Tracking Time-Varying Multipath Channels forActive Sonar Applications

arXiv:2602.15555 · eess.SP · Submitted 2026-02-17 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.

Rosa: We're looking at the paper: "Tracking Time-Varying Multipath Channels forActive Sonar Applications".

Dev: Here's the gist of the paper. Reliable detection and tracking in active sonar require accurate and efficient learning of the acoustic multipath background environment The proposed framework enables simultaneous tracking of the time-varying multipath background and detection of a target.

Rosa: Core Methodology The paper proposes a framework for learning and tracking the multipath background directly in the raw measurement domain instead of transforming measurements into the range–Doppler domain Starting from a wideband Doppler linearization of the impulse response of a time-varying multipath channel, a state-space model with a heteroscedastic measurement equation is derived.

Dev: This model enables channel tracking using an extended Kalman filter (EKF), and unknown model parameters are learned from the marginalized likelihood.

Rosa: The statistical adequacy of the proposed models is assessed via a p-value significance test.

Title and authors: Rosa: Now we’re looking at the title again, "Tracking Time-Varying Multipath Channels for Active Sonar Applications." It sounds very specific to the problem they are trying to solve.

Dev: It's a clear statement that they are focused on tracking channels that change over time in active sonar systems. They aren't just looking at a static environment; the channel itself is dynamic.

Taro: So, what does "Active Sonar Applications" mean in this context? Is this just for underwater acoustics, or does it have broader implications?

Rosa: It’s specifically aimed at active sonar, which means systems that send out a signal and listen for the return. This immediately brings up the issue of multipath—the signal bouncing off surfaces, creating multiple returns.

Dev: The core idea is tackling that time-varying nature of those returns directly in the measurement domain, avoiding that costly transformation step into range-Doppler space first.

Taro: So, if we think about the naive question a listener might have—"how do you handle the sheer complexity of acoustic propagation?"—this paper suggests modeling it as a state-space problem driven by random walks.

Rosa: They use that wideband Doppler linearization to create this model, which is then fed into an extended Kalman filter for tracking. This allows them to update the channel estimate in real time based on incoming data streams.

Dev: The numbers and caveats here are about the complexity of defining those discrete-time linear models, where they define the sampled transmitted waveform and its time derivative component as sn and un.

Taro: I wonder about the latency involved here. Since it's a state-space model, how fast can this EKF loop actually run in a real-time system?

Rosa: That’s where the engineering comes in. They are defining zero-padded Toeplitz matrices S and U of size N by Nl, which depend on the sampling period Delta t and the length of the sampled delay grid, Nl.

Dev: So they're dealing with these matrix sizes, where n equals zero through N minus one, and l equals zero through Nl minus one. That’s a lot of computation if you aren't careful.

Taro: And what about the performance limitations? The paper itself flags that it can favor faster, non-structured adaptive updates resulting in fewer target detections if not handled right.

Rosa: They acknowledge that this approach might limit performance in some cases and could favor faster, non-structured adaptive updates which might result in fewer target detections. But they propose a different detection method instead of the traditional one where you estimate the CIR first and then compute residuals for target detection.

Dev: Instead, they use a detector based on changes in the marginal likelihood of the measurements, which is derived from that tracking process and related concepts two, ten <ref:2602.15555#pg1>.

Taro: So, what does this change for someone who only listens to the show? It means that instead of waiting for a perfect background model before you can even look for a target, you can start looking sooner.

Rosa: It shifts the focus from post-processing estimation to a joint tracking and detection framework where both happen simultaneously. This is what they call "simultaneous tracking of the time-varying multipath background and detection of a target."

Dev: So, we’re moving toward a system that learns the environment and hunts for targets at the same time instead of doing them as two separate steps.

The paper's summary: Rosa: So, let’s summarize what they actually did in "Tracking Time-Varying Multipath Channels for Active Sonar Applications." They developed a framework to learn and track the multipath background directly in the raw measurement domain.

Dev: Instead of doing that transformation to range-Doppler space, which is computationally heavy, they start from a wideband Doppler linearization of the impulse response of the time-varying multipath channel.

Taro: So, if we simplify this for a listener—what’s the core mechanism? It’s using that linearization to create a state-space model with a heteroscedastic measurement equation.

Rosa: Right. This state-space model is then used to track the channel using an extended Kalman filter, or EKF, and they learn the unknown parameters from the marginalized likelihood of the measurements two <ref:2602.15555#pg1,using an extended Kalman filter>.

Dev: They define a discrete time approximation in vector form for a continuous time model, where xb is approximated by Sa plus U Lambda a r.

Taro: That vector representation involves these Nl length vectors collecting multipath amplitudes and log time-scaling parameters. It’s how they capture the complexity of the environment in terms of these basis coefficients.

Rosa: They then decompose the environmental fluctuations captured by those time scaling parameters into c and d, modeling them as zero-mean random variables with specific normal distributions.

Dev: The key idea is separating 'd', which models common Doppler effects, from 'c', which models the unique Doppler effect for each signal path. This decomposition allows them to use different statistical assumptions for tracking those components.

Taro: So, what’s the big picture implication? It means they’re not treating the entire acoustic environment as one monolithic thing but breaking it down into these statistically modeled components.

Rosa: Exactly. By separating c and d, they can use different distributions to model how those effects behave statistically, which is a lot more flexible than assuming everything is just one kind of noise.

Dev: And this decomposition directly enables the subsequent steps for target detection by designing a detector based on the changes in marginal likelihood of the measurements two <ref:2602.15555#pg1>.

Taro: So, it’s a way to build robustness into the detection stage by having a better understanding of how the background is structured statistically.

Rosa: Right. And this entire process is what they call enabling simultaneous tracking of both things at once in their framework for active sonar applications. This is the main contribution they highlight on page two of that work two <ref:2602.15555#pg1>.

The paper's improvements: Dev: Let’s talk about the specific improvements the paper suggests to this approach. They are focusing on how to make it more robust and statistically rigorous.

Rosa: One major improvement is testing four different covariance structures—M0, Mc, Md, and Mcd—to assess model significance via a p-value test. This helps them choose the most statistically adequate covariance structure for their tracking model.

Taro: So they’re not just picking one; they’re doing a proper statistical check to see which one makes sense for the data they've collected in simulations.

Dev: They also employ a sequential likelihood ratio test, or SLRT, to perform target detection by comparing the log marginal likelihoods under background-only and target-present hypotheses. This is defined by the stopping rule in equation 33c.

Taro: So, if we think about the naive question—"how do you improve detection performance beyond just having a better tracking model?"—this paper suggests this SLRT approach gives you a way to jointly learn background and detect the target.

Rosa: It allows for joint background learning and target detection, using that stopping rule in equation 33c. This is how they achieve the goal of simultaneous tracking and detection.

Dev: The simulation results show that models including the common-mode component, specifically Md and Mcd, consistently yield shorter mean time to detection or MTD and higher probability of detection or Pd across various scenarios.

Taro: So for someone who only listens to the show, what does that mean? It means if you use one of those improved models with the common-mode component, you get better metrics in terms of MTD and Pd compared to simpler ones.

Rosa: The implication is that incorporating these specific components actually boosts detection performance when dealing with real-world data. This is because they capture important physical effects like platform or surface wave motion through those components.

Dev: So, the caveat here is that this improved model depends on the statistical adequacy of those chosen covariance structures, and they rely on a p-value test to validate that choice before deployment.

Conclusion: Rosa: To wrap up our discussion on "Tracking Time-Varying Multipath Channels for Active Sonar Applications," the main implication is that this wideband Doppler linearization-based model reliably describes Doppler effects due to platform and surface wave motion.

Dev: This enables efficient background learning in the raw acoustic measurement domain, which is crucial for bolstering the detection of weak target(s) embedded in that multipath background.

Taro: So, what’s the final word from your perspective on what this means for real-world deployment? Does this framework feel mature enough to actually move into a track-before-detect system?

Rosa: The results indicate that the proposed model can reliably describe Doppler effects caused by platform and surface wave motion, which opens up efficient background learning in the raw acoustic measurement domain.

Dev: This capability is significant because it helps boost target detection when dealing with weak targets that are buried deep within the multipath clutter.

Taro: I think it sets a good foundation for future work, focusing on integrating this model into a track-before-detect framework for joint background learning, target detection, and tracking. That’s where the real complexity lies.

Rosa: It does sound like that's the next logical step in developing more sophisticated systems. We’ll keep an eye on how they integrate this model with those other concepts.

Dev: We’ve covered a lot about the technical specifics of "Tracking Time-Varying Multipath Channels for Active Sonar Applications," from the modeling to the specific statistical tests used.

Taro: Yeah, it was a solid piece of work that tackles a very hard problem head-on by proposing this raw measurement domain approach.

Rosa: That’s what we have on this paper for today. We’ve discussed how to model time-varying channels and how to use statistical tests to validate those models.

Division of Automatic Control, Linköping University · Division of Communication Systems, KTH Royal Institute of Technology · FOI-Swedish Defence Research Agency

eess.SP

Submitted: 2026-02-17

Updated: 2026-02-17

Comments: Submitted for possible publication in IEEE FUSION

DOI: 10.23919/FUSION67272.2026.11726709

Code: https://github.com/ASHKoul/Bcg_SLRT

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 78/100

The gist: Reliable detection and tracking in active sonar require accurate and efficient learning of the acoustic multipath background environment The proposed framework enables simultaneous tracking of the

Terminology

Summary

Reliable detection and tracking in active sonar require accurate and efficient learning of the acoustic multipath background environment The proposed framework enables simultaneous tracking of the time-varying multipath background and detection of a target.

Core Methodology

The paper proposes a framework for learning and tracking the multipath background directly in the raw measurement domain instead of transforming measurements into the range–Doppler domain Starting from a wideband Doppler linearization of the impulse response of a time-varying multipath channel, a state-space model with a heteroscedastic measurement equation is derived. This model enables channel tracking using an extended Kalman filter (EKF), and unknown model parameters are learned from the marginalized likelihood. The statistical adequacy of the proposed models is assessed via a p-value significance test.

Signal Modeling

The signal model considers a bistatic setup where received data consists of a time-varying multipath background and a possible target return. A general continuous-time channel model is defined by the summation of arrivals, where the amplitude and delay are approximated as functions of time. The multipath background, denoted as xb(t), is approximated using a first-order wideband Doppler linearization which results in a bi-linear model driven by known waveforms.

State-Space Representation

The time-varying multipath channel is modeled using a low-dimensional parameterization where the multipath amplitudes are defined as a random-walk process. The resulting state-space model under the background-only hypothesis is given by equations describing the evolution of basis coefficients and measurement vectors.

Evaluation and Results

The evaluation methodology involves testing four covariance structures (M0, Mc, Md, Mcd) to assess model significance via a p-value test. The sequential likelihood ratio test (SLRT) is employed for target detection by comparing the log marginal likelihoods under background-only and target-present hypotheses. Simulation results show that models including the common-mode component, specifically Md and Mcd, yield consistently shorter mean time to detection (MTD) and higher probability of detection (Pd) across various scenarios.

Conclusion

The presented results indicate that the proposed wide-band Doppler linearization-based model can reliably describe Doppler effects due to platform and surface wave motion. This enables efficient background learning in the raw acoustic measurement domain to bolster the detection of weak target(s) embedded in the multipath background. Future research will focus on integrating the model into a track-before-detect framework for joint background learning, target detection, and tracking.

Appendix A

Wideband linearization of time-varying multipath signal define F(r, t) ≜ s(e r t) with r = lnβ. Then ∂F(r, t) / ∂r = e r ts˙(e r t) = t dF(r, t) dt. With F(0, t) = s(t), the solution to the PDE is F(r, t) = e r (t d / dt) s(t), s(βt) = e lnβ (t d / dt) s(t). Using the series expansion, e a = 1 + a + a 2/2! + a 3/3! + · · · and taking the first order approximation, (35) can be written as s(βt) ≈ s(t) + lnβts˙(t).

References

[1] N. J. Willis, Bistatic radar. Raleigh, NC: SciTech Publishing, Jun. 2005

[3] B. Kalyan and A. Balasuriya, “Sonar based automatic target detection scheme for underwater environments using CFAR techniques: A comparative study,” in Proc. of the Int. Sym. on Underwater Tech., Taipei, Taiwan, Apr. 2004, pp. 33–37

[4] G. Delyon, “Clutter map detector for active diver detection sonar,” IET Radar, Sonar &; Navigation, vol. 14, no. 1, pp. 177–186, Jan. 2020

[5] A. Baldacci and G. Haralabus, “Signal processing for an active sonar system suitable for advanced sensor technology applications and environmental adaptation schemes,” in 14th European Signal Proc. Conf., Florence, Italy, Sep. 2006, pp. 1–5

[6] M. Stojanovic and J. Preisig, “Underwater acoustic communication channels: Propagation models and statistical characterization,” IEEE Commun. Mag., vol. 47, no. 1, pp. 84–89, Feb. 2009

[7] D. Bossér, G. Hendeby, M. L Nordenvaad, and I Skog, “Broadband passive sonar track-before-detect using raw acoustic data,” IEEE J. Oceanic Eng., Jul. 2025

[8] A Kullberg, I Skog, and M L Nordenvaad, “MAP-based bearings-only tracking at low SNR,” in OCEANS 2025 Brest, Brest, France, Jun. 2025

[9] C Yang, Q Ling, X Sheng, M Mu, and A Jakobsson, “Detecting weak underwater targets using block updating of sparse and structured channel impulse responses,” Remote Sensing, vol. 16, no. 3, p. 476, Jan. 2024

[10] I Skog, M L Nordenvaad, and G Hendeby, “Signals-ofopportunity-based hydrophone array shape and orientation estimation,” IEEE J. Oceanic Eng., vol. 49, no. 3, pp. 679–691, Jul. 2024

[11] J Durbin, Time series analysis by state space methods, 2nd ed., Oxford: Oxford University Press, May 2012

[13] J Li, X Wang, and S Huang, “The waymark v3.0 manual and user’s guide,” Xiamen University, Xiamen, China, Tech. Rep., 2025

[14] K Hasselmann et al., “Measurements of wind-wave growth and swell decay during the joint north sea wave project (JONSWAP),” Ergaenzungsheft zur Deutschen Hydrographischen Zeitschrift, Reihe A, 1973

[12] M B Porter, “The BELLHOP Manual and User’s Guide: PRELIMINARY DRAFT,” Heat, Light, and Sound Research, Inc., La Jolla, CA, USA, Tech. Rep., vol. 260, 2011

[8] A Kullberg et al., “MAP-based bearings-only tracking at low SNR,” in OCEANS 2025 Brest, Brest, France, Jun. 2025 pp. 1–6

[13] J Li et al., “The waymark v3.0 manual and user’s guide,” Xiamen University, Xiamen, China, Tech. Rep., 2025

[14] K Hasselmann et al.

Improvements for AI systems

  1. Improve target detection by explicitly modeling time-varying background fluctuations using an extended Kalman filter (EKF) in raw measurement domain, which enables simultaneous tracking of the time-varying multipath background and detection of a target.

  2. Enhance channel tracking accuracy by employing a state-space model with a heteroscedastic measurement equation to track the multipath channel, allowing for parameter learning from the marginalized likelihood.

  3. Improve statistical rigor in background modeling by assessing model adequacy through a p-value significance test, enabling the selection of the most statistically adequate covariance structure (M0, Mc, Md, or Mcd).

  4. Boost target detection performance by implementing a sequential likelihood ratio test (SLRT) that allows for joint background learning and target detection, using the stopping rule defined in equation 33c.

  5. Develop a low-dimensional parameterization for channel states by defining the multipath amplitudes as a basis expansion, which is necessary when the full vector of unknown parameters becomes "computationally expensive when Nl >> Np."

Related papers