21 cm信号的观测提供了一种测量大尺度结构中的重子声波振荡和探测暗能量的新途径。然而，由于前景辐射比 21 cm信号约强四个数量级，致使观测该信号存在极大的困难。普遍使用的前景扣除算法因依赖频谱的光滑性假设存在一定局限性。另外，对仪器造成的系统误差的分析研究还处于起步阶段。在本项目中，我们将：（1）扩展我们最近提出的一种贝叶斯框架下不依赖于先验假设的源分离半盲算法（HIEMICA），使其适用于大天区面积巡天中的曲面天场并在强前景干扰下稳定可靠地重建 21 cm 信号；（2） 针对天籁和 CHIME 实验中蝶形和柱形天线干涉阵列，建立全过程仿真模拟观测以检验各种信号提取算法；（3） 探索和发展吉布斯采样和哈密顿蒙特卡洛技术，通过拟合更复杂的前景模型精确扣除前景污染；（4）基于解析计算和时间流仿真数据分析，量化估算仪器系统误差的影响。以上研究结果将直接应用于天籁和 CHIME 的观测。

Observations of the 21 cm signal provide a new way to measure the baryon acoustic oscillations in the large scale structure and to probe the dark energy. However, since the foreground emission is about four orders of magnitude stronger than the signal, the measurement of such the weak signal thus is significantly difficult. Due to the assumption of the spectral smoothness, the commonly used techniques for foreground removal have some limitation. In addition, the study of the impact of instrumental systematic errors is still at the beginning stage. In this project, we aim to: (1) extend our recently proposed foreground removal technique (HIEMICA), which is the Bayesian semi-blind source separation approach but without any prior assumptions about foregrounds, such that it can be applicable to the case of the curved sky in large sky area survey and can reliably recover the 21 cm signal under the strong foreground contamination; (2) build an end-to-end simulation of interferometric observations for dish and cylinder arrays in the Tianlai and CHIME experiments to test various approaches for signal recovery; (3) explore and develop the Gibbs sampling and the Hamiltonian Monte Carlo techniques to accurately subtract the foreground contamination by fitting more complicated foreground models; (4) quantitatively assess the impact of the instrumental systematic errors. All of the above results will be directly applied to the real data analysis in the Tianlai and CHIME observations.