Acoustic Shooter Localization With Consistency-Function Fusion in Sparse Sensor Networks

Acoustic localization of supersonic rifle fire in urban environments in military and security operations faces challenges from multipath propagation and non-line-of-sight detections. Physics-based methods predominate because the geometric ground-truth data required for machine learning is hard to collect at scale. Building on consistency-function fusion – an approach that identifies shooter hypotheses by counting supporting sensor observations rather than minimizing measurement residuals – this paper compares two sparse-network architectures: a wearable multi-channel-array system and a single-microphone deployment exploiting geometric shockwave constraints. We introduce two algorithmic refinements: a caliber-aware deceleration correction that removes a systematic position bias growing quadratically with shooter range, and a coplanar-geometry recovery for the multi-channel pipeline that handles trajectories wide of the sensor field. Under noiseless evaluation on the Aberdeen 2006 geometry the refinements drive position error from ∼17 m to ∼1 m for the multi-channel pipeline and from ∼19 m to ∼2 m for the single-channel pipeline. A high-fidelity simulator with a noise budget representing the Aberdeen hardware reproduces the published position MAE values to within a few percent for both architectures. Under mobile-deployment noise the multi-channel pipeline retains both better position accuracy (23 vs 26 m allrange MAE) and a 17 percentage-point localization-rate advantage that widens to nearly 29 points with moderate acoustic multipath. Architecture choice thus emerges as an environment-dependent trade-off – single-channel for cost-constrained line-of-sight deployments, multi-channel for cluttered environments – rather than a universal hierarchy.