Improvements on the particle-filter-based GLONASS phase inter-frequency bias estimation approach

• Published in: GPS solutions 2018-07, Vol.22 (3), Article 71
• Main Authors: Tian, Yumiao, Ge, Maorong, Neitzel, Frank, Yuan, Linguo, Huang, Dingfa, Zhou, Letao, Yan, Haoming
• Format: Article
• Language: English
• Subjects: Atmospheric Sciences; Automotive Engineering; Earth and Environmental Science; Earth Sciences; Electrical Engineering; Geophysics/Geodesy; Original Article; Space Exploration and Astronautics; Space Sciences (including Extraterrestrial Physics
• ISSN: 1080-5370; 1521-1886
• DOI: 10.1007/s10291-018-0735-9

The carrier phase inter-frequency bias (IFB) of GLONASS between receivers of different types is usually not zero. This bias must be estimated and removed in data processing so that the integer double difference (DD) ambiguities can be fixed successfully. Recently, the particle filter approach has been proposed to estimate the IFB rate in real time. In this approach, the IFB rate samples are first generated and used to correct the phase IFB in the GLONASS observations. Then, the weights of the rate samples are updated with a function related to RATIO which is for ambiguity acceptance testing in integer ambiguity resolution. Afterwards, the IFB rate is estimated according to the weighted particles. This approach can estimate IFB accurately with short convergence time and without prior information. However, when the system noise is set too low, the estimated results are unstable due to the serious problem of particle diversity-loss, even though the system model is accurate. Additionally, the computational burden is dependent on the number of particles, which has to be optimized for the computation at hand. Therefore, this study proposes two improvements for the IFB estimation in regard to the above two aspects. The first improvement is to solve the noise setting problem by employing a regularized particle filter (RPF). The second improvement optimizes the number of particles in the resampling step according to the standard deviation (STD) of the weighted particles via a controlling function. The two improvements result in significantly better performances. The regularization method allows for the system noise to be set as zero without disturbing the estimates, and consequently, more precise estimates can be achieved. In addition, the approach using the controlling function for adapting the number of particles has comparable performance in precision but the computation load is largely reduced.