Decentralized microphone pair selection for steered response power based localization in wireless acoustic sensor networks

In distributed microphone arrays, using all microphones for sound source localization (SSL) significantly increases computational complexity. In wireless acoustic sensor networks (WASNs), this challenge is further compounded by the energy and bandwidth required to exchange microphone signals or intermediate features, which can affect network lifetime. In this work, we propose three methods for microphone pair selection in steered response power (SRP) based SSL using massive arrays of spatially distributed microphones. The first method selects microphone pairs that exhibit a strong cross-correlation peak at zero lag. Such pairs typically correspond to geometries where the source lies approximately in the broadside direction of the pair, although the method itself does not rely on explicit geometric information. The second and third methods rely on the sparsity of the cross-correlation function, where we evaluate two different sparsity measures and select pairs with high sparsity values. To make the approach fully distributed, we approximate cross-correlations locally at each node and apply a gossip algorithm for decentralized top- K pair selection. The evaluation of our algorithm using simulations based on both simulated and real measured room impulse responses (RIRs) demonstrates that we achieve a localization performance using only the selected microphone pairs that is comparable to the performance when all microphones are employed.