Hearing Aid Direction-of-Arrival Estimation Using Precomputed RTFs

The patent pre-calculates and stores relative transfer functions (RTFs) for multiple directions and frequencies in a database before actual DOA estimation. During operation, the system only needs to retrieve pre-computed RTFs and perform simple correlation operations, rather than calculating transfer functions in real-time. This preliminary preparation significantly reduces online computational complexity while maintaining high estimation accuracy through the maximum likelihood framework.

The patent divides the frequency spectrum into multiple frequency bins and processes each bin separately using short-time Fourier transform. The DOA estimation is performed independently for each frequency bin, allowing parallel processing and reducing the computational burden on any single processing stage. This segmentation enables efficient handling of wideband signals while maintaining precision across different frequencies.

The patent assumes sound sources are located in the frontal half-plane and optimizes the RTF database to cover only this region with higher resolution, while using coarser resolution or fewer frequency bins for other directions. This local quality optimization concentrates memory resources on the most important listening region (frontal hemisphere) where users typically focus attention, achieving high accuracy where needed while reducing overall storage requirements.

The system dynamically adjusts the resolution and density of RTF database entries based on the specific acoustic situation and frequency bin being processed. For frequencies and directions with similar acoustic characteristics, the system uses shared or interpolated RTF values, reducing redundant storage. The patent also employs parameter quantization to represent RTFs with sufficient but not excessive precision, balancing accuracy requirements with memory constraints.