Leaky adaptive filtering approaches for music-driven sound zone generation

Sound zones aim to create distinct listening areas within the same physical space. Such systems have potential applications in environments where listeners may move or where the acoustic environment may change. However, most existing sound zone methods are designed for static conditions, which limits their performance in dynamic scenarios—particularly when reproducing music signals that exhibit strong temporal correlation. To address this challenge, some methods including closed-loop algorithms have been proposed to continuously adapt filters in real time, using acoustic pressure at control points. Unfortunately, these methods suffer from slow convergence speeds due to correlated input signals. In this paper, three leaky gradient descent-based adaptive filtering approaches for generating sound zones in a potentially dynamic context are proposed to solve the sound zone problem. Their results are compared to the ones obtained with an existing adaptive filtering method: the filtered-x least mean squares. The leaky filtered-x affine projection algorithm (LFx-APA) enhances convergence speed and robustness by projecting over multiple past measurements, making it well-suited for non-stationary audio signals. Simulations performed in a reverberant room demonstrate that the LFx-APA achieves higher performance in terms of contrast without increasing reproduction error when the audio content is non-stationary.