PaperlayerView exact revision on arXiv

Source-linked AI summary

Evaluation of Position-related Information in Multipath Components for Indoor Positioning

Erik Leitinger, Paul Meissner, Christoph Rüdisser, Gregor Dumphart, Klaus Witrisal

arXiv:1409.1467v2cs.IT

TL;DR

Indoor positioning must extract reliable information despite harsh multipath and diverse transmission configurations. The paper develops a unified EFIM-based framework for CRLB analysis across bistatic, monostatic, and cooperative setups, including clock offsets. Its results quantify how environment properties, system parameters, diffuse multipath, and path overlap shape available position-related information.

  • Problem

    Indoor positioning needs robust location information despite harsh multipath and the limitations of position-dependent signal-strength measurements.

  • Method

    The paper uses a geometric-stochastic channel model and equivalent Fisher information to derive CRLBs for bistatic, monostatic, cooperative, and clock-offset configurations.

  • Results

    The framework quantifies how diffuse multipath, path overlap, synchronization, cooperation, and monostatic geometry affect position-related information.

  • Takeaways & Limitations

    Position-related information can guide positioning-system design, including channel-model parametrization, antenna placement, and signal-parameter choices.

Abstract

from arXiv · show

Location awareness is a key factor for a wealth of wireless indoor applications. Its provision requires the careful fusion of diverse information sources. For agents that use radio signals for localization, this information may either come from signal transmissions with respect to fixed anchors, from cooperative transmissions inbetween agents, or from radar-like monostatic transmissions. Using a-priori knowledge of a floor plan of the environment, specular multipath components can be exploited, based on a geometric-stochastic channel model. In this paper, a unified framework is presented for the quantification of this type of position-related information, using the concept of equivalent Fisher information. We derive analytical results for the Cramér-Rao lower bound of multipath-assisted positioning, considering bistatic transmissions between agents and fixed anchors, monostatic transmissions from agents, cooperative measurements inbetween agents, and combinations thereof, including the effect of clock offsets. Awareness of this information enables highly accurate and robust indoor positioning. Computational results show the applicability of the framework for the characterization of the localization capabilities of a given environment, quantifying the influence of different system setups, signal parameters, and the impact of path overlap.

I. INTRODUCTION

The paper frames indoor positioning as a problem of extracting reliable location information from harsh multipath environments. It introduces a unified information-based framework covering geometric multipath, multiple measurement setups, clock offsets, and diffuse-multipath effects.

  • Motivation: Indoor location awareness remains difficult because harsh multipath conditions and position-dependent signal-strength variance limit robust accuracy.The paper motivates fusing multiple information sources or exploiting multipath-aware sensing.
  • Geometric-stochastic channel model: Specular multipath components can be modeled as signals from virtual anchors computed from a known floor plan.The geometric model represents wall reflections through mirror images of anchors, while diffuse multipath is modeled statistically.
  • Information quantification: Equivalent Fisher information matrices enable analytical evaluation of Cramér-Rao bounds for position-related information.The CRLB provides a lower bound on the covariance of an unbiased estimator, and EFIMs support blockwise inversion of the FIM.
  • Measurement scenarios: The framework covers bistatic, monostatic, and cooperative measurements, including known or unknown clock offsets.The scenarios include fixed-anchor transmissions, agent cooperation, and self-referenced monostatic measurements.
  • Implications and validation: The position-related FIM reflects both localization accuracy and robustness, increasing with the number of useful multipath components.The paper also introduces SINR as a measure of position-related information and validates the bounds with real measurements.
  • Geometric-stochastic channel model: A geometric-stochastic model with known diffuse-multipath statistics supports closed-form analysis of diffuse interference on the positioning bound.The diffuse-multipath power-delay profile represents the power ratio between deterministic MPCs and diffuse components.

III. CRAM´ER-RAO LOWER BOUND

The section derives the CRLB for multipath-assisted positioning from the received-signal model, using Fisher information and equivalent Fisher information to isolate position-related information. It also characterizes delay information under orthogonal MPCs through SINR, bandwidth, and diffuse-multipath effects.

  • The position estimate uses known virtual-anchor positions from the waveform, while complex MPC amplitudes are treated as nuisance parameters.
  • The Fisher information matrix is computed from the received-signal likelihood under additive white Gaussian noise and Gaussian diffuse multipath.The sampled signal contains delayed transmit pulses, and its noise covariance combines AWGN with diffuse-multipath covariance.
  • 2) Orthogonal MPCs: The CRLB can be evaluated numerically without further assumptions, but whitening by the inverse noise covariance limits direct insight into the result.Assuming non-overlapping MPCs makes the deterministic components orthogonal and simplifies the information structure.
  • 2) Orthogonal MPCs: With negligible diffuse multipath, or a block-spectrum pulse, whitening provides no bandwidth extension; when AWGN is negligible, the spectrum corresponds to the absolute bandwidth.
  • 2) Orthogonal MPCs: 7 dB is the maximum approximate SINR gain from whitening at R = 1, while γk reaches 4 dB at R = 0.6 and INR of 15 dB.
  • 2) Orthogonal MPCs: The delay information of each MPC is quantified by the product of its SINR, effective bandwidth, and bandwidth extension factor.The interference term depends on the diffuse-multipath power delay profile at the MPC delay and scales with pulse duration.

C. Position Error Bound

The position error bound is obtained from the position-relevant block of the Fisher information matrix through blockwise inversion, yielding an equivalent Fisher information matrix. Spatial delay gradients connect MPC delays to agent and anchor positions across positioning scenarios.

  • The position error bound is the square root of the trace of the CRLB’s position submatrix.
  • Blockwise inversion produces the equivalent Fisher information matrix, which retains the information relevant to the position error bound.
  • The spatial delay-gradient matrix H = ∂τ/∂p describes how position changes affect the vector of MPC delays.Its form depends on whether derivatives are taken with respect to agent, anchor, or cooperating-agent positions.
  • Fixed anchors and cooperating agents are represented through their positions and associated virtual anchors in the gradient formulation.
  • Each MPC delay is defined by the distance between its virtual anchor and the agent, linking measured delays to environment geometry.
  • The gradient matrices stack transposed MPC gradient vectors for individual links and combine derivatives across all agents.

1) Bistatic scenario:

The bistatic-gradient analysis differentiates MPC delays with respect to agent and anchor positions. These gradients are represented geometrically through vectors toward virtual anchors and support the CRLB derivations for multiple positioning scenarios.

  • The agent-gradient case differentiates each MPC delay with respect to the agent position and forms a corresponding gradient matrix.
  • The anchor-gradient case differentiates MPC delays with respect to the anchor position, including cooperative agents acting as anchors.
  • In the monostatic case, the agent also acts as the anchor and moves synchronously, so the gradient terms interact.
  • Single reflections and rectangular-corner reflections form important monostatic virtual-anchor configurations.
  • Twice the spatial sensitivity of delays occurs for the specified monostatic reflections compared with bistatic cases, whereas second-order parallel-wall reflections can have zero gradient magnitude.
  • The CRLB derivation covers Multipath-Sync, Multipath-NSync, and Multipath-Coop scenarios using a Jacobian over position, clock offsets, delays, and complex amplitudes.

A. Derivation of the CRLB for Multipath-Sync

The derivation uses additive equivalent Fisher information matrices to quantify position information from multipath components, including bandwidth, diffuse multipath, path overlap, and unknown clock offsets.

  • Multipath-Sync: Independent anchor measurements produce additive position-related EFIMs under the assumed signaling scheme.The derivation combines the EFIM contributions from the different anchors.
  • Multipath-Sync: Each deterministic multipath component adds positive EFIM information along its ranging direction.The contribution magnitude depends on its extended SINR and is affected by diffuse multipath and interference whitening.
  • Multipath-Sync: Increased effective bandwidth scales the EFIM and decreases the position error bound.This scaling is part of the analytical characterization of multipath-assisted positioning.
  • Path overlap: When two multipath components arrive much closer than the pulse duration, their position-related information is entirely lost.For delays comparable to the pulse duration, the components remain correlated but can still provide partial information; widely separated components are treated as orthogonal.
  • Multipath-NSync: Unknown clock offsets reduce position information and increase the PEB unless the condition c = 0 holds.Multipath-NSync can theoretically match Multipath-Sync only under that rather unlikely condition.

C. Derivation of the CRLB for Multipath-Coop

The cooperative derivation combines bistatic measurements between agents and anchors with monostatic measurements into a joint EFIM for all agent positions.

  • Multipath-Coop: Each agent performs monostatic measurements and bistatic measurements with other agents and fixed anchors.The framework distributes received and transmitted signals so every agent can exploit the available measurements.
  • Multipath-Coop: The cooperative EFIM is constructed from stacked agent positions, signal parameters, and independent measurements.Spatial delay gradients connect the received signal parameters to the agents’ positions.
  • Multipath-Coop: The resulting EFIM separates contributions from other agents, fixed anchors, and each agent’s monostatic measurement.Its off-diagonal blocks represent interactions between cooperating agents.
  • Multipath-Coop: Uncertainty about cooperating agents acting as anchors negatively affects localization performance.Repeated measurements between agents introduce factors of two; omitting those repetitions removes the corresponding factors.

VI. RESULTS

The computational results validate the framework with measured data and use synthetic environments to examine anchors, bandwidth, path overlap, and channel-model assumptions.

  • Validation with Measurement Data: The validation uses measured MPC SINRs and compares them with values from the parametric channel model.The model produces realistic SINRs in many cases and therefore valid performance bounds, although its global diffuse-multipath model cannot describe local behavior.
  • Validation with Measurement Data: The validation environment achieves a PEB below 10 cm across almost the entire area using estimated SINRs and synchronized measurements.The geometrically decomposed CRLB ellipses closely match those from a multipath-assisted tracking algorithm.
  • Multipath-Sync: A single anchor yields a PEB below 10 cm for most of the synthetic room, including positions where the anchor is partly invisible.The spatial pattern reflects visibility regions associated with virtual-anchor modeled multipath components.
  • Multipath-Sync: Considering path overlap exposes adverse effects from room symmetries, while two anchors substantially reduce both the PEB and the overlap impact.The comparison uses the full PEB with path overlap for one versus two fixed anchors.
  • Multipath-Sync: PEB increases vastly as pulse duration increases from 0.5 ns to 2 ns in the evaluated cumulative distributions.The no-path-overlap results reflect Fisher-information scaling with bandwidth and increased diffuse-multipath interference.
  • Multipath-Sync: Algorithms based on the presented signal model obtain centimeter-level accuracy for 90% of estimates.The reported algorithms are described as closely approaching the derived bounds.

2) Multipath-NSync:

The paper evaluates multipath-assisted positioning across synchronized, nonsynchronized, and cooperative measurement setups using PEB analyses. Clock-offset estimation reduces position-related information, while synchronization, bandwidth, and complementary measurements mitigate path-overlap effects.

  • Multipath-NSync: Unknown clock offsets reduce positioning performance because some delay information is used for clock-offset estimation.
  • Multipath-NSync: A second anchor counteracts the information loss from clock-offset estimation, with additional gains when the anchors are synchronized.
  • Multipath-NSync: Path overlap has a larger impact in the nonsynchronized two-anchor case than in the synchronized case.
  • Multipath-NSync: Increasing VA order generally increases information, but higher-order reflections can worsen performance where path overlap becomes unresolvable, especially near walls and corners.
  • Multipath-Coop: Cooperative positioning combines monostatic and cooperative information to provide excellent performance across the room, although uncertain cooperating-agent positions limit the cooperative contribution.
  • Multipath-Coop: In Multipath-Coop, second-order MPCs do not improve overall performance because monostatic reflections create unresolvable overlap, whereas higher VA order remains beneficial for cooperative measurements.
  • Multipath-Coop: Larger bandwidth reduces unresolvable path overlap in monostatic measurements, producing a clear advantage for Multipath-Coop.
  • Conclusions: The framework quantifies how diffuse multipath, path overlap, synchronization, cooperation, and bandwidth shape position-related information for indoor localization.

APPENDIX A FIM FOR ORTHOGONAL MPCS

Appendix A develops the signal and interference model used to obtain the Fisher information for orthogonal multipath components. It incorporates diffuse multipath through covariance modeling and relates each component’s information to bandwidth, SINR, and whitening.

  • Signal and covariance model: The received-signal covariance combines additive white Gaussian noise with diffuse multipath interference.
  • Signal and covariance model: A circulant signal matrix and a unitary DFT diagonalization enable frequency-domain evaluation of the covariance inverse.
  • Fisher-information calculation: The Woodbury identity provides the inverse covariance expression used in the likelihood and Fisher-information calculations.
  • Fisher-information calculation: The transmitted-signal autocorrelation samples the nonstationary power-delay profile near each MPC delay, motivating a stationary PDP approximation for each component.
  • Fisher-information calculation: Each MPC’s information depends on mean-square bandwidth, SINR, and a bandwidth-extension factor determined by diffuse-multipath interference relative to noise.

APPENDIX B JACOBIAN OF VA POSITION W.R.T. ANCHOR POSITION

Appendix B derives the Jacobian of a virtual-anchor position with respect to its physical anchor by composing wall reflections. The resulting expressions expose how reflection order and wall geometry affect position sensitivity.

  • Virtual-anchor construction: A virtual anchor is constructed by repeatedly mirroring the physical position across the walls encountered by the multipath component.
  • Reflection operators: The mirror operation uses a matrix associated with each wall angle and offset, with equivalent rotation and sign-flip representations.
  • Jacobian derivation: The Jacobian with respect to the physical position is obtained from the leading product of mirror matrices, with transposition reversing multiplication order.
  • Jacobian derivation: An effective wall angle summarizes the alternating contribution of successive reflection-wall angles.
  • Monostatic specialization: The monostatic gradient is transformed into a magnitude-times-unit-vector form whose arguments depend on reflection-order parity.

APPENDIX D DERIVATION OF THE NSYNC CRLB

Appendix D derives the equivalent Fisher information for nonsynchronized anchors by transforming the signal-parameter information and eliminating clock-offset parameters through blockwise inversion. The resulting EFIM remains additive across anchors under the stated constructions.

  • Synchronized anchors: The synchronized-anchor derivation partitions the transformation matrix into spatial and clock-offset submatrices before forming the 3 × 3 EFIM.
  • Synchronized anchors: When path overlap is neglected, the transformed expression reduces to the simplified result used in the main derivation.
  • Asynchronous anchors: For asynchronous anchors, the clock-offset parameter becomes a vector and the blockwise inversion is applied twice to derive the position EFIM and anchor-wise additivity.
  • Synchronized anchors: Blockwise inversion yields additive EFIM contributions from the individual anchors.
  • Cooperative setup: The cooperative EFIM is decomposed into contributions from independent transmissions between agents and between agents and fixed anchors.

1) Off-diagonal blocks η̸ = η′:

The section concerns off-diagonal information blocks associated with monostatic and bistatic measurements. The supplied passages also contain MPC SINR comparisons and several location-specific values.

  • 1) Off-diagonal blocks η̸ = η′:: Off-diagonal contributions include monostatic measurements, bistatic measurements between agents, and bistatic measurements between agents and fixed anchors.
  • 1) Off-diagonal blocks η̸ = η′:: The reported MPC comparison uses measured and modeled SINR in decibels.
  • 1) Off-diagonal blocks η̸ = η′:: The listed location labels include left wall, upper wall, lower wall, right window, and upper window.
  • 1) Off-diagonal blocks η̸ = η′:: Other reported parameters include 29.5 dB at 1 m and 1 ns, with an interval shown as (0.5 ns,2 ns).
Loading 1409.1467v2…