3-D GPR Survey with a Modular System: Reducing Positioning Inaccuracies and Linear Noise Lieven Verdonck and Frank Vermeulen Department of Archaeology Ghent University Ghent, Belgium Lieven.Verdonck@UGent.be Abstract—Recently, the use of ground-penetrating radar (GPR) arrays with a large number of antenna elements in a fixed configuration has become more common. The investment needed for these systems is significant. Although gradually expandable modular systems, consisting of antennas which can be used independently, do not match the fast acquisition of detailed datasets by means of multi-channel arrays, they can help finding a compromise between increased acquisition speed and (limited) resources. In modular systems, the separation between transmitter-receiver pairs is often larger than the sampling distance prescribed by the Nyquist theorem. As a consequence, additional profiles have to be recorded in between, which requires a high positioning precision. As a completely identical response for the different antennas in an array is difficult to achieve, stripes can occur in the horizontal slices, especially when ringing occurs. This complicates the interpretation of features in the direction of the survey lines. In this paper, a three- dimensional frequency-wavenumber filter is proposed, consisting in a combination of a circular filter and a fan filter. The application of this filter to GPR data collected at the Roman town Mariana (Corsica, France) showed a reduction of the stripe patterns, allowing a more reliable characterization of subtle archaeological structures. Keywords-3-D ground-penetrating radar; modular antenna array; ringing; 3-D frequency-wavenumber filtering; archaeological prospection I. INTRODUCTION In the last few years, the use of multi-channel ground- penetrating radar (GPR) systems has become more common. Also in archaeology, field tests have been conducted with stepped frequency continuous wave and pulsed antenna arrays, enabling rapid data collection over large areas at sampling intervals approaching the Nyquist sampling theorem [1],[2]. Most of these arrays consist of antennas fixed in a large frame, arranged so that the cross-line spacing is smaller than the physical size of the antenna. For example, each receiver antenna can be made to record signals of two adjacent transmitter antennas [1],[3], or the transmitter-receiver combinations are rotated 45° with respect to the survey direction [4]. Modular systems, where each single antenna can be used independently, do not equal the combined acquisition speed and dense sampling of multi-channel GPR systems. However, when the available resources do not allow the high levels of investment demanded by large multi-channel systems, survey speed can be increased by gradually expanding a modular system while spreading investment over time. Furthermore, large multi-channel arrays can be difficult to operate in restricted spaces and on sites with a rough surface or otherwise inaccessible for systems towed behind a vehicle. As a consequence, it would be necessary to have a maneuverable single-channel instrument at one’s disposal, as well as a multi- channel system [5]. Modular systems can be a solution when surveys often have to be carried out in difficult environments. This paper presents a full-resolution 3-D survey with a pulsed modular GPR system, conducted at the Roman town Mariana in Corsica (France). The required positioning precision is discussed and three-dimensional frequency-wavenumber (f-k) filtering is applied to suppress stripes along the survey lines, which can be a consequence of using multi-antenna systems. II. DATA ACQUISITION AND POSITIONING An area of 85 by 35 m was surveyed in October 2010 with a Sensors & Software Spidar network consisting of three pulseEKKO PRO 500 MHz antennas. A single antenna casing is approximately 0.23 m wide. The antennas were fixed onto a wheeled frame towed behind an all-terrain vehicle (ATV), so that the distance between the antenna midpoints was 0.25 m (Fig. 1). Figure 1. Three antennas fixed onto a frame towed behind an ATV, with a distance between the antenna midpoints of 0.25 m. The total station prism is mounted as lowly as possible above the center of the middle antenna, to minimize errors due to uneven topography. An odometer wheel triggers the system every 0.05 m.