The frame of ATSC 3.0 consists of bootstrap, preamble and subframe in order, and the bootstrap signal can be used for initial synchronizations at the receiver. Also, it contains the signaling information needed for decoding of the preamble. Therefore, the bootstrap signals have to be reliably decoded even under a poor condition. In this paper, we derive the maximum-likelihood (ML) decision rule for the detection of the bootstrap signals and propose an improved decoding algorithm of bootstrap signals for ATSC 3.0 systems. The proposed algorithm improves the reliability of the channel estimate by averaging the channel estimates for previous symbols. The simulation results show that the proposed algorithm can improve symbol error rate (SER) performance by enhancing the reliability of the channel estimates.