Compared with existing systems, fifth generation (5G) mobile systems are expected to support diverse services and massive connectivity. Thus, multiple access schemes play a critical role. Recently, a codebook-based non-orthogonal access mode called sparse code multiple access (SCMA) has been proved to enable massive connectivity by allowing SCMA layer overloading. Message Passing Algorithm (MPA) can iteratively detect the multiplexed SCMA codewords. However, reducing MPA computation complexity is still a problem since the 5G ultra low latency and energy consumption targets. As the conditional channel probability (CCP) calculation takes up 60% computation of logarithm-domain MPA, reducing the CCP calculation is the tipping point of MPA optimization. In this paper, we proposed a dynamic shrunk square searching (DSSS) algorithm, based on the signal uncertainty theory, to dynamically shrink the state space before the CCP calculation. As the DSSS exploits both the structure of state space and noise characteristic, it could alleviate unnecessary CCP calculation while maintaining the near optimal decoding performance in block error (BLER) sense.