Experimental Results on Observation Filter Estimation for Microphone Virtualization

Active Noise Control (ANC) systems exploit the superposition principle to generate a quiet acoustic zone around a confined area affected by a disturbance signal detected by a monitoring microphone. In automotive applications, the problem of noise within the car cabin has become a hot research topic. Nowadays, most of the commercial solutions generate a silence zone around the position of the monitoring microphone only. Especially for the automotive environment, to weaken disturbance signal at a different position with respect to the monitoring microphone is fundamental. The estimation of the acoustic channel between the monitoring microphone and the physical region in which the noise cancellation is targeted, referred to as observation filter, is necessary. This approach is usually known as microphone virtualization, or Virtual Microphone Technique (VMT) since the audio signal detected by the virtual microphone (e.g., targeting driver’s ears) has to be retrieved starting from the monitoring ones. In this paper, a performance comparison between adaptive and fixed approaches to the estimation of the observation filter is presented. The disturbance signals are acquired in an experimental measurement campaign on a realistic car interior. Eight microphones are employed to acquire four different driving scenarios at various pace of the car. Experimental results show that, due the insufficient spectral coherence between the monitoring and virtual microphones, system performance is physically limited. A specific road and car pace scenario typically exhibits significant robustness to road mismatch. In the low frequency regime, both estimation approaches perform well. However, the fixed one guarantees improved broadband performance.