In the field of single image super resolution, the accuracy of high-frequency information and the clarity of the reconstructed images are closely linked. Current mainstream models rely on dense connections or module stacking to extract high-frequency information, optimizing image reconstruction through mechanisms like spatial domain attention. While effective, these methods often lead to redundancy in high-frequency information and a subsequent degradation in accuracy. To address this issue, this study proposes an image super resolution network that modulates high-frequency information. The network integrates self modulation and mutual modulation strategies to enhance the image reconstruction process. Specifically, the self modulation module combines a multi-scale sampling network with a high-frequency guided spatial attention mechanism, which effectively filters out redundant information while simultaneously calibrating data in both spatial and frequency domains. The mutual modulation module further refines the accuracy of high-frequency information by establishing a nonlinear relationship between the self modulated outputs. Through extensive experimental validation, it is demonstrated that the proposed model has superior performance in terms of quantization metrics and maintains parameter efficiency compared to current popular models. It also delivers outstanding visual results, fully showcasing its advantages in single image super resolution reconstruction.