LARGE_SCALE_SPARSE_SUBSPACE_ID is the main framework of alternating minimization. It initializes the input driving signal to a sparse signal and alternates between system identification and estimation of an input signal. Inputs y : multidimensional time series of measured outputs block_size: the size of the hankel matrices n : order of the system the hankel matrices should have 2*n/(number of observations) number of blocks of rows MAXITER_L1 : maximum number of iterations of the reweighted L1 minimization MAXITER_ALL : maximum number of iterations of the alternating minimization lambda : sparsity penalty s_const : bound of the l2 norm of the LDS impulse response plot_is_on : (true or false) show figures Outputs A, B, C : matrices of deterministic state space LDS u : sparse estimated input signal x_0 : Initial Condition cv_f : Values of Cost Function ------------------------------------------------------ Problem Description : Blind Deconvolution assuming a LDS of the form x_(k+1) = A x_k + B u_k y_k = C x_k (1) and that the input u_k is sparse and bounded, C*A*B has also bounded energy The goal of this function is to : minimize || u||o subject to (1), with respect u, A, B,C,D