PaperlayerView exact revision on arXiv

Source-linked AI summary

Fundamental Limits of Wideband Localization - Part II: Cooperative Networks

Yuan Shen, Henk Wymeersch, Moe Z. Win

arXiv:1006.0890v1cs.ITcs.NI

TL;DR

Wideband cooperative localization must be bounded under harsh propagation conditions and limited anchor deployment. The paper applies Fisher-information analysis to received waveforms, deriving SPEB-based limits, RI decompositions, geometric EFIM interpretations, and scaling laws that unify anchors with agents. These results characterize how cooperation contributes to localization information in cooperative networks.

  • Problem

    Fundamental bounds for cooperative localization were less developed than for non-cooperative localization, despite the need for accurate positioning with limited infrastructure and harsh radio conditions.

  • Method

    The paper uses received waveforms, Fisher information, equivalent Fisher information, and squared position error bounds to analyze wideband cooperative networks.

  • Results

    The framework decomposes localization information into pairwise ranging information, provides a geometric EFIM interpretation, and derives SPEB scaling laws for dense and extended networks.

  • Takeaways & Limitations

    Anchors and agents can be analyzed in a unified localization framework, with cooperation represented as a source of ranging information alongside anchor information and prior position knowledge.

Abstract

from arXiv · show

The availability of positional information is of great importance in many commercial, governmental, and military applications. Localization is commonly accomplished through the use of radio communication between mobile devices (agents) and fixed infrastructure (anchors). However, precise determination of agent positions is a challenging task, especially in harsh environments due to radio blockage or limited anchor deployment. In these situations, cooperation among agents can significantly improve localization accuracy and reduce localization outage probabilities. A general framework of analyzing the fundamental limits of wideband localization has been developed in Part I of the paper. Here, we build on this framework and establish the fundamental limits of wideband cooperative location-aware networks. Our analysis is based on the waveforms received at the nodes, in conjunction with Fisher information inequality. We provide a geometrical interpretation of equivalent Fisher information for cooperative networks. This approach allows us to succinctly derive fundamental performance limits and their scaling behaviors, and to treat anchors and agents in a unified way from the perspective of localization accuracy. Our results yield important insights into how and when cooperation is beneficial.

I. INTRODUCTION

The paper develops fundamental limits for wideband cooperative localization, motivated by harsh environments and limited anchor coverage. It uses waveform-based Fisher-information analysis to derive error bounds, geometric interpretations, and scaling laws for networks in which agents assist one another.

  • Motivation: Wideband transmission provides fine delay resolution and robustness for accurate, reliable range measurements in harsh environments.The paper presents wideband or UWB transmission as particularly well-suited to localization under difficult propagation conditions.
  • Motivation: Cooperation helps localize agents that cannot trilaterate from their neighboring anchors alone.In the illustrated network, each agent reaches only two anchors, but inter-agent cooperation enables both positions to be determined.
  • Contributions: The paper derives fundamental cooperative-localization limits using the squared position error bound (SPEB).This extends the Part I framework from individual-agent localization to wideband cooperative networks.
  • Contributions: Equivalent Fisher information decomposes network localization information into ranging-information (RI) building blocks for node pairs.The formulation accounts for information from anchors, agent cooperation, and prior position knowledge.
  • Contributions: Anchors and agents are treated uniformly by modeling anchors as agents with infinite a priori position knowledge.The framework also gives a geometric EFIM interpretation and derives SPEB scaling laws for dense and extended networks.
  • System model: The signal model considers synchronous networks with fixed topology, known anchor positions, unknown agent positions, and received waveforms from neighboring nodes.The main analysis focuses on two-dimensional positions and wideband multipath observations with additive white Gaussian noise.

C. Joint PDF of Observations and Parameters

The paper factors the joint distribution of received observations and localization parameters under conditional independence assumptions. This factorization separates anchor information, cooperative-agent information, and prior position information in the Fisher information matrix.

  • Joint distribution: The joint distribution is written as a likelihood function multiplied by an a priori parameter distribution.The likelihood models observations conditioned on parameters, while the prior captures parameter knowledge before observation.
  • Observation model: Conditionally independent received waveforms allow the likelihood to be expressed through per-link factors.The factorization applies because received waveforms are independent conditioned on the parameter vector.
  • Parameter model: Conditionally independent multipath parameters allow the prior to factor into channel-parameter terms and the joint prior over agent positions.The channel terms are conditioned on the agents’ positions.
  • Fisher information: The resulting Fisher information separates contributions from anchors, agent cooperation, and a priori knowledge of agent positions.These appear as distinct summation groups or terms in the Fisher information matrix.
  • Assumption: The analysis assumes independent multipath parameters conditioned on node positions, while correlated channels can also be handled but yield a higher SPEB.The independent-channel model is used as the main analytical setting.

III. EVALUATION OF FIM

The paper reduces cooperative localization analysis to equivalent Fisher information and ranging information, exposing how received waveforms, channel knowledge, and cooperation shape position accuracy.

  • Equivalent Fisher Information Matrix: The EFIM preserves the information needed to bound the mean-square error of the parameters of interest through the information inequality.It is obtained through a Schur complement of the full Fisher information matrix.
  • Ranging Information: Each received waveform contributes one-dimensional ranging information along its propagation direction, with intensity determined by signal, bandwidth, multipath, and channel knowledge.The ranging information has the form λ Jr(φ), where λ is the ranging information intensity.
  • Cooperative EFIM: Without a priori position knowledge, the EFIM separates anchor information from highly inter-related cooperative information represented by pairwise ranging-information matrices.Anchor contributions are block diagonal, whereas cooperation depends on agents’ position errors.
  • Waveform and Channel Effects: When channel parameters are unknown a priori, NLOS signals do not contribute to localization accuracy under the stated model.For LOS signals, the ranging information intensity depends on effective bandwidth, first-path SNR, and path overlap.
  • Waveform and Channel Effects: For LOS signals, path overlap degrades localization accuracy unless its coefficient is zero, and later multipath components need not be processed for the ranging information.The first contiguous cluster determines the ranging information intensity.

C. Fisher Information Analysis with A Priori Position Knowledge

With a priori position knowledge, the paper derives an EFIM that augments anchor and cooperative localization information with information from the position prior. It also shows that anchors are the limiting case of agents with infinitely precise prior positions.

  • A Priori Position Knowledge: The EFIM with a priori position knowledge retains anchor and cooperation structure while averaging ranging information over possible agent positions.The prior contributes an additive positive semi-definite component ΞP, which improves localization.
  • Unified Node Treatment: Agents and anchors can be treated uniformly because an anchor is equivalent to an agent with infinite a priori position knowledge.This unified view supports network analysis and allows agents to combine information from anchors and cooperating agents in the same framework.
  • Structured EFIM: Under the stated position-distribution and waveform-independence conditions, the EFIM has a structured 2Na × 2Na form expressible using ranging information.The special case assumes each prior position distribution is concentrated in a small area relative to distances to other nodes.

D. Discussions

The discussion extends the framework to non-cooperative localization, temporal cooperation, recursive network construction, and three-dimensional scenarios. It also characterizes how cooperation affects scaling in growing networks.

  • Non-Cooperative Localization: Removing agent cooperation reduces the EFIM to anchor-derived information for each agent, with independent position priors contributing through ΞP.The non-cooperative case discards the cooperation matrices from the cooperative EFIM.
  • Temporal Cooperation: A moving agent can be modeled as multiple cooperating agents over time, with navigation measurements supplying distances between successive positions.The framework can combine cooperation across both space and time.
  • Network Construction: The EFIM structure supports recursive updates when agents enter or leave a cooperative network.Adding an agent incorporates its anchor information and cooperation with existing agents; removing one eliminates its rows, columns, and corresponding cooperation terms.
  • Three-Dimensional Extension: The framework extends to three-dimensional localization, where each node position includes x_k, y_k, and z_k coordinates.The squared position error bound is correspondingly defined for the three-dimensional position vector.

4) Extension to 3D Localization:

The EFIM is represented geometrically as an information ellipse characterized by eigenvalues and orientation. This representation explains how additional ranging information changes localization accuracy according to anchor direction.

  • Information-ellipse representation: The EFIM is characterized by eigenvalues µ and η, with µ ≥ η, and a rotation angle ϑ.The corresponding eigenvectors define the ellipse axes in the rotated coordinate system.
  • Information-ellipse representation: The squared position error bound is independent of the coordinate system, and rotating to the eigenbasis decouples localization information along orthogonal axes.In this coordinate system, the EFIM is diagonal.
  • Non-cooperative update: A ranging-information contribution is a degenerate information ellipse, adding information along its associated direction.The resulting ellipse is characterized by updated parameters ˜µ, ˜η, and ˜ϑ.
  • Non-cooperative update: For fixed ranging information intensity, the minimum SPEB occurs when the new anchor lies along the original ellipse’s minor axis.This choice maximizes the area reduction of the updated information ellipse.
  • Non-cooperative update: The maximum SPEB occurs when the new anchor lies along the major axis, while increasing ranging information indefinitely approaches the corresponding eigenvalue-based limits.The major-axis direction minimizes the area reduction from the added information.

B. Interpretation for Cooperative Localization

Cooperative localization combines anchor information with inter-agent ranging information, whose effectiveness is limited by the assisting agent’s directional uncertainty. Its scaling laws show that cooperation contributes alongside anchors, with benefits depending on network geometry and propagation loss.

  • Cooperative EFIM: The individual-agent EFIM can be bounded by weighted sums of ranging information from neighboring anchors and agents, simplifying network analysis and design.Anchor information is not inter-related among agents, whereas cooperative information is highly inter-related because it depends on position error.
  • Cooperative EFIM: Cooperation provides an effective RII ˜ν1,2 = ξ1,2ν1,2, where 0 ≤ ξ1,2 ≤ 1, rather than the full inter-agent RII ν1,2.The reduction results from uncertainty in the assisting agent’s position.
  • Cooperative EFIM: Greater directional uncertainty of agent 2 reduces cooperation’s effectiveness, while the effective RII increases monotonically with ν1,2 for fixed directional position error.Its maximum equals the inverse directional position error bound of agent 2 based on anchors.
  • Cooperative EFIM: When the assisting agent is certain along the inter-agent direction, it can provide the same ranging information as an anchor; anchors are therefore special agents with zero SPEB.This equivalence is stated for the limiting case µ2 = +∞ and φ1,2 = ϑ2.
  • Scaling laws: In dense networks, SPEB scales as Θ(1/Nb) without cooperation and Θ(1/(Nb + Na)) with cooperation.The cooperative gain is Θ(1 + Na/Nb), so it is most pronounced when anchors are limited.
  • Scaling laws: In extended networks, SPEB scales as Θ(1/log Nb) for b = 1 and Θ(1) for b > 1 without cooperation, with cooperative scaling based on total nodes.For b = 1, the cooperative gain is Θ(log(Nb + Na)/log Nb); for b > 1, cooperation yields a constant gain.
  • Scaling laws: For amplitude loss exponent b > 1, adding nodes eventually gives diminishing benefit because their ranging information decays rapidly, although cooperation reaches a smaller limiting SPEB.This behavior is reported for extended networks.

V. NUMERICAL RESULTS

Numerical examples show that cooperation substantially improves localization as agent count grows, while anchor geometry creates a nonmonotonic SPEB trade-off between directional diversity and path loss.

  • Benefit of Cooperation: The effective RII increases from 0 toward 1/Δ2(φ1,2) as the cooperating agent’s RII increases.For fixed RII, the maximum occurs when the second agent lies along the relevant direction, while the minimum occurs at a perpendicular angle.
  • Benefit of Cooperation: As the number of agents increases, the average SPEB decreases significantly, roughly proportional to the number of agents.The simulation uses uniformly distributed agents in a 20 m by 20 m area with free-space path loss and D = 10.
  • Benefit of Cooperation: Anchor set II yields lower SPEB than set I because its anchors cover the area better.Set II places anchors at distance D from the center, whereas set I places them at distance √2D.
  • Benefit of Cooperation: The upper and lower SPEB approximations coincide for two cooperating agents and diverge with more agents, while converging to a positive ratio.Thus, both approximations decrease at the same asymptotic rate despite becoming looser as cooperation expands.
  • Anchor Deployment: Mean SPEB first decreases and then increases as anchors move farther from the area center.Centralized anchors provide nearly collinear range information, whereas distant anchors lose RII through path loss; intelligent deployment can outperform random placement.

APPENDIX B PROOF OF THEOREM 1

The proof derives the cooperative EFIM by separating prior-position and channel effects into block structures, then expressing localization information through range-related information matrices.

  • EFIM Structure: The derivation first establishes the EFIM structure and then derives the detailed range-information terms.The EFIM is assembled from block matrices representing agent priors, channel information, and cooperation.
  • EFIM Structure: When agent positions lack a priori knowledge, the likelihood separates received waveforms from channel-parameter distributions.This separation enables construction of the corresponding EFIM blocks.
  • Range Information: The cooperative EFIM depends on range-information matrices indexed by agent-anchor and agent-agent links.Each link contribution is built from Fisher-information blocks and channel-parameter elimination terms.
  • Range Information: For each link, the distance gradient is represented by the unit direction vector q_kj, while λ_kj captures the scalar range-information contribution.The resulting range information has directional and scalar components tied to inter-node distance and bearing.

APPENDIX C PROOF OF THEOREM 2

The proof characterizes how channel knowledge and prior position knowledge enter the EFIM, establishes the role of contiguous LOS path clusters, and shows that a perfectly known agent functions as an anchor.

  • Channel Information: With unknown a priori channel parameters, NLOS links have zero RII, whereas LOS links retain RII through the channel-information structure.The derivation uses the path-overlap coefficient and limiting Fisher information for known channel components.
  • Channel Information: Only the first contiguous cluster of paths in an LOS received waveform contains localization information.The resulting expression depends only on the first contiguous-cluster paths.
  • Prior Position Knowledge: With prior position knowledge, the EFIM gains an additional Ξ_P term and generally requires expectation over the random position parameter.Under stated conditions, these expectations can instead be evaluated at the mean position.
  • Anchor-Agent Equivalence: An agent with infinite a priori position knowledge is effectively an anchor, and its link information becomes fully utilizable after EFI elimination.Eliminating that agent reduces the EFIM dimension while preserving the cooperative structure.
  • SPEB Approximations: The lower approximation ignores cooperation among the remaining agents, whereas the upper approximation treats more agents as equivalent to anchors.These constructions respectively omit cooperative information or double selected diagonal terms while removing off-diagonal terms.

APPENDIX G PROOF OF THE SCALING LAWS

The scaling-law proof uses order statistics for random directions and link-information values to establish high-probability bounds on cooperative localization error.

  • Random Angular Geometry: For uniformly random directions, order-statistics arguments ensure sufficiently populated angular sectors with probability approaching one.The proof partitions directions into angular intervals and bounds the probability that required sectors are underpopulated.
  • Random Angular Geometry: The angular-event failure probability decreases exponentially with the number of nodes.The proof explicitly concludes that ε in the angular bound decreases exponentially with N.

A. Proof of Theorem 5

The proof establishes SPEB scaling for non-cooperative and cooperative dense networks using EFIM bounds and RII lower and upper bounds. Non-cooperation yields Θ(1/Nb), while cooperation yields Θ(1/(Nb + Na)) with probability approaching one.

  • Outage behavior: The outage probability of the dense-network scaling law decreases exponentially with Nb.Both component outage probabilities decrease exponentially with the number of anchors.
  • Cooperative case: The cooperative upper approximation treats all other agents as anchors, yielding the lower bound Ω(1/(Nb + Na)).
  • RII bounds: For dense networks, anchor RII values are bounded away from zero with probability approaching one under bounded path-loss and fading conditions.A common lower bound ˜ν is constructed for anchor RII and effective cooperative RII.
  • Cooperative case: Cooperative SPEB scales as Θ(1/(Nb + Na)) as both Nb and Na grow.Upper and lower EFIM approximations provide the matching bounds.

B. Proof of Theorem 6

The proof analyzes extended networks under distance-dependent path loss and derives scaling laws for non-cooperative and cooperative localization. The resulting behavior depends on the amplitude loss exponent, with logarithmic scaling at b = 1 and bounded SPEB for b > 1.

  • Model and outage: Anchor RII is bounded between positive constants under the extended-network path-loss and fading model, while large-node outage is dominated by spatial topology.The RII bounds use SNR(r) proportional to 1/r^(2b).
  • Loss-exponent regimes: For b > 1, non-cooperative SPEB scales as Θ(1) with probability approaching one as Nb increases.The corresponding EFIM trace remains Θ(1) at fixed anchor and agent densities.
  • Loss-exponent regimes: For 0 < b < 1, non-cooperative and cooperative SPEB scale as Θ(1/Nb^(1-b)) and Θ(1/(Nb + Na)^(1-b)), respectively.These scaling laws are stated for the two network configurations under the sublinear loss-exponent regime.
  • Non-cooperative case: Non-cooperative SPEB scales as Θ(1/log Nb) when b = 1.The proof obtains matching lower and upper bounds with probability approaching one.
  • Cooperative case: Cooperative SPEB scales as Θ(1/log(Nb + Na)) when b = 1.The lower and upper bounds follow from cooperative EFIM approximations.
Loading 1006.0890v1…