Types of GPR radargram signal processing
Performing a GPR survey is only part of the work. What follows is the GPR data analysis. There’s a couple of data processing techniques that can and should be performed before proper conclusions can be drawn from the raw signals. Some of them are essential and others depend only on the required level of fidelity. Let’s start.
Post processing gain
Most raw field data collected with GPR needs post processing gain to be able to visualize the radar pulses. Most GPR manufacturers today record un-gained radargrams without any filters, recording just the native raw field pulses. A few manufacturers allow the user to record gain into the raw field data as well as to apply filters.
All data recorded without gaining in the field will require post processing gain. Even for field data where the manufacturer allows for gains to be recorded in the raw field data, some post processing gain adjustment.
Bandpass filtering
Bandpass filtering is a process to remove unwanted frequencies in the raw radargrams. Bandpass filtering requires that radargram pulses first be converted to the spectral domain. All recorded signals can be decomposed into a combination of individual pure frequencies which have different amplitudes and phases. The combination of the amplitudes at different frequencies and the phase that those frequency components arrive at the receiving antenna defines the unique radar signals that get recorded.
Although manufacturers often list transmit antennas as a single frequency most GPR antenna emit energy over a large band of frequencies. The central frequency of this band where the peak energy is transmitting from the antenna is often reported by the manufacturers as the antenna frequency.
Background removal
GPR radargrams often have a constant noise infiltrating the recording of raw pulses. The noise manifests itself as bands of constant horizontal reflections across the radargrams. To remove the horizontal bands found in raw radargrams, one of the most popular filter processes is to calculate the average pulse across the entire radargram and then subtract this average pulse from each individually recorded pulse. The motivation for applying this filter is that it can help to illuminate changes only in the pulse that is different from the average pulse across the profile.
Hilbert transform
The simpler way to understand the Hilbert transform is to think of a rectified signal — a pulse’s absolute value — that connects all the peak responses.
The Hilbert transformed radar pulses are useful in imaging situations to simply show regions of weak or strong reflections and where the imaging is not dependent on the phase of the radar pulse. The Hilbert transformed pulses more closely represents the structure of the subsurface than a raw pulse since the oscillating components of the GPR pulse have been removed. A reflection from any flat interface for example, will be mapped as a single strong reflection.
Deconvolution
Deconvolution is a signal processing method to help reduce multiple reflections and echoes recorded on radargrams as well as to minimize the effects of the transmitted pulse. The pulses that are transmitted by GPR antenna have a defined impulse response function. One kind of deconvolution filter is designed to remove the transmitted impulse response function from the recorded radargram.
Resampling
Resampling is a basic signal process that has to be done to radargrams before any other signal process. GPR profiles are sometimes collected over the ground with user inserted markers used to set navigation tags in the radargram. The GPR control units are normally set to time modes, and a constant number of scans are recorded per unit time.
If the velocity of the antenna over the ground varies, the number of scans per unit distance marker can vary. Before any processes such as migration or other filters that require knowledge of the spatial density of the scans, it is normally required to resample the radargram scans to a constant number of scans per unit distance.
Smoothing / Stacking
Radargram scans can have high frequency noises which can be removed by bandpass filtering. Sometimes, what appears to be completely random noise has in fact small signals in the scans which when stacked together will remove the random noises and enhance the small recorded signals.
Often the GPR control units can be adjusted to stack signals. Some of the commercial control units mistakenly say stacking when in fact they are actually doing a smoothing process — where N number of scans are added over the ground in a moving average. It is imperative to make GPR surveyors understand what the equipment being used is actually doing.
To “instantaneously” add N number of scans in the exact same location is the traditional (seismic) stacking process, which can dramatically increase the signal-to-noise ratios. However, stacking has also been loosely used (by some manufacturers) when the actual process is a smoothing operation.
Taken from the book “Geotechnologies and the Environment”
Thanks for reading! If you found our post to be useful, be sure to give a round of claps! Also, check our website for more info on our products and services!
