FDE Implementations for a Low-Cost GPS/INS Module

This paper describes the implementation details of fault detection and exclusion (FDE) methods for possible use in a low-cost GPS/INS module. The use of a coupled system has been previously considered for improving navigation in GPS difficult environments. Such environments adversely affect the satellite signals received by a navigation system. Integration with an INS allows for higher output rate as well as improved coasting through GPS signal blockage. Previous work has shown that the inclusion of a FDE algorithm allows for the removal of faulty GPS measurements that tend to corrupt the navigation solution. This work gives analysis of several FDE methods both from the standpoint of performance and efficiency. The resulting goal of this work is the details of a real-time GPS/INS module with FDE improvements. The module design requirements are specified for use in low-cost applications. Since operation in vehicular environments is desired, meter-level accuracy is investigated so that approximate lane-level information could be available to the user of the navigation system. The FDE methods under consideration both have snapshot and sequential implementations. The snapshot methods are performed independently between measurement epochs and thus do not suffer from undetected errors that corrupt the states. Sequential methods are able to detect a wider variety of errors but are delayed in acting on failure conditions. These two implementations are compared for the normalized innovation method and the direct consistency check method. The normalized innovation technique is a comparison of the resulting new information provided by a measurement to a normalized threshold. The threshold is set to detect measurements that do not statistically conform to the expected accuracy. The direct consistency check algorithm performs a comparison of a measurement to what the measurement is expected to be given the removal of the measurement from the estimation. Faults are then detected and removed when inconsistencies are found. These methods are considered for use in the GPS difficult environments. The comparison of these methods is accomplished by the design and assembly of the GPS/INS module. This module is then used to log the required data for initial post-processing. Various data collection locations considered include open-sky, urban canyon, and heavy foliage areas. The integration and FDE algorithms are then run on the same data sets and fault detection occurrences compared among the methods. Processing time is also monitored for the post-processing to generate efficiency results. The details are then given to implement the chosen method in real-time on a low-cost GPS/INS module. Due to the quickly changing nature of the GPS errors in difficult environments, the snapshot methods tend to provide faster detection of errors and thus improved performance. For faster implementation, the normalized innovation technique is selected to reduce load on the embedded navigation system. This choice allows for more flexibility in extending the module use. The result of this work is a navigation system implementable in real-time that provides improved positioning in GPS difficult environments. Many applications such as vehicle navigation and control benefit from improved performance in these situations. The inclusion of the low-cost requirement allows for more ubiquitous use of these results.