Wolfson Microelectronics (2008-12)

Title: Investigation and implementation of adaptive noise suppression techniques for portable applications
Research Engineer: Matthew Robbetts
Sponsor: Wolfson Microelectronics, Edinburgh
Academic Supervision: Stephan Weiss and Prof Bob Stewart, University of Strathclyde
Industrial Supervision: TBA

The proposed project addresses several problems in connection with communication via a mobile handset, including Tx and Rx noise cancellation, as well as potentially acoustic echo cancellation.

TX noise cancellation. This refers to noise cancellation applied to the near end speech signal as picked up by the mobile handset. The aim is to reduce the influence of noise (general background noise, cocktail party effect etc) from the transmitted signal. Currently Wolfson are investigating adaptive methods based on the acqusition of both speech and background noise by two microphones in order to achieve this. A major challenge is to extract a reference for either the noise or speech signal from the scenario and perform adaptive cancellation thereafter. Methods to be investigated include various adaptive techniques, including adaptive beamforming, as well as non-linear approaches such as double-talk detection. Parameteric methods such as denoising may also offer a benefit in noise suppression.
While in most cases no explicit noise reference may be available, some application components may have the benefit of such a reference signal, such as noise cancellation of a camera zoom drive on a mobile device. Instead of an acoustic reference, an electrical reference signal of the camera drive may suffice to converge the adaptive fnoise cancellation filter.

RX noise cancellation. The far end speech signal is corrupted by ambient noise at the near end speaker's ear. Therefore the aim is to use active noise cancellation in order to improve the intelligibility of the far end signal for the near end user. At present, Wolfson apply fixed feedforward filters from the handset's microphone array to the loudspeaker. These feedforward filters are averaged to provide good performance over a range of positions of the handset with respect to the ear.
We will investigate adaptive methods without explicit reference signal in order to perform active ambient noise cancellation. We may also consider robust spatial averaging methods which are optimised with respect to the human auditory system as applied for spatial sound effect in home cinema applications.