Sensitivity of the deconvolution of acoustic transients to Green’s function mismatch

Abstract

In this paper various measurements of acoustic time series recorded in the Atlantic over source-to-receiver ranges of 600 to 12 900 m are analyzed. The transmitted source signature with a monitor hydrophone mounted on the source array was also recorded. The signature distortion introduced by propagation effects was treated by the use of single-channel deconvolution. In situations where the Green’s function structure is simple (e.g., direct arrival and surface reflection), single-channel deconvolution gave satisfactory results. When multipath effects (due to interaction with layered bottom sediments) were present, it was difficult to get a good source estimate. A way was developed of perturbing the Green’s functions such that the source estimates were guaranteed to improve. In most cases it was found that very small changes could produce significant improvement in the source estimates. This sensitivity was quantified by using the correlation coefficient. This sensitivity is not to be confused with the well-known fact that the single-channel deconvolution problem is ill-posed. That issue was treated separately.

A waveguide invariant adaptive matched filter for active sonar target depth classification

“…This work uses the STFT of the data computed using a window function w(t) of length T and centered at s sðx; s; r; zÞ ¼ ð 1 À1 w t À s T sðt; r; zÞ expðÀjxtÞdt: (3) However, if the length of the window is larger than the delay spread of the channel and s is chosen around the target range based on the two-way travel time s $ 2r=c, where c is the average water column sound speed, Eqs.…”

“…1,2 However, these methods are often sensitive to mismatch in the channel Green's function. 3 Gupta et al avoid blind deconvolution by using training data to jointly estimate the channel impulse response and classify the received signal. 4 Other techniques have been proposed which use target features that are invariant to effects of multipath propagation.…”

Classification of distant targets situated near channel bottoms

“…One direct method to recover the target's scattered signatures is referred to as deterministic deconvolution. 7,8 This method is sensitive to the Green's-function mismatch. A more generalized method was developed recently by using the minimum entropy deconvolution ͑MED͒ method 9 to estimate the channel response function, which requires that the channel's Green's function be sufficiently ''sparse.''…”

Blind Deconvolution to Improve Classification of Transient Source Signals in Multipath

“…Given the inherent uncertainties in underwater environmental data, it is virtually impossible to model a Green's function to represent multipath with sufficient accuracy for deconvolution to produce the correct source signal [1]. However, the source signal may be estimated using blind deconvolution.…”

“…The time series arriving at they'th hydrophone of an array is X: = S*gj+ Hj , (1) where s is the signal leaving the target of interest, g is the Green's function describing the propagation from the target to thejth hydrophone, and n is the noise on thejth hydrophone.…”