Performance Analysis of a Deep Simple Recurrent Unit Recurrent Neural Network (SRU-RNN) in MEMS Gyroscope De-Noising

Area of Science:

• Robotics and Control Systems
• Signal Processing
• Machine Learning Applications

Background:

• Microelectromechanical System (MEMS) Inertial Measurement Units (IMUs) are crucial for navigation systems due to their size and power efficiency.
• Manufacturing limitations in MEMS IMUs lead to complex noise and errors, hindering navigation accuracy.
• Effective noise modeling and suppression are vital for enhancing MEMS IMU-based navigation system performance.

Purpose of the Study:

• To introduce and evaluate a deep learning approach for de-noising MEMS gyroscope signals.
• To investigate the impact of training data length on the prediction performance of the Simple Recurrent Unit (SRU).
• To analyze and compensate for key noise parameters influencing MEMS gyroscope accuracy using Allan Variance.

Main Methods:

• Utilized a Simple Recurrent Unit Recurrent Neural Network (SRU-RNN) for de-noising raw MEMS gyroscope signals.
• Experimentally evaluated the SRU-RNN method using a MEMS IMU (MSI3200) from MT Microsystem.
• Compared SRU performance with varying training data lengths and applied Allan Variance to quantify noise parameters (quantization noise, angle random walk, bias instability).

Main Results:

• Determined optimal training data lengths (500, 3000, 3000 for three axes) for reliable SRU prediction performance.
• Achieved significant improvements in gyroscope accuracy across axes: X-axis (0.6%–12.5%), Y-axis (60.5%–34.1%), and Z-axis (11.3%–35.7%) for key noise parameters.
• Demonstrated substantial reductions in attitude errors, with improvements of 19.2%, 82.1%, and 69.4% for the respective axes.

Conclusions:

• The SRU-RNN deep learning method effectively de-noises MEMS gyroscope signals, enhancing navigation system accuracy.
• Specific training data lengths are sufficient for stable and reliable prediction performance.
• Compensation of quantization noise, angle random walk, and bias instability significantly improves MEMS gyroscope performance and reduces attitude errors.