Leaky adaptive filtering approaches for music-driven sound zone generation

Area of Science:

• Acoustics and Signal Processing
• Adaptive Filtering
• Spatial Audio

Background:

• Sound zones create distinct listening areas but struggle in dynamic environments.
• Existing methods fail with music due to temporal correlation and slow convergence.
• Closed-loop algorithms adapt in real-time but are often too slow.

Purpose of the Study:

• To develop adaptive filtering methods for robust sound zone generation in dynamic acoustic conditions.
• To improve convergence speed and performance for non-stationary audio signals.
• To address limitations of existing sound zone technologies in real-world applications.

Main Methods:

• Proposed three leaky gradient descent-based adaptive filtering approaches.
• Introduced the leaky filtered-x affine projection algorithm (LFx-APA).
• Compared LFx-APA against the filtered-x least mean squares (FxLMS) algorithm.

Main Results:

• LFx-APA demonstrated enhanced convergence speed and robustness.
• The algorithm performed well with non-stationary audio signals.
• Simulations in a reverberant room showed improved contrast without increased reproduction error.

Conclusions:

• Leaky gradient descent-based methods, particularly LFx-APA, offer superior sound zone generation in dynamic scenarios.
• LFx-APA is well-suited for non-stationary audio due to its ability to project over multiple past measurements.
• The proposed methods significantly advance the performance of sound zone systems in challenging acoustic environments.