Research Topic

Adaptive Feedback Cancellation in Hearing Aids


Whenever a sound signal is recorded by a microphone, processed (e.g. amplified) and then immediately played back by a loudspeaker nearby the microphone, it is unavoidable that the loudspeaker signal is fed back to the microphone. The described phenomenon is called acoustic feedback and it appears in case of, e.g., public address systems or hearing aids.

The acoustic feedback creates a closed signal loop, which can cause problems. Thus, under certain conditions, which concerns the magnitude and the phase
of the feedback path and the amplification of the system, the so called howling effect occurs. This effect is highly annoying for the hearing aid user and his environment. Also the acoustic feedback limits the possible amplification by the hearing aid.

An effective feedback cancellation algorithm can improve the user comfort and the performance of an hearing aid significantly. This can help to increase the acceptance of hearing aids, which is an important factor. Statistics show that although 14 to 16 million Germans suffer from hearing loss, only about 3.5
million wear a hearing aid.

Recent Results

Figure 1: Adaptive feedback cancellation scheme

The idea of the adaptive feedback cancellation (AFC) approach is to estimate the feedback path by an adaptive filter and then to subtract the estimated feedback from the recorded signal before it is amplified and played back. Figure 1 illustrates this idea. Due to the adaptive filter, the method is able to adapt fast changes of the feedback path, which is often the case in hearing aid scenarios. To calculate the adaptive filter coefficients the normalized least mean squared (NLMS) algorithm is used which minimizes the power of the error between the microphone signal and the estimated feedback signal, the error is denoted as e(n).

It can be shown, that ^f(n) equals f(n) if the loudspeaker signal u(n) and the original sound signal x(n) are uncorrelated. This is the case in an echo cancellation scenario. However, in our case u(n) is an amplified and slightly delayed version of x(n). This leads to a bias, which depends on the correlation in the adaptation.

An idea to overcome this problem is to split the signals into multiple frequency bands by using a filterbank and adapt in each band separately. A filterbank contains several band-pass filters which split the input signal in several frequency bands. Since we want to adapt in each band separately, we apply the NLMS algorithm in each band. The advantage is that we can adapt only in these bands where the feedback is strong. This leads to less adaptation where the correlation occurs. Therefore, the bias is decreased, the sound signal is less distorted and the possible gain is increased. The remaining task is to automatically control the adaptation in each band, so that the feedback is reduced as much as possible, the howling is prevented and the sound quality is preserved for all different scenarios and sound signals.

Key Research Area

R9) Communication Systems


Falco Strasser


Dolivostr. 15

D-64293 Darmstadt



+49 6151 16 - 24372


+49 6151 16 - 24404




strasser (at) gsc.tu...

 Print |  Impressum |  Sitemap |  Search |  Contact |  Privacy Policy
zum Seitenanfangzum Seitenanfang