Large-scale multiple input/multiple output system identification in room acoustics

Abstract

In audio reproduction scenarios, room acoustics may be described as a MIMO system response from multiple loudspeaker to multiple microphones in the listening space. This system response may, e.g., be used for an equalization of the listening room and must be identified from observing the available loudspeaker and microphone signals in real-world systems. For few transducers this task is mostly solved, but massive multichannel reproduction with dozens to hundreds of loudspeakers left many research questions open. This contribution points out the fundamental challenges, previous solutions and recent advances. As a key issue, the so-called nonuniqueness problem for MIMO system identification by adaptive filtering will be discussed along with decorrelation schemes for the loudspeaker signals to alleviate this problem. Successful adaptation algorithms suitable for these scenarios imply considerable computational demands and require additional measures to ensure robustness. Recently emerging system models in spatial transform domains allow for approximative models and seem to be promising for robust real-time implementations.