Empirical Mode Decomposition Technique With Conditional Mutual Information for Denoising Operational Sensor Data

Abstract

This paper presents a new approach for denoising sensor signals using the Empirical Mode Decomposition (EMD) technique and the Information-theoretic method. The EMD technique is applied to decompose a noisy sensor signal into the so-called intrinsic mode functions (IMFs). These functions are of the same length and in the same time domain as the original signal. Therefore, the EMD technique preserves varying frequency in time. Assuming the given signal is corrupted by high-frequency (HF) Gaussian noise implies that most of the noise should be captured by the first few modes. Therefore, our proposition is to separate the modes into HF and low-frequency (LF) groups. We applied an information-theoretic method, namely, mutual information to determine the cutoff for separating the modes. A denoising procedure is applied only to the HF group using a shrinkage approach. Upon denoising, this group is combined with the original LF group to obtain the overall denoised signal. We illustrate our approach with simulated and real-world cargo radiation data sets. The results are compared to two popular denoising techniques in the literature, namely discrete Fourier transform (DFT) and discrete wavelet transform (DWT). We found that our approach performs better than DWT and DFT in most cases, and comparatively to DWT in some cases in terms of: 1) mean square error; 2) recomputed signal-to-noise ratio; and 3) visual quality of the denoised signals.