PaperlayerView exact revision on arXiv

Source-linked AI summary

Fundamental Limits of Wideband Localization - Part I: A General Framework

Yuan Shen, Moe Z. Win

arXiv:1006.0888v1cs.ITcs.NI

TL;DR

Accurate localization is needed in applications where harsh multipath environments challenge conventional positioning. The paper analyzes received waveforms directly and uses SPEB and EFI to express localization information and its fundamental accuracy limit. It further characterizes contributions from anchors, prior knowledge, antenna arrays, and clock asynchronism.

  • Problem

    Localization accuracy must be fundamentally characterized for wideband wireless networks operating in harsh multipath and NLOS environments.

  • Method

    The paper derives SPEB from received waveforms using EFI, unifying anchor information with prior knowledge of channel parameters and agent position.

  • Results

    The framework expresses anchor contributions as weighted ranging-direction matrices and shows that antenna-array AOA does not improve accuracy beyond TOA alone.

  • Takeaways & Limitations

    SPEB and EFI provide a canonical framework for assessing localization limits while incorporating measurements, prior knowledge, multipath, NLOS propagation, and synchronization effects.

Abstract

from arXiv · show

The availability of positional information is of great importance in many commercial, public safety, and military applications. The coming years will see the emergence of location-aware networks with sub-meter accuracy, relying on accurate range measurements provided by wide bandwidth transmissions. In this two-part paper, we determine the fundamental limits of localization accuracy of wideband wireless networks in harsh multipath environments. We first develop a general framework to characterize the localization accuracy of a given node here and then extend our analysis to cooperative location-aware networks in Part II. In this paper, we characterize localization accuracy in terms of a performance measure called the squared position error bound (SPEB), and introduce the notion of equivalent Fisher information to derive the SPEB in a succinct expression. This methodology provides insights into the essence of the localization problem by unifying localization information from individual anchors and information from a priori knowledge of the agent's position in a canonical form. Our analysis begins with the received waveforms themselves rather than utilizing only the signal metrics extracted from these waveforms, such as time-of-arrival and received signal strength. Hence, our framework exploits all the information inherent in the received waveforms, and the resulting SPEB serves as a fundamental limit of localization accuracy.

I. INTRODUCTION

Wideband location-aware networks address the need for accurate positioning where GPS is ineffective, but harsh multipath and NLOS propagation make localization uncertain. The paper develops waveform-based fundamental limits using SPEB and EFI, including effects of channel knowledge, antenna arrays, and clock asynchronism.

  • GPS is limited in buildings, urban canyons, under tree canopies, and caves because its signals cannot penetrate most obstacles.
  • Wideband signals support accurate localization through fine delay resolution and robust range measurements in harsh environments.
  • Received-waveform analysis is necessary because extracted metrics can discard localization information and depend heavily on measurement-specific models.
  • The paper derives fundamental localization limits in multipath and NLOS conditions using the squared position error bound (SPEB).
  • Equivalent Fisher information (EFI) combines information from different anchors as a weighted sum of their direction matrices.
  • The framework quantifies prior channel and position knowledge, antenna-array limits, and clock-asynchronism effects on localization accuracy.

B. Error Bounds on Position Estimation

The framework models localization from received waveforms and nuisance multipath parameters, then uses Fisher information and the SPEB to bound position-estimation accuracy.

  • Model and information bound: The parameter vector includes the agent’s position and nuisance multipath parameters associated with each received waveform.The received waveforms are represented as a vector obtained through Karhunen–Loève expansions.
  • Model and information bound: The information inequality lower-bounds the mean squared error matrix of position estimates under deterministic, hybrid, and Bayesian parameter settings.The Fisher information matrix is derived for deterministic and random parameter estimation to evaluate the SPEB.
  • Model and information bound: The squared position error bound (SPEB) is defined as a measure of the limits of position accuracy derived from the Fisher information matrix.For three-dimensional localization, the corresponding position-information submatrix is 3 × 3 rather than 2 × 2.
  • Fisher information construction: When received waveforms from different anchors are independent, their likelihood information can be decomposed across anchors.The likelihood ratio is written using this independence, while LOS and NLOS contributions are represented separately in the information matrices.
  • Fisher information construction: A priori knowledge is incorporated through the joint PDF of the agent’s position and channel parameters, adding separate information from position and conditional multipath knowledge.The resulting prior-information terms include the Fisher information of the agent’s position and the joint position–channel information.

D. Equivalent Fisher Information Matrix

Equivalent Fisher information reduces the high-dimensional Fisher information matrix to the position-relevant information while preserving the bound needed for localization accuracy.

  • Equivalent Fisher information: The EFIM is obtained from a block-partitioned Fisher information matrix through the Schur complement of the nuisance-parameter block.This avoids directly inverting the full high-dimensional matrix when only a small position submatrix is needed.
  • Equivalent Fisher information: The EFIM retains the information needed to derive the information inequality and lower-bound the estimation error for the parameters of interest.For two-dimensional localization, the goal is a 2 × 2 EFIM for the agent’s position.
  • Equivalent Fisher information: The EFIM for the agent’s position is reduced to a 2 × 2 matrix when a priori knowledge is unavailable.The reduction is applied to the original Fisher information matrix to eliminate nuisance dimensions.
  • NLOS and resolvability: Without a priori knowledge, NLOS signals do not contribute to the EFIM for the agent’s position.When the first LOS path is resolvable, the ranging information reaches its maximum; overlapping paths instead degrade first-path arrival-time estimation.
  • Ranging information: Only the first contiguous cluster of LOS paths contains localization information, where paths remain connected by interarrival gaps shorter than the signal duration.The first contiguous cluster is the first group of non-disjoint paths.
  • Ranging information: Ranging information is represented as λJr(φ), combining a nonnegative ranging-information intensity with a ranging-direction matrix.The ranging-direction matrix is one-dimensional along direction φ and has one nonzero eigenvalue equal to 1.
  • Ranging information: Each anchor contributes one-dimensional ranging information along its direction to the agent, with intensity determined by effective bandwidth, first-path SNR, and path overlap.Path overlap reduces ranging-information intensity and therefore leads to a higher SPEB unless the first path does not overlap other paths.

B. EFIM with A Priori Knowledge

With a priori channel or position knowledge, the EFIM provides a canonical representation of localization information and explains when NLOS paths contribute. The framework also extends to wideband antenna arrays with unknown orientation.

  • A Priori Channel Knowledge: Theorem 2 expresses the position EFIM as a weighted sum of individual anchors’ direction matrices when channel parameters are known a priori and anchor parameter sets are independent.This generalizes the deterministic-parameter result to hybrid parameter estimation.
  • A Priori Channel Knowledge: A priori channel knowledge increases the ranging information intensity, making NLOS signals contribute positively to the EFIM.Without such knowledge, the NLOS ranging information intensity reduces to zero.
  • A Priori Channel Knowledge: LOS signals can be represented within the same Bayesian framework as NLOS signals assigned infinite a priori Fisher information for their first-path biases.This unifies LOS and NLOS modeling under one formulation.
  • A Priori Position Knowledge: When the position prior satisfies the stated condition, the position EFIM combines anchor-derived information with information from the agent’s a priori position knowledge.The condition holds when the position distribution is concentrated in a small area relative to anchor distances, including far-field scenarios.
  • Antenna Arrays: Wideband antenna-array localization jointly models the agent’s position and array orientation and derives both the SPEB and squared orientation error bound.The array model represents antenna positions using a reference point, orientation, and known relative antenna locations.

B. EFIM without A Priori Knowledge

Without a priori knowledge, the antenna-array EFIM separates individual-antenna information from the loss caused by uncertain orientation. The analysis identifies when arrays improve localization and how the reference point affects the SPEB.

  • Array EFIM: Theorem 3 derives the position and orientation EFIMs for an Na-antenna array when a priori knowledge is unavailable.The resulting position information includes the effects of array geometry and orientation uncertainty.
  • Array EFIM: Uncertain orientation reduces the array’s position EFIM, so it is bounded above by the sum of the individual-antenna EFIMs.Equality occurs for q = 0 or orientation-aware localization.
  • Array EFIM: The array EFIM depends on per-anchor, per-antenna ranging information and array geometry, so jointly processing antenna waveforms is unnecessary in this model.The analysis therefore finds no additional position-accuracy increase from indirectly obtained AOA beyond the TOA information already used by individual antennas.
  • Reference Point: The orientation center is the unique reference point minimizing the SPEB in orientation-unaware localization.In orientation-aware localization, the equality condition is achieved independently of the reference point.
  • Reference Point: The SPEB at any reference point equals the orientation-center SPEB plus an orientation-induced error proportional to squared distance from that center and the SOEB.Near-field scenarios may provide additional spatial-diversity gain from multiple antennas.

C. EFIM with A Priori Knowledge

The framework extends a priori knowledge and asynchronous-clock analysis to antenna arrays and far-field configurations. It characterizes how channel, position, orientation, and clock knowledge affect the EFIM and SPEB.

  • A Priori Array Knowledge: With a priori channel knowledge and independent channel-parameter sets, the antenna-array ranging information intensity is given by the specialized expression in Proposition 4.The result applies to the array setting with independent parameters across anchors and antennas.
  • A Priori Array Knowledge: When position and orientation priors satisfy the stated condition, the array EFIMs incorporate a priori position and orientation information alongside channel-derived information.The derivation uses an approximation based on the mean position and orientation under that condition.
  • Far-Field Arrays: In far-field scenarios, the array center becomes the orientation center and therefore achieves the minimum SPEB.Its location can be determined from array geometry alone, without received waveforms or anchor-topology knowledge.
  • Far-Field Arrays: In far-field scenarios, an Na-antenna array has localization performance equivalent to a single antenna with Na measurements, regardless of array geometry.Multiple antennas at each anchor can be viewed as individual anchors and reduce the agent’s SPEB as their number increases.
  • Clock Asynchronism: Theorem 4 derives position and time-offset EFIMs when channel parameters and the time offset are known a priori with independent anchor channel parameters.Equality with the corresponding known-offset limit occurs when the offset is accurately known or time-offset-independent.
  • Clock Asynchronism: An uncertain agent-anchor time offset increases or preserves the SPEB relative to localization with no offset or a known offset.The degradation results from uncertainty in the additional time-offset parameter.

B. Localization with Antenna Arrays

The antenna-array analysis derives EFIMs for position, orientation, and time offset under varying a priori knowledge and array configurations. The framework also connects waveform-based analysis to signal metrics and establishes SPEB achievability at high SNR.

  • B. Localization with Antenna Arrays: Theorem 5 gives the overall EFIM for position, orientation, and time offset with an Na-antenna array and a priori channel knowledge.Orientation-aware and orientation-unaware localization correspond to Ξ = ∞ and Ξ = 0, respectively.
  • B. Localization with Antenna Arrays: Theorem 5 supports deriving individual EFIMs from the overall 4×4 EFIM by applying equivalent Fisher information again.This separates information about position, orientation, and time offset within the joint information matrix.
  • B. Localization with Antenna Arrays: Corollary 8 provides separate EFIMs for position, orientation, and time offset when agent state and channel parameters are known in far-field scenarios.The result assumes mutually independent channel parameters across anchors and antennas.
  • B. Localization with Antenna Arrays: The framework uses complete received waveforms and implicitly or explicitly incorporates TOA, AOA, TDOA, and joint TOA/AOA or TDOA/AOA metrics.This unifies waveform-based localization information with analyses based on extracted signal metrics.
  • B. Localization with Antenna Arrays: The SPEB is achievable asymptotically at high SNR by MAP and ML estimates, with high SNR attainable through good-correlation sequences or repeated transmissions.Thus, the bound can serve as an attainable localization benchmark under the stated regime.
  • B. Localization with Antenna Arrays: The analytical results extend to three-dimensional localization by replacing the position vector with [x y z]^T and using a corresponding 3×3 EFIM.The paper states that the extension is straightforward within the same EFIM framework.

VII. NUMERICAL RESULTS

Numerical examples examine path overlap, prior channel knowledge, multipath density, ranging outage, array reference points, and time offsets. They show how these factors change SPEB, SOEB, POC, and RAO in the analyzed networks.

  • A. Effect of Path-Overlap: Path overlap increases SPEB because it reduces first-path estimation ability and ranging information, while separations beyond the approximately 4 ns pulse width recover the non-overlapping result.The full-parameter and partial-parameter models agree beyond the pulse width; excluding unknown amplitudes gives a loose bound when paths overlap.
  • B. Improvement from A Priori Channel Knowledge: A priori knowledge of amplitudes and NLOS biases decreases SPEB by increasing ranging information, and NLOS components can benefit localization when NLOS biases are known.With increasingly known amplitudes, the full-parameter result converges to the partial-parameter result.
  • C. Path-Overlap Coefficient: POC decreases monotonically from 1 to 0 as path inter-arrival time increases, while for fixed arrival rate it increases with the number of multipath components.Denser multipath creates more interference with earlier paths and reduces ranging information.
  • D. Outage in Ranging Ability: RAO decreases from 1 to 0 as the POC threshold increases or the path arrival rate decreases.The paper presents RAO as a channel-quality measure for ranging and a guide for transmitted-waveform design.
  • E. SPEB and SOEB for Wideband Antenna Array Systems: A priori orientation knowledge improves localization, while a priori reference-point knowledge improves orientation accuracy in the antenna-array example.The array center has the best localization accuracy, and orientation-aware localization makes SPEB independent of the reference point.
  • F. Effect of Time Offset: Both SPEB and STEB decrease with a priori time-offset knowledge, and the known-offset case yields the same SPEB as a system without a time offset.At φ1 = 0, the time offset has no effect on SPEB in the analyzed topology.
  • VIII. CONCLUSION: The framework provides fundamental localization limits by expressing anchor contributions and a priori position information through equivalent Fisher information and weighted ranging-direction matrices.The conclusions emphasize received-waveform analysis and use as design guidelines and network benchmarks.

APPENDIX A FISHER INFORMATION MATRIX DERIVATION

The appendix derives the Fisher information matrix for wideband channels with multipath, shadowing, fading, and random path arrivals. It models channel statistics and incorporates a priori channel knowledge into the parameter-information structure.

  • Fisher Information Matrix Derivation: The parameter transformation from θ to η is bijective when the agent is localizable, enabling an alternative Fisher information matrix expression through the Jacobian.Localizability is defined by position determination from signal metrics, generally requiring M ≥3 anchors or special cases with M = 2.
  • Fisher Information Matrix Derivation: The channel model uses Poisson multipath arrivals, log-normal shadowing, Nakagami small-scale fading, and an exponential power-delay profile for IEEE 802.15.4a exposition.The analysis remains valid for any wideband channels described by the paper’s general waveform model.
  • Fisher Information Matrix Derivation: The appendix derives the joint PDF of multipath parameters and received-signal strength conditioned on anchor-agent distance, then obtains the multipath-parameter PDF by integrating over RSS.Equation (53) characterizes a priori knowledge of channel parameters.

APPENDIX C PROOFS OF THE RESULTS IN SECTION III

The proofs reduce localization information to the first contiguous cluster of multipath components and derive equivalent Fisher information expressions for channel-known and channel-unknown cases. They also establish properties of the path-overlap coefficient.

  • Proofs of the Results in Section III: The equivalent Fisher information for position is obtained by partitioning the channel FIM and applying EFI to eliminate nuisance multipath parameters.The derivation yields a 2×2 position EFIM.
  • Proofs of the Results in Section III: Only the first contiguous cluster of LOS paths contains localization information in the derived EFIM.The resulting expression depends only on the first cluster’s paths, not on later separated components.
  • Proofs of the Results in Section III: The path-overlap coefficient satisfies 0 ≤ χk ≤ 1 because it is a nonnegative quadratic-form contribution bounded by the anchor EFIM contribution.This establishes the coefficient’s admissible range within the information analysis.
  • Proofs of the Results in Section III: When the first contiguous cluster contains one path, the proof treats the case as a special instance with cluster length ˜Lk = 1.The derivation uses continuity and time limitation of the waveform in this reduction.
  • Proofs of the Results in Section III: With a priori channel knowledge, the proof substitutes prior information into the partitioned FIM and obtains separate LOS and NLOS ranging-information expressions.The resulting coefficients are given for both LOS and NLOS signals.

D. Proof of Corollary 2

The proof shows how a priori channel knowledge changes the RII contributions of NLOS signals and derives an EFIM representation when positional uncertainty is averaged or approximated.

  • A priori channel knowledge increases the RII associated with channel information.
  • Without a priori channel knowledge, the RIIs in (65a) and (65b) reduce to (16) and zero, respectively.
  • NLOS signals are equivalent to LOS signals when the NLOS channel parameter b_k^(1) has infinite a priori knowledge.
  • When positional prior knowledge is available, the EFIM is obtained by taking expectations over the agent position.
  • If condition (22) holds, the EFIM can be approximated by evaluating its functions at the expected position.

APPENDIX D PROOFS OF THE RESULTS IN SECTION IV

The appendix derives EFIM expressions from joint observations and prior knowledge, then establishes uniqueness and reference-point independence for the orientation center and bound.

  • The joint likelihood combines random observations and parameters with deterministic but unknown position and orientation.
  • The FIM separates information from observations and a priori knowledge, with block decompositions across antennas and anchors.
  • The proof establishes a unique orientation center p* by showing only one geometric vector makes q* equal to zero.
  • The equivalent Fisher information for orientation is independent of the reference point, defining the squared orientation error bound.
  • The SPEB for any reference point follows from the 3 × 3 EFIM and the Schur complement.

D. Proof of Proposition 5

The proof specializes the EFIM to antenna-array localization with orientation uncertainty, time offset, and prior knowledge of array-center and orientation parameters.

  • The array-center reference point is the orientation center p*.
  • With a time offset, equivalent Fisher information produces a 3 × 3 EFIM and leads to equations (34) and (35).
  • The orientation-unaware antenna-array model uses a reduced parameter set for the orientation-aware special case.
  • In far-field scenarios, the array center equals the orientation center, simplifying the EFIM with prior knowledge of position and channel parameters.
Loading 1006.0888v1…