Noise Attenuation

Swell noise attenuation
Spectrum offers several methodologies for the attenuation of swell noise, and often combines them in an iterative sequence to reduce the effects of the noise without harming the primary energy.

One of the most commonly used methods works in the time-frequency domain. The transform is separated into amplitude and phase components for each frequency sub-band, and the median spectral amplitude within each frequency sub-band is calculated for the ensemble. A threshold value is derived from these and compared to each sample. Any samples that exceed the threshold by a user controlled level are set to the threshold value. This method can be very effective at removing noise bursts in the gathers without affecting nearby samples or traces. We offer two modules that work by this method (NOISERM and TFCLEAN)

SwellNoiseB4
SwellNoiseAft

Example of swell noise removal achieved through the use of frequency sub-band thresholding

Swell2-11
Swell2-12


Example of swell noise removal. Frequency sub-band thresholding measures the amplitude envelope of frequencies attributable to signal and scales the frequencies attributable to noise relative to them

A second method (SUPPRESS) decomposes seismic traces into noise and signal components by filtering the user defined frequency band (noise component) by a band-pass filter and subtracting the filter result from the original input trace (signal component). The signal and noise envelopes are calculated as a complex trace attribute, which is optimally smoothed over a user defined window. The envelope of the noise component is compared to that of the signal and in time zones where the noise exceeds a specified level the noise component is scaled down to match the signal envelope level. Both components are then summed to produce the output traces. This method is useful for organized noise such as swell noise, ground roll and air blast.

Spectrum also has an option in their FX filter module to remove swell noise by working on a limited bandwidth. It can be regarded as a Wiener noise removal filter. A prediction filter is applied to the data and discrepancies between predicted and actual signal are removed as noise. This can work very effectively when applied several times in different domains.

FKxKy Filtering
FK filters can be specified as polygons or fans and can be applied in either the X-T or FK domain. Polygons are specified in terms of F and K co-ordinates. Fans are defined by “pass” and “reject” slopes in ms per trace. The level in the reject zone can be defined in terms of “dB down” and has options for spatial and temporal padding to unwrap noise.

FK filters are usually applied to pre-stack data either in the shot or receiver domain for noise attenuation or on CDP gathers for multiple removal. For land data, statics are usually applied to a near surface floating datum and NMO applied prior to F-K filtering is applied – this prevents attenuation of primary data along the K0 axis. An AGC can be applied to the data prior to applying the F-K filter and removed afterwards.

NoiseAttFK-BEFORE2
NoiseAttFK-AFTER2

The standard FK filter is a 2D filter – FKK filtering applies a 3D filter. When 3D land data is acquired a shot record is fired into a number of parallel receiver lines with the source point typically towards the center of the spread as shown in this diagram.

FK Noise Shot + Cone Diagram

In this case, the noise is a 3 dimensional problem emanating from the source point equally in all directions and as a result the noise is a cone in 3D space. Each of the receiver lines represents a 2D vertical slice through the noise cone.

There are two ways of attacking this noise in FK space, either by treating each individual receiver line as a 2D sub-line, or treating the whole record as a 3D problem. Since the noise is different on each receiver line as azimuth and offset are varying irregularly it is not possible to design a simple 2D FK filter which can attack all of the noise. It is much better to treat the whole record as one and look for a 3D solution. However, a problem arises in that the data is poorly sampled in the cross-line direction (perpendicular to the receiver line direction) and as a result severe aliasing occurs in the cross-line direction limiting the effectiveness of FK filtering.

The spatial sampling problem has been overcome with the use of cross-spread gathers which are a truly 3D pre-stack gather with even spatial distribution in both inline and cross-line directions. Following a sort to this domain, three dimensional FK filters can be effectively and efficiently applied overcoming the aliasing problems.

The pre-filtering sort job takes all the shot traces from a single source line which contributes to a single receiver line and groups them together. Within the cross-gather record the traces are rearranged into midpoint inline by cross-line order to form a sub-cube of data.

InlineCrosslineDiagram

The data is FKK filtered by a combination of SPA modules before being sorted back to the original shot domain as acquired.

FKK-BEFORE
FKK-AFTER


Example of a shot record before and after FKK filtering in the cross-gather domain. Note the linear noise which has been removed from the upper shot record. This application of filtering was achieved on data with NMO applied. Image courtesy of Tiway Oil.

FKK1
FKK2


Shot domain FK filter application (“Before”) vs FkxKy filter application (“After”). Image courtesy of Tiway Oil.

Diffractive Noise Attenuation
The diffracted noise is treated as a spike, on the assumption that this noise is higher amplitude and frequency than the data around it. The despike is applied to the NMO corrected gathers such that any non-hyperbolic signal is removed. The despike threshold is set by the user and the application window contoured to follow multiple horizons (typically 2 X W.B.). The application is frequency dependent so that only those frequencies found within the multiple data are attenuated.

DiffractedNoise1
DiffractedNoise2


A Radon de-multiple stack before and after diffracted noise attenuation