Using orthogonal least squares identification for adaptive nonlinear filtering of GSM signals

The miniaturization of GSM handsets creates nonlinear acoustical echoes between the microphone and the loudspeaker when the signal level is high. Nonlinear adaptive filtering can tackle this problem but the computational complexity has to be reduced by restricting the number of coefficients introduced by the nonlinear models. This paper compares the performance of different nonlinear models. In a first training stage we use the OLS (orthogonal least squares) identification method to find models using the fewest coefficients along with a good fitting accuracy. In a second filtering stage these parsimonious models are used to adaptively filter the GSM signals