sparsity and conpressive sensing

I need help with the correct matlab implementation of Iterative Hard Thresholding.

ITH proceeds as follows: Input: ent matrix A, measurement vector y, sparsity level 8. Initialization: s-sparse vector ro, typically ro 0. Iteration: repeat until a stopping criterion is met: n+1 Output: the s sparse vector zn Let us put ITH to the test. Consider y Ar, where A is a 100 x 400 Gaussian random matrix and r is an s-sparse vector of length 400. The locations of the entries of r are chosen uniformly at random and the non-zero coefficients of zo are normal-distributed. For s 1, 2 find the sparsest solution to the system Az IT For each fixed s y using repeat the experiment 10 times. Create a graph plotting s versus the relative reconstruction error llr zll2 (averaged over the ten experiments for each s). Starting with which value of s (approximately) does ITH fail to recover z? How well does this the theoretical prediction regarding required number of measurements via L1-minimization? (In the theoretical prediction we had in class you can assume that the constant is close to 1)