This paper presents a bandwidth expansion method for dual-mass micro-electro-mechanical system (MEMS) gyroscopes based on the pole-zero temperature compensation method. When the sense loop operates under open conditions, the mechanical sensitivity of the gyroscope structure conflicts with the bandwidth and is governed by the frequency difference between the drive and sense modes (min { $\Delta\omega1, \Delta\omega2$ }), which is proven to change with temperature during the experiment. The pole-zero temperature compensation proportional controller (PZTCPC) is proposed to expand the bandwidth under different temperatures based on the pole-zero compensation method. The force rebalancing combs stimulation method (FRCSM) is used to achieve accurate gyroscope bandwidth characteristics. The mechanical bandwidth of the gyroscope is proven to be approximately 13 Hz when the sense mode loop is open, and the simulation results show that the PZTCPC method expands the bandwidth to greater than 91.7 Hz after the sense mode loop is closed. The FRCSM experiments indicate that gyroscope bandwidth is expanded to 95 Hz at -40°C 94 Hz at 20°C and 92 Hz at 60°C, while the bandwidths at -40°C, 20°C and 60°C are all 93 Hz with the turntable method. The experimental curves match the simulation curves well and verify the simulation results. The new limiting condition of the closed-loop bandwidth is the trough generated by conjugate zeros, formed by superposition of in-phase and anti-phase sense modes.

The immense evolution in Large Language Models (LLMs) has underscored the importance of massive, heterogeneous, and high-quality data. A data recipe is a mixture of data from different sources for training LLMs, which plays a vital role in LLMs' performance. Existing open-source tools for LLM data processing are mostly tailored for specific data recipes. To continuously uncover the potential of LLMs, incorporate data from new sources, and improve LLMs' performance, we build a new system named Data-Juicer, with which we can efficiently generate diverse data recipes, explore different possibilities in forming data mixtures, and evaluate their effects on model performance. Different from traditional data-analytics pipelines, Data-Juicer faces some unique challenges. Firstly, the possible data sources for forming data recipes are truly heterogeneous and massive with various qualities. Secondly, it is extremely expensive to precisely evaluate data recipes' impact on LLMs' performance. Thirdly, the end users of Data-Juicer, model developers, need sufficient flexibility to configure and evaluate different data recipes. Data-Juicer features a fine-grained abstraction of pipelines for constructing data recipes, with over 50 built-in operators for easy composition and extension. By incorporating visualization and auto-evaluation capabilities, Data-Juicer enables a timely feedback loop for both LLM pre-training and fine-tuning. Further, Data-Juicer is optimized and integrated with ecosystems for LLM training, evaluation, and distributed computing. The data recipes derived with Data-Juicer gain notable improvements on state-of-the-art LLMs, by up to 7.45% increase in averaged score across 16 benchmarks and 17.5% higher win rate in pair-wise GPT-4 evaluations. Our system, recipes, and tutorials are released, calling for broader data-centric research on training and understanding LLMs.

This study presents a new microelectromechanical system, a vibration ring gyroscope with a double U-beam (DUVRG), which was designed using a combination of mathematical analysis and the finite element method. First, a ring vibration resonator with eight double U-beam structures was developed, and 24 capacitive electrodes were designed for drive and sense according to the advantageous characteristics of a thin-shell vibrating gyroscope. Then, based on the elastic mechanics and thin-shell theory, a mathematical stiffness model of the double U-beam was established. The maximum mode resonant frequency error calculated by the DUVRG stiffness model, finite element analysis (FEA) and experiments was 0.04%. DUVRG structures were manufactured by an efficient fabrication process using silicon-on-glass (SOG) and deep reactive ion etching (DRIE), and the FEA value and theoretical calculation had differences of 5.33% and 5.36% with the measured resonant frequency value, respectively. Finally, the static and dynamic performance of the fabricated DUVRG was tested, and the bias instability and angular random walk were less than 8.86 (°)/h and 0.776 (°)/√h, respectively.

Inductively-coupled wireless power transfer (IWPT) is promising to power a freely moving capsule robot (CR) continuously. Existing IWPT system for this application normally adopts a single-layer Helmholtz coil (SHC), which can generate a uniform alternating magnetic field (AMF) to cover the CR workspace, but leads to a low power transfer efficiency (PTE). To improve the PTE, this paper proposes a novel double-layer Helmholtz coil (DHC). The DHC is composed of two in-parallel connected SHCs that have identical coil turns and winding directions but different diameters. To design and optimize the DHC, a dedicated analytical model, is built and verified. A comparison between the DHC and SHC is also conducted, which shows that the DHC can not only generate a uniform AMF to cover a larger region, but its power consumption is lower when the generated AMF is identical, thus improving the PTE. Finally, the DHC is used to power a three-dimensional receiving coil, and the achieved PTE is in a range of 7.58%∼17.41%, higher than that when using the SHC, which is in a range of 6.74%∼14.92%.

Wireless power transmission (WPT) for a capsule robot (CR) has always been a challenging issue because the unpredictable CR attitude can destabilize the induced power of the receiving coil onboard the CR. This article proposes a new WPT system for stably powering the CR, and it mainly consists of a rotatable two-dimensional transmitting coil (R2DTC), a one-dimensional receiving coil (1DRC), and an attitude detection module (ADM). The R2DTC can excite a uniform and directional alternating magnetic field (AMF) covering the CR workspace, and the AMF amplitude and direction are controllable in the 3-D space. The 1DRC has a shape of thin-walled ring, for ease of being used as a CR outer shell. The ADM is fixed together with the 1DRC; it can detect the 1DRC attitude in real time and is implemented in a compact size. Referring to the 1DRC attitude, the R2DTC can regulate the AMF to stabilize the 1DRC induced power. This article presents design and implementation details of the WPT system, as well as experiments for its performance evaluation. The R2DTC is verified to supply stable power of 500 mW to the attitude-changing 1DRC at most cases, with a transfer efficiency ranging from 8.15% to 13.53%.

This paper presents a sense mode closed-loop controlling system for micro-electro-mechanical system (MEMS) gyroscopes, and the controller is designed to achieve low error, wide bandwidth and low noise capability. The gyroscope monitoring system includes four independent closed-loop, and is simulated by Simulink soft to prove the speedability and stability of the sensing closed-loop. The system is realized through 3 analog PCBs and is tested on the turntable in temperature oven. The bias stability, angular random walking value and bias temperature coefficients improved from 2.168 °/h, 0.155°/√h and 10.59 °/h/℃ to 0.415 °/h, 0.0414°/√h and 3.59°/h/℃. And the bandwidth value is improved from 13Hz to 104Hz. Meanwhile, scale factor nonlinearity, asymmetry, repeatability and temperature coefficient parameters are enhanced from 660 ppm, 430ppm, 403ppm and 180 ppm/℃ to 59.3ppm, 62.4ppm, 50.4ppm and 28.7 ppm/℃ respectively.