Speaker
Description
In multi-motor battery electric vehicles the loss of combustion masking makes the order-structured tonal noise of each drive unit perceptually salient, and units that share the same machine and single-speed gear ratio, such as the two wheel motors of one axle or the twin units of a symmetric dual-motor vehicle, hold their same-numbered orders within about one percent of each other whenever both are active. This spectral degeneracy prevents tracking-filter separation for any small number of channels, including the binaural pair used in standard NVH practice. We propose a spatial decomposition framework for third-order Ambisonic recordings. Phase-coherent tracked-harmonic extraction, driven by one instantaneous phase per shaft, is applied independently to every spherical-harmonic channel, and full-sphere direction-of-arrival maps attribute each separated component to its source. Where the order families are degenerate, a constrained adaptive spatial prefilter, whose spherical-harmonic steering vectors are independent of frequency, places nulls on the interfering motor and on the background before extraction. The framework is validated on a real anechoic-chamber recording and on two-motor scenes constructed from measured electric-drive order data and measured room impulse responses, and is benchmarked against binaural processing of the same fields. Where per-channel tracking, binaural processing and even exact-phase joint Vold-Kalman estimation remain limited, with weaker-motor correlations of 0.60–0.84, spatial-null preprocessing restores the separation to 0.94, and the recovered components retain their directions of arrival.