Blind deconvolution for robust signal estimation and approximate source localization.
ABSTRACT Synthetic time reversal (STR) is a technique for blind deconvolution in an unknown multipath environment that relies on generic features (rays or modes) of multipath sound propagation. This paper describes how ray-based STR signal estimates may be improved and how ray-based STR sound-channel impulse-response estimates may be exploited for approximate source localization in underwater environments. Findings are based on simulations and underwater experiments involving source-array ranges from 100 m to 1 km in 60 -m-deep water and chirp signals with a bandwidth of 1.5-4.0 kHz. Signal estimation performance is quantified by the correlation coefficient between the source-broadcast and the STR-estimated signals for a variable number N of array elements, 2 ≤ N ≤ 32, and a range of signal-to-noise ratio (SNR), -5 dB ≤ SNR ≤ 30 dB. At high SNR, STR-estimated signals are found to have cross-correlation coefficients of ∼90% with as few as four array elements, and similar performance may be achieved at a SNR of nearly 0 dB with 32 array elements. When the broadband STR-estimated impulse response is used for source localization via a simple ray-based backpropagation scheme, the results are less ambiguous than those obtained from conventional broadband matched field processing.
- SourceAvailable from: Paul Hursky[show abstract] [hide abstract]
ABSTRACT: The authors demonstrate MFP using data-derived modes and the sound speed profile, using no a priori bottom information. Mode shapes can be estimated directly from vertical line array data, without a priori knowledge of the environment and without using numerical wave field models. However, it is difficult to make much headway with data-derived modes alone, without wave numbers, since only a few modes at a few frequencies may be captured, and only at depths sampled by the array. Using a measured sound speed profile, the authors derive self-consistent, complete sets of modes, wave numbers, and bottom parameters from data-derived modes. Bottom parameters enable modes to be calculated at all frequencies, not just those at which modes were derived from data. This process is demonstrated on SWellEx-96 experiment data. Modes, wave numbers, and bottom parameters are derived from one track and MFP based on this information is demonstrated on another track.The Journal of the Acoustical Society of America 05/2001; 109(4):1355-66. · 1.65 Impact Factor
- [show abstract] [hide abstract]
ABSTRACT: Many acoustical applications require the analysis of a signal that is corrupted by an unknown filtering function. Examples arise in the areas of noise or vibration control, room acoustics, structural vibration analysis, and speech processing. Here, the observed signal can be modeled as the convolution of the desired signal with an unknown system impulse response. Blind deconvolution refers to the process of learning the inverse of this unknown impulse response and applying it to the observed signal to remove the filtering effects. Unlike classical deconvolution, which requires prior knowledge of the impulse response, blind deconvolution requires only reasonable prior estimates of the input signal's statistics. The significant contribution of this work lies in experimental verification of a blind deconvolution algorithm in the context of acoustical system identification. Previous experimental work concerning blind deconvolution in acoustics has been minimal, as previous literature concerning blind deconvolution uses computer simulated data. This paper examines experiments involving three classical acoustic systems: driven pipe, driven pipe with open side branch, and driven pipe with Helmholtz resonator side branch. Experimental results confirm that the deconvolution algorithm learns these systems' inverse impulse responses, and that application of these learned inverses removes the effects of the filters.The Journal of the Acoustical Society of America 11/2003; 114(4 Pt 1):1988-96. · 1.65 Impact Factor
- [show abstract] [hide abstract]
ABSTRACT: To find the position of an acoustic source in a room, the relative delay between two (or more) microphone signals for the direct sound must be determined. The generalized cross-correlation method is the most popular technique to do so and is well explained in a landmark paper by Knapp and Carter. In this paper, a new approach is proposed that is based on eigenvalue decomposition. Indeed, the eigenvector corresponding to the minimum eigenvalue of the covariance matrix of the microphone signals contains the impulse responses between the source and the microphone signals (and therefore all the information we need for time delay estimation). In experiments, the proposed algorithm performs well and is very accurate.The Journal of the Acoustical Society of America 02/2000; 107(1):384-91. · 1.65 Impact Factor
Shima H Abadi