This paper presents a novel approach that overcomes the limitations of existing noise-removing diffusion probabilistic model (DDPM)-based methods for compensating missing values in probabilistic time series data. Existing DDPM-based methods suffer from high time-series modeling time complexity and ineffective handling of dependencies in time series data. To address these issues, this paper utilizes a state-space model (SSM), such as Mamba, as the backbone of the DDPM's noise-removing module and designs several SSM-based blocks for time series data modeling. Experimental results demonstrate that the proposed methodology achieves state-of-the-art time series missing value compensation performance on a variety of real-world datasets. The code and datasets are publicly available.
