PaperlayerView exact revision on arXiv

Source-linked AI summary

Underwater Acoustic Channel Library

Zhengnan Li, Mandar Chitre, Diego A. Cuji, James Preisig, Andrew C. Singer, Milica Stojanovic, Paul van Walree

arXiv:2609.03207v1eess.SP

TL;DR

Underwater acoustic communications lacks standardized, widely available channel models despite the need for realistic field-based evaluation. The paper presents an open-access library of time-varying channels, noise models, and replay software, and reports that replayed channels preserve communication-relevant characteristics of measured data.

  • Problem

    Underwater acoustic communications has no widely accepted standard channel models, while realistic field data and experimental facilities remain costly or inaccessible.

  • Method

    The paper constructs an open-access library from field recordings using time-varying impulse responses with multipath, Doppler, and site-specific noise models, plus a replay interface.

  • Results

    Validation against original at-sea recordings confirms that replayed channels faithfully preserve communication-relevant characteristics across the measured data.

  • Takeaways & Limitations

    The library provides a uniform and reproducible framework for comparing communication, networking, and signal-processing algorithms.

Abstract

from arXiv · show

The development of communication systems critically depends on realistic channel models, yet there are no widely accepted standards in the underwater acoustic communications community. The situation is in stark contrast to terrestrial radio communications, where channel models have been standardized and are widely available. To address this gap, we present an open-access library of underwater acoustic channels derived from field experiments conducted across geographically distinct locations and varying propagation conditions, including shallow and deep water, short and long range, and fixed and mobile platforms. Each channel is described by a time-varying impulse response extracted from at-sea recordings using an adaptive algorithm that separately identifies the multipath structure and the Doppler-induced phase and delay drift. Each channel is also accompanied by a site-specific ambient noise model, which captures the statistics of colored Gaussian noise and impulsive noise. Spatial diversity reception across an array of hydrophones is supported for most channels, while time diversity is included for single-hydrophone scenarios. The library is accompanied by a simple replay interface through which a user supplies an arbitrary transmit signal, passes it through a chosen channel, adds noise at a desired signal-to-noise ratio (SNR), and obtains the received signal. The models are validated by comparing the output SNR of a baseline receiver operating on replayed signals with its performance on the original at-sea recordings, demonstrating close agreement across all channels. The library, including all channel impulse responses, noise model parameters, and replay software, is freely available for download as an open-source package.

I. INTRODUCTION

Underwater acoustic research lacks standardized, widely accessible channel models despite the importance of realistic field conditions. The paper introduces an open-access library intended to support uniform, reproducible comparison of communication and signal-processing methods.

  • Motivation: Field testing remains essential but is costly, while equipment and signal recordings are not widely accessible.Oversimplified simulations can misrepresent underwater acoustic propagation characteristics.
  • Motivation: No standardized underwater acoustic channel models have emerged, unlike in terrestrial radio communications.The lack of common models limits the availability of a shared framework for research and comparison.
  • Purpose: The library is designed to test and compare candidate algorithms within a common, uniform, and reproducible framework.It is presented as an initial step toward adoption of standard underwater acoustic channel models, while field testing remains necessary for new advancements.
  • Library: The library provides field-derived, time-varying channel impulse responses and noise characteristics across diverse environments, ranges, platforms, and locations.The models include multipath and Doppler effects and are stored as matrices describing impulse-response evolution over time.
  • Library: Users can pass arbitrary signals through selected channels, add noise, and obtain received signals through a simple replay interface.The implementation is available as open-source code in MATLAB, Python, and Julia.

B. Red Channel

The repository’s channels were extracted from geographically diverse experiments spanning shallow and deep water, short and long ranges, and fixed or mobile measurement configurations. The experiments used varied receive arrays, probe signals, and propagation settings.

  • Red Channel: The Red channel was recorded near Singapore in November 2024 from a drifting boat using a three-channel vertical array.The transmitter-receiver range varied between 100 m and 400 m during the drifts.
  • Yellow Channel: The Yellow channel came from the 2011 KAM’11 experiment off Kauai, Hawaii, which studied interactions among ocean variability, acoustic propagation, and underwater communications.The experiment collected acoustic and environmental data for these interactions.
  • Purple Channel: The Purple channel was derived from the 2008 SPACE’08 experiment at the Martha’s Vineyard Coastal Observatory using multiple receive-array configurations.The configurations included 24-element arrays and a 12-element array with different inter-element spacings.
  • Green Channel: The Green channel was recorded in a Norwegian archipelago in November 2024 over a 270 m range using a single hydrophone.Repeated transmissions formed time-diversity channels for this single-hydrophone scenario.
  • Black and Pink Channels: The Black and Pink channels represent deep vertical-link and longer-range experiments in the Mariana Trench and off Kochi, Japan, respectively.The Pink transmission distance was around 14 km, while the Black transmitter was located 8718 m below the surface.
  • Brown Channel: The Brown channel spans approximately 3250 km across the Pacific Ocean between a source off San Diego and a receiver array east of Hawaii.Its receiver used a 20-element autonomous vertical line array in deep water.

III. CHANNEL EXTRACTION AND COMPRESSION

The extraction pipeline adaptively estimates a time-varying multipath impulse response while separately tracking phase and motion-induced delay drift. The resulting channel representation can then be compressed according to its temporal variation.

  • A. Basic Channel Estimation and Phase Tracking: Basic channel extraction models the complex baseband received signal using transmitted data, a channel impulse response, a separate phase term, and noise.The phase is separated because it typically varies more rapidly than the rest of the channel response.
  • A. Basic Channel Estimation and Phase Tracking: The received signal is demodulated, band-limited, sampled, and synchronized before impulse-response estimation.The sampling rate is fs = Ns/T = 1/Ts, and synchronization aligns the relevant cross-correlation peak at lag zero.
  • A. Basic Channel Estimation and Phase Tracking: The channel impulse-response vector spans the delay samples containing appreciable multipath energy and is indexed by observation time.The associated data vector uses a zero-padded transmitted sequence to support adaptive estimation.
  • A. Basic Channel Estimation and Phase Tracking: The adaptive algorithm tracks time variation in both the channel impulse response and phase as new samples arrive.The library uses LMS or RLS algorithms for channel impulse-response estimation and a PLL for phase updates.

B. Delay Tracking

The method separates Doppler-induced delay drift from the underlying time-varying channel by adaptively resampling the received signal before channel estimation. This produces drift-free channel estimates while retaining phase information for reconstruction.

  • Delay-drift problem: Motion causes the channel impulse response to drift and can eventually move multipath energy outside the pre-allocated channel window.The drift creates long-term storage problems and may require a longer impulse-response vector.
  • Channel model: The received signal is modeled with multipath gains and time-varying delays decomposed into common drift and path-specific deviations.The common delay term captures dominant drift when transmitter and receiver move toward or away from each other.
  • Drift-free extraction: Compensating time-varying delay leaves a drift-free channel response for estimation instead of the drifting response.The algorithm exploits the relationship between channel variation and time-varying delays.
  • Adaptive resampling: Delay tracking resamples the received signal using an adaptively estimated phase to compensate for time dilation or compression.The resampled signal and transmitted sequence are then supplied to the channel estimator.
  • Implementation: The full algorithm combines LMS- or RLS-based adaptive channel estimation with phase and delay tracking, applied independently across array elements.Array elements are synchronized to preserve inter-element timing.
  • Example: In the Blue-channel example, delay compensation suppresses response drift, keeping high-energy multipath within about 7 ms and all low-energy arrivals within about 17 ms.Without compensation, the response drifts across the 30 ms allotted delay window.

C. Compression

The library compresses channel impulse-response matrices to reduce storage requirements, using drift-free channel estimates and spectral occupancy to determine temporal sampling needs.

  • Storage challenge: Channel impulse responses are stored as matrices whose row count grows with observation duration, making compression essential for multi-hour recordings.Each row contains a channel vector, so long records can produce very large matrices.
  • Compression principle: Preventing channel drift slows apparent time variability and reduces the number of rows required for storage.The minimum temporal sampling rate can be determined from the scattering function or delay-Doppler spreading function.
  • Sampling example: Compared with roughly 10 kHz sampling in the uncompacted representation, the Blue-channel example implies a compression ratio of 5000.This reduction is described as substantial.
  • Library storage: All library channels are compressed and normalized, while phase samples stored at the original rate enable reconstruction of the full channel.Channels without delay tracking store the phase estimate instead.

IV. NOISE

The library models ambient noise beyond white Gaussian assumptions by generating site-specific colored Gaussian or impulsive noise with spatial and temporal dependencies.

  • Noise motivation: White Gaussian noise poorly represents ocean ambient noise because underwater noise is colored and has a frequency-dependent power spectral density.Ambient noise also exhibits directionality and spatiotemporal correlation across hydrophones.
  • Site-specific models: The library generates synthetic noise whose probability distribution and power spectral density resemble those observed in each environment.The approach supports Gaussian or impulsive noise and models dependencies across time and hydrophones.
  • Replay use: Stored noise parameters can be selected with a channel, scaled, and added to the received signal at a specified signal-to-noise ratio (SNR).The Brown channel lacks a site-specific model because its recordings contained only active transmissions.

A. Gaussian noise

The Gaussian-noise model estimates spatiotemporal covariance from multihydrophone recordings and synthesizes colored noise by filtering independent white Gaussian samples.

  • Model basis: Gaussian noise generation reduces to estimating the zero-mean noise covariance and generating random noise with the same covariance.Zero mean follows from hydrophones measuring dynamic pressure.
  • Covariance estimation: For M hydrophones sampled at fs, the model estimates band-filtered noise covariance from recorded sequences.The recordings occupy the same acoustic band as the communication signals used for channel extraction.
  • Covariance estimation: The covariance estimate includes cross-hydrophone and time-lag structure over lags from −L to L.L is the maximum discrete lag with non-negligible covariance.
  • Noise synthesis: Synthetic colored Gaussian noise is produced by generating standard spatiotemporal white Gaussian noise and applying spatiotemporal filtering.Filter or mixing coefficients can be obtained by covariance decomposition or optimization.
  • Noise synthesis: The synthesized noise combines mixing coefficients βij(kTs) with independent ηj(nTs) samples distributed as N(0, 1).The white samples are band-limited to the channel’s acoustic bandwidth.

B. Impulsive noise

The library models impulsive ambient noise with α-stable sub-Gaussian processes, estimating distribution and covariance parameters from recordings for synthetic generation.

  • The α-stable sub-Gaussian process models impulsive ambient noise and reduces to Gaussian noise when α = 2.Its characteristic exponent α controls the distribution family.
  • The method characterizes αSG noise using the characteristic exponent α and the covariance of an underlying multivariate Gaussian distribution.The exponent can be estimated with a fractile estimator.
  • Generated impulsive noise uses symmetric α-stable variates and mixing coefficients βij(kTs) chosen to match the ambient-noise covariance.The stable-distribution stability property supports this linear mixing approach.
  • Stored α and βij(kTs) parameters are used to generate impulsive noise samples with distributions similar to the recordings.The generator draws stable random variates and mixes them using the estimated coefficients.
  • The Red channel uses colored impulsive noise, while the other listed site-specific models are colored Gaussian noise; Brown has no site-specific model.Generic colored Gaussian noise can be used for Brown.

C. Illustration of Noise Models

Noise and channel replay models are assessed against recordings using spectral, distributional, spatial, and receiver-performance comparisons.

  • Noise-model illustrations: Noise parameters were estimated from signal-free ambient recordings after bandpass filtering matched to each communication signal or a wider band for impulsive noise.The Red impulsive-noise model used the wider-band filtering condition.
  • Receiver validation: Validation compares a multichannel decision-feedback equalizer on at-sea recordings and replayed signals using output SNR as the performance metric.The procedure adds site-specific synthetic noise to replay signals at the at-sea SNR.
  • Noise-model illustrations: The Blue and Red examples show close agreement between recorded and generated noise in spectral characteristics, distributions, and time-domain samples.Blue uses Gaussian noise, whereas Red uses impulsive noise whose pseudo PSD is compared with the recording.
  • Receiver validation: 1 dB for single-hydrophone reception and 2 dB for array processing are the average differences between replay and at-sea output SNRs.Replay SNRs are generally slightly higher, indicating somewhat more benign replay channels for the modulation scheme used.
  • Receiver validation: The validation uses the same receiver algorithm and parameter settings for at-sea and replay signals, while covering channels with varied complexities and performance levels.Receiver parameters were coarsely adjusted per channel without fine-tuning each case.
  • Spatial replay validation: The Purple vertical-array delay-angle spectra show close agreement in dominant arrival angles and relative intensities between at-sea and replayed signals.This comparison tests preservation of inter-element delay structure.

VI. USING THE LIBRARY

The library provides replay and noise-generation functions that transform user signals through stored time-varying channels while preserving supported array and temporal structure.

  • Library functions: The library offers replay, unpack, and noisegen functions for channel processing, impulse-response reconstruction, and synthetic noise generation in MATLAB, Python, and Julia.Replay passes a user-designed signal through measured channel data.
  • A. Channel Replay: The replay function accepts a passband input and returns the corresponding received signal as if transmitted through the measured channel.The output is a matrix whose columns contain time-domain samples for selected hydrophone elements.
  • A. Channel Replay: The input is converted to baseband, lowpass-filtered, and resampled to match the channel’s delay-domain sampling rate before time-varying convolution.The stored response is decompressed from its compressed time grid to the signal time grid.
  • A. Channel Replay: The replay pipeline applies stored phase estimates, restores delay drift through spline interpolation, and resamples and up-converts the output to passband.These operations reconstruct the time-varying channel behavior from the stored representation.
  • A. Channel Replay: Unit-power normalization preserves power differences across array elements and time, while multichannel replay applies each element’s stored channel estimate independently.Synchronizing elements to the first during extraction preserves inter-element delay structure.
  • Noise addition: Noise is added after channel replay, with σn adjustable to obtain a desired SNR; impulsive noise uses pseudo-power based on its α-stable scale parameter.The normalization preserves natural array-element power differences.
  • A. Channel Replay: The replay command accepts input, sampling frequency, hydrophone selection, channel data, and an optional start index controlling the channel-trace starting point.If start is omitted, a random starting time is selected.

B. Noise Generation

The noisegen function generates synthetic underwater acoustic noise and supports three modes of operation.

  • The noisegen function supports three modes of operation for generating synthetic underwater acoustic noise.

1) Generic colored Gaussian noise:

The library provides noise generation for experimentally characterized underwater acoustic environments, including colored Gaussian and impulsive noise with spatial and temporal structure. Channel unpacking reconstructs time-varying impulse responses from compressed representations, including phase and delay drift when tracked.

  • Generic colored Gaussian noise:: Experimentally measured noise statistics generate colored Gaussian noise with the spatiotemporal correlation structure observed during channel measurements.The model uses estimated β coefficients and related noise parameters, with α = 2.
  • Generic colored Gaussian noise:: For α < 2, the noise generator produces colored impulsive noise from a symmetric α-stable distribution with captured spatiotemporal correlation.The generator selects between Gaussian and impulsive noise based on α.
  • Channel unpacking: The unpack function reconstructs the full time-varying impulse response from its compressed representation for visualization or custom signal processing.Its output can be represented as a K × M × Nt array, indexed by delay taps, hydrophone elements, and time snapshots.
  • Channel unpacking: With delay tracking enabled, unpacking resamples channel estimates, restores residual phase, and reinserts delay drift by interpolation.Without delay tracking, it resamples the estimates and restores phase without reinserting drift.
  • Channel unpacking: The unpacked impulse response contains channel drift, and its delay-time magnitude heat map reveals temporal evolution of the multipath structure.This representation exposes how multipath changes over time.

VII. GALLERY OF CHANNELS

The gallery demonstrates that the library covers diverse underwater acoustic environments and propagation conditions, with impulse responses exhibiting environment-specific multipath, Doppler, reverberation, and scattering. The collection supports reproducible evaluation and is intended to expand through community contributions, while statistical models remain future work.

  • VII. GALLERY OF CHANNELS: Figure 11 presents time-varying impulse responses from eight channel collections spanning geographically distinct environments and experimental configurations.The examples include mobile, coastal, wind-dominated, fjord, Mariana Trench, and trans-Pacific channels.
  • VII. GALLERY OF CHANNELS: The channels span frequencies from 75 Hz to 25 kHz and ranges from 60 m to 3250 km.These examples cover both short- and long-range propagation conditions.
  • VII. GALLERY OF CHANNELS: Impulse responses reveal propagation characteristics shaped by water depth, distance, boundary conditions, geology, frequency band, and platform mobility.Examples include rough-surface scattering, transmitter-motion Doppler, fjord reverberation, and a benign vertical link.
  • VII. GALLERY OF CHANNELS: The collection provides a testbed for algorithm development and performance evaluation across challenges ranging from dispersive shallow-water channels to stable deep-water paths.It is intended to represent a wide spectrum of underwater acoustic communication scenarios.
  • VIII. CONCLUSION: Validation against original at-sea recordings confirms that replayed channels preserve communication-relevant characteristics of the measured data.The library combines field-derived time-varying impulse responses, site-specific noise models, and a reproducible replay interface.
  • VIII. CONCLUSION: Researchers can contribute new channels from their own field recordings using open-sourced channel-estimation and noise-extraction scripts.The library is designed to grow through contributions from different environments and experimental configurations.
  • VIII. CONCLUSION: Future work is to complement replay-based channels with statistical models that can generate unlimited time series and explore conditions beyond individual recordings.The paper identifies standardization of these statistical models as a natural next step.
Loading 2609.03207v1…