Investigations on Sparse System Identification with l0-LMS, Zero-Attracting LMS and Linearized Bregman Iterations
Sprache des Vortragstitels:
Englisch
Original Tagungtitel:
International Conference on Computer Aided Systems Theory (EUROCAST 2017)
Sprache des Tagungstitel:
Englisch
Original Kurzfassung:
In many system identification applications, the impulse response vector is of sparse nature. Dedicated sparse identification algorithms like the l0-LMS, Zero-Attracting LMS and Linearized Bregman Iterations outperform the traditional LMS algorithm. This imporved performance is achieved by extending the cost function with a L0 or L1 part of the estimated impulse response vector. In this presentation the three algorithms are compared regarding mean-square error of the estimated impulse response, convergence speed, and resulting sparsity.