Implicit large eddy simulations(ILES) of a low Reynolds number turbulent channel flow are performed using advection upstream vector splitting method (AUSM). Various interpolation methods of monotone upstream-centered conservation law (MUSCL) with/without limiters are tested in a coarse grid to diagnose the reliability of ILES. The results are compared with the empirical correlation and direct numerical simulation (DNS) statistics. Quantitative analysis of a priori test reveals that away from the wall, the effective sub-grid scale (SGS) dissipation of MUSCL schemes is greater than the SGS dissipation obtained from the filtered DNS data. Near the wall, the numerical dissipation behaves in a much more inappropriate manner. Two types of blending functions are proposed to reduce this inadequate behavior near the wall. Simulations with these two blending functions show that the skin friction coefficient is much better predicted.