Empirical Analysis of Waterbed-like Performance Trade-Off for a Feedforward Occlusion Effect Reduction System

While wearing earbuds, the wearers often perceive their own voice as boomy. This so-called occlusion effect (OE) poses a major problem. Various countermeasures based on active noise cancellation (ANC) have been proposed. In a previous publication, we investigated the use of a vibration sensor in a time-invariant feedforward ANC scheme to reduce the OE. However, a potential drawback is the leakage of sensor noise into the ear canal. In this publication, we employ an improved filter design method which allows us to constrain the noise leakage. We show empirically that constraining the noise leakage also results in limited OE reduction performance. This poses a trade-off similar to what is known for feedback systems as the waterbed effect. Furthermore, we show that the trade-off is rooted in the fact that we can only use causal filters in practice. Lastly, we give an example on how the improved filter design method, further extended by an outer microphone as reference sensor, can be of use in an application-oriented filter design. The analysis is done via simulations based on measurements with a hardware prototype.