Fault Tolerant Navigation using Decentralized INS/GNSS Tight Integration Filters
Inchara Lakshminarayan and Demoz Gebre-Egziabher, Department of Aerospace Engineering and Mechanics, University of Minnesota
Alternate Number 2
The most significant development in the field of navigation has been the advent of Global Navigation Satellite Systems (GNSS), and in an ideal world we would have continued, precise, accurate and reliable GNSS at all times. But we are still working towards this and currently, faults in GNSS are very common and hence it cannot be used for standalone applications. It is most frequently used with an Inertial Navigation System (INS) to generate position, velocity, attitude and time estimates. One common problem in INS/GNSS systems is a temporary/prolonged GNSS outage. The addition of more GNSS constellations by various countries (like USA, Russia, EU, China, India and Japan) to the already existing satellite network provides greater satellite coverage all over the world. This mitigates the issue of continued signal availability to some extent and improves our chances of tracking at least 4 satellites, which is the necessary minimum required for GNSS navigation, in a region at a given time. Yet, there is still the question of reliability of the position solution, detection and reconfiguration after occurrence of faults. Decentralized filtering techniques have proven to enhance development of reliable and fault tolerant algorithms for multisensory navigation  .
Decentralized filters consist of one or more parallel filters that generate a partial/complete navigation solution and a master filter to fuse them to generate an optimal solution. If the output from any one/more of the parallel filters are corrupted then that filter is discarded during the fusion stage. Since each parallel filter takes input from any of the one or more available sensors on board, failure of a parallel filter consequently points us to a failed sensor. In this way, decentralized filtering helps in identifying a sensor fault and graceful degradation of the fault with subsequent reconfiguration of the navigation system. In this paper, we use this decentralized filtering technique to detect satellite unavailability and faults in the GNSS system to maintain a reliable and consistent navigation solution at all times. For this purpose, we employ the INS/GNSS tight integration architecture where the INS and GNSS are reduced to basic sensor functions. It is known to be more accurate and robust than it’s loosely coupled counterpart and can also be used to aid the INS solution when insufficient satellite signals are tracked . The recent development of sensor modules that can provide raw GNSS measurements (pseudorange and pseudorates), enable us to extract the unprocessed signals transmitted from satellites. This reduces processing errors at the sensor level leading to enhanced accuracy. Further, by splitting up one centralized Kalman filter with tight integration structure into multiple parallel filters and fusing them, we can pin-point exactly which satellite is sending corrupted data.
For example, if there are ‘K’ number of satellites in view, then each parallel filter would input pseudorange measurements from only K-1 satellites. Each K-1 bundle of raw GNSS measurements will be used for the measurement update of the parallel filters. A similar concept of measurement separation has been successfully implemented by Honeywell to improve accuracy, integrity, continuity and availability for the Honeywell Inertial GPS Hybrid system . Finally, a master filter fuses all the information from the parallel filters. In our case, the master filter has limited knowledge of the correlation between the banks of filters. In such cases, the covariance intersection(CI) or it’s variant the Bounded Covariance Inflation(BCI) algorithms can be used.  Shows that CI and BCI can be used to fuse information even from black-box type filters that provide no information of error-statistics or correlation. This is done by upper bounding the known covariances to compensate for the unknown correlation in a way that still provides an optimal solution . The ability of CI to fuse information from stand-alone filters can also be exploited to fuse satellite data from multiple constellations (such as GPS, GLONASS etc.).
In this paper, a decentralized INS/GNSS tight integration filter will be implemented and fused using the fusion techniques presented in  (CI, BCI etc). Results of this implementation in MATLAB will be exhibited using raw data obtained from sensor modules (such as swift piksi etc). The trade-offs involved to provide the best estimate will be presented. Finally, the advantages and disadvantages between the centralized and decentralized versions of the tight integration filter will be discussed.
 I. Lakshminarayan and D. Gebre-Egziabher, "Decentralized Filtering for Automatic Reconfiguration of Integrated Navigation Systems," in 29th International Technical Meeting of the ION Satellite Division, ION GNSS+ 2016, Portland, Oregon.
 N. A. Carlson, "Federated Square Root Filter for Decentralized Parallel Processors," IEEE Transactions on Aerospace and Electronic Systems, vol. 26, no. 3, p. 517 – 525, August 1990.
 P. D. Groves, Principles of GNSS, Inertial, and Multisensor Integrated Navigation System, Artech House, 2013.
 C. Call, M. Ibis, J. McDonald and K. Vanderwerf, "Performance of Honeywell's Inertial/GPS Hybrid (HIGH) for RNP Operations," in Proceedings of IEEE/ION PLANS, April 2006.
 L. Chen, P. O. Arambel and R. K. Mehra, "Estimation Under Unknown Correlation: Covariance Intersection Revisited," in IEEE Transactions on Automatic Control, 2002.