Several years prior to this project’s initiation,  it was determined that discrimination performance using single-component electromagnetic induction (EMI) sensors in the field had been uniformly poor relative to expectations, predominantly due to stringent requirements on positional accuracy and signal-to-noise ratio (SNR). This led to the development of a number of multi-transmitter, multi-receiver EMI sensors that could be deployed in a cued-interrogation mode so that data could be collected without moving the platform. Initials tests with these systems have shown excellent discrimination potential. However, at many sites, it is believed that the most cost-effective discrimination strategy will be to deploy a time-domain EMI one-pass detection and discrimination system where positional and orientation accuracy, data density, and SNR are maximized.

The objective of this project was to conduct a feasibility study for a cart- or towed-array system capable of discriminating buried UXO during a moving one-pass survey.

Technical Approach

The researchers focused on estimating the positional and orientation accuracies that can be achieved in a dynamic survey mode and on simulating the discrimination performance of a number of candidate EMI sensors when subjected to different position and orientation errors.

For sensor positioning, the project team used the commercial off-the-shelf, Novatel Synchronized Position Attitude & Navigation (SPAN) system, which comprises a GPS receiver combined with a Honeywell HG1700 tactical grade Inertial Motion Unit (IMU). The data are post-processed in a commercially available package called Inertial Explorer.


Two positional tests were conducted, the first using a Rotating Amusement Park (RAP) carousel and the second using an EM61 towed array. The RAP was instrumented with a precision optical encoder, and the SPAN system was installed on a tilt-table at a fixed position on the RAP. The known distance of the SPAN system from the axis of revolution of the RAP ride, the tilt table angle, and the optical encoder information provided accurate ground-truth. The GPS and IMU data from the SPAN were augmented by an optical encoder attached to a bicycle wheel that was attached to the RAP and which was in constant contact with the ground. For one run, when the system was moving at a brisk 2 meters per second, the position estimated from the IMU/GPS combination had a root-mean-square (RMS) error of between 1.1 and 1.8 cm.

For the second position test, a laser prism was mounted on the rear left corner of an EM61 towed-array, about 1.5 m away from the SPAN system, which was mounted on the tow-bar in line with the center of the towed-array. The SPAN-predicted position of the prism was compared to the prism position measured with a Trimble SCS930 Universal Total Station. Data were collected at various speeds between 0.2 and 1.3 meters per second, on both smooth and rough ground, with the array traveling along, straight, or curved paths and with ramps sometimes present that one side of the cart had to travel over. Among all the tests, the RMS horizontal error varied from a minimum of 0.6 cm to a maximum of 1.7 cm. These represent upper bounds on the horizontal positional error of the SPAN system. If the error was entirely due to the SPAN heading, the RMS heading error would vary from a minimum of 0.2° to a maximum of 0.6°. RMS errors of the 3-D positions (including elevation) varied from 0.8 to 3.2 cm. The runs with larger errors typically exhibited systematic biases in the elevation angles. This indicates that there may be 1-2 cm magnitude systematic differences in the SPAN elevations along adjacent transect paths.

Numerical investigation of simulation performance used two approaches. The first was semi-analytical and based on estimating the expected errors in the positions and polarizabilities of a buried metallic object. The second was based on Monte-Carlo simulations of systems subjected to different amounts of sensor, position, and orientation noise. Five different sensor systems were simulated, including two Geonics EM63 equivalents, one with a single receiver coil, the other with three orthogonal receiver coils. The remaining three systems were based on the 25 transmitter / 25 receiver TEMTADS system developed at the Naval Research Laboratory. The first was deployed in the standard cued mode, with the other two deployed in dynamic mode, one with a single large transmitter around the array and the other with four 1 m by 1 m transmitters that fire sequentially. Both the semi-analytical and Monte-Carlo approaches demonstrated the excellent performance of the cued-system. They also revealed that the dynamic variants of that system have significantly improved performance over an EM63 equivalent. In particular, the multiple receiver coils make the TEMTADS more tolerant of position and orientation errors than the EM63. The four-transmitter TEMTADS with 5 cm position error and 2° orientation error outperforms an EM63 with 1 cm and 1° error.


The project team hypothesizes that, at many sites, one-pass detection and discrimination with a suitably modified TEMTADS would be feasible. Furthermore, depending on anomaly density, this one-pass survey would be more cost-effective than the two-pass mode required when the TEMTADS is deployed in a cued-mode.