Pith. sign in

REVIEW 2 major objections 5 minor 66 references

Message Passing for Track-Before-Detect

T0 review · 2 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A detector-free tracker beats detect-then-track on weak targets

desk verdict Solid BP-TBD extension with a genuine modeling advance, but the 'detector-free' claim is undercut by MP-seeded track birth; needs a revised framing and TBD baselines before it can be called state-of-the-art. read the letter →

arxiv 2506.18177 v3 pith:K3ABZD6L submitted 2025-06-22 eess.SP

Mingchao Liang , Florian Meyer This is my paper · ORCID
classification eess.SP
keywords track-before-detectmultiobjecttrackingbeliefpropagationfactorgraphsuperpositionalsignalmodelamplitudefluctuationsmeasurementcorrelationspassivesonar
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that track-before-detect (TBD) can work directly on raw, correlated sensor data without a frontend detector, by modeling the data as a superposition of object signals plus noise and doing Bayesian inference with belief propagation. The proposed model keeps correlations across measurement channels, lets signal amplitudes fluctuate over time, and treats noise power as unknown, so the likelihood reflects the physics of radar and sonar data more faithfully than earlier TBD models. Because the resulting factor graph is processed with scalable message passing, the method handles an unknown, time-varying number of objects. If correct, this removes two central obstacles to practical TBD: the need to assume independent measurements and the need to know signal amplitudes.

What carries the argument

The mechanism is a factor graph representation of the joint posterior of the kinematic states, Bernoulli-Gaussian signal-power states, and noise-power states. The measurement likelihood factorizes over dictionaries and snapshots, with each factor a zero-mean complex Gaussian whose covariance contains the sum of object contribution outer products plus noise; this covariance structure carries the correlation information. Belief propagation on this graph replaces intractable joint marginalization with local message updates, and moment matching collapses the Gaussian-mixture messages to single Gaussians, so each update costs O($M^{3}$) and the overall cost is O($M^{3}$ I J N_k), linear in the number of potential objects.

What would settle it

Run the proposed BP-TBD on the same synthetic radar scenario but initialize potential objects on all grid cells (full-grid birth) instead of only on matching-pursuit detections, and compare the GOSPA; if the full-grid version fails to track weak objects or is computationally prohibitive, the claim that the method is detector-free TBD is weakened.

Watch

Extended reading notes

Core claim

The paper's central claim is that a comprehensive superpositional measurement model accounting for correlated measurements, fluctuating complex amplitudes, unknown noise power, multiple snapshots, and multiple signal dictionaries can be paired with belief propagation to produce a scalable track-before-detect method that outperforms conventional detect-then-track approaches. The authors construct the joint posterior as a factor graph whose factors encode object dynamics, birth, signal-power states, and noise-power states, then compute marginal posteriors by iterated sum-product message passing. Intractable messages are approximated by particles or by single Gaussians obtained through moment matching, keeping the per-step cost linear in the number of potential objects, dictionaries, and snapshots. In experiments with synthetic LFMCW radar and real passive acoustic data, the method achieves lower GOSPA error than matching-pursuit and sparse-Bayesian-learning detectors paired with BP-based tracking.

Load-bearing premise

The experimental demonstrations initialize new potential objects only on grid cells where a matching-pursuit detector fires, so the claimed detector-free advantage is not actually tested for object birth in regions where that detector produces no detection.

Editorial extensions

If this is right

  • TBD can operate directly on raw radar or sonar data with correlated channels, eliminating matched-filter preprocessing or decorrelation steps.
  • Unknown noise power is estimated online, making the method adaptable to changing noise levels and colored noise, as shown by the noise amplification factor bη > 1.
  • The multi-dictionary, multi-snapshot formulation makes the same method applicable to multi-sensor fusion, multi-frequency beamforming, and multi-waveform localization.
  • Linear scaling in the number of potential objects, dictionaries, and snapshots allows the approach to scale to large surveillance regions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The experimental reliance on matching-pursuit detections for birth seeding means the practical gain may be limited to refining tracks after a conventional detector has already found a seed; a full-grid birth variant would be the true test of detector-free TBD.
  • The same factor-graph and moment-matching machinery could extend to other superpositional sensing modalities, such as MIMO communications or distributed acoustic arrays, where correlation structure carries information.
  • The Bernoulli-Gaussian amplitude model could be extended to other fluctuation statistics or to amplitude correlations across snapshots if future applications require it.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper proposes a track-before-detect (TBD) method for multi-object tracking based on a superpositional signal model that accounts for correlated sensor measurements, amplitude fluctuations (Swerling 1), and unknown noise power. The joint posterior is represented by a factor graph and solved with belief propagation, using particle representations and single-Gaussian moment matching for intractable messages. The method is evaluated on a synthetic multi-sensor radar scenario and two real passive-acoustic scenarios, comparing GOSPA against two detect-then-track baselines (MP+Tracking and SBL+Tracking). The paper claims that the proposed method 'eliminates the need for an object detector' and outperforms state-of-the-art conventional MOT methods.

Significance. If fully substantiated, the work would be a valuable contribution: it extends the superpositional signal model to multiple snapshots and dictionaries, explicitly handles correlated measurements without a conditional-independence assumption, and reports linear complexity in the number of potential objects. The real-data acoustic experiments are a useful addition to a literature dominated by synthetic studies. However, the central claim of detector-free TBD is not supported by the reported implementation, because track birth is seeded by a matching-pursuit detector rather than by the full-grid birth model. This issue is load-bearing and limits the significance of the experimental demonstration.

major comments (2)
  1. [IV-A (also IV-B and IV-C)] The implementation initializes new potential objects only on grid cells where the matching-pursuit detector fires, as stated: 'instead of initializing new POs for all grid cells, we only initialize new POs on cells with a MP detection.' This means the method still requires a frontend object detector for track initiation, contradicting the abstract's claim of 'eliminating the need for an object detector' and the introduction's 'without requiring a frontend object detector.' The full-grid birth model of Section II-C is described but never evaluated, so the experiments provide no evidence that the method can detect targets missed by the MP detector. The GOSPA comparison is therefore between two DTT-style variants (MP-seeded BP-TBD versus MP/SBL plus tracking) rather than TBD versus DTT. This is the central claim and it is not supported by the reported implementation.
  2. [III-C, Eqs. (12)-(14)] The covariance formulas used for the Gaussian-approximated messages are explicitly imported from previous work ('Based on derivations previously presented in [16], [22], [34]'), but no derivation or adaptation is provided for the new multi-snapshot, multi-dictionary, correlated-measurement model. Since these formulas are the core of the message computation, the paper should either (i) provide a derivation that shows how the correlated measurement model leads to these specific covariance expressions, or (ii) clearly delineate which aspects of the derivation are new. Without this, the novelty and correctness of the proposed inference method are difficult to assess.
minor comments (5)
  1. [II-A] The notation 'xk,n ∈ {1,...,Nk}' for the kinematic state is a typo; the state should be defined in the continuous state space, not as an index.
  2. [IV-A] The expected number of new objects is set to µB = 10−6, which is effectively zero; the paper should clarify whether the full birth model is active at all in the synthetic experiments, given that MP-seeded initialization is used instead.
  3. [IV-A (and IV-B, IV-C)] The mechanism for initializing MP-seeded POs is not specified: it is unclear whether the method uses the birth model's prior fB(x, γ) for those cells, and with what existence probability. This should be clarified to make the implementation reproducible.
  4. [IV-C] The per-PO DFT bin selection based on estimated range rate is a potentially information-discarding preprocessing step; the paper should discuss how this interacts with the claim of operating directly on raw sensor data without a frontend detector.
  5. [IV] The paper does not report the number of initialized POs or computational runtimes, making it difficult to verify the scalability claim in practice.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation; the detector-seeded birth implementation is a demonstration gap, not a circular step.

full rationale

The derivation chain in this paper is a standard sum-product belief-propagation construction over a factor graph: the measurement model (Eqs. 1-3), the state-transition and birth models (Sec. II-B, II-C), and the message update equations (Eqs. 5-11) are derived from explicit statistical assumptions, and no equation reduces to a fitted quantity or to a previously declared output. The moment-matching covariance approximations (Eqs. 12-17) are stated to be 'based on derivations previously presented in [16], [22], [34]'; two of these are the authors' own prior works, but they are published, checkable derivations with stated assumptions that do not themselves include the target result, so this self-citation is not circular in the load-bearing sense prohibited by the rules. The experimental comparison against MP+Tracking and SBL+Tracking uses independent baselines and an external metric (GOSPA). The one substantive weakness is that the experiments do not exercise the detector-free birth model of Section II-C: Sections IV-A, IV-B, and IV-C state, e.g., 'instead of initializing new POs for all grid cells, we only initialize new POs on cells with a MP detection,' so the implemented BP-TBD is a detector-seeded hybrid and the claim of eliminating the object detector is not fully demonstrated. That is a mismatch between the claim and the evaluated implementation, not an equation-level circular reduction, and therefore it does not increase the circularity score.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central algorithm is standard Bayesian filtering with a superpositional Gaussian likelihood; it introduces no new physical entities. The main load-bearing postulates are the Gaussian amplitude and noise model, the Poisson birth model, and the moment-matching approximation used to make messages tractable. Several experimental knobs (bη, cγ, cη, µB) are set by hand, and they influence the reported performance.

free parameters (5)
  • bη (noise amplification factor) = 1 (synthetic radar), 10 (high-SNR acoustic), 5 (low-SNR acoustic)
    Introduced in Eq. (17) to improve robustness against model mismatch or colored noise; the value is chosen per experiment and is not estimated from the model.
  • cγ (signal-power gamma transition scale) = 10^3 (synthetic, low-SNR acoustic), 10^4 (high-SNR acoustic)
    Controls the temporal fluctuation of signal power; set by hand for each scenario.
  • cη (noise-power gamma transition scale) = 100 (synthetic, high-SNR acoustic), 10 (low-SNR acoustic)
    Controls the temporal variation of noise power; set by hand for each scenario.
  • µB (expected number of new objects per time step) = 10^-6 (synthetic radar), 10^-4 (passive acoustic)
    Birth model intensity; chosen to set the rate of new object appearance.
  • ps, Tdec, Tpru (survival probability, declaration and pruning thresholds) = ps=0.9 or 0.95, Tdec=0.5, Tpru=10^-2
    Standard track-management parameters set manually; they affect which objects are declared and retained.
assumptions (5)
  • domain assumption Measurement vectors are a linear superposition of per-object contributions plus additive independent complex Gaussian noise (Eq. (2)).
    The likelihood in Eq. (3) is built on this superposition and independence; if real sensors violate it, the likelihood is misspecified.
  • domain assumption Complex amplitudes are zero-mean circular Gaussian and independent across objects, snapshots, and dictionaries, with variance γ (fast-fluctuation model).
    This makes the marginal likelihood Gaussian with the covariance in Eq. (3); it also rules out deterministic, coherent, or constant-amplitude signals across snapshots.
  • domain assumption Measurement noise is complex Gaussian with covariance η I_M and independent across snapshots and dictionaries.
    Used in the covariance expression; the paper compensates for colored noise by a hand-set bη rather than modeling color explicitly.
  • domain assumption First-order Markov state transitions, gamma transition densities for signal and noise powers, and Poisson birth with at most one new object per grid cell.
    These priors are standard in MOT and determine the prediction and birth messages in Section III-A.
  • ad hoc to paper Moment matching of intractable Gaussian-mixture BP messages by a single Gaussian is sufficiently accurate for tracking.
    The covariance formulas (12)-(14) are imported from [16,22,34] and are not re-derived or validated here; the whole algorithm relies on this approximation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Message Passing for Track-Before-Detect." pith.science (2026). https://pith.science/paper/K3ABZD6L

@misc{pith2026250618177,
  author       = {Pith},
  title        = {Pith review of: Message Passing for Track-Before-Detect},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/K3ABZD6L}},
  note         = {Machine review of arXiv:2506.18177}
}
read the original abstract

Accurately tracking an unknown and time-varying number of objects in complex environments is a significant challenge but a fundamental capability in a variety of applications, including applied ocean sciences, surveillance, autonomous driving, and wireless communications. Conventional Bayesian multiobject tracking (MOT) methods typically employ a detect-then-track (DTT) approach, where a frontend detector preprocesses raw sensor data to extract measurements for MOT. The irreversible nature of this preprocessing step can discard valuable object-related information, particularly impairing the ability to resolve weak or closely spaced objects. The track-before-detect (TBD) paradigm offers an alternative by operating directly on sensor data. However, existing TBD approaches introduce simplifications to facilitate the development of inference methods, such as assuming known signal amplitudes or conditional independence between sensor measurements given object states. These assumptions can lead to suboptimal performance and limit the applicability of the resulting TBD methods in realistic scenarios. This paper introduces a novel TBD method that is based on a comprehensive signal model for sensor data. The new model accounts for sensor data correlations and amplitude fluctuations, enabling the accurate representation of the physics of the data-generating process in TBD. The proposed model is suitable for a wide range of problems in active and passive radar, active and passive sonar, as well as integrated sensing and communication systems. Based on a factor graph representation of the new measurement model, a scalable belief propagation (BP) method is developed to perform efficient Bayesian inference. Experimental results, performed with both synthetic and real data, demonstrate that the proposed method outperforms state-of-the-art conventional MOT methods.

Figures

Figures reproduced from arXiv: 2506.18177 by the authors.

Figure 1
Figure 1. Factor graph representing the joint posterior distr [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Visualization of the synthetic radar tracking scenar [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 4
Figure 4. Estimated and true noise variance η (i) k averaged over 100 simulations runs for (a) the sensor at [0 0]T and (b) the sensor at [50 50]T. An SNR of −3 dB was considered. The qualitative tracking result for −3 dB SNR at a single time step is shown in Fig. 2b. Here, the background is the “Bartlett” spectrum, i.e., a average of the output of matched filtering defined as QI i=1 PJ j=1 k [PITH_FULL_IMAGE:figures/full_fi… view at source ↗
Figures from the paper (6 more)
Figure 3
Figure 3. Figure 3: GOSPA error of different tracking methods for synthe [PITH_FULL_IMAGE:figures/full_fig_p009_3.png]
Figure 5
Figure 5. Figure 5: Bartlett spectrum of high tonal set in dB at time [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: GOSPA error in the high-SNR passive acoustic scenari [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Range estimates of different methods in the high-SNR [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Bartlett spectrum in dB at time k = 1 in the low-SNR passive acoustic scenario with virtual array processing. The red circle is the true position of the source. 5 10 15 20 25 30 35 40 45 time step k 0 100 200 300 400 500 GOSPA [m] MP + Tracking SBL + Tracking BP-TBD (p…
Figure 9
Figure 9. Figure 9: GOSPA error of different tracking methods in the low- [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 64 canonical work pages

  1. [16]

    A BP method for t rack-before- detect,

    M. Liang, T. Kropfreiter, and F. Meyer, “A BP method for t rack-before- detect,” IEEE Signal Process. Lett. , vol. 30, pp. 1137–1141, July 2023

    work page 2023

  2. [22]

    Information e xchange track- before-detect multi-Bernoulli filter for superpositional sensors,

    E. S. Davies and Á. F. García-Fernández, “Information e xchange track- before-detect multi-Bernoulli filter for superpositional sensors,” IEEE Trans. Signal Process. , vol. 72, pp. 607–621, Jan. 2024

    work page 2024

  3. [34]

    Direct multipath -based SLAM,

    M. Liang, E. Leitinger, and F. Meyer, “Direct multipath -based SLAM,” IEEE Trans. Signal Process. , vol. 73, pp. 2336–2352, Mar. 2025

    work page 2025

  4. [1]

    Bar-Shalom, P

    Y . Bar-Shalom, P . K. Willett, and X. Tian, Tracking and Data Fusion: A Handbook of Algorithms . Storrs, CT: Y aakov Bar-Shalom, 2011

    work page 2011

  5. [2]

    Mahler, Statistical Multisource-Multitarget Information Fusion

    R. Mahler, Statistical Multisource-Multitarget Information Fusion . Nor- wood, MA: Artech House, 2007

    work page 2007

  6. [3]

    Multiple hypothesis tracking for multi ple target tracking,

    S. S. Blackman, “Multiple hypothesis tracking for multi ple target tracking,” IEEE Trans. Aerosp. Electron. Syst. , vol. 19, pp. 5–18, Jan. 2004

    work page 2004

  7. [4]

    Marginal multi-Bernoulli filters: RFS d erivation of MHT, JIPDA and association-based MeMBer,

    J. L. Williams, “Marginal multi-Bernoulli filters: RFS d erivation of MHT, JIPDA and association-based MeMBer,” IEEE Trans. Aerosp. Electron. Syst. , vol. 51, pp. 1664–1687, July 2015

    work page 2015

  8. [5]

    A multisensor m ulti- Bernoulli filter,

    A. A. Saucan, M. J. Coates, and M. Rabbat, “A multisensor m ulti- Bernoulli filter,” IEEE Trans. Signal Process. , vol. 65, pp. 5495–5509, Oct. 2017

    work page 2017

Show all 66 references
  1. [6]

    Poisson multi-Bernoulli mixture filter: Direct derivatio n and implemen- tation,

    Á. F. García-Fernández, J. L. Williams, K. Granström, an d L. Svensson, “Poisson multi-Bernoulli mixture filter: Direct derivatio n and implemen- tation,” IEEE Trans. Aerosp. Electron. Syst., vol. 54, pp. 1883–1901, Feb. 2018

    1901

  2. [7]

    Mono-camera 3D multi-object tracking using deep learning detections and PMBM filtering,

    S. Scheidegger, J. Benjaminsson, E. Rosenberg, A. Krish nan, and K. Granström, “Mono-camera 3D multi-object tracking using deep learning detections and PMBM filtering,” in Proc. IEEE IV-18, pp. 433– 440, June 2018

    2018

  3. [8]

    Message passing algorithms for scalable mult itarget tracking,

    F. Meyer, T. Kropfreiter, J. L. Williams, R. Lau, F. Hlawa tsch, P . Braca, and M. Z. Win, “Message passing algorithms for scalable mult itarget tracking,” Proc. IEEE , vol. 106, pp. 221–259, Feb. 2018

    2018

  4. [9]

    Scalable detection and trac king of geometric extended objects,

    F. Meyer and J. L. Williams, “Scalable detection and trac king of geometric extended objects,” IEEE Trans. Signal Process. , vol. 69, pp. 6283–6298, Oct. 2021

    2021

  5. [10]

    A Bayesian approach to trac king multiple targets using sensor arrays and particle filters,

    M. Orton and W. Fitzgerald, “A Bayesian approach to trac king multiple targets using sensor arrays and particle filters,” vol. 50, p p. 216–223, Feb. 2002

    2002

  6. [11]

    Multitarget tracking of distributed targets using histogram-PMHT,

    R. L. Streit, M. L. Graham, and M. J. Walsh, “Multitarget tracking of distributed targets using histogram-PMHT,” Digital Signal Processing , vol. 12, pp. 394–404, Nov. 2002

    2002

  7. [12]

    Multitarget particle filter t rack before detect application,

    Y . Boers and J. Driessen, “Multitarget particle filter t rack before detect application,” IEE Proc. Radar Sonar Navig. , vol. 151, pp. 351–357, Dec. 2004

    2004

  8. [13]

    S. J. Davey and H. X. Gaetjens, Track-Before-Detect Using Expectation Maximisation. Singapore: Springer, 2018

    2018

  9. [14]

    Neural enhanced belief propagat ion for multiobject tracking,

    M. Liang and F. Meyer, “Neural enhanced belief propagat ion for multiobject tracking,” IEEE Trans. Signal Process. , vol. 72, pp. 15–30, Sept. 2023

    2023

  10. [15]

    A multi-target track-before-de tect particle filter using superpositional data in non-Gaussian noise,

    N. Ito and S. Godsill, “A multi-target track-before-de tect particle filter using superpositional data in non-Gaussian noise,” vol. 27 , pp. 1075– 1079, June 2020

    2020

  11. [17]

    Joint dete ction and estimation of multiple objects from image observations,

    B.-N. V o, B.-T. V o, N.-T. Pham, and D. Suter, “Joint dete ction and estimation of multiple objects from image observations,” IEEE Trans. Signal Process. , vol. 58, pp. 5129–5141, Oct. 2010

    2010

  12. [18]

    A tutorial on Bernoulli filters: Theory, implementation and applications,

    B. Ristic, B.-T. V o, B.-N. V o, and A. Farina, “A tutorial on Bernoulli filters: Theory, implementation and applications,” IEEE Trans. Signal Process., vol. 61, pp. 3406–3430, Apr. 2013

    2013

  13. [19]

    Bernoulli track-before-detect filter for maritime radar,

    B. Ristic, L. Rosenberg, D. Y . Kim, X. Wang, and J. Willia ms, “Bernoulli track-before-detect filter for maritime radar,” IET Radar , Sonar Navig. , vol. 14, pp. 356–363, Mar. 2020

    2020

  14. [20]

    A scalabl e track-before- detect method with Poisson/multi-Bernoulli model,

    T. Kropfreiter, J. L. Williams, and F. Meyer, “A scalabl e track-before- detect method with Poisson/multi-Bernoulli model,” in Proc. FUSION- 21, pp. 1–8, Nov. 2021

    2021

  15. [21]

    A Bernou lli track- before-detect filter for interacting targets in maritime ra dar,

    D. Y . Kim, B. Ristic, R. Guan, and L. Rosenberg, “A Bernou lli track- before-detect filter for interacting targets in maritime ra dar,” IEEE Trans. Aerosp. Electron. Syst. , vol. 57, pp. 1981–1991, June 2021

    1981

  16. [23]

    A scal able algorithm for tracking an unknown number of targets using multiple sen sors,

    F. Meyer, P . Braca, P . Willett, and F. Hlawatsch, “A scal able algorithm for tracking an unknown number of targets using multiple sen sors,” IEEE Trans. Signal Process. , vol. 65, pp. 3478–3493, July 2017

    2017

  17. [24]

    A new architecture for ne ural enhanced multiobject tracking,

    S. Wei, M. Liang, and F. Meyer, “A new architecture for ne ural enhanced multiobject tracking,” in Proc. FUSION-24, pp. 1–8, July 2024

    2024

  18. [25]

    A tutorial on Bernoulli filters: Theory, implementation and applications,

    B. Ristic, B.-T. V o, B.-N. V o, and A. Farina, “A tutorial on Bernoulli filters: Theory, implementation and applications,” IEEE Trans. Signal Process., vol. 61, no. 13, pp. 3406–3430, 2013

    2013

  19. [26]

    A particle filter for track-bef ore-detect,

    D. Salmond and H. Birch, “A particle filter for track-bef ore-detect,” in Proc. ACC-01, pp. 3755–3760, June 2001

    2001

  20. [27]

    Long-time coher ent integration for low SNR target via particle filter in track-before-detec t,

    O. Rabaste, C. Riché, and A. Lepoutre, “Long-time coher ent integration for low SNR target via particle filter in track-before-detec t,” in Proc. FUSION-12, pp. 127–134, July 2012

    2012

  21. [28]

    Using phase to i m- prove track-before-detect,

    S. J. Davey, M. G. Rutten, and B. Cheung, “Using phase to i m- prove track-before-detect,” IEEE Trans. Aerosp. Electron. Syst. , vol. 48, pp. 832–849, Jan. 2012

    2012

  22. [29]

    Coherent track-before -detect with micro-Doppler signature estimation in array radars,

    K. Kim, M. Üney, and B. Mulgrew, “Coherent track-before -detect with micro-Doppler signature estimation in array radars,” IET Radar , Sonar Navig., vol. 14, pp. 572–585, Mar. 2020

    2020

  23. [30]

    Coher ent long-time integration and Bayesian detection with Bernoulli track-b efore-detect,

    M. Üney, P . Horridge, B. Mulgrew, and S. Maskell, “Coher ent long-time integration and Bayesian detection with Bernoulli track-b efore-detect,” IEEE Signal Process. Lett. , vol. 30, pp. 239–243, Mar. 2023

    2023

  24. [31]

    Sequen tial detection and estimation of multipath channel parameters using belie f propaga- tion,

    X. Li, E. Leitinger, A. V enus, and F. Tufvesson, “Sequen tial detection and estimation of multipath channel parameters using belie f propaga- tion,” IEEE Trans. Wireless Commun. , vol. 21, pp. 8385–8402, Apr. 2022

    2022

  25. [32]

    Multitarget l ikelihood com- putation for track-before-detect applications with ampli tude fluctuations of type Swerling 0, 1, and 3,

    A. Lepoutre, O. Rabaste, and F. Le Gland, “Multitarget l ikelihood com- putation for track-before-detect applications with ampli tude fluctuations of type Swerling 0, 1, and 3,” IEEE Trans. Aerosp. Electron. Syst. , vol. 52, pp. 1089–1107, June 2016

    2016

  26. [33]

    A belief propagat ion approach for direct multipath-based SLAM,

    M. Liang, E. Leitinger, and F. Meyer, “A belief propagat ion approach for direct multipath-based SLAM,” in Proc. IEEE Asilomar-23, pp. 588– 593, Nov. 2023

    2023

  27. [35]

    V ariational message passing-based multiobject tracking for MIMO-rada rs using raw sensor signals,

    A. M. Westerkam, J. Möderl, E. Leitinger, and T. Pederse n, “V ariational message passing-based multiobject tracking for MIMO-rada rs using raw sensor signals,” arXiv preprint arXiv:2503.15246 , 2025

    2025 arXiv

  28. [36]

    A passi ve, multi-static radar system,

    S. Carson, D. Kilfoyle, M. Potter, and J. V ance, “A passi ve, multi-static radar system,” in Proc. IET RADAR-07 , pp. 1–4, Oct. 2007

    2007

  29. [37]

    A block- sparse Bayesian learning algorithm with dictionary parame ter estimation for multi-sensor data fusion,

    J. Möderl, A. M. Westerkam, A. V enus, and E. Leitinger, “ A block- sparse Bayesian learning algorithm with dictionary parame ter estimation for multi-sensor data fusion,” arXiv preprint arXiv:2503.12913 , 2025

    2025

  30. [38]

    Detectab ility of low- level broad-band signals using adaptive matched-field proc essing with vertical aperture arrays,

    N. Booth, A. Abawi, P . Schey, and W. Hodgkiss, “Detectab ility of low- level broad-band signals using adaptive matched-field proc essing with vertical aperture arrays,” IEEE J. Ocean. Eng. , vol. 25, pp. 296–313, July 2000

    2000

  31. [39]

    Sparse Bayesian learning with multiple d ictionaries,

    S. Nannuru, K. L. Gemba, P . Gerstoft, W. S. Hodgkiss, and C. F. Mecklenbräuker, “Sparse Bayesian learning with multiple d ictionaries,” Signal Processing, vol. 159, pp. 159–170, June 2019

    2019

  32. [40]

    Improved targ et localization in multiwaveform multiband hybrid multistatic radar netwo rks,

    M. Temiz, H. Griffiths, and M. A. Ritchie, “Improved targ et localization in multiwaveform multiband hybrid multistatic radar netwo rks,” IEEE Sensors J. , vol. 22, pp. 20785–20796, Nov. 2022

    2022

  33. [41]

    Maximum likelihood detectio n and estima- tion of Bernoulli - Gaussian processes,

    J. Kormylo and J. Mendel, “Maximum likelihood detectio n and estima- tion of Bernoulli - Gaussian processes,” IEEE Trans. Inf. Theory, vol. 28, pp. 482–488, May 1982

    1982

  34. [42]

    V ariationa l Bayesian in- ference of line spectra,

    M.-A. Badiu, T. L. Hansen, and B. H. Fleury, “V ariationa l Bayesian in- ference of line spectra,” IEEE Trans. Signal Process. , vol. 65, pp. 2247– 2261, Jan. 2017

    2017

  35. [43]

    Superfast line spectral estimation,

    T. L. Hansen, B. H. Fleury, and B. D. Rao, “Superfast line spectral estimation,” IEEE Trans. Signal Process. , vol. 66, pp. 2511–2526, Feb. 2018. 13

    2018

  36. [44]

    Fact or graphs and the sum-product algorithm,

    F. R. Kschischang, B. J. Frey, and H.-A. Loeliger, “Fact or graphs and the sum-product algorithm,” IEEE Trans. Inf. Theory , vol. 47, pp. 498–519, Feb. 2001

    2001

  37. [45]

    An introduction to factor graphs,

    H.-A. Loeliger, “An introduction to factor graphs,” IEEE Signal Process. Mag., vol. 21, pp. 28–41, Jan. 2004

    2004

  38. [46]

    A tutorial on particle filters for online nonlinear/non-Gaussian Baye sian tracking,

    M. S. Arulampalam, S. Maskell, N. Gordon, and T. Clapp, “ A tutorial on particle filters for online nonlinear/non-Gaussian Baye sian tracking,” IEEE Trans. Signal Process. , vol. 50, pp. 174–188, Feb. 2002

    2002

  39. [47]

    Particle belief propagati on,

    A. Ihler and D. McAllester, “Particle belief propagati on,” in Proc. AISTATS-09, vol. 5, pp. 256–263, Apr. 2009

    2009

  40. [48]

    An approach of directly tracking multiple objects,

    M. Liang and F. Meyer, “An approach of directly tracking multiple objects,” in Proc. IEEE Asilomar-24 , pp. 995–999, Oct. 2024

    2024

  41. [49]

    M. I. Skolnik, Introduction to Radar Systems . New Y ork: Mcgraw-Hill, 3rd ed., 2002

    2002

  42. [50]

    Multitask compressive s ensing,

    S. Ji, D. Dunson, and L. Carin, “Multitask compressive s ensing,” IEEE Trans. Signal Process. , vol. 57, pp. 92–106, Sept. 2009

    2009

  43. [51]

    Multimodal spar se Bayesian dictionary learning applied to multimodal data classificat ion,

    I. Fedorov, B. D. Rao, and T. Q. Nguyen, “Multimodal spar se Bayesian dictionary learning applied to multimodal data classificat ion,” in Proc. ICASSP-17, pp. 2237–2241, Mar. 2017

    2017

  44. [52]

    Multi- frequency sparse Bayesian learning for robust matched field processing,

    K. L. Gemba, S. Nannuru, P . Gerstoft, and W. S. Hodgkiss, “Multi- frequency sparse Bayesian learning for robust matched field processing,” J. Acoust. Soc. Am. , vol. 141, pp. 3411–3420, May 2017

    2017

  45. [53]

    Bar-Shalom, T

    Y . Bar-Shalom, T. Kirubarajan, and X.-R. Li, Estimation with Applica- tions to Tracking and Navigation . New Y ork, NY: Wiley, 2002

    2002

  46. [54]

    New prediction for extend ed targets with random matrices,

    K. Granström and U. Orguner, “New prediction for extend ed targets with random matrices,” IEEE Trans. Aerosp. Electron. Syst. , vol. 50, pp. 1577–1589, July 2014

    2014

  47. [55]

    Koller and N

    D. Koller and N. Friedman, Probabilistic Graphical Models: Principles and Techniques. Cambridge, MA: MIT Press, 2009

    2009

  48. [56]

    Richardson and R

    T. Richardson and R. Urbanke, Modern Coding Theory . Cambridge University Press, 2008

    2008

  49. [57]

    Cooperative local ization in wireless networks,

    H. Wymeersch, J. Lien, and M. Z. Win, “Cooperative local ization in wireless networks,” Proc. IEEE , vol. 97, pp. 427–450, Feb. 2009

    2009

  50. [58]

    A belief propagation algorithm for multipath-based S LAM,

    E. Leitinger, F. Meyer, F. Hlawatsch, K. Witrisal, F. Tu fvesson, and M. Z. Win, “A belief propagation algorithm for multipath-based S LAM,” IEEE Trans. Wireless Commun. , vol. 18, pp. 5613–5629, Dec. 2019

    2019

  51. [59]

    Levanon and E

    N. Levanon and E. Mozeson, Radar signals. John Wiley & Sons, 2004

    2004

  52. [60]

    Joint multitarget detection and tracking with mmWave radar,

    J. Zhu, M. Xu, R. Guo, F. Wang, G. Zheng, and F. Qu, “Joint multitarget detection and tracking with mmWave radar,” arXiv preprint arXiv:2412.17211, 2024

    2024 arXiv

  53. [61]

    Matching pursuits with time-fr equency dic- tionaries,

    S. Mallat and Z. Zhang, “Matching pursuits with time-fr equency dic- tionaries,” IEEE Trans. Signal Process. , vol. 41, pp. 3397–3415, Dec. 1993

    1993

  54. [62]

    Gen- eralized optimal sub-pattern assignment metric,

    A. S. Rahmathullah, Á. F. García-Fernández, and L. Sven sson, “Gen- eralized optimal sub-pattern assignment metric,” in Proc. FUSION-17 , pp. 1–8, July 2017

    2017

  55. [63]

    F. B. Jensen, W. A. Kuperman, M. B. Porter, H. Schmidt, an d A. Tolstoy, Computational ocean acoustics , vol. 2011. Springer, 2011

    2011

  56. [64]

    The KRAKEN normal mode program,

    M. B. Porter, “The KRAKEN normal mode program,” SACLANT Un- dersea Research Centre Memorandum (SM-245) and Naval Resea rch Laboratory Memorandum Report No. 6920 , 1991

    1991

  57. [65]

    Mirages in shallow water matched field processing,

    G. D’Spain, J. Murray, W. Hodgkiss, N. Booth, and P . Sche y, “Mirages in shallow water matched field processing,” J. Acoust. Soc. Am. , vol. 105, pp. 3245–3265, June 1999

    1999

  58. [66]

    Range-coherent match ed field processing for low signal-to-noise ratio localization,

    F. Hunter Akins and W. Kuperman, “Range-coherent match ed field processing for low signal-to-noise ratio localization,” J. Acoust. Soc. Am., vol. 150, pp. 270–280, July 2021

    2021

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.