This thesis aims to develop an improved approach to enhance the gradient de-cent algorithm applied by L2 sensitivity minimization subject to sparse normal-form realization. In this thesis, the enhanced approach is focused on the convergence speed of searching the optimal filter parameters of using gradient decent algorithm, such that the convergence speed can be significantly faster than that of [1]. We adjust the step size in the algorithm to speed up the convergence. Finally, the numerical examples are performed to verify our proposed approach.
