Active noise control is common in headphones, but augmented reality glasses face a different problem: there is nowhere to put a microphone at the user's ear. The authors first confirm that virtual sensing—estimating the sound at the ear from elsewhere on the frame—is required for this geometry. They then analyse the Auxiliary Filter method, which predicts the ear signal using a nearby reference microphone.
The paper's theoretical analysis shows that a fixed auxiliary filter is highly sensitive to the direction of the incoming noise. Simulations match this prediction: when noise arrives from a different angle than the calibration direction, cancellation performance drops sharply. This rules out a single fixed filter for real-world use, where noise sources move.
To fix this, the authors propose storing a small set of auxiliary filters, each tuned to a different angle, and automatically selecting the correct one based on an estimate of the noise direction of arrival. In simulation, this filter bank approach substantially recovers performance compared to the fixed filter, while avoiding the complexity of continuous adaptive recalibration. The result is a practical step toward making AR glasses usable in noisy, dynamic environments.