Anti-noise and denoising performance of global rainbow processing

Abstract

This study simulated and experimentally evaluates the effect of signal noise of 5–60 dB on the inversion accuracy of the global rainbow signals of droplets with three typical size distributions (normal, log-normal, and bimodal normal), using a local minimum-based algorithm for parameter retrieval and empirical mode decomposition for denoising. For the simulated noise-free global rainbow signals of droplets with normal size distributions, the inversion algorithm has the highest accuracy, with the maximum relative error of droplet mean diameter D and the maximum absolute error of refractive index n being only 0.34% and 1.3 × 10−5. The empirical mode decomposition provides the best denoising and stability when preprocessing global rainbow signals of bimodal-normal size distributed droplets, reducing the errors of D and n from 51.4%, 9 × 10−4 to 7.9%, 2.4 × 10−4 even at a signal-to-noise ratio of 5 dB. Experiments were conducted based on a typical global rainbow measurement system. The errors of D and the n retrieved before and after denoising the noisy experimental signal are less than 5.2%, 1.4 × 10−4 and 2.4%, 0.51×10−4 respectively. Experimental results verify the feasibility and effectiveness of the noisy global rainbow processing procedures consisting of an original inversion algorithm and EMD denoising.