The most promising method for acoustic feedback control in public address systems is adaptive feedback cancellation (AFC). This approach is based on making an estimation of the room impulse response, which is used to subtract an estimate of the feedback from the microphone signal. A postfilter is often used to improve the performance of such an AFC system by reducing residual feedback or increasing stability. The postfilter implemented in this thesis focuses on the latter. It consists of two modules: howling detection and howling suppression. The final design makes use of the peak to harmonic power ratio criterion (PHPR) and second order digital Butterworth notch filter respectively. Simulation measurements show a successful suppression of howling as long as the loudspeaker signal does not
get saturated.

This report is written as part of the Bachelor Graduation Project in the third year of the Electrical Engineering Bachelor programme at the Delft University of Technology. This report is made by a subgroup that is part of a larger project dedicated to finding a solution to frequency feedback in electroacoustic systems, also known as the Larsen effect.

This problem has a long history of literature and many solutions to this problem have been presented. For this project, the focus lies on the application of Adaptive Feedback Cancellation (AFC). This technique uses estimations of the Room Impulse Response (RIR) to minimise the feedback component. This report focuses on estimating such a RIR in the case of a PA system. Several methods will be discussed and compared by implementing and testing them in a simulation environment built in Matlab. The different methods are subjected to a set of performance measurements that provide the necessary information to see for each method and situation if and how the results measure up against a set of predetermined system requirements. From the researched algorithms, one is chosen as the one to be implemented in the final design of the group. At last, some considerations and recommendations are given for implementing one of the algorithms into software and hardware components.

For people with hearing impairment, it is important to have good speech intelligibility, while also being able to localise the sound sources. Many beam-forming algorithms for hearing aids have been proposed, that minimise the noise, in combination with spatial scene preservation of the target and the interferers. By constraining the spatial cues, the limited degrees of freedom available for the design of the filter are expended, and this, to some extent degrades the noise reduction performance. Most of these methods try to preserve both the interaural time difference (ITD) and the interaural level difference (ILD) cues of the noise components over the entire frequency spectrum. However not all frequencies rely on both the ITDs and the ILDs for the localisation of sound. More specifically, the ITDs are dominant cues in the frequencies below 1.5 kHz and the ILDs are dominant cues in the frequencies above 1.5 kHz. Based on these facts, in this thesis we try to preserve only the ILD cues of the noise components at frequencies above 1.5 kHz, while keeping the target signal undistorted. We investigate whether doing so saves the degrees of freedom that can be used to improve the noise reduction performance, in contrast to preserving both the cues. The thesis proposes two methods to preserve only the ILD cues of the interferers. The first method preserves the ILD cues perfectly, while the second method achieves a relaxed preservation of the ILD cues. Both methods show similar performance in anechoic and reverberant environments, and show that the noise reduction performance improves only mildly, when only the ILD cues of the noise components in the higher frequencies are preserved.