Single-Channel Blind Signal Separation of the MHD Linear Vibration Sensor Based on Singular Spectrum Analysis and Fast Independent Component Analysis

Mengjie Xu, Jianhan Wang, Jiahui Mo, Xingfei Li, Lei Yang, Feng Ji2022 · DOI

TL;DR

This paper introduces a novel method for separating useful linear vibration signals from noisy measurements obtained by Magnetohydrodynamic (MHD) sensors. By combining Singular Spectrum Analysis (SSA) for noise reduction and Fast Independent Component Analysis (FastICA) for signal separation, the authors achieve a significant improvement in signal-to-noise ratio (SNR), making it easier to detect subtle vibrations in critical applications like spacecraft.

What Problem Does This Paper Address?

Spacecraft engineering demands highly accurate measurements of mechanical environments, including vibrations. MHD linear vibration sensors offer advantages like no mechanical wear and good dynamic performance, making them suitable for precise measurements. However, in practical scenarios, their output signals are often heavily contaminated by various noise sources, particularly narrowband interference. This noise significantly degrades the signal-to-noise ratio (SNR), making it difficult to extract the weak, useful vibration signals, especially for micro-vibration detection. Existing methods like EEMD and VMD have limitations such as modal aliasing or long computation times, making them less ideal for real-time applications. This paper aims to address this challenge by developing an effective single-channel blind signal separation technique for MHD vibration sensors.

Methods

The core of this paper is a two-stage signal processing approach:

  1. Singular Spectrum Analysis (SSA) for Denoising: SSA is employed first to suppress narrowband interference and white noise. The process involves:
  2. Decomposition: Embedding the time series into a trajectory matrix.
  3. Singular Value Decomposition (SVD): Decomposing the trajectory matrix to identify principal components.
  4. Reconstruction: Grouping components based on their importance (contribution rate of eigenvalues) to reconstruct a denoised signal, effectively separating useful signal components from noise.
  5. Fast Independent Component Analysis (FastICA) for Separation: After denoising with SSA, FastICA is applied to the cleaner signal. FastICA is a fast optimization algorithm that aims to maximize the non-Gaussianity of the output signals. It works by:
  6. Centering and Whitening: Preprocessing the signal to have zero mean and unit variance.
  7. Iterative Optimization: Using a fixed-point and Newton iteration to find a separation matrix that maximizes the non-Gaussianity of the estimated source signals. This effectively separates the independent components, which are assumed to be the useful vibration signal and residual noise.

The authors highlight that this combined SSA-FastICA approach is particularly suited for single-channel signals where direct ICA might be challenging.

Datasets

The paper utilizes experimental data collected from a self-developed MHD linear vibration sensor prototype. To validate the results, a quartz flexible accelerometer was used as a comparison sensor. Four groups of experimental data were collected under different voltage and frequency conditions (0.6 V, 10 Hz; 0.6 V, 60 Hz; 0.8 V, 5 Hz; and 0.8 V, 60 Hz). These datasets were used for both denoising evaluation and signal separation verification. No new benchmark datasets were introduced.

Key Insights and Ideas

  • Synergistic Denoising and Separation: The paper demonstrates that a two-stage approach, where SSA first denoises and then FastICA separates, is more effective than using either method alone or other common denoising techniques for this specific application.
  • SSA's Effectiveness for Narrowband Noise: SSA proves to be a powerful tool for suppressing the prominent narrowband interference characteristic of MHD sensor signals, significantly improving the SNR before separation.
  • FastICA's Suitability for Weak Signals: After SSA pre-processing, FastICA can effectively extract the underlying independent vibration signal even when it's weak and mixed with residual noise.
  • High Similarity to Reference Sensor: The separated MHD vibration signals show a high similarity (over 98%) to the signals from a standard quartz flexible accelerometer, validating the accuracy of the proposed method.

Key Results

  • SNR Improvement: The SSA algorithm alone achieved a higher SNR compared to EEMD, VMD, and wavelet threshold denoising. The SNR of the signal after SSA processing was 1.6 to 32 times higher than the other methods.
  • Effective Separation: The combined SSA-FastICA method effectively separated the useful linear vibration signals from the noisy collected signals.
  • High Similarity: The similarity factor between the separated useful signals and the corresponding collected signals (after SSA processing) was consistently high, exceeding 98% in the best cases.
  • Repeatability: Analysis of standard error (SE) for both the quartz accelerometer and the processed MHD sensor signals indicated good repeatability for the MHD sensor, especially after SSA-FastICA processing.

Limitations and Future Work

  • Constant Speed Measurement Limitation: The self-made MHD linear vibration sensor has a limitation in that it is not suitable for constant speed measurements.
  • Need for Noise Reduction: The paper emphasizes that for signals with significant noise, direct separation by ICA is insufficient, and pre-denoising is crucial.
  • Future Work: While not explicitly detailed, the success of this method suggests potential for application to other types of sensors with similar noise challenges and for real-time implementation in dynamic measurement systems. The authors' focus on improving dynamic measurement error implies a broader goal of enhancing the overall accuracy of such systems.

Summary updated

BibTeX
@misc{xu2022single,
  author = {Mengjie Xu and Jianhan Wang and Jiahui Mo and Xingfei Li and Lei Yang and Feng Ji},
  title = {{Single-Channel Blind Signal Separation of the MHD Linear Vibration Sensor Based on Singular Spectrum Analysis and Fast Independent Component Analysis}},
  year = {2022},
  doi = {10.3390/s22249657},
}

Loading related content...

Figure