High-density surface electromyography (HD-sEMG) has become a valuable signal source in human-machine interface research due to its ability to non-invasively support motor unit (MU) decomposition. Although various filters have been proposed to enhance HD-sEMG signal quality, their effectiveness in improving MU decomposition performance remains unclear. In this study, we systematically evaluate five state-of-the-art filtering methods: Infinite Impulse Response (IIR), Wavelet Decomposition (WD), Variational Mode Decomposition (VMD), Ensemble Empirical Mode Decomposition (EEMD), Independent Vector Analysis (IVA), and a Hybrid filter. Their effects are assessed using two widely used MU decomposition algorithms, fastICA and LIBD-CKC. Both simulation and experimental results show that most filters, including WD, EEMD, VMD, and IVA, do not improve decomposition accuracy and can even degrade performance. The Hybrid filter provides meaningful improvement with statistically and practically relevant gains, although it incurs significant computational cost. Therefore, unless exceptionally high decomposition precision is required, advanced filtering methods are not recommended. In practical applications such as fingertip press estimation in this paper, HD-sEMG signal can be used directly without additional filtering.