A novel approach to robust channel estimation and its performance with uncoded and chaotic coded speech

High QoS in communication demands channel estimation in real time. The normalized least mean squares (NLMS) algorithm is employed due to its simplicity. Several modifications have been considered to improve the adaptation performance based on the spectral properties of the driving signal including orthogonalisation procedures like the recursive LMS, the fast affine projection or frequency domain methods. These methods are extremely sensitive to observation noise. A new finite correlation LMS (FCLS) algorithm is proposed which yields increased noise robustness in noise contaminated environments. Under the set of conditions considered an adaptation improvement of about 23 dB is obtained by replacing NLMS by FCLS. Further improvements can be obtained if chaotic coding is used