The AoUFML (average of unmatched-filter maximum-likelihood) estimator of Apartsin, Cooper, and Intrator (IEEE MLSP 2010) improves time-of-arrival (ToA) estimation in the semi-coherent, below-threshold SNR regime by averaging two ToA estimates obtained from a pair of phase-shifted unmatched filters, with the phase shift adapted to the estimated SNR. Its validation covered only a Gaussian-modulated sinusoidal pulse in additive white Gaussian noise, a single echo, a single pulse, and an assumed-reliable SNR estimate. The authors' own 2011-2014 follow-up program generalized the combination rule (regression, then machine-learning fusion of a bank of unmatched filters) and settled the multiple-echo and multi-pulse extensions. No work examines the estimator under colored, non-Gaussian, or clutter noise, characterizes the optimal phase selection for other standard waveforms (Ormsby, Klauder, Ricker), or analyzes sensitivity to SNR-estimation errors. The surviving open problem is this three-part characterization.

Background

Time-of-arrival (ToA) estimation — recovering the delay t0t_0 of a known pulse in a received signal x(t)=α s(t−t0)+n(t)x(t) = \alpha\, s(t - t_0) + n(t) — underlies ranging in radar, sonar, and reflective seismology. The classical receiver correlates x(t)x(t) with the transmitted waveform s(t)s(t) (a matched filter, the linear filter that maximizes peak signal-to-mean-noise ratio) and takes the location of the global maximum of the correlator output as the maximum-likelihood estimate of t0t_0; this is the matched-filter maximum-likelihood (MFML) estimator. Under additive white Gaussian noise (AWGN) and sufficiently high signal-to-noise ratio (SNR), the MFML estimator is asymptotically efficient, attaining the Cramér–Rao bound, the fundamental lower bound on the variance of any unbiased estimator.

Below a critical SNR, however, the MFML estimator suffers the threshold effect, studied in radar theory since Woodward (1953): for a narrowband, low-bandwidth pulse, the autocorrelation function of s(t)s(t) has sizeable secondary peaks (side lobes), and when the noise is strong enough a side lobe occasionally overtakes the main peak, producing large outlier estimates and an abrupt growth of the root-mean-square error (RMSE). Receivers are called coherent, semi-coherent, or noncoherent according to whether the operating SNR is above, near, or below this threshold; the approximate threshold SNR RR satisfies the implicit relation R≃2ln⁡(F⋅B⋅R)R \simeq 2\ln(F \cdot B \cdot R), where FF is the detection interval and BB is the signal bandwidth, so the threshold SNR increases logarithmically with the time–bandwidth product. The source paper additionally notes that the height of the autocorrelation side lobes is affected by the pulse bandwidth, so the threshold effect is considerable for low-frequency narrowband pulses.

Apartsin, Cooper, and Intrator (2010) proposed an estimator for the semi-coherent regime. They construct a pair of phase-shifted unmatched (PSU) filters — correlation templates that differ from the transmitted waveform — by advancing and delaying the phase of the pulse by ±φ\pm\varphi. Cross-correlating the received signal with each filter yields two unmatched-filter maximum-likelihood (UFML) estimates t^φ±\hat t^{\pm}_{\varphi}; because the side lobes of the cross-correlation are asymmetric, each UFML estimate is biased, with biases of equal magnitude and opposite sign. In the semi-coherent regime the two noise-induced errors are only partially correlated, so their average, the AoUFML estimator, t^φ=t^φ++t^φ−2,\hat t_{\varphi} = \frac{\hat t^{+}_{\varphi} + \hat t^{-}_{\varphi}}{2}, has lower RMSE than the MFML estimator. The phase shift φ\varphi controls the trade-off (a larger φ\varphi decorrelates the two filtered noises but lowers the output peak), and its optimal value depends on the SNR, the pulse central frequency, and the pulse bandwidth; the authors tabulate it from simulations and select it using an SNR estimate produced by prior techniques such as those reviewed by Pauluzzi and Beaulieu (2000).

Validation in that paper was by simulation, for a Gaussian-modulated sinusoidal pulse in AWGN, a single echo, and a single transmitted pulse; the fusion of repeated transmissions (multi-ping fusion, where a ping is one transmitted acoustic pulse) had previously been demonstrated only for matched-filter estimators, by Yu, Neretti, and Intrator (2006). The authors' subsequent 2011–2014 program replaced the fixed averaging by regression over a bank of phase-shifted unmatched filters and then by machine-learning fusion, and extended the framework to multiple echoes and to adaptive control of the number of transmitted pulses. Three characterization questions were never taken up, and they form the open problem below. For seismic context: the Ormsby wavelet has a trapezoidal band-limited spectrum, the Klauder wavelet is the autocorrelation of a linear frequency-modulated chirp, and the Ricker wavelet is the zero-phase 'Mexican hat' pulse — all standard source wavelets whose autocorrelation side-lobe structures differ markedly from that of a Gaussian-modulated sinusoid. Clutter here means unwanted environment-generated echoes (ground, sea floor, vegetation) that add structured, non-white disturbance to the receiver noise.

Problem Statement

Characterize the SNR-dependent unmatched-filter ToA estimation framework — the phase-shifted-unmatched-filter construction of Apartsin, Cooper, and Intrator (MLSP 2010) and its bank-of-filters successors — under the realistic operating conditions that the original validation excluded. Of the five limitations of the original validation, two (multiple echoes and multi-ping / multi-pulse estimation) were subsequently addressed by the authors' 2014 follow-up work; the following three linked questions remain open and constitute the problem:

(a) Noise realism. Analyze the estimator's performance — RMSE as a function of SNR, and the threshold behavior — for each of the following three disturbance regimes, in place of AWGN: (i) colored noise, (ii) non-Gaussian noise (for example impulsive), and (iii) noise accompanied by clutter. In particular, determine whether the design principle survives: the useful partial decorrelation of the two filtered noises (which are phase-orthogonal at φ=π/2\varphi = \pi/2 only in the white-noise case) and the SNR-dependent selection of the phase shift.

(b) Waveform generality. Characterize the optimal phase shift (or, for the bank-of-filters variants, the bank configuration) as a function of SNR for source waveforms other than the Gaussian-modulated sinusoid — at minimum the Ormsby, Klauder, and Ricker wavelets used in seismology — yielding design curves analogous to the source paper's characterization by central frequency and bandwidth.

(c) SNR-estimate sensitivity. Quantify how errors in the SNR estimate (bias and variance of the kind produced by practical estimators) propagate to the adaptive phase selection and to the resulting ToA accuracy, and determine the margin of SNR-estimation error within which the unmatched-filter approach retains its advantage over the matched-filter ML estimator.

The common scientific objective is a defensible statement of when, and under which design choices, SNR-dependent unmatched-filter fusion improves ToA estimation in the regimes a practitioner actually encounters.

References

  1. Alexander Apartsin, Leon N. Cooper, Nathan Intrator, "SNR-dependent filtering for Time Of Arrival estimation in high noise", 2010 IEEE International Workshop on Machine Learning for Signal Processing (MLSP), Kittilä, Finland, pp. 427–431, 2010. DOI: 10.1109/MLSP.2010.5588848, https://doi.org/10.1109/mlsp.2010.5588848 (open-access copy: https://www.math.tau.ac.il/~nin/papers/ApartsinMLSP2010.pdf)
  2. Sasha Apartsin, Leon N. Cooper, Nathan Intrator, "Biosonar-Inspired Source Localization in Low SNR", Proceedings of the 4th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2011), pp. 399–404, 2011. DOI: 10.5220/0003126803990404, https://doi.org/10.5220/0003126803990404
  3. Alexander Apartsin, Leon N. Cooper, Nathan Intrator, "Semi-coherent time of arrival estimation using regression", Journal of the Acoustical Society of America 132(2), pp. 832–837, 2012. DOI: 10.1121/1.4730885, https://doi.org/10.1121/1.4730885
  4. Alexander Apartsin, Leon N. Cooper, Nathan Intrator, "Time-of-Flight Estimation in the Presence of Outliers Part I—Single Echo Processing", IEEE Transactions on Geoscience and Remote Sensing 52(6), pp. 3382–3392, 2014. DOI: 10.1109/TGRS.2013.2272737, https://doi.org/10.1109/TGRS.2013.2272737
  5. Alexander Apartsin, Leon N. Cooper, Nathan Intrator, "Time-of-Flight Estimation in the Presence of Outliers. Part II—Multiple Echo Processing", IEEE Transactions on Geoscience and Remote Sensing 52(7), pp. 3843–3850, 2014. DOI: 10.1109/TGRS.2013.2276919, https://doi.org/10.1109/TGRS.2013.2276919
  6. Alexander Apartsin, Leon N. Cooper, Nathan Intrator, "Energy-Efficient Time-of-Flight Estimation in the Presence of Outliers: A Machine Learning Approach", IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 7(4), pp. 1306–1313, 2014. DOI: 10.1109/JSTARS.2013.2295324, https://doi.org/10.1109/JSTARS.2013.2295324
  7. Leon N. Cooper, Nathan Intrator, "Time of Flight Estimation in the Presence of Outliers: A Biosonar-Inspired Machine Learning Approach", technical report, DTIC accession ADA597037, 2013. DOI: 10.21236/ADA597037, https://doi.org/10.21236/ada597037
  8. Alexander Apartsin, "Fusion of biased estimators using machine learning for time of flight estimation in the presence of outliers", PhD thesis, Tel Aviv University, 2012. DBLP record phd/il/Apartsin12, https://dblp.org/rec/phd/il/Apartsin12.html
  9. Jie He, Yishuang Geng, Cheng Xu, Zhishuai Han, Shihong Duan, "Height dependent TOA ranging error model for near ground localization applications", 2014 IEEE 25th Annual International Symposium on Personal, Indoor, and Mobile Radio Communication (PIMRC), pp. 2045–2050, 2014. DOI: 10.1109/PIMRC.2014.7136508, https://doi.org/10.1109/PIMRC.2014.7136508
  10. Lian Yu, Nicola Neretti, Nathan Intrator, "Multiple ping sonar accuracy improvement using robust motion estimation and ping fusion", Journal of the Acoustical Society of America 119(4), pp. 2106–2113, 2006. DOI: 10.1121/1.2169913, https://doi.org/10.1121/1.2169913
  11. D. R. Pauluzzi, N. C. Beaulieu, "A comparison of SNR estimation techniques for the AWGN channel", IEEE Transactions on Communications 48(10), pp. 1681–1691, 2000. DOI: 10.1109/26.871393, https://doi.org/10.1109/26.871393
  12. P. M. Woodward, "Probability and Information Theory with Applications to Radar", McGraw-Hill, New York / Pergamon Press, London, 1953. DOI: 10.1016/c2013-0-05390-x, https://doi.org/10.1016/c2013-0-05390-x
  13. Vladimir A. Prokhorov, Lev V. Kuzmin, Andrey A. Krivenko, Pavel A. Vladyka, Elena V. Efremova, "UWB Chaotic Pulse-Based Ranging: Time-of-Flight Approach", Technologies 12(12), article 269, 2024. DOI: 10.3390/technologies12120269, https://doi.org/10.3390/technologies12120269
  14. Ibrahima Mbaye, Saba Al–Rubaye, Christopher Conrad, Gökhan İnalhan, "Integrated Sensing and New Radio Communications for Air Vehicle Positioning", Vehicles 6(3), pp. 1665–1689, 2024. DOI: 10.3390/vehicles6030080, https://doi.org/10.3390/vehicles6030080
  15. Breno C. Pinheiro, Ubirajara F. Moreno, João T. B. de Sousa, Orlando C. Rodriguez, "Kernel-Function-Based Models for Acoustic Localization of Underwater Vehicles", IEEE Journal of Oceanic Engineering 42(3), pp. 603–618, 2017. DOI: 10.1109/JOE.2016.2578218, https://doi.org/10.1109/JOE.2016.2578218
  16. Dorel Aiordăchioaie, "Some Aspects of Time Delay Estimation Based on Cross-Correlation", Circuits, Systems, and Signal Processing 45, pp. 2139–2163, 2025 (published online September 4, 2025; issue dated March 2026). DOI: 10.1007/s00034-025-03323-2, https://doi.org/10.1007/s00034-025-03323-2