intro algorithm

Adaptive feedback cancellation - algorithm details

The FDIVAFC adaptive feedback cancellation scheme studied in the HearCom project is based on the observation that many audio signals can be closely approximated by a low-order autoregressive (AR) random process. This is the same principle that lies at the basis of the most powerful speech compression standards that are used nowadays, in GSM mobile telephony for instance. The FDIVAFC-algorithm in fact is an instrumental-variable (IV) based approach, which aims at prewhitening the hearing aid loudspeaker and microphone signal. In this way, a bias-free estimate of the feedback path can be computed using standard adaptive filtering techniques. This in general gives a better performance than with classical adaptive filtering approaches. Once the estimate of the feedback path is available, the feedback signal can be electrically reconstructed in the hearing aid device and be subtracted from the recorded microphone signal, leading to a significant reduction of the loop gain and, hence, a suppression of the feedback signal and increased stability.

Compared to standard adaptive feedback suppression algorithms the FDIVAFC-approach achieves more Added Stable Gain (ASG), i.e. it can better prevent feedback, as shown in the figure below for two behind-the-ear hearing aids with different gain settings (G) :

FDIVAFCresults.gif

Apparently, depending on the processing delay that is set in the hearing aid, the FDIVAFC-algorithm offers 2.5 to 11 dB extra amplification before feedback occurs compared to the standard approach.

References:

Spriet, A., Moonen, M., Proudler, I. and Wouters, J. (2005). Adaptive feedback cancellation in hearing aids with linear prediction of the desired signal. IEEE Transactions on Signal Processing, 53 (2005), pages 3749-3763 

Spriet, A., Proudler, I., Moonen, M. and Wouters, J. (2005). An instrumental variable method for adaptive feedback cancellation in hearing aids. In IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 129–132, Philadelphia, Pennsylvania.