MEMS IMU Based INS/GNSS Integration: Design Strategies and System Performance Evaluation

Application of MEMS sensor in navigation is increasingly becoming important due to its advantages in terms of the quickly improving precision, robustness, high dynamic response and lower costs of development and usage. Moreover by employing the optimal estimation technique of Kalman filtering, the performance of MEMS based INS has been greatly enhanced by the integration of GNSS. This paper focuses on the integration design of INS with GNSS based on MEMS IMU sensors. The sophisticated MEMS IMU error models are employed to evaluate the MEMS error impacts on INS performance. Through the use of the MEMS IMU error model, the MEMS IMU raw measurements can be simulated conveniently. Moreover it provides the fundamental model for the inertial sensor error compensation before INS calculation. The Kalman filter based integration configuration is also designed by combining GNSS solutions in this paper. Different experimental scenarios are designed to evaluate the performance of the proposed integration configuration by various simulation tests. The experimental results indicated that the performance of the MEMS based INS is greatly enhanced by integration of GNSS compared with its stand-alone usage and by employing the MEMS IMU error model to compensate the deterministic sensor errors. INTRODUCTION As an independent means of navigation, strapdown inertial navigation system (INS) providing position/velocity/attitude information via the measurements from inertial sensors has various advantages, such as totally autonomous, high dynamic response, good short-term accuracy and robust performance when exposed to interference/jamming, etc. In the last decade, as the rapid development of MicroElectro-Mechanical System (MEMS) drastically reducing the cost of previously expensive inertial sensors, the lowcost MEMS-based strapdown INS has been a subject of great interest. However its usage as a stand-alone navigation system is limited due to time-dependent growth of the MEMS sensor bias/noises. In comparison to MEMS-based INS, the Global Navigation Satellite System (GNSS), such as GPS, GLONASS and Galileo, is capable of delivering position and velocity information ascertained with time-independent precision, while the performance becomes unreliable when the system is exposed to high dynamics, interference from communication equipment and jamming. Because of the aforementioned complementary characteristics, GNSS and INS are commonly coupled with Kalman filter technique to augment the over-all performance by overcoming the shortcomings of each individual system. The purpose of this study is to determine the MEMSbased INS/GNSS integration design strategies by evaluating the impacts of those aforementioned issues on the integrated INS/GNSS system. Although the fundamental principle is the same, a practical design of INS/GNSS integration varies from the system requirements as well the selected IMU sensors. For example, MEMS presents the unique noise characteristics and generally containing high-level noises inherent in the output data, therefore the issues of the advanced techniques to simulate and also compensate MEMS raw measurement errors should be addressed. Firstly, to simulate the raw MEMS IMU measurements, the advanced error modelling technique is utilised by this study to estimate and evaluate the MEMS sensor noises. With the comprehensive error models, the noises of MEMS can be characterised in terms of random noise, bias, quantization error, scale factor correction and alignment error, etc. Secondly, by employing the sensor error model in INS calculation, the compensation of the deterministic noise terms from the raw MEMS measurements can be performed, such as the bias, scale factor errors, and the quality of the MEMS measurements for INS algorithm can be thus greatly improved. Finally, a loosely coupled Kalman filtering design for INS/GNSS integration is presented. Particularly, the design of Kalman filter process model which is suitable for the MEMS sensor noise characteristics is addressed in the paper. In additional, with the results from MEMS error modelling, the random noise compensation of MEMS sensors is achieved for the Kalman filter design. The structure employed in this study is to perform the theoretical analysis which provides the fundamental verification of the design strategies, followed by the experimental validations in simulation. With its well-convinced advantages in theoretical analysis and the simplicity tuning the parameters, the simulation in Matlab is used for this study in order to validate the design and potential performances. It consists of firstly the generation of simulated GNSS and MEMS raw measurements corresponding to the reference flight trajectories. Secondly, the INS algorithm utilising the raw IMU measurements derives the navigation attitude, velocity and position solutions. Finally, the integrated INS/GNSS Kalman filter is design to deliver the optimal navigational solutions through fusing both of MEMSbased INS and GNSS data. Experimental results indicate that mitigation of the raw MEMS IMU measurement noises by MEMS error modelling technique improves the accuracy of the INS solutions which in turn enhances the integrated INS/GNSS performance. Moreover this error model helps to achieve the more accurate statistical noise estimation of Kalman filter, which is well-known critical to the integrated Kalman filter’s performance. The results also show that due to the large IMU noises in the MEMSbased INS/GNSS integration design, the formerly complicated Kalman filter dynamic model can be greatly simplified by reducing the dimension of the error state vector without degrading the performance. With using the simplified models, the processing load of Kalman filter can be greatly reduced. MEMS IMU BASED INS/GNSS INTEGRATION STRUCTURE IN SIMULATION It is worth mentioning that the simulation of INS/GNSS integration algorithms is a mandatory step prior to realtime implementation in order to validate the design and assess the performance [4]. The following scenario is proposed as shown in Figure 1. Figure 1. The structure of the integrated INS/GNSS navigation based on MEMS IMU Flight profile Generator : Given the flight trajectories, this software package is used to generate the flight profile typically providing detailed information such as attitude, velocity and position. In this research, a realistic flight profile simulator, i.e. Microsoft Flight SimulatorTM or XPlane is employed. IMU Raw Measurements : Given the inputs of flight profile, this software package generates the corresponding error-free IMU measurements, i.e. the true outputs of triad gyro and accelerometer. By modelling the inertial sensor errors, and then combining them to those error-free gyro and accelerometer measurements, the raw IMU measurements therefore can be obtained. In this study, to characterise the MEMS IMU sensor error features, a more sophisticated IMU sensor error model is employed [1], which will be detailed in the following section. INS : Inertial Navigation System software package is an autonomous process of computing position location by doubly integrating the raw measurements of triad gyro and accelerometer of a point, whose position is to be determined [1]. First integration is to determine a threedimensional velocity vector and again to obtain a threedimensional position vector. INS is indispensable in the integration of INS/GNSS, providing the INS navigation solutions for the integrated filter, system errors can be estimated and corrected by the integrated Kalman filter. GNSS measurements : This software generates the GNSS velocity and positioning raw measurements and solutions, in terms of velocity/position and pseudorangerate/pseudorange corresponding to the flight profile. Optionally, it provides the GNSS shortage / blockage information. Integrated Kalman filter : Optimal estimation technology, i.e. so-called Kalman filtering, is employed through combining the navigational information from both INS and GNSS to deliver the optimal estimations of the navigational states. In the integrated INS/GNSS, the INS error equations are commonly utilised to construct the Kalman filter; hence Kalman filter’s states are the navigational solution errors. Through compensating these optimally estimated navigational errors, the optimal solutions can be derived thereafter. Another obvious advantage of the integrated Kalman filter is, during the blockage/shortage of the GNSS signal, it directly forwards the INS stand-alone solutions as Kalman filtering outputs at the typical INS output rate, e.g. 100Hz. Once the GNSS signal is available again, Kalman filter corrects the INS stand-alone solutions at the typical GNSS measurement rate, e.g. 1Hz. MEMS IMU ERROR MODEL DESIGN Comparing with traditional IMU sensors, e.g. nuclear magnetic gyro (NMR), electrostatic gyro (ESG), Ring Laser Gyro (RLG), the pendulum accelerometers .etc., which costs rise dramatically with the increasing precision, MEMS has greatly reduced the costs therefore a mass production has been permitted. Nevertheless, MEMS fabrication process is continuously improving and its performance is always better. Whereas due to the large time-growing noises inherently existing in the MEMS IMU sensor, its applications are limited significantly, usually constrained to low-precision and short-term applications. Hence careful studies of the characteristics of MEMS IMU sensor noises are potentially beneficial to enhance the performance. In this study, a set of more sophisticated IMU sensor models [1], which is expressed respectively as Gyro and accelerometer sensor model, is employed to investigate the MEMS IMU sensor performance. While manufacturing an inertial MEMS sensor, it is virtually impossible to control tolerances with actual production technology. However, manufactured component inaccuracies are generally stable and analytically predictable, which provides the basis for modelling the inaccuracies. The error effects of each sensor can be measured as part of manufacturing/testing operations, in turn the relative error effects will be provided in the specifications in terms of various errors/noise parameters, which are organised by a general inertial sensor model. For example, the gyro error model is described by: ) )( ( 1 lg Rand Quant Bias n A Scal W Puls F F I to ω δ ω δ ω δ ω ω + + + + Ω = (1) Where, ω Error free angular rate output Puls ω , Gyro output in pulses/second , Nominal pulse weight, i.e. scale factor to W Ω , Scale factor correction matrix Scal F , Alignment matrix n A F lg Bias ω δ , Gyro bias vector Quant ω δ , Quantization error Rand ω δ , Gyro random error Comparing with gyro error model, the similar form of the accelerometer error model can be characterised as: ) )( ( 1 lg Rand Quant niso Size Bias SF n A Scal W SF a a a a a a G G I A a to Puls δ δ δ δ δ + + + + + + =