Title:Line Spectral Estimation Based on Compressed Sensing with Deterministic Sub-Nyquist Sampling

Abstract:As an alternative to the traditional sampling theory, compressed sensing allows acquiring much smaller amount of data, still estimating the spectra of frequency-sparse signals accurately. However, compressed sensing usually requires random sampling in data acquisition, which is difficult to implement in hardware. In this paper, we propose a deterministic and simple sampling scheme, that is, sampling at three sub-Nyquist rates which have coprime undersampled ratios. This sampling method turns out to be valid through numerical experiments. A complex-valued multitask algorithm based on variational Bayesian inference is proposed to estimate the spectra of frequency-sparse signals after sampling. Simulations show that this method is feasible and robust at quite low sampling rates.