Seismic Data Deblending Using Non-Blended Reference Dataset
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional deblending methods for seismic data are inefficient and reliant on specific data acquisition techniques, failing to effectively separate reflections from blended datasets, particularly when dealing with low-amplitude cross-talk noise and primary energy attenuation.
Innovation Solution
A method utilizing a non-blended dataset from the same survey area to calculate a model dataset, which is then used to deblend a blended dataset by interpolating and subtracting cross-talk noise, employing sparseness weights and filters to mitigate noise and preserve primary signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If conventional deblending methods are used, then some cross-talk noise can be removed, but primary energy is attenuated and low-amplitude cross-talk noise remains ineffective
Solution Approach 1:
The patent introduces a non-blended dataset as an intermediary reference to guide the deblending process. This reference dataset provides clean signal templates that mediate between the blended input and the deblended output, enabling selective removal of cross-talk without attenuating primary energy. The non-blended data acts as a mediator that preserves signal integrity while eliminating noise.
Solution Approach 2:
The patent changes the parameter of data representation by transforming blended seismic data into a domain where sparse signals can be identified and separated. By applying sparsity constraints and using the non-blended reference to guide parameter optimization, the method achieves effective separation of overlapping signals while preserving primary energy components.
2Loss of time
If simultaneous source acquisition is used, then survey time is reduced, but blended data requires additional deblending processing
Solution Approach 1:
The patent applies preliminary action by using a non-blended dataset acquired in the same survey area as a reference guide before performing deblending. This pre-existing clean data establishes signal templates and characteristics in advance, making the subsequent deblending process more efficient and less complex. The preliminary reference data simplifies the separation task by providing known good signal patterns.
Solution Approach 2:
The patent uses copying by creating a model of the expected signal behavior from the non-blended dataset and applying it to the blended data. The non-blended data serves as a template that is copied and adapted to guide the extraction of individual source signals from the blended mixture, reducing the complexity of the inverse problem.
3Object-affected harmful factors
If impulsive denoise methods are used, then strong cross-talk energy is removed, but low-amplitude cross-talk noise is not removed and primary energy is attenuated
Solution Approach 1:
The patent implements feedback by iteratively comparing the deblended result with the non-blended reference dataset and adjusting the separation parameters accordingly. The non-blended data provides feedback on what the correct signal should look like, allowing the algorithm to refine its separation and avoid attenuating primary energy while removing both strong and low-amplitude cross-talk noise.
Solution Approach 2:
The patent changes parameters by using sparsity constraints and optimization criteria that are guided by the non-blended reference. Instead of fixed thresholding, the method dynamically adjusts separation parameters based on the reference data, enabling precise removal of cross-talk across different amplitude levels while preserving primary signals.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
A non-blended dataset related to a same surveyed area as a blended dataset is used to deblend the blended dataset. The non-blended dataset may be used to calculate a model dataset emulating the blended dataset, or may be transformed in a model domain and used to derive sparseness weights, model domain masking, scaling or shaping functions used to deblend the blended dataset.