Adaptive Feedback Cancellation in Hearing Aids

Acoustic feedback is a well-known phenomenon in hearing aids and public address systems. Under certain conditions it causes the so-called howling effect, which is highly annoying for the hearing aid user and limits the maximum amplification of the hearing aid. The most common choice to prevent howling is the adaptive feedback cancellation algorithm, which is able to completely eliminate the feedback signal. However, standard adaptive feedback cancellation algorithms suffer from a biased adaptation if the input signal is spectrally colored, as it is for speech and music signals. Due to this bias distortion artifacts (entrainment) are generated and consequently, the sound quality is significantly reduced. Most of the known methods to reduce the bias have focused on speech signals. However, those methods do not cope with music, since the tonality and correlation are much stronger for such signals. This leads to a higher bias and consequently, to stronger entrainment for music than for speech. Other methods, which deal with music signals, work only satisfactorily when using a very slow adaptation speed. This reduces the ability to react fast to feedback path changes. Hence, howling occurs for a longer time when the feedback path is changing. In this thesis, a new sub-band adaptive feedback cancellation system for hearing aid applications is proposed. It combines decorrelation methods with a new realization of a non-parametric variable step size. The adaptation is realized in sub-bands which decreases the computational complexity and increases the adaptation performance of the system simultaneously. The applied decorrelation methods, prediction error filter and frequency shift, are well known approaches to reduce the bias. However, the combination of both is first proposed in this thesis. To apply the proposed step size in the context of adaptive feedback cancellation, a method to estimate the signal power of the desired input signal, i.e., without feedback, also referred to as source signal power is necessary. This estimate is theoretically derived and it is demonstrated that it is a reliabe estimate if the decorrelation methods are additionally applied. In order to further improve the performance of the system three additional control methods are derived: The first one is an impulse detection to detect wideband impulses, which could lead to misadaptation. Secondly, a modified estimate of the source signal power to stabilize the system in case of jarring voices is proposed. Lastly, a correlation detection, which is applied to improve the trade-off between adaptation stability and tracking behavior, is developed. The complete system is optimized and evaluated for several speech and music signals as well as for different feedback scenarios in simulations with feedback paths measured under realistic situations. Additionally, the system is tested by real-time simulations with hearing aid dummies and a torso and head simulator. For both simulation setups hearing loss compensation methods as applied in realistic hearing aids are used. The performance is measured in terms of being able to prevent entrainment (adaptation stability) and reacting to feedback path changes (tracking behavior). The complete adaptive feedback cancellation system shows an excellent performance. Furthermore, the system relies only on few parameters, shows a low computational complexity, and therefore has a strong practical relevance