Large eddy simulation (LES) is a widely used method for modeling turbulent flows in which explicit computations of large flow scales, called resolved flow scales (RFS), are performed on a computational mesh. In contrast, dynamics at smalle scales, called sub-filter scales (SFS), are modeled. In the Approximate Deconvolution Method (ADM) proposed by Stolz and Adams [1], SFS are modeled through an explicit form of the sub-filter stress tensor evaluated with the solution in the RFS range. It has been observed that the accuracy of the smallest RFS is significantly influenced by induced filters associated with general discretization methods. ADM allows for an approximate reconstruction of the RFS, but it necessitates explicit knowledge of the spatial filter for the deconvolution process. The selection of this filter is critical and can markedly impact the accuracy of the reconstructed RFS. No universal methodology is currently available to specify the form of this filter. To address this issue, various approaches have been proposed in the literature. Here, we will assume a generic filter and investigate the total dissipation arising in turbulent Taylor-Green flow. This study compares various filter selections in combination with a high-order half-staggered spatial discretization method [2]. The final contribution of this research will introduce an ADM filter designed to improve comparison to direct numerical simulation (DNS) and adapt the total dissipation to improve stability while retaining accuracy. The outcomes of this investigation will have potential to enhance the precision of LES studies, thereby advancing our understanding of turbulent flows. [1] S. Stolz, and N.A. Adams, An approximate deconvolution procedure for large-eddy simulation. Phys. Fluids, Vol. 11(7), pp. 1699-1701, 1999. [2] A. Tyliszczak, A high-order compact difference algorithm for half-staggered grids for laminar and turbulent incompressible flows. J. Comput. Phys, Vol. 276, pp. 438-467, 2014.