A high-robustness radial intensity-orientated mode decomposition with reliable noise elimination

Abstract

Mode decomposition (MD) provides profound evidence to reveal the internal modal transmission mechanism. However, the indelible noise has always been the main stubborn hindrance in practical MD. In the complex superposition case with a large number of modes, the traditional MD is not capable enough to distinguish the real modal intensity and the annoying noise, sustaining an unacceptable accuracy and fluctuation. This paper proposes a radial intensity-orientated MD (RIO-MD) method with reliable noise elimination. Our approach focuses on the inherent modal radial features in Polar coordinates, getting rid of the traditional two-dimensional image processing in Cartesian ones. The RIO-MD introduces the inherent radial intensity relationship into MD for better extracting mode coefficients. Based on the expectable real radial modal intensity, the RIO-MD enables to recognize and extraction of the three kinds of stubborn noise, including interference pattern noise, device noise, and random noise. The RIO-MD works well in mode decomposition case. The values of correlation coefficients (C) between the experimental and reconstructed image are higher than 93%. The mean square error (MSE) is lower than 3 × 10−3. Both the C and MSE keep stable, with the standard deviation 30 times lower than the other widely used methods, demonstrating the high-robustness of the RIO-MD. Due to the reliable noise recognition, the RIO-MD shows great possibility in mode number expansion.