PaperlayerView exact revision on arXiv

Source-linked AI summary

Semblance Distance Analysis for Channel Quality Assessment in Distributed Acoustic Sensing

Reinhardt Rading, Markus Hoffmann, Norbert Meyners, Holger Schlarb, Céline Hadziioannou, Oliver Gerberding, Katharina-Sophie Isleif

arXiv:2610.30401v1eess.SPphysics.app-ph

TL;DR

DAS channel quality is difficult to assess amid spatially heterogeneous noise and variable ground coupling, especially when noise is nonstationary. The paper proposes measuring each channel’s semblance distance from a locally coherent reference. Across its validation scenarios, the method identifies consistent low-noise and noise-dominated fiber regions; the study focuses primarily on frequencies up to approximately 48 Hz.

  • Problem

    Spatially heterogeneous noise and variable ground coupling make DAS channel quality difficult to assess, while conventional methods may be unreliable in nonstationary environments.

    Source: §I · p. 2

  • Method

    The semblance-distance framework measures each channel’s deviation from a locally coherent reference to quantify inter-channel dissimilarity.

    Source: §I · pp. 2–3, C. Semblance Distance (SemD) Analysis · p. 5

  • Results

    Across the validated datasets, the algorithm identified the same spatial regions of the fiber as low-noise or noise dominated.

    Source: §VII · pp. 12–13

  • Takeaways & Limitations

    Semblance distance provides a channel-quality assessment for distinguishing well-coupled, high-quality segments from noise- or poor-coupling-dominated sections.

    Source: §VII · pp. 12–13

    The study primarily focused on frequencies from the sub-Hz range up to approximately 48 Hz.

    Source: §VII · p. 13

Abstract

from arXiv · show

Distributed Acoustic Sensing (DAS) enables dense, kilometer-scale measurements for seismic and environmental monitoring, but its performance is limited by spatially heterogeneous noise and fiber to ground coupling. Identifying high-quality sensing channels remains a challenge, particularly in nonstationary and complex noise environments where conventional signal to noise ratio (SNR) or correlation based approaches are insufficient. In this work, we introduce a statistical framework for DAS channel quality assessment based on a semblance distance metric, which quantifies the deviation of individual channels from a locally coherent reference signal. Unlike traditional pairwise similarity measures, the proposed method formulates channel selection as a distance estimation problem in a coherence enhanced signal space, enabling identification of incoherent noise and coupling induced distortions without requiring prior knowledge of installation conditions or noise statistics. The method is validated using multiple independent datasets, including recordings of the 2023 Turkish-Syrian earthquakes, a controlled seismic source experiment, and long-term continuous measurements. Across all scenarios, the approach identifies regions of low noise and strong coupling, with results that correlate with independent SNR and root mean square error (RMSE) metrics. Furthermore, the method detects subtle sub-decibel temporal variations in coupling conditions over seasonal timescales, demonstrating its capability for continuous fiber health monitoring. These results establish semblance distance analysis as a physically interpretable tool for DAS channel selection with direct applications in seismic imaging, structural health monitoring, and noise mitigation in high sensitivity systems such as gravitational wave detectors.

I. INTRODUCTION

DAS turns optical fibers into dense sensing arrays, but spatially varying noise and ground coupling make channel quality uneven. This section motivates a semblance-distance framework that assesses channels by their deviation from a locally coherent reference.

I. INTRODUCTION

  • DAS uses Rayleigh-backscatter phase changes to measure strain along optical fibers, creating dense, kilometer-scale arrays with selectable virtual sensor channels.Selecting channels with strong coupling and high SNR allows noisy or poorly coupled sensors to be excluded.Source: §I · pp. 1–2
  • Spatially heterogeneous noise and variable fiber-to-ground coupling alter channel sensitivity and can distort amplitude and phase, complicating reliable signal capture.Noise is time varying and may lack well-defined spectral signatures; protective coatings also affect strain transfer.Source: §I · p. 2
  • High-SNR channel selection is relevant to seismic wavefield reconstruction, gravitational-wave detector noise cancellation, and detection of weak signals in planetary and forensic seismology.The authors state that selected channels can minimize noise in reconstructed wavefields.Source: §I · pp. 1–2
  • Conventional cross-correlation, coherence, and SNR methods may be unreliable for nonstationary noise and may miss minute variations or nonlinear effects.Conventional SNR estimates often assume noise stationarity, which the authors say does not hold for DAS measurements.Source: §I · p. 2
  • The proposed framework quantifies inter-channel dissimilarity as a semblance distance from a locally coherent reference, rather than relying on pairwise similarity.It is intended to identify high-SNR channels and segments affected by poor coupling or noise contamination.Source: §I · pp. 2–3
  • In the Hamburg network example, earthquake signals are clear from 1,200 to 8,500 m, while sections below 1,200 m and beyond 10,000 m are noise dominated.The caption describes a 12.5 km dark-fiber network recording the February 6, 2023, magnitude 7.8 Turkish-Syrian earthquake.Source: Figure 2 · p. 3

II. THEORY

The theory models each DAS channel as an event signal plus noise, then constructs a locally coherent reference to assess channel-wise deviations. Semblance distance is used to identify noise- and coupling-related departures, with estimation cost increasing for longer records.

II. THEORY

  • The channel noise N_i is divided into spatially coherent N_c and channel-dependent incoherent N_in components.Each channel output combines the coherent event signal X_i with these noise contributions.Source: §II · p. 3
  • The modified semblance operator uses Hilbert-transform phase information to form a locally coherent reference while suppressing unwanted spatially coherent environmental noise.The resulting reference is a noise-reduced estimate of the coherent signal.Source: §II · pp. 3–4
  • Semblance distance measures time-series dissimilarity from the coherent reference; zero denotes identical series, while larger values indicate greater dissimilarity from noise or coupling artifacts.Poorly coupled channels have large SemD values, whereas well-coupled channels remain coherent with the reference and have minimal values.Source: §II · p. 4
  • Euclidean distance is selected because it measures sample-by-sample deviation from the coherent reference, while cosine and correlation distances are insensitive to amplitude scaling.The passage also contrasts Euclidean distance with covariance-intensive Mahalanobis distance and distribution-based Wasserstein distance.Source: §II · p. 4
  • With time alignment, SemD computation scales as O(M^2), so increasing temporal sample count improves estimate reliability but raises computational cost.Short transient-event records remain tractable, while continuous long-term monitoring is the more computationally demanding case.Source: §II · pp. 4–5
  • The earthquake time series have semblance coherence above 0.2, compared with below 0.005 for the next-day no-event reference.The reference was recorded during a period without significant seismic or anthropogenic activity.Source: Figure 3 · p. 5

A. Fiber Distance Selection

Channel selection first confines analysis to a fiber segment suited to the signal's spatial coherence scale. A local coherent reference then supports channel comparison, correction, and validation.

A. Fiber Distance Selection

  • For low-frequency events, select channels within a spatially confined region and avoid distances much larger than the signal's spatial coherence length.The selected sensing length should be considered relative to the wavelength of interest.Source: A. Fiber Distance Selection · p. 5

B. Local Coherent Signal Estimation

  • After selecting channels, suppress incoherent contributions with the modified semblance method to produce a local coherent reference.This reference serves as the benchmark for comparing and correcting individual channels.Source: B. Local Coherent Signal Estimation · p. 5
  • The coherent reference also enables RMSE validation and identification of channels whose responses deviate from it.RMSE is computed between each channel and the coherent estimate.Source: B. Local Coherent Signal Estimation · p. 5

C. Semblance Distance (SemD) Analysis

Semblance distance (SemD) measures each channel’s deviation from a locally noise-reduced coherent reference or from the ensemble behavior of selected channels. The WAVE network’s varied installation conditions and event recordings provide the setting for evaluating channel quality with SNR and RMSE validation.

C. Semblance Distance (SemD) Analysis

  • The direct-reference approach compares each channel’s semblance-transformed series with the locally noise-reduced coherent reference, yielding one distance value per channel.The reference is Aref(t), and the resulting vector has length N.Source: C. Semblance Distance (SemD) Analysis · p. 5
  • Pairwise approaches use time alignment to compensate for shifts or phase misalignments, then average an N × N dissimilarity matrix into one distance value per channel.The last approach applies time alignment directly and is described as slightly outperforming the second approach, particularly for low-frequency, long-term coupling changes.Source: C. Semblance Distance (SemD) Analysis · p. 5
  • Larger SemD indicates reduced coherence associated with instrumental noise or poor mechanical coupling, while values approaching zero indicate agreement with the ensemble response.Source: C. Semblance Distance (SemD) Analysis · p. 5

IV. EXPERIMENTAL SETUP

  • The 15 km WAVE fiber spans cable trays, campus noise exposure, and tunnel sections, including portions buried about 20 m underground.These differing environments yield spatially variable and partly unknown coupling conditions.Source: §IV · pp. 5–6
  • Validation uses earthquake, controlled stone-drop, and continuous recordings, with channel quality checked against SNR and RMSE across sub-Hz to 48 Hz.Earthquake data use 60-minute windows downsampled to 10 Hz; the stone-drop analysis uses a five-minute window at 100 Hz, and the continuous data span February–May 2025.Source: §IV · p. 6

V. RESULTS

In the 2023 earthquake recordings, SemD identifies a quiet, well-coupled underground central fiber section and noisier exposed distal sections. SNR and RMSE profiles broadly corroborate this spatial pattern and distinguish earthquake recordings from the no-event baseline.

A. TWIN TURKISH-SYRIAN EARTHQUAKES 2023

  • From about 1110 to 8000 m, the underground fiber has the lowest earthquake-period SemD, averaging about −54.6 dB; exposed campus sections show higher values from −53 to −36 dB.Localized SemD spikes within the central region align with short exposed fiber sections.Source: A. TWIN TURKISH-SYRIAN EARTHQUAKES 2023 · p. 7, Figure 6 · p. 7
  • The approximately 0.1 dB lower SemD during the Mw 7.8 event than during the Mw 7.5 event may reflect small thermally induced coupling and strain-transfer changes.The events were about nine hours apart; the proposed thermal explanation is qualified as likely.Source: A. TWIN TURKISH-SYRIAN EARTHQUAKES 2023 · p. 7
  • The quiet underground section’s no-event SemD is about −58 dB, roughly 3 dB below the earthquake cases, as the array predominantly records background and instrument noise.Source: A. TWIN TURKISH-SYRIAN EARTHQUAKES 2023 · p. 7
  • Earthquake RMSE is on the order of 10^-8, compared with about 10^-9 during the no-event period, indicating sensitivity to coherent seismic energy.The earthquake RMSE profiles are spatially consistent across the two events, while distal sections show spikes and unstable behavior.Source: Figure 7 · p. 7, A. TWIN TURKISH-SYRIAN EARTHQUAKES 2023 · p. 8
  • Central-section SNR is predominantly 0–6 dB during both earthquakes, with only a 0.1 dB difference between events at 3880 m (6.4 versus 6.3 dB).The profiles are nearly identical, while the same region is slightly negative during the no-event period.Source: A. TWIN TURKISH-SYRIAN EARTHQUAKES 2023 · p. 8
  • Distal sections show unstable, frequently negative earthquake SNR; at 10,050 m, values are −33.1 dB and −36.2 dB for the two events, with noise power exceeding signal power by more than three orders of magnitude.The cited distal ranges are 0–1100 m and 8000–12,500 m.Source: A. TWIN TURKISH-SYRIAN EARTHQUAKES 2023 · p. 8, Figure 8 · p. 8

B. CONTROLLED STONE DROP EXPERIMENT

A controlled stone-drop experiment tests whether SemD distinguishes channel quality under a local impulsive source, complementing the distant-earthquake analysis. Its spatial pattern agrees with SNR validation.

B. CONTROLLED STONE DROP EXPERIMENT

  • During the stone-drop event, SemD is below 10^-5 from about 1100 to 8000 m and approaches 10^-4 in the noisier sections from 200 to 1100 m and beyond 8000 m.The central section is identified as the quietest and best coupled, while the outer sections indicate elevated noise or poor coupling.Source: B. CONTROLLED STONE DROP EXPERIMENT · p. 9
  • SNR is spatially stable at −3 to −8 dB in the central section, but falls below −20 dB and sometimes approaches −40 dB in the outer sections.Source: B. CONTROLLED STONE DROP EXPERIMENT · p. 9
  • Higher SNR coincides with lower SemD, while negative SNR accompanies large, variable SemD, supporting their shared identification of noisy or poorly coupled sections.Source: B. CONTROLLED STONE DROP EXPERIMENT · p. 9

C. NOISY CHANNELS IN THE FIBER NETWORK 2025

After the dark-fiber network was extended to 15 km, SemD identifies quiet and noisy regions under the updated geometry. The results distinguish tunnel-routed and underground sections from an anthropogenically exposed campus segment.

C. NOISY CHANNELS IN THE FIBER NETWORK 2025

  • In the 15 km array, the 6200–13,000 m segment is the lowest-noise, best-coupled region, with strong spatial coherence around 10^-8.Source: Figure 10 · p. 9, C. NOISY CHANNELS IN THE FIBER NETWORK… · p. 9
  • The newly added 1500–3500 m tunnel-routed section has similarly low SemD despite not being deeply buried, while the 3500–5300 m campus segment exposed to anthropogenic activity has elevated distances.The tunnel segment is shielded from direct surface disturbance; the campus segment faces sources including traffic, mechanical infrastructure, and surface vibrations.Source: C. NOISY CHANNELS IN THE FIBER NETWORK… · p. 9, Figure 10 · p. 9
  • Localized SemD spikes in the quiet regions correspond to routing transitions where the fiber is temporarily exposed to the external environment.The paper reports that even minor exposure to ambient disturbances can measurably increase inter-channel dissimilarity.Source: C. NOISY CHANNELS IN THE FIBER NETWORK… · p. 9

D. TEMPORAL EVOLUTION OF COUPLING CONDITIONS

From February to May, semblance distance profiles show a gradual rise alongside persistent spatial patterns, consistent with changing fiber–ground coupling. The metric resolves sub-decibel temporal changes and recurring spatial anomalies.

D. TEMPORAL EVOLUTION OF COUPLING CONDITIONS

  • Average semblance distance rose from -63.6 dB in February to -62.3 dB in May, increasing roughly 0.3–0.5 dB per month.The profiles retain persistent spatial features while the baseline rises.Source: Figure 11 · p. 10, D. TEMPORAL EVOLUTION OF COUPLING… · p. 10
  • Soil temperature rose from 5.13 °C in February to 12.92 °C in May, accompanying the distance trend and potentially modifying soil–cable contact and coupling stiffness.Proposed mechanisms include soil expansion, reduced ground stiffness, moisture changes, and altered contact pressure.Source: D. TEMPORAL EVOLUTION OF COUPLING… · p. 10, Figure 12 · p. 10
  • Anomalies near 3,500–4,500 m, 8,000–9,000 m, and 14,500 m recur at fixed locations, with amplitudes increasing slightly in later months.The authors associate their fixed positions with infrastructure features such as road crossings, splices, conduit transitions, or cable bends.Source: D. TEMPORAL EVOLUTION OF COUPLING… · p. 10
  • The method resolves gradual 0.3–0.4 dB step changes that are difficult to see in raw strain data, supporting its use for tracking slow cable changes.The authors describe this as fiber health and coupling monitoring.Source: D. TEMPORAL EVOLUTION OF COUPLING… · p. 10

E. COMPARISON OF SEMBLANCE DISTANCE METHOD TO CONVENTIONAL METHODS

Across the four-month comparison, semblance distance tracks the independently measured soil-temperature ranking more closely than SNR or RMS and agrees inversely with channel-level SNR. It also resolves localized, repeatable spatial variations that conventional metrics can smooth or obscure.

E. COMPARISON OF SEMBLANCE DISTANCE METHOD TO CONVENTIONAL METHODS

  • Channel-level SNR and SemD are strongly negatively correlated in all four months (r = -0.77 to -0.73, p < 0.0001), indicating agreement on relative channel quality.The correlation decreases slightly from February to May.Source: E. COMPARISON OF SEMBLANCE DISTANCE… · p. 11, Figure 13 · p. 11
  • SemD reproduces the true monthly temperature ranking exactly (Spearman ρ = 1.00), while SNR and RMS show only a weak positive association (ρ = 0.20).The comparison uses the true ranking from independently measured mean soil temperature; the table summarizes monthly rankings.Source: Table 2 · p. 11, E. COMPARISON OF SEMBLANCE DISTANCE… · p. 12
  • SemD reveals periodic microoscillations at consistent spatial locations across all four months, while SNR and RMS smooth or average out these localized variations.The oscillations gradually increase in amplitude from February to May.Source: E. COMPARISON OF SEMBLANCE DISTANCE… · p. 11
  • SemD resolves monthly differences of approximately 0.4 dB, 0.3 dB, and 0.5 dB, whereas SNR and RMS fail to reproduce the expected temperature trend.SNR incorrectly ranks May as the second-coldest month, April as the third-coldest, and March as the warmest; RMS shows a similar inconsistency.Source: E. COMPARISON OF SEMBLANCE DISTANCE… · pp. 11–12, Figure 14 · p. 11
  • Semblance preserves phase-aligned spatial coherence, while correlation can obscure small deviations, entropy lacks spatial-structure discrimination, and gradient energy misses slow baseline drifts.Semblance also provides coherence-based denoising without explicit filtering.Source: E. COMPARISON OF SEMBLANCE DISTANCE… · p. 12

VI. APPLICATIONS

The authors describe potential uses of the algorithm across infrastructure, planetary, high-sensitivity research, and seismic-imaging applications. These uses center on detecting channel-specific anomalies or selecting higher-quality DAS channels.

VI. APPLICATIONS

  • For structural health monitoring, channel-specific deviations may help detect and localize strain irregularities associated with cracks, faults, or structural degradation.The examples include bridges, tunnels, aircraft surfaces, pipelines, and high-rise buildings.Source: §VI · p. 12
  • For planetary exploration, the algorithm could help identify high-quality channels and channel-specific events relevant to interpreting weak seismic signals.The passage discusses proposed lunar and planetary DAS deployments.Source: §VI · p. 12
  • For gravitational-wave observatories, the algorithm could help select informative DAS channels for environmental monitoring and noise subtraction.The stated application is sensor-based noise cancellation.Source: §VI · p. 12
  • For seismic imaging and subsurface characterization, the algorithm could identify high-SNR channels and exclude unreliable measurements.The passage connects channel selection with imaging and inversion quality.Source: §VI · p. 12

VII. CONCLUSION

The study presents semblance-distance analysis for identifying low-noise and poorly coupled DAS channels, validated across earthquake, controlled-excitation, and long-duration background recordings. The authors identify higher-frequency testing and real-time temporal analysis as areas for further development.

VII. CONCLUSION

  • Across independent earthquake, stone-drop, and quiet-period recordings, the algorithm identified the same fiber regions as low-noise or noise-dominated.The datasets came from a campus-scale DAS network and included the 2023 Mw 7.8 and Mw 7.5 Turkish-Syrian earthquakes.Source: §VII · pp. 12–13
  • The current study primarily covers frequencies from the sub-Hz range to approximately 48 Hz; testing at higher frequencies remains future work.The authors also propose real-time temporal analysis and machine-learning integration as further development directions.Source: §VII · p. 13
Loading 2610.30401v1…