Efficient Segmentation for Large-Scale Filtering Based on Overlap-Save Method

In modern communication systems, especially in fiber optics, implementing large-scale filters is inevitable. Finite impulse response implementation of these filters leads to intractable complexity, but advanced methods such as Overlap-Save (OLS) and Overlap-Add exist that significantly reduce complexity by implementing filters in the frequency domain. However, these methods require the Fast Fourier Transform (FFT) size to be larger than the filter size, limiting their applicability due to hardware constraints on the FFT size. In this paper, to remove this limitation, we propose a novel approach called low-complexity OLS-based segmented filter (LOSF). LOSF not only allows the FFT size to be smaller than the filter but also efficiently handles digital processing tasks, minimizing the number of FFT blocks required. We demonstrate the efficacy of LOSF through theoretical analysis and simulation studies, showcasing its ability to implement arbitrarily large filters within practical hardware constraints. Simulation results indicate that LOSF outperforms existing methods, achieving significant complexity reduction while maintaining filter performance.