Adaptation of Acoustic Echo Cancellers Incorporating a Memoryless Nonlinearity

With acoustic echo cancellation for hands-free communications, dominant nonlinearities occur at the loudspeaker and its amplifier. This contribution addresses memoryless nonlinearities as they are common, e.g., in loudspeaker amplifiers of mobile phone terminals. An NLMS (normalized least mean square)-type adaptation algorithm is investigated which allows simultaneous identification of a nonlinearity modeled by a polynomial, followed by a linear FIR system. For the analysis of the proposed algorithm, we introduce a signal-independent distance measure which provides more insight into the adaptation behavior than the commonly used ERLE (echo return loss enhancement) measure. A difficult issue with cascaded adaptive filters is the adaptation stepsize control, which is discussed in detail. A special stepsize control mechanism is derived and experiments with speech signals show that it is suitable for practical use.