1: \begin{abstract}

2: Terahertz ultra-massive MIMO (THz UM-MIMO) is envisioned as one of

3: the key enablers of 6G wireless networks, for which channel estimation

4: is highly challenging. Traditional analytical estimation methods are

5: no longer effective, as the enlarged array aperture and the small

6: wavelength result in a mixture of far-field and near-field paths,

7: constituting a hybrid-field channel. Deep learning (DL)-based methods,

8: despite the competitive performance, generally lack theoretical guarantees

9: and scale poorly with the size of the array. In this paper, we propose

10: a general DL framework for THz UM-MIMO channel estimation, which leverages

11: existing iterative channel estimators and is with provable guarantees.

12: Each iteration is implemented by a fixed point network (FPN), consisting

13: of a closed-form linear estimator and a DL-based non-linear estimator.

14: The proposed method perfectly matches the THz UM-MIMO channel estimation

15: due to several unique advantages. First, the complexity is low and

16: adaptive. It enjoys provable linear convergence with a low per-iteration

17: co\textcolor{black}{st and monotonically increasing accuracy, which

18: enables an adaptive accuracy-complexity tradeoff. Second, it is robust

19: to practical distribution shifts and can directly generalize to a

20: variety of heavily out-of-distribution scenarios with almost no performance

21: loss, which is suitable for the complicated THz channel conditions.

22: For practical usage, the proposed framework is further extended to

23: wideband THz UM-MIMO systems with beam squint effect. Theoretical

24: analysis and extensive simulation results are provided to illustrate

25: the advantages over the state-of-the-art methods in estimation accuracy,

26: convergence rate, complexity, and robustness. }

27: \end{abstract}

28: