Recent advancements in point cloud upsampling have highlighted challenges in preserving geometric details and uniform distribution, particularly for high-ratio tasks. To address the limitations of existing methods in capturing global topological structures, this paper propose a novel point cloud upsampling method that decouples the process into topology-based point interpolation and point correction, fully leveraging the original topological information to maintain structural consistency and avoiding the information distortion caused by feature-space interpolation in previous approaches.Simultaneously, we design collaborative feature and topological encoders to effectively couple point features with topological structures. Additionally, a symmetric interpolator is introduced to associate interpolated features with point coordinates. Finally, a correction module performs point-wise adjustments based on the corresponding features.The framework constructs a cascaded self-supervised point cloud upsampling pipeline that progressively inserts topological points and incrementally refines their positions, achieving high-quality upsampled results with fine-grained structures.Experimental results demonstrate that our method not only surpasses all existing unsupervised methods but also matches the performance of state-of-the-art supervised approaches.