Abstract:
Sparse systems are those systems, the impulse response of which contains a signi_cant number of zero or near-zero coe_cients. Traditional least mean square (LMS) algorithm based system identi_cation schemes are not e_ective when applied for modeling such systems. A few sparse adaptive algorithms have been recently developed to overcome this limitation of conventional LMS algorithms. The underlying principle of most of these sparse adaptive algorithms is the concept of zero attraction, where by the near zero coe_cients of the model are forced to zero. In order to achieve an improved modeling accuracy in sparse system identi_cation scenarios, a new Polynomial zero attracting LMS (PZA-LMS) algorithm has been developed in this thesis. Attempts were also made to improve existing adaptive algorithms by improving their robustness when employed in modeling acoustic paths. In addition, a hybrid sprase adaptive algorithm and a least angle regression (LARS) algorithm based sparse adaptive algorithm were also designed in this work. Further, in an endeavour to reduce the critial parameter dependency of the adaptive algorithms on the modeling accuracy, the sparse system identi_cation and sparse signal reconstruction tasks have been formulated as a multi-objective optimization problem.