Blind Source Unmixing Algorithm of MEMS Magnetic Signal of Multisource Sensor in PDR Positioning

• Published in: IEEE sensors journal Vol. 23; no. 18; pp. 20916 - 20927
• Main Authors: Wang, Jinhua, Meng, Qinggang, Li, Mingduo, Wang, Sainan, Hou, Jinliang
• Format: Journal Article
• Language: English
• Published: New York The Institute of Electrical and Electronics Engineers, Inc. (IEEE) 15.09.2023
• Subjects: Algorithms; Dead reckoning; Independent component analysis; Magnetic signals; Microelectromechanical systems; Noise; Noise reduction; Separation
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2023.3291698

Pedestrian dead reckoning (PDR) positioning with micro-electromechanical system (MEMS) is currently intelligent autonomous positioning, which is a front research field in indoor positioning. However, it is the problem of low positioning accuracy, as the signals measured using MEMS positioning devices contain various interference noises. A blind source unmixing algorithm was used to denoise the observed magnetic signal in order to solve the problem of noise reduction in the measured signal. The noise disturbance law was analyzed through experimental measured signals, and a new idea of blind source unmixing algorithm for denoising observed magnetic signals was proposed in this article. The dynamic and static magnetic values were collected using a self-made HBEQ-1 MEMS device and FVM-400 magnetometer in the banding area; fast independent component analysis (FastICA), nonnegative matrix factorization (NMF), and reconstruction independent component analysis (RICA) algorithm were applied in denoising experiment of the analog signal. The experimental results show that the separation index of the three algorithms is all less than 0.1 in the simulation test, the NMF algorithm with the best separation performance achieved 0.001; meanwhile, there are some differences at the peak point between the unmixing signal and the source signal. Then, the experimental results with the NMF algorithm for dynamic magnetic value show that the wave similarity between the unmixing signal and the known static signal is 0.99 without prior knowledge; it is down to 0.98 after adding prior knowledge of known walking noise and device vibration noise. PDR positioning experiments were conducted for denoising the effectiveness of the NMF algorithm, in which its heading angle was calculated using the magnetic signals before and after unmixing. Its results show that the trajectory direction after denoising is close to the actual trajectory, which can be improved the accuracy of PDR positioning.