Federated Geometry Filter for GNSS Network Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for processing data from a network of Global Navigation Satellite Systems (GNSS) reference stations are inefficient, leading to long processing times that become impractical as the number of reference stations increases, especially when using centralized Kalman filters.
Innovation Solution
Implementing an ionosphere-free, federated geometry filter that reduces computation time by distributing processing between local station geometry filters and a master filter, allowing for parallel operation and reduced states, thereby enabling real-time processing even with a large number of reference stations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a centralized Kalman filter is used to process data from multiple reference stations, then the processing can be performed with a single system, but the processing time becomes extremely long as the number of reference stations increases
Solution Approach 1:
The patent divides the centralized Kalman filter into multiple distributed Kalman filters, each operating at individual reference stations. Each filter processes data independently and only exchanges essential information (common state estimates) with others, transforming a monolithic system into modular segments that scale linearly with the number of stations rather than cubically.
Solution Approach 2:
The patent introduces a communication network as an intermediary between reference stations, enabling them to share common state estimates (satellite clock errors, orbit errors) without requiring direct central coordination. This intermediary layer allows distributed processing while maintaining system coherence through periodic information exchange.
2Reliability
If the number of reference stations is increased to improve coverage and accuracy, then the positioning reliability improves, but the processing time becomes impractical
Solution Approach 1:
By segmenting the processing into distributed filters at each reference station, the system can handle large numbers of stations without exponential processing time increases. Each station maintains its own filter state and only communicates necessary corrections, enabling real-time processing even with extensive networks.
Solution Approach 2:
The patent applies partial processing at each reference station (local state estimation) rather than complete centralized processing. Each station performs sufficient processing to generate local corrections and common state estimates, then stops there, avoiding the excessive computational burden of full centralized processing while still achieving reliable positioning.
3Measurement precision
If a centralized Kalman filter with many states is used to estimate common parameters, then the estimation accuracy improves, but the computation load becomes cubically proportional to the number of reference stations
Solution Approach 1:
The patent segments the large state vector into local states (unique to each station) and common states (shared across stations). Each distributed filter maintains only local states, reducing individual computation load from cubic to linear complexity while common states are estimated through information exchange, preserving overall estimation accuracy.
Solution Approach 2:
The patent assigns different computational qualities to different parts of the system: local filters perform detailed processing for station-specific parameters, while common parameters are estimated through distributed consensus. This local quality differentiation allows accurate estimation without uniform high computational cost across the entire system.
Data Source
AI summary
Methods and apparatus for processing of data from a network of GNSS reference stations are presented. An ionosphere-free, federated geometry filter is employed so that computation time increases only linearly with the increase in number of reference stations, significantly reducing processing time as compared to a centralized filter approach.