The seismic data used for this study consist of 201 CMP gathers recorded along the Blake Outer Ridge with a non-linear streamer resulting in a 100 m group spacing at the near offset traces and a 50 m group spacing at the far offset traces (Figure ).
The seafloor reflection occurs at about 4.4 seconds, indicating a water depth of over 3 km, and decreases in depth gradually along the line (Figures and ). It is followed by a strong bottom simulating reflection at approximately 5.2 seconds vertical traveltime. A fairly bright ``flat spot'' reflection is apparent beneath the BSR at 5.4 seconds, which does not parallel the seafloor or BSR reflection events. Examining the raw prestack data, a traveltime kink is very striking in all reflections near the central offset of the CMP gathers. This kink occurs exactly at the transition zone between the nonlinear cable group spacing of 100 m at the near offsets, and 50 m group spacing at the far offsets. Adding the far offset traces in pairs to simulate a constant cable group spacing of 100 m at all offsets eliminates the traveltime kink in the data (Figure ), but causes adverse waveform and amplitude changes with offset.
The group-corrected data have less spatial resolution than the original data, based on a decrease from 48 to 36 traces. Furthermore, the group summation causes a significant loss in temporal frequency content at the far offsets, due to spatial averaging of moveout delayed reflections across a twice longer effective group array. One possibility to compensate for this low frequency group array response is to filter (deconvolve) the far offset data in the k-x domain. However, the decreased number of summed far offset traces result in a very short data series, which makes accurate spectral estimates and the application of spatial deconvolutional filters difficult. Another possibility to account for this frequency loss is to deconvolve the far offsets as function of the Snell parameter p in the slant stack domain. Unfortunately, the small number of traces tends to introduce large edge effects in the slant stack spectrum. These edge effects could be minimized by using a least-squares slant stack which, however, would smear the notches in the spectrum, and the resulting deconvolution would overemphasize those portions of the spectrum. Therefore, it seems unreasonable to correct the raw data for the different group spacings by summing the far offset traces in pairs. A simple linear interpolation of the near offset traces after NMO, with source wavelet deconvolution and amplitude calibration to compensate for the group array response appears to be a better method for suppressing the nonlinear cable effects.