DSP-based Tightly-coupled GPS/MEMS-IMU Integration Using Sequential Processing

The fast development of sensor technology has promoted the development of consumer navigation devices that are embedded with the GPS/INS integration. While most of these applications adopt the loosely-coupled integration to achieve better navigation performance, the performance improvement, however, is still limited especially in the GPS challenged environment. When integrating a more advanced fusion algorithm, for example tightly-coupled integration, designers often encounter many difficulties in system implementation due to the sophisticated integration structure. This paper explores the design of a real-time tightly-coupled integrated system based on a DSP platform, utilizing one commercial GPS receiver and low-cost MEMS inertial sensors, which appeals to the increasing demand of more accurate and robust navigation systems for low-cost commercial applications. The tightly-coupled architecture in the proposed system employs a common extended Kalman Filter with the popular INS error model. To differ from the common way of using the system process model based on the perturbation analysis, we utilize one that is based on the first-order linearization. This brings the advantages that, without making small misalignment angle assumptions, the accurate IMU self-alignment and calibration process are not necessarily required. Furthermore, the sequential processing technique is adopted in the measurement update of the Kalman Filter, which improves efficiency of the onboard realization of GPS/INS integration. Two main implementation challenges caused by the high communication expense between the sensor and the processor and the heavy processing load of the integration task are solved in the system design. On one hand, various hardware resources are utilized to synchronize the data from GPS and IMU, as well as to guarantee an efficient system response to both subsystems. On the other hand, the implementation of both the time update (prediction) and the measurement update of the filter are well treated to fulfill the system's real-time capability. Matrix partitioning method is adopted to reduce the multiplication burden of bulky matrices based on the unique feather of the system transition matrix, while the sequential measurement update we use avoids the huge matrix inversion. Field experiment with a regional train was conducted with a selected trajectory which involves a GPS challenged environment to validate the system feasibility. With experimental data, the real-time performance of the system is analyzed and discussed. Moreover, further considerations are addressed as well.