English

Quantization for spectral super-resolution

Information Theory 2022-03-02 v2 math.IT

Abstract

We show that the method of distributed noise-shaping beta-quantization offers superior performance for the problem of spectral super-resolution with quantization whenever there is redundancy in the number of measurements. More precisely, we define the oversampling ratio λ\lambda as the largest integer such that M/λ14/Δ\lfloor M/\lambda\rfloor - 1\geq 4/\Delta, where MM denotes the number of Fourier measurements and Δ\Delta is the minimum separation distance associated with the atomic measure to be resolved. We prove that for any number K2K\geq 2 of quantization levels available for the real and imaginary parts of the measurements, our quantization method combined with either TV-min/BLASSO or ESPRIT guarantees reconstruction accuracy of order O(M1/4λ5/4Kλ/2)O(M^{1/4}\lambda^{5/4} K^{- \lambda/2}) and O(M3/2λ1/2Kλ)O(M^{3/2} \lambda^{1/2} K^{- \lambda}) respectively, where the implicit constants are independent of MM, KK and λ\lambda. In contrast, naive rounding or memoryless scalar quantization for the same alphabet offers a guarantee of order O(M1K1)O(M^{-1}K^{-1}) only, regardless of the reconstruction algorithm.

Keywords

Cite

@article{arxiv.2103.00079,
  title  = {Quantization for spectral super-resolution},
  author = {C. Sinan Güntürk and Weilin Li},
  journal= {arXiv preprint arXiv:2103.00079},
  year   = {2022}
}

Comments

29 pages, 2 figures, to appear in Constructive Approximation