Source-linked AI summary
Set-membership localization of intermittent RF sources using a fleet of collaborating UAVs
Jacques Bois, Raul de Lacerda, Michel Kieffer
TL;DR
Intermittent emissions make it difficult to localize RF sources and certify that an observed area is empty, because nondetection can reflect either absence or silence. The paper develops distributed set-membership estimation for collaborating UAVs, producing source-location sets and source-free regions under stated assumptions. Simulations report slightly better localization accuracy and convergence speed than a Bayesian baseline.
Problem
With intermittent RF emissions, nondetection cannot reliably distinguish an empty region from a source in a silent period, limiting deterministic source-absence certification.
Source: §1 · p. 1Method
Results
Simulations show that SMA slightly outperforms BBA in localization accuracy and convergence speed.
Source: §5 · p. 5Takeaways & Limitations
The approach evaluates source-location sets and source-free areas when assumptions on emission periods and detection probability hold.
Source: §5 · p. 5The formulation assumes static, frequency-separable sources with periodic emissions and periods bounded by a maximum period.
Source: §2 · p. 1
Abstract
from arXiv · showhide
This paper proposes a set-membership approach (SMA) to localize radio frequency (RF) sources observed by a collaborating fleet of Unmanned Aerial Vehicles (UAVs). Considering frequency-separable RF transmitters with intermittent and periodic emission patterns, %and unknown but bounded periods the SMA evaluates set estimates of the source locations and a set free of sources. Simulation results show that SMA outperforms a Bayesian baseline approach in terms of localization accuracy and convergence speed.
1. INTRODUCTION
Intermittent RF emissions make detection and localization difficult, especially when silence cannot distinguish an empty region from an emitter between transmissions. The paper introduces distributed set-membership localization and source-absence certification for collaborating UAVs.
1. INTRODUCTION
- Unknown waveform fluctuations degrade matched-filtering and time-frequency detection, while intermittent emissions complicate signal association and source separation.Standard probabilistic formulations use empirical emission probabilities or fixed missed-detection rates.Source: §1 · p. 1
- Probabilistic approaches struggle to certify source absence because nondetection may mean either an empty zone or an emitter in a silent period.Existing state-machine and Bernoulli-based methods provide confidence probabilities but not deterministic clearance guarantees.Source: §1 · p. 1
- Collaborating mobile UAVs can mitigate spatial occlusion and improve coverage, but their coordination requires information fusion and path planning.The introduction lists distributed filtering, flocking, optimization, and reinforcement learning as coordination approaches.Source: §1 · p. 1
- The paper introduces distributed SMA for static, frequency-separable RF sources with intermittent periodic emissions, estimating locations and certifying source-free areas.The paper compares its approach with a Bayesian baseline.Source: §1 · p. 1
2. HYPOTHESES AND PROBLEM FORMULATION
The formulation considers static, frequency-separable sources with periodic RF emissions of bounded but unknown parameters, observed by a communicating UAV fleet. It defines the information and set estimates used to balance localization and source-absence uncertainty.
2. HYPOTHESES AND PROBLEM FORMULATION
- The model includes an unknown number of static sources in a bounded 2-D search area, emitting separable periodic signals with unknown duration, period, and offset.Each period lies between the emission duration and a maximum period.Source: §2 · p. 1
- Each UAV explores the search area in discrete time and is assumed to know its own location and orientation.UAV state is expressed in the shared reference frame.Source: §2 · p. 1
- A source emitting within a UAV’s field of detection is assumed always detected when its emission overlaps the sensing interval.The sensing interval spans multiple time slots as defined by the model.Source: §2 · p. 1
- The fleet assumes perfectly synchronized clocks, negligible communication latency and errors, and a limited communication range.UAVs exchange information with one-hop neighbors.Source: §2 · p. 1
2.1. Set estimates and problem formulation
- Each UAV’s available information includes its state, known-source list, measurements, and corresponding information received from neighbors.This information is denoted Ii(k).Source: §2.1 · p. 2
- Using that information, a UAV maintains set estimates for detected source locations and a set of locations it can declare source-free.The fleet’s objective trades off source-localization uncertainty against uncertainty about source absence.Source: §2.1 · p. 2
3. ITERATIVE SET-MEMBERSHIP ESTIMATION
The iterative SMA updates source-location sets from detections and uses periods of nondetection to determine when regions can be certified source-free. UAVs exchange and fuse these estimates and observation histories.
3. ITERATIVE SET-MEMBERSHIP ESTIMATION
- At each time step, UAVs update their known-source lists and set estimates using newly detected sources and observations of previously detected sources.A newly detected source receives a location set; later detections intersect it with the current field of detection and regions not already certified source-free.Source: §3 · p. 2
3.1. Proving the absence of a source
- Continuous UAV motion can leave locations observed only in separated intervals, so accumulated observation time alone may not rule out an intermittent source.A static UAV could wait longer than the maximum emission period to certify absence, but that delays source detection.Source: §3.1 · p. 2
3.1.1. Time intervals without detection and source detectability
- A location is source-free if every possible emission period and initial offset would have produced a detectable emission during at least one recorded no-detection interval.This follows under the model’s assumption that emissions in the field of detection are always detected.Source: §3.1.1 · p. 2
3.1.2. Proving the absence of a source at a point of X0
- The method checks detectability across a discrete set of candidate periods, using interval modulo operations to test whether all possible offsets are covered.The continuous-period criterion is extended in Appendix A.1.Source: §3.1.2 · p. 2
3.1.3. Proving the absence of sources in subsets of X0
- For region-level certification, the search area is partitioned into boxes and each box accumulates observation intervals when it lies wholly within the field of detection and outside known source sets.Cell size trades evaluation complexity against accuracy; overly large cells may leave the interval list empty.Source: §3.1.3 · p. 3
3.2. Communications and information fusion
- UAVs broadcast known sources, location and source-free sets, and observation intervals; neighbors combine source-free sets and intersect source-location estimates.The illustrated UAV trajectory is lawn-mower shaped.Source: §3.2 · p. 3, Figure 1 · p. 3
4. SIMULATION AND RESULTS
Simulations compare SMA with a Bayesian baseline under varying UAV speeds and fleet sizes, measuring detection, localization error, and uncertainty about source absence. SMA has lower localization error and performs better than the baseline in the reported comparisons, while speed affects detection and absence certification differently.
4. SIMULATION AND RESULTS
- The simulation uses ten uniformly placed sources in a 100 m square, with emission periods uniformly drawn from 2 to 16 seconds.UAVs start at random locations and follow constant-speed lawn-mower trajectories.Source: §4 · p. 3
4.1. Bayesian baseline algorithm (BBA)
- BBA maintains a posterior location probability map per source and declares cells clear when the mean source-presence probability falls below a threshold.The metrics include detected-source proportion, average localization error, and uncertainty about source absence, averaged across UAVs and ten runs.Source: §4.1 · p. 4, §4.2 · p. 4
4.3. Results
- Faster UAV movement and larger fleets reduce the time to detect all sources, with detection trends shared by SMA and BBA under the same trajectories.The speed comparison uses two UAVs; the fleet-size comparison fixes speed at 1.25 m/s.Source: Figure 2 · p. 4, §4.3 · p. 4
- SMA provides lower average localization error than BBA, and it performs consistently better as fleet size varies.The comparisons cover two-UAV speed variations and fleet sizes at 1.25 m/s.Source: Figure 3 · p. 4, §4.3 · p. 4, Figure 4 · p. 5
- The best speed for reducing the potentially source-containing set is 1.25 m/s for SMA and 2.5 m/s for BBA.The optimal speed balances source-detection delay, detection accuracy, and the ability to show source absence.Source: Figure 3 · p. 4, §4.3 · p. 4
- SMA performs better than BBA for the presented scenario across tested absence thresholds from 10^-1p0 to 10^-8p0.BBA’s source-absence uncertainty depends on an arbitrary threshold and does not have the same meaning as deterministic uncertainty.Source: Figure 5 · p. 5, §4.3 · p. 5
5. CONCLUSION
The SMA estimates locations of periodically emitting RF sources and identifies source-free areas when its period and detection-probability hypotheses hold. Simulations show slightly better localization accuracy and convergence speed than the BBA, while moving sources with sporadic, non-periodic emissions remain an open challenge.
5. CONCLUSION
- The SMA estimates locations of periodically emitting RF sources and identifies subsets of the search area clear of sources when its period and detection-probability hypotheses hold.Source: §5 · p. 5
- The SMA slightly outperforms the BBA in localization accuracy and convergence speed in simulations.The authors also identify deterministic localization and certification of source absence as direct applications in search and rescue, surveillance, electromagnetic intelligence, and connected object monitoring.Source: §5 · p. 5
- Extension to moving sources with sporadic, non-periodic emissions remains an open challenge.Source: §5 · p. 5
A.1. Extending the criterion of target absence from discrete periods to period intervals
This appendix extends the target-absence criterion from a discrete period to an interval of possible periods. It considers small variations around a period satisfying the criterion and derives conditions for the criterion to remain satisfied across cases determined by the observed intervals.
A.1. Extending the criterion of target absence from discrete periods to period intervals
- The extension seeks an interval of periods containing τ for which the target-absence criterion remains satisfied, by testing small variations of τ across cases based on the number of observation intervals.Source: §A.1 · p. 5
- For one observation interval, the criterion holds for every τ′ from zero to that interval’s width.For two intervals, if either interval’s width is at least τ, the criterion holds up to the larger width.Source: §A.1.1 · p. 5, §A.1.2 · p. 5
- For the general case, the method selects a smallest subset of observation intervals satisfying the criterion and partitions it by whether each interval modulo τ yields one or two intervals.Figure 7 illustrates a case with at least three observation intervals; its colors distinguish intervals inside and outside the minimal subset and the two partition groups.Source: §A.1.3 · p. 7, Figure 7 · p. 7
A.2. Fusion of probability maps between UAVs
This section describes how UAVs update exchanged probability maps under an independence assumption for their observations and trajectories. The map represents each cell’s probability of containing a target and its complementary probability of being empty.
A.2. Fusion of probability maps between UAVs
- The map assigns each cell a target probability and defines its empty-cell probability as the complement; fusion assumes independent observations and trajectories across the two UAVs.Source: §A.2 · p. 8
- Under that independence assumption, the joint likelihood of the two UAVs’ observations given a cell containing a target factors into the product of their individual likelihoods.Source: §A.2 · p. 8
- The fusion equations express the updated cell probability as odds, the ratio of target probability to its complement.Source: §A.2 · p. 8