytical predictions. The source signal estimate variance is roughly inversely proportional to the number of sensors and roughly proportional to both the additive noise variance and microphone position error variance. Beyond a threshold the advantages of increased channel count and precise knowledge of the sensor locations are outweighed by other limitations.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.