CENTRALIZED CASCADE MULTI-CHANNEL NOISE REDUCTION AND ACOUSTIC FEEDBACK CANCELLATION IN A WIRELESS ACOUSTIC SENSOR AND ACTUATOR NETWORK

This paper presents a centralized cascade multi-channel noise reduction (NR) and acoustic feedback cancellation (AFC) algorithm for speech applications in a wireless acoustic sensor and actuator net- work (WASAN). The algorithm consists of a multi-channel Wiener filter (MWF) based NR stage, where M microphone and L loud- speaker signals in the network are used to estimate, for each node, the speech component in its reference microphone and loudspeaker signal. For each node then a prediction error method based AFC stage is applied using these estimates to estimate the desired signal. Closed-loop simulations show that the proposed centralized algorithm outperforms a node working in isolation, in terms of the signal-to-interference-plus-noise ratio improvement (∆SINR) and short-time objective intelligibility (STOI) metric. In particular, the performance is improved when for instance one of the nodes has a lower input signal-to-noise ratio (iSNR) than the other nodes. In addition, it is also shown that a single-channel local adaptive filter per node is sufficient in the AFC stage, instead of a K-channel adaptive filter. Based on this and on the availability of a distributed algorithm for the NR stage, the presented algorithm is viewed as a stepping stone towards a fully distributed NR and AFC algorithm.