Spatial Active Noise Control via Virtual Acoustic Sensing

Engineered an advanced spatial acoustic signal processing framework enabling Active Noise Control (ANC) in unmonitored spatial zones without requiring intrusive physical microphones at human ear locations.

Technical Innovations & Experiments

• Virtual Sensing Framework Formulation:
  • Addressed the fundamental spatial constraint of ANC where physical sensors cannot be placed inside the human ear canal or head volume during active operation.
  • Reconstructed the acoustic pressure field at target virtual microphone positions using pressure measurements streamed from a sparse array of 4 remote physical microphones.
• ObsTasNet Deep Observation-Filter:
  • Adapted and trained an ObsTasNet (Time-domain Audio Separation Network) observation-filter on 400 rigorously simulated 3D acoustic room environments exhibiting varying reverberation times ((T_{60})) and reflection coefficients.
  • Learned the spatio-temporal Green’s function transfer mapping between physical and virtual monitoring positions directly from raw pressure time-series.
• Inference Speedup & Benchmarking:
  • Benchmarked against classical analytical kernel interpolation and boundary element methods (BEM).
  • Achieved a 12.6× acceleration in virtual microphone pressure estimation latency, meeting the hard real-time latency thresholds necessary for stable adaptive feedback anti-noise generation.