| Feedback cancellation is an important component of hearing aid signal processing and is almost a necessity when dealing with patients with moderate to severe hearing losses. Most contemporary implementations of feedback cancellation schemes involve the use of adaptive filters which are used to continuously model the feedback path and then subtract it from the input. An ideal model would mean we can perfectly cancel the feedback signal at the input and provide stability over any amount of gain. Most contemporary schemes also model the feedback path based on time-domain analysis suffer from the inability to properly handle pure tones and large changes to the feedback path. These abilities are important for handling many real-life scenarios such as listening to music or talking on a cell phone. The drawbacks of time-domain schemes also make the examination of transformation-domain strategies, where we would operate on the signal in frequency bands and thus avoiding time correlations, compelling. This project examines the advantages of a frequency-domain power normalization-based feedback-cancellation scheme. The thesis presents the performance of such a scheme in the presence of various signals including white noise, speech shaped noise, speech, music and pure tones. This work also presents a comparative analysis of the performance of time and frequency-domain cancellers to these inputs under varying acoustic and ambient noise conditions. The frequency-domain algorithm was found to be better than the equivalent time-domain system for real-life inputs like speech and music. An additional advantage of entrainment avoidance was also noted. Overall, the frequency-domain method seems to offer a more stable and reliable method for handling tonal inputs and large variations in feedback model. |
