Source-linked AI summary
Statistical framework for estimating GNSS bias
Juha Vierinen, Anthea J. Coster, William C. Rideout, Philip J. Erickson, Johannes Norberg
TL;DR
GNSS bias estimation is complicated by instrumental delays, limited-angle observations, and assumptions about ionospheric altitude profiles. The paper develops a weighted statistical framework that estimates biases from scaled TEC differences and applies it across receivers and GNSS systems. Compared with MAPGPS, the new method produces smaller day-to-day receiver-bias variability and more self-consistent vertical TEC maps.
Problem
Instrumental delays, signal-loss errors, and limited-angle GNSS observations complicate estimation of non-ionospheric receiver and satellite biases.
Method
The framework models TEC-difference variance with structure functions and uses weighted linear least squares to estimate biases, including multiple bias terms when needed.
Results
The new method yields smaller day-to-day receiver-bias variability and a more self-consistent vertical TEC map than MAPGPS.
Takeaways & Limitations
The framework supports bias estimation for single stations and large GNSS networks, including GPS and GLONASS measurements.
Takeaways & Limitations
The temporal structure function is not estimated from data in this study and is identified as future work.
Abstract
from arXiv · showhide
We present a statistical framework for estimating global navigation satellite system (GNSS) non-ionospheric differential time delay bias. The biases are estimated by examining differences of measured line integrated electron densities (TEC) that are scaled to equivalent vertical integrated densities. The spatio-temporal variability, instrumentation dependent errors, and errors due to inaccurate ionospheric altitude profile assumptions are modeled as structure functions. These structure functions determine how the TEC differences are weighted in the linear least-squares minimization procedure, which is used to produce the bias estimates. A method for automatic detection and removal of outlier measurements that do not fit into a model of receiver bias is also described. The same statistical framework can be used for a single receiver station, but it also scales to a large global network of receivers. In addition to the Global Positioning System (GPS), the method is also applicable to other dual frequency GNSS systems, such as GLONASS (Globalnaya Navigazionnaya Sputnikovaya Sistema). The use of the framework is demonstrated in practice through several examples. A specific implementation of the methods presented here are used to compute GPS receiver biases for measurements in the MIT Haystack Madrigal distributed database system. Results of the new algorithm are compared with the current MIT Haystack Observatory MAPGPS bias determination algorithm. The new method is found to produce estimates of receiver bias that have reduced day-to-day variability and more consistent coincident vertical TEC values.
1 Introduction
GNSS measurements provide line-integrated ionospheric electron density, but estimating hardware biases and three-dimensional electron density requires simplifying assumptions. The paper introduces a statistical framework that scales slant TEC measurements to vertical TEC and estimates biases through weighted differences.
- Measurement basis: Dual-frequency GNSS receivers infer line-integrated ionospheric electron density from propagation-time differences between radio frequencies.Instrumental frequency-dependent delays and signal-loss discontinuities introduce non-ionospheric errors that must be characterized.
- Inference challenge: Recovering the three-dimensional electron-density function from GNSS line integrals is a limited-angle tomography problem requiring additional smoothness assumptions.Full tomographic solvers are computationally costly, motivating use of vertical total electron content.
- Vertical TEC model: Vertical TEC processing converts slanted path integrals into equivalent vertical integrals using an elevation-dependent mapping factor v(α).The mapping factor can be derived from assumed altitude profiles, including Chapman profiles or slab profiles with exponential ramps.
- Model trade-off: The simplified vertical TEC model is less flexible than full tomography but yields an over-determined, well-posed problem useful for ionospheric studies.More advanced mapping functions may also incorporate time, location, solar activity, and ray azimuth.
- Bias estimation: WLLSID estimates receiver and satellite biases from many scaled TEC differences, weighting them by variance determined by temporal spacing, geographic distance, and elevation angle.The framework is compared with the existing MIT Haystack MAPGPS bias-determination scheme.
2 Receiver bias estimation
Receiver-bias estimation compares vertically scaled TEC measurements and models their expected differences with structure functions. These functions account for spatial and temporal variability, receiver noise, and errors from vertical-TEC scaling assumptions.
- 2 Receiver bias estimation: Vertically scaled TEC measurements are compared pairwise, regardless of whether they share a receiver, satellite, or observation time.The comparison uses mapping-function values to convert slanted TEC into equivalent vertical TEC.
- 2 Receiver bias estimation: Pairs with nearby times and pierce points are assumed to have similar equivalent vertical TEC.This similarity motivates treating their difference as a normally distributed random variable.
- 2 Receiver bias estimation: The structure function variance depends on pierce-point distance, time separation, receiver noise, and elevation-dependent scaling errors.Scaling errors arise because slanted paths sample geographic areas and because assumed ionospheric altitude profiles are imperfect.
- 2.1 Geographic distance: The spatial structure function used here assumes TEC-difference standard deviation grows by 0.5 TEC units per 100 km between pierce points.The paper notes that a more complex location- and time-dependent function could better represent sunrise, sunset, and regional gradients.
- 2.2 Temporal distance: The temporal structure function models larger vertical-TEC differences for measurements taken farther apart in time.Its estimation from data is identified as future work.
- 2.3 Model and receiver errors: Low-elevation measurements receive heavier modeling-error penalties because multipath and vertical-TEC scaling errors increase toward the horizon.The chosen elevation structure function makes the variance grow rapidly as elevation approaches the horizon.
3 Generalized linear least-squares solution
The generalized solution casts TEC differences as a linear statistical inverse problem for receiver and satellite biases. Structure-function variances whiten the system for sparse least-squares estimation, followed by iterative residual-based outlier removal.
- 3 Generalized linear least-squares solution: Assuming independent structure-function components, the full variance is obtained by adding their contributions.This produces the covariance used for the difference measurements.
- 3 Generalized linear least-squares solution: The measurement vector contains differences between vertically scaled measurements, while the unknown vector contains receiver and satellite biases.The theory matrix provides the forward model linking measurements to these bias parameters.
- 3 Generalized linear least-squares solution: The bias-estimation problem is a linear statistical inverse problem with a closed-form maximum-likelihood solution.The paper identifies the unknown vector as the receiver and satellite biases.
- 3 Generalized linear least-squares solution: Large sparse systems are solved with sparse linear least-squares methods after each measurement difference is scaled by the square root of its variance.This whitening transformation projects the covariance matrix to the identity matrix.
- 3 Generalized linear least-squares solution: Residuals exceeding a 4σ threshold identify measurements that do not consistently fit the receiver-bias model.Problematic measurements are removed and the maximum-likelihood solution is recomputed iteratively.
4 Special cases
The framework supports practical special cases for known satellite bias, single-receiver estimation, and multiple receiver-bias terms. Multiple biases accommodate discontinuities, drift, and receiver instability while preserving comparable or more consistent TEC estimates.
- Known satellite bias: Known satellite biases can be subtracted before solving, reducing the unknowns; this special case is appropriate for GPS receivers.The resulting theory matrix has at most two non-zero elements per row.
- Single receiver bias estimation: For a single receiver with known satellite bias, the overdetermined bias problem can be solved using time differences and resembles scalloping.A comparison with standard MAPGPS produced quite similar results in the example.
- Multiple biases: The framework assigns independent bias parameters to continuous TEC-curve segments after phase-lock loss, provided enough overlapping measurements are available.This avoids forcing discontinuous curves to remain continuous through realignment.
- Multiple biases: Multiple bias terms can represent receiver-bias variation across satellite passes, including effects associated with temperature-dependent receiver behavior.Each receiver may have many unknown bias parameters rather than one fixed parameter.
- Multiple biases: In a 19-receiver China example, multiple biases recovered measurements from stations whose discrete TEC jumps made a constant receiver bias untenable.The standard MAPGPS algorithm produced a poor fit for these unstable receivers.
5 Comparison
The new WLLSID bias algorithm was implemented in MAPGPS and compared with the existing method using coincident vertical TEC measurements and day-to-day receiver-bias changes. It produced more self-consistent TEC estimates and lower day-to-day variability in the reported comparison.
- Implementation and setup: The WLLSID implementation replaced MAPGPS bias routines for daily analysis of data from over 5000 receivers.The comparison assumed fixed 24-hour receiver biases, known satellite biases, and selected neighboring-station measurement differences within five minutes.
- Self consistency comparison: The comparison evaluated self-consistency using the mean absolute difference between coincident vertical TEC measurements from different receivers.Coincident measurements had pierce points within 50 km and acquisition times within 30 seconds.
- Self consistency comparison: 1.62 TEC units versus 2.25 TEC units: WLLSID's figure of merit was about 30% better than MAPGPS across 192360 coincidences from 5220 GPS receivers.Smaller figure-of-merit values indicate more consistent vertical TEC measurements.
- Self consistency comparison: The new method produced significantly more coincident vertical TEC differences below 1 TEC unit than the old method.Some large disagreements remained, at least partly because the comparison included measurements down to 10° elevation.
- Receiver bias day-to-day change: WLLSID reduced day-to-day receiver-bias variability to σ_b = 1.3 TEC units from MAPGPS's σ_b = 1.6 TEC units.The MAPGPS mean change was δb = −0.2 ± 0.05(2σ), whereas WLLSID's was δb = 0.02 ± 0.05(2σ).
6 Conclusions
The framework estimates GNSS receiver biases through weighted measurement differences and scales to large receiver networks. Compared with MAPGPS, it produces less day-to-day receiver-bias variability and more self-consistent vertical TEC maps, while leaving weighting and difference-selection strategies open for improvement.
- The framework converts GNSS measurement differences into a linear model solved by least squares, with sparse-matrix implementation for extremely large receiver networks.
- Compared with MAPGPS, the new method produces smaller day-to-day variability in receiver bias and a more self-consistent vertical TEC map.
- The weighting structure function is not guaranteed to be optimal, motivating future data-derived improvements.
- Only two measurement-difference types were explored, leaving other difference-selection strategies for future work.
- Special cases cover known satellite bias, single receivers with known satellite bias, and multiple bias terms per receiver for GPS and GLONASS-related measurements.