PaperlayerView exact revision on arXiv

Source-linked AI summary

Spatial sparse sampling-based iterative optimization framework for GNSS Direct Position Estimation

Wei Gao, Rong Yang, Jihong Huang, Xingqun Zhan, Yonggang Zhang

arXiv:2609.10241v1eess.SP

TL;DR

CAF-peak grid search can require many correlation evaluations, while simpler approaches may discard correlation-domain information. SS-DPE instead uses spatially sparse correlation measurements and their coherence and gradient; urban vehicle experiments report performance comparable to GS-DPE with fewer spatial points and reduced computational cost.

  • Problem

    Conventional GS-DPE relies on CAF maximization and extensive correlation evaluations, while lighter approaches may sacrifice correlation-domain information.

    Source: §I · p. 1

  • Method

    SS-DPE estimates PVT from sparsely sampled correlation measurements, characterizing their spatial coherence and gradient and minimizing measurement residuals.

    Source: §I · p. 2

  • Results

    In urban vehicle tests, SS-DPE and GS-DPE showed comparable performance, with SS-DPE using substantially fewer spatial points and computational cost.

    Source: B. PVT results analysis · p. 12

  • Takeaways & Limitations

    The framework demonstrates PVT estimation using sparse correlation samples while improving information utilization efficiency and reducing computation.

    Source: §I · p. 2, B. PVT results analysis · p. 12

Abstract

from arXiv · show

Direct position estimation (DPE), a promising technique in Global Navigation Satellite Systems (GNSS) receivers, enables estimation of position, velocity, and time (PVT) solutions directly from correlator outputs. The conventional grid search (GS)-based DPE is computationally intensive, as it relies solely on locating the peak of the cross ambiguity function (CAF), and it does not fully leverage the PVT information present in the correlation values. This paper proposes an iterative optimization DPE framework that capitalizes on spatial coherence and spatial gradient via spatial sparse sampling (SS). In SS-DPE framework, the correlation outputs of spatial sampled PVT points all serve as measurements, where the spatial gradient and spatial coherence are derived to capture the information density and diversity of different correlation values. The analytical Cram\'er-Rao Bound (CRB) is derived and indicates that both spatial coherence and gradient determine the theoretical performance limit via the noise covariance and Jacobian matrices analysis. The proposed theoretical framework not only validates the feasibility of sparse sampling but also guides weight optimization to distinct correlation values, effectively integrating these insights with a general gradient-based optimization. Theoretical derivations are validated via Monte Carlo simulations. Field experiments further demonstrate the practical feasibility of the proposed SS-DPE optimization framework. Comparative analysis shows that the proposed SS-DPE achieves comparable PVT estimation accuracy to the conventional GS-DPE while consumes only sparsely sampled correlation values, improving the information utilization efficiency and reducing the computation load.

I. INTRODUCTION

The introduction presents SS-DPE as a sparse-sampling alternative to CAF-peak grid search, using correlation values and their spatial properties to estimate PVT. It first establishes a correlation measurement model from GNSS signal samples and connects it to the paper’s coherence, gradient, and CRB analyses.

I. INTRODUCTION

  • Conventional GS-DPE selects PVT parameters by maximizing the CAF, requiring a large search range and fine steps that increase computation.Coarse-to-fine grids and external sensor information can reduce the burden, while simpler approaches may sacrifice correlation-domain information.Source: §I · p. 1
  • SS-DPE replaces CAF maximization with correlation-measurement residual minimization, using spatial coherence and gradient to estimate PVT from sparse points.The paper’s contributions include deriving a CRB to guide sampling-point selection and using all sampled correlation measurements to improve information utilization efficiency and reduce computation.Source: §I · p. 2

II. THE CORRELATION MEASUREMENT MODEL OF SPATIAL SAMPLING POINTS IN DPE RECEIVER

  • The paper models sparse-point correlations as measurements, characterizing spatial coherence and gradient as the diversity and density of their PVT information.Its framework derives a CRB from these spatial properties, then uses CRB-guided point selection and gradient-based optimization; Monte Carlo simulations and urban vehicle experiments assess the framework.Source: §I · p. 2

A. The linear transformation from raw signal samples to correlation measurementss

  • Correlation values are linear combinations of intermediate-frequency samples, and the paper states that this transformation preserves the PVT information in those samples.The measurement model relates correlation values to the PVT states being estimated.Source: A. The linear transformation from raw… · p. 3

B. The spatial coherence for spatial sampling points

The paper characterizes spatial coherence through the covariance of correlation-measurement noise. It describes coherence within satellite channels and negligible cross-covariance between different satellites.

B. The spatial coherence for spatial sampling points

  • The measurement-noise covariance is approximately block diagonal, with each block describing spatial coherence among sampling points for one satellite channel.The cross-covariance between correlation measurements from different satellites is approximately zero.Source: B. The spatial coherence for spatial… · p. 4
  • Within a satellite channel, spatial coherence between sampling points is characterized by their equivalent code-phase distance.The covariance of correlation-measurement noise mathematically captures the diversity of PVT information across sampling points.Source: B. The spatial coherence for spatial… · p. 4
  • A smaller equivalent code-phase distance means greater spatial coherence and less diversity of PVT information between two points.The coherence arises in the cross-covariance of measurement noise between points.Source: B. The spatial coherence for spatial… · p. 4

C. The spatial gradient for spatial sampling points

The paper uses spatial gradient to characterize PVT information density in correlation measurements. It represents that gradient through a Jacobian, combining the correlation function’s intrinsic slope with its projection onto PVT coordinates.

C. The spatial gradient for spatial sampling points

  • Spatial gradient complements spatial coherence by characterizing the PVT information density in correlation measurements.The gradient is described through the Jacobian of the measurement function.Source: C. The spatial gradient for spatial… · pp. 4–5
  • The paper derives Jacobian terms from the code phase and carrier Doppler derivatives with respect to PVT parameters, with similar analysis for velocity and clock drift.The code-phase and Doppler derivatives are assembled into a Jacobian for the available correlation measurements.Source: C. The spatial gradient for spatial… · p. 5
  • For position and clock bias, the spatial gradient combines the correlation function’s derivative with respect to code-phase error and the code phase’s derivative with respect to PVT parameters.The first term captures the correlation function’s intrinsic gradient; the second projects it onto the position-clock-bias coordinates.Source: C. The spatial gradient for spatial… · p. 5

III. THE CRAM´ER-RAO BOUND ANALYSIS FOR THE SPATIAL SAMPLING-BASED DPE

The section frames the Cramér-Rao Bound as the theoretical performance limit for PVT estimation using spatial information in SS-DPE.

III. THE CRAM´ER-RAO BOUND ANALYSIS FOR THE SPATIAL SAMPLING-BASED DPE

  • Correlation values from all spatial sampling points collectively provide the information used for PVT estimation.The passage describes these values collectively, rather than any one sampled point, as the information for estimation.Source: §III · p. 5
  • The Cramér-Rao Bound represents the theoretical performance limit for PVT parameter estimation in SS-DPE.The bound concerns estimation using the spatial information in the framework.Source: §III · p. 5
  • The section considers the bound for PVT estimation using spatial information from the correlation measurements.Source: §III · p. 5

A. The theoretical Cram´er-Rao Bound for PVT estimation

The paper derives a Cramér-Rao Bound for estimating PVT from spatial correlation measurements. The bound depends jointly on measurement-noise covariance and the Jacobian of the correlation measurements.

A. The theoretical Cram´er-Rao Bound for PVT estimation

  • The inverse Fisher information matrix is the Cramér-Rao Bound matrix, which describes the covariance of estimation results.The bound is derived for PVT estimation from the correlation values.Source: A. The theoretical Cram´er-Rao Bound… · p. 5
  • The correlation measurements are modeled as a function of PVT state Γ with total measurement noise ν, providing the basis for deriving the bound.The measurement vector is z, with total measurement function h(Γ).Source: A. The theoretical Cram´er-Rao Bound… · p. 5
  • The PVT accuracy bound is determined jointly by noise covariance R and correlation-measurement Jacobian J(Γ), representing information diversity and density, respectively.The covariance captures diversity across sampling points, while the Jacobian captures information density at individual points.Source: A. The theoretical Cram´er-Rao Bound… · p. 6

B. The sensitivity analysis for Cram´er-Rao Bound

The sensitivity analysis assesses the contributions of the Jacobian and noise covariance to the Cramér-Rao Bound. It uses derivatives of the bound’s trace and Frobenius norms to quantify those contributions.

B. The sensitivity analysis for Cram´er-Rao Bound

  • The trace of the Cramér-Rao Bound matrix summarizes overall PVT estimation performance, and its derivatives with respect to J(Γ) and R describe their contributions.The analysis focuses on the bound’s diagonal elements for individual PVT parameters and its trace for an overall description.Source: B. The sensitivity analysis for… · p. 6
  • Frobenius norms convert the matrix-form derivatives into scalar measures, and weights WJ and WR represent the proportional contributions of J(Γ) and R.The derivatives are taken with respect to the Jacobian and noise covariance matrix.Source: B. The sensitivity analysis for… · p. 6

IV. THE ITERATIVE OPTIMIZATION FRAMEWORK FOR SS-DPE RECEIVERS

The SS-DPE framework uses spatially sampled correlation measurements in an iterative optimization process. It minimizes weighted squared residuals and addresses singular noise covariance through a blockwise pseudo-inverse procedure.

IV. THE ITERATIVE OPTIMIZATION FRAMEWORK FOR SS-DPE RECEIVERS

  • SS-DPE optimizes PVT using spatially sampled correlations and an iterative framework informed by spatial coherence and gradient analysis.The framework covers the optimization criterion, iteration path, covariance inversion, and sampling strategy.Source: §IV · p. 6

A. The optimization criterion and iteration path determination

  • The criterion minimizes weighted squared residuals from all correlation values, using the measurement model and spatial coherence rather than maximizing the CAF.The iteration direction vectors determine the PVT convergence path, while damping μ balances optimization efficiency and stability.Source: A. The optimization criterion and… · pp. 6–7

B. Pseudo-Inverse solution for noise covariance matrix

  • Spatial coherence can make covariance R singular, so the method computes pseudo-inverses blockwise and regularizes each block with ε before inversion.The resulting pseudo-inverse is used to calculate the Hessian and optimization gradient.Source: B. Pseudo-Inverse solution for noise… · p. 7

C. The selection for spatial sparse sampling points

Spatial sampling choices govern the coherence and gradient used by SS-DPE, shaping information diversity, computational load, and convergence behavior. The paper also describes the iterative implementation and states that Monte Carlo simulations verify its theoretical framework.

C. The selection for spatial sparse sampling points

  • Sampling points are selected along PVT dimensions and combined into a spatial set; α sets the number of intervals and β the step size and sampling range.The search center is typically the previous epoch’s estimate.Source: C. The selection for spatial sparse… · p. 7
  • The larger SS-DPE step size βSS = 150m than GS-DPE βGS = 2m increases information diversity, while fewer intervals reduce required correlations and computation load.The number of required correlation values is KSSM, with KSS = 2(αSS + 1)4; sampling choices also affect CRB and convergence, and a larger step can diminish gradients and potentially trigger divergence.Source: C. The selection for spatial sparse… · pp. 7–8

D. Implementation process for iterative optimization

  • The iterative implementation uses a Levenberg-Marquardt unconstrained optimizer and independently optimizes position-clock bias and velocity-clock drift.Source: D. Implementation process for… · p. 8

SIMULATION VERIFICATION

  • Monte Carlo simulation verification is conducted for the proposed measurement model, CRB analysis, and iterative optimization framework.Source: §V · p. 8

A. Correlation measurement model analysis

The section validates the correlation measurement model against Monte Carlo simulations and outlines how sampled PVT points contribute to iterative SS-DPE estimation.

A. Correlation measurement model analysis

  • The iterative procedure constructs sampled points and local signals, computes code-phase and Doppler quantities, then forms covariance and Jacobian matrices for PVT updates.The algorithm accepts or rejects each update based on the optimization criterion.Source: A. Correlation measurement model… · p. 8
  • The validation uses four satellite channels and αSS = 2, βSS = 100m sampling around the true position-clock bias.This setup produces 4 × 3^4 correlation values.Source: A. Correlation measurement model… · p. 8
  • The theoretical measurement model closely matches simulated correlation measurements, validating its effectiveness.The Fig. 3 curves average 500 Monte Carlo runs.Source: A. Correlation measurement model… · p. 9, Figure 3 · p. 9

B. Spatial coherence analysis

The covariance analysis characterizes coherence among sampled points, while the Jacobian analysis tests the spatial-gradient model against numerical derivatives.

B. Spatial coherence analysis

  • The Monte Carlo covariance matrix has strong same-satellite diagonal blocks and nearly negligible cross-satellite off-diagonal blocks, matching the theoretical diagonal-block textures.This agreement supports the theoretical covariance calculation; the diagonal-block textures may relate to individual satellite geometry.Source: B. Spatial coherence analysis · p. 9, Figure 4 · p. 9

C. Spatial gradient analysis

  • The theoretical Jacobian coefficients closely agree with finite-difference estimates for x- and y-position measurements.The numerical perturbation is 5m.Source: C. Spatial gradient analysis · p. 9, Figure 5 · p. 9
  • Within a satellite, Jacobian coefficients oscillate between positive and negative values because the code-phase derivative contains a sign(Δτm) term.Differences in satellite geometry mainly account for amplitude differences across satellites.Source: C. Spatial gradient analysis · pp. 9–10

D. The Cram´er-Rao Bound analysis

The CRB analysis finds SS-DPE theoretically comparable to GS-DPE and shows that sampling choices affect the bound. The iteration example converges from each tested sampled point to the same position.

D. The Cram´er-Rao Bound analysis

  • With fixed sampling intervals number αSS = 2, CRB increases monotonically as sampling step size βSS grows; at fixed αSS = 4, CRB is lower than with αSS = 2.This occurs because farther sampling points contain less PVT information, whereas, at the same βSS, more sampling points over a wider range provide richer PVT information.Source: D. The Cram´er-Rao Bound analysis · p. 10
  • For velocity-clock drift, CRB first decreases and then increases as βSS grows, with local fluctuations linked to Doppler-derivative behavior inside and outside the main peak.The derivative initially increases and then decreases within the main peak, and oscillates outside it.Source: D. The Cram´er-Rao Bound analysis · p. 10
  • Jacobian weight exceeds 92% for position-clock bias but ranges from 20% to 80% for velocity-clock drift, indicating that sampling should reflect the parameter-specific weighting.The paper attributes the difference to the magnitude of the code-phase derivative term.Source: D. The Cram´er-Rao Bound analysis · p. 10
  • The SS-DPE CRB curves are very close to GS-DPE across varying C/N0; at 45 dB-Hz, SS-DPE reaches 2.3m for position-clock bias and 0.06m/s for velocity-clock drift.A slight CRB gap remains because correlation values are linear combinations of original IF signal points, limiting performance below the original points.Source: D. The Cram´er-Rao Bound analysis · p. 10, Figure 7 · p. 10

E. Iteration convergence and computational complexity

  • With αSS = 2 and βSS = 100m, nine spatial points yield 36 correlations, and all nine initial points converge to (5.86m, 5.87m), near the true position (0m, 0m).During iteration, Jacobian magnitude remains essentially constant while its sign alternates; the combined Jacobian determines the update direction and step.Source: E. Iteration convergence and… · p. 11, Figure 8 · p. 11

VI. EXPERIMENT VERIFICATION

In the vehicle test, SS-DPE achieved PVT accuracy comparable to GS-DPE using sparse spatial sampling. The results also identify a need for further theoretical support to reduce sampling points.

A. Experimental setup

  • A Trimble Zephyr 3 antenna and Labsat 3 recorded GPS L1 C/A IF; an Xsens MTi-670 supplied reference PVT during the 150s viaduct-heavy route.The comparison used the same subset of good-quality satellites for both algorithms.Source: A. Experimental setup · pp. 11–12, Figure 10 · p. 11

VII. CONCLUSION

  • In the vehicle test, 9 spatial sampling points yielded 5m position accuracy and 2m/s velocity accuracy, comparable to GS-DPE with less than 0.05% of its correlation computations.This reduced computation accompanied comparable PVT estimation performance.Source: Figure 11 · p. 12, §VII · p. 13

B. PVT results analysis

  • Both methods stayed within ±5m horizontal position error; velocity errors were within ±2m/s horizontally and ±3m/s vertically, while vertical position showed approximately 15m bias attributed to satellite geometry and atmospheric delays.The comparison used optimized spatial sampling incorporating spatial coherence and gradient.Source: B. PVT results analysis · p. 12, Figure 11 · p. 12

C. Iterative path analysis

  • At 75s, SS-DPE converged to the reference from 9 points even with the energy peak outside the sampling range, versus GS-DPE's 90,601-point plane-covering grid.The SS-DPE path began at the central sampling point.Source: Figure 12 · p. 12, C. Iterative path analysis · p. 12

VII. CONCLUSION

  • The spatial coherence matrix has a block-diagonal structure, while the spatial gradient alternates signs; together, these properties determine the theoretical CRB.Further theoretical support for optimizing spatial sampling to reconstruct PVT with fewer points remains future work.Source: §VII · p. 13
Loading 2609.10241v1…