1: \begin{abstract}

2: In this work, we investigate dynamic oversampling techniques for

3: large-scale multiple-antenna systems equipped with low-cost and

4: low-power 1-bit analog-to-digital converters at the base stations.

5: To compensate for the performance loss caused by the coarse

6: quantization, oversampling is applied at the receiver. Unlike

7: existing works that use uniform oversampling, which samples the

8: signal at a constant rate, a novel dynamic oversampling scheme is

9: proposed. The basic idea is to perform time-varying nonuniform

10: oversampling, which selects samples with nonuniform patterns that

11: vary over time. We consider two system design criteria: a design

12: that maximizes the achievable sum rate and another design that

13: minimizes the mean square error of detected symbols. Dynamic

14: oversampling is carried out using a dimension reduction matrix

15: $\mathbf{\Delta}$, which can be computed by the generalized

16: eigenvalue decomposition or by novel submatrix-level feature

17: selection algorithms. Moreover, the proposed scheme is analyzed in

18: terms of convergence, computational complexity and power consumption

19: at the receiver. Simulations show that systems with the proposed

20: dynamic oversampling outperform those with uniform oversampling in

21: terms of computational cost, achievable sum rate and symbol error

22: rate performance.

23: \end{abstract}

24: