Sparse channel estimation problem is one of challenge technical issues in broadband wireless communications. Square error criterion based adaptive sparse channel estimation (SEC-ASCE) algorithms, e.g., zero-attracting least mean square (ZA-LMS) and reweighted ZA LMS (RZA-LMS), have been proposed to mitigate noises as well as to exploit the channel sparsity. However, the conventional SEC-ASCE algorithms are vulnerable to performance deteriorate due to 1) random scaling of input training signal, and 2) unable to balance between convergence speed and transient-state mean square error (MSE) performance. In this paper, a mixed square/fourth error criterion (SFEC) based ASCE algorithms (SEFC-ASCE), i.e., zero-attracting least mean square/fourth error (ZA-LMS/F) and reweighted ZA-LMS/F (RZA-LMS/F), are proposed to enhance estimation performance while without exhausting a lot computational complexity. First, regularization parameters of the ZA-LMS/F and RZA-LMS/F algorithms are selected by means of Monte-Carlo simulations. Second, lower bounds of the proposed channel estimation algorithms are derived and analyzed. Finally, simulation results are given to show that the proposed sparse LMS/F-type algorithms achieve better estimation performance than the conventional algorithms.