In this paper, we are interested in adaptive blind channel identification of sparse single input multiple output (SIMO) systems. A generalized Laplacian distribution is considered to enhance the sparsity of the channel coefficients with a maximum a posteriori (MAP) approach. The resulting cost function is composed of the classical deterministic maximum likelihood (ML) term and an additive $\ell_p$ norm of the channel coefficient vector which represents the sparsity penalization. The proposed adaptive optimization algorithm is based on a simple gradient step. Simulations show that our method outperforms the existing adaptive versions of cross-relation (CR) method.